Commit | Line | Data |
---|---|---|
6aa8b732 AK |
1 | /* |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * | |
4 | * This module enables machines with Intel VT-x extensions to run virtual | |
5 | * machines without emulation or binary translation. | |
6 | * | |
7 | * Copyright (C) 2006 Qumranet, Inc. | |
9611c187 | 8 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
6aa8b732 AK |
9 | * |
10 | * Authors: | |
11 | * Avi Kivity <avi@qumranet.com> | |
12 | * Yaniv Kamay <yaniv@qumranet.com> | |
13 | * | |
14 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
15 | * the COPYING file in the top-level directory. | |
16 | * | |
17 | */ | |
18 | ||
85f455f7 | 19 | #include "irq.h" |
1d737c8a | 20 | #include "mmu.h" |
00b27a3e | 21 | #include "cpuid.h" |
d62caabb | 22 | #include "lapic.h" |
e495606d | 23 | |
edf88417 | 24 | #include <linux/kvm_host.h> |
6aa8b732 | 25 | #include <linux/module.h> |
9d8f549d | 26 | #include <linux/kernel.h> |
6aa8b732 AK |
27 | #include <linux/mm.h> |
28 | #include <linux/highmem.h> | |
e8edc6e0 | 29 | #include <linux/sched.h> |
c7addb90 | 30 | #include <linux/moduleparam.h> |
e9bda3b3 | 31 | #include <linux/mod_devicetable.h> |
af658dca | 32 | #include <linux/trace_events.h> |
5a0e3ad6 | 33 | #include <linux/slab.h> |
cafd6659 | 34 | #include <linux/tboot.h> |
f4124500 | 35 | #include <linux/hrtimer.h> |
5fdbf976 | 36 | #include "kvm_cache_regs.h" |
35920a35 | 37 | #include "x86.h" |
e495606d | 38 | |
28b835d6 | 39 | #include <asm/cpu.h> |
6aa8b732 | 40 | #include <asm/io.h> |
3b3be0d1 | 41 | #include <asm/desc.h> |
13673a90 | 42 | #include <asm/vmx.h> |
6210e37b | 43 | #include <asm/virtext.h> |
a0861c02 | 44 | #include <asm/mce.h> |
952f07ec | 45 | #include <asm/fpu/internal.h> |
d7cd9796 | 46 | #include <asm/perf_event.h> |
81908bf4 | 47 | #include <asm/debugreg.h> |
8f536b76 | 48 | #include <asm/kexec.h> |
dab2087d | 49 | #include <asm/apic.h> |
efc64404 | 50 | #include <asm/irq_remapping.h> |
6aa8b732 | 51 | |
229456fc | 52 | #include "trace.h" |
25462f7f | 53 | #include "pmu.h" |
229456fc | 54 | |
4ecac3fd | 55 | #define __ex(x) __kvm_handle_fault_on_reboot(x) |
5e520e62 AK |
56 | #define __ex_clear(x, reg) \ |
57 | ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg) | |
4ecac3fd | 58 | |
6aa8b732 AK |
59 | MODULE_AUTHOR("Qumranet"); |
60 | MODULE_LICENSE("GPL"); | |
61 | ||
e9bda3b3 JT |
62 | static const struct x86_cpu_id vmx_cpu_id[] = { |
63 | X86_FEATURE_MATCH(X86_FEATURE_VMX), | |
64 | {} | |
65 | }; | |
66 | MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id); | |
67 | ||
476bc001 | 68 | static bool __read_mostly enable_vpid = 1; |
736caefe | 69 | module_param_named(vpid, enable_vpid, bool, 0444); |
2384d2b3 | 70 | |
476bc001 | 71 | static bool __read_mostly flexpriority_enabled = 1; |
736caefe | 72 | module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO); |
4c9fc8ef | 73 | |
476bc001 | 74 | static bool __read_mostly enable_ept = 1; |
736caefe | 75 | module_param_named(ept, enable_ept, bool, S_IRUGO); |
d56f546d | 76 | |
476bc001 | 77 | static bool __read_mostly enable_unrestricted_guest = 1; |
3a624e29 NK |
78 | module_param_named(unrestricted_guest, |
79 | enable_unrestricted_guest, bool, S_IRUGO); | |
80 | ||
83c3a331 XH |
81 | static bool __read_mostly enable_ept_ad_bits = 1; |
82 | module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO); | |
83 | ||
a27685c3 | 84 | static bool __read_mostly emulate_invalid_guest_state = true; |
c1f8bc04 | 85 | module_param(emulate_invalid_guest_state, bool, S_IRUGO); |
04fa4d32 | 86 | |
476bc001 | 87 | static bool __read_mostly vmm_exclusive = 1; |
b923e62e DX |
88 | module_param(vmm_exclusive, bool, S_IRUGO); |
89 | ||
476bc001 | 90 | static bool __read_mostly fasteoi = 1; |
58fbbf26 KT |
91 | module_param(fasteoi, bool, S_IRUGO); |
92 | ||
5a71785d | 93 | static bool __read_mostly enable_apicv = 1; |
01e439be | 94 | module_param(enable_apicv, bool, S_IRUGO); |
83d4c286 | 95 | |
abc4fc58 AG |
96 | static bool __read_mostly enable_shadow_vmcs = 1; |
97 | module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO); | |
801d3424 NHE |
98 | /* |
99 | * If nested=1, nested virtualization is supported, i.e., guests may use | |
100 | * VMX and be a hypervisor for its own guests. If nested=0, guests may not | |
101 | * use VMX instructions. | |
102 | */ | |
476bc001 | 103 | static bool __read_mostly nested = 0; |
801d3424 NHE |
104 | module_param(nested, bool, S_IRUGO); |
105 | ||
20300099 WL |
106 | static u64 __read_mostly host_xss; |
107 | ||
843e4330 KH |
108 | static bool __read_mostly enable_pml = 1; |
109 | module_param_named(pml, enable_pml, bool, S_IRUGO); | |
110 | ||
64903d61 HZ |
111 | #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL |
112 | ||
5037878e GN |
113 | #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD) |
114 | #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE) | |
cdc0e244 AK |
115 | #define KVM_VM_CR0_ALWAYS_ON \ |
116 | (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE) | |
4c38609a AK |
117 | #define KVM_CR4_GUEST_OWNED_BITS \ |
118 | (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \ | |
52ce3c21 | 119 | | X86_CR4_OSXMMEXCPT | X86_CR4_TSD) |
4c38609a | 120 | |
cdc0e244 AK |
121 | #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE) |
122 | #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE) | |
123 | ||
78ac8b47 AK |
124 | #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM)) |
125 | ||
f4124500 JK |
126 | #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5 |
127 | ||
4b8d54f9 ZE |
128 | /* |
129 | * These 2 parameters are used to config the controls for Pause-Loop Exiting: | |
130 | * ple_gap: upper bound on the amount of time between two successive | |
131 | * executions of PAUSE in a loop. Also indicate if ple enabled. | |
00c25bce | 132 | * According to test, this time is usually smaller than 128 cycles. |
4b8d54f9 ZE |
133 | * ple_window: upper bound on the amount of time a guest is allowed to execute |
134 | * in a PAUSE loop. Tests indicate that most spinlocks are held for | |
135 | * less than 2^12 cycles | |
136 | * Time is measured based on a counter that runs at the same rate as the TSC, | |
137 | * refer SDM volume 3b section 21.6.13 & 22.1.3. | |
138 | */ | |
b4a2d31d RK |
139 | #define KVM_VMX_DEFAULT_PLE_GAP 128 |
140 | #define KVM_VMX_DEFAULT_PLE_WINDOW 4096 | |
141 | #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2 | |
142 | #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0 | |
143 | #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \ | |
144 | INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW | |
145 | ||
4b8d54f9 ZE |
146 | static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP; |
147 | module_param(ple_gap, int, S_IRUGO); | |
148 | ||
149 | static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW; | |
150 | module_param(ple_window, int, S_IRUGO); | |
151 | ||
b4a2d31d RK |
152 | /* Default doubles per-vcpu window every exit. */ |
153 | static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW; | |
154 | module_param(ple_window_grow, int, S_IRUGO); | |
155 | ||
156 | /* Default resets per-vcpu window every exit to ple_window. */ | |
157 | static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK; | |
158 | module_param(ple_window_shrink, int, S_IRUGO); | |
159 | ||
160 | /* Default is to compute the maximum so we can never overflow. */ | |
161 | static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX; | |
162 | static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX; | |
163 | module_param(ple_window_max, int, S_IRUGO); | |
164 | ||
83287ea4 AK |
165 | extern const ulong vmx_return; |
166 | ||
8bf00a52 | 167 | #define NR_AUTOLOAD_MSRS 8 |
ff2f6fe9 | 168 | #define VMCS02_POOL_SIZE 1 |
61d2ef2c | 169 | |
a2fa3e9f GH |
170 | struct vmcs { |
171 | u32 revision_id; | |
172 | u32 abort; | |
173 | char data[0]; | |
174 | }; | |
175 | ||
d462b819 NHE |
176 | /* |
177 | * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also | |
178 | * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs | |
179 | * loaded on this CPU (so we can clear them if the CPU goes down). | |
180 | */ | |
181 | struct loaded_vmcs { | |
182 | struct vmcs *vmcs; | |
183 | int cpu; | |
184 | int launched; | |
185 | struct list_head loaded_vmcss_on_cpu_link; | |
186 | }; | |
187 | ||
26bb0981 AK |
188 | struct shared_msr_entry { |
189 | unsigned index; | |
190 | u64 data; | |
d5696725 | 191 | u64 mask; |
26bb0981 AK |
192 | }; |
193 | ||
a9d30f33 NHE |
194 | /* |
195 | * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a | |
196 | * single nested guest (L2), hence the name vmcs12. Any VMX implementation has | |
197 | * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is | |
198 | * stored in guest memory specified by VMPTRLD, but is opaque to the guest, | |
199 | * which must access it using VMREAD/VMWRITE/VMCLEAR instructions. | |
200 | * More than one of these structures may exist, if L1 runs multiple L2 guests. | |
201 | * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the | |
202 | * underlying hardware which will be used to run L2. | |
203 | * This structure is packed to ensure that its layout is identical across | |
204 | * machines (necessary for live migration). | |
205 | * If there are changes in this struct, VMCS12_REVISION must be changed. | |
206 | */ | |
22bd0358 | 207 | typedef u64 natural_width; |
a9d30f33 NHE |
208 | struct __packed vmcs12 { |
209 | /* According to the Intel spec, a VMCS region must start with the | |
210 | * following two fields. Then follow implementation-specific data. | |
211 | */ | |
212 | u32 revision_id; | |
213 | u32 abort; | |
22bd0358 | 214 | |
27d6c865 NHE |
215 | u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */ |
216 | u32 padding[7]; /* room for future expansion */ | |
217 | ||
22bd0358 NHE |
218 | u64 io_bitmap_a; |
219 | u64 io_bitmap_b; | |
220 | u64 msr_bitmap; | |
221 | u64 vm_exit_msr_store_addr; | |
222 | u64 vm_exit_msr_load_addr; | |
223 | u64 vm_entry_msr_load_addr; | |
224 | u64 tsc_offset; | |
225 | u64 virtual_apic_page_addr; | |
226 | u64 apic_access_addr; | |
705699a1 | 227 | u64 posted_intr_desc_addr; |
22bd0358 | 228 | u64 ept_pointer; |
608406e2 WV |
229 | u64 eoi_exit_bitmap0; |
230 | u64 eoi_exit_bitmap1; | |
231 | u64 eoi_exit_bitmap2; | |
232 | u64 eoi_exit_bitmap3; | |
81dc01f7 | 233 | u64 xss_exit_bitmap; |
22bd0358 NHE |
234 | u64 guest_physical_address; |
235 | u64 vmcs_link_pointer; | |
236 | u64 guest_ia32_debugctl; | |
237 | u64 guest_ia32_pat; | |
238 | u64 guest_ia32_efer; | |
239 | u64 guest_ia32_perf_global_ctrl; | |
240 | u64 guest_pdptr0; | |
241 | u64 guest_pdptr1; | |
242 | u64 guest_pdptr2; | |
243 | u64 guest_pdptr3; | |
36be0b9d | 244 | u64 guest_bndcfgs; |
22bd0358 NHE |
245 | u64 host_ia32_pat; |
246 | u64 host_ia32_efer; | |
247 | u64 host_ia32_perf_global_ctrl; | |
248 | u64 padding64[8]; /* room for future expansion */ | |
249 | /* | |
250 | * To allow migration of L1 (complete with its L2 guests) between | |
251 | * machines of different natural widths (32 or 64 bit), we cannot have | |
252 | * unsigned long fields with no explict size. We use u64 (aliased | |
253 | * natural_width) instead. Luckily, x86 is little-endian. | |
254 | */ | |
255 | natural_width cr0_guest_host_mask; | |
256 | natural_width cr4_guest_host_mask; | |
257 | natural_width cr0_read_shadow; | |
258 | natural_width cr4_read_shadow; | |
259 | natural_width cr3_target_value0; | |
260 | natural_width cr3_target_value1; | |
261 | natural_width cr3_target_value2; | |
262 | natural_width cr3_target_value3; | |
263 | natural_width exit_qualification; | |
264 | natural_width guest_linear_address; | |
265 | natural_width guest_cr0; | |
266 | natural_width guest_cr3; | |
267 | natural_width guest_cr4; | |
268 | natural_width guest_es_base; | |
269 | natural_width guest_cs_base; | |
270 | natural_width guest_ss_base; | |
271 | natural_width guest_ds_base; | |
272 | natural_width guest_fs_base; | |
273 | natural_width guest_gs_base; | |
274 | natural_width guest_ldtr_base; | |
275 | natural_width guest_tr_base; | |
276 | natural_width guest_gdtr_base; | |
277 | natural_width guest_idtr_base; | |
278 | natural_width guest_dr7; | |
279 | natural_width guest_rsp; | |
280 | natural_width guest_rip; | |
281 | natural_width guest_rflags; | |
282 | natural_width guest_pending_dbg_exceptions; | |
283 | natural_width guest_sysenter_esp; | |
284 | natural_width guest_sysenter_eip; | |
285 | natural_width host_cr0; | |
286 | natural_width host_cr3; | |
287 | natural_width host_cr4; | |
288 | natural_width host_fs_base; | |
289 | natural_width host_gs_base; | |
290 | natural_width host_tr_base; | |
291 | natural_width host_gdtr_base; | |
292 | natural_width host_idtr_base; | |
293 | natural_width host_ia32_sysenter_esp; | |
294 | natural_width host_ia32_sysenter_eip; | |
295 | natural_width host_rsp; | |
296 | natural_width host_rip; | |
297 | natural_width paddingl[8]; /* room for future expansion */ | |
298 | u32 pin_based_vm_exec_control; | |
299 | u32 cpu_based_vm_exec_control; | |
300 | u32 exception_bitmap; | |
301 | u32 page_fault_error_code_mask; | |
302 | u32 page_fault_error_code_match; | |
303 | u32 cr3_target_count; | |
304 | u32 vm_exit_controls; | |
305 | u32 vm_exit_msr_store_count; | |
306 | u32 vm_exit_msr_load_count; | |
307 | u32 vm_entry_controls; | |
308 | u32 vm_entry_msr_load_count; | |
309 | u32 vm_entry_intr_info_field; | |
310 | u32 vm_entry_exception_error_code; | |
311 | u32 vm_entry_instruction_len; | |
312 | u32 tpr_threshold; | |
313 | u32 secondary_vm_exec_control; | |
314 | u32 vm_instruction_error; | |
315 | u32 vm_exit_reason; | |
316 | u32 vm_exit_intr_info; | |
317 | u32 vm_exit_intr_error_code; | |
318 | u32 idt_vectoring_info_field; | |
319 | u32 idt_vectoring_error_code; | |
320 | u32 vm_exit_instruction_len; | |
321 | u32 vmx_instruction_info; | |
322 | u32 guest_es_limit; | |
323 | u32 guest_cs_limit; | |
324 | u32 guest_ss_limit; | |
325 | u32 guest_ds_limit; | |
326 | u32 guest_fs_limit; | |
327 | u32 guest_gs_limit; | |
328 | u32 guest_ldtr_limit; | |
329 | u32 guest_tr_limit; | |
330 | u32 guest_gdtr_limit; | |
331 | u32 guest_idtr_limit; | |
332 | u32 guest_es_ar_bytes; | |
333 | u32 guest_cs_ar_bytes; | |
334 | u32 guest_ss_ar_bytes; | |
335 | u32 guest_ds_ar_bytes; | |
336 | u32 guest_fs_ar_bytes; | |
337 | u32 guest_gs_ar_bytes; | |
338 | u32 guest_ldtr_ar_bytes; | |
339 | u32 guest_tr_ar_bytes; | |
340 | u32 guest_interruptibility_info; | |
341 | u32 guest_activity_state; | |
342 | u32 guest_sysenter_cs; | |
343 | u32 host_ia32_sysenter_cs; | |
0238ea91 JK |
344 | u32 vmx_preemption_timer_value; |
345 | u32 padding32[7]; /* room for future expansion */ | |
22bd0358 | 346 | u16 virtual_processor_id; |
705699a1 | 347 | u16 posted_intr_nv; |
22bd0358 NHE |
348 | u16 guest_es_selector; |
349 | u16 guest_cs_selector; | |
350 | u16 guest_ss_selector; | |
351 | u16 guest_ds_selector; | |
352 | u16 guest_fs_selector; | |
353 | u16 guest_gs_selector; | |
354 | u16 guest_ldtr_selector; | |
355 | u16 guest_tr_selector; | |
608406e2 | 356 | u16 guest_intr_status; |
22bd0358 NHE |
357 | u16 host_es_selector; |
358 | u16 host_cs_selector; | |
359 | u16 host_ss_selector; | |
360 | u16 host_ds_selector; | |
361 | u16 host_fs_selector; | |
362 | u16 host_gs_selector; | |
363 | u16 host_tr_selector; | |
a9d30f33 NHE |
364 | }; |
365 | ||
366 | /* | |
367 | * VMCS12_REVISION is an arbitrary id that should be changed if the content or | |
368 | * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and | |
369 | * VMPTRLD verifies that the VMCS region that L1 is loading contains this id. | |
370 | */ | |
371 | #define VMCS12_REVISION 0x11e57ed0 | |
372 | ||
373 | /* | |
374 | * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region | |
375 | * and any VMCS region. Although only sizeof(struct vmcs12) are used by the | |
376 | * current implementation, 4K are reserved to avoid future complications. | |
377 | */ | |
378 | #define VMCS12_SIZE 0x1000 | |
379 | ||
ff2f6fe9 NHE |
380 | /* Used to remember the last vmcs02 used for some recently used vmcs12s */ |
381 | struct vmcs02_list { | |
382 | struct list_head list; | |
383 | gpa_t vmptr; | |
384 | struct loaded_vmcs vmcs02; | |
385 | }; | |
386 | ||
ec378aee NHE |
387 | /* |
388 | * The nested_vmx structure is part of vcpu_vmx, and holds information we need | |
389 | * for correct emulation of VMX (i.e., nested VMX) on this vcpu. | |
390 | */ | |
391 | struct nested_vmx { | |
392 | /* Has the level1 guest done vmxon? */ | |
393 | bool vmxon; | |
3573e22c | 394 | gpa_t vmxon_ptr; |
a9d30f33 NHE |
395 | |
396 | /* The guest-physical address of the current VMCS L1 keeps for L2 */ | |
397 | gpa_t current_vmptr; | |
398 | /* The host-usable pointer to the above */ | |
399 | struct page *current_vmcs12_page; | |
400 | struct vmcs12 *current_vmcs12; | |
8de48833 | 401 | struct vmcs *current_shadow_vmcs; |
012f83cb AG |
402 | /* |
403 | * Indicates if the shadow vmcs must be updated with the | |
404 | * data hold by vmcs12 | |
405 | */ | |
406 | bool sync_shadow_vmcs; | |
ff2f6fe9 NHE |
407 | |
408 | /* vmcs02_list cache of VMCSs recently used to run L2 guests */ | |
409 | struct list_head vmcs02_pool; | |
410 | int vmcs02_num; | |
fe3ef05c | 411 | u64 vmcs01_tsc_offset; |
644d711a NHE |
412 | /* L2 must run next, and mustn't decide to exit to L1. */ |
413 | bool nested_run_pending; | |
fe3ef05c NHE |
414 | /* |
415 | * Guest pages referred to in vmcs02 with host-physical pointers, so | |
416 | * we must keep them pinned while L2 runs. | |
417 | */ | |
418 | struct page *apic_access_page; | |
a7c0b07d | 419 | struct page *virtual_apic_page; |
705699a1 WV |
420 | struct page *pi_desc_page; |
421 | struct pi_desc *pi_desc; | |
422 | bool pi_pending; | |
423 | u16 posted_intr_nv; | |
b3897a49 | 424 | u64 msr_ia32_feature_control; |
f4124500 JK |
425 | |
426 | struct hrtimer preemption_timer; | |
427 | bool preemption_timer_expired; | |
2996fca0 JK |
428 | |
429 | /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */ | |
430 | u64 vmcs01_debugctl; | |
b9c237bb | 431 | |
5c614b35 WL |
432 | u16 vpid02; |
433 | u16 last_vpid; | |
434 | ||
b9c237bb WV |
435 | u32 nested_vmx_procbased_ctls_low; |
436 | u32 nested_vmx_procbased_ctls_high; | |
437 | u32 nested_vmx_true_procbased_ctls_low; | |
438 | u32 nested_vmx_secondary_ctls_low; | |
439 | u32 nested_vmx_secondary_ctls_high; | |
440 | u32 nested_vmx_pinbased_ctls_low; | |
441 | u32 nested_vmx_pinbased_ctls_high; | |
442 | u32 nested_vmx_exit_ctls_low; | |
443 | u32 nested_vmx_exit_ctls_high; | |
444 | u32 nested_vmx_true_exit_ctls_low; | |
445 | u32 nested_vmx_entry_ctls_low; | |
446 | u32 nested_vmx_entry_ctls_high; | |
447 | u32 nested_vmx_true_entry_ctls_low; | |
448 | u32 nested_vmx_misc_low; | |
449 | u32 nested_vmx_misc_high; | |
450 | u32 nested_vmx_ept_caps; | |
99b83ac8 | 451 | u32 nested_vmx_vpid_caps; |
ec378aee NHE |
452 | }; |
453 | ||
01e439be | 454 | #define POSTED_INTR_ON 0 |
ebbfc765 FW |
455 | #define POSTED_INTR_SN 1 |
456 | ||
01e439be YZ |
457 | /* Posted-Interrupt Descriptor */ |
458 | struct pi_desc { | |
459 | u32 pir[8]; /* Posted interrupt requested */ | |
6ef1522f FW |
460 | union { |
461 | struct { | |
462 | /* bit 256 - Outstanding Notification */ | |
463 | u16 on : 1, | |
464 | /* bit 257 - Suppress Notification */ | |
465 | sn : 1, | |
466 | /* bit 271:258 - Reserved */ | |
467 | rsvd_1 : 14; | |
468 | /* bit 279:272 - Notification Vector */ | |
469 | u8 nv; | |
470 | /* bit 287:280 - Reserved */ | |
471 | u8 rsvd_2; | |
472 | /* bit 319:288 - Notification Destination */ | |
473 | u32 ndst; | |
474 | }; | |
475 | u64 control; | |
476 | }; | |
477 | u32 rsvd[6]; | |
01e439be YZ |
478 | } __aligned(64); |
479 | ||
a20ed54d YZ |
480 | static bool pi_test_and_set_on(struct pi_desc *pi_desc) |
481 | { | |
482 | return test_and_set_bit(POSTED_INTR_ON, | |
483 | (unsigned long *)&pi_desc->control); | |
484 | } | |
485 | ||
486 | static bool pi_test_and_clear_on(struct pi_desc *pi_desc) | |
487 | { | |
488 | return test_and_clear_bit(POSTED_INTR_ON, | |
489 | (unsigned long *)&pi_desc->control); | |
490 | } | |
491 | ||
492 | static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc) | |
493 | { | |
494 | return test_and_set_bit(vector, (unsigned long *)pi_desc->pir); | |
495 | } | |
496 | ||
ebbfc765 FW |
497 | static inline void pi_clear_sn(struct pi_desc *pi_desc) |
498 | { | |
499 | return clear_bit(POSTED_INTR_SN, | |
500 | (unsigned long *)&pi_desc->control); | |
501 | } | |
502 | ||
503 | static inline void pi_set_sn(struct pi_desc *pi_desc) | |
504 | { | |
505 | return set_bit(POSTED_INTR_SN, | |
506 | (unsigned long *)&pi_desc->control); | |
507 | } | |
508 | ||
509 | static inline int pi_test_on(struct pi_desc *pi_desc) | |
510 | { | |
511 | return test_bit(POSTED_INTR_ON, | |
512 | (unsigned long *)&pi_desc->control); | |
513 | } | |
514 | ||
515 | static inline int pi_test_sn(struct pi_desc *pi_desc) | |
516 | { | |
517 | return test_bit(POSTED_INTR_SN, | |
518 | (unsigned long *)&pi_desc->control); | |
519 | } | |
520 | ||
a2fa3e9f | 521 | struct vcpu_vmx { |
fb3f0f51 | 522 | struct kvm_vcpu vcpu; |
313dbd49 | 523 | unsigned long host_rsp; |
29bd8a78 | 524 | u8 fail; |
9d58b931 | 525 | bool nmi_known_unmasked; |
51aa01d1 | 526 | u32 exit_intr_info; |
1155f76a | 527 | u32 idt_vectoring_info; |
6de12732 | 528 | ulong rflags; |
26bb0981 | 529 | struct shared_msr_entry *guest_msrs; |
a2fa3e9f GH |
530 | int nmsrs; |
531 | int save_nmsrs; | |
a547c6db | 532 | unsigned long host_idt_base; |
a2fa3e9f | 533 | #ifdef CONFIG_X86_64 |
44ea2b17 AK |
534 | u64 msr_host_kernel_gs_base; |
535 | u64 msr_guest_kernel_gs_base; | |
a2fa3e9f | 536 | #endif |
2961e876 GN |
537 | u32 vm_entry_controls_shadow; |
538 | u32 vm_exit_controls_shadow; | |
d462b819 NHE |
539 | /* |
540 | * loaded_vmcs points to the VMCS currently used in this vcpu. For a | |
541 | * non-nested (L1) guest, it always points to vmcs01. For a nested | |
542 | * guest (L2), it points to a different VMCS. | |
543 | */ | |
544 | struct loaded_vmcs vmcs01; | |
545 | struct loaded_vmcs *loaded_vmcs; | |
546 | bool __launched; /* temporary, used in vmx_vcpu_run */ | |
61d2ef2c AK |
547 | struct msr_autoload { |
548 | unsigned nr; | |
549 | struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS]; | |
550 | struct vmx_msr_entry host[NR_AUTOLOAD_MSRS]; | |
551 | } msr_autoload; | |
a2fa3e9f GH |
552 | struct { |
553 | int loaded; | |
554 | u16 fs_sel, gs_sel, ldt_sel; | |
b2da15ac AK |
555 | #ifdef CONFIG_X86_64 |
556 | u16 ds_sel, es_sel; | |
557 | #endif | |
152d3f2f LV |
558 | int gs_ldt_reload_needed; |
559 | int fs_reload_needed; | |
da8999d3 | 560 | u64 msr_host_bndcfgs; |
d974baa3 | 561 | unsigned long vmcs_host_cr4; /* May not match real cr4 */ |
d77c26fc | 562 | } host_state; |
9c8cba37 | 563 | struct { |
7ffd92c5 | 564 | int vm86_active; |
78ac8b47 | 565 | ulong save_rflags; |
f5f7b2fe AK |
566 | struct kvm_segment segs[8]; |
567 | } rmode; | |
568 | struct { | |
569 | u32 bitmask; /* 4 bits per segment (1 bit per field) */ | |
7ffd92c5 AK |
570 | struct kvm_save_segment { |
571 | u16 selector; | |
572 | unsigned long base; | |
573 | u32 limit; | |
574 | u32 ar; | |
f5f7b2fe | 575 | } seg[8]; |
2fb92db1 | 576 | } segment_cache; |
2384d2b3 | 577 | int vpid; |
04fa4d32 | 578 | bool emulation_required; |
3b86cd99 JK |
579 | |
580 | /* Support for vnmi-less CPUs */ | |
581 | int soft_vnmi_blocked; | |
582 | ktime_t entry_time; | |
583 | s64 vnmi_blocked_time; | |
a0861c02 | 584 | u32 exit_reason; |
4e47c7a6 | 585 | |
01e439be YZ |
586 | /* Posted interrupt descriptor */ |
587 | struct pi_desc pi_desc; | |
588 | ||
ec378aee NHE |
589 | /* Support for a guest hypervisor (nested VMX) */ |
590 | struct nested_vmx nested; | |
a7653ecd RK |
591 | |
592 | /* Dynamic PLE window. */ | |
593 | int ple_window; | |
594 | bool ple_window_dirty; | |
843e4330 KH |
595 | |
596 | /* Support for PML */ | |
597 | #define PML_ENTITY_NUM 512 | |
598 | struct page *pml_pg; | |
2680d6da OH |
599 | |
600 | u64 current_tsc_ratio; | |
a2fa3e9f GH |
601 | }; |
602 | ||
2fb92db1 AK |
603 | enum segment_cache_field { |
604 | SEG_FIELD_SEL = 0, | |
605 | SEG_FIELD_BASE = 1, | |
606 | SEG_FIELD_LIMIT = 2, | |
607 | SEG_FIELD_AR = 3, | |
608 | ||
609 | SEG_FIELD_NR = 4 | |
610 | }; | |
611 | ||
a2fa3e9f GH |
612 | static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu) |
613 | { | |
fb3f0f51 | 614 | return container_of(vcpu, struct vcpu_vmx, vcpu); |
a2fa3e9f GH |
615 | } |
616 | ||
efc64404 FW |
617 | static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu) |
618 | { | |
619 | return &(to_vmx(vcpu)->pi_desc); | |
620 | } | |
621 | ||
22bd0358 NHE |
622 | #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x) |
623 | #define FIELD(number, name) [number] = VMCS12_OFFSET(name) | |
624 | #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \ | |
625 | [number##_HIGH] = VMCS12_OFFSET(name)+4 | |
626 | ||
4607c2d7 | 627 | |
fe2b201b | 628 | static unsigned long shadow_read_only_fields[] = { |
4607c2d7 AG |
629 | /* |
630 | * We do NOT shadow fields that are modified when L0 | |
631 | * traps and emulates any vmx instruction (e.g. VMPTRLD, | |
632 | * VMXON...) executed by L1. | |
633 | * For example, VM_INSTRUCTION_ERROR is read | |
634 | * by L1 if a vmx instruction fails (part of the error path). | |
635 | * Note the code assumes this logic. If for some reason | |
636 | * we start shadowing these fields then we need to | |
637 | * force a shadow sync when L0 emulates vmx instructions | |
638 | * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified | |
639 | * by nested_vmx_failValid) | |
640 | */ | |
641 | VM_EXIT_REASON, | |
642 | VM_EXIT_INTR_INFO, | |
643 | VM_EXIT_INSTRUCTION_LEN, | |
644 | IDT_VECTORING_INFO_FIELD, | |
645 | IDT_VECTORING_ERROR_CODE, | |
646 | VM_EXIT_INTR_ERROR_CODE, | |
647 | EXIT_QUALIFICATION, | |
648 | GUEST_LINEAR_ADDRESS, | |
649 | GUEST_PHYSICAL_ADDRESS | |
650 | }; | |
fe2b201b | 651 | static int max_shadow_read_only_fields = |
4607c2d7 AG |
652 | ARRAY_SIZE(shadow_read_only_fields); |
653 | ||
fe2b201b | 654 | static unsigned long shadow_read_write_fields[] = { |
a7c0b07d | 655 | TPR_THRESHOLD, |
4607c2d7 AG |
656 | GUEST_RIP, |
657 | GUEST_RSP, | |
658 | GUEST_CR0, | |
659 | GUEST_CR3, | |
660 | GUEST_CR4, | |
661 | GUEST_INTERRUPTIBILITY_INFO, | |
662 | GUEST_RFLAGS, | |
663 | GUEST_CS_SELECTOR, | |
664 | GUEST_CS_AR_BYTES, | |
665 | GUEST_CS_LIMIT, | |
666 | GUEST_CS_BASE, | |
667 | GUEST_ES_BASE, | |
36be0b9d | 668 | GUEST_BNDCFGS, |
4607c2d7 AG |
669 | CR0_GUEST_HOST_MASK, |
670 | CR0_READ_SHADOW, | |
671 | CR4_READ_SHADOW, | |
672 | TSC_OFFSET, | |
673 | EXCEPTION_BITMAP, | |
674 | CPU_BASED_VM_EXEC_CONTROL, | |
675 | VM_ENTRY_EXCEPTION_ERROR_CODE, | |
676 | VM_ENTRY_INTR_INFO_FIELD, | |
677 | VM_ENTRY_INSTRUCTION_LEN, | |
678 | VM_ENTRY_EXCEPTION_ERROR_CODE, | |
679 | HOST_FS_BASE, | |
680 | HOST_GS_BASE, | |
681 | HOST_FS_SELECTOR, | |
682 | HOST_GS_SELECTOR | |
683 | }; | |
fe2b201b | 684 | static int max_shadow_read_write_fields = |
4607c2d7 AG |
685 | ARRAY_SIZE(shadow_read_write_fields); |
686 | ||
772e0318 | 687 | static const unsigned short vmcs_field_to_offset_table[] = { |
22bd0358 | 688 | FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id), |
705699a1 | 689 | FIELD(POSTED_INTR_NV, posted_intr_nv), |
22bd0358 NHE |
690 | FIELD(GUEST_ES_SELECTOR, guest_es_selector), |
691 | FIELD(GUEST_CS_SELECTOR, guest_cs_selector), | |
692 | FIELD(GUEST_SS_SELECTOR, guest_ss_selector), | |
693 | FIELD(GUEST_DS_SELECTOR, guest_ds_selector), | |
694 | FIELD(GUEST_FS_SELECTOR, guest_fs_selector), | |
695 | FIELD(GUEST_GS_SELECTOR, guest_gs_selector), | |
696 | FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector), | |
697 | FIELD(GUEST_TR_SELECTOR, guest_tr_selector), | |
608406e2 | 698 | FIELD(GUEST_INTR_STATUS, guest_intr_status), |
22bd0358 NHE |
699 | FIELD(HOST_ES_SELECTOR, host_es_selector), |
700 | FIELD(HOST_CS_SELECTOR, host_cs_selector), | |
701 | FIELD(HOST_SS_SELECTOR, host_ss_selector), | |
702 | FIELD(HOST_DS_SELECTOR, host_ds_selector), | |
703 | FIELD(HOST_FS_SELECTOR, host_fs_selector), | |
704 | FIELD(HOST_GS_SELECTOR, host_gs_selector), | |
705 | FIELD(HOST_TR_SELECTOR, host_tr_selector), | |
706 | FIELD64(IO_BITMAP_A, io_bitmap_a), | |
707 | FIELD64(IO_BITMAP_B, io_bitmap_b), | |
708 | FIELD64(MSR_BITMAP, msr_bitmap), | |
709 | FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr), | |
710 | FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr), | |
711 | FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr), | |
712 | FIELD64(TSC_OFFSET, tsc_offset), | |
713 | FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr), | |
714 | FIELD64(APIC_ACCESS_ADDR, apic_access_addr), | |
705699a1 | 715 | FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr), |
22bd0358 | 716 | FIELD64(EPT_POINTER, ept_pointer), |
608406e2 WV |
717 | FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0), |
718 | FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1), | |
719 | FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2), | |
720 | FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3), | |
81dc01f7 | 721 | FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap), |
22bd0358 NHE |
722 | FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address), |
723 | FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer), | |
724 | FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl), | |
725 | FIELD64(GUEST_IA32_PAT, guest_ia32_pat), | |
726 | FIELD64(GUEST_IA32_EFER, guest_ia32_efer), | |
727 | FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl), | |
728 | FIELD64(GUEST_PDPTR0, guest_pdptr0), | |
729 | FIELD64(GUEST_PDPTR1, guest_pdptr1), | |
730 | FIELD64(GUEST_PDPTR2, guest_pdptr2), | |
731 | FIELD64(GUEST_PDPTR3, guest_pdptr3), | |
36be0b9d | 732 | FIELD64(GUEST_BNDCFGS, guest_bndcfgs), |
22bd0358 NHE |
733 | FIELD64(HOST_IA32_PAT, host_ia32_pat), |
734 | FIELD64(HOST_IA32_EFER, host_ia32_efer), | |
735 | FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl), | |
736 | FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control), | |
737 | FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control), | |
738 | FIELD(EXCEPTION_BITMAP, exception_bitmap), | |
739 | FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask), | |
740 | FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match), | |
741 | FIELD(CR3_TARGET_COUNT, cr3_target_count), | |
742 | FIELD(VM_EXIT_CONTROLS, vm_exit_controls), | |
743 | FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count), | |
744 | FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count), | |
745 | FIELD(VM_ENTRY_CONTROLS, vm_entry_controls), | |
746 | FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count), | |
747 | FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field), | |
748 | FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code), | |
749 | FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len), | |
750 | FIELD(TPR_THRESHOLD, tpr_threshold), | |
751 | FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control), | |
752 | FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error), | |
753 | FIELD(VM_EXIT_REASON, vm_exit_reason), | |
754 | FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info), | |
755 | FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code), | |
756 | FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field), | |
757 | FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code), | |
758 | FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len), | |
759 | FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info), | |
760 | FIELD(GUEST_ES_LIMIT, guest_es_limit), | |
761 | FIELD(GUEST_CS_LIMIT, guest_cs_limit), | |
762 | FIELD(GUEST_SS_LIMIT, guest_ss_limit), | |
763 | FIELD(GUEST_DS_LIMIT, guest_ds_limit), | |
764 | FIELD(GUEST_FS_LIMIT, guest_fs_limit), | |
765 | FIELD(GUEST_GS_LIMIT, guest_gs_limit), | |
766 | FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit), | |
767 | FIELD(GUEST_TR_LIMIT, guest_tr_limit), | |
768 | FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit), | |
769 | FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit), | |
770 | FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes), | |
771 | FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes), | |
772 | FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes), | |
773 | FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes), | |
774 | FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes), | |
775 | FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes), | |
776 | FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes), | |
777 | FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes), | |
778 | FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info), | |
779 | FIELD(GUEST_ACTIVITY_STATE, guest_activity_state), | |
780 | FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs), | |
781 | FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs), | |
0238ea91 | 782 | FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value), |
22bd0358 NHE |
783 | FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask), |
784 | FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask), | |
785 | FIELD(CR0_READ_SHADOW, cr0_read_shadow), | |
786 | FIELD(CR4_READ_SHADOW, cr4_read_shadow), | |
787 | FIELD(CR3_TARGET_VALUE0, cr3_target_value0), | |
788 | FIELD(CR3_TARGET_VALUE1, cr3_target_value1), | |
789 | FIELD(CR3_TARGET_VALUE2, cr3_target_value2), | |
790 | FIELD(CR3_TARGET_VALUE3, cr3_target_value3), | |
791 | FIELD(EXIT_QUALIFICATION, exit_qualification), | |
792 | FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address), | |
793 | FIELD(GUEST_CR0, guest_cr0), | |
794 | FIELD(GUEST_CR3, guest_cr3), | |
795 | FIELD(GUEST_CR4, guest_cr4), | |
796 | FIELD(GUEST_ES_BASE, guest_es_base), | |
797 | FIELD(GUEST_CS_BASE, guest_cs_base), | |
798 | FIELD(GUEST_SS_BASE, guest_ss_base), | |
799 | FIELD(GUEST_DS_BASE, guest_ds_base), | |
800 | FIELD(GUEST_FS_BASE, guest_fs_base), | |
801 | FIELD(GUEST_GS_BASE, guest_gs_base), | |
802 | FIELD(GUEST_LDTR_BASE, guest_ldtr_base), | |
803 | FIELD(GUEST_TR_BASE, guest_tr_base), | |
804 | FIELD(GUEST_GDTR_BASE, guest_gdtr_base), | |
805 | FIELD(GUEST_IDTR_BASE, guest_idtr_base), | |
806 | FIELD(GUEST_DR7, guest_dr7), | |
807 | FIELD(GUEST_RSP, guest_rsp), | |
808 | FIELD(GUEST_RIP, guest_rip), | |
809 | FIELD(GUEST_RFLAGS, guest_rflags), | |
810 | FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions), | |
811 | FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp), | |
812 | FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip), | |
813 | FIELD(HOST_CR0, host_cr0), | |
814 | FIELD(HOST_CR3, host_cr3), | |
815 | FIELD(HOST_CR4, host_cr4), | |
816 | FIELD(HOST_FS_BASE, host_fs_base), | |
817 | FIELD(HOST_GS_BASE, host_gs_base), | |
818 | FIELD(HOST_TR_BASE, host_tr_base), | |
819 | FIELD(HOST_GDTR_BASE, host_gdtr_base), | |
820 | FIELD(HOST_IDTR_BASE, host_idtr_base), | |
821 | FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp), | |
822 | FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip), | |
823 | FIELD(HOST_RSP, host_rsp), | |
824 | FIELD(HOST_RIP, host_rip), | |
825 | }; | |
22bd0358 NHE |
826 | |
827 | static inline short vmcs_field_to_offset(unsigned long field) | |
828 | { | |
a2ae9df7 PB |
829 | BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX); |
830 | ||
831 | if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) || | |
832 | vmcs_field_to_offset_table[field] == 0) | |
833 | return -ENOENT; | |
834 | ||
22bd0358 NHE |
835 | return vmcs_field_to_offset_table[field]; |
836 | } | |
837 | ||
a9d30f33 NHE |
838 | static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu) |
839 | { | |
840 | return to_vmx(vcpu)->nested.current_vmcs12; | |
841 | } | |
842 | ||
843 | static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr) | |
844 | { | |
54bf36aa | 845 | struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT); |
32cad84f | 846 | if (is_error_page(page)) |
a9d30f33 | 847 | return NULL; |
32cad84f | 848 | |
a9d30f33 NHE |
849 | return page; |
850 | } | |
851 | ||
852 | static void nested_release_page(struct page *page) | |
853 | { | |
854 | kvm_release_page_dirty(page); | |
855 | } | |
856 | ||
857 | static void nested_release_page_clean(struct page *page) | |
858 | { | |
859 | kvm_release_page_clean(page); | |
860 | } | |
861 | ||
bfd0a56b | 862 | static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu); |
4e1096d2 | 863 | static u64 construct_eptp(unsigned long root_hpa); |
4610c9cc DX |
864 | static void kvm_cpu_vmxon(u64 addr); |
865 | static void kvm_cpu_vmxoff(void); | |
f53cd63c | 866 | static bool vmx_xsaves_supported(void); |
776e58ea | 867 | static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr); |
b246dd5d OW |
868 | static void vmx_set_segment(struct kvm_vcpu *vcpu, |
869 | struct kvm_segment *var, int seg); | |
870 | static void vmx_get_segment(struct kvm_vcpu *vcpu, | |
871 | struct kvm_segment *var, int seg); | |
d99e4152 GN |
872 | static bool guest_state_valid(struct kvm_vcpu *vcpu); |
873 | static u32 vmx_segment_access_rights(struct kvm_segment *var); | |
c3114420 | 874 | static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx); |
16f5b903 | 875 | static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx); |
a255d479 | 876 | static int alloc_identity_pagetable(struct kvm *kvm); |
75880a01 | 877 | |
6aa8b732 AK |
878 | static DEFINE_PER_CPU(struct vmcs *, vmxarea); |
879 | static DEFINE_PER_CPU(struct vmcs *, current_vmcs); | |
d462b819 NHE |
880 | /* |
881 | * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed | |
882 | * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it. | |
883 | */ | |
884 | static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu); | |
3444d7da | 885 | static DEFINE_PER_CPU(struct desc_ptr, host_gdt); |
6aa8b732 | 886 | |
bf9f6ac8 FW |
887 | /* |
888 | * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we | |
889 | * can find which vCPU should be waken up. | |
890 | */ | |
891 | static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu); | |
892 | static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock); | |
893 | ||
3e7c73e9 AK |
894 | static unsigned long *vmx_io_bitmap_a; |
895 | static unsigned long *vmx_io_bitmap_b; | |
5897297b AK |
896 | static unsigned long *vmx_msr_bitmap_legacy; |
897 | static unsigned long *vmx_msr_bitmap_longmode; | |
8d14695f YZ |
898 | static unsigned long *vmx_msr_bitmap_legacy_x2apic; |
899 | static unsigned long *vmx_msr_bitmap_longmode_x2apic; | |
3af18d9c | 900 | static unsigned long *vmx_msr_bitmap_nested; |
4607c2d7 AG |
901 | static unsigned long *vmx_vmread_bitmap; |
902 | static unsigned long *vmx_vmwrite_bitmap; | |
fdef3ad1 | 903 | |
110312c8 | 904 | static bool cpu_has_load_ia32_efer; |
8bf00a52 | 905 | static bool cpu_has_load_perf_global_ctrl; |
110312c8 | 906 | |
2384d2b3 SY |
907 | static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS); |
908 | static DEFINE_SPINLOCK(vmx_vpid_lock); | |
909 | ||
1c3d14fe | 910 | static struct vmcs_config { |
6aa8b732 AK |
911 | int size; |
912 | int order; | |
913 | u32 revision_id; | |
1c3d14fe YS |
914 | u32 pin_based_exec_ctrl; |
915 | u32 cpu_based_exec_ctrl; | |
f78e0e2e | 916 | u32 cpu_based_2nd_exec_ctrl; |
1c3d14fe YS |
917 | u32 vmexit_ctrl; |
918 | u32 vmentry_ctrl; | |
919 | } vmcs_config; | |
6aa8b732 | 920 | |
efff9e53 | 921 | static struct vmx_capability { |
d56f546d SY |
922 | u32 ept; |
923 | u32 vpid; | |
924 | } vmx_capability; | |
925 | ||
6aa8b732 AK |
926 | #define VMX_SEGMENT_FIELD(seg) \ |
927 | [VCPU_SREG_##seg] = { \ | |
928 | .selector = GUEST_##seg##_SELECTOR, \ | |
929 | .base = GUEST_##seg##_BASE, \ | |
930 | .limit = GUEST_##seg##_LIMIT, \ | |
931 | .ar_bytes = GUEST_##seg##_AR_BYTES, \ | |
932 | } | |
933 | ||
772e0318 | 934 | static const struct kvm_vmx_segment_field { |
6aa8b732 AK |
935 | unsigned selector; |
936 | unsigned base; | |
937 | unsigned limit; | |
938 | unsigned ar_bytes; | |
939 | } kvm_vmx_segment_fields[] = { | |
940 | VMX_SEGMENT_FIELD(CS), | |
941 | VMX_SEGMENT_FIELD(DS), | |
942 | VMX_SEGMENT_FIELD(ES), | |
943 | VMX_SEGMENT_FIELD(FS), | |
944 | VMX_SEGMENT_FIELD(GS), | |
945 | VMX_SEGMENT_FIELD(SS), | |
946 | VMX_SEGMENT_FIELD(TR), | |
947 | VMX_SEGMENT_FIELD(LDTR), | |
948 | }; | |
949 | ||
26bb0981 AK |
950 | static u64 host_efer; |
951 | ||
6de4f3ad AK |
952 | static void ept_save_pdptrs(struct kvm_vcpu *vcpu); |
953 | ||
4d56c8a7 | 954 | /* |
8c06585d | 955 | * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it |
4d56c8a7 AK |
956 | * away by decrementing the array size. |
957 | */ | |
6aa8b732 | 958 | static const u32 vmx_msr_index[] = { |
05b3e0c2 | 959 | #ifdef CONFIG_X86_64 |
44ea2b17 | 960 | MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, |
6aa8b732 | 961 | #endif |
8c06585d | 962 | MSR_EFER, MSR_TSC_AUX, MSR_STAR, |
6aa8b732 | 963 | }; |
6aa8b732 | 964 | |
5bb16016 | 965 | static inline bool is_exception_n(u32 intr_info, u8 vector) |
6aa8b732 AK |
966 | { |
967 | return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK | | |
968 | INTR_INFO_VALID_MASK)) == | |
5bb16016 JK |
969 | (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK); |
970 | } | |
971 | ||
6f05485d JK |
972 | static inline bool is_debug(u32 intr_info) |
973 | { | |
974 | return is_exception_n(intr_info, DB_VECTOR); | |
975 | } | |
976 | ||
977 | static inline bool is_breakpoint(u32 intr_info) | |
978 | { | |
979 | return is_exception_n(intr_info, BP_VECTOR); | |
980 | } | |
981 | ||
5bb16016 JK |
982 | static inline bool is_page_fault(u32 intr_info) |
983 | { | |
984 | return is_exception_n(intr_info, PF_VECTOR); | |
6aa8b732 AK |
985 | } |
986 | ||
31299944 | 987 | static inline bool is_no_device(u32 intr_info) |
2ab455cc | 988 | { |
5bb16016 | 989 | return is_exception_n(intr_info, NM_VECTOR); |
2ab455cc AL |
990 | } |
991 | ||
31299944 | 992 | static inline bool is_invalid_opcode(u32 intr_info) |
7aa81cc0 | 993 | { |
5bb16016 | 994 | return is_exception_n(intr_info, UD_VECTOR); |
7aa81cc0 AL |
995 | } |
996 | ||
31299944 | 997 | static inline bool is_external_interrupt(u32 intr_info) |
6aa8b732 AK |
998 | { |
999 | return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK)) | |
1000 | == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK); | |
1001 | } | |
1002 | ||
31299944 | 1003 | static inline bool is_machine_check(u32 intr_info) |
a0861c02 AK |
1004 | { |
1005 | return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK | | |
1006 | INTR_INFO_VALID_MASK)) == | |
1007 | (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK); | |
1008 | } | |
1009 | ||
31299944 | 1010 | static inline bool cpu_has_vmx_msr_bitmap(void) |
25c5f225 | 1011 | { |
04547156 | 1012 | return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS; |
25c5f225 SY |
1013 | } |
1014 | ||
31299944 | 1015 | static inline bool cpu_has_vmx_tpr_shadow(void) |
6e5d865c | 1016 | { |
04547156 | 1017 | return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW; |
6e5d865c YS |
1018 | } |
1019 | ||
35754c98 | 1020 | static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu) |
6e5d865c | 1021 | { |
35754c98 | 1022 | return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu); |
6e5d865c YS |
1023 | } |
1024 | ||
31299944 | 1025 | static inline bool cpu_has_secondary_exec_ctrls(void) |
f78e0e2e | 1026 | { |
04547156 SY |
1027 | return vmcs_config.cpu_based_exec_ctrl & |
1028 | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS; | |
f78e0e2e SY |
1029 | } |
1030 | ||
774ead3a | 1031 | static inline bool cpu_has_vmx_virtualize_apic_accesses(void) |
f78e0e2e | 1032 | { |
04547156 SY |
1033 | return vmcs_config.cpu_based_2nd_exec_ctrl & |
1034 | SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES; | |
1035 | } | |
1036 | ||
8d14695f YZ |
1037 | static inline bool cpu_has_vmx_virtualize_x2apic_mode(void) |
1038 | { | |
1039 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
1040 | SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE; | |
1041 | } | |
1042 | ||
83d4c286 YZ |
1043 | static inline bool cpu_has_vmx_apic_register_virt(void) |
1044 | { | |
1045 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
1046 | SECONDARY_EXEC_APIC_REGISTER_VIRT; | |
1047 | } | |
1048 | ||
c7c9c56c YZ |
1049 | static inline bool cpu_has_vmx_virtual_intr_delivery(void) |
1050 | { | |
1051 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
1052 | SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY; | |
1053 | } | |
1054 | ||
01e439be YZ |
1055 | static inline bool cpu_has_vmx_posted_intr(void) |
1056 | { | |
d6a858d1 PB |
1057 | return IS_ENABLED(CONFIG_X86_LOCAL_APIC) && |
1058 | vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR; | |
01e439be YZ |
1059 | } |
1060 | ||
1061 | static inline bool cpu_has_vmx_apicv(void) | |
1062 | { | |
1063 | return cpu_has_vmx_apic_register_virt() && | |
1064 | cpu_has_vmx_virtual_intr_delivery() && | |
1065 | cpu_has_vmx_posted_intr(); | |
1066 | } | |
1067 | ||
04547156 SY |
1068 | static inline bool cpu_has_vmx_flexpriority(void) |
1069 | { | |
1070 | return cpu_has_vmx_tpr_shadow() && | |
1071 | cpu_has_vmx_virtualize_apic_accesses(); | |
f78e0e2e SY |
1072 | } |
1073 | ||
e799794e MT |
1074 | static inline bool cpu_has_vmx_ept_execute_only(void) |
1075 | { | |
31299944 | 1076 | return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT; |
e799794e MT |
1077 | } |
1078 | ||
e799794e MT |
1079 | static inline bool cpu_has_vmx_ept_2m_page(void) |
1080 | { | |
31299944 | 1081 | return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT; |
e799794e MT |
1082 | } |
1083 | ||
878403b7 SY |
1084 | static inline bool cpu_has_vmx_ept_1g_page(void) |
1085 | { | |
31299944 | 1086 | return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT; |
878403b7 SY |
1087 | } |
1088 | ||
4bc9b982 SY |
1089 | static inline bool cpu_has_vmx_ept_4levels(void) |
1090 | { | |
1091 | return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT; | |
1092 | } | |
1093 | ||
83c3a331 XH |
1094 | static inline bool cpu_has_vmx_ept_ad_bits(void) |
1095 | { | |
1096 | return vmx_capability.ept & VMX_EPT_AD_BIT; | |
1097 | } | |
1098 | ||
31299944 | 1099 | static inline bool cpu_has_vmx_invept_context(void) |
d56f546d | 1100 | { |
31299944 | 1101 | return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT; |
d56f546d SY |
1102 | } |
1103 | ||
31299944 | 1104 | static inline bool cpu_has_vmx_invept_global(void) |
d56f546d | 1105 | { |
31299944 | 1106 | return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT; |
d56f546d SY |
1107 | } |
1108 | ||
518c8aee GJ |
1109 | static inline bool cpu_has_vmx_invvpid_single(void) |
1110 | { | |
1111 | return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT; | |
1112 | } | |
1113 | ||
b9d762fa GJ |
1114 | static inline bool cpu_has_vmx_invvpid_global(void) |
1115 | { | |
1116 | return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT; | |
1117 | } | |
1118 | ||
31299944 | 1119 | static inline bool cpu_has_vmx_ept(void) |
d56f546d | 1120 | { |
04547156 SY |
1121 | return vmcs_config.cpu_based_2nd_exec_ctrl & |
1122 | SECONDARY_EXEC_ENABLE_EPT; | |
d56f546d SY |
1123 | } |
1124 | ||
31299944 | 1125 | static inline bool cpu_has_vmx_unrestricted_guest(void) |
3a624e29 NK |
1126 | { |
1127 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
1128 | SECONDARY_EXEC_UNRESTRICTED_GUEST; | |
1129 | } | |
1130 | ||
31299944 | 1131 | static inline bool cpu_has_vmx_ple(void) |
4b8d54f9 ZE |
1132 | { |
1133 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
1134 | SECONDARY_EXEC_PAUSE_LOOP_EXITING; | |
1135 | } | |
1136 | ||
35754c98 | 1137 | static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu) |
f78e0e2e | 1138 | { |
35754c98 | 1139 | return flexpriority_enabled && lapic_in_kernel(vcpu); |
f78e0e2e SY |
1140 | } |
1141 | ||
31299944 | 1142 | static inline bool cpu_has_vmx_vpid(void) |
2384d2b3 | 1143 | { |
04547156 SY |
1144 | return vmcs_config.cpu_based_2nd_exec_ctrl & |
1145 | SECONDARY_EXEC_ENABLE_VPID; | |
2384d2b3 SY |
1146 | } |
1147 | ||
31299944 | 1148 | static inline bool cpu_has_vmx_rdtscp(void) |
4e47c7a6 SY |
1149 | { |
1150 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
1151 | SECONDARY_EXEC_RDTSCP; | |
1152 | } | |
1153 | ||
ad756a16 MJ |
1154 | static inline bool cpu_has_vmx_invpcid(void) |
1155 | { | |
1156 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
1157 | SECONDARY_EXEC_ENABLE_INVPCID; | |
1158 | } | |
1159 | ||
31299944 | 1160 | static inline bool cpu_has_virtual_nmis(void) |
f08864b4 SY |
1161 | { |
1162 | return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS; | |
1163 | } | |
1164 | ||
f5f48ee1 SY |
1165 | static inline bool cpu_has_vmx_wbinvd_exit(void) |
1166 | { | |
1167 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
1168 | SECONDARY_EXEC_WBINVD_EXITING; | |
1169 | } | |
1170 | ||
abc4fc58 AG |
1171 | static inline bool cpu_has_vmx_shadow_vmcs(void) |
1172 | { | |
1173 | u64 vmx_msr; | |
1174 | rdmsrl(MSR_IA32_VMX_MISC, vmx_msr); | |
1175 | /* check if the cpu supports writing r/o exit information fields */ | |
1176 | if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) | |
1177 | return false; | |
1178 | ||
1179 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
1180 | SECONDARY_EXEC_SHADOW_VMCS; | |
1181 | } | |
1182 | ||
843e4330 KH |
1183 | static inline bool cpu_has_vmx_pml(void) |
1184 | { | |
1185 | return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML; | |
1186 | } | |
1187 | ||
64903d61 HZ |
1188 | static inline bool cpu_has_vmx_tsc_scaling(void) |
1189 | { | |
1190 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
1191 | SECONDARY_EXEC_TSC_SCALING; | |
1192 | } | |
1193 | ||
04547156 SY |
1194 | static inline bool report_flexpriority(void) |
1195 | { | |
1196 | return flexpriority_enabled; | |
1197 | } | |
1198 | ||
fe3ef05c NHE |
1199 | static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit) |
1200 | { | |
1201 | return vmcs12->cpu_based_vm_exec_control & bit; | |
1202 | } | |
1203 | ||
1204 | static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit) | |
1205 | { | |
1206 | return (vmcs12->cpu_based_vm_exec_control & | |
1207 | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) && | |
1208 | (vmcs12->secondary_vm_exec_control & bit); | |
1209 | } | |
1210 | ||
f5c4368f | 1211 | static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12) |
644d711a NHE |
1212 | { |
1213 | return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS; | |
1214 | } | |
1215 | ||
f4124500 JK |
1216 | static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12) |
1217 | { | |
1218 | return vmcs12->pin_based_vm_exec_control & | |
1219 | PIN_BASED_VMX_PREEMPTION_TIMER; | |
1220 | } | |
1221 | ||
155a97a3 NHE |
1222 | static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12) |
1223 | { | |
1224 | return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT); | |
1225 | } | |
1226 | ||
81dc01f7 WL |
1227 | static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12) |
1228 | { | |
1229 | return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) && | |
1230 | vmx_xsaves_supported(); | |
1231 | } | |
1232 | ||
f2b93280 WV |
1233 | static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12) |
1234 | { | |
1235 | return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE); | |
1236 | } | |
1237 | ||
5c614b35 WL |
1238 | static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12) |
1239 | { | |
1240 | return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID); | |
1241 | } | |
1242 | ||
82f0dd4b WV |
1243 | static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12) |
1244 | { | |
1245 | return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT); | |
1246 | } | |
1247 | ||
608406e2 WV |
1248 | static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12) |
1249 | { | |
1250 | return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY); | |
1251 | } | |
1252 | ||
705699a1 WV |
1253 | static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12) |
1254 | { | |
1255 | return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR; | |
1256 | } | |
1257 | ||
644d711a NHE |
1258 | static inline bool is_exception(u32 intr_info) |
1259 | { | |
1260 | return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK)) | |
1261 | == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK); | |
1262 | } | |
1263 | ||
533558bc JK |
1264 | static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason, |
1265 | u32 exit_intr_info, | |
1266 | unsigned long exit_qualification); | |
7c177938 NHE |
1267 | static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu, |
1268 | struct vmcs12 *vmcs12, | |
1269 | u32 reason, unsigned long qualification); | |
1270 | ||
8b9cf98c | 1271 | static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr) |
7725f0ba AK |
1272 | { |
1273 | int i; | |
1274 | ||
a2fa3e9f | 1275 | for (i = 0; i < vmx->nmsrs; ++i) |
26bb0981 | 1276 | if (vmx_msr_index[vmx->guest_msrs[i].index] == msr) |
a75beee6 ED |
1277 | return i; |
1278 | return -1; | |
1279 | } | |
1280 | ||
2384d2b3 SY |
1281 | static inline void __invvpid(int ext, u16 vpid, gva_t gva) |
1282 | { | |
1283 | struct { | |
1284 | u64 vpid : 16; | |
1285 | u64 rsvd : 48; | |
1286 | u64 gva; | |
1287 | } operand = { vpid, 0, gva }; | |
1288 | ||
4ecac3fd | 1289 | asm volatile (__ex(ASM_VMX_INVVPID) |
2384d2b3 SY |
1290 | /* CF==1 or ZF==1 --> rc = -1 */ |
1291 | "; ja 1f ; ud2 ; 1:" | |
1292 | : : "a"(&operand), "c"(ext) : "cc", "memory"); | |
1293 | } | |
1294 | ||
1439442c SY |
1295 | static inline void __invept(int ext, u64 eptp, gpa_t gpa) |
1296 | { | |
1297 | struct { | |
1298 | u64 eptp, gpa; | |
1299 | } operand = {eptp, gpa}; | |
1300 | ||
4ecac3fd | 1301 | asm volatile (__ex(ASM_VMX_INVEPT) |
1439442c SY |
1302 | /* CF==1 or ZF==1 --> rc = -1 */ |
1303 | "; ja 1f ; ud2 ; 1:\n" | |
1304 | : : "a" (&operand), "c" (ext) : "cc", "memory"); | |
1305 | } | |
1306 | ||
26bb0981 | 1307 | static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr) |
a75beee6 ED |
1308 | { |
1309 | int i; | |
1310 | ||
8b9cf98c | 1311 | i = __find_msr_index(vmx, msr); |
a75beee6 | 1312 | if (i >= 0) |
a2fa3e9f | 1313 | return &vmx->guest_msrs[i]; |
8b6d44c7 | 1314 | return NULL; |
7725f0ba AK |
1315 | } |
1316 | ||
6aa8b732 AK |
1317 | static void vmcs_clear(struct vmcs *vmcs) |
1318 | { | |
1319 | u64 phys_addr = __pa(vmcs); | |
1320 | u8 error; | |
1321 | ||
4ecac3fd | 1322 | asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0" |
16d8f72f | 1323 | : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr) |
6aa8b732 AK |
1324 | : "cc", "memory"); |
1325 | if (error) | |
1326 | printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n", | |
1327 | vmcs, phys_addr); | |
1328 | } | |
1329 | ||
d462b819 NHE |
1330 | static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs) |
1331 | { | |
1332 | vmcs_clear(loaded_vmcs->vmcs); | |
1333 | loaded_vmcs->cpu = -1; | |
1334 | loaded_vmcs->launched = 0; | |
1335 | } | |
1336 | ||
7725b894 DX |
1337 | static void vmcs_load(struct vmcs *vmcs) |
1338 | { | |
1339 | u64 phys_addr = __pa(vmcs); | |
1340 | u8 error; | |
1341 | ||
1342 | asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0" | |
16d8f72f | 1343 | : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr) |
7725b894 DX |
1344 | : "cc", "memory"); |
1345 | if (error) | |
2844d849 | 1346 | printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n", |
7725b894 DX |
1347 | vmcs, phys_addr); |
1348 | } | |
1349 | ||
2965faa5 | 1350 | #ifdef CONFIG_KEXEC_CORE |
8f536b76 ZY |
1351 | /* |
1352 | * This bitmap is used to indicate whether the vmclear | |
1353 | * operation is enabled on all cpus. All disabled by | |
1354 | * default. | |
1355 | */ | |
1356 | static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE; | |
1357 | ||
1358 | static inline void crash_enable_local_vmclear(int cpu) | |
1359 | { | |
1360 | cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap); | |
1361 | } | |
1362 | ||
1363 | static inline void crash_disable_local_vmclear(int cpu) | |
1364 | { | |
1365 | cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap); | |
1366 | } | |
1367 | ||
1368 | static inline int crash_local_vmclear_enabled(int cpu) | |
1369 | { | |
1370 | return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap); | |
1371 | } | |
1372 | ||
1373 | static void crash_vmclear_local_loaded_vmcss(void) | |
1374 | { | |
1375 | int cpu = raw_smp_processor_id(); | |
1376 | struct loaded_vmcs *v; | |
1377 | ||
1378 | if (!crash_local_vmclear_enabled(cpu)) | |
1379 | return; | |
1380 | ||
1381 | list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu), | |
1382 | loaded_vmcss_on_cpu_link) | |
1383 | vmcs_clear(v->vmcs); | |
1384 | } | |
1385 | #else | |
1386 | static inline void crash_enable_local_vmclear(int cpu) { } | |
1387 | static inline void crash_disable_local_vmclear(int cpu) { } | |
2965faa5 | 1388 | #endif /* CONFIG_KEXEC_CORE */ |
8f536b76 | 1389 | |
d462b819 | 1390 | static void __loaded_vmcs_clear(void *arg) |
6aa8b732 | 1391 | { |
d462b819 | 1392 | struct loaded_vmcs *loaded_vmcs = arg; |
d3b2c338 | 1393 | int cpu = raw_smp_processor_id(); |
6aa8b732 | 1394 | |
d462b819 NHE |
1395 | if (loaded_vmcs->cpu != cpu) |
1396 | return; /* vcpu migration can race with cpu offline */ | |
1397 | if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs) | |
6aa8b732 | 1398 | per_cpu(current_vmcs, cpu) = NULL; |
8f536b76 | 1399 | crash_disable_local_vmclear(cpu); |
d462b819 | 1400 | list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link); |
5a560f8b XG |
1401 | |
1402 | /* | |
1403 | * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link | |
1404 | * is before setting loaded_vmcs->vcpu to -1 which is done in | |
1405 | * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist | |
1406 | * then adds the vmcs into percpu list before it is deleted. | |
1407 | */ | |
1408 | smp_wmb(); | |
1409 | ||
d462b819 | 1410 | loaded_vmcs_init(loaded_vmcs); |
8f536b76 | 1411 | crash_enable_local_vmclear(cpu); |
6aa8b732 AK |
1412 | } |
1413 | ||
d462b819 | 1414 | static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs) |
8d0be2b3 | 1415 | { |
e6c7d321 XG |
1416 | int cpu = loaded_vmcs->cpu; |
1417 | ||
1418 | if (cpu != -1) | |
1419 | smp_call_function_single(cpu, | |
1420 | __loaded_vmcs_clear, loaded_vmcs, 1); | |
8d0be2b3 AK |
1421 | } |
1422 | ||
dd5f5341 | 1423 | static inline void vpid_sync_vcpu_single(int vpid) |
2384d2b3 | 1424 | { |
dd5f5341 | 1425 | if (vpid == 0) |
2384d2b3 SY |
1426 | return; |
1427 | ||
518c8aee | 1428 | if (cpu_has_vmx_invvpid_single()) |
dd5f5341 | 1429 | __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0); |
2384d2b3 SY |
1430 | } |
1431 | ||
b9d762fa GJ |
1432 | static inline void vpid_sync_vcpu_global(void) |
1433 | { | |
1434 | if (cpu_has_vmx_invvpid_global()) | |
1435 | __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0); | |
1436 | } | |
1437 | ||
dd5f5341 | 1438 | static inline void vpid_sync_context(int vpid) |
b9d762fa GJ |
1439 | { |
1440 | if (cpu_has_vmx_invvpid_single()) | |
dd5f5341 | 1441 | vpid_sync_vcpu_single(vpid); |
b9d762fa GJ |
1442 | else |
1443 | vpid_sync_vcpu_global(); | |
1444 | } | |
1445 | ||
1439442c SY |
1446 | static inline void ept_sync_global(void) |
1447 | { | |
1448 | if (cpu_has_vmx_invept_global()) | |
1449 | __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0); | |
1450 | } | |
1451 | ||
1452 | static inline void ept_sync_context(u64 eptp) | |
1453 | { | |
089d034e | 1454 | if (enable_ept) { |
1439442c SY |
1455 | if (cpu_has_vmx_invept_context()) |
1456 | __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0); | |
1457 | else | |
1458 | ept_sync_global(); | |
1459 | } | |
1460 | } | |
1461 | ||
8a86aea9 PB |
1462 | static __always_inline void vmcs_check16(unsigned long field) |
1463 | { | |
1464 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000, | |
1465 | "16-bit accessor invalid for 64-bit field"); | |
1466 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001, | |
1467 | "16-bit accessor invalid for 64-bit high field"); | |
1468 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000, | |
1469 | "16-bit accessor invalid for 32-bit high field"); | |
1470 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000, | |
1471 | "16-bit accessor invalid for natural width field"); | |
1472 | } | |
1473 | ||
1474 | static __always_inline void vmcs_check32(unsigned long field) | |
1475 | { | |
1476 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0, | |
1477 | "32-bit accessor invalid for 16-bit field"); | |
1478 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000, | |
1479 | "32-bit accessor invalid for natural width field"); | |
1480 | } | |
1481 | ||
1482 | static __always_inline void vmcs_check64(unsigned long field) | |
1483 | { | |
1484 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0, | |
1485 | "64-bit accessor invalid for 16-bit field"); | |
1486 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001, | |
1487 | "64-bit accessor invalid for 64-bit high field"); | |
1488 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000, | |
1489 | "64-bit accessor invalid for 32-bit field"); | |
1490 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000, | |
1491 | "64-bit accessor invalid for natural width field"); | |
1492 | } | |
1493 | ||
1494 | static __always_inline void vmcs_checkl(unsigned long field) | |
1495 | { | |
1496 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0, | |
1497 | "Natural width accessor invalid for 16-bit field"); | |
1498 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000, | |
1499 | "Natural width accessor invalid for 64-bit field"); | |
1500 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001, | |
1501 | "Natural width accessor invalid for 64-bit high field"); | |
1502 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000, | |
1503 | "Natural width accessor invalid for 32-bit field"); | |
1504 | } | |
1505 | ||
1506 | static __always_inline unsigned long __vmcs_readl(unsigned long field) | |
6aa8b732 | 1507 | { |
5e520e62 | 1508 | unsigned long value; |
6aa8b732 | 1509 | |
5e520e62 AK |
1510 | asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0") |
1511 | : "=a"(value) : "d"(field) : "cc"); | |
6aa8b732 AK |
1512 | return value; |
1513 | } | |
1514 | ||
96304217 | 1515 | static __always_inline u16 vmcs_read16(unsigned long field) |
6aa8b732 | 1516 | { |
8a86aea9 PB |
1517 | vmcs_check16(field); |
1518 | return __vmcs_readl(field); | |
6aa8b732 AK |
1519 | } |
1520 | ||
96304217 | 1521 | static __always_inline u32 vmcs_read32(unsigned long field) |
6aa8b732 | 1522 | { |
8a86aea9 PB |
1523 | vmcs_check32(field); |
1524 | return __vmcs_readl(field); | |
6aa8b732 AK |
1525 | } |
1526 | ||
96304217 | 1527 | static __always_inline u64 vmcs_read64(unsigned long field) |
6aa8b732 | 1528 | { |
8a86aea9 | 1529 | vmcs_check64(field); |
05b3e0c2 | 1530 | #ifdef CONFIG_X86_64 |
8a86aea9 | 1531 | return __vmcs_readl(field); |
6aa8b732 | 1532 | #else |
8a86aea9 | 1533 | return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32); |
6aa8b732 AK |
1534 | #endif |
1535 | } | |
1536 | ||
8a86aea9 PB |
1537 | static __always_inline unsigned long vmcs_readl(unsigned long field) |
1538 | { | |
1539 | vmcs_checkl(field); | |
1540 | return __vmcs_readl(field); | |
1541 | } | |
1542 | ||
e52de1b8 AK |
1543 | static noinline void vmwrite_error(unsigned long field, unsigned long value) |
1544 | { | |
1545 | printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n", | |
1546 | field, value, vmcs_read32(VM_INSTRUCTION_ERROR)); | |
1547 | dump_stack(); | |
1548 | } | |
1549 | ||
8a86aea9 | 1550 | static __always_inline void __vmcs_writel(unsigned long field, unsigned long value) |
6aa8b732 AK |
1551 | { |
1552 | u8 error; | |
1553 | ||
4ecac3fd | 1554 | asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0" |
d77c26fc | 1555 | : "=q"(error) : "a"(value), "d"(field) : "cc"); |
e52de1b8 AK |
1556 | if (unlikely(error)) |
1557 | vmwrite_error(field, value); | |
6aa8b732 AK |
1558 | } |
1559 | ||
8a86aea9 | 1560 | static __always_inline void vmcs_write16(unsigned long field, u16 value) |
6aa8b732 | 1561 | { |
8a86aea9 PB |
1562 | vmcs_check16(field); |
1563 | __vmcs_writel(field, value); | |
6aa8b732 AK |
1564 | } |
1565 | ||
8a86aea9 | 1566 | static __always_inline void vmcs_write32(unsigned long field, u32 value) |
6aa8b732 | 1567 | { |
8a86aea9 PB |
1568 | vmcs_check32(field); |
1569 | __vmcs_writel(field, value); | |
6aa8b732 AK |
1570 | } |
1571 | ||
8a86aea9 | 1572 | static __always_inline void vmcs_write64(unsigned long field, u64 value) |
6aa8b732 | 1573 | { |
8a86aea9 PB |
1574 | vmcs_check64(field); |
1575 | __vmcs_writel(field, value); | |
7682f2d0 | 1576 | #ifndef CONFIG_X86_64 |
6aa8b732 | 1577 | asm volatile (""); |
8a86aea9 | 1578 | __vmcs_writel(field+1, value >> 32); |
6aa8b732 AK |
1579 | #endif |
1580 | } | |
1581 | ||
8a86aea9 | 1582 | static __always_inline void vmcs_writel(unsigned long field, unsigned long value) |
2ab455cc | 1583 | { |
8a86aea9 PB |
1584 | vmcs_checkl(field); |
1585 | __vmcs_writel(field, value); | |
2ab455cc AL |
1586 | } |
1587 | ||
8a86aea9 | 1588 | static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask) |
2ab455cc | 1589 | { |
8a86aea9 PB |
1590 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000, |
1591 | "vmcs_clear_bits does not support 64-bit fields"); | |
1592 | __vmcs_writel(field, __vmcs_readl(field) & ~mask); | |
2ab455cc AL |
1593 | } |
1594 | ||
8a86aea9 | 1595 | static __always_inline void vmcs_set_bits(unsigned long field, u32 mask) |
2ab455cc | 1596 | { |
8a86aea9 PB |
1597 | BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000, |
1598 | "vmcs_set_bits does not support 64-bit fields"); | |
1599 | __vmcs_writel(field, __vmcs_readl(field) | mask); | |
2ab455cc AL |
1600 | } |
1601 | ||
2961e876 GN |
1602 | static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val) |
1603 | { | |
1604 | vmcs_write32(VM_ENTRY_CONTROLS, val); | |
1605 | vmx->vm_entry_controls_shadow = val; | |
1606 | } | |
1607 | ||
1608 | static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val) | |
1609 | { | |
1610 | if (vmx->vm_entry_controls_shadow != val) | |
1611 | vm_entry_controls_init(vmx, val); | |
1612 | } | |
1613 | ||
1614 | static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx) | |
1615 | { | |
1616 | return vmx->vm_entry_controls_shadow; | |
1617 | } | |
1618 | ||
1619 | ||
1620 | static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val) | |
1621 | { | |
1622 | vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val); | |
1623 | } | |
1624 | ||
1625 | static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val) | |
1626 | { | |
1627 | vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val); | |
1628 | } | |
1629 | ||
1630 | static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val) | |
1631 | { | |
1632 | vmcs_write32(VM_EXIT_CONTROLS, val); | |
1633 | vmx->vm_exit_controls_shadow = val; | |
1634 | } | |
1635 | ||
1636 | static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val) | |
1637 | { | |
1638 | if (vmx->vm_exit_controls_shadow != val) | |
1639 | vm_exit_controls_init(vmx, val); | |
1640 | } | |
1641 | ||
1642 | static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx) | |
1643 | { | |
1644 | return vmx->vm_exit_controls_shadow; | |
1645 | } | |
1646 | ||
1647 | ||
1648 | static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val) | |
1649 | { | |
1650 | vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val); | |
1651 | } | |
1652 | ||
1653 | static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val) | |
1654 | { | |
1655 | vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val); | |
1656 | } | |
1657 | ||
2fb92db1 AK |
1658 | static void vmx_segment_cache_clear(struct vcpu_vmx *vmx) |
1659 | { | |
1660 | vmx->segment_cache.bitmask = 0; | |
1661 | } | |
1662 | ||
1663 | static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg, | |
1664 | unsigned field) | |
1665 | { | |
1666 | bool ret; | |
1667 | u32 mask = 1 << (seg * SEG_FIELD_NR + field); | |
1668 | ||
1669 | if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) { | |
1670 | vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS); | |
1671 | vmx->segment_cache.bitmask = 0; | |
1672 | } | |
1673 | ret = vmx->segment_cache.bitmask & mask; | |
1674 | vmx->segment_cache.bitmask |= mask; | |
1675 | return ret; | |
1676 | } | |
1677 | ||
1678 | static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg) | |
1679 | { | |
1680 | u16 *p = &vmx->segment_cache.seg[seg].selector; | |
1681 | ||
1682 | if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL)) | |
1683 | *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector); | |
1684 | return *p; | |
1685 | } | |
1686 | ||
1687 | static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg) | |
1688 | { | |
1689 | ulong *p = &vmx->segment_cache.seg[seg].base; | |
1690 | ||
1691 | if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE)) | |
1692 | *p = vmcs_readl(kvm_vmx_segment_fields[seg].base); | |
1693 | return *p; | |
1694 | } | |
1695 | ||
1696 | static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg) | |
1697 | { | |
1698 | u32 *p = &vmx->segment_cache.seg[seg].limit; | |
1699 | ||
1700 | if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT)) | |
1701 | *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit); | |
1702 | return *p; | |
1703 | } | |
1704 | ||
1705 | static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg) | |
1706 | { | |
1707 | u32 *p = &vmx->segment_cache.seg[seg].ar; | |
1708 | ||
1709 | if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR)) | |
1710 | *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes); | |
1711 | return *p; | |
1712 | } | |
1713 | ||
abd3f2d6 AK |
1714 | static void update_exception_bitmap(struct kvm_vcpu *vcpu) |
1715 | { | |
1716 | u32 eb; | |
1717 | ||
fd7373cc | 1718 | eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) | |
54a20552 | 1719 | (1u << NM_VECTOR) | (1u << DB_VECTOR) | (1u << AC_VECTOR); |
fd7373cc JK |
1720 | if ((vcpu->guest_debug & |
1721 | (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) == | |
1722 | (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) | |
1723 | eb |= 1u << BP_VECTOR; | |
7ffd92c5 | 1724 | if (to_vmx(vcpu)->rmode.vm86_active) |
abd3f2d6 | 1725 | eb = ~0; |
089d034e | 1726 | if (enable_ept) |
1439442c | 1727 | eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */ |
02daab21 AK |
1728 | if (vcpu->fpu_active) |
1729 | eb &= ~(1u << NM_VECTOR); | |
36cf24e0 NHE |
1730 | |
1731 | /* When we are running a nested L2 guest and L1 specified for it a | |
1732 | * certain exception bitmap, we must trap the same exceptions and pass | |
1733 | * them to L1. When running L2, we will only handle the exceptions | |
1734 | * specified above if L1 did not want them. | |
1735 | */ | |
1736 | if (is_guest_mode(vcpu)) | |
1737 | eb |= get_vmcs12(vcpu)->exception_bitmap; | |
1738 | ||
abd3f2d6 AK |
1739 | vmcs_write32(EXCEPTION_BITMAP, eb); |
1740 | } | |
1741 | ||
2961e876 GN |
1742 | static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx, |
1743 | unsigned long entry, unsigned long exit) | |
8bf00a52 | 1744 | { |
2961e876 GN |
1745 | vm_entry_controls_clearbit(vmx, entry); |
1746 | vm_exit_controls_clearbit(vmx, exit); | |
8bf00a52 GN |
1747 | } |
1748 | ||
61d2ef2c AK |
1749 | static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr) |
1750 | { | |
1751 | unsigned i; | |
1752 | struct msr_autoload *m = &vmx->msr_autoload; | |
1753 | ||
8bf00a52 GN |
1754 | switch (msr) { |
1755 | case MSR_EFER: | |
1756 | if (cpu_has_load_ia32_efer) { | |
2961e876 GN |
1757 | clear_atomic_switch_msr_special(vmx, |
1758 | VM_ENTRY_LOAD_IA32_EFER, | |
8bf00a52 GN |
1759 | VM_EXIT_LOAD_IA32_EFER); |
1760 | return; | |
1761 | } | |
1762 | break; | |
1763 | case MSR_CORE_PERF_GLOBAL_CTRL: | |
1764 | if (cpu_has_load_perf_global_ctrl) { | |
2961e876 | 1765 | clear_atomic_switch_msr_special(vmx, |
8bf00a52 GN |
1766 | VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL, |
1767 | VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL); | |
1768 | return; | |
1769 | } | |
1770 | break; | |
110312c8 AK |
1771 | } |
1772 | ||
61d2ef2c AK |
1773 | for (i = 0; i < m->nr; ++i) |
1774 | if (m->guest[i].index == msr) | |
1775 | break; | |
1776 | ||
1777 | if (i == m->nr) | |
1778 | return; | |
1779 | --m->nr; | |
1780 | m->guest[i] = m->guest[m->nr]; | |
1781 | m->host[i] = m->host[m->nr]; | |
1782 | vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr); | |
1783 | vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr); | |
1784 | } | |
1785 | ||
2961e876 GN |
1786 | static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx, |
1787 | unsigned long entry, unsigned long exit, | |
1788 | unsigned long guest_val_vmcs, unsigned long host_val_vmcs, | |
1789 | u64 guest_val, u64 host_val) | |
8bf00a52 GN |
1790 | { |
1791 | vmcs_write64(guest_val_vmcs, guest_val); | |
1792 | vmcs_write64(host_val_vmcs, host_val); | |
2961e876 GN |
1793 | vm_entry_controls_setbit(vmx, entry); |
1794 | vm_exit_controls_setbit(vmx, exit); | |
8bf00a52 GN |
1795 | } |
1796 | ||
61d2ef2c AK |
1797 | static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr, |
1798 | u64 guest_val, u64 host_val) | |
1799 | { | |
1800 | unsigned i; | |
1801 | struct msr_autoload *m = &vmx->msr_autoload; | |
1802 | ||
8bf00a52 GN |
1803 | switch (msr) { |
1804 | case MSR_EFER: | |
1805 | if (cpu_has_load_ia32_efer) { | |
2961e876 GN |
1806 | add_atomic_switch_msr_special(vmx, |
1807 | VM_ENTRY_LOAD_IA32_EFER, | |
8bf00a52 GN |
1808 | VM_EXIT_LOAD_IA32_EFER, |
1809 | GUEST_IA32_EFER, | |
1810 | HOST_IA32_EFER, | |
1811 | guest_val, host_val); | |
1812 | return; | |
1813 | } | |
1814 | break; | |
1815 | case MSR_CORE_PERF_GLOBAL_CTRL: | |
1816 | if (cpu_has_load_perf_global_ctrl) { | |
2961e876 | 1817 | add_atomic_switch_msr_special(vmx, |
8bf00a52 GN |
1818 | VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL, |
1819 | VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL, | |
1820 | GUEST_IA32_PERF_GLOBAL_CTRL, | |
1821 | HOST_IA32_PERF_GLOBAL_CTRL, | |
1822 | guest_val, host_val); | |
1823 | return; | |
1824 | } | |
1825 | break; | |
7099e2e1 RK |
1826 | case MSR_IA32_PEBS_ENABLE: |
1827 | /* PEBS needs a quiescent period after being disabled (to write | |
1828 | * a record). Disabling PEBS through VMX MSR swapping doesn't | |
1829 | * provide that period, so a CPU could write host's record into | |
1830 | * guest's memory. | |
1831 | */ | |
1832 | wrmsrl(MSR_IA32_PEBS_ENABLE, 0); | |
110312c8 AK |
1833 | } |
1834 | ||
61d2ef2c AK |
1835 | for (i = 0; i < m->nr; ++i) |
1836 | if (m->guest[i].index == msr) | |
1837 | break; | |
1838 | ||
e7fc6f93 | 1839 | if (i == NR_AUTOLOAD_MSRS) { |
60266204 | 1840 | printk_once(KERN_WARNING "Not enough msr switch entries. " |
e7fc6f93 GN |
1841 | "Can't add msr %x\n", msr); |
1842 | return; | |
1843 | } else if (i == m->nr) { | |
61d2ef2c AK |
1844 | ++m->nr; |
1845 | vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr); | |
1846 | vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr); | |
1847 | } | |
1848 | ||
1849 | m->guest[i].index = msr; | |
1850 | m->guest[i].value = guest_val; | |
1851 | m->host[i].index = msr; | |
1852 | m->host[i].value = host_val; | |
1853 | } | |
1854 | ||
33ed6329 AK |
1855 | static void reload_tss(void) |
1856 | { | |
33ed6329 AK |
1857 | /* |
1858 | * VT restores TR but not its size. Useless. | |
1859 | */ | |
89cbc767 | 1860 | struct desc_ptr *gdt = this_cpu_ptr(&host_gdt); |
a5f61300 | 1861 | struct desc_struct *descs; |
33ed6329 | 1862 | |
d359192f | 1863 | descs = (void *)gdt->address; |
33ed6329 AK |
1864 | descs[GDT_ENTRY_TSS].type = 9; /* available TSS */ |
1865 | load_TR_desc(); | |
33ed6329 AK |
1866 | } |
1867 | ||
92c0d900 | 1868 | static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset) |
2cc51560 | 1869 | { |
844a5fe2 PB |
1870 | u64 guest_efer = vmx->vcpu.arch.efer; |
1871 | u64 ignore_bits = 0; | |
1872 | ||
1873 | if (!enable_ept) { | |
1874 | /* | |
1875 | * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing | |
1876 | * host CPUID is more efficient than testing guest CPUID | |
1877 | * or CR4. Host SMEP is anyway a requirement for guest SMEP. | |
1878 | */ | |
1879 | if (boot_cpu_has(X86_FEATURE_SMEP)) | |
1880 | guest_efer |= EFER_NX; | |
1881 | else if (!(guest_efer & EFER_NX)) | |
1882 | ignore_bits |= EFER_NX; | |
1883 | } | |
3a34a881 | 1884 | |
51c6cf66 | 1885 | /* |
844a5fe2 | 1886 | * LMA and LME handled by hardware; SCE meaningless outside long mode. |
51c6cf66 | 1887 | */ |
844a5fe2 | 1888 | ignore_bits |= EFER_SCE; |
51c6cf66 AK |
1889 | #ifdef CONFIG_X86_64 |
1890 | ignore_bits |= EFER_LMA | EFER_LME; | |
1891 | /* SCE is meaningful only in long mode on Intel */ | |
1892 | if (guest_efer & EFER_LMA) | |
1893 | ignore_bits &= ~(u64)EFER_SCE; | |
1894 | #endif | |
84ad33ef AK |
1895 | |
1896 | clear_atomic_switch_msr(vmx, MSR_EFER); | |
f6577a5f AL |
1897 | |
1898 | /* | |
1899 | * On EPT, we can't emulate NX, so we must switch EFER atomically. | |
1900 | * On CPUs that support "load IA32_EFER", always switch EFER | |
1901 | * atomically, since it's faster than switching it manually. | |
1902 | */ | |
1903 | if (cpu_has_load_ia32_efer || | |
1904 | (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) { | |
84ad33ef AK |
1905 | if (!(guest_efer & EFER_LMA)) |
1906 | guest_efer &= ~EFER_LME; | |
54b98bff AL |
1907 | if (guest_efer != host_efer) |
1908 | add_atomic_switch_msr(vmx, MSR_EFER, | |
1909 | guest_efer, host_efer); | |
84ad33ef | 1910 | return false; |
844a5fe2 PB |
1911 | } else { |
1912 | guest_efer &= ~ignore_bits; | |
1913 | guest_efer |= host_efer & ignore_bits; | |
1914 | ||
1915 | vmx->guest_msrs[efer_offset].data = guest_efer; | |
1916 | vmx->guest_msrs[efer_offset].mask = ~ignore_bits; | |
84ad33ef | 1917 | |
844a5fe2 PB |
1918 | return true; |
1919 | } | |
51c6cf66 AK |
1920 | } |
1921 | ||
2d49ec72 GN |
1922 | static unsigned long segment_base(u16 selector) |
1923 | { | |
89cbc767 | 1924 | struct desc_ptr *gdt = this_cpu_ptr(&host_gdt); |
2d49ec72 GN |
1925 | struct desc_struct *d; |
1926 | unsigned long table_base; | |
1927 | unsigned long v; | |
1928 | ||
1929 | if (!(selector & ~3)) | |
1930 | return 0; | |
1931 | ||
d359192f | 1932 | table_base = gdt->address; |
2d49ec72 GN |
1933 | |
1934 | if (selector & 4) { /* from ldt */ | |
1935 | u16 ldt_selector = kvm_read_ldt(); | |
1936 | ||
1937 | if (!(ldt_selector & ~3)) | |
1938 | return 0; | |
1939 | ||
1940 | table_base = segment_base(ldt_selector); | |
1941 | } | |
1942 | d = (struct desc_struct *)(table_base + (selector & ~7)); | |
1943 | v = get_desc_base(d); | |
1944 | #ifdef CONFIG_X86_64 | |
1945 | if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11)) | |
1946 | v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32; | |
1947 | #endif | |
1948 | return v; | |
1949 | } | |
1950 | ||
1951 | static inline unsigned long kvm_read_tr_base(void) | |
1952 | { | |
1953 | u16 tr; | |
1954 | asm("str %0" : "=g"(tr)); | |
1955 | return segment_base(tr); | |
1956 | } | |
1957 | ||
04d2cc77 | 1958 | static void vmx_save_host_state(struct kvm_vcpu *vcpu) |
33ed6329 | 1959 | { |
04d2cc77 | 1960 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
26bb0981 | 1961 | int i; |
04d2cc77 | 1962 | |
a2fa3e9f | 1963 | if (vmx->host_state.loaded) |
33ed6329 AK |
1964 | return; |
1965 | ||
a2fa3e9f | 1966 | vmx->host_state.loaded = 1; |
33ed6329 AK |
1967 | /* |
1968 | * Set host fs and gs selectors. Unfortunately, 22.2.3 does not | |
1969 | * allow segment selectors with cpl > 0 or ti == 1. | |
1970 | */ | |
d6e88aec | 1971 | vmx->host_state.ldt_sel = kvm_read_ldt(); |
152d3f2f | 1972 | vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel; |
9581d442 | 1973 | savesegment(fs, vmx->host_state.fs_sel); |
152d3f2f | 1974 | if (!(vmx->host_state.fs_sel & 7)) { |
a2fa3e9f | 1975 | vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel); |
152d3f2f LV |
1976 | vmx->host_state.fs_reload_needed = 0; |
1977 | } else { | |
33ed6329 | 1978 | vmcs_write16(HOST_FS_SELECTOR, 0); |
152d3f2f | 1979 | vmx->host_state.fs_reload_needed = 1; |
33ed6329 | 1980 | } |
9581d442 | 1981 | savesegment(gs, vmx->host_state.gs_sel); |
a2fa3e9f GH |
1982 | if (!(vmx->host_state.gs_sel & 7)) |
1983 | vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel); | |
33ed6329 AK |
1984 | else { |
1985 | vmcs_write16(HOST_GS_SELECTOR, 0); | |
152d3f2f | 1986 | vmx->host_state.gs_ldt_reload_needed = 1; |
33ed6329 AK |
1987 | } |
1988 | ||
b2da15ac AK |
1989 | #ifdef CONFIG_X86_64 |
1990 | savesegment(ds, vmx->host_state.ds_sel); | |
1991 | savesegment(es, vmx->host_state.es_sel); | |
1992 | #endif | |
1993 | ||
33ed6329 AK |
1994 | #ifdef CONFIG_X86_64 |
1995 | vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE)); | |
1996 | vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE)); | |
1997 | #else | |
a2fa3e9f GH |
1998 | vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel)); |
1999 | vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel)); | |
33ed6329 | 2000 | #endif |
707c0874 AK |
2001 | |
2002 | #ifdef CONFIG_X86_64 | |
c8770e7b AK |
2003 | rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base); |
2004 | if (is_long_mode(&vmx->vcpu)) | |
44ea2b17 | 2005 | wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base); |
707c0874 | 2006 | #endif |
da8999d3 LJ |
2007 | if (boot_cpu_has(X86_FEATURE_MPX)) |
2008 | rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs); | |
26bb0981 AK |
2009 | for (i = 0; i < vmx->save_nmsrs; ++i) |
2010 | kvm_set_shared_msr(vmx->guest_msrs[i].index, | |
d5696725 AK |
2011 | vmx->guest_msrs[i].data, |
2012 | vmx->guest_msrs[i].mask); | |
33ed6329 AK |
2013 | } |
2014 | ||
a9b21b62 | 2015 | static void __vmx_load_host_state(struct vcpu_vmx *vmx) |
33ed6329 | 2016 | { |
a2fa3e9f | 2017 | if (!vmx->host_state.loaded) |
33ed6329 AK |
2018 | return; |
2019 | ||
e1beb1d3 | 2020 | ++vmx->vcpu.stat.host_state_reload; |
a2fa3e9f | 2021 | vmx->host_state.loaded = 0; |
c8770e7b AK |
2022 | #ifdef CONFIG_X86_64 |
2023 | if (is_long_mode(&vmx->vcpu)) | |
2024 | rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base); | |
2025 | #endif | |
152d3f2f | 2026 | if (vmx->host_state.gs_ldt_reload_needed) { |
d6e88aec | 2027 | kvm_load_ldt(vmx->host_state.ldt_sel); |
33ed6329 | 2028 | #ifdef CONFIG_X86_64 |
9581d442 | 2029 | load_gs_index(vmx->host_state.gs_sel); |
9581d442 AK |
2030 | #else |
2031 | loadsegment(gs, vmx->host_state.gs_sel); | |
33ed6329 | 2032 | #endif |
33ed6329 | 2033 | } |
0a77fe4c AK |
2034 | if (vmx->host_state.fs_reload_needed) |
2035 | loadsegment(fs, vmx->host_state.fs_sel); | |
b2da15ac AK |
2036 | #ifdef CONFIG_X86_64 |
2037 | if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) { | |
2038 | loadsegment(ds, vmx->host_state.ds_sel); | |
2039 | loadsegment(es, vmx->host_state.es_sel); | |
2040 | } | |
b2da15ac | 2041 | #endif |
152d3f2f | 2042 | reload_tss(); |
44ea2b17 | 2043 | #ifdef CONFIG_X86_64 |
c8770e7b | 2044 | wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base); |
44ea2b17 | 2045 | #endif |
da8999d3 LJ |
2046 | if (vmx->host_state.msr_host_bndcfgs) |
2047 | wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs); | |
b1a74bf8 SS |
2048 | /* |
2049 | * If the FPU is not active (through the host task or | |
2050 | * the guest vcpu), then restore the cr0.TS bit. | |
2051 | */ | |
3c6dffa9 | 2052 | if (!fpregs_active() && !vmx->vcpu.guest_fpu_loaded) |
b1a74bf8 | 2053 | stts(); |
89cbc767 | 2054 | load_gdt(this_cpu_ptr(&host_gdt)); |
33ed6329 AK |
2055 | } |
2056 | ||
a9b21b62 AK |
2057 | static void vmx_load_host_state(struct vcpu_vmx *vmx) |
2058 | { | |
2059 | preempt_disable(); | |
2060 | __vmx_load_host_state(vmx); | |
2061 | preempt_enable(); | |
2062 | } | |
2063 | ||
28b835d6 FW |
2064 | static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu) |
2065 | { | |
2066 | struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu); | |
2067 | struct pi_desc old, new; | |
2068 | unsigned int dest; | |
2069 | ||
2070 | if (!kvm_arch_has_assigned_device(vcpu->kvm) || | |
2071 | !irq_remapping_cap(IRQ_POSTING_CAP)) | |
2072 | return; | |
2073 | ||
2074 | do { | |
2075 | old.control = new.control = pi_desc->control; | |
2076 | ||
2077 | /* | |
2078 | * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there | |
2079 | * are two possible cases: | |
2080 | * 1. After running 'pre_block', context switch | |
2081 | * happened. For this case, 'sn' was set in | |
2082 | * vmx_vcpu_put(), so we need to clear it here. | |
2083 | * 2. After running 'pre_block', we were blocked, | |
2084 | * and woken up by some other guy. For this case, | |
2085 | * we don't need to do anything, 'pi_post_block' | |
2086 | * will do everything for us. However, we cannot | |
2087 | * check whether it is case #1 or case #2 here | |
2088 | * (maybe, not needed), so we also clear sn here, | |
2089 | * I think it is not a big deal. | |
2090 | */ | |
2091 | if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) { | |
2092 | if (vcpu->cpu != cpu) { | |
2093 | dest = cpu_physical_id(cpu); | |
2094 | ||
2095 | if (x2apic_enabled()) | |
2096 | new.ndst = dest; | |
2097 | else | |
2098 | new.ndst = (dest << 8) & 0xFF00; | |
2099 | } | |
2100 | ||
2101 | /* set 'NV' to 'notification vector' */ | |
2102 | new.nv = POSTED_INTR_VECTOR; | |
2103 | } | |
2104 | ||
2105 | /* Allow posting non-urgent interrupts */ | |
2106 | new.sn = 0; | |
2107 | } while (cmpxchg(&pi_desc->control, old.control, | |
2108 | new.control) != old.control); | |
2109 | } | |
6aa8b732 AK |
2110 | /* |
2111 | * Switches to specified vcpu, until a matching vcpu_put(), but assumes | |
2112 | * vcpu mutex is already taken. | |
2113 | */ | |
15ad7146 | 2114 | static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
6aa8b732 | 2115 | { |
a2fa3e9f | 2116 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
4610c9cc | 2117 | u64 phys_addr = __pa(per_cpu(vmxarea, cpu)); |
6aa8b732 | 2118 | |
4610c9cc DX |
2119 | if (!vmm_exclusive) |
2120 | kvm_cpu_vmxon(phys_addr); | |
d462b819 NHE |
2121 | else if (vmx->loaded_vmcs->cpu != cpu) |
2122 | loaded_vmcs_clear(vmx->loaded_vmcs); | |
6aa8b732 | 2123 | |
d462b819 NHE |
2124 | if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) { |
2125 | per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs; | |
2126 | vmcs_load(vmx->loaded_vmcs->vmcs); | |
6aa8b732 AK |
2127 | } |
2128 | ||
d462b819 | 2129 | if (vmx->loaded_vmcs->cpu != cpu) { |
89cbc767 | 2130 | struct desc_ptr *gdt = this_cpu_ptr(&host_gdt); |
6aa8b732 AK |
2131 | unsigned long sysenter_esp; |
2132 | ||
a8eeb04a | 2133 | kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
92fe13be | 2134 | local_irq_disable(); |
8f536b76 | 2135 | crash_disable_local_vmclear(cpu); |
5a560f8b XG |
2136 | |
2137 | /* | |
2138 | * Read loaded_vmcs->cpu should be before fetching | |
2139 | * loaded_vmcs->loaded_vmcss_on_cpu_link. | |
2140 | * See the comments in __loaded_vmcs_clear(). | |
2141 | */ | |
2142 | smp_rmb(); | |
2143 | ||
d462b819 NHE |
2144 | list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link, |
2145 | &per_cpu(loaded_vmcss_on_cpu, cpu)); | |
8f536b76 | 2146 | crash_enable_local_vmclear(cpu); |
92fe13be DX |
2147 | local_irq_enable(); |
2148 | ||
6aa8b732 AK |
2149 | /* |
2150 | * Linux uses per-cpu TSS and GDT, so set these when switching | |
2151 | * processors. | |
2152 | */ | |
d6e88aec | 2153 | vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */ |
d359192f | 2154 | vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */ |
6aa8b732 AK |
2155 | |
2156 | rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp); | |
2157 | vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */ | |
ff2c3a18 | 2158 | |
d462b819 | 2159 | vmx->loaded_vmcs->cpu = cpu; |
6aa8b732 | 2160 | } |
28b835d6 | 2161 | |
2680d6da OH |
2162 | /* Setup TSC multiplier */ |
2163 | if (kvm_has_tsc_control && | |
2164 | vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio) { | |
2165 | vmx->current_tsc_ratio = vcpu->arch.tsc_scaling_ratio; | |
2166 | vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio); | |
2167 | } | |
2168 | ||
28b835d6 FW |
2169 | vmx_vcpu_pi_load(vcpu, cpu); |
2170 | } | |
2171 | ||
2172 | static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu) | |
2173 | { | |
2174 | struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu); | |
2175 | ||
2176 | if (!kvm_arch_has_assigned_device(vcpu->kvm) || | |
2177 | !irq_remapping_cap(IRQ_POSTING_CAP)) | |
2178 | return; | |
2179 | ||
2180 | /* Set SN when the vCPU is preempted */ | |
2181 | if (vcpu->preempted) | |
2182 | pi_set_sn(pi_desc); | |
6aa8b732 AK |
2183 | } |
2184 | ||
2185 | static void vmx_vcpu_put(struct kvm_vcpu *vcpu) | |
2186 | { | |
28b835d6 FW |
2187 | vmx_vcpu_pi_put(vcpu); |
2188 | ||
a9b21b62 | 2189 | __vmx_load_host_state(to_vmx(vcpu)); |
4610c9cc | 2190 | if (!vmm_exclusive) { |
d462b819 NHE |
2191 | __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs); |
2192 | vcpu->cpu = -1; | |
4610c9cc DX |
2193 | kvm_cpu_vmxoff(); |
2194 | } | |
6aa8b732 AK |
2195 | } |
2196 | ||
5fd86fcf AK |
2197 | static void vmx_fpu_activate(struct kvm_vcpu *vcpu) |
2198 | { | |
81231c69 AK |
2199 | ulong cr0; |
2200 | ||
5fd86fcf AK |
2201 | if (vcpu->fpu_active) |
2202 | return; | |
2203 | vcpu->fpu_active = 1; | |
81231c69 AK |
2204 | cr0 = vmcs_readl(GUEST_CR0); |
2205 | cr0 &= ~(X86_CR0_TS | X86_CR0_MP); | |
2206 | cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP); | |
2207 | vmcs_writel(GUEST_CR0, cr0); | |
5fd86fcf | 2208 | update_exception_bitmap(vcpu); |
edcafe3c | 2209 | vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS; |
36cf24e0 NHE |
2210 | if (is_guest_mode(vcpu)) |
2211 | vcpu->arch.cr0_guest_owned_bits &= | |
2212 | ~get_vmcs12(vcpu)->cr0_guest_host_mask; | |
edcafe3c | 2213 | vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits); |
5fd86fcf AK |
2214 | } |
2215 | ||
edcafe3c AK |
2216 | static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu); |
2217 | ||
fe3ef05c NHE |
2218 | /* |
2219 | * Return the cr0 value that a nested guest would read. This is a combination | |
2220 | * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by | |
2221 | * its hypervisor (cr0_read_shadow). | |
2222 | */ | |
2223 | static inline unsigned long nested_read_cr0(struct vmcs12 *fields) | |
2224 | { | |
2225 | return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) | | |
2226 | (fields->cr0_read_shadow & fields->cr0_guest_host_mask); | |
2227 | } | |
2228 | static inline unsigned long nested_read_cr4(struct vmcs12 *fields) | |
2229 | { | |
2230 | return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) | | |
2231 | (fields->cr4_read_shadow & fields->cr4_guest_host_mask); | |
2232 | } | |
2233 | ||
5fd86fcf AK |
2234 | static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu) |
2235 | { | |
36cf24e0 NHE |
2236 | /* Note that there is no vcpu->fpu_active = 0 here. The caller must |
2237 | * set this *before* calling this function. | |
2238 | */ | |
edcafe3c | 2239 | vmx_decache_cr0_guest_bits(vcpu); |
81231c69 | 2240 | vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP); |
5fd86fcf | 2241 | update_exception_bitmap(vcpu); |
edcafe3c AK |
2242 | vcpu->arch.cr0_guest_owned_bits = 0; |
2243 | vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits); | |
36cf24e0 NHE |
2244 | if (is_guest_mode(vcpu)) { |
2245 | /* | |
2246 | * L1's specified read shadow might not contain the TS bit, | |
2247 | * so now that we turned on shadowing of this bit, we need to | |
2248 | * set this bit of the shadow. Like in nested_vmx_run we need | |
2249 | * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet | |
2250 | * up-to-date here because we just decached cr0.TS (and we'll | |
2251 | * only update vmcs12->guest_cr0 on nested exit). | |
2252 | */ | |
2253 | struct vmcs12 *vmcs12 = get_vmcs12(vcpu); | |
2254 | vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) | | |
2255 | (vcpu->arch.cr0 & X86_CR0_TS); | |
2256 | vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12)); | |
2257 | } else | |
2258 | vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0); | |
5fd86fcf AK |
2259 | } |
2260 | ||
6aa8b732 AK |
2261 | static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu) |
2262 | { | |
78ac8b47 | 2263 | unsigned long rflags, save_rflags; |
345dcaa8 | 2264 | |
6de12732 AK |
2265 | if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) { |
2266 | __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail); | |
2267 | rflags = vmcs_readl(GUEST_RFLAGS); | |
2268 | if (to_vmx(vcpu)->rmode.vm86_active) { | |
2269 | rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS; | |
2270 | save_rflags = to_vmx(vcpu)->rmode.save_rflags; | |
2271 | rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS; | |
2272 | } | |
2273 | to_vmx(vcpu)->rflags = rflags; | |
78ac8b47 | 2274 | } |
6de12732 | 2275 | return to_vmx(vcpu)->rflags; |
6aa8b732 AK |
2276 | } |
2277 | ||
2278 | static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
2279 | { | |
6de12732 AK |
2280 | __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail); |
2281 | to_vmx(vcpu)->rflags = rflags; | |
78ac8b47 AK |
2282 | if (to_vmx(vcpu)->rmode.vm86_active) { |
2283 | to_vmx(vcpu)->rmode.save_rflags = rflags; | |
053de044 | 2284 | rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM; |
78ac8b47 | 2285 | } |
6aa8b732 AK |
2286 | vmcs_writel(GUEST_RFLAGS, rflags); |
2287 | } | |
2288 | ||
37ccdcbe | 2289 | static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu) |
2809f5d2 GC |
2290 | { |
2291 | u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO); | |
2292 | int ret = 0; | |
2293 | ||
2294 | if (interruptibility & GUEST_INTR_STATE_STI) | |
48005f64 | 2295 | ret |= KVM_X86_SHADOW_INT_STI; |
2809f5d2 | 2296 | if (interruptibility & GUEST_INTR_STATE_MOV_SS) |
48005f64 | 2297 | ret |= KVM_X86_SHADOW_INT_MOV_SS; |
2809f5d2 | 2298 | |
37ccdcbe | 2299 | return ret; |
2809f5d2 GC |
2300 | } |
2301 | ||
2302 | static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask) | |
2303 | { | |
2304 | u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO); | |
2305 | u32 interruptibility = interruptibility_old; | |
2306 | ||
2307 | interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS); | |
2308 | ||
48005f64 | 2309 | if (mask & KVM_X86_SHADOW_INT_MOV_SS) |
2809f5d2 | 2310 | interruptibility |= GUEST_INTR_STATE_MOV_SS; |
48005f64 | 2311 | else if (mask & KVM_X86_SHADOW_INT_STI) |
2809f5d2 GC |
2312 | interruptibility |= GUEST_INTR_STATE_STI; |
2313 | ||
2314 | if ((interruptibility != interruptibility_old)) | |
2315 | vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility); | |
2316 | } | |
2317 | ||
6aa8b732 AK |
2318 | static void skip_emulated_instruction(struct kvm_vcpu *vcpu) |
2319 | { | |
2320 | unsigned long rip; | |
6aa8b732 | 2321 | |
5fdbf976 | 2322 | rip = kvm_rip_read(vcpu); |
6aa8b732 | 2323 | rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN); |
5fdbf976 | 2324 | kvm_rip_write(vcpu, rip); |
6aa8b732 | 2325 | |
2809f5d2 GC |
2326 | /* skipping an emulated instruction also counts */ |
2327 | vmx_set_interrupt_shadow(vcpu, 0); | |
6aa8b732 AK |
2328 | } |
2329 | ||
0b6ac343 NHE |
2330 | /* |
2331 | * KVM wants to inject page-faults which it got to the guest. This function | |
2332 | * checks whether in a nested guest, we need to inject them to L1 or L2. | |
0b6ac343 | 2333 | */ |
e011c663 | 2334 | static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr) |
0b6ac343 NHE |
2335 | { |
2336 | struct vmcs12 *vmcs12 = get_vmcs12(vcpu); | |
2337 | ||
e011c663 | 2338 | if (!(vmcs12->exception_bitmap & (1u << nr))) |
0b6ac343 NHE |
2339 | return 0; |
2340 | ||
533558bc JK |
2341 | nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason, |
2342 | vmcs_read32(VM_EXIT_INTR_INFO), | |
2343 | vmcs_readl(EXIT_QUALIFICATION)); | |
0b6ac343 NHE |
2344 | return 1; |
2345 | } | |
2346 | ||
298101da | 2347 | static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr, |
ce7ddec4 JR |
2348 | bool has_error_code, u32 error_code, |
2349 | bool reinject) | |
298101da | 2350 | { |
77ab6db0 | 2351 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
8ab2d2e2 | 2352 | u32 intr_info = nr | INTR_INFO_VALID_MASK; |
77ab6db0 | 2353 | |
e011c663 GN |
2354 | if (!reinject && is_guest_mode(vcpu) && |
2355 | nested_vmx_check_exception(vcpu, nr)) | |
0b6ac343 NHE |
2356 | return; |
2357 | ||
8ab2d2e2 | 2358 | if (has_error_code) { |
77ab6db0 | 2359 | vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code); |
8ab2d2e2 JK |
2360 | intr_info |= INTR_INFO_DELIVER_CODE_MASK; |
2361 | } | |
77ab6db0 | 2362 | |
7ffd92c5 | 2363 | if (vmx->rmode.vm86_active) { |
71f9833b SH |
2364 | int inc_eip = 0; |
2365 | if (kvm_exception_is_soft(nr)) | |
2366 | inc_eip = vcpu->arch.event_exit_inst_len; | |
2367 | if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE) | |
a92601bb | 2368 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
77ab6db0 JK |
2369 | return; |
2370 | } | |
2371 | ||
66fd3f7f GN |
2372 | if (kvm_exception_is_soft(nr)) { |
2373 | vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, | |
2374 | vmx->vcpu.arch.event_exit_inst_len); | |
8ab2d2e2 JK |
2375 | intr_info |= INTR_TYPE_SOFT_EXCEPTION; |
2376 | } else | |
2377 | intr_info |= INTR_TYPE_HARD_EXCEPTION; | |
2378 | ||
2379 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info); | |
298101da AK |
2380 | } |
2381 | ||
4e47c7a6 SY |
2382 | static bool vmx_rdtscp_supported(void) |
2383 | { | |
2384 | return cpu_has_vmx_rdtscp(); | |
2385 | } | |
2386 | ||
ad756a16 MJ |
2387 | static bool vmx_invpcid_supported(void) |
2388 | { | |
2389 | return cpu_has_vmx_invpcid() && enable_ept; | |
2390 | } | |
2391 | ||
a75beee6 ED |
2392 | /* |
2393 | * Swap MSR entry in host/guest MSR entry array. | |
2394 | */ | |
8b9cf98c | 2395 | static void move_msr_up(struct vcpu_vmx *vmx, int from, int to) |
a75beee6 | 2396 | { |
26bb0981 | 2397 | struct shared_msr_entry tmp; |
a2fa3e9f GH |
2398 | |
2399 | tmp = vmx->guest_msrs[to]; | |
2400 | vmx->guest_msrs[to] = vmx->guest_msrs[from]; | |
2401 | vmx->guest_msrs[from] = tmp; | |
a75beee6 ED |
2402 | } |
2403 | ||
8d14695f YZ |
2404 | static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu) |
2405 | { | |
2406 | unsigned long *msr_bitmap; | |
2407 | ||
670125bd WV |
2408 | if (is_guest_mode(vcpu)) |
2409 | msr_bitmap = vmx_msr_bitmap_nested; | |
8a9781f7 | 2410 | else if (vcpu->arch.apic_base & X2APIC_ENABLE) { |
8d14695f YZ |
2411 | if (is_long_mode(vcpu)) |
2412 | msr_bitmap = vmx_msr_bitmap_longmode_x2apic; | |
2413 | else | |
2414 | msr_bitmap = vmx_msr_bitmap_legacy_x2apic; | |
2415 | } else { | |
2416 | if (is_long_mode(vcpu)) | |
2417 | msr_bitmap = vmx_msr_bitmap_longmode; | |
2418 | else | |
2419 | msr_bitmap = vmx_msr_bitmap_legacy; | |
2420 | } | |
2421 | ||
2422 | vmcs_write64(MSR_BITMAP, __pa(msr_bitmap)); | |
2423 | } | |
2424 | ||
e38aea3e AK |
2425 | /* |
2426 | * Set up the vmcs to automatically save and restore system | |
2427 | * msrs. Don't touch the 64-bit msrs if the guest is in legacy | |
2428 | * mode, as fiddling with msrs is very expensive. | |
2429 | */ | |
8b9cf98c | 2430 | static void setup_msrs(struct vcpu_vmx *vmx) |
e38aea3e | 2431 | { |
26bb0981 | 2432 | int save_nmsrs, index; |
e38aea3e | 2433 | |
a75beee6 ED |
2434 | save_nmsrs = 0; |
2435 | #ifdef CONFIG_X86_64 | |
8b9cf98c | 2436 | if (is_long_mode(&vmx->vcpu)) { |
8b9cf98c | 2437 | index = __find_msr_index(vmx, MSR_SYSCALL_MASK); |
a75beee6 | 2438 | if (index >= 0) |
8b9cf98c RR |
2439 | move_msr_up(vmx, index, save_nmsrs++); |
2440 | index = __find_msr_index(vmx, MSR_LSTAR); | |
a75beee6 | 2441 | if (index >= 0) |
8b9cf98c RR |
2442 | move_msr_up(vmx, index, save_nmsrs++); |
2443 | index = __find_msr_index(vmx, MSR_CSTAR); | |
a75beee6 | 2444 | if (index >= 0) |
8b9cf98c | 2445 | move_msr_up(vmx, index, save_nmsrs++); |
4e47c7a6 | 2446 | index = __find_msr_index(vmx, MSR_TSC_AUX); |
1cea0ce6 | 2447 | if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu)) |
4e47c7a6 | 2448 | move_msr_up(vmx, index, save_nmsrs++); |
a75beee6 | 2449 | /* |
8c06585d | 2450 | * MSR_STAR is only needed on long mode guests, and only |
a75beee6 ED |
2451 | * if efer.sce is enabled. |
2452 | */ | |
8c06585d | 2453 | index = __find_msr_index(vmx, MSR_STAR); |
f6801dff | 2454 | if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE)) |
8b9cf98c | 2455 | move_msr_up(vmx, index, save_nmsrs++); |
a75beee6 ED |
2456 | } |
2457 | #endif | |
92c0d900 AK |
2458 | index = __find_msr_index(vmx, MSR_EFER); |
2459 | if (index >= 0 && update_transition_efer(vmx, index)) | |
26bb0981 | 2460 | move_msr_up(vmx, index, save_nmsrs++); |
e38aea3e | 2461 | |
26bb0981 | 2462 | vmx->save_nmsrs = save_nmsrs; |
5897297b | 2463 | |
8d14695f YZ |
2464 | if (cpu_has_vmx_msr_bitmap()) |
2465 | vmx_set_msr_bitmap(&vmx->vcpu); | |
e38aea3e AK |
2466 | } |
2467 | ||
6aa8b732 AK |
2468 | /* |
2469 | * reads and returns guest's timestamp counter "register" | |
be7b263e HZ |
2470 | * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset |
2471 | * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3 | |
6aa8b732 | 2472 | */ |
be7b263e | 2473 | static u64 guest_read_tsc(struct kvm_vcpu *vcpu) |
6aa8b732 AK |
2474 | { |
2475 | u64 host_tsc, tsc_offset; | |
2476 | ||
4ea1636b | 2477 | host_tsc = rdtsc(); |
6aa8b732 | 2478 | tsc_offset = vmcs_read64(TSC_OFFSET); |
be7b263e | 2479 | return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset; |
6aa8b732 AK |
2480 | } |
2481 | ||
d5c1785d NHE |
2482 | /* |
2483 | * Like guest_read_tsc, but always returns L1's notion of the timestamp | |
2484 | * counter, even if a nested guest (L2) is currently running. | |
2485 | */ | |
48d89b92 | 2486 | static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc) |
d5c1785d | 2487 | { |
886b470c | 2488 | u64 tsc_offset; |
d5c1785d | 2489 | |
d5c1785d NHE |
2490 | tsc_offset = is_guest_mode(vcpu) ? |
2491 | to_vmx(vcpu)->nested.vmcs01_tsc_offset : | |
2492 | vmcs_read64(TSC_OFFSET); | |
2493 | return host_tsc + tsc_offset; | |
2494 | } | |
2495 | ||
ba904635 WA |
2496 | static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu) |
2497 | { | |
2498 | return vmcs_read64(TSC_OFFSET); | |
2499 | } | |
2500 | ||
6aa8b732 | 2501 | /* |
99e3e30a | 2502 | * writes 'offset' into guest's timestamp counter offset register |
6aa8b732 | 2503 | */ |
99e3e30a | 2504 | static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset) |
6aa8b732 | 2505 | { |
27fc51b2 | 2506 | if (is_guest_mode(vcpu)) { |
7991825b | 2507 | /* |
27fc51b2 NHE |
2508 | * We're here if L1 chose not to trap WRMSR to TSC. According |
2509 | * to the spec, this should set L1's TSC; The offset that L1 | |
2510 | * set for L2 remains unchanged, and still needs to be added | |
2511 | * to the newly set TSC to get L2's TSC. | |
7991825b | 2512 | */ |
27fc51b2 NHE |
2513 | struct vmcs12 *vmcs12; |
2514 | to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset; | |
2515 | /* recalculate vmcs02.TSC_OFFSET: */ | |
2516 | vmcs12 = get_vmcs12(vcpu); | |
2517 | vmcs_write64(TSC_OFFSET, offset + | |
2518 | (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ? | |
2519 | vmcs12->tsc_offset : 0)); | |
2520 | } else { | |
489223ed YY |
2521 | trace_kvm_write_tsc_offset(vcpu->vcpu_id, |
2522 | vmcs_read64(TSC_OFFSET), offset); | |
27fc51b2 NHE |
2523 | vmcs_write64(TSC_OFFSET, offset); |
2524 | } | |
6aa8b732 AK |
2525 | } |
2526 | ||
58ea6767 | 2527 | static void vmx_adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, s64 adjustment) |
e48672fa ZA |
2528 | { |
2529 | u64 offset = vmcs_read64(TSC_OFFSET); | |
489223ed | 2530 | |
e48672fa | 2531 | vmcs_write64(TSC_OFFSET, offset + adjustment); |
7991825b NHE |
2532 | if (is_guest_mode(vcpu)) { |
2533 | /* Even when running L2, the adjustment needs to apply to L1 */ | |
2534 | to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment; | |
489223ed YY |
2535 | } else |
2536 | trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset, | |
2537 | offset + adjustment); | |
e48672fa ZA |
2538 | } |
2539 | ||
801d3424 NHE |
2540 | static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu) |
2541 | { | |
2542 | struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0); | |
2543 | return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31))); | |
2544 | } | |
2545 | ||
2546 | /* | |
2547 | * nested_vmx_allowed() checks whether a guest should be allowed to use VMX | |
2548 | * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for | |
2549 | * all guests if the "nested" module option is off, and can also be disabled | |
2550 | * for a single guest by disabling its VMX cpuid bit. | |
2551 | */ | |
2552 | static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu) | |
2553 | { | |
2554 | return nested && guest_cpuid_has_vmx(vcpu); | |
2555 | } | |
2556 | ||
b87a51ae NHE |
2557 | /* |
2558 | * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be | |
2559 | * returned for the various VMX controls MSRs when nested VMX is enabled. | |
2560 | * The same values should also be used to verify that vmcs12 control fields are | |
2561 | * valid during nested entry from L1 to L2. | |
2562 | * Each of these control msrs has a low and high 32-bit half: A low bit is on | |
2563 | * if the corresponding bit in the (32-bit) control field *must* be on, and a | |
2564 | * bit in the high half is on if the corresponding bit in the control field | |
2565 | * may be on. See also vmx_control_verify(). | |
b87a51ae | 2566 | */ |
b9c237bb | 2567 | static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx) |
b87a51ae NHE |
2568 | { |
2569 | /* | |
2570 | * Note that as a general rule, the high half of the MSRs (bits in | |
2571 | * the control fields which may be 1) should be initialized by the | |
2572 | * intersection of the underlying hardware's MSR (i.e., features which | |
2573 | * can be supported) and the list of features we want to expose - | |
2574 | * because they are known to be properly supported in our code. | |
2575 | * Also, usually, the low half of the MSRs (bits which must be 1) can | |
2576 | * be set to 0, meaning that L1 may turn off any of these bits. The | |
2577 | * reason is that if one of these bits is necessary, it will appear | |
2578 | * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control | |
2579 | * fields of vmcs01 and vmcs02, will turn these bits off - and | |
2580 | * nested_vmx_exit_handled() will not pass related exits to L1. | |
2581 | * These rules have exceptions below. | |
2582 | */ | |
2583 | ||
2584 | /* pin-based controls */ | |
eabeaacc | 2585 | rdmsr(MSR_IA32_VMX_PINBASED_CTLS, |
b9c237bb WV |
2586 | vmx->nested.nested_vmx_pinbased_ctls_low, |
2587 | vmx->nested.nested_vmx_pinbased_ctls_high); | |
2588 | vmx->nested.nested_vmx_pinbased_ctls_low |= | |
2589 | PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR; | |
2590 | vmx->nested.nested_vmx_pinbased_ctls_high &= | |
2591 | PIN_BASED_EXT_INTR_MASK | | |
2592 | PIN_BASED_NMI_EXITING | | |
2593 | PIN_BASED_VIRTUAL_NMIS; | |
2594 | vmx->nested.nested_vmx_pinbased_ctls_high |= | |
2595 | PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR | | |
0238ea91 | 2596 | PIN_BASED_VMX_PREEMPTION_TIMER; |
d62caabb | 2597 | if (kvm_vcpu_apicv_active(&vmx->vcpu)) |
705699a1 WV |
2598 | vmx->nested.nested_vmx_pinbased_ctls_high |= |
2599 | PIN_BASED_POSTED_INTR; | |
b87a51ae | 2600 | |
3dbcd8da | 2601 | /* exit controls */ |
c0dfee58 | 2602 | rdmsr(MSR_IA32_VMX_EXIT_CTLS, |
b9c237bb WV |
2603 | vmx->nested.nested_vmx_exit_ctls_low, |
2604 | vmx->nested.nested_vmx_exit_ctls_high); | |
2605 | vmx->nested.nested_vmx_exit_ctls_low = | |
2606 | VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR; | |
e0ba1a6f | 2607 | |
b9c237bb | 2608 | vmx->nested.nested_vmx_exit_ctls_high &= |
b87a51ae | 2609 | #ifdef CONFIG_X86_64 |
c0dfee58 | 2610 | VM_EXIT_HOST_ADDR_SPACE_SIZE | |
b87a51ae | 2611 | #endif |
f4124500 | 2612 | VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT; |
b9c237bb WV |
2613 | vmx->nested.nested_vmx_exit_ctls_high |= |
2614 | VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR | | |
f4124500 | 2615 | VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER | |
e0ba1a6f BD |
2616 | VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT; |
2617 | ||
a87036ad | 2618 | if (kvm_mpx_supported()) |
b9c237bb | 2619 | vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS; |
b87a51ae | 2620 | |
2996fca0 | 2621 | /* We support free control of debug control saving. */ |
b9c237bb WV |
2622 | vmx->nested.nested_vmx_true_exit_ctls_low = |
2623 | vmx->nested.nested_vmx_exit_ctls_low & | |
2996fca0 JK |
2624 | ~VM_EXIT_SAVE_DEBUG_CONTROLS; |
2625 | ||
b87a51ae NHE |
2626 | /* entry controls */ |
2627 | rdmsr(MSR_IA32_VMX_ENTRY_CTLS, | |
b9c237bb WV |
2628 | vmx->nested.nested_vmx_entry_ctls_low, |
2629 | vmx->nested.nested_vmx_entry_ctls_high); | |
2630 | vmx->nested.nested_vmx_entry_ctls_low = | |
2631 | VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR; | |
2632 | vmx->nested.nested_vmx_entry_ctls_high &= | |
57435349 JK |
2633 | #ifdef CONFIG_X86_64 |
2634 | VM_ENTRY_IA32E_MODE | | |
2635 | #endif | |
2636 | VM_ENTRY_LOAD_IA32_PAT; | |
b9c237bb WV |
2637 | vmx->nested.nested_vmx_entry_ctls_high |= |
2638 | (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER); | |
a87036ad | 2639 | if (kvm_mpx_supported()) |
b9c237bb | 2640 | vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS; |
57435349 | 2641 | |
2996fca0 | 2642 | /* We support free control of debug control loading. */ |
b9c237bb WV |
2643 | vmx->nested.nested_vmx_true_entry_ctls_low = |
2644 | vmx->nested.nested_vmx_entry_ctls_low & | |
2996fca0 JK |
2645 | ~VM_ENTRY_LOAD_DEBUG_CONTROLS; |
2646 | ||
b87a51ae NHE |
2647 | /* cpu-based controls */ |
2648 | rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, | |
b9c237bb WV |
2649 | vmx->nested.nested_vmx_procbased_ctls_low, |
2650 | vmx->nested.nested_vmx_procbased_ctls_high); | |
2651 | vmx->nested.nested_vmx_procbased_ctls_low = | |
2652 | CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR; | |
2653 | vmx->nested.nested_vmx_procbased_ctls_high &= | |
a294c9bb JK |
2654 | CPU_BASED_VIRTUAL_INTR_PENDING | |
2655 | CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING | | |
b87a51ae NHE |
2656 | CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING | |
2657 | CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING | | |
2658 | CPU_BASED_CR3_STORE_EXITING | | |
2659 | #ifdef CONFIG_X86_64 | |
2660 | CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING | | |
2661 | #endif | |
2662 | CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING | | |
5f3d45e7 MD |
2663 | CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG | |
2664 | CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING | | |
2665 | CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING | | |
2666 | CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS; | |
b87a51ae NHE |
2667 | /* |
2668 | * We can allow some features even when not supported by the | |
2669 | * hardware. For example, L1 can specify an MSR bitmap - and we | |
2670 | * can use it to avoid exits to L1 - even when L0 runs L2 | |
2671 | * without MSR bitmaps. | |
2672 | */ | |
b9c237bb WV |
2673 | vmx->nested.nested_vmx_procbased_ctls_high |= |
2674 | CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR | | |
560b7ee1 | 2675 | CPU_BASED_USE_MSR_BITMAPS; |
b87a51ae | 2676 | |
3dcdf3ec | 2677 | /* We support free control of CR3 access interception. */ |
b9c237bb WV |
2678 | vmx->nested.nested_vmx_true_procbased_ctls_low = |
2679 | vmx->nested.nested_vmx_procbased_ctls_low & | |
3dcdf3ec JK |
2680 | ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING); |
2681 | ||
b87a51ae NHE |
2682 | /* secondary cpu-based controls */ |
2683 | rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2, | |
b9c237bb WV |
2684 | vmx->nested.nested_vmx_secondary_ctls_low, |
2685 | vmx->nested.nested_vmx_secondary_ctls_high); | |
2686 | vmx->nested.nested_vmx_secondary_ctls_low = 0; | |
2687 | vmx->nested.nested_vmx_secondary_ctls_high &= | |
d6851fbe | 2688 | SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES | |
b3a2a907 | 2689 | SECONDARY_EXEC_RDTSCP | |
f2b93280 | 2690 | SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE | |
5c614b35 | 2691 | SECONDARY_EXEC_ENABLE_VPID | |
82f0dd4b | 2692 | SECONDARY_EXEC_APIC_REGISTER_VIRT | |
608406e2 | 2693 | SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY | |
81dc01f7 | 2694 | SECONDARY_EXEC_WBINVD_EXITING | |
8b3e34e4 XG |
2695 | SECONDARY_EXEC_XSAVES | |
2696 | SECONDARY_EXEC_PCOMMIT; | |
c18911a2 | 2697 | |
afa61f75 NHE |
2698 | if (enable_ept) { |
2699 | /* nested EPT: emulate EPT also to L1 */ | |
b9c237bb | 2700 | vmx->nested.nested_vmx_secondary_ctls_high |= |
0790ec17 | 2701 | SECONDARY_EXEC_ENABLE_EPT; |
b9c237bb | 2702 | vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT | |
d3134dbf JK |
2703 | VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT | |
2704 | VMX_EPT_INVEPT_BIT; | |
b9c237bb | 2705 | vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept; |
afa61f75 | 2706 | /* |
4b855078 BD |
2707 | * For nested guests, we don't do anything specific |
2708 | * for single context invalidation. Hence, only advertise | |
2709 | * support for global context invalidation. | |
afa61f75 | 2710 | */ |
b9c237bb | 2711 | vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT; |
afa61f75 | 2712 | } else |
b9c237bb | 2713 | vmx->nested.nested_vmx_ept_caps = 0; |
afa61f75 | 2714 | |
089d7b6e WL |
2715 | if (enable_vpid) |
2716 | vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT | | |
2717 | VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT; | |
2718 | else | |
2719 | vmx->nested.nested_vmx_vpid_caps = 0; | |
99b83ac8 | 2720 | |
0790ec17 RK |
2721 | if (enable_unrestricted_guest) |
2722 | vmx->nested.nested_vmx_secondary_ctls_high |= | |
2723 | SECONDARY_EXEC_UNRESTRICTED_GUEST; | |
2724 | ||
c18911a2 | 2725 | /* miscellaneous data */ |
b9c237bb WV |
2726 | rdmsr(MSR_IA32_VMX_MISC, |
2727 | vmx->nested.nested_vmx_misc_low, | |
2728 | vmx->nested.nested_vmx_misc_high); | |
2729 | vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA; | |
2730 | vmx->nested.nested_vmx_misc_low |= | |
2731 | VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE | | |
f4124500 | 2732 | VMX_MISC_ACTIVITY_HLT; |
b9c237bb | 2733 | vmx->nested.nested_vmx_misc_high = 0; |
b87a51ae NHE |
2734 | } |
2735 | ||
2736 | static inline bool vmx_control_verify(u32 control, u32 low, u32 high) | |
2737 | { | |
2738 | /* | |
2739 | * Bits 0 in high must be 0, and bits 1 in low must be 1. | |
2740 | */ | |
2741 | return ((control & high) | low) == control; | |
2742 | } | |
2743 | ||
2744 | static inline u64 vmx_control_msr(u32 low, u32 high) | |
2745 | { | |
2746 | return low | ((u64)high << 32); | |
2747 | } | |
2748 | ||
cae50139 | 2749 | /* Returns 0 on success, non-0 otherwise. */ |
b87a51ae NHE |
2750 | static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata) |
2751 | { | |
b9c237bb WV |
2752 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
2753 | ||
b87a51ae | 2754 | switch (msr_index) { |
b87a51ae NHE |
2755 | case MSR_IA32_VMX_BASIC: |
2756 | /* | |
2757 | * This MSR reports some information about VMX support. We | |
2758 | * should return information about the VMX we emulate for the | |
2759 | * guest, and the VMCS structure we give it - not about the | |
2760 | * VMX support of the underlying hardware. | |
2761 | */ | |
3dbcd8da | 2762 | *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS | |
b87a51ae NHE |
2763 | ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) | |
2764 | (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT); | |
2765 | break; | |
2766 | case MSR_IA32_VMX_TRUE_PINBASED_CTLS: | |
2767 | case MSR_IA32_VMX_PINBASED_CTLS: | |
b9c237bb WV |
2768 | *pdata = vmx_control_msr( |
2769 | vmx->nested.nested_vmx_pinbased_ctls_low, | |
2770 | vmx->nested.nested_vmx_pinbased_ctls_high); | |
b87a51ae NHE |
2771 | break; |
2772 | case MSR_IA32_VMX_TRUE_PROCBASED_CTLS: | |
b9c237bb WV |
2773 | *pdata = vmx_control_msr( |
2774 | vmx->nested.nested_vmx_true_procbased_ctls_low, | |
2775 | vmx->nested.nested_vmx_procbased_ctls_high); | |
3dcdf3ec | 2776 | break; |
b87a51ae | 2777 | case MSR_IA32_VMX_PROCBASED_CTLS: |
b9c237bb WV |
2778 | *pdata = vmx_control_msr( |
2779 | vmx->nested.nested_vmx_procbased_ctls_low, | |
2780 | vmx->nested.nested_vmx_procbased_ctls_high); | |
b87a51ae NHE |
2781 | break; |
2782 | case MSR_IA32_VMX_TRUE_EXIT_CTLS: | |
b9c237bb WV |
2783 | *pdata = vmx_control_msr( |
2784 | vmx->nested.nested_vmx_true_exit_ctls_low, | |
2785 | vmx->nested.nested_vmx_exit_ctls_high); | |
2996fca0 | 2786 | break; |
b87a51ae | 2787 | case MSR_IA32_VMX_EXIT_CTLS: |
b9c237bb WV |
2788 | *pdata = vmx_control_msr( |
2789 | vmx->nested.nested_vmx_exit_ctls_low, | |
2790 | vmx->nested.nested_vmx_exit_ctls_high); | |
b87a51ae NHE |
2791 | break; |
2792 | case MSR_IA32_VMX_TRUE_ENTRY_CTLS: | |
b9c237bb WV |
2793 | *pdata = vmx_control_msr( |
2794 | vmx->nested.nested_vmx_true_entry_ctls_low, | |
2795 | vmx->nested.nested_vmx_entry_ctls_high); | |
2996fca0 | 2796 | break; |
b87a51ae | 2797 | case MSR_IA32_VMX_ENTRY_CTLS: |
b9c237bb WV |
2798 | *pdata = vmx_control_msr( |
2799 | vmx->nested.nested_vmx_entry_ctls_low, | |
2800 | vmx->nested.nested_vmx_entry_ctls_high); | |
b87a51ae NHE |
2801 | break; |
2802 | case MSR_IA32_VMX_MISC: | |
b9c237bb WV |
2803 | *pdata = vmx_control_msr( |
2804 | vmx->nested.nested_vmx_misc_low, | |
2805 | vmx->nested.nested_vmx_misc_high); | |
b87a51ae NHE |
2806 | break; |
2807 | /* | |
2808 | * These MSRs specify bits which the guest must keep fixed (on or off) | |
2809 | * while L1 is in VMXON mode (in L1's root mode, or running an L2). | |
2810 | * We picked the standard core2 setting. | |
2811 | */ | |
2812 | #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE) | |
2813 | #define VMXON_CR4_ALWAYSON X86_CR4_VMXE | |
2814 | case MSR_IA32_VMX_CR0_FIXED0: | |
2815 | *pdata = VMXON_CR0_ALWAYSON; | |
2816 | break; | |
2817 | case MSR_IA32_VMX_CR0_FIXED1: | |
2818 | *pdata = -1ULL; | |
2819 | break; | |
2820 | case MSR_IA32_VMX_CR4_FIXED0: | |
2821 | *pdata = VMXON_CR4_ALWAYSON; | |
2822 | break; | |
2823 | case MSR_IA32_VMX_CR4_FIXED1: | |
2824 | *pdata = -1ULL; | |
2825 | break; | |
2826 | case MSR_IA32_VMX_VMCS_ENUM: | |
5381417f | 2827 | *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */ |
b87a51ae NHE |
2828 | break; |
2829 | case MSR_IA32_VMX_PROCBASED_CTLS2: | |
b9c237bb WV |
2830 | *pdata = vmx_control_msr( |
2831 | vmx->nested.nested_vmx_secondary_ctls_low, | |
2832 | vmx->nested.nested_vmx_secondary_ctls_high); | |
b87a51ae NHE |
2833 | break; |
2834 | case MSR_IA32_VMX_EPT_VPID_CAP: | |
afa61f75 | 2835 | /* Currently, no nested vpid support */ |
089d7b6e WL |
2836 | *pdata = vmx->nested.nested_vmx_ept_caps | |
2837 | ((u64)vmx->nested.nested_vmx_vpid_caps << 32); | |
b87a51ae NHE |
2838 | break; |
2839 | default: | |
b87a51ae | 2840 | return 1; |
b3897a49 NHE |
2841 | } |
2842 | ||
b87a51ae NHE |
2843 | return 0; |
2844 | } | |
2845 | ||
6aa8b732 AK |
2846 | /* |
2847 | * Reads an msr value (of 'msr_index') into 'pdata'. | |
2848 | * Returns 0 on success, non-0 otherwise. | |
2849 | * Assumes vcpu_load() was already called. | |
2850 | */ | |
609e36d3 | 2851 | static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
6aa8b732 | 2852 | { |
26bb0981 | 2853 | struct shared_msr_entry *msr; |
6aa8b732 | 2854 | |
609e36d3 | 2855 | switch (msr_info->index) { |
05b3e0c2 | 2856 | #ifdef CONFIG_X86_64 |
6aa8b732 | 2857 | case MSR_FS_BASE: |
609e36d3 | 2858 | msr_info->data = vmcs_readl(GUEST_FS_BASE); |
6aa8b732 AK |
2859 | break; |
2860 | case MSR_GS_BASE: | |
609e36d3 | 2861 | msr_info->data = vmcs_readl(GUEST_GS_BASE); |
6aa8b732 | 2862 | break; |
44ea2b17 AK |
2863 | case MSR_KERNEL_GS_BASE: |
2864 | vmx_load_host_state(to_vmx(vcpu)); | |
609e36d3 | 2865 | msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base; |
44ea2b17 | 2866 | break; |
26bb0981 | 2867 | #endif |
6aa8b732 | 2868 | case MSR_EFER: |
609e36d3 | 2869 | return kvm_get_msr_common(vcpu, msr_info); |
af24a4e4 | 2870 | case MSR_IA32_TSC: |
be7b263e | 2871 | msr_info->data = guest_read_tsc(vcpu); |
6aa8b732 AK |
2872 | break; |
2873 | case MSR_IA32_SYSENTER_CS: | |
609e36d3 | 2874 | msr_info->data = vmcs_read32(GUEST_SYSENTER_CS); |
6aa8b732 AK |
2875 | break; |
2876 | case MSR_IA32_SYSENTER_EIP: | |
609e36d3 | 2877 | msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP); |
6aa8b732 AK |
2878 | break; |
2879 | case MSR_IA32_SYSENTER_ESP: | |
609e36d3 | 2880 | msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP); |
6aa8b732 | 2881 | break; |
0dd376e7 | 2882 | case MSR_IA32_BNDCFGS: |
a87036ad | 2883 | if (!kvm_mpx_supported()) |
93c4adc7 | 2884 | return 1; |
609e36d3 | 2885 | msr_info->data = vmcs_read64(GUEST_BNDCFGS); |
0dd376e7 | 2886 | break; |
cae50139 JK |
2887 | case MSR_IA32_FEATURE_CONTROL: |
2888 | if (!nested_vmx_allowed(vcpu)) | |
2889 | return 1; | |
609e36d3 | 2890 | msr_info->data = to_vmx(vcpu)->nested.msr_ia32_feature_control; |
cae50139 JK |
2891 | break; |
2892 | case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC: | |
2893 | if (!nested_vmx_allowed(vcpu)) | |
2894 | return 1; | |
609e36d3 | 2895 | return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data); |
20300099 WL |
2896 | case MSR_IA32_XSS: |
2897 | if (!vmx_xsaves_supported()) | |
2898 | return 1; | |
609e36d3 | 2899 | msr_info->data = vcpu->arch.ia32_xss; |
20300099 | 2900 | break; |
4e47c7a6 | 2901 | case MSR_TSC_AUX: |
81b1b9ca | 2902 | if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated) |
4e47c7a6 SY |
2903 | return 1; |
2904 | /* Otherwise falls through */ | |
6aa8b732 | 2905 | default: |
609e36d3 | 2906 | msr = find_msr_entry(to_vmx(vcpu), msr_info->index); |
3bab1f5d | 2907 | if (msr) { |
609e36d3 | 2908 | msr_info->data = msr->data; |
3bab1f5d | 2909 | break; |
6aa8b732 | 2910 | } |
609e36d3 | 2911 | return kvm_get_msr_common(vcpu, msr_info); |
6aa8b732 AK |
2912 | } |
2913 | ||
6aa8b732 AK |
2914 | return 0; |
2915 | } | |
2916 | ||
cae50139 JK |
2917 | static void vmx_leave_nested(struct kvm_vcpu *vcpu); |
2918 | ||
6aa8b732 AK |
2919 | /* |
2920 | * Writes msr value into into the appropriate "register". | |
2921 | * Returns 0 on success, non-0 otherwise. | |
2922 | * Assumes vcpu_load() was already called. | |
2923 | */ | |
8fe8ab46 | 2924 | static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
6aa8b732 | 2925 | { |
a2fa3e9f | 2926 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
26bb0981 | 2927 | struct shared_msr_entry *msr; |
2cc51560 | 2928 | int ret = 0; |
8fe8ab46 WA |
2929 | u32 msr_index = msr_info->index; |
2930 | u64 data = msr_info->data; | |
2cc51560 | 2931 | |
6aa8b732 | 2932 | switch (msr_index) { |
3bab1f5d | 2933 | case MSR_EFER: |
8fe8ab46 | 2934 | ret = kvm_set_msr_common(vcpu, msr_info); |
2cc51560 | 2935 | break; |
16175a79 | 2936 | #ifdef CONFIG_X86_64 |
6aa8b732 | 2937 | case MSR_FS_BASE: |
2fb92db1 | 2938 | vmx_segment_cache_clear(vmx); |
6aa8b732 AK |
2939 | vmcs_writel(GUEST_FS_BASE, data); |
2940 | break; | |
2941 | case MSR_GS_BASE: | |
2fb92db1 | 2942 | vmx_segment_cache_clear(vmx); |
6aa8b732 AK |
2943 | vmcs_writel(GUEST_GS_BASE, data); |
2944 | break; | |
44ea2b17 AK |
2945 | case MSR_KERNEL_GS_BASE: |
2946 | vmx_load_host_state(vmx); | |
2947 | vmx->msr_guest_kernel_gs_base = data; | |
2948 | break; | |
6aa8b732 AK |
2949 | #endif |
2950 | case MSR_IA32_SYSENTER_CS: | |
2951 | vmcs_write32(GUEST_SYSENTER_CS, data); | |
2952 | break; | |
2953 | case MSR_IA32_SYSENTER_EIP: | |
f5b42c33 | 2954 | vmcs_writel(GUEST_SYSENTER_EIP, data); |
6aa8b732 AK |
2955 | break; |
2956 | case MSR_IA32_SYSENTER_ESP: | |
f5b42c33 | 2957 | vmcs_writel(GUEST_SYSENTER_ESP, data); |
6aa8b732 | 2958 | break; |
0dd376e7 | 2959 | case MSR_IA32_BNDCFGS: |
a87036ad | 2960 | if (!kvm_mpx_supported()) |
93c4adc7 | 2961 | return 1; |
0dd376e7 LJ |
2962 | vmcs_write64(GUEST_BNDCFGS, data); |
2963 | break; | |
af24a4e4 | 2964 | case MSR_IA32_TSC: |
8fe8ab46 | 2965 | kvm_write_tsc(vcpu, msr_info); |
6aa8b732 | 2966 | break; |
468d472f SY |
2967 | case MSR_IA32_CR_PAT: |
2968 | if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) { | |
4566654b NA |
2969 | if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data)) |
2970 | return 1; | |
468d472f SY |
2971 | vmcs_write64(GUEST_IA32_PAT, data); |
2972 | vcpu->arch.pat = data; | |
2973 | break; | |
2974 | } | |
8fe8ab46 | 2975 | ret = kvm_set_msr_common(vcpu, msr_info); |
4e47c7a6 | 2976 | break; |
ba904635 WA |
2977 | case MSR_IA32_TSC_ADJUST: |
2978 | ret = kvm_set_msr_common(vcpu, msr_info); | |
4e47c7a6 | 2979 | break; |
cae50139 JK |
2980 | case MSR_IA32_FEATURE_CONTROL: |
2981 | if (!nested_vmx_allowed(vcpu) || | |
2982 | (to_vmx(vcpu)->nested.msr_ia32_feature_control & | |
2983 | FEATURE_CONTROL_LOCKED && !msr_info->host_initiated)) | |
2984 | return 1; | |
2985 | vmx->nested.msr_ia32_feature_control = data; | |
2986 | if (msr_info->host_initiated && data == 0) | |
2987 | vmx_leave_nested(vcpu); | |
2988 | break; | |
2989 | case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC: | |
2990 | return 1; /* they are read-only */ | |
20300099 WL |
2991 | case MSR_IA32_XSS: |
2992 | if (!vmx_xsaves_supported()) | |
2993 | return 1; | |
2994 | /* | |
2995 | * The only supported bit as of Skylake is bit 8, but | |
2996 | * it is not supported on KVM. | |
2997 | */ | |
2998 | if (data != 0) | |
2999 | return 1; | |
3000 | vcpu->arch.ia32_xss = data; | |
3001 | if (vcpu->arch.ia32_xss != host_xss) | |
3002 | add_atomic_switch_msr(vmx, MSR_IA32_XSS, | |
3003 | vcpu->arch.ia32_xss, host_xss); | |
3004 | else | |
3005 | clear_atomic_switch_msr(vmx, MSR_IA32_XSS); | |
3006 | break; | |
4e47c7a6 | 3007 | case MSR_TSC_AUX: |
81b1b9ca | 3008 | if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated) |
4e47c7a6 SY |
3009 | return 1; |
3010 | /* Check reserved bit, higher 32 bits should be zero */ | |
3011 | if ((data >> 32) != 0) | |
3012 | return 1; | |
3013 | /* Otherwise falls through */ | |
6aa8b732 | 3014 | default: |
8b9cf98c | 3015 | msr = find_msr_entry(vmx, msr_index); |
3bab1f5d | 3016 | if (msr) { |
8b3c3104 | 3017 | u64 old_msr_data = msr->data; |
3bab1f5d | 3018 | msr->data = data; |
2225fd56 AK |
3019 | if (msr - vmx->guest_msrs < vmx->save_nmsrs) { |
3020 | preempt_disable(); | |
8b3c3104 AH |
3021 | ret = kvm_set_shared_msr(msr->index, msr->data, |
3022 | msr->mask); | |
2225fd56 | 3023 | preempt_enable(); |
8b3c3104 AH |
3024 | if (ret) |
3025 | msr->data = old_msr_data; | |
2225fd56 | 3026 | } |
3bab1f5d | 3027 | break; |
6aa8b732 | 3028 | } |
8fe8ab46 | 3029 | ret = kvm_set_msr_common(vcpu, msr_info); |
6aa8b732 AK |
3030 | } |
3031 | ||
2cc51560 | 3032 | return ret; |
6aa8b732 AK |
3033 | } |
3034 | ||
5fdbf976 | 3035 | static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg) |
6aa8b732 | 3036 | { |
5fdbf976 MT |
3037 | __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail); |
3038 | switch (reg) { | |
3039 | case VCPU_REGS_RSP: | |
3040 | vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP); | |
3041 | break; | |
3042 | case VCPU_REGS_RIP: | |
3043 | vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP); | |
3044 | break; | |
6de4f3ad AK |
3045 | case VCPU_EXREG_PDPTR: |
3046 | if (enable_ept) | |
3047 | ept_save_pdptrs(vcpu); | |
3048 | break; | |
5fdbf976 MT |
3049 | default: |
3050 | break; | |
3051 | } | |
6aa8b732 AK |
3052 | } |
3053 | ||
6aa8b732 AK |
3054 | static __init int cpu_has_kvm_support(void) |
3055 | { | |
6210e37b | 3056 | return cpu_has_vmx(); |
6aa8b732 AK |
3057 | } |
3058 | ||
3059 | static __init int vmx_disabled_by_bios(void) | |
3060 | { | |
3061 | u64 msr; | |
3062 | ||
3063 | rdmsrl(MSR_IA32_FEATURE_CONTROL, msr); | |
cafd6659 | 3064 | if (msr & FEATURE_CONTROL_LOCKED) { |
23f3e991 | 3065 | /* launched w/ TXT and VMX disabled */ |
cafd6659 SW |
3066 | if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX) |
3067 | && tboot_enabled()) | |
3068 | return 1; | |
23f3e991 | 3069 | /* launched w/o TXT and VMX only enabled w/ TXT */ |
cafd6659 | 3070 | if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX) |
23f3e991 | 3071 | && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX) |
f9335afe SW |
3072 | && !tboot_enabled()) { |
3073 | printk(KERN_WARNING "kvm: disable TXT in the BIOS or " | |
23f3e991 | 3074 | "activate TXT before enabling KVM\n"); |
cafd6659 | 3075 | return 1; |
f9335afe | 3076 | } |
23f3e991 JC |
3077 | /* launched w/o TXT and VMX disabled */ |
3078 | if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX) | |
3079 | && !tboot_enabled()) | |
3080 | return 1; | |
cafd6659 SW |
3081 | } |
3082 | ||
3083 | return 0; | |
6aa8b732 AK |
3084 | } |
3085 | ||
7725b894 DX |
3086 | static void kvm_cpu_vmxon(u64 addr) |
3087 | { | |
3088 | asm volatile (ASM_VMX_VMXON_RAX | |
3089 | : : "a"(&addr), "m"(addr) | |
3090 | : "memory", "cc"); | |
3091 | } | |
3092 | ||
13a34e06 | 3093 | static int hardware_enable(void) |
6aa8b732 AK |
3094 | { |
3095 | int cpu = raw_smp_processor_id(); | |
3096 | u64 phys_addr = __pa(per_cpu(vmxarea, cpu)); | |
cafd6659 | 3097 | u64 old, test_bits; |
6aa8b732 | 3098 | |
1e02ce4c | 3099 | if (cr4_read_shadow() & X86_CR4_VMXE) |
10474ae8 AG |
3100 | return -EBUSY; |
3101 | ||
d462b819 | 3102 | INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu)); |
bf9f6ac8 FW |
3103 | INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu)); |
3104 | spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu)); | |
8f536b76 ZY |
3105 | |
3106 | /* | |
3107 | * Now we can enable the vmclear operation in kdump | |
3108 | * since the loaded_vmcss_on_cpu list on this cpu | |
3109 | * has been initialized. | |
3110 | * | |
3111 | * Though the cpu is not in VMX operation now, there | |
3112 | * is no problem to enable the vmclear operation | |
3113 | * for the loaded_vmcss_on_cpu list is empty! | |
3114 | */ | |
3115 | crash_enable_local_vmclear(cpu); | |
3116 | ||
6aa8b732 | 3117 | rdmsrl(MSR_IA32_FEATURE_CONTROL, old); |
cafd6659 SW |
3118 | |
3119 | test_bits = FEATURE_CONTROL_LOCKED; | |
3120 | test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX; | |
3121 | if (tboot_enabled()) | |
3122 | test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX; | |
3123 | ||
3124 | if ((old & test_bits) != test_bits) { | |
6aa8b732 | 3125 | /* enable and lock */ |
cafd6659 SW |
3126 | wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits); |
3127 | } | |
375074cc | 3128 | cr4_set_bits(X86_CR4_VMXE); |
10474ae8 | 3129 | |
4610c9cc DX |
3130 | if (vmm_exclusive) { |
3131 | kvm_cpu_vmxon(phys_addr); | |
3132 | ept_sync_global(); | |
3133 | } | |
10474ae8 | 3134 | |
89cbc767 | 3135 | native_store_gdt(this_cpu_ptr(&host_gdt)); |
3444d7da | 3136 | |
10474ae8 | 3137 | return 0; |
6aa8b732 AK |
3138 | } |
3139 | ||
d462b819 | 3140 | static void vmclear_local_loaded_vmcss(void) |
543e4243 AK |
3141 | { |
3142 | int cpu = raw_smp_processor_id(); | |
d462b819 | 3143 | struct loaded_vmcs *v, *n; |
543e4243 | 3144 | |
d462b819 NHE |
3145 | list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu), |
3146 | loaded_vmcss_on_cpu_link) | |
3147 | __loaded_vmcs_clear(v); | |
543e4243 AK |
3148 | } |
3149 | ||
710ff4a8 EH |
3150 | |
3151 | /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot() | |
3152 | * tricks. | |
3153 | */ | |
3154 | static void kvm_cpu_vmxoff(void) | |
6aa8b732 | 3155 | { |
4ecac3fd | 3156 | asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc"); |
6aa8b732 AK |
3157 | } |
3158 | ||
13a34e06 | 3159 | static void hardware_disable(void) |
710ff4a8 | 3160 | { |
4610c9cc | 3161 | if (vmm_exclusive) { |
d462b819 | 3162 | vmclear_local_loaded_vmcss(); |
4610c9cc DX |
3163 | kvm_cpu_vmxoff(); |
3164 | } | |
375074cc | 3165 | cr4_clear_bits(X86_CR4_VMXE); |
710ff4a8 EH |
3166 | } |
3167 | ||
1c3d14fe | 3168 | static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt, |
d77c26fc | 3169 | u32 msr, u32 *result) |
1c3d14fe YS |
3170 | { |
3171 | u32 vmx_msr_low, vmx_msr_high; | |
3172 | u32 ctl = ctl_min | ctl_opt; | |
3173 | ||
3174 | rdmsr(msr, vmx_msr_low, vmx_msr_high); | |
3175 | ||
3176 | ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */ | |
3177 | ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */ | |
3178 | ||
3179 | /* Ensure minimum (required) set of control bits are supported. */ | |
3180 | if (ctl_min & ~ctl) | |
002c7f7c | 3181 | return -EIO; |
1c3d14fe YS |
3182 | |
3183 | *result = ctl; | |
3184 | return 0; | |
3185 | } | |
3186 | ||
110312c8 AK |
3187 | static __init bool allow_1_setting(u32 msr, u32 ctl) |
3188 | { | |
3189 | u32 vmx_msr_low, vmx_msr_high; | |
3190 | ||
3191 | rdmsr(msr, vmx_msr_low, vmx_msr_high); | |
3192 | return vmx_msr_high & ctl; | |
3193 | } | |
3194 | ||
002c7f7c | 3195 | static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf) |
6aa8b732 AK |
3196 | { |
3197 | u32 vmx_msr_low, vmx_msr_high; | |
d56f546d | 3198 | u32 min, opt, min2, opt2; |
1c3d14fe YS |
3199 | u32 _pin_based_exec_control = 0; |
3200 | u32 _cpu_based_exec_control = 0; | |
f78e0e2e | 3201 | u32 _cpu_based_2nd_exec_control = 0; |
1c3d14fe YS |
3202 | u32 _vmexit_control = 0; |
3203 | u32 _vmentry_control = 0; | |
3204 | ||
10166744 | 3205 | min = CPU_BASED_HLT_EXITING | |
1c3d14fe YS |
3206 | #ifdef CONFIG_X86_64 |
3207 | CPU_BASED_CR8_LOAD_EXITING | | |
3208 | CPU_BASED_CR8_STORE_EXITING | | |
3209 | #endif | |
d56f546d SY |
3210 | CPU_BASED_CR3_LOAD_EXITING | |
3211 | CPU_BASED_CR3_STORE_EXITING | | |
1c3d14fe YS |
3212 | CPU_BASED_USE_IO_BITMAPS | |
3213 | CPU_BASED_MOV_DR_EXITING | | |
a7052897 | 3214 | CPU_BASED_USE_TSC_OFFSETING | |
59708670 SY |
3215 | CPU_BASED_MWAIT_EXITING | |
3216 | CPU_BASED_MONITOR_EXITING | | |
fee84b07 AK |
3217 | CPU_BASED_INVLPG_EXITING | |
3218 | CPU_BASED_RDPMC_EXITING; | |
443381a8 | 3219 | |
f78e0e2e | 3220 | opt = CPU_BASED_TPR_SHADOW | |
25c5f225 | 3221 | CPU_BASED_USE_MSR_BITMAPS | |
f78e0e2e | 3222 | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS; |
1c3d14fe YS |
3223 | if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS, |
3224 | &_cpu_based_exec_control) < 0) | |
002c7f7c | 3225 | return -EIO; |
6e5d865c YS |
3226 | #ifdef CONFIG_X86_64 |
3227 | if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW)) | |
3228 | _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING & | |
3229 | ~CPU_BASED_CR8_STORE_EXITING; | |
3230 | #endif | |
f78e0e2e | 3231 | if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) { |
d56f546d SY |
3232 | min2 = 0; |
3233 | opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES | | |
8d14695f | 3234 | SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE | |
2384d2b3 | 3235 | SECONDARY_EXEC_WBINVD_EXITING | |
d56f546d | 3236 | SECONDARY_EXEC_ENABLE_VPID | |
3a624e29 | 3237 | SECONDARY_EXEC_ENABLE_EPT | |
4b8d54f9 | 3238 | SECONDARY_EXEC_UNRESTRICTED_GUEST | |
4e47c7a6 | 3239 | SECONDARY_EXEC_PAUSE_LOOP_EXITING | |
ad756a16 | 3240 | SECONDARY_EXEC_RDTSCP | |
83d4c286 | 3241 | SECONDARY_EXEC_ENABLE_INVPCID | |
c7c9c56c | 3242 | SECONDARY_EXEC_APIC_REGISTER_VIRT | |
abc4fc58 | 3243 | SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY | |
20300099 | 3244 | SECONDARY_EXEC_SHADOW_VMCS | |
843e4330 | 3245 | SECONDARY_EXEC_XSAVES | |
8b3e34e4 | 3246 | SECONDARY_EXEC_ENABLE_PML | |
64903d61 HZ |
3247 | SECONDARY_EXEC_PCOMMIT | |
3248 | SECONDARY_EXEC_TSC_SCALING; | |
d56f546d SY |
3249 | if (adjust_vmx_controls(min2, opt2, |
3250 | MSR_IA32_VMX_PROCBASED_CTLS2, | |
f78e0e2e SY |
3251 | &_cpu_based_2nd_exec_control) < 0) |
3252 | return -EIO; | |
3253 | } | |
3254 | #ifndef CONFIG_X86_64 | |
3255 | if (!(_cpu_based_2nd_exec_control & | |
3256 | SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) | |
3257 | _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW; | |
3258 | #endif | |
83d4c286 YZ |
3259 | |
3260 | if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW)) | |
3261 | _cpu_based_2nd_exec_control &= ~( | |
8d14695f | 3262 | SECONDARY_EXEC_APIC_REGISTER_VIRT | |
c7c9c56c YZ |
3263 | SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE | |
3264 | SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY); | |
83d4c286 | 3265 | |
d56f546d | 3266 | if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) { |
a7052897 MT |
3267 | /* CR3 accesses and invlpg don't need to cause VM Exits when EPT |
3268 | enabled */ | |
5fff7d27 GN |
3269 | _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING | |
3270 | CPU_BASED_CR3_STORE_EXITING | | |
3271 | CPU_BASED_INVLPG_EXITING); | |
d56f546d SY |
3272 | rdmsr(MSR_IA32_VMX_EPT_VPID_CAP, |
3273 | vmx_capability.ept, vmx_capability.vpid); | |
3274 | } | |
1c3d14fe | 3275 | |
81908bf4 | 3276 | min = VM_EXIT_SAVE_DEBUG_CONTROLS; |
1c3d14fe YS |
3277 | #ifdef CONFIG_X86_64 |
3278 | min |= VM_EXIT_HOST_ADDR_SPACE_SIZE; | |
3279 | #endif | |
a547c6db | 3280 | opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT | |
da8999d3 | 3281 | VM_EXIT_ACK_INTR_ON_EXIT | VM_EXIT_CLEAR_BNDCFGS; |
1c3d14fe YS |
3282 | if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS, |
3283 | &_vmexit_control) < 0) | |
002c7f7c | 3284 | return -EIO; |
1c3d14fe | 3285 | |
01e439be YZ |
3286 | min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING; |
3287 | opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR; | |
3288 | if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS, | |
3289 | &_pin_based_exec_control) < 0) | |
3290 | return -EIO; | |
3291 | ||
3292 | if (!(_cpu_based_2nd_exec_control & | |
3293 | SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) || | |
3294 | !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT)) | |
3295 | _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR; | |
3296 | ||
c845f9c6 | 3297 | min = VM_ENTRY_LOAD_DEBUG_CONTROLS; |
da8999d3 | 3298 | opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS; |
1c3d14fe YS |
3299 | if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS, |
3300 | &_vmentry_control) < 0) | |
002c7f7c | 3301 | return -EIO; |
6aa8b732 | 3302 | |
c68876fd | 3303 | rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high); |
1c3d14fe YS |
3304 | |
3305 | /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */ | |
3306 | if ((vmx_msr_high & 0x1fff) > PAGE_SIZE) | |
002c7f7c | 3307 | return -EIO; |
1c3d14fe YS |
3308 | |
3309 | #ifdef CONFIG_X86_64 | |
3310 | /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */ | |
3311 | if (vmx_msr_high & (1u<<16)) | |
002c7f7c | 3312 | return -EIO; |
1c3d14fe YS |
3313 | #endif |
3314 | ||
3315 | /* Require Write-Back (WB) memory type for VMCS accesses. */ | |
3316 | if (((vmx_msr_high >> 18) & 15) != 6) | |
002c7f7c | 3317 | return -EIO; |
1c3d14fe | 3318 | |
002c7f7c YS |
3319 | vmcs_conf->size = vmx_msr_high & 0x1fff; |
3320 | vmcs_conf->order = get_order(vmcs_config.size); | |
3321 | vmcs_conf->revision_id = vmx_msr_low; | |
1c3d14fe | 3322 | |
002c7f7c YS |
3323 | vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control; |
3324 | vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control; | |
f78e0e2e | 3325 | vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control; |
002c7f7c YS |
3326 | vmcs_conf->vmexit_ctrl = _vmexit_control; |
3327 | vmcs_conf->vmentry_ctrl = _vmentry_control; | |
1c3d14fe | 3328 | |
110312c8 AK |
3329 | cpu_has_load_ia32_efer = |
3330 | allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS, | |
3331 | VM_ENTRY_LOAD_IA32_EFER) | |
3332 | && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS, | |
3333 | VM_EXIT_LOAD_IA32_EFER); | |
3334 | ||
8bf00a52 GN |
3335 | cpu_has_load_perf_global_ctrl = |
3336 | allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS, | |
3337 | VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) | |
3338 | && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS, | |
3339 | VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL); | |
3340 | ||
3341 | /* | |
3342 | * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL | |
3343 | * but due to arrata below it can't be used. Workaround is to use | |
3344 | * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL. | |
3345 | * | |
3346 | * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32] | |
3347 | * | |
3348 | * AAK155 (model 26) | |
3349 | * AAP115 (model 30) | |
3350 | * AAT100 (model 37) | |
3351 | * BC86,AAY89,BD102 (model 44) | |
3352 | * BA97 (model 46) | |
3353 | * | |
3354 | */ | |
3355 | if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) { | |
3356 | switch (boot_cpu_data.x86_model) { | |
3357 | case 26: | |
3358 | case 30: | |
3359 | case 37: | |
3360 | case 44: | |
3361 | case 46: | |
3362 | cpu_has_load_perf_global_ctrl = false; | |
3363 | printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL " | |
3364 | "does not work properly. Using workaround\n"); | |
3365 | break; | |
3366 | default: | |
3367 | break; | |
3368 | } | |
3369 | } | |
3370 | ||
20300099 WL |
3371 | if (cpu_has_xsaves) |
3372 | rdmsrl(MSR_IA32_XSS, host_xss); | |
3373 | ||
1c3d14fe | 3374 | return 0; |
c68876fd | 3375 | } |
6aa8b732 AK |
3376 | |
3377 | static struct vmcs *alloc_vmcs_cpu(int cpu) | |
3378 | { | |
3379 | int node = cpu_to_node(cpu); | |
3380 | struct page *pages; | |
3381 | struct vmcs *vmcs; | |
3382 | ||
96db800f | 3383 | pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order); |
6aa8b732 AK |
3384 | if (!pages) |
3385 | return NULL; | |
3386 | vmcs = page_address(pages); | |
1c3d14fe YS |
3387 | memset(vmcs, 0, vmcs_config.size); |
3388 | vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */ | |
6aa8b732 AK |
3389 | return vmcs; |
3390 | } | |
3391 | ||
3392 | static struct vmcs *alloc_vmcs(void) | |
3393 | { | |
d3b2c338 | 3394 | return alloc_vmcs_cpu(raw_smp_processor_id()); |
6aa8b732 AK |
3395 | } |
3396 | ||
3397 | static void free_vmcs(struct vmcs *vmcs) | |
3398 | { | |
1c3d14fe | 3399 | free_pages((unsigned long)vmcs, vmcs_config.order); |
6aa8b732 AK |
3400 | } |
3401 | ||
d462b819 NHE |
3402 | /* |
3403 | * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded | |
3404 | */ | |
3405 | static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs) | |
3406 | { | |
3407 | if (!loaded_vmcs->vmcs) | |
3408 | return; | |
3409 | loaded_vmcs_clear(loaded_vmcs); | |
3410 | free_vmcs(loaded_vmcs->vmcs); | |
3411 | loaded_vmcs->vmcs = NULL; | |
3412 | } | |
3413 | ||
39959588 | 3414 | static void free_kvm_area(void) |
6aa8b732 AK |
3415 | { |
3416 | int cpu; | |
3417 | ||
3230bb47 | 3418 | for_each_possible_cpu(cpu) { |
6aa8b732 | 3419 | free_vmcs(per_cpu(vmxarea, cpu)); |
3230bb47 ZA |
3420 | per_cpu(vmxarea, cpu) = NULL; |
3421 | } | |
6aa8b732 AK |
3422 | } |
3423 | ||
fe2b201b BD |
3424 | static void init_vmcs_shadow_fields(void) |
3425 | { | |
3426 | int i, j; | |
3427 | ||
3428 | /* No checks for read only fields yet */ | |
3429 | ||
3430 | for (i = j = 0; i < max_shadow_read_write_fields; i++) { | |
3431 | switch (shadow_read_write_fields[i]) { | |
3432 | case GUEST_BNDCFGS: | |
a87036ad | 3433 | if (!kvm_mpx_supported()) |
fe2b201b BD |
3434 | continue; |
3435 | break; | |
3436 | default: | |
3437 | break; | |
3438 | } | |
3439 | ||
3440 | if (j < i) | |
3441 | shadow_read_write_fields[j] = | |
3442 | shadow_read_write_fields[i]; | |
3443 | j++; | |
3444 | } | |
3445 | max_shadow_read_write_fields = j; | |
3446 | ||
3447 | /* shadowed fields guest access without vmexit */ | |
3448 | for (i = 0; i < max_shadow_read_write_fields; i++) { | |
3449 | clear_bit(shadow_read_write_fields[i], | |
3450 | vmx_vmwrite_bitmap); | |
3451 | clear_bit(shadow_read_write_fields[i], | |
3452 | vmx_vmread_bitmap); | |
3453 | } | |
3454 | for (i = 0; i < max_shadow_read_only_fields; i++) | |
3455 | clear_bit(shadow_read_only_fields[i], | |
3456 | vmx_vmread_bitmap); | |
3457 | } | |
3458 | ||
6aa8b732 AK |
3459 | static __init int alloc_kvm_area(void) |
3460 | { | |
3461 | int cpu; | |
3462 | ||
3230bb47 | 3463 | for_each_possible_cpu(cpu) { |
6aa8b732 AK |
3464 | struct vmcs *vmcs; |
3465 | ||
3466 | vmcs = alloc_vmcs_cpu(cpu); | |
3467 | if (!vmcs) { | |
3468 | free_kvm_area(); | |
3469 | return -ENOMEM; | |
3470 | } | |
3471 | ||
3472 | per_cpu(vmxarea, cpu) = vmcs; | |
3473 | } | |
3474 | return 0; | |
3475 | } | |
3476 | ||
14168786 GN |
3477 | static bool emulation_required(struct kvm_vcpu *vcpu) |
3478 | { | |
3479 | return emulate_invalid_guest_state && !guest_state_valid(vcpu); | |
3480 | } | |
3481 | ||
91b0aa2c | 3482 | static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg, |
d99e4152 | 3483 | struct kvm_segment *save) |
6aa8b732 | 3484 | { |
d99e4152 GN |
3485 | if (!emulate_invalid_guest_state) { |
3486 | /* | |
3487 | * CS and SS RPL should be equal during guest entry according | |
3488 | * to VMX spec, but in reality it is not always so. Since vcpu | |
3489 | * is in the middle of the transition from real mode to | |
3490 | * protected mode it is safe to assume that RPL 0 is a good | |
3491 | * default value. | |
3492 | */ | |
3493 | if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS) | |
b32a9918 NA |
3494 | save->selector &= ~SEGMENT_RPL_MASK; |
3495 | save->dpl = save->selector & SEGMENT_RPL_MASK; | |
d99e4152 | 3496 | save->s = 1; |
6aa8b732 | 3497 | } |
d99e4152 | 3498 | vmx_set_segment(vcpu, save, seg); |
6aa8b732 AK |
3499 | } |
3500 | ||
3501 | static void enter_pmode(struct kvm_vcpu *vcpu) | |
3502 | { | |
3503 | unsigned long flags; | |
a89a8fb9 | 3504 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
6aa8b732 | 3505 | |
d99e4152 GN |
3506 | /* |
3507 | * Update real mode segment cache. It may be not up-to-date if sement | |
3508 | * register was written while vcpu was in a guest mode. | |
3509 | */ | |
3510 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES); | |
3511 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS); | |
3512 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS); | |
3513 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS); | |
3514 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS); | |
3515 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS); | |
3516 | ||
7ffd92c5 | 3517 | vmx->rmode.vm86_active = 0; |
6aa8b732 | 3518 | |
2fb92db1 AK |
3519 | vmx_segment_cache_clear(vmx); |
3520 | ||
f5f7b2fe | 3521 | vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR); |
6aa8b732 AK |
3522 | |
3523 | flags = vmcs_readl(GUEST_RFLAGS); | |
78ac8b47 AK |
3524 | flags &= RMODE_GUEST_OWNED_EFLAGS_BITS; |
3525 | flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS; | |
6aa8b732 AK |
3526 | vmcs_writel(GUEST_RFLAGS, flags); |
3527 | ||
66aee91a RR |
3528 | vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) | |
3529 | (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME)); | |
6aa8b732 AK |
3530 | |
3531 | update_exception_bitmap(vcpu); | |
3532 | ||
91b0aa2c GN |
3533 | fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]); |
3534 | fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]); | |
3535 | fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]); | |
3536 | fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]); | |
3537 | fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]); | |
3538 | fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]); | |
6aa8b732 AK |
3539 | } |
3540 | ||
f5f7b2fe | 3541 | static void fix_rmode_seg(int seg, struct kvm_segment *save) |
6aa8b732 | 3542 | { |
772e0318 | 3543 | const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg]; |
d99e4152 GN |
3544 | struct kvm_segment var = *save; |
3545 | ||
3546 | var.dpl = 0x3; | |
3547 | if (seg == VCPU_SREG_CS) | |
3548 | var.type = 0x3; | |
3549 | ||
3550 | if (!emulate_invalid_guest_state) { | |
3551 | var.selector = var.base >> 4; | |
3552 | var.base = var.base & 0xffff0; | |
3553 | var.limit = 0xffff; | |
3554 | var.g = 0; | |
3555 | var.db = 0; | |
3556 | var.present = 1; | |
3557 | var.s = 1; | |
3558 | var.l = 0; | |
3559 | var.unusable = 0; | |
3560 | var.type = 0x3; | |
3561 | var.avl = 0; | |
3562 | if (save->base & 0xf) | |
3563 | printk_once(KERN_WARNING "kvm: segment base is not " | |
3564 | "paragraph aligned when entering " | |
3565 | "protected mode (seg=%d)", seg); | |
3566 | } | |
6aa8b732 | 3567 | |
d99e4152 GN |
3568 | vmcs_write16(sf->selector, var.selector); |
3569 | vmcs_write32(sf->base, var.base); | |
3570 | vmcs_write32(sf->limit, var.limit); | |
3571 | vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var)); | |
6aa8b732 AK |
3572 | } |
3573 | ||
3574 | static void enter_rmode(struct kvm_vcpu *vcpu) | |
3575 | { | |
3576 | unsigned long flags; | |
a89a8fb9 | 3577 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
6aa8b732 | 3578 | |
f5f7b2fe AK |
3579 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR); |
3580 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES); | |
3581 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS); | |
3582 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS); | |
3583 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS); | |
c6ad1153 GN |
3584 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS); |
3585 | vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS); | |
f5f7b2fe | 3586 | |
7ffd92c5 | 3587 | vmx->rmode.vm86_active = 1; |
6aa8b732 | 3588 | |
776e58ea GN |
3589 | /* |
3590 | * Very old userspace does not call KVM_SET_TSS_ADDR before entering | |
4918c6ca | 3591 | * vcpu. Warn the user that an update is overdue. |
776e58ea | 3592 | */ |
4918c6ca | 3593 | if (!vcpu->kvm->arch.tss_addr) |
776e58ea GN |
3594 | printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be " |
3595 | "called before entering vcpu\n"); | |
776e58ea | 3596 | |
2fb92db1 AK |
3597 | vmx_segment_cache_clear(vmx); |
3598 | ||
4918c6ca | 3599 | vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr); |
6aa8b732 | 3600 | vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1); |
6aa8b732 AK |
3601 | vmcs_write32(GUEST_TR_AR_BYTES, 0x008b); |
3602 | ||
3603 | flags = vmcs_readl(GUEST_RFLAGS); | |
78ac8b47 | 3604 | vmx->rmode.save_rflags = flags; |
6aa8b732 | 3605 | |
053de044 | 3606 | flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM; |
6aa8b732 AK |
3607 | |
3608 | vmcs_writel(GUEST_RFLAGS, flags); | |
66aee91a | 3609 | vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME); |
6aa8b732 AK |
3610 | update_exception_bitmap(vcpu); |
3611 | ||
d99e4152 GN |
3612 | fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]); |
3613 | fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]); | |
3614 | fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]); | |
3615 | fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]); | |
3616 | fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]); | |
3617 | fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]); | |
b246dd5d | 3618 | |
8668a3c4 | 3619 | kvm_mmu_reset_context(vcpu); |
6aa8b732 AK |
3620 | } |
3621 | ||
401d10de AS |
3622 | static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer) |
3623 | { | |
3624 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
26bb0981 AK |
3625 | struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER); |
3626 | ||
3627 | if (!msr) | |
3628 | return; | |
401d10de | 3629 | |
44ea2b17 AK |
3630 | /* |
3631 | * Force kernel_gs_base reloading before EFER changes, as control | |
3632 | * of this msr depends on is_long_mode(). | |
3633 | */ | |
3634 | vmx_load_host_state(to_vmx(vcpu)); | |
f6801dff | 3635 | vcpu->arch.efer = efer; |
401d10de | 3636 | if (efer & EFER_LMA) { |
2961e876 | 3637 | vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE); |
401d10de AS |
3638 | msr->data = efer; |
3639 | } else { | |
2961e876 | 3640 | vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE); |
401d10de AS |
3641 | |
3642 | msr->data = efer & ~EFER_LME; | |
3643 | } | |
3644 | setup_msrs(vmx); | |
3645 | } | |
3646 | ||
05b3e0c2 | 3647 | #ifdef CONFIG_X86_64 |
6aa8b732 AK |
3648 | |
3649 | static void enter_lmode(struct kvm_vcpu *vcpu) | |
3650 | { | |
3651 | u32 guest_tr_ar; | |
3652 | ||
2fb92db1 AK |
3653 | vmx_segment_cache_clear(to_vmx(vcpu)); |
3654 | ||
6aa8b732 | 3655 | guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES); |
4d283ec9 | 3656 | if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) { |
bd80158a JK |
3657 | pr_debug_ratelimited("%s: tss fixup for long mode. \n", |
3658 | __func__); | |
6aa8b732 | 3659 | vmcs_write32(GUEST_TR_AR_BYTES, |
4d283ec9 AL |
3660 | (guest_tr_ar & ~VMX_AR_TYPE_MASK) |
3661 | | VMX_AR_TYPE_BUSY_64_TSS); | |
6aa8b732 | 3662 | } |
da38f438 | 3663 | vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA); |
6aa8b732 AK |
3664 | } |
3665 | ||
3666 | static void exit_lmode(struct kvm_vcpu *vcpu) | |
3667 | { | |
2961e876 | 3668 | vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE); |
da38f438 | 3669 | vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA); |
6aa8b732 AK |
3670 | } |
3671 | ||
3672 | #endif | |
3673 | ||
dd5f5341 | 3674 | static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid) |
2384d2b3 | 3675 | { |
dd5f5341 | 3676 | vpid_sync_context(vpid); |
dd180b3e XG |
3677 | if (enable_ept) { |
3678 | if (!VALID_PAGE(vcpu->arch.mmu.root_hpa)) | |
3679 | return; | |
4e1096d2 | 3680 | ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa)); |
dd180b3e | 3681 | } |
2384d2b3 SY |
3682 | } |
3683 | ||
dd5f5341 WL |
3684 | static void vmx_flush_tlb(struct kvm_vcpu *vcpu) |
3685 | { | |
3686 | __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid); | |
3687 | } | |
3688 | ||
e8467fda AK |
3689 | static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu) |
3690 | { | |
3691 | ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits; | |
3692 | ||
3693 | vcpu->arch.cr0 &= ~cr0_guest_owned_bits; | |
3694 | vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits; | |
3695 | } | |
3696 | ||
aff48baa AK |
3697 | static void vmx_decache_cr3(struct kvm_vcpu *vcpu) |
3698 | { | |
3699 | if (enable_ept && is_paging(vcpu)) | |
3700 | vcpu->arch.cr3 = vmcs_readl(GUEST_CR3); | |
3701 | __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail); | |
3702 | } | |
3703 | ||
25c4c276 | 3704 | static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu) |
399badf3 | 3705 | { |
fc78f519 AK |
3706 | ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits; |
3707 | ||
3708 | vcpu->arch.cr4 &= ~cr4_guest_owned_bits; | |
3709 | vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits; | |
399badf3 AK |
3710 | } |
3711 | ||
1439442c SY |
3712 | static void ept_load_pdptrs(struct kvm_vcpu *vcpu) |
3713 | { | |
d0d538b9 GN |
3714 | struct kvm_mmu *mmu = vcpu->arch.walk_mmu; |
3715 | ||
6de4f3ad AK |
3716 | if (!test_bit(VCPU_EXREG_PDPTR, |
3717 | (unsigned long *)&vcpu->arch.regs_dirty)) | |
3718 | return; | |
3719 | ||
1439442c | 3720 | if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) { |
d0d538b9 GN |
3721 | vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]); |
3722 | vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]); | |
3723 | vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]); | |
3724 | vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]); | |
1439442c SY |
3725 | } |
3726 | } | |
3727 | ||
8f5d549f AK |
3728 | static void ept_save_pdptrs(struct kvm_vcpu *vcpu) |
3729 | { | |
d0d538b9 GN |
3730 | struct kvm_mmu *mmu = vcpu->arch.walk_mmu; |
3731 | ||
8f5d549f | 3732 | if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) { |
d0d538b9 GN |
3733 | mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0); |
3734 | mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1); | |
3735 | mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2); | |
3736 | mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3); | |
8f5d549f | 3737 | } |
6de4f3ad AK |
3738 | |
3739 | __set_bit(VCPU_EXREG_PDPTR, | |
3740 | (unsigned long *)&vcpu->arch.regs_avail); | |
3741 | __set_bit(VCPU_EXREG_PDPTR, | |
3742 | (unsigned long *)&vcpu->arch.regs_dirty); | |
8f5d549f AK |
3743 | } |
3744 | ||
5e1746d6 | 3745 | static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4); |
1439442c SY |
3746 | |
3747 | static void ept_update_paging_mode_cr0(unsigned long *hw_cr0, | |
3748 | unsigned long cr0, | |
3749 | struct kvm_vcpu *vcpu) | |
3750 | { | |
5233dd51 MT |
3751 | if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail)) |
3752 | vmx_decache_cr3(vcpu); | |
1439442c SY |
3753 | if (!(cr0 & X86_CR0_PG)) { |
3754 | /* From paging/starting to nonpaging */ | |
3755 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, | |
65267ea1 | 3756 | vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) | |
1439442c SY |
3757 | (CPU_BASED_CR3_LOAD_EXITING | |
3758 | CPU_BASED_CR3_STORE_EXITING)); | |
3759 | vcpu->arch.cr0 = cr0; | |
fc78f519 | 3760 | vmx_set_cr4(vcpu, kvm_read_cr4(vcpu)); |
1439442c SY |
3761 | } else if (!is_paging(vcpu)) { |
3762 | /* From nonpaging to paging */ | |
3763 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, | |
65267ea1 | 3764 | vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) & |
1439442c SY |
3765 | ~(CPU_BASED_CR3_LOAD_EXITING | |
3766 | CPU_BASED_CR3_STORE_EXITING)); | |
3767 | vcpu->arch.cr0 = cr0; | |
fc78f519 | 3768 | vmx_set_cr4(vcpu, kvm_read_cr4(vcpu)); |
1439442c | 3769 | } |
95eb84a7 SY |
3770 | |
3771 | if (!(cr0 & X86_CR0_WP)) | |
3772 | *hw_cr0 &= ~X86_CR0_WP; | |
1439442c SY |
3773 | } |
3774 | ||
6aa8b732 AK |
3775 | static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
3776 | { | |
7ffd92c5 | 3777 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
3a624e29 NK |
3778 | unsigned long hw_cr0; |
3779 | ||
5037878e | 3780 | hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK); |
3a624e29 | 3781 | if (enable_unrestricted_guest) |
5037878e | 3782 | hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST; |
218e763f | 3783 | else { |
5037878e | 3784 | hw_cr0 |= KVM_VM_CR0_ALWAYS_ON; |
1439442c | 3785 | |
218e763f GN |
3786 | if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE)) |
3787 | enter_pmode(vcpu); | |
6aa8b732 | 3788 | |
218e763f GN |
3789 | if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE)) |
3790 | enter_rmode(vcpu); | |
3791 | } | |
6aa8b732 | 3792 | |
05b3e0c2 | 3793 | #ifdef CONFIG_X86_64 |
f6801dff | 3794 | if (vcpu->arch.efer & EFER_LME) { |
707d92fa | 3795 | if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) |
6aa8b732 | 3796 | enter_lmode(vcpu); |
707d92fa | 3797 | if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) |
6aa8b732 AK |
3798 | exit_lmode(vcpu); |
3799 | } | |
3800 | #endif | |
3801 | ||
089d034e | 3802 | if (enable_ept) |
1439442c SY |
3803 | ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu); |
3804 | ||
02daab21 | 3805 | if (!vcpu->fpu_active) |
81231c69 | 3806 | hw_cr0 |= X86_CR0_TS | X86_CR0_MP; |
02daab21 | 3807 | |
6aa8b732 | 3808 | vmcs_writel(CR0_READ_SHADOW, cr0); |
1439442c | 3809 | vmcs_writel(GUEST_CR0, hw_cr0); |
ad312c7c | 3810 | vcpu->arch.cr0 = cr0; |
14168786 GN |
3811 | |
3812 | /* depends on vcpu->arch.cr0 to be set to a new value */ | |
3813 | vmx->emulation_required = emulation_required(vcpu); | |
6aa8b732 AK |
3814 | } |
3815 | ||
1439442c SY |
3816 | static u64 construct_eptp(unsigned long root_hpa) |
3817 | { | |
3818 | u64 eptp; | |
3819 | ||
3820 | /* TODO write the value reading from MSR */ | |
3821 | eptp = VMX_EPT_DEFAULT_MT | | |
3822 | VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT; | |
b38f9934 XH |
3823 | if (enable_ept_ad_bits) |
3824 | eptp |= VMX_EPT_AD_ENABLE_BIT; | |
1439442c SY |
3825 | eptp |= (root_hpa & PAGE_MASK); |
3826 | ||
3827 | return eptp; | |
3828 | } | |
3829 | ||
6aa8b732 AK |
3830 | static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3) |
3831 | { | |
1439442c SY |
3832 | unsigned long guest_cr3; |
3833 | u64 eptp; | |
3834 | ||
3835 | guest_cr3 = cr3; | |
089d034e | 3836 | if (enable_ept) { |
1439442c SY |
3837 | eptp = construct_eptp(cr3); |
3838 | vmcs_write64(EPT_POINTER, eptp); | |
59ab5a8f JK |
3839 | if (is_paging(vcpu) || is_guest_mode(vcpu)) |
3840 | guest_cr3 = kvm_read_cr3(vcpu); | |
3841 | else | |
3842 | guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr; | |
7c93be44 | 3843 | ept_load_pdptrs(vcpu); |
1439442c SY |
3844 | } |
3845 | ||
2384d2b3 | 3846 | vmx_flush_tlb(vcpu); |
1439442c | 3847 | vmcs_writel(GUEST_CR3, guest_cr3); |
6aa8b732 AK |
3848 | } |
3849 | ||
5e1746d6 | 3850 | static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) |
6aa8b732 | 3851 | { |
085e68ee BS |
3852 | /* |
3853 | * Pass through host's Machine Check Enable value to hw_cr4, which | |
3854 | * is in force while we are in guest mode. Do not let guests control | |
3855 | * this bit, even if host CR4.MCE == 0. | |
3856 | */ | |
3857 | unsigned long hw_cr4 = | |
3858 | (cr4_read_shadow() & X86_CR4_MCE) | | |
3859 | (cr4 & ~X86_CR4_MCE) | | |
3860 | (to_vmx(vcpu)->rmode.vm86_active ? | |
3861 | KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON); | |
1439442c | 3862 | |
5e1746d6 NHE |
3863 | if (cr4 & X86_CR4_VMXE) { |
3864 | /* | |
3865 | * To use VMXON (and later other VMX instructions), a guest | |
3866 | * must first be able to turn on cr4.VMXE (see handle_vmon()). | |
3867 | * So basically the check on whether to allow nested VMX | |
3868 | * is here. | |
3869 | */ | |
3870 | if (!nested_vmx_allowed(vcpu)) | |
3871 | return 1; | |
1a0d74e6 JK |
3872 | } |
3873 | if (to_vmx(vcpu)->nested.vmxon && | |
3874 | ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) | |
5e1746d6 NHE |
3875 | return 1; |
3876 | ||
ad312c7c | 3877 | vcpu->arch.cr4 = cr4; |
bc23008b AK |
3878 | if (enable_ept) { |
3879 | if (!is_paging(vcpu)) { | |
3880 | hw_cr4 &= ~X86_CR4_PAE; | |
3881 | hw_cr4 |= X86_CR4_PSE; | |
3882 | } else if (!(cr4 & X86_CR4_PAE)) { | |
3883 | hw_cr4 &= ~X86_CR4_PAE; | |
3884 | } | |
3885 | } | |
1439442c | 3886 | |
656ec4a4 RK |
3887 | if (!enable_unrestricted_guest && !is_paging(vcpu)) |
3888 | /* | |
3889 | * SMEP/SMAP is disabled if CPU is in non-paging mode in | |
3890 | * hardware. However KVM always uses paging mode without | |
3891 | * unrestricted guest. | |
3892 | * To emulate this behavior, SMEP/SMAP needs to be manually | |
3893 | * disabled when guest switches to non-paging mode. | |
3894 | */ | |
3895 | hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP); | |
3896 | ||
1439442c SY |
3897 | vmcs_writel(CR4_READ_SHADOW, cr4); |
3898 | vmcs_writel(GUEST_CR4, hw_cr4); | |
5e1746d6 | 3899 | return 0; |
6aa8b732 AK |
3900 | } |
3901 | ||
6aa8b732 AK |
3902 | static void vmx_get_segment(struct kvm_vcpu *vcpu, |
3903 | struct kvm_segment *var, int seg) | |
3904 | { | |
a9179499 | 3905 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
6aa8b732 AK |
3906 | u32 ar; |
3907 | ||
c6ad1153 | 3908 | if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) { |
f5f7b2fe | 3909 | *var = vmx->rmode.segs[seg]; |
a9179499 | 3910 | if (seg == VCPU_SREG_TR |
2fb92db1 | 3911 | || var->selector == vmx_read_guest_seg_selector(vmx, seg)) |
f5f7b2fe | 3912 | return; |
1390a28b AK |
3913 | var->base = vmx_read_guest_seg_base(vmx, seg); |
3914 | var->selector = vmx_read_guest_seg_selector(vmx, seg); | |
3915 | return; | |
a9179499 | 3916 | } |
2fb92db1 AK |
3917 | var->base = vmx_read_guest_seg_base(vmx, seg); |
3918 | var->limit = vmx_read_guest_seg_limit(vmx, seg); | |
3919 | var->selector = vmx_read_guest_seg_selector(vmx, seg); | |
3920 | ar = vmx_read_guest_seg_ar(vmx, seg); | |
03617c18 | 3921 | var->unusable = (ar >> 16) & 1; |
6aa8b732 AK |
3922 | var->type = ar & 15; |
3923 | var->s = (ar >> 4) & 1; | |
3924 | var->dpl = (ar >> 5) & 3; | |
03617c18 GN |
3925 | /* |
3926 | * Some userspaces do not preserve unusable property. Since usable | |
3927 | * segment has to be present according to VMX spec we can use present | |
3928 | * property to amend userspace bug by making unusable segment always | |
3929 | * nonpresent. vmx_segment_access_rights() already marks nonpresent | |
3930 | * segment as unusable. | |
3931 | */ | |
3932 | var->present = !var->unusable; | |
6aa8b732 AK |
3933 | var->avl = (ar >> 12) & 1; |
3934 | var->l = (ar >> 13) & 1; | |
3935 | var->db = (ar >> 14) & 1; | |
3936 | var->g = (ar >> 15) & 1; | |
6aa8b732 AK |
3937 | } |
3938 | ||
a9179499 AK |
3939 | static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg) |
3940 | { | |
a9179499 AK |
3941 | struct kvm_segment s; |
3942 | ||
3943 | if (to_vmx(vcpu)->rmode.vm86_active) { | |
3944 | vmx_get_segment(vcpu, &s, seg); | |
3945 | return s.base; | |
3946 | } | |
2fb92db1 | 3947 | return vmx_read_guest_seg_base(to_vmx(vcpu), seg); |
a9179499 AK |
3948 | } |
3949 | ||
b09408d0 | 3950 | static int vmx_get_cpl(struct kvm_vcpu *vcpu) |
2e4d2653 | 3951 | { |
b09408d0 MT |
3952 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
3953 | ||
ae9fedc7 | 3954 | if (unlikely(vmx->rmode.vm86_active)) |
2e4d2653 | 3955 | return 0; |
ae9fedc7 PB |
3956 | else { |
3957 | int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS); | |
4d283ec9 | 3958 | return VMX_AR_DPL(ar); |
69c73028 | 3959 | } |
69c73028 AK |
3960 | } |
3961 | ||
653e3108 | 3962 | static u32 vmx_segment_access_rights(struct kvm_segment *var) |
6aa8b732 | 3963 | { |
6aa8b732 AK |
3964 | u32 ar; |
3965 | ||
f0495f9b | 3966 | if (var->unusable || !var->present) |
6aa8b732 AK |
3967 | ar = 1 << 16; |
3968 | else { | |
3969 | ar = var->type & 15; | |
3970 | ar |= (var->s & 1) << 4; | |
3971 | ar |= (var->dpl & 3) << 5; | |
3972 | ar |= (var->present & 1) << 7; | |
3973 | ar |= (var->avl & 1) << 12; | |
3974 | ar |= (var->l & 1) << 13; | |
3975 | ar |= (var->db & 1) << 14; | |
3976 | ar |= (var->g & 1) << 15; | |
3977 | } | |
653e3108 AK |
3978 | |
3979 | return ar; | |
3980 | } | |
3981 | ||
3982 | static void vmx_set_segment(struct kvm_vcpu *vcpu, | |
3983 | struct kvm_segment *var, int seg) | |
3984 | { | |
7ffd92c5 | 3985 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
772e0318 | 3986 | const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg]; |
653e3108 | 3987 | |
2fb92db1 AK |
3988 | vmx_segment_cache_clear(vmx); |
3989 | ||
1ecd50a9 GN |
3990 | if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) { |
3991 | vmx->rmode.segs[seg] = *var; | |
3992 | if (seg == VCPU_SREG_TR) | |
3993 | vmcs_write16(sf->selector, var->selector); | |
3994 | else if (var->s) | |
3995 | fix_rmode_seg(seg, &vmx->rmode.segs[seg]); | |
d99e4152 | 3996 | goto out; |
653e3108 | 3997 | } |
1ecd50a9 | 3998 | |
653e3108 AK |
3999 | vmcs_writel(sf->base, var->base); |
4000 | vmcs_write32(sf->limit, var->limit); | |
4001 | vmcs_write16(sf->selector, var->selector); | |
3a624e29 NK |
4002 | |
4003 | /* | |
4004 | * Fix the "Accessed" bit in AR field of segment registers for older | |
4005 | * qemu binaries. | |
4006 | * IA32 arch specifies that at the time of processor reset the | |
4007 | * "Accessed" bit in the AR field of segment registers is 1. And qemu | |
0fa06071 | 4008 | * is setting it to 0 in the userland code. This causes invalid guest |
3a624e29 NK |
4009 | * state vmexit when "unrestricted guest" mode is turned on. |
4010 | * Fix for this setup issue in cpu_reset is being pushed in the qemu | |
4011 | * tree. Newer qemu binaries with that qemu fix would not need this | |
4012 | * kvm hack. | |
4013 | */ | |
4014 | if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR)) | |
f924d66d | 4015 | var->type |= 0x1; /* Accessed */ |
3a624e29 | 4016 | |
f924d66d | 4017 | vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var)); |
d99e4152 GN |
4018 | |
4019 | out: | |
98eb2f8b | 4020 | vmx->emulation_required = emulation_required(vcpu); |
6aa8b732 AK |
4021 | } |
4022 | ||
6aa8b732 AK |
4023 | static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l) |
4024 | { | |
2fb92db1 | 4025 | u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS); |
6aa8b732 AK |
4026 | |
4027 | *db = (ar >> 14) & 1; | |
4028 | *l = (ar >> 13) & 1; | |
4029 | } | |
4030 | ||
89a27f4d | 4031 | static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 4032 | { |
89a27f4d GN |
4033 | dt->size = vmcs_read32(GUEST_IDTR_LIMIT); |
4034 | dt->address = vmcs_readl(GUEST_IDTR_BASE); | |
6aa8b732 AK |
4035 | } |
4036 | ||
89a27f4d | 4037 | static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 4038 | { |
89a27f4d GN |
4039 | vmcs_write32(GUEST_IDTR_LIMIT, dt->size); |
4040 | vmcs_writel(GUEST_IDTR_BASE, dt->address); | |
6aa8b732 AK |
4041 | } |
4042 | ||
89a27f4d | 4043 | static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 4044 | { |
89a27f4d GN |
4045 | dt->size = vmcs_read32(GUEST_GDTR_LIMIT); |
4046 | dt->address = vmcs_readl(GUEST_GDTR_BASE); | |
6aa8b732 AK |
4047 | } |
4048 | ||
89a27f4d | 4049 | static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 4050 | { |
89a27f4d GN |
4051 | vmcs_write32(GUEST_GDTR_LIMIT, dt->size); |
4052 | vmcs_writel(GUEST_GDTR_BASE, dt->address); | |
6aa8b732 AK |
4053 | } |
4054 | ||
648dfaa7 MG |
4055 | static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg) |
4056 | { | |
4057 | struct kvm_segment var; | |
4058 | u32 ar; | |
4059 | ||
4060 | vmx_get_segment(vcpu, &var, seg); | |
07f42f5f | 4061 | var.dpl = 0x3; |
0647f4aa GN |
4062 | if (seg == VCPU_SREG_CS) |
4063 | var.type = 0x3; | |
648dfaa7 MG |
4064 | ar = vmx_segment_access_rights(&var); |
4065 | ||
4066 | if (var.base != (var.selector << 4)) | |
4067 | return false; | |
89efbed0 | 4068 | if (var.limit != 0xffff) |
648dfaa7 | 4069 | return false; |
07f42f5f | 4070 | if (ar != 0xf3) |
648dfaa7 MG |
4071 | return false; |
4072 | ||
4073 | return true; | |
4074 | } | |
4075 | ||
4076 | static bool code_segment_valid(struct kvm_vcpu *vcpu) | |
4077 | { | |
4078 | struct kvm_segment cs; | |
4079 | unsigned int cs_rpl; | |
4080 | ||
4081 | vmx_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
b32a9918 | 4082 | cs_rpl = cs.selector & SEGMENT_RPL_MASK; |
648dfaa7 | 4083 | |
1872a3f4 AK |
4084 | if (cs.unusable) |
4085 | return false; | |
4d283ec9 | 4086 | if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK)) |
648dfaa7 MG |
4087 | return false; |
4088 | if (!cs.s) | |
4089 | return false; | |
4d283ec9 | 4090 | if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) { |
648dfaa7 MG |
4091 | if (cs.dpl > cs_rpl) |
4092 | return false; | |
1872a3f4 | 4093 | } else { |
648dfaa7 MG |
4094 | if (cs.dpl != cs_rpl) |
4095 | return false; | |
4096 | } | |
4097 | if (!cs.present) | |
4098 | return false; | |
4099 | ||
4100 | /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */ | |
4101 | return true; | |
4102 | } | |
4103 | ||
4104 | static bool stack_segment_valid(struct kvm_vcpu *vcpu) | |
4105 | { | |
4106 | struct kvm_segment ss; | |
4107 | unsigned int ss_rpl; | |
4108 | ||
4109 | vmx_get_segment(vcpu, &ss, VCPU_SREG_SS); | |
b32a9918 | 4110 | ss_rpl = ss.selector & SEGMENT_RPL_MASK; |
648dfaa7 | 4111 | |
1872a3f4 AK |
4112 | if (ss.unusable) |
4113 | return true; | |
4114 | if (ss.type != 3 && ss.type != 7) | |
648dfaa7 MG |
4115 | return false; |
4116 | if (!ss.s) | |
4117 | return false; | |
4118 | if (ss.dpl != ss_rpl) /* DPL != RPL */ | |
4119 | return false; | |
4120 | if (!ss.present) | |
4121 | return false; | |
4122 | ||
4123 | return true; | |
4124 | } | |
4125 | ||
4126 | static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg) | |
4127 | { | |
4128 | struct kvm_segment var; | |
4129 | unsigned int rpl; | |
4130 | ||
4131 | vmx_get_segment(vcpu, &var, seg); | |
b32a9918 | 4132 | rpl = var.selector & SEGMENT_RPL_MASK; |
648dfaa7 | 4133 | |
1872a3f4 AK |
4134 | if (var.unusable) |
4135 | return true; | |
648dfaa7 MG |
4136 | if (!var.s) |
4137 | return false; | |
4138 | if (!var.present) | |
4139 | return false; | |
4d283ec9 | 4140 | if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) { |
648dfaa7 MG |
4141 | if (var.dpl < rpl) /* DPL < RPL */ |
4142 | return false; | |
4143 | } | |
4144 | ||
4145 | /* TODO: Add other members to kvm_segment_field to allow checking for other access | |
4146 | * rights flags | |
4147 | */ | |
4148 | return true; | |
4149 | } | |
4150 | ||
4151 | static bool tr_valid(struct kvm_vcpu *vcpu) | |
4152 | { | |
4153 | struct kvm_segment tr; | |
4154 | ||
4155 | vmx_get_segment(vcpu, &tr, VCPU_SREG_TR); | |
4156 | ||
1872a3f4 AK |
4157 | if (tr.unusable) |
4158 | return false; | |
b32a9918 | 4159 | if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */ |
648dfaa7 | 4160 | return false; |
1872a3f4 | 4161 | if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */ |
648dfaa7 MG |
4162 | return false; |
4163 | if (!tr.present) | |
4164 | return false; | |
4165 | ||
4166 | return true; | |
4167 | } | |
4168 | ||
4169 | static bool ldtr_valid(struct kvm_vcpu *vcpu) | |
4170 | { | |
4171 | struct kvm_segment ldtr; | |
4172 | ||
4173 | vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR); | |
4174 | ||
1872a3f4 AK |
4175 | if (ldtr.unusable) |
4176 | return true; | |
b32a9918 | 4177 | if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */ |
648dfaa7 MG |
4178 | return false; |
4179 | if (ldtr.type != 2) | |
4180 | return false; | |
4181 | if (!ldtr.present) | |
4182 | return false; | |
4183 | ||
4184 | return true; | |
4185 | } | |
4186 | ||
4187 | static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu) | |
4188 | { | |
4189 | struct kvm_segment cs, ss; | |
4190 | ||
4191 | vmx_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
4192 | vmx_get_segment(vcpu, &ss, VCPU_SREG_SS); | |
4193 | ||
b32a9918 NA |
4194 | return ((cs.selector & SEGMENT_RPL_MASK) == |
4195 | (ss.selector & SEGMENT_RPL_MASK)); | |
648dfaa7 MG |
4196 | } |
4197 | ||
4198 | /* | |
4199 | * Check if guest state is valid. Returns true if valid, false if | |
4200 | * not. | |
4201 | * We assume that registers are always usable | |
4202 | */ | |
4203 | static bool guest_state_valid(struct kvm_vcpu *vcpu) | |
4204 | { | |
c5e97c80 GN |
4205 | if (enable_unrestricted_guest) |
4206 | return true; | |
4207 | ||
648dfaa7 | 4208 | /* real mode guest state checks */ |
f13882d8 | 4209 | if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) { |
648dfaa7 MG |
4210 | if (!rmode_segment_valid(vcpu, VCPU_SREG_CS)) |
4211 | return false; | |
4212 | if (!rmode_segment_valid(vcpu, VCPU_SREG_SS)) | |
4213 | return false; | |
4214 | if (!rmode_segment_valid(vcpu, VCPU_SREG_DS)) | |
4215 | return false; | |
4216 | if (!rmode_segment_valid(vcpu, VCPU_SREG_ES)) | |
4217 | return false; | |
4218 | if (!rmode_segment_valid(vcpu, VCPU_SREG_FS)) | |
4219 | return false; | |
4220 | if (!rmode_segment_valid(vcpu, VCPU_SREG_GS)) | |
4221 | return false; | |
4222 | } else { | |
4223 | /* protected mode guest state checks */ | |
4224 | if (!cs_ss_rpl_check(vcpu)) | |
4225 | return false; | |
4226 | if (!code_segment_valid(vcpu)) | |
4227 | return false; | |
4228 | if (!stack_segment_valid(vcpu)) | |
4229 | return false; | |
4230 | if (!data_segment_valid(vcpu, VCPU_SREG_DS)) | |
4231 | return false; | |
4232 | if (!data_segment_valid(vcpu, VCPU_SREG_ES)) | |
4233 | return false; | |
4234 | if (!data_segment_valid(vcpu, VCPU_SREG_FS)) | |
4235 | return false; | |
4236 | if (!data_segment_valid(vcpu, VCPU_SREG_GS)) | |
4237 | return false; | |
4238 | if (!tr_valid(vcpu)) | |
4239 | return false; | |
4240 | if (!ldtr_valid(vcpu)) | |
4241 | return false; | |
4242 | } | |
4243 | /* TODO: | |
4244 | * - Add checks on RIP | |
4245 | * - Add checks on RFLAGS | |
4246 | */ | |
4247 | ||
4248 | return true; | |
4249 | } | |
4250 | ||
d77c26fc | 4251 | static int init_rmode_tss(struct kvm *kvm) |
6aa8b732 | 4252 | { |
40dcaa9f | 4253 | gfn_t fn; |
195aefde | 4254 | u16 data = 0; |
1f755a82 | 4255 | int idx, r; |
6aa8b732 | 4256 | |
40dcaa9f | 4257 | idx = srcu_read_lock(&kvm->srcu); |
4918c6ca | 4258 | fn = kvm->arch.tss_addr >> PAGE_SHIFT; |
195aefde IE |
4259 | r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE); |
4260 | if (r < 0) | |
10589a46 | 4261 | goto out; |
195aefde | 4262 | data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE; |
464d17c8 SY |
4263 | r = kvm_write_guest_page(kvm, fn++, &data, |
4264 | TSS_IOPB_BASE_OFFSET, sizeof(u16)); | |
195aefde | 4265 | if (r < 0) |
10589a46 | 4266 | goto out; |
195aefde IE |
4267 | r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE); |
4268 | if (r < 0) | |
10589a46 | 4269 | goto out; |
195aefde IE |
4270 | r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE); |
4271 | if (r < 0) | |
10589a46 | 4272 | goto out; |
195aefde | 4273 | data = ~0; |
10589a46 MT |
4274 | r = kvm_write_guest_page(kvm, fn, &data, |
4275 | RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1, | |
4276 | sizeof(u8)); | |
10589a46 | 4277 | out: |
40dcaa9f | 4278 | srcu_read_unlock(&kvm->srcu, idx); |
1f755a82 | 4279 | return r; |
6aa8b732 AK |
4280 | } |
4281 | ||
b7ebfb05 SY |
4282 | static int init_rmode_identity_map(struct kvm *kvm) |
4283 | { | |
f51770ed | 4284 | int i, idx, r = 0; |
ba049e93 | 4285 | kvm_pfn_t identity_map_pfn; |
b7ebfb05 SY |
4286 | u32 tmp; |
4287 | ||
089d034e | 4288 | if (!enable_ept) |
f51770ed | 4289 | return 0; |
a255d479 TC |
4290 | |
4291 | /* Protect kvm->arch.ept_identity_pagetable_done. */ | |
4292 | mutex_lock(&kvm->slots_lock); | |
4293 | ||
f51770ed | 4294 | if (likely(kvm->arch.ept_identity_pagetable_done)) |
a255d479 | 4295 | goto out2; |
a255d479 | 4296 | |
b927a3ce | 4297 | identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT; |
a255d479 TC |
4298 | |
4299 | r = alloc_identity_pagetable(kvm); | |
f51770ed | 4300 | if (r < 0) |
a255d479 TC |
4301 | goto out2; |
4302 | ||
40dcaa9f | 4303 | idx = srcu_read_lock(&kvm->srcu); |
b7ebfb05 SY |
4304 | r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE); |
4305 | if (r < 0) | |
4306 | goto out; | |
4307 | /* Set up identity-mapping pagetable for EPT in real mode */ | |
4308 | for (i = 0; i < PT32_ENT_PER_PAGE; i++) { | |
4309 | tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | | |
4310 | _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE); | |
4311 | r = kvm_write_guest_page(kvm, identity_map_pfn, | |
4312 | &tmp, i * sizeof(tmp), sizeof(tmp)); | |
4313 | if (r < 0) | |
4314 | goto out; | |
4315 | } | |
4316 | kvm->arch.ept_identity_pagetable_done = true; | |
f51770ed | 4317 | |
b7ebfb05 | 4318 | out: |
40dcaa9f | 4319 | srcu_read_unlock(&kvm->srcu, idx); |
a255d479 TC |
4320 | |
4321 | out2: | |
4322 | mutex_unlock(&kvm->slots_lock); | |
f51770ed | 4323 | return r; |
b7ebfb05 SY |
4324 | } |
4325 | ||
6aa8b732 AK |
4326 | static void seg_setup(int seg) |
4327 | { | |
772e0318 | 4328 | const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg]; |
3a624e29 | 4329 | unsigned int ar; |
6aa8b732 AK |
4330 | |
4331 | vmcs_write16(sf->selector, 0); | |
4332 | vmcs_writel(sf->base, 0); | |
4333 | vmcs_write32(sf->limit, 0xffff); | |
d54d07b2 GN |
4334 | ar = 0x93; |
4335 | if (seg == VCPU_SREG_CS) | |
4336 | ar |= 0x08; /* code segment */ | |
3a624e29 NK |
4337 | |
4338 | vmcs_write32(sf->ar_bytes, ar); | |
6aa8b732 AK |
4339 | } |
4340 | ||
f78e0e2e SY |
4341 | static int alloc_apic_access_page(struct kvm *kvm) |
4342 | { | |
4484141a | 4343 | struct page *page; |
f78e0e2e SY |
4344 | int r = 0; |
4345 | ||
79fac95e | 4346 | mutex_lock(&kvm->slots_lock); |
c24ae0dc | 4347 | if (kvm->arch.apic_access_page_done) |
f78e0e2e | 4348 | goto out; |
1d8007bd PB |
4349 | r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, |
4350 | APIC_DEFAULT_PHYS_BASE, PAGE_SIZE); | |
f78e0e2e SY |
4351 | if (r) |
4352 | goto out; | |
72dc67a6 | 4353 | |
73a6d941 | 4354 | page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT); |
4484141a XG |
4355 | if (is_error_page(page)) { |
4356 | r = -EFAULT; | |
4357 | goto out; | |
4358 | } | |
4359 | ||
c24ae0dc TC |
4360 | /* |
4361 | * Do not pin the page in memory, so that memory hot-unplug | |
4362 | * is able to migrate it. | |
4363 | */ | |
4364 | put_page(page); | |
4365 | kvm->arch.apic_access_page_done = true; | |
f78e0e2e | 4366 | out: |
79fac95e | 4367 | mutex_unlock(&kvm->slots_lock); |
f78e0e2e SY |
4368 | return r; |
4369 | } | |
4370 | ||
b7ebfb05 SY |
4371 | static int alloc_identity_pagetable(struct kvm *kvm) |
4372 | { | |
a255d479 TC |
4373 | /* Called with kvm->slots_lock held. */ |
4374 | ||
b7ebfb05 SY |
4375 | int r = 0; |
4376 | ||
a255d479 TC |
4377 | BUG_ON(kvm->arch.ept_identity_pagetable_done); |
4378 | ||
1d8007bd PB |
4379 | r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, |
4380 | kvm->arch.ept_identity_map_addr, PAGE_SIZE); | |
b7ebfb05 | 4381 | |
b7ebfb05 SY |
4382 | return r; |
4383 | } | |
4384 | ||
991e7a0e | 4385 | static int allocate_vpid(void) |
2384d2b3 SY |
4386 | { |
4387 | int vpid; | |
4388 | ||
919818ab | 4389 | if (!enable_vpid) |
991e7a0e | 4390 | return 0; |
2384d2b3 SY |
4391 | spin_lock(&vmx_vpid_lock); |
4392 | vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS); | |
991e7a0e | 4393 | if (vpid < VMX_NR_VPIDS) |
2384d2b3 | 4394 | __set_bit(vpid, vmx_vpid_bitmap); |
991e7a0e WL |
4395 | else |
4396 | vpid = 0; | |
2384d2b3 | 4397 | spin_unlock(&vmx_vpid_lock); |
991e7a0e | 4398 | return vpid; |
2384d2b3 SY |
4399 | } |
4400 | ||
991e7a0e | 4401 | static void free_vpid(int vpid) |
cdbecfc3 | 4402 | { |
991e7a0e | 4403 | if (!enable_vpid || vpid == 0) |
cdbecfc3 LJ |
4404 | return; |
4405 | spin_lock(&vmx_vpid_lock); | |
991e7a0e | 4406 | __clear_bit(vpid, vmx_vpid_bitmap); |
cdbecfc3 LJ |
4407 | spin_unlock(&vmx_vpid_lock); |
4408 | } | |
4409 | ||
8d14695f YZ |
4410 | #define MSR_TYPE_R 1 |
4411 | #define MSR_TYPE_W 2 | |
4412 | static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, | |
4413 | u32 msr, int type) | |
25c5f225 | 4414 | { |
3e7c73e9 | 4415 | int f = sizeof(unsigned long); |
25c5f225 SY |
4416 | |
4417 | if (!cpu_has_vmx_msr_bitmap()) | |
4418 | return; | |
4419 | ||
4420 | /* | |
4421 | * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals | |
4422 | * have the write-low and read-high bitmap offsets the wrong way round. | |
4423 | * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff. | |
4424 | */ | |
25c5f225 | 4425 | if (msr <= 0x1fff) { |
8d14695f YZ |
4426 | if (type & MSR_TYPE_R) |
4427 | /* read-low */ | |
4428 | __clear_bit(msr, msr_bitmap + 0x000 / f); | |
4429 | ||
4430 | if (type & MSR_TYPE_W) | |
4431 | /* write-low */ | |
4432 | __clear_bit(msr, msr_bitmap + 0x800 / f); | |
4433 | ||
25c5f225 SY |
4434 | } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) { |
4435 | msr &= 0x1fff; | |
8d14695f YZ |
4436 | if (type & MSR_TYPE_R) |
4437 | /* read-high */ | |
4438 | __clear_bit(msr, msr_bitmap + 0x400 / f); | |
4439 | ||
4440 | if (type & MSR_TYPE_W) | |
4441 | /* write-high */ | |
4442 | __clear_bit(msr, msr_bitmap + 0xc00 / f); | |
4443 | ||
4444 | } | |
4445 | } | |
4446 | ||
4447 | static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap, | |
4448 | u32 msr, int type) | |
4449 | { | |
4450 | int f = sizeof(unsigned long); | |
4451 | ||
4452 | if (!cpu_has_vmx_msr_bitmap()) | |
4453 | return; | |
4454 | ||
4455 | /* | |
4456 | * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals | |
4457 | * have the write-low and read-high bitmap offsets the wrong way round. | |
4458 | * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff. | |
4459 | */ | |
4460 | if (msr <= 0x1fff) { | |
4461 | if (type & MSR_TYPE_R) | |
4462 | /* read-low */ | |
4463 | __set_bit(msr, msr_bitmap + 0x000 / f); | |
4464 | ||
4465 | if (type & MSR_TYPE_W) | |
4466 | /* write-low */ | |
4467 | __set_bit(msr, msr_bitmap + 0x800 / f); | |
4468 | ||
4469 | } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) { | |
4470 | msr &= 0x1fff; | |
4471 | if (type & MSR_TYPE_R) | |
4472 | /* read-high */ | |
4473 | __set_bit(msr, msr_bitmap + 0x400 / f); | |
4474 | ||
4475 | if (type & MSR_TYPE_W) | |
4476 | /* write-high */ | |
4477 | __set_bit(msr, msr_bitmap + 0xc00 / f); | |
4478 | ||
25c5f225 | 4479 | } |
25c5f225 SY |
4480 | } |
4481 | ||
f2b93280 WV |
4482 | /* |
4483 | * If a msr is allowed by L0, we should check whether it is allowed by L1. | |
4484 | * The corresponding bit will be cleared unless both of L0 and L1 allow it. | |
4485 | */ | |
4486 | static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1, | |
4487 | unsigned long *msr_bitmap_nested, | |
4488 | u32 msr, int type) | |
4489 | { | |
4490 | int f = sizeof(unsigned long); | |
4491 | ||
4492 | if (!cpu_has_vmx_msr_bitmap()) { | |
4493 | WARN_ON(1); | |
4494 | return; | |
4495 | } | |
4496 | ||
4497 | /* | |
4498 | * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals | |
4499 | * have the write-low and read-high bitmap offsets the wrong way round. | |
4500 | * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff. | |
4501 | */ | |
4502 | if (msr <= 0x1fff) { | |
4503 | if (type & MSR_TYPE_R && | |
4504 | !test_bit(msr, msr_bitmap_l1 + 0x000 / f)) | |
4505 | /* read-low */ | |
4506 | __clear_bit(msr, msr_bitmap_nested + 0x000 / f); | |
4507 | ||
4508 | if (type & MSR_TYPE_W && | |
4509 | !test_bit(msr, msr_bitmap_l1 + 0x800 / f)) | |
4510 | /* write-low */ | |
4511 | __clear_bit(msr, msr_bitmap_nested + 0x800 / f); | |
4512 | ||
4513 | } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) { | |
4514 | msr &= 0x1fff; | |
4515 | if (type & MSR_TYPE_R && | |
4516 | !test_bit(msr, msr_bitmap_l1 + 0x400 / f)) | |
4517 | /* read-high */ | |
4518 | __clear_bit(msr, msr_bitmap_nested + 0x400 / f); | |
4519 | ||
4520 | if (type & MSR_TYPE_W && | |
4521 | !test_bit(msr, msr_bitmap_l1 + 0xc00 / f)) | |
4522 | /* write-high */ | |
4523 | __clear_bit(msr, msr_bitmap_nested + 0xc00 / f); | |
4524 | ||
4525 | } | |
4526 | } | |
4527 | ||
5897297b AK |
4528 | static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only) |
4529 | { | |
4530 | if (!longmode_only) | |
8d14695f YZ |
4531 | __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, |
4532 | msr, MSR_TYPE_R | MSR_TYPE_W); | |
4533 | __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, | |
4534 | msr, MSR_TYPE_R | MSR_TYPE_W); | |
4535 | } | |
4536 | ||
4537 | static void vmx_enable_intercept_msr_read_x2apic(u32 msr) | |
4538 | { | |
4539 | __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic, | |
4540 | msr, MSR_TYPE_R); | |
4541 | __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic, | |
4542 | msr, MSR_TYPE_R); | |
4543 | } | |
4544 | ||
4545 | static void vmx_disable_intercept_msr_read_x2apic(u32 msr) | |
4546 | { | |
4547 | __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic, | |
4548 | msr, MSR_TYPE_R); | |
4549 | __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic, | |
4550 | msr, MSR_TYPE_R); | |
4551 | } | |
4552 | ||
4553 | static void vmx_disable_intercept_msr_write_x2apic(u32 msr) | |
4554 | { | |
4555 | __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic, | |
4556 | msr, MSR_TYPE_W); | |
4557 | __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic, | |
4558 | msr, MSR_TYPE_W); | |
5897297b AK |
4559 | } |
4560 | ||
d62caabb | 4561 | static bool vmx_get_enable_apicv(void) |
d50ab6c1 | 4562 | { |
d62caabb | 4563 | return enable_apicv; |
d50ab6c1 PB |
4564 | } |
4565 | ||
705699a1 WV |
4566 | static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu) |
4567 | { | |
4568 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
4569 | int max_irr; | |
4570 | void *vapic_page; | |
4571 | u16 status; | |
4572 | ||
4573 | if (vmx->nested.pi_desc && | |
4574 | vmx->nested.pi_pending) { | |
4575 | vmx->nested.pi_pending = false; | |
4576 | if (!pi_test_and_clear_on(vmx->nested.pi_desc)) | |
4577 | return 0; | |
4578 | ||
4579 | max_irr = find_last_bit( | |
4580 | (unsigned long *)vmx->nested.pi_desc->pir, 256); | |
4581 | ||
4582 | if (max_irr == 256) | |
4583 | return 0; | |
4584 | ||
4585 | vapic_page = kmap(vmx->nested.virtual_apic_page); | |
4586 | if (!vapic_page) { | |
4587 | WARN_ON(1); | |
4588 | return -ENOMEM; | |
4589 | } | |
4590 | __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page); | |
4591 | kunmap(vmx->nested.virtual_apic_page); | |
4592 | ||
4593 | status = vmcs_read16(GUEST_INTR_STATUS); | |
4594 | if ((u8)max_irr > ((u8)status & 0xff)) { | |
4595 | status &= ~0xff; | |
4596 | status |= (u8)max_irr; | |
4597 | vmcs_write16(GUEST_INTR_STATUS, status); | |
4598 | } | |
4599 | } | |
4600 | return 0; | |
4601 | } | |
4602 | ||
21bc8dc5 RK |
4603 | static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu) |
4604 | { | |
4605 | #ifdef CONFIG_SMP | |
4606 | if (vcpu->mode == IN_GUEST_MODE) { | |
28b835d6 FW |
4607 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
4608 | ||
4609 | /* | |
4610 | * Currently, we don't support urgent interrupt, | |
4611 | * all interrupts are recognized as non-urgent | |
4612 | * interrupt, so we cannot post interrupts when | |
4613 | * 'SN' is set. | |
4614 | * | |
4615 | * If the vcpu is in guest mode, it means it is | |
4616 | * running instead of being scheduled out and | |
4617 | * waiting in the run queue, and that's the only | |
4618 | * case when 'SN' is set currently, warning if | |
4619 | * 'SN' is set. | |
4620 | */ | |
4621 | WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc)); | |
4622 | ||
21bc8dc5 RK |
4623 | apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), |
4624 | POSTED_INTR_VECTOR); | |
4625 | return true; | |
4626 | } | |
4627 | #endif | |
4628 | return false; | |
4629 | } | |
4630 | ||
705699a1 WV |
4631 | static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu, |
4632 | int vector) | |
4633 | { | |
4634 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
4635 | ||
4636 | if (is_guest_mode(vcpu) && | |
4637 | vector == vmx->nested.posted_intr_nv) { | |
4638 | /* the PIR and ON have been set by L1. */ | |
21bc8dc5 | 4639 | kvm_vcpu_trigger_posted_interrupt(vcpu); |
705699a1 WV |
4640 | /* |
4641 | * If a posted intr is not recognized by hardware, | |
4642 | * we will accomplish it in the next vmentry. | |
4643 | */ | |
4644 | vmx->nested.pi_pending = true; | |
4645 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
4646 | return 0; | |
4647 | } | |
4648 | return -1; | |
4649 | } | |
a20ed54d YZ |
4650 | /* |
4651 | * Send interrupt to vcpu via posted interrupt way. | |
4652 | * 1. If target vcpu is running(non-root mode), send posted interrupt | |
4653 | * notification to vcpu and hardware will sync PIR to vIRR atomically. | |
4654 | * 2. If target vcpu isn't running(root mode), kick it to pick up the | |
4655 | * interrupt from PIR in next vmentry. | |
4656 | */ | |
4657 | static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector) | |
4658 | { | |
4659 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
4660 | int r; | |
4661 | ||
705699a1 WV |
4662 | r = vmx_deliver_nested_posted_interrupt(vcpu, vector); |
4663 | if (!r) | |
4664 | return; | |
4665 | ||
a20ed54d YZ |
4666 | if (pi_test_and_set_pir(vector, &vmx->pi_desc)) |
4667 | return; | |
4668 | ||
4669 | r = pi_test_and_set_on(&vmx->pi_desc); | |
4670 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
21bc8dc5 | 4671 | if (r || !kvm_vcpu_trigger_posted_interrupt(vcpu)) |
a20ed54d YZ |
4672 | kvm_vcpu_kick(vcpu); |
4673 | } | |
4674 | ||
4675 | static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu) | |
4676 | { | |
4677 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
4678 | ||
4679 | if (!pi_test_and_clear_on(&vmx->pi_desc)) | |
4680 | return; | |
4681 | ||
4682 | kvm_apic_update_irr(vcpu, vmx->pi_desc.pir); | |
4683 | } | |
4684 | ||
a3a8ff8e NHE |
4685 | /* |
4686 | * Set up the vmcs's constant host-state fields, i.e., host-state fields that | |
4687 | * will not change in the lifetime of the guest. | |
4688 | * Note that host-state that does change is set elsewhere. E.g., host-state | |
4689 | * that is set differently for each CPU is set in vmx_vcpu_load(), not here. | |
4690 | */ | |
a547c6db | 4691 | static void vmx_set_constant_host_state(struct vcpu_vmx *vmx) |
a3a8ff8e NHE |
4692 | { |
4693 | u32 low32, high32; | |
4694 | unsigned long tmpl; | |
4695 | struct desc_ptr dt; | |
d974baa3 | 4696 | unsigned long cr4; |
a3a8ff8e | 4697 | |
b1a74bf8 | 4698 | vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */ |
a3a8ff8e NHE |
4699 | vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */ |
4700 | ||
d974baa3 | 4701 | /* Save the most likely value for this task's CR4 in the VMCS. */ |
1e02ce4c | 4702 | cr4 = cr4_read_shadow(); |
d974baa3 AL |
4703 | vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */ |
4704 | vmx->host_state.vmcs_host_cr4 = cr4; | |
4705 | ||
a3a8ff8e | 4706 | vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */ |
b2da15ac AK |
4707 | #ifdef CONFIG_X86_64 |
4708 | /* | |
4709 | * Load null selectors, so we can avoid reloading them in | |
4710 | * __vmx_load_host_state(), in case userspace uses the null selectors | |
4711 | * too (the expected case). | |
4712 | */ | |
4713 | vmcs_write16(HOST_DS_SELECTOR, 0); | |
4714 | vmcs_write16(HOST_ES_SELECTOR, 0); | |
4715 | #else | |
a3a8ff8e NHE |
4716 | vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */ |
4717 | vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */ | |
b2da15ac | 4718 | #endif |
a3a8ff8e NHE |
4719 | vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */ |
4720 | vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */ | |
4721 | ||
4722 | native_store_idt(&dt); | |
4723 | vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */ | |
a547c6db | 4724 | vmx->host_idt_base = dt.address; |
a3a8ff8e | 4725 | |
83287ea4 | 4726 | vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */ |
a3a8ff8e NHE |
4727 | |
4728 | rdmsr(MSR_IA32_SYSENTER_CS, low32, high32); | |
4729 | vmcs_write32(HOST_IA32_SYSENTER_CS, low32); | |
4730 | rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl); | |
4731 | vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */ | |
4732 | ||
4733 | if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) { | |
4734 | rdmsr(MSR_IA32_CR_PAT, low32, high32); | |
4735 | vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32)); | |
4736 | } | |
4737 | } | |
4738 | ||
bf8179a0 NHE |
4739 | static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx) |
4740 | { | |
4741 | vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS; | |
4742 | if (enable_ept) | |
4743 | vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE; | |
fe3ef05c NHE |
4744 | if (is_guest_mode(&vmx->vcpu)) |
4745 | vmx->vcpu.arch.cr4_guest_owned_bits &= | |
4746 | ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask; | |
bf8179a0 NHE |
4747 | vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits); |
4748 | } | |
4749 | ||
01e439be YZ |
4750 | static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx) |
4751 | { | |
4752 | u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl; | |
4753 | ||
d62caabb | 4754 | if (!kvm_vcpu_apicv_active(&vmx->vcpu)) |
01e439be YZ |
4755 | pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR; |
4756 | return pin_based_exec_ctrl; | |
4757 | } | |
4758 | ||
d62caabb AS |
4759 | static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu) |
4760 | { | |
4761 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
4762 | ||
4763 | vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx)); | |
4764 | } | |
4765 | ||
bf8179a0 NHE |
4766 | static u32 vmx_exec_control(struct vcpu_vmx *vmx) |
4767 | { | |
4768 | u32 exec_control = vmcs_config.cpu_based_exec_ctrl; | |
d16c293e PB |
4769 | |
4770 | if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT) | |
4771 | exec_control &= ~CPU_BASED_MOV_DR_EXITING; | |
4772 | ||
35754c98 | 4773 | if (!cpu_need_tpr_shadow(&vmx->vcpu)) { |
bf8179a0 NHE |
4774 | exec_control &= ~CPU_BASED_TPR_SHADOW; |
4775 | #ifdef CONFIG_X86_64 | |
4776 | exec_control |= CPU_BASED_CR8_STORE_EXITING | | |
4777 | CPU_BASED_CR8_LOAD_EXITING; | |
4778 | #endif | |
4779 | } | |
4780 | if (!enable_ept) | |
4781 | exec_control |= CPU_BASED_CR3_STORE_EXITING | | |
4782 | CPU_BASED_CR3_LOAD_EXITING | | |
4783 | CPU_BASED_INVLPG_EXITING; | |
4784 | return exec_control; | |
4785 | } | |
4786 | ||
4787 | static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx) | |
4788 | { | |
4789 | u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl; | |
35754c98 | 4790 | if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu)) |
bf8179a0 NHE |
4791 | exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES; |
4792 | if (vmx->vpid == 0) | |
4793 | exec_control &= ~SECONDARY_EXEC_ENABLE_VPID; | |
4794 | if (!enable_ept) { | |
4795 | exec_control &= ~SECONDARY_EXEC_ENABLE_EPT; | |
4796 | enable_unrestricted_guest = 0; | |
ad756a16 MJ |
4797 | /* Enable INVPCID for non-ept guests may cause performance regression. */ |
4798 | exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID; | |
bf8179a0 NHE |
4799 | } |
4800 | if (!enable_unrestricted_guest) | |
4801 | exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST; | |
4802 | if (!ple_gap) | |
4803 | exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING; | |
d62caabb | 4804 | if (!kvm_vcpu_apicv_active(&vmx->vcpu)) |
c7c9c56c YZ |
4805 | exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT | |
4806 | SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY); | |
8d14695f | 4807 | exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE; |
abc4fc58 AG |
4808 | /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD |
4809 | (handle_vmptrld). | |
4810 | We can NOT enable shadow_vmcs here because we don't have yet | |
4811 | a current VMCS12 | |
4812 | */ | |
4813 | exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS; | |
a3eaa864 KH |
4814 | |
4815 | if (!enable_pml) | |
4816 | exec_control &= ~SECONDARY_EXEC_ENABLE_PML; | |
843e4330 | 4817 | |
8b3e34e4 XG |
4818 | /* Currently, we allow L1 guest to directly run pcommit instruction. */ |
4819 | exec_control &= ~SECONDARY_EXEC_PCOMMIT; | |
4820 | ||
bf8179a0 NHE |
4821 | return exec_control; |
4822 | } | |
4823 | ||
ce88decf XG |
4824 | static void ept_set_mmio_spte_mask(void) |
4825 | { | |
4826 | /* | |
4827 | * EPT Misconfigurations can be generated if the value of bits 2:0 | |
4828 | * of an EPT paging-structure entry is 110b (write/execute). | |
885032b9 | 4829 | * Also, magic bits (0x3ull << 62) is set to quickly identify mmio |
ce88decf XG |
4830 | * spte. |
4831 | */ | |
885032b9 | 4832 | kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull); |
ce88decf XG |
4833 | } |
4834 | ||
f53cd63c | 4835 | #define VMX_XSS_EXIT_BITMAP 0 |
6aa8b732 AK |
4836 | /* |
4837 | * Sets up the vmcs for emulated real mode. | |
4838 | */ | |
8b9cf98c | 4839 | static int vmx_vcpu_setup(struct vcpu_vmx *vmx) |
6aa8b732 | 4840 | { |
2e4ce7f5 | 4841 | #ifdef CONFIG_X86_64 |
6aa8b732 | 4842 | unsigned long a; |
2e4ce7f5 | 4843 | #endif |
6aa8b732 | 4844 | int i; |
6aa8b732 | 4845 | |
6aa8b732 | 4846 | /* I/O */ |
3e7c73e9 AK |
4847 | vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a)); |
4848 | vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b)); | |
6aa8b732 | 4849 | |
4607c2d7 AG |
4850 | if (enable_shadow_vmcs) { |
4851 | vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap)); | |
4852 | vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap)); | |
4853 | } | |
25c5f225 | 4854 | if (cpu_has_vmx_msr_bitmap()) |
5897297b | 4855 | vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy)); |
25c5f225 | 4856 | |
6aa8b732 AK |
4857 | vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */ |
4858 | ||
6aa8b732 | 4859 | /* Control */ |
01e439be | 4860 | vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx)); |
6e5d865c | 4861 | |
bf8179a0 | 4862 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx)); |
6aa8b732 | 4863 | |
8b3e34e4 | 4864 | if (cpu_has_secondary_exec_ctrls()) |
bf8179a0 NHE |
4865 | vmcs_write32(SECONDARY_VM_EXEC_CONTROL, |
4866 | vmx_secondary_exec_control(vmx)); | |
f78e0e2e | 4867 | |
d62caabb | 4868 | if (kvm_vcpu_apicv_active(&vmx->vcpu)) { |
c7c9c56c YZ |
4869 | vmcs_write64(EOI_EXIT_BITMAP0, 0); |
4870 | vmcs_write64(EOI_EXIT_BITMAP1, 0); | |
4871 | vmcs_write64(EOI_EXIT_BITMAP2, 0); | |
4872 | vmcs_write64(EOI_EXIT_BITMAP3, 0); | |
4873 | ||
4874 | vmcs_write16(GUEST_INTR_STATUS, 0); | |
01e439be | 4875 | |
0bcf261c | 4876 | vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR); |
01e439be | 4877 | vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc))); |
c7c9c56c YZ |
4878 | } |
4879 | ||
4b8d54f9 ZE |
4880 | if (ple_gap) { |
4881 | vmcs_write32(PLE_GAP, ple_gap); | |
a7653ecd RK |
4882 | vmx->ple_window = ple_window; |
4883 | vmx->ple_window_dirty = true; | |
4b8d54f9 ZE |
4884 | } |
4885 | ||
c3707958 XG |
4886 | vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0); |
4887 | vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0); | |
6aa8b732 AK |
4888 | vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */ |
4889 | ||
9581d442 AK |
4890 | vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */ |
4891 | vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */ | |
a547c6db | 4892 | vmx_set_constant_host_state(vmx); |
05b3e0c2 | 4893 | #ifdef CONFIG_X86_64 |
6aa8b732 AK |
4894 | rdmsrl(MSR_FS_BASE, a); |
4895 | vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */ | |
4896 | rdmsrl(MSR_GS_BASE, a); | |
4897 | vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */ | |
4898 | #else | |
4899 | vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */ | |
4900 | vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */ | |
4901 | #endif | |
4902 | ||
2cc51560 ED |
4903 | vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0); |
4904 | vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0); | |
61d2ef2c | 4905 | vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host)); |
2cc51560 | 4906 | vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0); |
61d2ef2c | 4907 | vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest)); |
6aa8b732 | 4908 | |
74545705 RK |
4909 | if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) |
4910 | vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat); | |
468d472f | 4911 | |
03916db9 | 4912 | for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) { |
6aa8b732 AK |
4913 | u32 index = vmx_msr_index[i]; |
4914 | u32 data_low, data_high; | |
a2fa3e9f | 4915 | int j = vmx->nmsrs; |
6aa8b732 AK |
4916 | |
4917 | if (rdmsr_safe(index, &data_low, &data_high) < 0) | |
4918 | continue; | |
432bd6cb AK |
4919 | if (wrmsr_safe(index, data_low, data_high) < 0) |
4920 | continue; | |
26bb0981 AK |
4921 | vmx->guest_msrs[j].index = i; |
4922 | vmx->guest_msrs[j].data = 0; | |
d5696725 | 4923 | vmx->guest_msrs[j].mask = -1ull; |
a2fa3e9f | 4924 | ++vmx->nmsrs; |
6aa8b732 | 4925 | } |
6aa8b732 | 4926 | |
2961e876 GN |
4927 | |
4928 | vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl); | |
6aa8b732 AK |
4929 | |
4930 | /* 22.2.1, 20.8.1 */ | |
2961e876 | 4931 | vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl); |
1c3d14fe | 4932 | |
e00c8cf2 | 4933 | vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL); |
bf8179a0 | 4934 | set_cr4_guest_host_mask(vmx); |
e00c8cf2 | 4935 | |
f53cd63c WL |
4936 | if (vmx_xsaves_supported()) |
4937 | vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP); | |
4938 | ||
e00c8cf2 AK |
4939 | return 0; |
4940 | } | |
4941 | ||
d28bc9dd | 4942 | static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) |
e00c8cf2 AK |
4943 | { |
4944 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
58cb628d | 4945 | struct msr_data apic_base_msr; |
d28bc9dd | 4946 | u64 cr0; |
e00c8cf2 | 4947 | |
7ffd92c5 | 4948 | vmx->rmode.vm86_active = 0; |
e00c8cf2 | 4949 | |
3b86cd99 JK |
4950 | vmx->soft_vnmi_blocked = 0; |
4951 | ||
ad312c7c | 4952 | vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val(); |
d28bc9dd NA |
4953 | kvm_set_cr8(vcpu, 0); |
4954 | ||
4955 | if (!init_event) { | |
4956 | apic_base_msr.data = APIC_DEFAULT_PHYS_BASE | | |
4957 | MSR_IA32_APICBASE_ENABLE; | |
4958 | if (kvm_vcpu_is_reset_bsp(vcpu)) | |
4959 | apic_base_msr.data |= MSR_IA32_APICBASE_BSP; | |
4960 | apic_base_msr.host_initiated = true; | |
4961 | kvm_set_apic_base(vcpu, &apic_base_msr); | |
4962 | } | |
e00c8cf2 | 4963 | |
2fb92db1 AK |
4964 | vmx_segment_cache_clear(vmx); |
4965 | ||
5706be0d | 4966 | seg_setup(VCPU_SREG_CS); |
66450a21 | 4967 | vmcs_write16(GUEST_CS_SELECTOR, 0xf000); |
f3531054 | 4968 | vmcs_writel(GUEST_CS_BASE, 0xffff0000ul); |
e00c8cf2 AK |
4969 | |
4970 | seg_setup(VCPU_SREG_DS); | |
4971 | seg_setup(VCPU_SREG_ES); | |
4972 | seg_setup(VCPU_SREG_FS); | |
4973 | seg_setup(VCPU_SREG_GS); | |
4974 | seg_setup(VCPU_SREG_SS); | |
4975 | ||
4976 | vmcs_write16(GUEST_TR_SELECTOR, 0); | |
4977 | vmcs_writel(GUEST_TR_BASE, 0); | |
4978 | vmcs_write32(GUEST_TR_LIMIT, 0xffff); | |
4979 | vmcs_write32(GUEST_TR_AR_BYTES, 0x008b); | |
4980 | ||
4981 | vmcs_write16(GUEST_LDTR_SELECTOR, 0); | |
4982 | vmcs_writel(GUEST_LDTR_BASE, 0); | |
4983 | vmcs_write32(GUEST_LDTR_LIMIT, 0xffff); | |
4984 | vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082); | |
4985 | ||
d28bc9dd NA |
4986 | if (!init_event) { |
4987 | vmcs_write32(GUEST_SYSENTER_CS, 0); | |
4988 | vmcs_writel(GUEST_SYSENTER_ESP, 0); | |
4989 | vmcs_writel(GUEST_SYSENTER_EIP, 0); | |
4990 | vmcs_write64(GUEST_IA32_DEBUGCTL, 0); | |
4991 | } | |
e00c8cf2 AK |
4992 | |
4993 | vmcs_writel(GUEST_RFLAGS, 0x02); | |
66450a21 | 4994 | kvm_rip_write(vcpu, 0xfff0); |
e00c8cf2 | 4995 | |
e00c8cf2 AK |
4996 | vmcs_writel(GUEST_GDTR_BASE, 0); |
4997 | vmcs_write32(GUEST_GDTR_LIMIT, 0xffff); | |
4998 | ||
4999 | vmcs_writel(GUEST_IDTR_BASE, 0); | |
5000 | vmcs_write32(GUEST_IDTR_LIMIT, 0xffff); | |
5001 | ||
443381a8 | 5002 | vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE); |
e00c8cf2 | 5003 | vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0); |
f3531054 | 5004 | vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0); |
e00c8cf2 | 5005 | |
e00c8cf2 AK |
5006 | setup_msrs(vmx); |
5007 | ||
6aa8b732 AK |
5008 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */ |
5009 | ||
d28bc9dd | 5010 | if (cpu_has_vmx_tpr_shadow() && !init_event) { |
f78e0e2e | 5011 | vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0); |
35754c98 | 5012 | if (cpu_need_tpr_shadow(vcpu)) |
f78e0e2e | 5013 | vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, |
d28bc9dd | 5014 | __pa(vcpu->arch.apic->regs)); |
f78e0e2e SY |
5015 | vmcs_write32(TPR_THRESHOLD, 0); |
5016 | } | |
5017 | ||
a73896cb | 5018 | kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu); |
6aa8b732 | 5019 | |
d62caabb | 5020 | if (kvm_vcpu_apicv_active(vcpu)) |
01e439be YZ |
5021 | memset(&vmx->pi_desc, 0, sizeof(struct pi_desc)); |
5022 | ||
2384d2b3 SY |
5023 | if (vmx->vpid != 0) |
5024 | vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid); | |
5025 | ||
d28bc9dd NA |
5026 | cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET; |
5027 | vmx_set_cr0(vcpu, cr0); /* enter rmode */ | |
5028 | vmx->vcpu.arch.cr0 = cr0; | |
5029 | vmx_set_cr4(vcpu, 0); | |
5690891b | 5030 | vmx_set_efer(vcpu, 0); |
d28bc9dd NA |
5031 | vmx_fpu_activate(vcpu); |
5032 | update_exception_bitmap(vcpu); | |
6aa8b732 | 5033 | |
dd5f5341 | 5034 | vpid_sync_context(vmx->vpid); |
6aa8b732 AK |
5035 | } |
5036 | ||
b6f1250e NHE |
5037 | /* |
5038 | * In nested virtualization, check if L1 asked to exit on external interrupts. | |
5039 | * For most existing hypervisors, this will always return true. | |
5040 | */ | |
5041 | static bool nested_exit_on_intr(struct kvm_vcpu *vcpu) | |
5042 | { | |
5043 | return get_vmcs12(vcpu)->pin_based_vm_exec_control & | |
5044 | PIN_BASED_EXT_INTR_MASK; | |
5045 | } | |
5046 | ||
77b0f5d6 BD |
5047 | /* |
5048 | * In nested virtualization, check if L1 has set | |
5049 | * VM_EXIT_ACK_INTR_ON_EXIT | |
5050 | */ | |
5051 | static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu) | |
5052 | { | |
5053 | return get_vmcs12(vcpu)->vm_exit_controls & | |
5054 | VM_EXIT_ACK_INTR_ON_EXIT; | |
5055 | } | |
5056 | ||
ea8ceb83 JK |
5057 | static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu) |
5058 | { | |
5059 | return get_vmcs12(vcpu)->pin_based_vm_exec_control & | |
5060 | PIN_BASED_NMI_EXITING; | |
5061 | } | |
5062 | ||
c9a7953f | 5063 | static void enable_irq_window(struct kvm_vcpu *vcpu) |
3b86cd99 JK |
5064 | { |
5065 | u32 cpu_based_vm_exec_control; | |
730dca42 | 5066 | |
3b86cd99 JK |
5067 | cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); |
5068 | cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING; | |
5069 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); | |
5070 | } | |
5071 | ||
c9a7953f | 5072 | static void enable_nmi_window(struct kvm_vcpu *vcpu) |
3b86cd99 JK |
5073 | { |
5074 | u32 cpu_based_vm_exec_control; | |
5075 | ||
c9a7953f JK |
5076 | if (!cpu_has_virtual_nmis() || |
5077 | vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) { | |
5078 | enable_irq_window(vcpu); | |
5079 | return; | |
5080 | } | |
3b86cd99 JK |
5081 | |
5082 | cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); | |
5083 | cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING; | |
5084 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); | |
5085 | } | |
5086 | ||
66fd3f7f | 5087 | static void vmx_inject_irq(struct kvm_vcpu *vcpu) |
85f455f7 | 5088 | { |
9c8cba37 | 5089 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
66fd3f7f GN |
5090 | uint32_t intr; |
5091 | int irq = vcpu->arch.interrupt.nr; | |
9c8cba37 | 5092 | |
229456fc | 5093 | trace_kvm_inj_virq(irq); |
2714d1d3 | 5094 | |
fa89a817 | 5095 | ++vcpu->stat.irq_injections; |
7ffd92c5 | 5096 | if (vmx->rmode.vm86_active) { |
71f9833b SH |
5097 | int inc_eip = 0; |
5098 | if (vcpu->arch.interrupt.soft) | |
5099 | inc_eip = vcpu->arch.event_exit_inst_len; | |
5100 | if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE) | |
a92601bb | 5101 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
85f455f7 ED |
5102 | return; |
5103 | } | |
66fd3f7f GN |
5104 | intr = irq | INTR_INFO_VALID_MASK; |
5105 | if (vcpu->arch.interrupt.soft) { | |
5106 | intr |= INTR_TYPE_SOFT_INTR; | |
5107 | vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, | |
5108 | vmx->vcpu.arch.event_exit_inst_len); | |
5109 | } else | |
5110 | intr |= INTR_TYPE_EXT_INTR; | |
5111 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr); | |
85f455f7 ED |
5112 | } |
5113 | ||
f08864b4 SY |
5114 | static void vmx_inject_nmi(struct kvm_vcpu *vcpu) |
5115 | { | |
66a5a347 JK |
5116 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
5117 | ||
0b6ac343 NHE |
5118 | if (is_guest_mode(vcpu)) |
5119 | return; | |
5120 | ||
3b86cd99 JK |
5121 | if (!cpu_has_virtual_nmis()) { |
5122 | /* | |
5123 | * Tracking the NMI-blocked state in software is built upon | |
5124 | * finding the next open IRQ window. This, in turn, depends on | |
5125 | * well-behaving guests: They have to keep IRQs disabled at | |
5126 | * least as long as the NMI handler runs. Otherwise we may | |
5127 | * cause NMI nesting, maybe breaking the guest. But as this is | |
5128 | * highly unlikely, we can live with the residual risk. | |
5129 | */ | |
5130 | vmx->soft_vnmi_blocked = 1; | |
5131 | vmx->vnmi_blocked_time = 0; | |
5132 | } | |
5133 | ||
487b391d | 5134 | ++vcpu->stat.nmi_injections; |
9d58b931 | 5135 | vmx->nmi_known_unmasked = false; |
7ffd92c5 | 5136 | if (vmx->rmode.vm86_active) { |
71f9833b | 5137 | if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE) |
a92601bb | 5138 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
66a5a347 JK |
5139 | return; |
5140 | } | |
f08864b4 SY |
5141 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, |
5142 | INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR); | |
f08864b4 SY |
5143 | } |
5144 | ||
3cfc3092 JK |
5145 | static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu) |
5146 | { | |
5147 | if (!cpu_has_virtual_nmis()) | |
5148 | return to_vmx(vcpu)->soft_vnmi_blocked; | |
9d58b931 AK |
5149 | if (to_vmx(vcpu)->nmi_known_unmasked) |
5150 | return false; | |
c332c83a | 5151 | return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI; |
3cfc3092 JK |
5152 | } |
5153 | ||
5154 | static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked) | |
5155 | { | |
5156 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
5157 | ||
5158 | if (!cpu_has_virtual_nmis()) { | |
5159 | if (vmx->soft_vnmi_blocked != masked) { | |
5160 | vmx->soft_vnmi_blocked = masked; | |
5161 | vmx->vnmi_blocked_time = 0; | |
5162 | } | |
5163 | } else { | |
9d58b931 | 5164 | vmx->nmi_known_unmasked = !masked; |
3cfc3092 JK |
5165 | if (masked) |
5166 | vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, | |
5167 | GUEST_INTR_STATE_NMI); | |
5168 | else | |
5169 | vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO, | |
5170 | GUEST_INTR_STATE_NMI); | |
5171 | } | |
5172 | } | |
5173 | ||
2505dc9f JK |
5174 | static int vmx_nmi_allowed(struct kvm_vcpu *vcpu) |
5175 | { | |
b6b8a145 JK |
5176 | if (to_vmx(vcpu)->nested.nested_run_pending) |
5177 | return 0; | |
ea8ceb83 | 5178 | |
2505dc9f JK |
5179 | if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked) |
5180 | return 0; | |
5181 | ||
5182 | return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & | |
5183 | (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI | |
5184 | | GUEST_INTR_STATE_NMI)); | |
5185 | } | |
5186 | ||
78646121 GN |
5187 | static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu) |
5188 | { | |
b6b8a145 JK |
5189 | return (!to_vmx(vcpu)->nested.nested_run_pending && |
5190 | vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) && | |
c4282df9 GN |
5191 | !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & |
5192 | (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS)); | |
78646121 GN |
5193 | } |
5194 | ||
cbc94022 IE |
5195 | static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr) |
5196 | { | |
5197 | int ret; | |
cbc94022 | 5198 | |
1d8007bd PB |
5199 | ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr, |
5200 | PAGE_SIZE * 3); | |
cbc94022 IE |
5201 | if (ret) |
5202 | return ret; | |
bfc6d222 | 5203 | kvm->arch.tss_addr = addr; |
1f755a82 | 5204 | return init_rmode_tss(kvm); |
cbc94022 IE |
5205 | } |
5206 | ||
0ca1b4f4 | 5207 | static bool rmode_exception(struct kvm_vcpu *vcpu, int vec) |
6aa8b732 | 5208 | { |
77ab6db0 | 5209 | switch (vec) { |
77ab6db0 | 5210 | case BP_VECTOR: |
c573cd22 JK |
5211 | /* |
5212 | * Update instruction length as we may reinject the exception | |
5213 | * from user space while in guest debugging mode. | |
5214 | */ | |
5215 | to_vmx(vcpu)->vcpu.arch.event_exit_inst_len = | |
5216 | vmcs_read32(VM_EXIT_INSTRUCTION_LEN); | |
d0bfb940 | 5217 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) |
0ca1b4f4 GN |
5218 | return false; |
5219 | /* fall through */ | |
5220 | case DB_VECTOR: | |
5221 | if (vcpu->guest_debug & | |
5222 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) | |
5223 | return false; | |
d0bfb940 JK |
5224 | /* fall through */ |
5225 | case DE_VECTOR: | |
77ab6db0 JK |
5226 | case OF_VECTOR: |
5227 | case BR_VECTOR: | |
5228 | case UD_VECTOR: | |
5229 | case DF_VECTOR: | |
5230 | case SS_VECTOR: | |
5231 | case GP_VECTOR: | |
5232 | case MF_VECTOR: | |
0ca1b4f4 GN |
5233 | return true; |
5234 | break; | |
77ab6db0 | 5235 | } |
0ca1b4f4 GN |
5236 | return false; |
5237 | } | |
5238 | ||
5239 | static int handle_rmode_exception(struct kvm_vcpu *vcpu, | |
5240 | int vec, u32 err_code) | |
5241 | { | |
5242 | /* | |
5243 | * Instruction with address size override prefix opcode 0x67 | |
5244 | * Cause the #SS fault with 0 error code in VM86 mode. | |
5245 | */ | |
5246 | if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) { | |
5247 | if (emulate_instruction(vcpu, 0) == EMULATE_DONE) { | |
5248 | if (vcpu->arch.halt_request) { | |
5249 | vcpu->arch.halt_request = 0; | |
5cb56059 | 5250 | return kvm_vcpu_halt(vcpu); |
0ca1b4f4 GN |
5251 | } |
5252 | return 1; | |
5253 | } | |
5254 | return 0; | |
5255 | } | |
5256 | ||
5257 | /* | |
5258 | * Forward all other exceptions that are valid in real mode. | |
5259 | * FIXME: Breaks guest debugging in real mode, needs to be fixed with | |
5260 | * the required debugging infrastructure rework. | |
5261 | */ | |
5262 | kvm_queue_exception(vcpu, vec); | |
5263 | return 1; | |
6aa8b732 AK |
5264 | } |
5265 | ||
a0861c02 AK |
5266 | /* |
5267 | * Trigger machine check on the host. We assume all the MSRs are already set up | |
5268 | * by the CPU and that we still run on the same CPU as the MCE occurred on. | |
5269 | * We pass a fake environment to the machine check handler because we want | |
5270 | * the guest to be always treated like user space, no matter what context | |
5271 | * it used internally. | |
5272 | */ | |
5273 | static void kvm_machine_check(void) | |
5274 | { | |
5275 | #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64) | |
5276 | struct pt_regs regs = { | |
5277 | .cs = 3, /* Fake ring 3 no matter what the guest ran on */ | |
5278 | .flags = X86_EFLAGS_IF, | |
5279 | }; | |
5280 | ||
5281 | do_machine_check(®s, 0); | |
5282 | #endif | |
5283 | } | |
5284 | ||
851ba692 | 5285 | static int handle_machine_check(struct kvm_vcpu *vcpu) |
a0861c02 AK |
5286 | { |
5287 | /* already handled by vcpu_run */ | |
5288 | return 1; | |
5289 | } | |
5290 | ||
851ba692 | 5291 | static int handle_exception(struct kvm_vcpu *vcpu) |
6aa8b732 | 5292 | { |
1155f76a | 5293 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
851ba692 | 5294 | struct kvm_run *kvm_run = vcpu->run; |
d0bfb940 | 5295 | u32 intr_info, ex_no, error_code; |
42dbaa5a | 5296 | unsigned long cr2, rip, dr6; |
6aa8b732 AK |
5297 | u32 vect_info; |
5298 | enum emulation_result er; | |
5299 | ||
1155f76a | 5300 | vect_info = vmx->idt_vectoring_info; |
88786475 | 5301 | intr_info = vmx->exit_intr_info; |
6aa8b732 | 5302 | |
a0861c02 | 5303 | if (is_machine_check(intr_info)) |
851ba692 | 5304 | return handle_machine_check(vcpu); |
a0861c02 | 5305 | |
e4a41889 | 5306 | if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR) |
1b6269db | 5307 | return 1; /* already handled by vmx_vcpu_run() */ |
2ab455cc AL |
5308 | |
5309 | if (is_no_device(intr_info)) { | |
5fd86fcf | 5310 | vmx_fpu_activate(vcpu); |
2ab455cc AL |
5311 | return 1; |
5312 | } | |
5313 | ||
7aa81cc0 | 5314 | if (is_invalid_opcode(intr_info)) { |
ae1f5767 JK |
5315 | if (is_guest_mode(vcpu)) { |
5316 | kvm_queue_exception(vcpu, UD_VECTOR); | |
5317 | return 1; | |
5318 | } | |
51d8b661 | 5319 | er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD); |
7aa81cc0 | 5320 | if (er != EMULATE_DONE) |
7ee5d940 | 5321 | kvm_queue_exception(vcpu, UD_VECTOR); |
7aa81cc0 AL |
5322 | return 1; |
5323 | } | |
5324 | ||
6aa8b732 | 5325 | error_code = 0; |
2e11384c | 5326 | if (intr_info & INTR_INFO_DELIVER_CODE_MASK) |
6aa8b732 | 5327 | error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE); |
bf4ca23e XG |
5328 | |
5329 | /* | |
5330 | * The #PF with PFEC.RSVD = 1 indicates the guest is accessing | |
5331 | * MMIO, it is better to report an internal error. | |
5332 | * See the comments in vmx_handle_exit. | |
5333 | */ | |
5334 | if ((vect_info & VECTORING_INFO_VALID_MASK) && | |
5335 | !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) { | |
5336 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
5337 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX; | |
80f0e95d | 5338 | vcpu->run->internal.ndata = 3; |
bf4ca23e XG |
5339 | vcpu->run->internal.data[0] = vect_info; |
5340 | vcpu->run->internal.data[1] = intr_info; | |
80f0e95d | 5341 | vcpu->run->internal.data[2] = error_code; |
bf4ca23e XG |
5342 | return 0; |
5343 | } | |
5344 | ||
6aa8b732 | 5345 | if (is_page_fault(intr_info)) { |
1439442c | 5346 | /* EPT won't cause page fault directly */ |
cf3ace79 | 5347 | BUG_ON(enable_ept); |
6aa8b732 | 5348 | cr2 = vmcs_readl(EXIT_QUALIFICATION); |
229456fc MT |
5349 | trace_kvm_page_fault(cr2, error_code); |
5350 | ||
3298b75c | 5351 | if (kvm_event_needs_reinjection(vcpu)) |
577bdc49 | 5352 | kvm_mmu_unprotect_page_virt(vcpu, cr2); |
dc25e89e | 5353 | return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0); |
6aa8b732 AK |
5354 | } |
5355 | ||
d0bfb940 | 5356 | ex_no = intr_info & INTR_INFO_VECTOR_MASK; |
0ca1b4f4 GN |
5357 | |
5358 | if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no)) | |
5359 | return handle_rmode_exception(vcpu, ex_no, error_code); | |
5360 | ||
42dbaa5a | 5361 | switch (ex_no) { |
54a20552 EN |
5362 | case AC_VECTOR: |
5363 | kvm_queue_exception_e(vcpu, AC_VECTOR, error_code); | |
5364 | return 1; | |
42dbaa5a JK |
5365 | case DB_VECTOR: |
5366 | dr6 = vmcs_readl(EXIT_QUALIFICATION); | |
5367 | if (!(vcpu->guest_debug & | |
5368 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) { | |
8246bf52 | 5369 | vcpu->arch.dr6 &= ~15; |
6f43ed01 | 5370 | vcpu->arch.dr6 |= dr6 | DR6_RTM; |
fd2a445a HD |
5371 | if (!(dr6 & ~DR6_RESERVED)) /* icebp */ |
5372 | skip_emulated_instruction(vcpu); | |
5373 | ||
42dbaa5a JK |
5374 | kvm_queue_exception(vcpu, DB_VECTOR); |
5375 | return 1; | |
5376 | } | |
5377 | kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1; | |
5378 | kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7); | |
5379 | /* fall through */ | |
5380 | case BP_VECTOR: | |
c573cd22 JK |
5381 | /* |
5382 | * Update instruction length as we may reinject #BP from | |
5383 | * user space while in guest debugging mode. Reading it for | |
5384 | * #DB as well causes no harm, it is not used in that case. | |
5385 | */ | |
5386 | vmx->vcpu.arch.event_exit_inst_len = | |
5387 | vmcs_read32(VM_EXIT_INSTRUCTION_LEN); | |
6aa8b732 | 5388 | kvm_run->exit_reason = KVM_EXIT_DEBUG; |
0a434bb2 | 5389 | rip = kvm_rip_read(vcpu); |
d0bfb940 JK |
5390 | kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip; |
5391 | kvm_run->debug.arch.exception = ex_no; | |
42dbaa5a JK |
5392 | break; |
5393 | default: | |
d0bfb940 JK |
5394 | kvm_run->exit_reason = KVM_EXIT_EXCEPTION; |
5395 | kvm_run->ex.exception = ex_no; | |
5396 | kvm_run->ex.error_code = error_code; | |
42dbaa5a | 5397 | break; |
6aa8b732 | 5398 | } |
6aa8b732 AK |
5399 | return 0; |
5400 | } | |
5401 | ||
851ba692 | 5402 | static int handle_external_interrupt(struct kvm_vcpu *vcpu) |
6aa8b732 | 5403 | { |
1165f5fe | 5404 | ++vcpu->stat.irq_exits; |
6aa8b732 AK |
5405 | return 1; |
5406 | } | |
5407 | ||
851ba692 | 5408 | static int handle_triple_fault(struct kvm_vcpu *vcpu) |
988ad74f | 5409 | { |
851ba692 | 5410 | vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN; |
988ad74f AK |
5411 | return 0; |
5412 | } | |
6aa8b732 | 5413 | |
851ba692 | 5414 | static int handle_io(struct kvm_vcpu *vcpu) |
6aa8b732 | 5415 | { |
bfdaab09 | 5416 | unsigned long exit_qualification; |
34c33d16 | 5417 | int size, in, string; |
039576c0 | 5418 | unsigned port; |
6aa8b732 | 5419 | |
bfdaab09 | 5420 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
039576c0 | 5421 | string = (exit_qualification & 16) != 0; |
cf8f70bf | 5422 | in = (exit_qualification & 8) != 0; |
e70669ab | 5423 | |
cf8f70bf | 5424 | ++vcpu->stat.io_exits; |
e70669ab | 5425 | |
cf8f70bf | 5426 | if (string || in) |
51d8b661 | 5427 | return emulate_instruction(vcpu, 0) == EMULATE_DONE; |
e70669ab | 5428 | |
cf8f70bf GN |
5429 | port = exit_qualification >> 16; |
5430 | size = (exit_qualification & 7) + 1; | |
e93f36bc | 5431 | skip_emulated_instruction(vcpu); |
cf8f70bf GN |
5432 | |
5433 | return kvm_fast_pio_out(vcpu, size, port); | |
6aa8b732 AK |
5434 | } |
5435 | ||
102d8325 IM |
5436 | static void |
5437 | vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall) | |
5438 | { | |
5439 | /* | |
5440 | * Patch in the VMCALL instruction: | |
5441 | */ | |
5442 | hypercall[0] = 0x0f; | |
5443 | hypercall[1] = 0x01; | |
5444 | hypercall[2] = 0xc1; | |
102d8325 IM |
5445 | } |
5446 | ||
b9c237bb | 5447 | static bool nested_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val) |
92fbc7b1 JK |
5448 | { |
5449 | unsigned long always_on = VMXON_CR0_ALWAYSON; | |
b9c237bb | 5450 | struct vmcs12 *vmcs12 = get_vmcs12(vcpu); |
92fbc7b1 | 5451 | |
b9c237bb | 5452 | if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high & |
92fbc7b1 JK |
5453 | SECONDARY_EXEC_UNRESTRICTED_GUEST && |
5454 | nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST)) | |
5455 | always_on &= ~(X86_CR0_PE | X86_CR0_PG); | |
5456 | return (val & always_on) == always_on; | |
5457 | } | |
5458 | ||
0fa06071 | 5459 | /* called to set cr0 as appropriate for a mov-to-cr0 exit. */ |
eeadf9e7 NHE |
5460 | static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val) |
5461 | { | |
eeadf9e7 | 5462 | if (is_guest_mode(vcpu)) { |
1a0d74e6 JK |
5463 | struct vmcs12 *vmcs12 = get_vmcs12(vcpu); |
5464 | unsigned long orig_val = val; | |
5465 | ||
eeadf9e7 NHE |
5466 | /* |
5467 | * We get here when L2 changed cr0 in a way that did not change | |
5468 | * any of L1's shadowed bits (see nested_vmx_exit_handled_cr), | |
1a0d74e6 JK |
5469 | * but did change L0 shadowed bits. So we first calculate the |
5470 | * effective cr0 value that L1 would like to write into the | |
5471 | * hardware. It consists of the L2-owned bits from the new | |
5472 | * value combined with the L1-owned bits from L1's guest_cr0. | |
eeadf9e7 | 5473 | */ |
1a0d74e6 JK |
5474 | val = (val & ~vmcs12->cr0_guest_host_mask) | |
5475 | (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask); | |
5476 | ||
b9c237bb | 5477 | if (!nested_cr0_valid(vcpu, val)) |
eeadf9e7 | 5478 | return 1; |
1a0d74e6 JK |
5479 | |
5480 | if (kvm_set_cr0(vcpu, val)) | |
5481 | return 1; | |
5482 | vmcs_writel(CR0_READ_SHADOW, orig_val); | |
eeadf9e7 | 5483 | return 0; |
1a0d74e6 JK |
5484 | } else { |
5485 | if (to_vmx(vcpu)->nested.vmxon && | |
5486 | ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON)) | |
5487 | return 1; | |
eeadf9e7 | 5488 | return kvm_set_cr0(vcpu, val); |
1a0d74e6 | 5489 | } |
eeadf9e7 NHE |
5490 | } |
5491 | ||
5492 | static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val) | |
5493 | { | |
5494 | if (is_guest_mode(vcpu)) { | |
1a0d74e6 JK |
5495 | struct vmcs12 *vmcs12 = get_vmcs12(vcpu); |
5496 | unsigned long orig_val = val; | |
5497 | ||
5498 | /* analogously to handle_set_cr0 */ | |
5499 | val = (val & ~vmcs12->cr4_guest_host_mask) | | |
5500 | (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask); | |
5501 | if (kvm_set_cr4(vcpu, val)) | |
eeadf9e7 | 5502 | return 1; |
1a0d74e6 | 5503 | vmcs_writel(CR4_READ_SHADOW, orig_val); |
eeadf9e7 NHE |
5504 | return 0; |
5505 | } else | |
5506 | return kvm_set_cr4(vcpu, val); | |
5507 | } | |
5508 | ||
5509 | /* called to set cr0 as approriate for clts instruction exit. */ | |
5510 | static void handle_clts(struct kvm_vcpu *vcpu) | |
5511 | { | |
5512 | if (is_guest_mode(vcpu)) { | |
5513 | /* | |
5514 | * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS | |
5515 | * but we did (!fpu_active). We need to keep GUEST_CR0.TS on, | |
5516 | * just pretend it's off (also in arch.cr0 for fpu_activate). | |
5517 | */ | |
5518 | vmcs_writel(CR0_READ_SHADOW, | |
5519 | vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS); | |
5520 | vcpu->arch.cr0 &= ~X86_CR0_TS; | |
5521 | } else | |
5522 | vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS)); | |
5523 | } | |
5524 | ||
851ba692 | 5525 | static int handle_cr(struct kvm_vcpu *vcpu) |
6aa8b732 | 5526 | { |
229456fc | 5527 | unsigned long exit_qualification, val; |
6aa8b732 AK |
5528 | int cr; |
5529 | int reg; | |
49a9b07e | 5530 | int err; |
6aa8b732 | 5531 | |
bfdaab09 | 5532 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
6aa8b732 AK |
5533 | cr = exit_qualification & 15; |
5534 | reg = (exit_qualification >> 8) & 15; | |
5535 | switch ((exit_qualification >> 4) & 3) { | |
5536 | case 0: /* mov to cr */ | |
1e32c079 | 5537 | val = kvm_register_readl(vcpu, reg); |
229456fc | 5538 | trace_kvm_cr_write(cr, val); |
6aa8b732 AK |
5539 | switch (cr) { |
5540 | case 0: | |
eeadf9e7 | 5541 | err = handle_set_cr0(vcpu, val); |
db8fcefa | 5542 | kvm_complete_insn_gp(vcpu, err); |
6aa8b732 AK |
5543 | return 1; |
5544 | case 3: | |
2390218b | 5545 | err = kvm_set_cr3(vcpu, val); |
db8fcefa | 5546 | kvm_complete_insn_gp(vcpu, err); |
6aa8b732 AK |
5547 | return 1; |
5548 | case 4: | |
eeadf9e7 | 5549 | err = handle_set_cr4(vcpu, val); |
db8fcefa | 5550 | kvm_complete_insn_gp(vcpu, err); |
6aa8b732 | 5551 | return 1; |
0a5fff19 GN |
5552 | case 8: { |
5553 | u8 cr8_prev = kvm_get_cr8(vcpu); | |
1e32c079 | 5554 | u8 cr8 = (u8)val; |
eea1cff9 | 5555 | err = kvm_set_cr8(vcpu, cr8); |
db8fcefa | 5556 | kvm_complete_insn_gp(vcpu, err); |
35754c98 | 5557 | if (lapic_in_kernel(vcpu)) |
0a5fff19 GN |
5558 | return 1; |
5559 | if (cr8_prev <= cr8) | |
5560 | return 1; | |
851ba692 | 5561 | vcpu->run->exit_reason = KVM_EXIT_SET_TPR; |
0a5fff19 GN |
5562 | return 0; |
5563 | } | |
4b8073e4 | 5564 | } |
6aa8b732 | 5565 | break; |
25c4c276 | 5566 | case 2: /* clts */ |
eeadf9e7 | 5567 | handle_clts(vcpu); |
4d4ec087 | 5568 | trace_kvm_cr_write(0, kvm_read_cr0(vcpu)); |
25c4c276 | 5569 | skip_emulated_instruction(vcpu); |
6b52d186 | 5570 | vmx_fpu_activate(vcpu); |
25c4c276 | 5571 | return 1; |
6aa8b732 AK |
5572 | case 1: /*mov from cr*/ |
5573 | switch (cr) { | |
5574 | case 3: | |
9f8fe504 AK |
5575 | val = kvm_read_cr3(vcpu); |
5576 | kvm_register_write(vcpu, reg, val); | |
5577 | trace_kvm_cr_read(cr, val); | |
6aa8b732 AK |
5578 | skip_emulated_instruction(vcpu); |
5579 | return 1; | |
5580 | case 8: | |
229456fc MT |
5581 | val = kvm_get_cr8(vcpu); |
5582 | kvm_register_write(vcpu, reg, val); | |
5583 | trace_kvm_cr_read(cr, val); | |
6aa8b732 AK |
5584 | skip_emulated_instruction(vcpu); |
5585 | return 1; | |
5586 | } | |
5587 | break; | |
5588 | case 3: /* lmsw */ | |
a1f83a74 | 5589 | val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f; |
4d4ec087 | 5590 | trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val); |
a1f83a74 | 5591 | kvm_lmsw(vcpu, val); |
6aa8b732 AK |
5592 | |
5593 | skip_emulated_instruction(vcpu); | |
5594 | return 1; | |
5595 | default: | |
5596 | break; | |
5597 | } | |
851ba692 | 5598 | vcpu->run->exit_reason = 0; |
a737f256 | 5599 | vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n", |
6aa8b732 AK |
5600 | (int)(exit_qualification >> 4) & 3, cr); |
5601 | return 0; | |
5602 | } | |
5603 | ||
851ba692 | 5604 | static int handle_dr(struct kvm_vcpu *vcpu) |
6aa8b732 | 5605 | { |
bfdaab09 | 5606 | unsigned long exit_qualification; |
16f8a6f9 NA |
5607 | int dr, dr7, reg; |
5608 | ||
5609 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); | |
5610 | dr = exit_qualification & DEBUG_REG_ACCESS_NUM; | |
5611 | ||
5612 | /* First, if DR does not exist, trigger UD */ | |
5613 | if (!kvm_require_dr(vcpu, dr)) | |
5614 | return 1; | |
6aa8b732 | 5615 | |
f2483415 | 5616 | /* Do not handle if the CPL > 0, will trigger GP on re-entry */ |
0a79b009 AK |
5617 | if (!kvm_require_cpl(vcpu, 0)) |
5618 | return 1; | |
16f8a6f9 NA |
5619 | dr7 = vmcs_readl(GUEST_DR7); |
5620 | if (dr7 & DR7_GD) { | |
42dbaa5a JK |
5621 | /* |
5622 | * As the vm-exit takes precedence over the debug trap, we | |
5623 | * need to emulate the latter, either for the host or the | |
5624 | * guest debugging itself. | |
5625 | */ | |
5626 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) { | |
851ba692 | 5627 | vcpu->run->debug.arch.dr6 = vcpu->arch.dr6; |
16f8a6f9 | 5628 | vcpu->run->debug.arch.dr7 = dr7; |
82b32774 | 5629 | vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu); |
851ba692 AK |
5630 | vcpu->run->debug.arch.exception = DB_VECTOR; |
5631 | vcpu->run->exit_reason = KVM_EXIT_DEBUG; | |
42dbaa5a JK |
5632 | return 0; |
5633 | } else { | |
7305eb5d | 5634 | vcpu->arch.dr6 &= ~15; |
6f43ed01 | 5635 | vcpu->arch.dr6 |= DR6_BD | DR6_RTM; |
42dbaa5a JK |
5636 | kvm_queue_exception(vcpu, DB_VECTOR); |
5637 | return 1; | |
5638 | } | |
5639 | } | |
5640 | ||
81908bf4 | 5641 | if (vcpu->guest_debug == 0) { |
8f22372f PB |
5642 | vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL, |
5643 | CPU_BASED_MOV_DR_EXITING); | |
81908bf4 PB |
5644 | |
5645 | /* | |
5646 | * No more DR vmexits; force a reload of the debug registers | |
5647 | * and reenter on this instruction. The next vmexit will | |
5648 | * retrieve the full state of the debug registers. | |
5649 | */ | |
5650 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT; | |
5651 | return 1; | |
5652 | } | |
5653 | ||
42dbaa5a JK |
5654 | reg = DEBUG_REG_ACCESS_REG(exit_qualification); |
5655 | if (exit_qualification & TYPE_MOV_FROM_DR) { | |
020df079 | 5656 | unsigned long val; |
4c4d563b JK |
5657 | |
5658 | if (kvm_get_dr(vcpu, dr, &val)) | |
5659 | return 1; | |
5660 | kvm_register_write(vcpu, reg, val); | |
020df079 | 5661 | } else |
5777392e | 5662 | if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg))) |
4c4d563b JK |
5663 | return 1; |
5664 | ||
6aa8b732 AK |
5665 | skip_emulated_instruction(vcpu); |
5666 | return 1; | |
5667 | } | |
5668 | ||
73aaf249 JK |
5669 | static u64 vmx_get_dr6(struct kvm_vcpu *vcpu) |
5670 | { | |
5671 | return vcpu->arch.dr6; | |
5672 | } | |
5673 | ||
5674 | static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val) | |
5675 | { | |
5676 | } | |
5677 | ||
81908bf4 PB |
5678 | static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu) |
5679 | { | |
81908bf4 PB |
5680 | get_debugreg(vcpu->arch.db[0], 0); |
5681 | get_debugreg(vcpu->arch.db[1], 1); | |
5682 | get_debugreg(vcpu->arch.db[2], 2); | |
5683 | get_debugreg(vcpu->arch.db[3], 3); | |
5684 | get_debugreg(vcpu->arch.dr6, 6); | |
5685 | vcpu->arch.dr7 = vmcs_readl(GUEST_DR7); | |
5686 | ||
5687 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT; | |
8f22372f | 5688 | vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING); |
81908bf4 PB |
5689 | } |
5690 | ||
020df079 GN |
5691 | static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val) |
5692 | { | |
5693 | vmcs_writel(GUEST_DR7, val); | |
5694 | } | |
5695 | ||
851ba692 | 5696 | static int handle_cpuid(struct kvm_vcpu *vcpu) |
6aa8b732 | 5697 | { |
06465c5a AK |
5698 | kvm_emulate_cpuid(vcpu); |
5699 | return 1; | |
6aa8b732 AK |
5700 | } |
5701 | ||
851ba692 | 5702 | static int handle_rdmsr(struct kvm_vcpu *vcpu) |
6aa8b732 | 5703 | { |
ad312c7c | 5704 | u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX]; |
609e36d3 | 5705 | struct msr_data msr_info; |
6aa8b732 | 5706 | |
609e36d3 PB |
5707 | msr_info.index = ecx; |
5708 | msr_info.host_initiated = false; | |
5709 | if (vmx_get_msr(vcpu, &msr_info)) { | |
59200273 | 5710 | trace_kvm_msr_read_ex(ecx); |
c1a5d4f9 | 5711 | kvm_inject_gp(vcpu, 0); |
6aa8b732 AK |
5712 | return 1; |
5713 | } | |
5714 | ||
609e36d3 | 5715 | trace_kvm_msr_read(ecx, msr_info.data); |
2714d1d3 | 5716 | |
6aa8b732 | 5717 | /* FIXME: handling of bits 32:63 of rax, rdx */ |
609e36d3 PB |
5718 | vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u; |
5719 | vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u; | |
6aa8b732 AK |
5720 | skip_emulated_instruction(vcpu); |
5721 | return 1; | |
5722 | } | |
5723 | ||
851ba692 | 5724 | static int handle_wrmsr(struct kvm_vcpu *vcpu) |
6aa8b732 | 5725 | { |
8fe8ab46 | 5726 | struct msr_data msr; |
ad312c7c ZX |
5727 | u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX]; |
5728 | u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u) | |
5729 | | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32); | |
6aa8b732 | 5730 | |
8fe8ab46 WA |
5731 | msr.data = data; |
5732 | msr.index = ecx; | |
5733 | msr.host_initiated = false; | |
854e8bb1 | 5734 | if (kvm_set_msr(vcpu, &msr) != 0) { |
59200273 | 5735 | trace_kvm_msr_write_ex(ecx, data); |
c1a5d4f9 | 5736 | kvm_inject_gp(vcpu, 0); |
6aa8b732 AK |
5737 | return 1; |
5738 | } | |
5739 | ||
59200273 | 5740 | trace_kvm_msr_write(ecx, data); |
6aa8b732 AK |
5741 | skip_emulated_instruction(vcpu); |
5742 | return 1; | |
5743 | } | |
5744 | ||
851ba692 | 5745 | static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu) |
6e5d865c | 5746 | { |
3842d135 | 5747 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
6e5d865c YS |
5748 | return 1; |
5749 | } | |
5750 | ||
851ba692 | 5751 | static int handle_interrupt_window(struct kvm_vcpu *vcpu) |
6aa8b732 | 5752 | { |
85f455f7 ED |
5753 | u32 cpu_based_vm_exec_control; |
5754 | ||
5755 | /* clear pending irq */ | |
5756 | cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); | |
5757 | cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING; | |
5758 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); | |
2714d1d3 | 5759 | |
3842d135 AK |
5760 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
5761 | ||
a26bf12a | 5762 | ++vcpu->stat.irq_window_exits; |
6aa8b732 AK |
5763 | return 1; |
5764 | } | |
5765 | ||
851ba692 | 5766 | static int handle_halt(struct kvm_vcpu *vcpu) |
6aa8b732 | 5767 | { |
d3bef15f | 5768 | return kvm_emulate_halt(vcpu); |
6aa8b732 AK |
5769 | } |
5770 | ||
851ba692 | 5771 | static int handle_vmcall(struct kvm_vcpu *vcpu) |
c21415e8 | 5772 | { |
0d9c055e | 5773 | return kvm_emulate_hypercall(vcpu); |
c21415e8 IM |
5774 | } |
5775 | ||
ec25d5e6 GN |
5776 | static int handle_invd(struct kvm_vcpu *vcpu) |
5777 | { | |
51d8b661 | 5778 | return emulate_instruction(vcpu, 0) == EMULATE_DONE; |
ec25d5e6 GN |
5779 | } |
5780 | ||
851ba692 | 5781 | static int handle_invlpg(struct kvm_vcpu *vcpu) |
a7052897 | 5782 | { |
f9c617f6 | 5783 | unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
a7052897 MT |
5784 | |
5785 | kvm_mmu_invlpg(vcpu, exit_qualification); | |
5786 | skip_emulated_instruction(vcpu); | |
5787 | return 1; | |
5788 | } | |
5789 | ||
fee84b07 AK |
5790 | static int handle_rdpmc(struct kvm_vcpu *vcpu) |
5791 | { | |
5792 | int err; | |
5793 | ||
5794 | err = kvm_rdpmc(vcpu); | |
5795 | kvm_complete_insn_gp(vcpu, err); | |
5796 | ||
5797 | return 1; | |
5798 | } | |
5799 | ||
851ba692 | 5800 | static int handle_wbinvd(struct kvm_vcpu *vcpu) |
e5edaa01 | 5801 | { |
f5f48ee1 | 5802 | kvm_emulate_wbinvd(vcpu); |
e5edaa01 ED |
5803 | return 1; |
5804 | } | |
5805 | ||
2acf923e DC |
5806 | static int handle_xsetbv(struct kvm_vcpu *vcpu) |
5807 | { | |
5808 | u64 new_bv = kvm_read_edx_eax(vcpu); | |
5809 | u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
5810 | ||
5811 | if (kvm_set_xcr(vcpu, index, new_bv) == 0) | |
5812 | skip_emulated_instruction(vcpu); | |
5813 | return 1; | |
5814 | } | |
5815 | ||
f53cd63c WL |
5816 | static int handle_xsaves(struct kvm_vcpu *vcpu) |
5817 | { | |
5818 | skip_emulated_instruction(vcpu); | |
5819 | WARN(1, "this should never happen\n"); | |
5820 | return 1; | |
5821 | } | |
5822 | ||
5823 | static int handle_xrstors(struct kvm_vcpu *vcpu) | |
5824 | { | |
5825 | skip_emulated_instruction(vcpu); | |
5826 | WARN(1, "this should never happen\n"); | |
5827 | return 1; | |
5828 | } | |
5829 | ||
851ba692 | 5830 | static int handle_apic_access(struct kvm_vcpu *vcpu) |
f78e0e2e | 5831 | { |
58fbbf26 KT |
5832 | if (likely(fasteoi)) { |
5833 | unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); | |
5834 | int access_type, offset; | |
5835 | ||
5836 | access_type = exit_qualification & APIC_ACCESS_TYPE; | |
5837 | offset = exit_qualification & APIC_ACCESS_OFFSET; | |
5838 | /* | |
5839 | * Sane guest uses MOV to write EOI, with written value | |
5840 | * not cared. So make a short-circuit here by avoiding | |
5841 | * heavy instruction emulation. | |
5842 | */ | |
5843 | if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) && | |
5844 | (offset == APIC_EOI)) { | |
5845 | kvm_lapic_set_eoi(vcpu); | |
5846 | skip_emulated_instruction(vcpu); | |
5847 | return 1; | |
5848 | } | |
5849 | } | |
51d8b661 | 5850 | return emulate_instruction(vcpu, 0) == EMULATE_DONE; |
f78e0e2e SY |
5851 | } |
5852 | ||
c7c9c56c YZ |
5853 | static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu) |
5854 | { | |
5855 | unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); | |
5856 | int vector = exit_qualification & 0xff; | |
5857 | ||
5858 | /* EOI-induced VM exit is trap-like and thus no need to adjust IP */ | |
5859 | kvm_apic_set_eoi_accelerated(vcpu, vector); | |
5860 | return 1; | |
5861 | } | |
5862 | ||
83d4c286 YZ |
5863 | static int handle_apic_write(struct kvm_vcpu *vcpu) |
5864 | { | |
5865 | unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); | |
5866 | u32 offset = exit_qualification & 0xfff; | |
5867 | ||
5868 | /* APIC-write VM exit is trap-like and thus no need to adjust IP */ | |
5869 | kvm_apic_write_nodecode(vcpu, offset); | |
5870 | return 1; | |
5871 | } | |
5872 | ||
851ba692 | 5873 | static int handle_task_switch(struct kvm_vcpu *vcpu) |
37817f29 | 5874 | { |
60637aac | 5875 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
37817f29 | 5876 | unsigned long exit_qualification; |
e269fb21 JK |
5877 | bool has_error_code = false; |
5878 | u32 error_code = 0; | |
37817f29 | 5879 | u16 tss_selector; |
7f3d35fd | 5880 | int reason, type, idt_v, idt_index; |
64a7ec06 GN |
5881 | |
5882 | idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK); | |
7f3d35fd | 5883 | idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK); |
64a7ec06 | 5884 | type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK); |
37817f29 IE |
5885 | |
5886 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); | |
5887 | ||
5888 | reason = (u32)exit_qualification >> 30; | |
64a7ec06 GN |
5889 | if (reason == TASK_SWITCH_GATE && idt_v) { |
5890 | switch (type) { | |
5891 | case INTR_TYPE_NMI_INTR: | |
5892 | vcpu->arch.nmi_injected = false; | |
654f06fc | 5893 | vmx_set_nmi_mask(vcpu, true); |
64a7ec06 GN |
5894 | break; |
5895 | case INTR_TYPE_EXT_INTR: | |
66fd3f7f | 5896 | case INTR_TYPE_SOFT_INTR: |
64a7ec06 GN |
5897 | kvm_clear_interrupt_queue(vcpu); |
5898 | break; | |
5899 | case INTR_TYPE_HARD_EXCEPTION: | |
e269fb21 JK |
5900 | if (vmx->idt_vectoring_info & |
5901 | VECTORING_INFO_DELIVER_CODE_MASK) { | |
5902 | has_error_code = true; | |
5903 | error_code = | |
5904 | vmcs_read32(IDT_VECTORING_ERROR_CODE); | |
5905 | } | |
5906 | /* fall through */ | |
64a7ec06 GN |
5907 | case INTR_TYPE_SOFT_EXCEPTION: |
5908 | kvm_clear_exception_queue(vcpu); | |
5909 | break; | |
5910 | default: | |
5911 | break; | |
5912 | } | |
60637aac | 5913 | } |
37817f29 IE |
5914 | tss_selector = exit_qualification; |
5915 | ||
64a7ec06 GN |
5916 | if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION && |
5917 | type != INTR_TYPE_EXT_INTR && | |
5918 | type != INTR_TYPE_NMI_INTR)) | |
5919 | skip_emulated_instruction(vcpu); | |
5920 | ||
7f3d35fd KW |
5921 | if (kvm_task_switch(vcpu, tss_selector, |
5922 | type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason, | |
5923 | has_error_code, error_code) == EMULATE_FAIL) { | |
acb54517 GN |
5924 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; |
5925 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
5926 | vcpu->run->internal.ndata = 0; | |
42dbaa5a | 5927 | return 0; |
acb54517 | 5928 | } |
42dbaa5a | 5929 | |
42dbaa5a JK |
5930 | /* |
5931 | * TODO: What about debug traps on tss switch? | |
5932 | * Are we supposed to inject them and update dr6? | |
5933 | */ | |
5934 | ||
5935 | return 1; | |
37817f29 IE |
5936 | } |
5937 | ||
851ba692 | 5938 | static int handle_ept_violation(struct kvm_vcpu *vcpu) |
1439442c | 5939 | { |
f9c617f6 | 5940 | unsigned long exit_qualification; |
1439442c | 5941 | gpa_t gpa; |
4f5982a5 | 5942 | u32 error_code; |
1439442c | 5943 | int gla_validity; |
1439442c | 5944 | |
f9c617f6 | 5945 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
1439442c | 5946 | |
1439442c SY |
5947 | gla_validity = (exit_qualification >> 7) & 0x3; |
5948 | if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) { | |
5949 | printk(KERN_ERR "EPT: Handling EPT violation failed!\n"); | |
5950 | printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n", | |
5951 | (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS), | |
f9c617f6 | 5952 | vmcs_readl(GUEST_LINEAR_ADDRESS)); |
1439442c SY |
5953 | printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n", |
5954 | (long unsigned int)exit_qualification); | |
851ba692 AK |
5955 | vcpu->run->exit_reason = KVM_EXIT_UNKNOWN; |
5956 | vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION; | |
596ae895 | 5957 | return 0; |
1439442c SY |
5958 | } |
5959 | ||
0be9c7a8 GN |
5960 | /* |
5961 | * EPT violation happened while executing iret from NMI, | |
5962 | * "blocked by NMI" bit has to be set before next VM entry. | |
5963 | * There are errata that may cause this bit to not be set: | |
5964 | * AAK134, BY25. | |
5965 | */ | |
bcd1c294 GN |
5966 | if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) && |
5967 | cpu_has_virtual_nmis() && | |
5968 | (exit_qualification & INTR_INFO_UNBLOCK_NMI)) | |
0be9c7a8 GN |
5969 | vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI); |
5970 | ||
1439442c | 5971 | gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS); |
229456fc | 5972 | trace_kvm_page_fault(gpa, exit_qualification); |
4f5982a5 XG |
5973 | |
5974 | /* It is a write fault? */ | |
81ed33e4 | 5975 | error_code = exit_qualification & PFERR_WRITE_MASK; |
25d92081 | 5976 | /* It is a fetch fault? */ |
81ed33e4 | 5977 | error_code |= (exit_qualification << 2) & PFERR_FETCH_MASK; |
4f5982a5 | 5978 | /* ept page table is present? */ |
81ed33e4 | 5979 | error_code |= (exit_qualification >> 3) & PFERR_PRESENT_MASK; |
4f5982a5 | 5980 | |
25d92081 YZ |
5981 | vcpu->arch.exit_qualification = exit_qualification; |
5982 | ||
4f5982a5 | 5983 | return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0); |
1439442c SY |
5984 | } |
5985 | ||
851ba692 | 5986 | static int handle_ept_misconfig(struct kvm_vcpu *vcpu) |
68f89400 | 5987 | { |
f735d4af | 5988 | int ret; |
68f89400 MT |
5989 | gpa_t gpa; |
5990 | ||
5991 | gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS); | |
e32edf4f | 5992 | if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) { |
68c3b4d1 | 5993 | skip_emulated_instruction(vcpu); |
931c33b1 | 5994 | trace_kvm_fast_mmio(gpa); |
68c3b4d1 MT |
5995 | return 1; |
5996 | } | |
68f89400 | 5997 | |
450869d6 | 5998 | ret = handle_mmio_page_fault(vcpu, gpa, true); |
b37fbea6 | 5999 | if (likely(ret == RET_MMIO_PF_EMULATE)) |
ce88decf XG |
6000 | return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) == |
6001 | EMULATE_DONE; | |
f8f55942 XG |
6002 | |
6003 | if (unlikely(ret == RET_MMIO_PF_INVALID)) | |
6004 | return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0); | |
6005 | ||
b37fbea6 | 6006 | if (unlikely(ret == RET_MMIO_PF_RETRY)) |
ce88decf XG |
6007 | return 1; |
6008 | ||
6009 | /* It is the real ept misconfig */ | |
f735d4af | 6010 | WARN_ON(1); |
68f89400 | 6011 | |
851ba692 AK |
6012 | vcpu->run->exit_reason = KVM_EXIT_UNKNOWN; |
6013 | vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG; | |
68f89400 MT |
6014 | |
6015 | return 0; | |
6016 | } | |
6017 | ||
851ba692 | 6018 | static int handle_nmi_window(struct kvm_vcpu *vcpu) |
f08864b4 SY |
6019 | { |
6020 | u32 cpu_based_vm_exec_control; | |
6021 | ||
6022 | /* clear pending NMI */ | |
6023 | cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); | |
6024 | cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING; | |
6025 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); | |
6026 | ++vcpu->stat.nmi_window_exits; | |
3842d135 | 6027 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
f08864b4 SY |
6028 | |
6029 | return 1; | |
6030 | } | |
6031 | ||
80ced186 | 6032 | static int handle_invalid_guest_state(struct kvm_vcpu *vcpu) |
ea953ef0 | 6033 | { |
8b3079a5 AK |
6034 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
6035 | enum emulation_result err = EMULATE_DONE; | |
80ced186 | 6036 | int ret = 1; |
49e9d557 AK |
6037 | u32 cpu_exec_ctrl; |
6038 | bool intr_window_requested; | |
b8405c18 | 6039 | unsigned count = 130; |
49e9d557 AK |
6040 | |
6041 | cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); | |
6042 | intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING; | |
ea953ef0 | 6043 | |
98eb2f8b | 6044 | while (vmx->emulation_required && count-- != 0) { |
bdea48e3 | 6045 | if (intr_window_requested && vmx_interrupt_allowed(vcpu)) |
49e9d557 AK |
6046 | return handle_interrupt_window(&vmx->vcpu); |
6047 | ||
de87dcdd AK |
6048 | if (test_bit(KVM_REQ_EVENT, &vcpu->requests)) |
6049 | return 1; | |
6050 | ||
991eebf9 | 6051 | err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE); |
ea953ef0 | 6052 | |
ac0a48c3 | 6053 | if (err == EMULATE_USER_EXIT) { |
94452b9e | 6054 | ++vcpu->stat.mmio_exits; |
80ced186 MG |
6055 | ret = 0; |
6056 | goto out; | |
6057 | } | |
1d5a4d9b | 6058 | |
de5f70e0 AK |
6059 | if (err != EMULATE_DONE) { |
6060 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
6061 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
6062 | vcpu->run->internal.ndata = 0; | |
6d77dbfc | 6063 | return 0; |
de5f70e0 | 6064 | } |
ea953ef0 | 6065 | |
8d76c49e GN |
6066 | if (vcpu->arch.halt_request) { |
6067 | vcpu->arch.halt_request = 0; | |
5cb56059 | 6068 | ret = kvm_vcpu_halt(vcpu); |
8d76c49e GN |
6069 | goto out; |
6070 | } | |
6071 | ||
ea953ef0 | 6072 | if (signal_pending(current)) |
80ced186 | 6073 | goto out; |
ea953ef0 MG |
6074 | if (need_resched()) |
6075 | schedule(); | |
6076 | } | |
6077 | ||
80ced186 MG |
6078 | out: |
6079 | return ret; | |
ea953ef0 MG |
6080 | } |
6081 | ||
b4a2d31d RK |
6082 | static int __grow_ple_window(int val) |
6083 | { | |
6084 | if (ple_window_grow < 1) | |
6085 | return ple_window; | |
6086 | ||
6087 | val = min(val, ple_window_actual_max); | |
6088 | ||
6089 | if (ple_window_grow < ple_window) | |
6090 | val *= ple_window_grow; | |
6091 | else | |
6092 | val += ple_window_grow; | |
6093 | ||
6094 | return val; | |
6095 | } | |
6096 | ||
6097 | static int __shrink_ple_window(int val, int modifier, int minimum) | |
6098 | { | |
6099 | if (modifier < 1) | |
6100 | return ple_window; | |
6101 | ||
6102 | if (modifier < ple_window) | |
6103 | val /= modifier; | |
6104 | else | |
6105 | val -= modifier; | |
6106 | ||
6107 | return max(val, minimum); | |
6108 | } | |
6109 | ||
6110 | static void grow_ple_window(struct kvm_vcpu *vcpu) | |
6111 | { | |
6112 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
6113 | int old = vmx->ple_window; | |
6114 | ||
6115 | vmx->ple_window = __grow_ple_window(old); | |
6116 | ||
6117 | if (vmx->ple_window != old) | |
6118 | vmx->ple_window_dirty = true; | |
7b46268d RK |
6119 | |
6120 | trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old); | |
b4a2d31d RK |
6121 | } |
6122 | ||
6123 | static void shrink_ple_window(struct kvm_vcpu *vcpu) | |
6124 | { | |
6125 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
6126 | int old = vmx->ple_window; | |
6127 | ||
6128 | vmx->ple_window = __shrink_ple_window(old, | |
6129 | ple_window_shrink, ple_window); | |
6130 | ||
6131 | if (vmx->ple_window != old) | |
6132 | vmx->ple_window_dirty = true; | |
7b46268d RK |
6133 | |
6134 | trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old); | |
b4a2d31d RK |
6135 | } |
6136 | ||
6137 | /* | |
6138 | * ple_window_actual_max is computed to be one grow_ple_window() below | |
6139 | * ple_window_max. (See __grow_ple_window for the reason.) | |
6140 | * This prevents overflows, because ple_window_max is int. | |
6141 | * ple_window_max effectively rounded down to a multiple of ple_window_grow in | |
6142 | * this process. | |
6143 | * ple_window_max is also prevented from setting vmx->ple_window < ple_window. | |
6144 | */ | |
6145 | static void update_ple_window_actual_max(void) | |
6146 | { | |
6147 | ple_window_actual_max = | |
6148 | __shrink_ple_window(max(ple_window_max, ple_window), | |
6149 | ple_window_grow, INT_MIN); | |
6150 | } | |
6151 | ||
bf9f6ac8 FW |
6152 | /* |
6153 | * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR. | |
6154 | */ | |
6155 | static void wakeup_handler(void) | |
6156 | { | |
6157 | struct kvm_vcpu *vcpu; | |
6158 | int cpu = smp_processor_id(); | |
6159 | ||
6160 | spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu)); | |
6161 | list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu), | |
6162 | blocked_vcpu_list) { | |
6163 | struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu); | |
6164 | ||
6165 | if (pi_test_on(pi_desc) == 1) | |
6166 | kvm_vcpu_kick(vcpu); | |
6167 | } | |
6168 | spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu)); | |
6169 | } | |
6170 | ||
f2c7648d TC |
6171 | static __init int hardware_setup(void) |
6172 | { | |
34a1cd60 TC |
6173 | int r = -ENOMEM, i, msr; |
6174 | ||
6175 | rdmsrl_safe(MSR_EFER, &host_efer); | |
6176 | ||
6177 | for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) | |
6178 | kvm_define_shared_msr(i, vmx_msr_index[i]); | |
6179 | ||
6180 | vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL); | |
6181 | if (!vmx_io_bitmap_a) | |
6182 | return r; | |
6183 | ||
6184 | vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL); | |
6185 | if (!vmx_io_bitmap_b) | |
6186 | goto out; | |
6187 | ||
6188 | vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL); | |
6189 | if (!vmx_msr_bitmap_legacy) | |
6190 | goto out1; | |
6191 | ||
6192 | vmx_msr_bitmap_legacy_x2apic = | |
6193 | (unsigned long *)__get_free_page(GFP_KERNEL); | |
6194 | if (!vmx_msr_bitmap_legacy_x2apic) | |
6195 | goto out2; | |
6196 | ||
6197 | vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL); | |
6198 | if (!vmx_msr_bitmap_longmode) | |
6199 | goto out3; | |
6200 | ||
6201 | vmx_msr_bitmap_longmode_x2apic = | |
6202 | (unsigned long *)__get_free_page(GFP_KERNEL); | |
6203 | if (!vmx_msr_bitmap_longmode_x2apic) | |
6204 | goto out4; | |
3af18d9c WV |
6205 | |
6206 | if (nested) { | |
6207 | vmx_msr_bitmap_nested = | |
6208 | (unsigned long *)__get_free_page(GFP_KERNEL); | |
6209 | if (!vmx_msr_bitmap_nested) | |
6210 | goto out5; | |
6211 | } | |
6212 | ||
34a1cd60 TC |
6213 | vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL); |
6214 | if (!vmx_vmread_bitmap) | |
3af18d9c | 6215 | goto out6; |
34a1cd60 TC |
6216 | |
6217 | vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL); | |
6218 | if (!vmx_vmwrite_bitmap) | |
3af18d9c | 6219 | goto out7; |
34a1cd60 TC |
6220 | |
6221 | memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE); | |
6222 | memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE); | |
6223 | ||
6224 | /* | |
6225 | * Allow direct access to the PC debug port (it is often used for I/O | |
6226 | * delays, but the vmexits simply slow things down). | |
6227 | */ | |
6228 | memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE); | |
6229 | clear_bit(0x80, vmx_io_bitmap_a); | |
6230 | ||
6231 | memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE); | |
6232 | ||
6233 | memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE); | |
6234 | memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE); | |
3af18d9c WV |
6235 | if (nested) |
6236 | memset(vmx_msr_bitmap_nested, 0xff, PAGE_SIZE); | |
34a1cd60 | 6237 | |
34a1cd60 TC |
6238 | if (setup_vmcs_config(&vmcs_config) < 0) { |
6239 | r = -EIO; | |
3af18d9c | 6240 | goto out8; |
baa03522 | 6241 | } |
f2c7648d TC |
6242 | |
6243 | if (boot_cpu_has(X86_FEATURE_NX)) | |
6244 | kvm_enable_efer_bits(EFER_NX); | |
6245 | ||
6246 | if (!cpu_has_vmx_vpid()) | |
6247 | enable_vpid = 0; | |
6248 | if (!cpu_has_vmx_shadow_vmcs()) | |
6249 | enable_shadow_vmcs = 0; | |
6250 | if (enable_shadow_vmcs) | |
6251 | init_vmcs_shadow_fields(); | |
6252 | ||
6253 | if (!cpu_has_vmx_ept() || | |
6254 | !cpu_has_vmx_ept_4levels()) { | |
6255 | enable_ept = 0; | |
6256 | enable_unrestricted_guest = 0; | |
6257 | enable_ept_ad_bits = 0; | |
6258 | } | |
6259 | ||
6260 | if (!cpu_has_vmx_ept_ad_bits()) | |
6261 | enable_ept_ad_bits = 0; | |
6262 | ||
6263 | if (!cpu_has_vmx_unrestricted_guest()) | |
6264 | enable_unrestricted_guest = 0; | |
6265 | ||
ad15a296 | 6266 | if (!cpu_has_vmx_flexpriority()) |
f2c7648d TC |
6267 | flexpriority_enabled = 0; |
6268 | ||
ad15a296 PB |
6269 | /* |
6270 | * set_apic_access_page_addr() is used to reload apic access | |
6271 | * page upon invalidation. No need to do anything if not | |
6272 | * using the APIC_ACCESS_ADDR VMCS field. | |
6273 | */ | |
6274 | if (!flexpriority_enabled) | |
f2c7648d | 6275 | kvm_x86_ops->set_apic_access_page_addr = NULL; |
f2c7648d TC |
6276 | |
6277 | if (!cpu_has_vmx_tpr_shadow()) | |
6278 | kvm_x86_ops->update_cr8_intercept = NULL; | |
6279 | ||
6280 | if (enable_ept && !cpu_has_vmx_ept_2m_page()) | |
6281 | kvm_disable_largepages(); | |
6282 | ||
6283 | if (!cpu_has_vmx_ple()) | |
6284 | ple_gap = 0; | |
6285 | ||
6286 | if (!cpu_has_vmx_apicv()) | |
6287 | enable_apicv = 0; | |
6288 | ||
64903d61 HZ |
6289 | if (cpu_has_vmx_tsc_scaling()) { |
6290 | kvm_has_tsc_control = true; | |
6291 | kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX; | |
6292 | kvm_tsc_scaling_ratio_frac_bits = 48; | |
6293 | } | |
6294 | ||
baa03522 TC |
6295 | vmx_disable_intercept_for_msr(MSR_FS_BASE, false); |
6296 | vmx_disable_intercept_for_msr(MSR_GS_BASE, false); | |
6297 | vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true); | |
6298 | vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false); | |
6299 | vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false); | |
6300 | vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false); | |
6301 | vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true); | |
6302 | ||
6303 | memcpy(vmx_msr_bitmap_legacy_x2apic, | |
6304 | vmx_msr_bitmap_legacy, PAGE_SIZE); | |
6305 | memcpy(vmx_msr_bitmap_longmode_x2apic, | |
6306 | vmx_msr_bitmap_longmode, PAGE_SIZE); | |
6307 | ||
04bb92e4 WL |
6308 | set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */ |
6309 | ||
baa03522 TC |
6310 | if (enable_apicv) { |
6311 | for (msr = 0x800; msr <= 0x8ff; msr++) | |
6312 | vmx_disable_intercept_msr_read_x2apic(msr); | |
6313 | ||
6314 | /* According SDM, in x2apic mode, the whole id reg is used. | |
6315 | * But in KVM, it only use the highest eight bits. Need to | |
6316 | * intercept it */ | |
6317 | vmx_enable_intercept_msr_read_x2apic(0x802); | |
6318 | /* TMCCT */ | |
6319 | vmx_enable_intercept_msr_read_x2apic(0x839); | |
6320 | /* TPR */ | |
6321 | vmx_disable_intercept_msr_write_x2apic(0x808); | |
6322 | /* EOI */ | |
6323 | vmx_disable_intercept_msr_write_x2apic(0x80b); | |
6324 | /* SELF-IPI */ | |
6325 | vmx_disable_intercept_msr_write_x2apic(0x83f); | |
6326 | } | |
6327 | ||
6328 | if (enable_ept) { | |
6329 | kvm_mmu_set_mask_ptes(0ull, | |
6330 | (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull, | |
6331 | (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull, | |
6332 | 0ull, VMX_EPT_EXECUTABLE_MASK); | |
6333 | ept_set_mmio_spte_mask(); | |
6334 | kvm_enable_tdp(); | |
6335 | } else | |
6336 | kvm_disable_tdp(); | |
6337 | ||
6338 | update_ple_window_actual_max(); | |
6339 | ||
843e4330 KH |
6340 | /* |
6341 | * Only enable PML when hardware supports PML feature, and both EPT | |
6342 | * and EPT A/D bit features are enabled -- PML depends on them to work. | |
6343 | */ | |
6344 | if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml()) | |
6345 | enable_pml = 0; | |
6346 | ||
6347 | if (!enable_pml) { | |
6348 | kvm_x86_ops->slot_enable_log_dirty = NULL; | |
6349 | kvm_x86_ops->slot_disable_log_dirty = NULL; | |
6350 | kvm_x86_ops->flush_log_dirty = NULL; | |
6351 | kvm_x86_ops->enable_log_dirty_pt_masked = NULL; | |
6352 | } | |
6353 | ||
bf9f6ac8 FW |
6354 | kvm_set_posted_intr_wakeup_handler(wakeup_handler); |
6355 | ||
f2c7648d | 6356 | return alloc_kvm_area(); |
34a1cd60 | 6357 | |
3af18d9c | 6358 | out8: |
34a1cd60 | 6359 | free_page((unsigned long)vmx_vmwrite_bitmap); |
3af18d9c | 6360 | out7: |
34a1cd60 | 6361 | free_page((unsigned long)vmx_vmread_bitmap); |
3af18d9c WV |
6362 | out6: |
6363 | if (nested) | |
6364 | free_page((unsigned long)vmx_msr_bitmap_nested); | |
34a1cd60 TC |
6365 | out5: |
6366 | free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic); | |
6367 | out4: | |
6368 | free_page((unsigned long)vmx_msr_bitmap_longmode); | |
6369 | out3: | |
6370 | free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic); | |
6371 | out2: | |
6372 | free_page((unsigned long)vmx_msr_bitmap_legacy); | |
6373 | out1: | |
6374 | free_page((unsigned long)vmx_io_bitmap_b); | |
6375 | out: | |
6376 | free_page((unsigned long)vmx_io_bitmap_a); | |
6377 | ||
6378 | return r; | |
f2c7648d TC |
6379 | } |
6380 | ||
6381 | static __exit void hardware_unsetup(void) | |
6382 | { | |
34a1cd60 TC |
6383 | free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic); |
6384 | free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic); | |
6385 | free_page((unsigned long)vmx_msr_bitmap_legacy); | |
6386 | free_page((unsigned long)vmx_msr_bitmap_longmode); | |
6387 | free_page((unsigned long)vmx_io_bitmap_b); | |
6388 | free_page((unsigned long)vmx_io_bitmap_a); | |
6389 | free_page((unsigned long)vmx_vmwrite_bitmap); | |
6390 | free_page((unsigned long)vmx_vmread_bitmap); | |
3af18d9c WV |
6391 | if (nested) |
6392 | free_page((unsigned long)vmx_msr_bitmap_nested); | |
34a1cd60 | 6393 | |
f2c7648d TC |
6394 | free_kvm_area(); |
6395 | } | |
6396 | ||
4b8d54f9 ZE |
6397 | /* |
6398 | * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE | |
6399 | * exiting, so only get here on cpu with PAUSE-Loop-Exiting. | |
6400 | */ | |
9fb41ba8 | 6401 | static int handle_pause(struct kvm_vcpu *vcpu) |
4b8d54f9 | 6402 | { |
b4a2d31d RK |
6403 | if (ple_gap) |
6404 | grow_ple_window(vcpu); | |
6405 | ||
4b8d54f9 ZE |
6406 | skip_emulated_instruction(vcpu); |
6407 | kvm_vcpu_on_spin(vcpu); | |
6408 | ||
6409 | return 1; | |
6410 | } | |
6411 | ||
87c00572 | 6412 | static int handle_nop(struct kvm_vcpu *vcpu) |
59708670 | 6413 | { |
87c00572 | 6414 | skip_emulated_instruction(vcpu); |
59708670 SY |
6415 | return 1; |
6416 | } | |
6417 | ||
87c00572 GS |
6418 | static int handle_mwait(struct kvm_vcpu *vcpu) |
6419 | { | |
6420 | printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n"); | |
6421 | return handle_nop(vcpu); | |
6422 | } | |
6423 | ||
5f3d45e7 MD |
6424 | static int handle_monitor_trap(struct kvm_vcpu *vcpu) |
6425 | { | |
6426 | return 1; | |
6427 | } | |
6428 | ||
87c00572 GS |
6429 | static int handle_monitor(struct kvm_vcpu *vcpu) |
6430 | { | |
6431 | printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n"); | |
6432 | return handle_nop(vcpu); | |
6433 | } | |
6434 | ||
ff2f6fe9 NHE |
6435 | /* |
6436 | * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12. | |
6437 | * We could reuse a single VMCS for all the L2 guests, but we also want the | |
6438 | * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this | |
6439 | * allows keeping them loaded on the processor, and in the future will allow | |
6440 | * optimizations where prepare_vmcs02 doesn't need to set all the fields on | |
6441 | * every entry if they never change. | |
6442 | * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE | |
6443 | * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first. | |
6444 | * | |
6445 | * The following functions allocate and free a vmcs02 in this pool. | |
6446 | */ | |
6447 | ||
6448 | /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */ | |
6449 | static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx) | |
6450 | { | |
6451 | struct vmcs02_list *item; | |
6452 | list_for_each_entry(item, &vmx->nested.vmcs02_pool, list) | |
6453 | if (item->vmptr == vmx->nested.current_vmptr) { | |
6454 | list_move(&item->list, &vmx->nested.vmcs02_pool); | |
6455 | return &item->vmcs02; | |
6456 | } | |
6457 | ||
6458 | if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) { | |
6459 | /* Recycle the least recently used VMCS. */ | |
d74c0e6b GT |
6460 | item = list_last_entry(&vmx->nested.vmcs02_pool, |
6461 | struct vmcs02_list, list); | |
ff2f6fe9 NHE |
6462 | item->vmptr = vmx->nested.current_vmptr; |
6463 | list_move(&item->list, &vmx->nested.vmcs02_pool); | |
6464 | return &item->vmcs02; | |
6465 | } | |
6466 | ||
6467 | /* Create a new VMCS */ | |
0fa24ce3 | 6468 | item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL); |
ff2f6fe9 NHE |
6469 | if (!item) |
6470 | return NULL; | |
6471 | item->vmcs02.vmcs = alloc_vmcs(); | |
6472 | if (!item->vmcs02.vmcs) { | |
6473 | kfree(item); | |
6474 | return NULL; | |
6475 | } | |
6476 | loaded_vmcs_init(&item->vmcs02); | |
6477 | item->vmptr = vmx->nested.current_vmptr; | |
6478 | list_add(&(item->list), &(vmx->nested.vmcs02_pool)); | |
6479 | vmx->nested.vmcs02_num++; | |
6480 | return &item->vmcs02; | |
6481 | } | |
6482 | ||
6483 | /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */ | |
6484 | static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr) | |
6485 | { | |
6486 | struct vmcs02_list *item; | |
6487 | list_for_each_entry(item, &vmx->nested.vmcs02_pool, list) | |
6488 | if (item->vmptr == vmptr) { | |
6489 | free_loaded_vmcs(&item->vmcs02); | |
6490 | list_del(&item->list); | |
6491 | kfree(item); | |
6492 | vmx->nested.vmcs02_num--; | |
6493 | return; | |
6494 | } | |
6495 | } | |
6496 | ||
6497 | /* | |
6498 | * Free all VMCSs saved for this vcpu, except the one pointed by | |
4fa7734c PB |
6499 | * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs |
6500 | * must be &vmx->vmcs01. | |
ff2f6fe9 NHE |
6501 | */ |
6502 | static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx) | |
6503 | { | |
6504 | struct vmcs02_list *item, *n; | |
4fa7734c PB |
6505 | |
6506 | WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01); | |
ff2f6fe9 | 6507 | list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) { |
4fa7734c PB |
6508 | /* |
6509 | * Something will leak if the above WARN triggers. Better than | |
6510 | * a use-after-free. | |
6511 | */ | |
6512 | if (vmx->loaded_vmcs == &item->vmcs02) | |
6513 | continue; | |
6514 | ||
6515 | free_loaded_vmcs(&item->vmcs02); | |
ff2f6fe9 NHE |
6516 | list_del(&item->list); |
6517 | kfree(item); | |
4fa7734c | 6518 | vmx->nested.vmcs02_num--; |
ff2f6fe9 | 6519 | } |
ff2f6fe9 NHE |
6520 | } |
6521 | ||
0658fbaa ACL |
6522 | /* |
6523 | * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(), | |
6524 | * set the success or error code of an emulated VMX instruction, as specified | |
6525 | * by Vol 2B, VMX Instruction Reference, "Conventions". | |
6526 | */ | |
6527 | static void nested_vmx_succeed(struct kvm_vcpu *vcpu) | |
6528 | { | |
6529 | vmx_set_rflags(vcpu, vmx_get_rflags(vcpu) | |
6530 | & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | | |
6531 | X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF)); | |
6532 | } | |
6533 | ||
6534 | static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu) | |
6535 | { | |
6536 | vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu) | |
6537 | & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF | | |
6538 | X86_EFLAGS_SF | X86_EFLAGS_OF)) | |
6539 | | X86_EFLAGS_CF); | |
6540 | } | |
6541 | ||
145c28dd | 6542 | static void nested_vmx_failValid(struct kvm_vcpu *vcpu, |
0658fbaa ACL |
6543 | u32 vm_instruction_error) |
6544 | { | |
6545 | if (to_vmx(vcpu)->nested.current_vmptr == -1ull) { | |
6546 | /* | |
6547 | * failValid writes the error number to the current VMCS, which | |
6548 | * can't be done there isn't a current VMCS. | |
6549 | */ | |
6550 | nested_vmx_failInvalid(vcpu); | |
6551 | return; | |
6552 | } | |
6553 | vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu) | |
6554 | & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | | |
6555 | X86_EFLAGS_SF | X86_EFLAGS_OF)) | |
6556 | | X86_EFLAGS_ZF); | |
6557 | get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error; | |
6558 | /* | |
6559 | * We don't need to force a shadow sync because | |
6560 | * VM_INSTRUCTION_ERROR is not shadowed | |
6561 | */ | |
6562 | } | |
145c28dd | 6563 | |
ff651cb6 WV |
6564 | static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator) |
6565 | { | |
6566 | /* TODO: not to reset guest simply here. */ | |
6567 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); | |
6568 | pr_warn("kvm: nested vmx abort, indicator %d\n", indicator); | |
6569 | } | |
6570 | ||
f4124500 JK |
6571 | static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer) |
6572 | { | |
6573 | struct vcpu_vmx *vmx = | |
6574 | container_of(timer, struct vcpu_vmx, nested.preemption_timer); | |
6575 | ||
6576 | vmx->nested.preemption_timer_expired = true; | |
6577 | kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu); | |
6578 | kvm_vcpu_kick(&vmx->vcpu); | |
6579 | ||
6580 | return HRTIMER_NORESTART; | |
6581 | } | |
6582 | ||
19677e32 BD |
6583 | /* |
6584 | * Decode the memory-address operand of a vmx instruction, as recorded on an | |
6585 | * exit caused by such an instruction (run by a guest hypervisor). | |
6586 | * On success, returns 0. When the operand is invalid, returns 1 and throws | |
6587 | * #UD or #GP. | |
6588 | */ | |
6589 | static int get_vmx_mem_address(struct kvm_vcpu *vcpu, | |
6590 | unsigned long exit_qualification, | |
f9eb4af6 | 6591 | u32 vmx_instruction_info, bool wr, gva_t *ret) |
19677e32 | 6592 | { |
f9eb4af6 EK |
6593 | gva_t off; |
6594 | bool exn; | |
6595 | struct kvm_segment s; | |
6596 | ||
19677e32 BD |
6597 | /* |
6598 | * According to Vol. 3B, "Information for VM Exits Due to Instruction | |
6599 | * Execution", on an exit, vmx_instruction_info holds most of the | |
6600 | * addressing components of the operand. Only the displacement part | |
6601 | * is put in exit_qualification (see 3B, "Basic VM-Exit Information"). | |
6602 | * For how an actual address is calculated from all these components, | |
6603 | * refer to Vol. 1, "Operand Addressing". | |
6604 | */ | |
6605 | int scaling = vmx_instruction_info & 3; | |
6606 | int addr_size = (vmx_instruction_info >> 7) & 7; | |
6607 | bool is_reg = vmx_instruction_info & (1u << 10); | |
6608 | int seg_reg = (vmx_instruction_info >> 15) & 7; | |
6609 | int index_reg = (vmx_instruction_info >> 18) & 0xf; | |
6610 | bool index_is_valid = !(vmx_instruction_info & (1u << 22)); | |
6611 | int base_reg = (vmx_instruction_info >> 23) & 0xf; | |
6612 | bool base_is_valid = !(vmx_instruction_info & (1u << 27)); | |
6613 | ||
6614 | if (is_reg) { | |
6615 | kvm_queue_exception(vcpu, UD_VECTOR); | |
6616 | return 1; | |
6617 | } | |
6618 | ||
6619 | /* Addr = segment_base + offset */ | |
6620 | /* offset = base + [index * scale] + displacement */ | |
f9eb4af6 | 6621 | off = exit_qualification; /* holds the displacement */ |
19677e32 | 6622 | if (base_is_valid) |
f9eb4af6 | 6623 | off += kvm_register_read(vcpu, base_reg); |
19677e32 | 6624 | if (index_is_valid) |
f9eb4af6 EK |
6625 | off += kvm_register_read(vcpu, index_reg)<<scaling; |
6626 | vmx_get_segment(vcpu, &s, seg_reg); | |
6627 | *ret = s.base + off; | |
19677e32 BD |
6628 | |
6629 | if (addr_size == 1) /* 32 bit */ | |
6630 | *ret &= 0xffffffff; | |
6631 | ||
f9eb4af6 EK |
6632 | /* Checks for #GP/#SS exceptions. */ |
6633 | exn = false; | |
6634 | if (is_protmode(vcpu)) { | |
6635 | /* Protected mode: apply checks for segment validity in the | |
6636 | * following order: | |
6637 | * - segment type check (#GP(0) may be thrown) | |
6638 | * - usability check (#GP(0)/#SS(0)) | |
6639 | * - limit check (#GP(0)/#SS(0)) | |
6640 | */ | |
6641 | if (wr) | |
6642 | /* #GP(0) if the destination operand is located in a | |
6643 | * read-only data segment or any code segment. | |
6644 | */ | |
6645 | exn = ((s.type & 0xa) == 0 || (s.type & 8)); | |
6646 | else | |
6647 | /* #GP(0) if the source operand is located in an | |
6648 | * execute-only code segment | |
6649 | */ | |
6650 | exn = ((s.type & 0xa) == 8); | |
6651 | } | |
6652 | if (exn) { | |
6653 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
6654 | return 1; | |
6655 | } | |
6656 | if (is_long_mode(vcpu)) { | |
6657 | /* Long mode: #GP(0)/#SS(0) if the memory address is in a | |
6658 | * non-canonical form. This is an only check for long mode. | |
6659 | */ | |
6660 | exn = is_noncanonical_address(*ret); | |
6661 | } else if (is_protmode(vcpu)) { | |
6662 | /* Protected mode: #GP(0)/#SS(0) if the segment is unusable. | |
6663 | */ | |
6664 | exn = (s.unusable != 0); | |
6665 | /* Protected mode: #GP(0)/#SS(0) if the memory | |
6666 | * operand is outside the segment limit. | |
6667 | */ | |
6668 | exn = exn || (off + sizeof(u64) > s.limit); | |
6669 | } | |
6670 | if (exn) { | |
6671 | kvm_queue_exception_e(vcpu, | |
6672 | seg_reg == VCPU_SREG_SS ? | |
6673 | SS_VECTOR : GP_VECTOR, | |
6674 | 0); | |
6675 | return 1; | |
6676 | } | |
6677 | ||
19677e32 BD |
6678 | return 0; |
6679 | } | |
6680 | ||
3573e22c BD |
6681 | /* |
6682 | * This function performs the various checks including | |
6683 | * - if it's 4KB aligned | |
6684 | * - No bits beyond the physical address width are set | |
6685 | * - Returns 0 on success or else 1 | |
4291b588 | 6686 | * (Intel SDM Section 30.3) |
3573e22c | 6687 | */ |
4291b588 BD |
6688 | static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason, |
6689 | gpa_t *vmpointer) | |
3573e22c BD |
6690 | { |
6691 | gva_t gva; | |
6692 | gpa_t vmptr; | |
6693 | struct x86_exception e; | |
6694 | struct page *page; | |
6695 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
6696 | int maxphyaddr = cpuid_maxphyaddr(vcpu); | |
6697 | ||
6698 | if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION), | |
f9eb4af6 | 6699 | vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva)) |
3573e22c BD |
6700 | return 1; |
6701 | ||
6702 | if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr, | |
6703 | sizeof(vmptr), &e)) { | |
6704 | kvm_inject_page_fault(vcpu, &e); | |
6705 | return 1; | |
6706 | } | |
6707 | ||
6708 | switch (exit_reason) { | |
6709 | case EXIT_REASON_VMON: | |
6710 | /* | |
6711 | * SDM 3: 24.11.5 | |
6712 | * The first 4 bytes of VMXON region contain the supported | |
6713 | * VMCS revision identifier | |
6714 | * | |
6715 | * Note - IA32_VMX_BASIC[48] will never be 1 | |
6716 | * for the nested case; | |
6717 | * which replaces physical address width with 32 | |
6718 | * | |
6719 | */ | |
bc39c4db | 6720 | if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) { |
3573e22c BD |
6721 | nested_vmx_failInvalid(vcpu); |
6722 | skip_emulated_instruction(vcpu); | |
6723 | return 1; | |
6724 | } | |
6725 | ||
6726 | page = nested_get_page(vcpu, vmptr); | |
6727 | if (page == NULL || | |
6728 | *(u32 *)kmap(page) != VMCS12_REVISION) { | |
6729 | nested_vmx_failInvalid(vcpu); | |
6730 | kunmap(page); | |
6731 | skip_emulated_instruction(vcpu); | |
6732 | return 1; | |
6733 | } | |
6734 | kunmap(page); | |
6735 | vmx->nested.vmxon_ptr = vmptr; | |
6736 | break; | |
4291b588 | 6737 | case EXIT_REASON_VMCLEAR: |
bc39c4db | 6738 | if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) { |
4291b588 BD |
6739 | nested_vmx_failValid(vcpu, |
6740 | VMXERR_VMCLEAR_INVALID_ADDRESS); | |
6741 | skip_emulated_instruction(vcpu); | |
6742 | return 1; | |
6743 | } | |
6744 | ||
6745 | if (vmptr == vmx->nested.vmxon_ptr) { | |
6746 | nested_vmx_failValid(vcpu, | |
6747 | VMXERR_VMCLEAR_VMXON_POINTER); | |
6748 | skip_emulated_instruction(vcpu); | |
6749 | return 1; | |
6750 | } | |
6751 | break; | |
6752 | case EXIT_REASON_VMPTRLD: | |
bc39c4db | 6753 | if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) { |
4291b588 BD |
6754 | nested_vmx_failValid(vcpu, |
6755 | VMXERR_VMPTRLD_INVALID_ADDRESS); | |
6756 | skip_emulated_instruction(vcpu); | |
6757 | return 1; | |
6758 | } | |
3573e22c | 6759 | |
4291b588 BD |
6760 | if (vmptr == vmx->nested.vmxon_ptr) { |
6761 | nested_vmx_failValid(vcpu, | |
6762 | VMXERR_VMCLEAR_VMXON_POINTER); | |
6763 | skip_emulated_instruction(vcpu); | |
6764 | return 1; | |
6765 | } | |
6766 | break; | |
3573e22c BD |
6767 | default: |
6768 | return 1; /* shouldn't happen */ | |
6769 | } | |
6770 | ||
4291b588 BD |
6771 | if (vmpointer) |
6772 | *vmpointer = vmptr; | |
3573e22c BD |
6773 | return 0; |
6774 | } | |
6775 | ||
ec378aee NHE |
6776 | /* |
6777 | * Emulate the VMXON instruction. | |
6778 | * Currently, we just remember that VMX is active, and do not save or even | |
6779 | * inspect the argument to VMXON (the so-called "VMXON pointer") because we | |
6780 | * do not currently need to store anything in that guest-allocated memory | |
6781 | * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their | |
6782 | * argument is different from the VMXON pointer (which the spec says they do). | |
6783 | */ | |
6784 | static int handle_vmon(struct kvm_vcpu *vcpu) | |
6785 | { | |
6786 | struct kvm_segment cs; | |
6787 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
8de48833 | 6788 | struct vmcs *shadow_vmcs; |
b3897a49 NHE |
6789 | const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED |
6790 | | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX; | |
ec378aee NHE |
6791 | |
6792 | /* The Intel VMX Instruction Reference lists a bunch of bits that | |
6793 | * are prerequisite to running VMXON, most notably cr4.VMXE must be | |
6794 | * set to 1 (see vmx_set_cr4() for when we allow the guest to set this). | |
6795 | * Otherwise, we should fail with #UD. We test these now: | |
6796 | */ | |
6797 | if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) || | |
6798 | !kvm_read_cr0_bits(vcpu, X86_CR0_PE) || | |
6799 | (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) { | |
6800 | kvm_queue_exception(vcpu, UD_VECTOR); | |
6801 | return 1; | |
6802 | } | |
6803 | ||
6804 | vmx_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
6805 | if (is_long_mode(vcpu) && !cs.l) { | |
6806 | kvm_queue_exception(vcpu, UD_VECTOR); | |
6807 | return 1; | |
6808 | } | |
6809 | ||
6810 | if (vmx_get_cpl(vcpu)) { | |
6811 | kvm_inject_gp(vcpu, 0); | |
6812 | return 1; | |
6813 | } | |
3573e22c | 6814 | |
4291b588 | 6815 | if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL)) |
3573e22c BD |
6816 | return 1; |
6817 | ||
145c28dd AG |
6818 | if (vmx->nested.vmxon) { |
6819 | nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION); | |
6820 | skip_emulated_instruction(vcpu); | |
6821 | return 1; | |
6822 | } | |
b3897a49 NHE |
6823 | |
6824 | if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES) | |
6825 | != VMXON_NEEDED_FEATURES) { | |
6826 | kvm_inject_gp(vcpu, 0); | |
6827 | return 1; | |
6828 | } | |
6829 | ||
8de48833 AG |
6830 | if (enable_shadow_vmcs) { |
6831 | shadow_vmcs = alloc_vmcs(); | |
6832 | if (!shadow_vmcs) | |
6833 | return -ENOMEM; | |
6834 | /* mark vmcs as shadow */ | |
6835 | shadow_vmcs->revision_id |= (1u << 31); | |
6836 | /* init shadow vmcs */ | |
6837 | vmcs_clear(shadow_vmcs); | |
6838 | vmx->nested.current_shadow_vmcs = shadow_vmcs; | |
6839 | } | |
ec378aee | 6840 | |
ff2f6fe9 NHE |
6841 | INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool)); |
6842 | vmx->nested.vmcs02_num = 0; | |
6843 | ||
f4124500 JK |
6844 | hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC, |
6845 | HRTIMER_MODE_REL); | |
6846 | vmx->nested.preemption_timer.function = vmx_preemption_timer_fn; | |
6847 | ||
ec378aee NHE |
6848 | vmx->nested.vmxon = true; |
6849 | ||
6850 | skip_emulated_instruction(vcpu); | |
a25eb114 | 6851 | nested_vmx_succeed(vcpu); |
ec378aee NHE |
6852 | return 1; |
6853 | } | |
6854 | ||
6855 | /* | |
6856 | * Intel's VMX Instruction Reference specifies a common set of prerequisites | |
6857 | * for running VMX instructions (except VMXON, whose prerequisites are | |
6858 | * slightly different). It also specifies what exception to inject otherwise. | |
6859 | */ | |
6860 | static int nested_vmx_check_permission(struct kvm_vcpu *vcpu) | |
6861 | { | |
6862 | struct kvm_segment cs; | |
6863 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
6864 | ||
6865 | if (!vmx->nested.vmxon) { | |
6866 | kvm_queue_exception(vcpu, UD_VECTOR); | |
6867 | return 0; | |
6868 | } | |
6869 | ||
6870 | vmx_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
6871 | if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) || | |
6872 | (is_long_mode(vcpu) && !cs.l)) { | |
6873 | kvm_queue_exception(vcpu, UD_VECTOR); | |
6874 | return 0; | |
6875 | } | |
6876 | ||
6877 | if (vmx_get_cpl(vcpu)) { | |
6878 | kvm_inject_gp(vcpu, 0); | |
6879 | return 0; | |
6880 | } | |
6881 | ||
6882 | return 1; | |
6883 | } | |
6884 | ||
e7953d7f AG |
6885 | static inline void nested_release_vmcs12(struct vcpu_vmx *vmx) |
6886 | { | |
9a2a05b9 PB |
6887 | if (vmx->nested.current_vmptr == -1ull) |
6888 | return; | |
6889 | ||
6890 | /* current_vmptr and current_vmcs12 are always set/reset together */ | |
6891 | if (WARN_ON(vmx->nested.current_vmcs12 == NULL)) | |
6892 | return; | |
6893 | ||
012f83cb | 6894 | if (enable_shadow_vmcs) { |
9a2a05b9 PB |
6895 | /* copy to memory all shadowed fields in case |
6896 | they were modified */ | |
6897 | copy_shadow_to_vmcs12(vmx); | |
6898 | vmx->nested.sync_shadow_vmcs = false; | |
7ec36296 XG |
6899 | vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, |
6900 | SECONDARY_EXEC_SHADOW_VMCS); | |
9a2a05b9 | 6901 | vmcs_write64(VMCS_LINK_POINTER, -1ull); |
012f83cb | 6902 | } |
705699a1 | 6903 | vmx->nested.posted_intr_nv = -1; |
e7953d7f AG |
6904 | kunmap(vmx->nested.current_vmcs12_page); |
6905 | nested_release_page(vmx->nested.current_vmcs12_page); | |
9a2a05b9 PB |
6906 | vmx->nested.current_vmptr = -1ull; |
6907 | vmx->nested.current_vmcs12 = NULL; | |
e7953d7f AG |
6908 | } |
6909 | ||
ec378aee NHE |
6910 | /* |
6911 | * Free whatever needs to be freed from vmx->nested when L1 goes down, or | |
6912 | * just stops using VMX. | |
6913 | */ | |
6914 | static void free_nested(struct vcpu_vmx *vmx) | |
6915 | { | |
6916 | if (!vmx->nested.vmxon) | |
6917 | return; | |
9a2a05b9 | 6918 | |
ec378aee | 6919 | vmx->nested.vmxon = false; |
5c614b35 | 6920 | free_vpid(vmx->nested.vpid02); |
9a2a05b9 | 6921 | nested_release_vmcs12(vmx); |
e7953d7f AG |
6922 | if (enable_shadow_vmcs) |
6923 | free_vmcs(vmx->nested.current_shadow_vmcs); | |
fe3ef05c NHE |
6924 | /* Unpin physical memory we referred to in current vmcs02 */ |
6925 | if (vmx->nested.apic_access_page) { | |
6926 | nested_release_page(vmx->nested.apic_access_page); | |
48d89b92 | 6927 | vmx->nested.apic_access_page = NULL; |
fe3ef05c | 6928 | } |
a7c0b07d WL |
6929 | if (vmx->nested.virtual_apic_page) { |
6930 | nested_release_page(vmx->nested.virtual_apic_page); | |
48d89b92 | 6931 | vmx->nested.virtual_apic_page = NULL; |
a7c0b07d | 6932 | } |
705699a1 WV |
6933 | if (vmx->nested.pi_desc_page) { |
6934 | kunmap(vmx->nested.pi_desc_page); | |
6935 | nested_release_page(vmx->nested.pi_desc_page); | |
6936 | vmx->nested.pi_desc_page = NULL; | |
6937 | vmx->nested.pi_desc = NULL; | |
6938 | } | |
ff2f6fe9 NHE |
6939 | |
6940 | nested_free_all_saved_vmcss(vmx); | |
ec378aee NHE |
6941 | } |
6942 | ||
6943 | /* Emulate the VMXOFF instruction */ | |
6944 | static int handle_vmoff(struct kvm_vcpu *vcpu) | |
6945 | { | |
6946 | if (!nested_vmx_check_permission(vcpu)) | |
6947 | return 1; | |
6948 | free_nested(to_vmx(vcpu)); | |
6949 | skip_emulated_instruction(vcpu); | |
a25eb114 | 6950 | nested_vmx_succeed(vcpu); |
ec378aee NHE |
6951 | return 1; |
6952 | } | |
6953 | ||
27d6c865 NHE |
6954 | /* Emulate the VMCLEAR instruction */ |
6955 | static int handle_vmclear(struct kvm_vcpu *vcpu) | |
6956 | { | |
6957 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
27d6c865 NHE |
6958 | gpa_t vmptr; |
6959 | struct vmcs12 *vmcs12; | |
6960 | struct page *page; | |
27d6c865 NHE |
6961 | |
6962 | if (!nested_vmx_check_permission(vcpu)) | |
6963 | return 1; | |
6964 | ||
4291b588 | 6965 | if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr)) |
27d6c865 | 6966 | return 1; |
27d6c865 | 6967 | |
9a2a05b9 | 6968 | if (vmptr == vmx->nested.current_vmptr) |
e7953d7f | 6969 | nested_release_vmcs12(vmx); |
27d6c865 NHE |
6970 | |
6971 | page = nested_get_page(vcpu, vmptr); | |
6972 | if (page == NULL) { | |
6973 | /* | |
6974 | * For accurate processor emulation, VMCLEAR beyond available | |
6975 | * physical memory should do nothing at all. However, it is | |
6976 | * possible that a nested vmx bug, not a guest hypervisor bug, | |
6977 | * resulted in this case, so let's shut down before doing any | |
6978 | * more damage: | |
6979 | */ | |
6980 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); | |
6981 | return 1; | |
6982 | } | |
6983 | vmcs12 = kmap(page); | |
6984 | vmcs12->launch_state = 0; | |
6985 | kunmap(page); | |
6986 | nested_release_page(page); | |
6987 | ||
6988 | nested_free_vmcs02(vmx, vmptr); | |
6989 | ||
6990 | skip_emulated_instruction(vcpu); | |
6991 | nested_vmx_succeed(vcpu); | |
6992 | return 1; | |
6993 | } | |
6994 | ||
cd232ad0 NHE |
6995 | static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch); |
6996 | ||
6997 | /* Emulate the VMLAUNCH instruction */ | |
6998 | static int handle_vmlaunch(struct kvm_vcpu *vcpu) | |
6999 | { | |
7000 | return nested_vmx_run(vcpu, true); | |
7001 | } | |
7002 | ||
7003 | /* Emulate the VMRESUME instruction */ | |
7004 | static int handle_vmresume(struct kvm_vcpu *vcpu) | |
7005 | { | |
7006 | ||
7007 | return nested_vmx_run(vcpu, false); | |
7008 | } | |
7009 | ||
49f705c5 NHE |
7010 | enum vmcs_field_type { |
7011 | VMCS_FIELD_TYPE_U16 = 0, | |
7012 | VMCS_FIELD_TYPE_U64 = 1, | |
7013 | VMCS_FIELD_TYPE_U32 = 2, | |
7014 | VMCS_FIELD_TYPE_NATURAL_WIDTH = 3 | |
7015 | }; | |
7016 | ||
7017 | static inline int vmcs_field_type(unsigned long field) | |
7018 | { | |
7019 | if (0x1 & field) /* the *_HIGH fields are all 32 bit */ | |
7020 | return VMCS_FIELD_TYPE_U32; | |
7021 | return (field >> 13) & 0x3 ; | |
7022 | } | |
7023 | ||
7024 | static inline int vmcs_field_readonly(unsigned long field) | |
7025 | { | |
7026 | return (((field >> 10) & 0x3) == 1); | |
7027 | } | |
7028 | ||
7029 | /* | |
7030 | * Read a vmcs12 field. Since these can have varying lengths and we return | |
7031 | * one type, we chose the biggest type (u64) and zero-extend the return value | |
7032 | * to that size. Note that the caller, handle_vmread, might need to use only | |
7033 | * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of | |
7034 | * 64-bit fields are to be returned). | |
7035 | */ | |
a2ae9df7 PB |
7036 | static inline int vmcs12_read_any(struct kvm_vcpu *vcpu, |
7037 | unsigned long field, u64 *ret) | |
49f705c5 NHE |
7038 | { |
7039 | short offset = vmcs_field_to_offset(field); | |
7040 | char *p; | |
7041 | ||
7042 | if (offset < 0) | |
a2ae9df7 | 7043 | return offset; |
49f705c5 NHE |
7044 | |
7045 | p = ((char *)(get_vmcs12(vcpu))) + offset; | |
7046 | ||
7047 | switch (vmcs_field_type(field)) { | |
7048 | case VMCS_FIELD_TYPE_NATURAL_WIDTH: | |
7049 | *ret = *((natural_width *)p); | |
a2ae9df7 | 7050 | return 0; |
49f705c5 NHE |
7051 | case VMCS_FIELD_TYPE_U16: |
7052 | *ret = *((u16 *)p); | |
a2ae9df7 | 7053 | return 0; |
49f705c5 NHE |
7054 | case VMCS_FIELD_TYPE_U32: |
7055 | *ret = *((u32 *)p); | |
a2ae9df7 | 7056 | return 0; |
49f705c5 NHE |
7057 | case VMCS_FIELD_TYPE_U64: |
7058 | *ret = *((u64 *)p); | |
a2ae9df7 | 7059 | return 0; |
49f705c5 | 7060 | default: |
a2ae9df7 PB |
7061 | WARN_ON(1); |
7062 | return -ENOENT; | |
49f705c5 NHE |
7063 | } |
7064 | } | |
7065 | ||
20b97fea | 7066 | |
a2ae9df7 PB |
7067 | static inline int vmcs12_write_any(struct kvm_vcpu *vcpu, |
7068 | unsigned long field, u64 field_value){ | |
20b97fea AG |
7069 | short offset = vmcs_field_to_offset(field); |
7070 | char *p = ((char *) get_vmcs12(vcpu)) + offset; | |
7071 | if (offset < 0) | |
a2ae9df7 | 7072 | return offset; |
20b97fea AG |
7073 | |
7074 | switch (vmcs_field_type(field)) { | |
7075 | case VMCS_FIELD_TYPE_U16: | |
7076 | *(u16 *)p = field_value; | |
a2ae9df7 | 7077 | return 0; |
20b97fea AG |
7078 | case VMCS_FIELD_TYPE_U32: |
7079 | *(u32 *)p = field_value; | |
a2ae9df7 | 7080 | return 0; |
20b97fea AG |
7081 | case VMCS_FIELD_TYPE_U64: |
7082 | *(u64 *)p = field_value; | |
a2ae9df7 | 7083 | return 0; |
20b97fea AG |
7084 | case VMCS_FIELD_TYPE_NATURAL_WIDTH: |
7085 | *(natural_width *)p = field_value; | |
a2ae9df7 | 7086 | return 0; |
20b97fea | 7087 | default: |
a2ae9df7 PB |
7088 | WARN_ON(1); |
7089 | return -ENOENT; | |
20b97fea AG |
7090 | } |
7091 | ||
7092 | } | |
7093 | ||
16f5b903 AG |
7094 | static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx) |
7095 | { | |
7096 | int i; | |
7097 | unsigned long field; | |
7098 | u64 field_value; | |
7099 | struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs; | |
c2bae893 MK |
7100 | const unsigned long *fields = shadow_read_write_fields; |
7101 | const int num_fields = max_shadow_read_write_fields; | |
16f5b903 | 7102 | |
282da870 JK |
7103 | preempt_disable(); |
7104 | ||
16f5b903 AG |
7105 | vmcs_load(shadow_vmcs); |
7106 | ||
7107 | for (i = 0; i < num_fields; i++) { | |
7108 | field = fields[i]; | |
7109 | switch (vmcs_field_type(field)) { | |
7110 | case VMCS_FIELD_TYPE_U16: | |
7111 | field_value = vmcs_read16(field); | |
7112 | break; | |
7113 | case VMCS_FIELD_TYPE_U32: | |
7114 | field_value = vmcs_read32(field); | |
7115 | break; | |
7116 | case VMCS_FIELD_TYPE_U64: | |
7117 | field_value = vmcs_read64(field); | |
7118 | break; | |
7119 | case VMCS_FIELD_TYPE_NATURAL_WIDTH: | |
7120 | field_value = vmcs_readl(field); | |
7121 | break; | |
a2ae9df7 PB |
7122 | default: |
7123 | WARN_ON(1); | |
7124 | continue; | |
16f5b903 AG |
7125 | } |
7126 | vmcs12_write_any(&vmx->vcpu, field, field_value); | |
7127 | } | |
7128 | ||
7129 | vmcs_clear(shadow_vmcs); | |
7130 | vmcs_load(vmx->loaded_vmcs->vmcs); | |
282da870 JK |
7131 | |
7132 | preempt_enable(); | |
16f5b903 AG |
7133 | } |
7134 | ||
c3114420 AG |
7135 | static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx) |
7136 | { | |
c2bae893 MK |
7137 | const unsigned long *fields[] = { |
7138 | shadow_read_write_fields, | |
7139 | shadow_read_only_fields | |
c3114420 | 7140 | }; |
c2bae893 | 7141 | const int max_fields[] = { |
c3114420 AG |
7142 | max_shadow_read_write_fields, |
7143 | max_shadow_read_only_fields | |
7144 | }; | |
7145 | int i, q; | |
7146 | unsigned long field; | |
7147 | u64 field_value = 0; | |
7148 | struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs; | |
7149 | ||
7150 | vmcs_load(shadow_vmcs); | |
7151 | ||
c2bae893 | 7152 | for (q = 0; q < ARRAY_SIZE(fields); q++) { |
c3114420 AG |
7153 | for (i = 0; i < max_fields[q]; i++) { |
7154 | field = fields[q][i]; | |
7155 | vmcs12_read_any(&vmx->vcpu, field, &field_value); | |
7156 | ||
7157 | switch (vmcs_field_type(field)) { | |
7158 | case VMCS_FIELD_TYPE_U16: | |
7159 | vmcs_write16(field, (u16)field_value); | |
7160 | break; | |
7161 | case VMCS_FIELD_TYPE_U32: | |
7162 | vmcs_write32(field, (u32)field_value); | |
7163 | break; | |
7164 | case VMCS_FIELD_TYPE_U64: | |
7165 | vmcs_write64(field, (u64)field_value); | |
7166 | break; | |
7167 | case VMCS_FIELD_TYPE_NATURAL_WIDTH: | |
7168 | vmcs_writel(field, (long)field_value); | |
7169 | break; | |
a2ae9df7 PB |
7170 | default: |
7171 | WARN_ON(1); | |
7172 | break; | |
c3114420 AG |
7173 | } |
7174 | } | |
7175 | } | |
7176 | ||
7177 | vmcs_clear(shadow_vmcs); | |
7178 | vmcs_load(vmx->loaded_vmcs->vmcs); | |
7179 | } | |
7180 | ||
49f705c5 NHE |
7181 | /* |
7182 | * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was | |
7183 | * used before) all generate the same failure when it is missing. | |
7184 | */ | |
7185 | static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu) | |
7186 | { | |
7187 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
7188 | if (vmx->nested.current_vmptr == -1ull) { | |
7189 | nested_vmx_failInvalid(vcpu); | |
7190 | skip_emulated_instruction(vcpu); | |
7191 | return 0; | |
7192 | } | |
7193 | return 1; | |
7194 | } | |
7195 | ||
7196 | static int handle_vmread(struct kvm_vcpu *vcpu) | |
7197 | { | |
7198 | unsigned long field; | |
7199 | u64 field_value; | |
7200 | unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); | |
7201 | u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO); | |
7202 | gva_t gva = 0; | |
7203 | ||
7204 | if (!nested_vmx_check_permission(vcpu) || | |
7205 | !nested_vmx_check_vmcs12(vcpu)) | |
7206 | return 1; | |
7207 | ||
7208 | /* Decode instruction info and find the field to read */ | |
27e6fb5d | 7209 | field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf)); |
49f705c5 | 7210 | /* Read the field, zero-extended to a u64 field_value */ |
a2ae9df7 | 7211 | if (vmcs12_read_any(vcpu, field, &field_value) < 0) { |
49f705c5 NHE |
7212 | nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT); |
7213 | skip_emulated_instruction(vcpu); | |
7214 | return 1; | |
7215 | } | |
7216 | /* | |
7217 | * Now copy part of this value to register or memory, as requested. | |
7218 | * Note that the number of bits actually copied is 32 or 64 depending | |
7219 | * on the guest's mode (32 or 64 bit), not on the given field's length. | |
7220 | */ | |
7221 | if (vmx_instruction_info & (1u << 10)) { | |
27e6fb5d | 7222 | kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf), |
49f705c5 NHE |
7223 | field_value); |
7224 | } else { | |
7225 | if (get_vmx_mem_address(vcpu, exit_qualification, | |
f9eb4af6 | 7226 | vmx_instruction_info, true, &gva)) |
49f705c5 NHE |
7227 | return 1; |
7228 | /* _system ok, as nested_vmx_check_permission verified cpl=0 */ | |
7229 | kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva, | |
7230 | &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL); | |
7231 | } | |
7232 | ||
7233 | nested_vmx_succeed(vcpu); | |
7234 | skip_emulated_instruction(vcpu); | |
7235 | return 1; | |
7236 | } | |
7237 | ||
7238 | ||
7239 | static int handle_vmwrite(struct kvm_vcpu *vcpu) | |
7240 | { | |
7241 | unsigned long field; | |
7242 | gva_t gva; | |
7243 | unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); | |
7244 | u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO); | |
49f705c5 NHE |
7245 | /* The value to write might be 32 or 64 bits, depending on L1's long |
7246 | * mode, and eventually we need to write that into a field of several | |
7247 | * possible lengths. The code below first zero-extends the value to 64 | |
7248 | * bit (field_value), and then copies only the approriate number of | |
7249 | * bits into the vmcs12 field. | |
7250 | */ | |
7251 | u64 field_value = 0; | |
7252 | struct x86_exception e; | |
7253 | ||
7254 | if (!nested_vmx_check_permission(vcpu) || | |
7255 | !nested_vmx_check_vmcs12(vcpu)) | |
7256 | return 1; | |
7257 | ||
7258 | if (vmx_instruction_info & (1u << 10)) | |
27e6fb5d | 7259 | field_value = kvm_register_readl(vcpu, |
49f705c5 NHE |
7260 | (((vmx_instruction_info) >> 3) & 0xf)); |
7261 | else { | |
7262 | if (get_vmx_mem_address(vcpu, exit_qualification, | |
f9eb4af6 | 7263 | vmx_instruction_info, false, &gva)) |
49f705c5 NHE |
7264 | return 1; |
7265 | if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, | |
27e6fb5d | 7266 | &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) { |
49f705c5 NHE |
7267 | kvm_inject_page_fault(vcpu, &e); |
7268 | return 1; | |
7269 | } | |
7270 | } | |
7271 | ||
7272 | ||
27e6fb5d | 7273 | field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf)); |
49f705c5 NHE |
7274 | if (vmcs_field_readonly(field)) { |
7275 | nested_vmx_failValid(vcpu, | |
7276 | VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT); | |
7277 | skip_emulated_instruction(vcpu); | |
7278 | return 1; | |
7279 | } | |
7280 | ||
a2ae9df7 | 7281 | if (vmcs12_write_any(vcpu, field, field_value) < 0) { |
49f705c5 NHE |
7282 | nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT); |
7283 | skip_emulated_instruction(vcpu); | |
7284 | return 1; | |
7285 | } | |
7286 | ||
7287 | nested_vmx_succeed(vcpu); | |
7288 | skip_emulated_instruction(vcpu); | |
7289 | return 1; | |
7290 | } | |
7291 | ||
63846663 NHE |
7292 | /* Emulate the VMPTRLD instruction */ |
7293 | static int handle_vmptrld(struct kvm_vcpu *vcpu) | |
7294 | { | |
7295 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
63846663 | 7296 | gpa_t vmptr; |
63846663 NHE |
7297 | |
7298 | if (!nested_vmx_check_permission(vcpu)) | |
7299 | return 1; | |
7300 | ||
4291b588 | 7301 | if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr)) |
63846663 | 7302 | return 1; |
63846663 NHE |
7303 | |
7304 | if (vmx->nested.current_vmptr != vmptr) { | |
7305 | struct vmcs12 *new_vmcs12; | |
7306 | struct page *page; | |
7307 | page = nested_get_page(vcpu, vmptr); | |
7308 | if (page == NULL) { | |
7309 | nested_vmx_failInvalid(vcpu); | |
7310 | skip_emulated_instruction(vcpu); | |
7311 | return 1; | |
7312 | } | |
7313 | new_vmcs12 = kmap(page); | |
7314 | if (new_vmcs12->revision_id != VMCS12_REVISION) { | |
7315 | kunmap(page); | |
7316 | nested_release_page_clean(page); | |
7317 | nested_vmx_failValid(vcpu, | |
7318 | VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID); | |
7319 | skip_emulated_instruction(vcpu); | |
7320 | return 1; | |
7321 | } | |
63846663 | 7322 | |
9a2a05b9 | 7323 | nested_release_vmcs12(vmx); |
63846663 NHE |
7324 | vmx->nested.current_vmptr = vmptr; |
7325 | vmx->nested.current_vmcs12 = new_vmcs12; | |
7326 | vmx->nested.current_vmcs12_page = page; | |
012f83cb | 7327 | if (enable_shadow_vmcs) { |
7ec36296 XG |
7328 | vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL, |
7329 | SECONDARY_EXEC_SHADOW_VMCS); | |
8a1b9dd0 AG |
7330 | vmcs_write64(VMCS_LINK_POINTER, |
7331 | __pa(vmx->nested.current_shadow_vmcs)); | |
012f83cb AG |
7332 | vmx->nested.sync_shadow_vmcs = true; |
7333 | } | |
63846663 NHE |
7334 | } |
7335 | ||
7336 | nested_vmx_succeed(vcpu); | |
7337 | skip_emulated_instruction(vcpu); | |
7338 | return 1; | |
7339 | } | |
7340 | ||
6a4d7550 NHE |
7341 | /* Emulate the VMPTRST instruction */ |
7342 | static int handle_vmptrst(struct kvm_vcpu *vcpu) | |
7343 | { | |
7344 | unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); | |
7345 | u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO); | |
7346 | gva_t vmcs_gva; | |
7347 | struct x86_exception e; | |
7348 | ||
7349 | if (!nested_vmx_check_permission(vcpu)) | |
7350 | return 1; | |
7351 | ||
7352 | if (get_vmx_mem_address(vcpu, exit_qualification, | |
f9eb4af6 | 7353 | vmx_instruction_info, true, &vmcs_gva)) |
6a4d7550 NHE |
7354 | return 1; |
7355 | /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */ | |
7356 | if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva, | |
7357 | (void *)&to_vmx(vcpu)->nested.current_vmptr, | |
7358 | sizeof(u64), &e)) { | |
7359 | kvm_inject_page_fault(vcpu, &e); | |
7360 | return 1; | |
7361 | } | |
7362 | nested_vmx_succeed(vcpu); | |
7363 | skip_emulated_instruction(vcpu); | |
7364 | return 1; | |
7365 | } | |
7366 | ||
bfd0a56b NHE |
7367 | /* Emulate the INVEPT instruction */ |
7368 | static int handle_invept(struct kvm_vcpu *vcpu) | |
7369 | { | |
b9c237bb | 7370 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
bfd0a56b NHE |
7371 | u32 vmx_instruction_info, types; |
7372 | unsigned long type; | |
7373 | gva_t gva; | |
7374 | struct x86_exception e; | |
7375 | struct { | |
7376 | u64 eptp, gpa; | |
7377 | } operand; | |
bfd0a56b | 7378 | |
b9c237bb WV |
7379 | if (!(vmx->nested.nested_vmx_secondary_ctls_high & |
7380 | SECONDARY_EXEC_ENABLE_EPT) || | |
7381 | !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) { | |
bfd0a56b NHE |
7382 | kvm_queue_exception(vcpu, UD_VECTOR); |
7383 | return 1; | |
7384 | } | |
7385 | ||
7386 | if (!nested_vmx_check_permission(vcpu)) | |
7387 | return 1; | |
7388 | ||
7389 | if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) { | |
7390 | kvm_queue_exception(vcpu, UD_VECTOR); | |
7391 | return 1; | |
7392 | } | |
7393 | ||
7394 | vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO); | |
27e6fb5d | 7395 | type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf); |
bfd0a56b | 7396 | |
b9c237bb | 7397 | types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6; |
bfd0a56b NHE |
7398 | |
7399 | if (!(types & (1UL << type))) { | |
7400 | nested_vmx_failValid(vcpu, | |
7401 | VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID); | |
2849eb4f | 7402 | skip_emulated_instruction(vcpu); |
bfd0a56b NHE |
7403 | return 1; |
7404 | } | |
7405 | ||
7406 | /* According to the Intel VMX instruction reference, the memory | |
7407 | * operand is read even if it isn't needed (e.g., for type==global) | |
7408 | */ | |
7409 | if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION), | |
f9eb4af6 | 7410 | vmx_instruction_info, false, &gva)) |
bfd0a56b NHE |
7411 | return 1; |
7412 | if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand, | |
7413 | sizeof(operand), &e)) { | |
7414 | kvm_inject_page_fault(vcpu, &e); | |
7415 | return 1; | |
7416 | } | |
7417 | ||
7418 | switch (type) { | |
bfd0a56b NHE |
7419 | case VMX_EPT_EXTENT_GLOBAL: |
7420 | kvm_mmu_sync_roots(vcpu); | |
77c3913b | 7421 | kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
bfd0a56b NHE |
7422 | nested_vmx_succeed(vcpu); |
7423 | break; | |
7424 | default: | |
4b855078 | 7425 | /* Trap single context invalidation invept calls */ |
bfd0a56b NHE |
7426 | BUG_ON(1); |
7427 | break; | |
7428 | } | |
7429 | ||
7430 | skip_emulated_instruction(vcpu); | |
7431 | return 1; | |
7432 | } | |
7433 | ||
a642fc30 PM |
7434 | static int handle_invvpid(struct kvm_vcpu *vcpu) |
7435 | { | |
99b83ac8 WL |
7436 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
7437 | u32 vmx_instruction_info; | |
7438 | unsigned long type, types; | |
7439 | gva_t gva; | |
7440 | struct x86_exception e; | |
7441 | int vpid; | |
7442 | ||
7443 | if (!(vmx->nested.nested_vmx_secondary_ctls_high & | |
7444 | SECONDARY_EXEC_ENABLE_VPID) || | |
7445 | !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) { | |
7446 | kvm_queue_exception(vcpu, UD_VECTOR); | |
7447 | return 1; | |
7448 | } | |
7449 | ||
7450 | if (!nested_vmx_check_permission(vcpu)) | |
7451 | return 1; | |
7452 | ||
7453 | vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO); | |
7454 | type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf); | |
7455 | ||
7456 | types = (vmx->nested.nested_vmx_vpid_caps >> 8) & 0x7; | |
7457 | ||
7458 | if (!(types & (1UL << type))) { | |
7459 | nested_vmx_failValid(vcpu, | |
7460 | VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID); | |
f6870ee9 | 7461 | skip_emulated_instruction(vcpu); |
99b83ac8 WL |
7462 | return 1; |
7463 | } | |
7464 | ||
7465 | /* according to the intel vmx instruction reference, the memory | |
7466 | * operand is read even if it isn't needed (e.g., for type==global) | |
7467 | */ | |
7468 | if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION), | |
7469 | vmx_instruction_info, false, &gva)) | |
7470 | return 1; | |
7471 | if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid, | |
7472 | sizeof(u32), &e)) { | |
7473 | kvm_inject_page_fault(vcpu, &e); | |
7474 | return 1; | |
7475 | } | |
7476 | ||
7477 | switch (type) { | |
7478 | case VMX_VPID_EXTENT_ALL_CONTEXT: | |
5c614b35 | 7479 | __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02); |
99b83ac8 WL |
7480 | nested_vmx_succeed(vcpu); |
7481 | break; | |
7482 | default: | |
7483 | /* Trap single context invalidation invvpid calls */ | |
7484 | BUG_ON(1); | |
7485 | break; | |
7486 | } | |
7487 | ||
7488 | skip_emulated_instruction(vcpu); | |
a642fc30 PM |
7489 | return 1; |
7490 | } | |
7491 | ||
843e4330 KH |
7492 | static int handle_pml_full(struct kvm_vcpu *vcpu) |
7493 | { | |
7494 | unsigned long exit_qualification; | |
7495 | ||
7496 | trace_kvm_pml_full(vcpu->vcpu_id); | |
7497 | ||
7498 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); | |
7499 | ||
7500 | /* | |
7501 | * PML buffer FULL happened while executing iret from NMI, | |
7502 | * "blocked by NMI" bit has to be set before next VM entry. | |
7503 | */ | |
7504 | if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) && | |
7505 | cpu_has_virtual_nmis() && | |
7506 | (exit_qualification & INTR_INFO_UNBLOCK_NMI)) | |
7507 | vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, | |
7508 | GUEST_INTR_STATE_NMI); | |
7509 | ||
7510 | /* | |
7511 | * PML buffer already flushed at beginning of VMEXIT. Nothing to do | |
7512 | * here.., and there's no userspace involvement needed for PML. | |
7513 | */ | |
7514 | return 1; | |
7515 | } | |
7516 | ||
8b3e34e4 XG |
7517 | static int handle_pcommit(struct kvm_vcpu *vcpu) |
7518 | { | |
7519 | /* we never catch pcommit instruct for L1 guest. */ | |
7520 | WARN_ON(1); | |
7521 | return 1; | |
7522 | } | |
7523 | ||
6aa8b732 AK |
7524 | /* |
7525 | * The exit handlers return 1 if the exit was handled fully and guest execution | |
7526 | * may resume. Otherwise they set the kvm_run parameter to indicate what needs | |
7527 | * to be done to userspace and return 0. | |
7528 | */ | |
772e0318 | 7529 | static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = { |
6aa8b732 AK |
7530 | [EXIT_REASON_EXCEPTION_NMI] = handle_exception, |
7531 | [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt, | |
988ad74f | 7532 | [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault, |
f08864b4 | 7533 | [EXIT_REASON_NMI_WINDOW] = handle_nmi_window, |
6aa8b732 | 7534 | [EXIT_REASON_IO_INSTRUCTION] = handle_io, |
6aa8b732 AK |
7535 | [EXIT_REASON_CR_ACCESS] = handle_cr, |
7536 | [EXIT_REASON_DR_ACCESS] = handle_dr, | |
7537 | [EXIT_REASON_CPUID] = handle_cpuid, | |
7538 | [EXIT_REASON_MSR_READ] = handle_rdmsr, | |
7539 | [EXIT_REASON_MSR_WRITE] = handle_wrmsr, | |
7540 | [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window, | |
7541 | [EXIT_REASON_HLT] = handle_halt, | |
ec25d5e6 | 7542 | [EXIT_REASON_INVD] = handle_invd, |
a7052897 | 7543 | [EXIT_REASON_INVLPG] = handle_invlpg, |
fee84b07 | 7544 | [EXIT_REASON_RDPMC] = handle_rdpmc, |
c21415e8 | 7545 | [EXIT_REASON_VMCALL] = handle_vmcall, |
27d6c865 | 7546 | [EXIT_REASON_VMCLEAR] = handle_vmclear, |
cd232ad0 | 7547 | [EXIT_REASON_VMLAUNCH] = handle_vmlaunch, |
63846663 | 7548 | [EXIT_REASON_VMPTRLD] = handle_vmptrld, |
6a4d7550 | 7549 | [EXIT_REASON_VMPTRST] = handle_vmptrst, |
49f705c5 | 7550 | [EXIT_REASON_VMREAD] = handle_vmread, |
cd232ad0 | 7551 | [EXIT_REASON_VMRESUME] = handle_vmresume, |
49f705c5 | 7552 | [EXIT_REASON_VMWRITE] = handle_vmwrite, |
ec378aee NHE |
7553 | [EXIT_REASON_VMOFF] = handle_vmoff, |
7554 | [EXIT_REASON_VMON] = handle_vmon, | |
f78e0e2e SY |
7555 | [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold, |
7556 | [EXIT_REASON_APIC_ACCESS] = handle_apic_access, | |
83d4c286 | 7557 | [EXIT_REASON_APIC_WRITE] = handle_apic_write, |
c7c9c56c | 7558 | [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced, |
e5edaa01 | 7559 | [EXIT_REASON_WBINVD] = handle_wbinvd, |
2acf923e | 7560 | [EXIT_REASON_XSETBV] = handle_xsetbv, |
37817f29 | 7561 | [EXIT_REASON_TASK_SWITCH] = handle_task_switch, |
a0861c02 | 7562 | [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check, |
68f89400 MT |
7563 | [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation, |
7564 | [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig, | |
4b8d54f9 | 7565 | [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause, |
87c00572 | 7566 | [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait, |
5f3d45e7 | 7567 | [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap, |
87c00572 | 7568 | [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor, |
bfd0a56b | 7569 | [EXIT_REASON_INVEPT] = handle_invept, |
a642fc30 | 7570 | [EXIT_REASON_INVVPID] = handle_invvpid, |
f53cd63c WL |
7571 | [EXIT_REASON_XSAVES] = handle_xsaves, |
7572 | [EXIT_REASON_XRSTORS] = handle_xrstors, | |
843e4330 | 7573 | [EXIT_REASON_PML_FULL] = handle_pml_full, |
8b3e34e4 | 7574 | [EXIT_REASON_PCOMMIT] = handle_pcommit, |
6aa8b732 AK |
7575 | }; |
7576 | ||
7577 | static const int kvm_vmx_max_exit_handlers = | |
50a3485c | 7578 | ARRAY_SIZE(kvm_vmx_exit_handlers); |
6aa8b732 | 7579 | |
908a7bdd JK |
7580 | static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu, |
7581 | struct vmcs12 *vmcs12) | |
7582 | { | |
7583 | unsigned long exit_qualification; | |
7584 | gpa_t bitmap, last_bitmap; | |
7585 | unsigned int port; | |
7586 | int size; | |
7587 | u8 b; | |
7588 | ||
908a7bdd | 7589 | if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS)) |
2f0a6397 | 7590 | return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING); |
908a7bdd JK |
7591 | |
7592 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); | |
7593 | ||
7594 | port = exit_qualification >> 16; | |
7595 | size = (exit_qualification & 7) + 1; | |
7596 | ||
7597 | last_bitmap = (gpa_t)-1; | |
7598 | b = -1; | |
7599 | ||
7600 | while (size > 0) { | |
7601 | if (port < 0x8000) | |
7602 | bitmap = vmcs12->io_bitmap_a; | |
7603 | else if (port < 0x10000) | |
7604 | bitmap = vmcs12->io_bitmap_b; | |
7605 | else | |
1d804d07 | 7606 | return true; |
908a7bdd JK |
7607 | bitmap += (port & 0x7fff) / 8; |
7608 | ||
7609 | if (last_bitmap != bitmap) | |
54bf36aa | 7610 | if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1)) |
1d804d07 | 7611 | return true; |
908a7bdd | 7612 | if (b & (1 << (port & 7))) |
1d804d07 | 7613 | return true; |
908a7bdd JK |
7614 | |
7615 | port++; | |
7616 | size--; | |
7617 | last_bitmap = bitmap; | |
7618 | } | |
7619 | ||
1d804d07 | 7620 | return false; |
908a7bdd JK |
7621 | } |
7622 | ||
644d711a NHE |
7623 | /* |
7624 | * Return 1 if we should exit from L2 to L1 to handle an MSR access access, | |
7625 | * rather than handle it ourselves in L0. I.e., check whether L1 expressed | |
7626 | * disinterest in the current event (read or write a specific MSR) by using an | |
7627 | * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps. | |
7628 | */ | |
7629 | static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu, | |
7630 | struct vmcs12 *vmcs12, u32 exit_reason) | |
7631 | { | |
7632 | u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX]; | |
7633 | gpa_t bitmap; | |
7634 | ||
cbd29cb6 | 7635 | if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS)) |
1d804d07 | 7636 | return true; |
644d711a NHE |
7637 | |
7638 | /* | |
7639 | * The MSR_BITMAP page is divided into four 1024-byte bitmaps, | |
7640 | * for the four combinations of read/write and low/high MSR numbers. | |
7641 | * First we need to figure out which of the four to use: | |
7642 | */ | |
7643 | bitmap = vmcs12->msr_bitmap; | |
7644 | if (exit_reason == EXIT_REASON_MSR_WRITE) | |
7645 | bitmap += 2048; | |
7646 | if (msr_index >= 0xc0000000) { | |
7647 | msr_index -= 0xc0000000; | |
7648 | bitmap += 1024; | |
7649 | } | |
7650 | ||
7651 | /* Then read the msr_index'th bit from this bitmap: */ | |
7652 | if (msr_index < 1024*8) { | |
7653 | unsigned char b; | |
54bf36aa | 7654 | if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1)) |
1d804d07 | 7655 | return true; |
644d711a NHE |
7656 | return 1 & (b >> (msr_index & 7)); |
7657 | } else | |
1d804d07 | 7658 | return true; /* let L1 handle the wrong parameter */ |
644d711a NHE |
7659 | } |
7660 | ||
7661 | /* | |
7662 | * Return 1 if we should exit from L2 to L1 to handle a CR access exit, | |
7663 | * rather than handle it ourselves in L0. I.e., check if L1 wanted to | |
7664 | * intercept (via guest_host_mask etc.) the current event. | |
7665 | */ | |
7666 | static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu, | |
7667 | struct vmcs12 *vmcs12) | |
7668 | { | |
7669 | unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); | |
7670 | int cr = exit_qualification & 15; | |
7671 | int reg = (exit_qualification >> 8) & 15; | |
1e32c079 | 7672 | unsigned long val = kvm_register_readl(vcpu, reg); |
644d711a NHE |
7673 | |
7674 | switch ((exit_qualification >> 4) & 3) { | |
7675 | case 0: /* mov to cr */ | |
7676 | switch (cr) { | |
7677 | case 0: | |
7678 | if (vmcs12->cr0_guest_host_mask & | |
7679 | (val ^ vmcs12->cr0_read_shadow)) | |
1d804d07 | 7680 | return true; |
644d711a NHE |
7681 | break; |
7682 | case 3: | |
7683 | if ((vmcs12->cr3_target_count >= 1 && | |
7684 | vmcs12->cr3_target_value0 == val) || | |
7685 | (vmcs12->cr3_target_count >= 2 && | |
7686 | vmcs12->cr3_target_value1 == val) || | |
7687 | (vmcs12->cr3_target_count >= 3 && | |
7688 | vmcs12->cr3_target_value2 == val) || | |
7689 | (vmcs12->cr3_target_count >= 4 && | |
7690 | vmcs12->cr3_target_value3 == val)) | |
1d804d07 | 7691 | return false; |
644d711a | 7692 | if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING)) |
1d804d07 | 7693 | return true; |
644d711a NHE |
7694 | break; |
7695 | case 4: | |
7696 | if (vmcs12->cr4_guest_host_mask & | |
7697 | (vmcs12->cr4_read_shadow ^ val)) | |
1d804d07 | 7698 | return true; |
644d711a NHE |
7699 | break; |
7700 | case 8: | |
7701 | if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING)) | |
1d804d07 | 7702 | return true; |
644d711a NHE |
7703 | break; |
7704 | } | |
7705 | break; | |
7706 | case 2: /* clts */ | |
7707 | if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) && | |
7708 | (vmcs12->cr0_read_shadow & X86_CR0_TS)) | |
1d804d07 | 7709 | return true; |
644d711a NHE |
7710 | break; |
7711 | case 1: /* mov from cr */ | |
7712 | switch (cr) { | |
7713 | case 3: | |
7714 | if (vmcs12->cpu_based_vm_exec_control & | |
7715 | CPU_BASED_CR3_STORE_EXITING) | |
1d804d07 | 7716 | return true; |
644d711a NHE |
7717 | break; |
7718 | case 8: | |
7719 | if (vmcs12->cpu_based_vm_exec_control & | |
7720 | CPU_BASED_CR8_STORE_EXITING) | |
1d804d07 | 7721 | return true; |
644d711a NHE |
7722 | break; |
7723 | } | |
7724 | break; | |
7725 | case 3: /* lmsw */ | |
7726 | /* | |
7727 | * lmsw can change bits 1..3 of cr0, and only set bit 0 of | |
7728 | * cr0. Other attempted changes are ignored, with no exit. | |
7729 | */ | |
7730 | if (vmcs12->cr0_guest_host_mask & 0xe & | |
7731 | (val ^ vmcs12->cr0_read_shadow)) | |
1d804d07 | 7732 | return true; |
644d711a NHE |
7733 | if ((vmcs12->cr0_guest_host_mask & 0x1) && |
7734 | !(vmcs12->cr0_read_shadow & 0x1) && | |
7735 | (val & 0x1)) | |
1d804d07 | 7736 | return true; |
644d711a NHE |
7737 | break; |
7738 | } | |
1d804d07 | 7739 | return false; |
644d711a NHE |
7740 | } |
7741 | ||
7742 | /* | |
7743 | * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we | |
7744 | * should handle it ourselves in L0 (and then continue L2). Only call this | |
7745 | * when in is_guest_mode (L2). | |
7746 | */ | |
7747 | static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu) | |
7748 | { | |
644d711a NHE |
7749 | u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO); |
7750 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
7751 | struct vmcs12 *vmcs12 = get_vmcs12(vcpu); | |
957c897e | 7752 | u32 exit_reason = vmx->exit_reason; |
644d711a | 7753 | |
542060ea JK |
7754 | trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason, |
7755 | vmcs_readl(EXIT_QUALIFICATION), | |
7756 | vmx->idt_vectoring_info, | |
7757 | intr_info, | |
7758 | vmcs_read32(VM_EXIT_INTR_ERROR_CODE), | |
7759 | KVM_ISA_VMX); | |
7760 | ||
644d711a | 7761 | if (vmx->nested.nested_run_pending) |
1d804d07 | 7762 | return false; |
644d711a NHE |
7763 | |
7764 | if (unlikely(vmx->fail)) { | |
bd80158a JK |
7765 | pr_info_ratelimited("%s failed vm entry %x\n", __func__, |
7766 | vmcs_read32(VM_INSTRUCTION_ERROR)); | |
1d804d07 | 7767 | return true; |
644d711a NHE |
7768 | } |
7769 | ||
7770 | switch (exit_reason) { | |
7771 | case EXIT_REASON_EXCEPTION_NMI: | |
7772 | if (!is_exception(intr_info)) | |
1d804d07 | 7773 | return false; |
644d711a NHE |
7774 | else if (is_page_fault(intr_info)) |
7775 | return enable_ept; | |
e504c909 | 7776 | else if (is_no_device(intr_info) && |
ccf9844e | 7777 | !(vmcs12->guest_cr0 & X86_CR0_TS)) |
1d804d07 | 7778 | return false; |
6f05485d JK |
7779 | else if (is_debug(intr_info) && |
7780 | vcpu->guest_debug & | |
7781 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) | |
7782 | return false; | |
7783 | else if (is_breakpoint(intr_info) && | |
7784 | vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) | |
7785 | return false; | |
644d711a NHE |
7786 | return vmcs12->exception_bitmap & |
7787 | (1u << (intr_info & INTR_INFO_VECTOR_MASK)); | |
7788 | case EXIT_REASON_EXTERNAL_INTERRUPT: | |
1d804d07 | 7789 | return false; |
644d711a | 7790 | case EXIT_REASON_TRIPLE_FAULT: |
1d804d07 | 7791 | return true; |
644d711a | 7792 | case EXIT_REASON_PENDING_INTERRUPT: |
3b656cf7 | 7793 | return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING); |
644d711a | 7794 | case EXIT_REASON_NMI_WINDOW: |
3b656cf7 | 7795 | return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING); |
644d711a | 7796 | case EXIT_REASON_TASK_SWITCH: |
1d804d07 | 7797 | return true; |
644d711a | 7798 | case EXIT_REASON_CPUID: |
bc613494 | 7799 | if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa) |
1d804d07 JP |
7800 | return false; |
7801 | return true; | |
644d711a NHE |
7802 | case EXIT_REASON_HLT: |
7803 | return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING); | |
7804 | case EXIT_REASON_INVD: | |
1d804d07 | 7805 | return true; |
644d711a NHE |
7806 | case EXIT_REASON_INVLPG: |
7807 | return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING); | |
7808 | case EXIT_REASON_RDPMC: | |
7809 | return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING); | |
b3a2a907 | 7810 | case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP: |
644d711a NHE |
7811 | return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING); |
7812 | case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR: | |
7813 | case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD: | |
7814 | case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD: | |
7815 | case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE: | |
7816 | case EXIT_REASON_VMOFF: case EXIT_REASON_VMON: | |
a642fc30 | 7817 | case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID: |
644d711a NHE |
7818 | /* |
7819 | * VMX instructions trap unconditionally. This allows L1 to | |
7820 | * emulate them for its L2 guest, i.e., allows 3-level nesting! | |
7821 | */ | |
1d804d07 | 7822 | return true; |
644d711a NHE |
7823 | case EXIT_REASON_CR_ACCESS: |
7824 | return nested_vmx_exit_handled_cr(vcpu, vmcs12); | |
7825 | case EXIT_REASON_DR_ACCESS: | |
7826 | return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING); | |
7827 | case EXIT_REASON_IO_INSTRUCTION: | |
908a7bdd | 7828 | return nested_vmx_exit_handled_io(vcpu, vmcs12); |
644d711a NHE |
7829 | case EXIT_REASON_MSR_READ: |
7830 | case EXIT_REASON_MSR_WRITE: | |
7831 | return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason); | |
7832 | case EXIT_REASON_INVALID_STATE: | |
1d804d07 | 7833 | return true; |
644d711a NHE |
7834 | case EXIT_REASON_MWAIT_INSTRUCTION: |
7835 | return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING); | |
5f3d45e7 MD |
7836 | case EXIT_REASON_MONITOR_TRAP_FLAG: |
7837 | return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG); | |
644d711a NHE |
7838 | case EXIT_REASON_MONITOR_INSTRUCTION: |
7839 | return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING); | |
7840 | case EXIT_REASON_PAUSE_INSTRUCTION: | |
7841 | return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) || | |
7842 | nested_cpu_has2(vmcs12, | |
7843 | SECONDARY_EXEC_PAUSE_LOOP_EXITING); | |
7844 | case EXIT_REASON_MCE_DURING_VMENTRY: | |
1d804d07 | 7845 | return false; |
644d711a | 7846 | case EXIT_REASON_TPR_BELOW_THRESHOLD: |
a7c0b07d | 7847 | return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW); |
644d711a NHE |
7848 | case EXIT_REASON_APIC_ACCESS: |
7849 | return nested_cpu_has2(vmcs12, | |
7850 | SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES); | |
82f0dd4b | 7851 | case EXIT_REASON_APIC_WRITE: |
608406e2 WV |
7852 | case EXIT_REASON_EOI_INDUCED: |
7853 | /* apic_write and eoi_induced should exit unconditionally. */ | |
1d804d07 | 7854 | return true; |
644d711a | 7855 | case EXIT_REASON_EPT_VIOLATION: |
2b1be677 NHE |
7856 | /* |
7857 | * L0 always deals with the EPT violation. If nested EPT is | |
7858 | * used, and the nested mmu code discovers that the address is | |
7859 | * missing in the guest EPT table (EPT12), the EPT violation | |
7860 | * will be injected with nested_ept_inject_page_fault() | |
7861 | */ | |
1d804d07 | 7862 | return false; |
644d711a | 7863 | case EXIT_REASON_EPT_MISCONFIG: |
2b1be677 NHE |
7864 | /* |
7865 | * L2 never uses directly L1's EPT, but rather L0's own EPT | |
7866 | * table (shadow on EPT) or a merged EPT table that L0 built | |
7867 | * (EPT on EPT). So any problems with the structure of the | |
7868 | * table is L0's fault. | |
7869 | */ | |
1d804d07 | 7870 | return false; |
644d711a NHE |
7871 | case EXIT_REASON_WBINVD: |
7872 | return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING); | |
7873 | case EXIT_REASON_XSETBV: | |
1d804d07 | 7874 | return true; |
81dc01f7 WL |
7875 | case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS: |
7876 | /* | |
7877 | * This should never happen, since it is not possible to | |
7878 | * set XSS to a non-zero value---neither in L1 nor in L2. | |
7879 | * If if it were, XSS would have to be checked against | |
7880 | * the XSS exit bitmap in vmcs12. | |
7881 | */ | |
7882 | return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES); | |
8b3e34e4 XG |
7883 | case EXIT_REASON_PCOMMIT: |
7884 | return nested_cpu_has2(vmcs12, SECONDARY_EXEC_PCOMMIT); | |
644d711a | 7885 | default: |
1d804d07 | 7886 | return true; |
644d711a NHE |
7887 | } |
7888 | } | |
7889 | ||
586f9607 AK |
7890 | static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2) |
7891 | { | |
7892 | *info1 = vmcs_readl(EXIT_QUALIFICATION); | |
7893 | *info2 = vmcs_read32(VM_EXIT_INTR_INFO); | |
7894 | } | |
7895 | ||
a3eaa864 | 7896 | static int vmx_create_pml_buffer(struct vcpu_vmx *vmx) |
843e4330 KH |
7897 | { |
7898 | struct page *pml_pg; | |
843e4330 KH |
7899 | |
7900 | pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
7901 | if (!pml_pg) | |
7902 | return -ENOMEM; | |
7903 | ||
7904 | vmx->pml_pg = pml_pg; | |
7905 | ||
7906 | vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg)); | |
7907 | vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1); | |
7908 | ||
843e4330 KH |
7909 | return 0; |
7910 | } | |
7911 | ||
a3eaa864 | 7912 | static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx) |
843e4330 | 7913 | { |
a3eaa864 KH |
7914 | if (vmx->pml_pg) { |
7915 | __free_page(vmx->pml_pg); | |
7916 | vmx->pml_pg = NULL; | |
7917 | } | |
843e4330 KH |
7918 | } |
7919 | ||
54bf36aa | 7920 | static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu) |
843e4330 | 7921 | { |
54bf36aa | 7922 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
843e4330 KH |
7923 | u64 *pml_buf; |
7924 | u16 pml_idx; | |
7925 | ||
7926 | pml_idx = vmcs_read16(GUEST_PML_INDEX); | |
7927 | ||
7928 | /* Do nothing if PML buffer is empty */ | |
7929 | if (pml_idx == (PML_ENTITY_NUM - 1)) | |
7930 | return; | |
7931 | ||
7932 | /* PML index always points to next available PML buffer entity */ | |
7933 | if (pml_idx >= PML_ENTITY_NUM) | |
7934 | pml_idx = 0; | |
7935 | else | |
7936 | pml_idx++; | |
7937 | ||
7938 | pml_buf = page_address(vmx->pml_pg); | |
7939 | for (; pml_idx < PML_ENTITY_NUM; pml_idx++) { | |
7940 | u64 gpa; | |
7941 | ||
7942 | gpa = pml_buf[pml_idx]; | |
7943 | WARN_ON(gpa & (PAGE_SIZE - 1)); | |
54bf36aa | 7944 | kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT); |
843e4330 KH |
7945 | } |
7946 | ||
7947 | /* reset PML index */ | |
7948 | vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1); | |
7949 | } | |
7950 | ||
7951 | /* | |
7952 | * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap. | |
7953 | * Called before reporting dirty_bitmap to userspace. | |
7954 | */ | |
7955 | static void kvm_flush_pml_buffers(struct kvm *kvm) | |
7956 | { | |
7957 | int i; | |
7958 | struct kvm_vcpu *vcpu; | |
7959 | /* | |
7960 | * We only need to kick vcpu out of guest mode here, as PML buffer | |
7961 | * is flushed at beginning of all VMEXITs, and it's obvious that only | |
7962 | * vcpus running in guest are possible to have unflushed GPAs in PML | |
7963 | * buffer. | |
7964 | */ | |
7965 | kvm_for_each_vcpu(i, vcpu, kvm) | |
7966 | kvm_vcpu_kick(vcpu); | |
7967 | } | |
7968 | ||
4eb64dce PB |
7969 | static void vmx_dump_sel(char *name, uint32_t sel) |
7970 | { | |
7971 | pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n", | |
7972 | name, vmcs_read32(sel), | |
7973 | vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR), | |
7974 | vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR), | |
7975 | vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR)); | |
7976 | } | |
7977 | ||
7978 | static void vmx_dump_dtsel(char *name, uint32_t limit) | |
7979 | { | |
7980 | pr_err("%s limit=0x%08x, base=0x%016lx\n", | |
7981 | name, vmcs_read32(limit), | |
7982 | vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT)); | |
7983 | } | |
7984 | ||
7985 | static void dump_vmcs(void) | |
7986 | { | |
7987 | u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS); | |
7988 | u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS); | |
7989 | u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); | |
7990 | u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL); | |
7991 | u32 secondary_exec_control = 0; | |
7992 | unsigned long cr4 = vmcs_readl(GUEST_CR4); | |
f3531054 | 7993 | u64 efer = vmcs_read64(GUEST_IA32_EFER); |
4eb64dce PB |
7994 | int i, n; |
7995 | ||
7996 | if (cpu_has_secondary_exec_ctrls()) | |
7997 | secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL); | |
7998 | ||
7999 | pr_err("*** Guest State ***\n"); | |
8000 | pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n", | |
8001 | vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW), | |
8002 | vmcs_readl(CR0_GUEST_HOST_MASK)); | |
8003 | pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n", | |
8004 | cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK)); | |
8005 | pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3)); | |
8006 | if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) && | |
8007 | (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA)) | |
8008 | { | |
845c5b40 PB |
8009 | pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n", |
8010 | vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1)); | |
8011 | pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n", | |
8012 | vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3)); | |
4eb64dce PB |
8013 | } |
8014 | pr_err("RSP = 0x%016lx RIP = 0x%016lx\n", | |
8015 | vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP)); | |
8016 | pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n", | |
8017 | vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7)); | |
8018 | pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n", | |
8019 | vmcs_readl(GUEST_SYSENTER_ESP), | |
8020 | vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP)); | |
8021 | vmx_dump_sel("CS: ", GUEST_CS_SELECTOR); | |
8022 | vmx_dump_sel("DS: ", GUEST_DS_SELECTOR); | |
8023 | vmx_dump_sel("SS: ", GUEST_SS_SELECTOR); | |
8024 | vmx_dump_sel("ES: ", GUEST_ES_SELECTOR); | |
8025 | vmx_dump_sel("FS: ", GUEST_FS_SELECTOR); | |
8026 | vmx_dump_sel("GS: ", GUEST_GS_SELECTOR); | |
8027 | vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT); | |
8028 | vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR); | |
8029 | vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT); | |
8030 | vmx_dump_sel("TR: ", GUEST_TR_SELECTOR); | |
8031 | if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) || | |
8032 | (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER))) | |
845c5b40 PB |
8033 | pr_err("EFER = 0x%016llx PAT = 0x%016llx\n", |
8034 | efer, vmcs_read64(GUEST_IA32_PAT)); | |
8035 | pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n", | |
8036 | vmcs_read64(GUEST_IA32_DEBUGCTL), | |
4eb64dce PB |
8037 | vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS)); |
8038 | if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) | |
845c5b40 PB |
8039 | pr_err("PerfGlobCtl = 0x%016llx\n", |
8040 | vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL)); | |
4eb64dce | 8041 | if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS) |
845c5b40 | 8042 | pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS)); |
4eb64dce PB |
8043 | pr_err("Interruptibility = %08x ActivityState = %08x\n", |
8044 | vmcs_read32(GUEST_INTERRUPTIBILITY_INFO), | |
8045 | vmcs_read32(GUEST_ACTIVITY_STATE)); | |
8046 | if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) | |
8047 | pr_err("InterruptStatus = %04x\n", | |
8048 | vmcs_read16(GUEST_INTR_STATUS)); | |
8049 | ||
8050 | pr_err("*** Host State ***\n"); | |
8051 | pr_err("RIP = 0x%016lx RSP = 0x%016lx\n", | |
8052 | vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP)); | |
8053 | pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n", | |
8054 | vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR), | |
8055 | vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR), | |
8056 | vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR), | |
8057 | vmcs_read16(HOST_TR_SELECTOR)); | |
8058 | pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n", | |
8059 | vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE), | |
8060 | vmcs_readl(HOST_TR_BASE)); | |
8061 | pr_err("GDTBase=%016lx IDTBase=%016lx\n", | |
8062 | vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE)); | |
8063 | pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n", | |
8064 | vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3), | |
8065 | vmcs_readl(HOST_CR4)); | |
8066 | pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n", | |
8067 | vmcs_readl(HOST_IA32_SYSENTER_ESP), | |
8068 | vmcs_read32(HOST_IA32_SYSENTER_CS), | |
8069 | vmcs_readl(HOST_IA32_SYSENTER_EIP)); | |
8070 | if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER)) | |
845c5b40 PB |
8071 | pr_err("EFER = 0x%016llx PAT = 0x%016llx\n", |
8072 | vmcs_read64(HOST_IA32_EFER), | |
8073 | vmcs_read64(HOST_IA32_PAT)); | |
4eb64dce | 8074 | if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) |
845c5b40 PB |
8075 | pr_err("PerfGlobCtl = 0x%016llx\n", |
8076 | vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL)); | |
4eb64dce PB |
8077 | |
8078 | pr_err("*** Control State ***\n"); | |
8079 | pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n", | |
8080 | pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control); | |
8081 | pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl); | |
8082 | pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n", | |
8083 | vmcs_read32(EXCEPTION_BITMAP), | |
8084 | vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK), | |
8085 | vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH)); | |
8086 | pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n", | |
8087 | vmcs_read32(VM_ENTRY_INTR_INFO_FIELD), | |
8088 | vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE), | |
8089 | vmcs_read32(VM_ENTRY_INSTRUCTION_LEN)); | |
8090 | pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n", | |
8091 | vmcs_read32(VM_EXIT_INTR_INFO), | |
8092 | vmcs_read32(VM_EXIT_INTR_ERROR_CODE), | |
8093 | vmcs_read32(VM_EXIT_INSTRUCTION_LEN)); | |
8094 | pr_err(" reason=%08x qualification=%016lx\n", | |
8095 | vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION)); | |
8096 | pr_err("IDTVectoring: info=%08x errcode=%08x\n", | |
8097 | vmcs_read32(IDT_VECTORING_INFO_FIELD), | |
8098 | vmcs_read32(IDT_VECTORING_ERROR_CODE)); | |
845c5b40 | 8099 | pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET)); |
8cfe9866 | 8100 | if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING) |
845c5b40 PB |
8101 | pr_err("TSC Multiplier = 0x%016llx\n", |
8102 | vmcs_read64(TSC_MULTIPLIER)); | |
4eb64dce PB |
8103 | if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) |
8104 | pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD)); | |
8105 | if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR) | |
8106 | pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV)); | |
8107 | if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT)) | |
845c5b40 | 8108 | pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER)); |
4eb64dce PB |
8109 | n = vmcs_read32(CR3_TARGET_COUNT); |
8110 | for (i = 0; i + 1 < n; i += 4) | |
8111 | pr_err("CR3 target%u=%016lx target%u=%016lx\n", | |
8112 | i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2), | |
8113 | i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2)); | |
8114 | if (i < n) | |
8115 | pr_err("CR3 target%u=%016lx\n", | |
8116 | i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2)); | |
8117 | if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING) | |
8118 | pr_err("PLE Gap=%08x Window=%08x\n", | |
8119 | vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW)); | |
8120 | if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID) | |
8121 | pr_err("Virtual processor ID = 0x%04x\n", | |
8122 | vmcs_read16(VIRTUAL_PROCESSOR_ID)); | |
8123 | } | |
8124 | ||
6aa8b732 AK |
8125 | /* |
8126 | * The guest has exited. See if we can fix it or if we need userspace | |
8127 | * assistance. | |
8128 | */ | |
851ba692 | 8129 | static int vmx_handle_exit(struct kvm_vcpu *vcpu) |
6aa8b732 | 8130 | { |
29bd8a78 | 8131 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
a0861c02 | 8132 | u32 exit_reason = vmx->exit_reason; |
1155f76a | 8133 | u32 vectoring_info = vmx->idt_vectoring_info; |
29bd8a78 | 8134 | |
8b89fe1f PB |
8135 | trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX); |
8136 | ||
843e4330 KH |
8137 | /* |
8138 | * Flush logged GPAs PML buffer, this will make dirty_bitmap more | |
8139 | * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before | |
8140 | * querying dirty_bitmap, we only need to kick all vcpus out of guest | |
8141 | * mode as if vcpus is in root mode, the PML buffer must has been | |
8142 | * flushed already. | |
8143 | */ | |
8144 | if (enable_pml) | |
54bf36aa | 8145 | vmx_flush_pml_buffer(vcpu); |
843e4330 | 8146 | |
80ced186 | 8147 | /* If guest state is invalid, start emulating */ |
14168786 | 8148 | if (vmx->emulation_required) |
80ced186 | 8149 | return handle_invalid_guest_state(vcpu); |
1d5a4d9b | 8150 | |
644d711a | 8151 | if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) { |
533558bc JK |
8152 | nested_vmx_vmexit(vcpu, exit_reason, |
8153 | vmcs_read32(VM_EXIT_INTR_INFO), | |
8154 | vmcs_readl(EXIT_QUALIFICATION)); | |
644d711a NHE |
8155 | return 1; |
8156 | } | |
8157 | ||
5120702e | 8158 | if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) { |
4eb64dce | 8159 | dump_vmcs(); |
5120702e MG |
8160 | vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY; |
8161 | vcpu->run->fail_entry.hardware_entry_failure_reason | |
8162 | = exit_reason; | |
8163 | return 0; | |
8164 | } | |
8165 | ||
29bd8a78 | 8166 | if (unlikely(vmx->fail)) { |
851ba692 AK |
8167 | vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY; |
8168 | vcpu->run->fail_entry.hardware_entry_failure_reason | |
29bd8a78 AK |
8169 | = vmcs_read32(VM_INSTRUCTION_ERROR); |
8170 | return 0; | |
8171 | } | |
6aa8b732 | 8172 | |
b9bf6882 XG |
8173 | /* |
8174 | * Note: | |
8175 | * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by | |
8176 | * delivery event since it indicates guest is accessing MMIO. | |
8177 | * The vm-exit can be triggered again after return to guest that | |
8178 | * will cause infinite loop. | |
8179 | */ | |
d77c26fc | 8180 | if ((vectoring_info & VECTORING_INFO_VALID_MASK) && |
1439442c | 8181 | (exit_reason != EXIT_REASON_EXCEPTION_NMI && |
60637aac | 8182 | exit_reason != EXIT_REASON_EPT_VIOLATION && |
b9bf6882 XG |
8183 | exit_reason != EXIT_REASON_TASK_SWITCH)) { |
8184 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
8185 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV; | |
8186 | vcpu->run->internal.ndata = 2; | |
8187 | vcpu->run->internal.data[0] = vectoring_info; | |
8188 | vcpu->run->internal.data[1] = exit_reason; | |
8189 | return 0; | |
8190 | } | |
3b86cd99 | 8191 | |
644d711a NHE |
8192 | if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked && |
8193 | !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis( | |
f5c4368f | 8194 | get_vmcs12(vcpu))))) { |
c4282df9 | 8195 | if (vmx_interrupt_allowed(vcpu)) { |
3b86cd99 | 8196 | vmx->soft_vnmi_blocked = 0; |
3b86cd99 | 8197 | } else if (vmx->vnmi_blocked_time > 1000000000LL && |
4531220b | 8198 | vcpu->arch.nmi_pending) { |
3b86cd99 JK |
8199 | /* |
8200 | * This CPU don't support us in finding the end of an | |
8201 | * NMI-blocked window if the guest runs with IRQs | |
8202 | * disabled. So we pull the trigger after 1 s of | |
8203 | * futile waiting, but inform the user about this. | |
8204 | */ | |
8205 | printk(KERN_WARNING "%s: Breaking out of NMI-blocked " | |
8206 | "state on VCPU %d after 1 s timeout\n", | |
8207 | __func__, vcpu->vcpu_id); | |
8208 | vmx->soft_vnmi_blocked = 0; | |
3b86cd99 | 8209 | } |
3b86cd99 JK |
8210 | } |
8211 | ||
6aa8b732 AK |
8212 | if (exit_reason < kvm_vmx_max_exit_handlers |
8213 | && kvm_vmx_exit_handlers[exit_reason]) | |
851ba692 | 8214 | return kvm_vmx_exit_handlers[exit_reason](vcpu); |
6aa8b732 | 8215 | else { |
2bc19dc3 MT |
8216 | WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason); |
8217 | kvm_queue_exception(vcpu, UD_VECTOR); | |
8218 | return 1; | |
6aa8b732 | 8219 | } |
6aa8b732 AK |
8220 | } |
8221 | ||
95ba8273 | 8222 | static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr) |
6e5d865c | 8223 | { |
a7c0b07d WL |
8224 | struct vmcs12 *vmcs12 = get_vmcs12(vcpu); |
8225 | ||
8226 | if (is_guest_mode(vcpu) && | |
8227 | nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) | |
8228 | return; | |
8229 | ||
95ba8273 | 8230 | if (irr == -1 || tpr < irr) { |
6e5d865c YS |
8231 | vmcs_write32(TPR_THRESHOLD, 0); |
8232 | return; | |
8233 | } | |
8234 | ||
95ba8273 | 8235 | vmcs_write32(TPR_THRESHOLD, irr); |
6e5d865c YS |
8236 | } |
8237 | ||
8d14695f YZ |
8238 | static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set) |
8239 | { | |
8240 | u32 sec_exec_control; | |
8241 | ||
8242 | /* | |
8243 | * There is not point to enable virtualize x2apic without enable | |
8244 | * apicv | |
8245 | */ | |
c7c9c56c | 8246 | if (!cpu_has_vmx_virtualize_x2apic_mode() || |
d62caabb | 8247 | !kvm_vcpu_apicv_active(vcpu)) |
8d14695f YZ |
8248 | return; |
8249 | ||
35754c98 | 8250 | if (!cpu_need_tpr_shadow(vcpu)) |
8d14695f YZ |
8251 | return; |
8252 | ||
8253 | sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL); | |
8254 | ||
8255 | if (set) { | |
8256 | sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES; | |
8257 | sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE; | |
8258 | } else { | |
8259 | sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE; | |
8260 | sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES; | |
8261 | } | |
8262 | vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control); | |
8263 | ||
8264 | vmx_set_msr_bitmap(vcpu); | |
8265 | } | |
8266 | ||
38b99173 TC |
8267 | static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa) |
8268 | { | |
8269 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
8270 | ||
8271 | /* | |
8272 | * Currently we do not handle the nested case where L2 has an | |
8273 | * APIC access page of its own; that page is still pinned. | |
8274 | * Hence, we skip the case where the VCPU is in guest mode _and_ | |
8275 | * L1 prepared an APIC access page for L2. | |
8276 | * | |
8277 | * For the case where L1 and L2 share the same APIC access page | |
8278 | * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear | |
8279 | * in the vmcs12), this function will only update either the vmcs01 | |
8280 | * or the vmcs02. If the former, the vmcs02 will be updated by | |
8281 | * prepare_vmcs02. If the latter, the vmcs01 will be updated in | |
8282 | * the next L2->L1 exit. | |
8283 | */ | |
8284 | if (!is_guest_mode(vcpu) || | |
8285 | !nested_cpu_has2(vmx->nested.current_vmcs12, | |
8286 | SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) | |
8287 | vmcs_write64(APIC_ACCESS_ADDR, hpa); | |
8288 | } | |
8289 | ||
c7c9c56c YZ |
8290 | static void vmx_hwapic_isr_update(struct kvm *kvm, int isr) |
8291 | { | |
8292 | u16 status; | |
8293 | u8 old; | |
8294 | ||
c7c9c56c YZ |
8295 | if (isr == -1) |
8296 | isr = 0; | |
8297 | ||
8298 | status = vmcs_read16(GUEST_INTR_STATUS); | |
8299 | old = status >> 8; | |
8300 | if (isr != old) { | |
8301 | status &= 0xff; | |
8302 | status |= isr << 8; | |
8303 | vmcs_write16(GUEST_INTR_STATUS, status); | |
8304 | } | |
8305 | } | |
8306 | ||
8307 | static void vmx_set_rvi(int vector) | |
8308 | { | |
8309 | u16 status; | |
8310 | u8 old; | |
8311 | ||
4114c27d WW |
8312 | if (vector == -1) |
8313 | vector = 0; | |
8314 | ||
c7c9c56c YZ |
8315 | status = vmcs_read16(GUEST_INTR_STATUS); |
8316 | old = (u8)status & 0xff; | |
8317 | if ((u8)vector != old) { | |
8318 | status &= ~0xff; | |
8319 | status |= (u8)vector; | |
8320 | vmcs_write16(GUEST_INTR_STATUS, status); | |
8321 | } | |
8322 | } | |
8323 | ||
8324 | static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr) | |
8325 | { | |
4114c27d WW |
8326 | if (!is_guest_mode(vcpu)) { |
8327 | vmx_set_rvi(max_irr); | |
8328 | return; | |
8329 | } | |
8330 | ||
c7c9c56c YZ |
8331 | if (max_irr == -1) |
8332 | return; | |
8333 | ||
963fee16 | 8334 | /* |
4114c27d WW |
8335 | * In guest mode. If a vmexit is needed, vmx_check_nested_events |
8336 | * handles it. | |
963fee16 | 8337 | */ |
4114c27d | 8338 | if (nested_exit_on_intr(vcpu)) |
963fee16 WL |
8339 | return; |
8340 | ||
963fee16 | 8341 | /* |
4114c27d | 8342 | * Else, fall back to pre-APICv interrupt injection since L2 |
963fee16 WL |
8343 | * is run without virtual interrupt delivery. |
8344 | */ | |
8345 | if (!kvm_event_needs_reinjection(vcpu) && | |
8346 | vmx_interrupt_allowed(vcpu)) { | |
8347 | kvm_queue_interrupt(vcpu, max_irr, false); | |
8348 | vmx_inject_irq(vcpu); | |
8349 | } | |
c7c9c56c YZ |
8350 | } |
8351 | ||
6308630b | 8352 | static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap) |
c7c9c56c | 8353 | { |
d62caabb | 8354 | if (!kvm_vcpu_apicv_active(vcpu)) |
3d81bc7e YZ |
8355 | return; |
8356 | ||
c7c9c56c YZ |
8357 | vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]); |
8358 | vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]); | |
8359 | vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]); | |
8360 | vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]); | |
8361 | } | |
8362 | ||
51aa01d1 | 8363 | static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx) |
cf393f75 | 8364 | { |
00eba012 AK |
8365 | u32 exit_intr_info; |
8366 | ||
8367 | if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY | |
8368 | || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)) | |
8369 | return; | |
8370 | ||
c5ca8e57 | 8371 | vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO); |
00eba012 | 8372 | exit_intr_info = vmx->exit_intr_info; |
a0861c02 AK |
8373 | |
8374 | /* Handle machine checks before interrupts are enabled */ | |
00eba012 | 8375 | if (is_machine_check(exit_intr_info)) |
a0861c02 AK |
8376 | kvm_machine_check(); |
8377 | ||
20f65983 | 8378 | /* We need to handle NMIs before interrupts are enabled */ |
00eba012 | 8379 | if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR && |
ff9d07a0 ZY |
8380 | (exit_intr_info & INTR_INFO_VALID_MASK)) { |
8381 | kvm_before_handle_nmi(&vmx->vcpu); | |
20f65983 | 8382 | asm("int $2"); |
ff9d07a0 ZY |
8383 | kvm_after_handle_nmi(&vmx->vcpu); |
8384 | } | |
51aa01d1 | 8385 | } |
20f65983 | 8386 | |
a547c6db YZ |
8387 | static void vmx_handle_external_intr(struct kvm_vcpu *vcpu) |
8388 | { | |
8389 | u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO); | |
3f62de5f | 8390 | register void *__sp asm(_ASM_SP); |
a547c6db YZ |
8391 | |
8392 | /* | |
8393 | * If external interrupt exists, IF bit is set in rflags/eflags on the | |
8394 | * interrupt stack frame, and interrupt will be enabled on a return | |
8395 | * from interrupt handler. | |
8396 | */ | |
8397 | if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK)) | |
8398 | == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) { | |
8399 | unsigned int vector; | |
8400 | unsigned long entry; | |
8401 | gate_desc *desc; | |
8402 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
8403 | #ifdef CONFIG_X86_64 | |
8404 | unsigned long tmp; | |
8405 | #endif | |
8406 | ||
8407 | vector = exit_intr_info & INTR_INFO_VECTOR_MASK; | |
8408 | desc = (gate_desc *)vmx->host_idt_base + vector; | |
8409 | entry = gate_offset(*desc); | |
8410 | asm volatile( | |
8411 | #ifdef CONFIG_X86_64 | |
8412 | "mov %%" _ASM_SP ", %[sp]\n\t" | |
8413 | "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t" | |
8414 | "push $%c[ss]\n\t" | |
8415 | "push %[sp]\n\t" | |
8416 | #endif | |
8417 | "pushf\n\t" | |
8418 | "orl $0x200, (%%" _ASM_SP ")\n\t" | |
8419 | __ASM_SIZE(push) " $%c[cs]\n\t" | |
8420 | "call *%[entry]\n\t" | |
8421 | : | |
8422 | #ifdef CONFIG_X86_64 | |
3f62de5f | 8423 | [sp]"=&r"(tmp), |
a547c6db | 8424 | #endif |
3f62de5f | 8425 | "+r"(__sp) |
a547c6db YZ |
8426 | : |
8427 | [entry]"r"(entry), | |
8428 | [ss]"i"(__KERNEL_DS), | |
8429 | [cs]"i"(__KERNEL_CS) | |
8430 | ); | |
8431 | } else | |
8432 | local_irq_enable(); | |
8433 | } | |
8434 | ||
6d396b55 PB |
8435 | static bool vmx_has_high_real_mode_segbase(void) |
8436 | { | |
8437 | return enable_unrestricted_guest || emulate_invalid_guest_state; | |
8438 | } | |
8439 | ||
da8999d3 LJ |
8440 | static bool vmx_mpx_supported(void) |
8441 | { | |
8442 | return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) && | |
8443 | (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS); | |
8444 | } | |
8445 | ||
55412b2e WL |
8446 | static bool vmx_xsaves_supported(void) |
8447 | { | |
8448 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
8449 | SECONDARY_EXEC_XSAVES; | |
8450 | } | |
8451 | ||
51aa01d1 AK |
8452 | static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx) |
8453 | { | |
c5ca8e57 | 8454 | u32 exit_intr_info; |
51aa01d1 AK |
8455 | bool unblock_nmi; |
8456 | u8 vector; | |
8457 | bool idtv_info_valid; | |
8458 | ||
8459 | idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK; | |
20f65983 | 8460 | |
cf393f75 | 8461 | if (cpu_has_virtual_nmis()) { |
9d58b931 AK |
8462 | if (vmx->nmi_known_unmasked) |
8463 | return; | |
c5ca8e57 AK |
8464 | /* |
8465 | * Can't use vmx->exit_intr_info since we're not sure what | |
8466 | * the exit reason is. | |
8467 | */ | |
8468 | exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO); | |
cf393f75 AK |
8469 | unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0; |
8470 | vector = exit_intr_info & INTR_INFO_VECTOR_MASK; | |
8471 | /* | |
7b4a25cb | 8472 | * SDM 3: 27.7.1.2 (September 2008) |
cf393f75 AK |
8473 | * Re-set bit "block by NMI" before VM entry if vmexit caused by |
8474 | * a guest IRET fault. | |
7b4a25cb GN |
8475 | * SDM 3: 23.2.2 (September 2008) |
8476 | * Bit 12 is undefined in any of the following cases: | |
8477 | * If the VM exit sets the valid bit in the IDT-vectoring | |
8478 | * information field. | |
8479 | * If the VM exit is due to a double fault. | |
cf393f75 | 8480 | */ |
7b4a25cb GN |
8481 | if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi && |
8482 | vector != DF_VECTOR && !idtv_info_valid) | |
cf393f75 AK |
8483 | vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, |
8484 | GUEST_INTR_STATE_NMI); | |
9d58b931 AK |
8485 | else |
8486 | vmx->nmi_known_unmasked = | |
8487 | !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) | |
8488 | & GUEST_INTR_STATE_NMI); | |
3b86cd99 JK |
8489 | } else if (unlikely(vmx->soft_vnmi_blocked)) |
8490 | vmx->vnmi_blocked_time += | |
8491 | ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time)); | |
51aa01d1 AK |
8492 | } |
8493 | ||
3ab66e8a | 8494 | static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu, |
83422e17 AK |
8495 | u32 idt_vectoring_info, |
8496 | int instr_len_field, | |
8497 | int error_code_field) | |
51aa01d1 | 8498 | { |
51aa01d1 AK |
8499 | u8 vector; |
8500 | int type; | |
8501 | bool idtv_info_valid; | |
8502 | ||
8503 | idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK; | |
668f612f | 8504 | |
3ab66e8a JK |
8505 | vcpu->arch.nmi_injected = false; |
8506 | kvm_clear_exception_queue(vcpu); | |
8507 | kvm_clear_interrupt_queue(vcpu); | |
37b96e98 GN |
8508 | |
8509 | if (!idtv_info_valid) | |
8510 | return; | |
8511 | ||
3ab66e8a | 8512 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
3842d135 | 8513 | |
668f612f AK |
8514 | vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK; |
8515 | type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK; | |
37b96e98 | 8516 | |
64a7ec06 | 8517 | switch (type) { |
37b96e98 | 8518 | case INTR_TYPE_NMI_INTR: |
3ab66e8a | 8519 | vcpu->arch.nmi_injected = true; |
668f612f | 8520 | /* |
7b4a25cb | 8521 | * SDM 3: 27.7.1.2 (September 2008) |
37b96e98 GN |
8522 | * Clear bit "block by NMI" before VM entry if a NMI |
8523 | * delivery faulted. | |
668f612f | 8524 | */ |
3ab66e8a | 8525 | vmx_set_nmi_mask(vcpu, false); |
37b96e98 | 8526 | break; |
37b96e98 | 8527 | case INTR_TYPE_SOFT_EXCEPTION: |
3ab66e8a | 8528 | vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field); |
66fd3f7f GN |
8529 | /* fall through */ |
8530 | case INTR_TYPE_HARD_EXCEPTION: | |
35920a35 | 8531 | if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) { |
83422e17 | 8532 | u32 err = vmcs_read32(error_code_field); |
851eb667 | 8533 | kvm_requeue_exception_e(vcpu, vector, err); |
35920a35 | 8534 | } else |
851eb667 | 8535 | kvm_requeue_exception(vcpu, vector); |
37b96e98 | 8536 | break; |
66fd3f7f | 8537 | case INTR_TYPE_SOFT_INTR: |
3ab66e8a | 8538 | vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field); |
66fd3f7f | 8539 | /* fall through */ |
37b96e98 | 8540 | case INTR_TYPE_EXT_INTR: |
3ab66e8a | 8541 | kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR); |
37b96e98 GN |
8542 | break; |
8543 | default: | |
8544 | break; | |
f7d9238f | 8545 | } |
cf393f75 AK |
8546 | } |
8547 | ||
83422e17 AK |
8548 | static void vmx_complete_interrupts(struct vcpu_vmx *vmx) |
8549 | { | |
3ab66e8a | 8550 | __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info, |
83422e17 AK |
8551 | VM_EXIT_INSTRUCTION_LEN, |
8552 | IDT_VECTORING_ERROR_CODE); | |
8553 | } | |
8554 | ||
b463a6f7 AK |
8555 | static void vmx_cancel_injection(struct kvm_vcpu *vcpu) |
8556 | { | |
3ab66e8a | 8557 | __vmx_complete_interrupts(vcpu, |
b463a6f7 AK |
8558 | vmcs_read32(VM_ENTRY_INTR_INFO_FIELD), |
8559 | VM_ENTRY_INSTRUCTION_LEN, | |
8560 | VM_ENTRY_EXCEPTION_ERROR_CODE); | |
8561 | ||
8562 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); | |
8563 | } | |
8564 | ||
d7cd9796 GN |
8565 | static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx) |
8566 | { | |
8567 | int i, nr_msrs; | |
8568 | struct perf_guest_switch_msr *msrs; | |
8569 | ||
8570 | msrs = perf_guest_get_msrs(&nr_msrs); | |
8571 | ||
8572 | if (!msrs) | |
8573 | return; | |
8574 | ||
8575 | for (i = 0; i < nr_msrs; i++) | |
8576 | if (msrs[i].host == msrs[i].guest) | |
8577 | clear_atomic_switch_msr(vmx, msrs[i].msr); | |
8578 | else | |
8579 | add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest, | |
8580 | msrs[i].host); | |
8581 | } | |
8582 | ||
a3b5ba49 | 8583 | static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu) |
6aa8b732 | 8584 | { |
a2fa3e9f | 8585 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
d974baa3 | 8586 | unsigned long debugctlmsr, cr4; |
104f226b AK |
8587 | |
8588 | /* Record the guest's net vcpu time for enforced NMI injections. */ | |
8589 | if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) | |
8590 | vmx->entry_time = ktime_get(); | |
8591 | ||
8592 | /* Don't enter VMX if guest state is invalid, let the exit handler | |
8593 | start emulation until we arrive back to a valid state */ | |
14168786 | 8594 | if (vmx->emulation_required) |
104f226b AK |
8595 | return; |
8596 | ||
a7653ecd RK |
8597 | if (vmx->ple_window_dirty) { |
8598 | vmx->ple_window_dirty = false; | |
8599 | vmcs_write32(PLE_WINDOW, vmx->ple_window); | |
8600 | } | |
8601 | ||
012f83cb AG |
8602 | if (vmx->nested.sync_shadow_vmcs) { |
8603 | copy_vmcs12_to_shadow(vmx); | |
8604 | vmx->nested.sync_shadow_vmcs = false; | |
8605 | } | |
8606 | ||
104f226b AK |
8607 | if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty)) |
8608 | vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]); | |
8609 | if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty)) | |
8610 | vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]); | |
8611 | ||
1e02ce4c | 8612 | cr4 = cr4_read_shadow(); |
d974baa3 AL |
8613 | if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) { |
8614 | vmcs_writel(HOST_CR4, cr4); | |
8615 | vmx->host_state.vmcs_host_cr4 = cr4; | |
8616 | } | |
8617 | ||
104f226b AK |
8618 | /* When single-stepping over STI and MOV SS, we must clear the |
8619 | * corresponding interruptibility bits in the guest state. Otherwise | |
8620 | * vmentry fails as it then expects bit 14 (BS) in pending debug | |
8621 | * exceptions being set, but that's not correct for the guest debugging | |
8622 | * case. */ | |
8623 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) | |
8624 | vmx_set_interrupt_shadow(vcpu, 0); | |
8625 | ||
d7cd9796 | 8626 | atomic_switch_perf_msrs(vmx); |
2a7921b7 | 8627 | debugctlmsr = get_debugctlmsr(); |
d7cd9796 | 8628 | |
d462b819 | 8629 | vmx->__launched = vmx->loaded_vmcs->launched; |
104f226b | 8630 | asm( |
6aa8b732 | 8631 | /* Store host registers */ |
b188c81f AK |
8632 | "push %%" _ASM_DX "; push %%" _ASM_BP ";" |
8633 | "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */ | |
8634 | "push %%" _ASM_CX " \n\t" | |
8635 | "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t" | |
313dbd49 | 8636 | "je 1f \n\t" |
b188c81f | 8637 | "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t" |
4ecac3fd | 8638 | __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t" |
313dbd49 | 8639 | "1: \n\t" |
d3edefc0 | 8640 | /* Reload cr2 if changed */ |
b188c81f AK |
8641 | "mov %c[cr2](%0), %%" _ASM_AX " \n\t" |
8642 | "mov %%cr2, %%" _ASM_DX " \n\t" | |
8643 | "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t" | |
d3edefc0 | 8644 | "je 2f \n\t" |
b188c81f | 8645 | "mov %%" _ASM_AX", %%cr2 \n\t" |
d3edefc0 | 8646 | "2: \n\t" |
6aa8b732 | 8647 | /* Check if vmlaunch of vmresume is needed */ |
e08aa78a | 8648 | "cmpl $0, %c[launched](%0) \n\t" |
6aa8b732 | 8649 | /* Load guest registers. Don't clobber flags. */ |
b188c81f AK |
8650 | "mov %c[rax](%0), %%" _ASM_AX " \n\t" |
8651 | "mov %c[rbx](%0), %%" _ASM_BX " \n\t" | |
8652 | "mov %c[rdx](%0), %%" _ASM_DX " \n\t" | |
8653 | "mov %c[rsi](%0), %%" _ASM_SI " \n\t" | |
8654 | "mov %c[rdi](%0), %%" _ASM_DI " \n\t" | |
8655 | "mov %c[rbp](%0), %%" _ASM_BP " \n\t" | |
05b3e0c2 | 8656 | #ifdef CONFIG_X86_64 |
e08aa78a AK |
8657 | "mov %c[r8](%0), %%r8 \n\t" |
8658 | "mov %c[r9](%0), %%r9 \n\t" | |
8659 | "mov %c[r10](%0), %%r10 \n\t" | |
8660 | "mov %c[r11](%0), %%r11 \n\t" | |
8661 | "mov %c[r12](%0), %%r12 \n\t" | |
8662 | "mov %c[r13](%0), %%r13 \n\t" | |
8663 | "mov %c[r14](%0), %%r14 \n\t" | |
8664 | "mov %c[r15](%0), %%r15 \n\t" | |
6aa8b732 | 8665 | #endif |
b188c81f | 8666 | "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */ |
c801949d | 8667 | |
6aa8b732 | 8668 | /* Enter guest mode */ |
83287ea4 | 8669 | "jne 1f \n\t" |
4ecac3fd | 8670 | __ex(ASM_VMX_VMLAUNCH) "\n\t" |
83287ea4 AK |
8671 | "jmp 2f \n\t" |
8672 | "1: " __ex(ASM_VMX_VMRESUME) "\n\t" | |
8673 | "2: " | |
6aa8b732 | 8674 | /* Save guest registers, load host registers, keep flags */ |
b188c81f | 8675 | "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t" |
40712fae | 8676 | "pop %0 \n\t" |
b188c81f AK |
8677 | "mov %%" _ASM_AX ", %c[rax](%0) \n\t" |
8678 | "mov %%" _ASM_BX ", %c[rbx](%0) \n\t" | |
8679 | __ASM_SIZE(pop) " %c[rcx](%0) \n\t" | |
8680 | "mov %%" _ASM_DX ", %c[rdx](%0) \n\t" | |
8681 | "mov %%" _ASM_SI ", %c[rsi](%0) \n\t" | |
8682 | "mov %%" _ASM_DI ", %c[rdi](%0) \n\t" | |
8683 | "mov %%" _ASM_BP ", %c[rbp](%0) \n\t" | |
05b3e0c2 | 8684 | #ifdef CONFIG_X86_64 |
e08aa78a AK |
8685 | "mov %%r8, %c[r8](%0) \n\t" |
8686 | "mov %%r9, %c[r9](%0) \n\t" | |
8687 | "mov %%r10, %c[r10](%0) \n\t" | |
8688 | "mov %%r11, %c[r11](%0) \n\t" | |
8689 | "mov %%r12, %c[r12](%0) \n\t" | |
8690 | "mov %%r13, %c[r13](%0) \n\t" | |
8691 | "mov %%r14, %c[r14](%0) \n\t" | |
8692 | "mov %%r15, %c[r15](%0) \n\t" | |
6aa8b732 | 8693 | #endif |
b188c81f AK |
8694 | "mov %%cr2, %%" _ASM_AX " \n\t" |
8695 | "mov %%" _ASM_AX ", %c[cr2](%0) \n\t" | |
c801949d | 8696 | |
b188c81f | 8697 | "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t" |
e08aa78a | 8698 | "setbe %c[fail](%0) \n\t" |
83287ea4 AK |
8699 | ".pushsection .rodata \n\t" |
8700 | ".global vmx_return \n\t" | |
8701 | "vmx_return: " _ASM_PTR " 2b \n\t" | |
8702 | ".popsection" | |
e08aa78a | 8703 | : : "c"(vmx), "d"((unsigned long)HOST_RSP), |
d462b819 | 8704 | [launched]"i"(offsetof(struct vcpu_vmx, __launched)), |
e08aa78a | 8705 | [fail]"i"(offsetof(struct vcpu_vmx, fail)), |
313dbd49 | 8706 | [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)), |
ad312c7c ZX |
8707 | [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])), |
8708 | [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])), | |
8709 | [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])), | |
8710 | [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])), | |
8711 | [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])), | |
8712 | [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])), | |
8713 | [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])), | |
05b3e0c2 | 8714 | #ifdef CONFIG_X86_64 |
ad312c7c ZX |
8715 | [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])), |
8716 | [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])), | |
8717 | [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])), | |
8718 | [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])), | |
8719 | [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])), | |
8720 | [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])), | |
8721 | [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])), | |
8722 | [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])), | |
6aa8b732 | 8723 | #endif |
40712fae AK |
8724 | [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)), |
8725 | [wordsize]"i"(sizeof(ulong)) | |
c2036300 LV |
8726 | : "cc", "memory" |
8727 | #ifdef CONFIG_X86_64 | |
b188c81f | 8728 | , "rax", "rbx", "rdi", "rsi" |
c2036300 | 8729 | , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15" |
b188c81f AK |
8730 | #else |
8731 | , "eax", "ebx", "edi", "esi" | |
c2036300 LV |
8732 | #endif |
8733 | ); | |
6aa8b732 | 8734 | |
2a7921b7 GN |
8735 | /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */ |
8736 | if (debugctlmsr) | |
8737 | update_debugctlmsr(debugctlmsr); | |
8738 | ||
aa67f609 AK |
8739 | #ifndef CONFIG_X86_64 |
8740 | /* | |
8741 | * The sysexit path does not restore ds/es, so we must set them to | |
8742 | * a reasonable value ourselves. | |
8743 | * | |
8744 | * We can't defer this to vmx_load_host_state() since that function | |
8745 | * may be executed in interrupt context, which saves and restore segments | |
8746 | * around it, nullifying its effect. | |
8747 | */ | |
8748 | loadsegment(ds, __USER_DS); | |
8749 | loadsegment(es, __USER_DS); | |
8750 | #endif | |
8751 | ||
6de4f3ad | 8752 | vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP) |
6de12732 | 8753 | | (1 << VCPU_EXREG_RFLAGS) |
aff48baa | 8754 | | (1 << VCPU_EXREG_PDPTR) |
2fb92db1 | 8755 | | (1 << VCPU_EXREG_SEGMENTS) |
aff48baa | 8756 | | (1 << VCPU_EXREG_CR3)); |
5fdbf976 MT |
8757 | vcpu->arch.regs_dirty = 0; |
8758 | ||
1155f76a AK |
8759 | vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD); |
8760 | ||
d462b819 | 8761 | vmx->loaded_vmcs->launched = 1; |
1b6269db | 8762 | |
51aa01d1 | 8763 | vmx->exit_reason = vmcs_read32(VM_EXIT_REASON); |
51aa01d1 | 8764 | |
e0b890d3 GN |
8765 | /* |
8766 | * the KVM_REQ_EVENT optimization bit is only on for one entry, and if | |
8767 | * we did not inject a still-pending event to L1 now because of | |
8768 | * nested_run_pending, we need to re-enable this bit. | |
8769 | */ | |
8770 | if (vmx->nested.nested_run_pending) | |
8771 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
8772 | ||
8773 | vmx->nested.nested_run_pending = 0; | |
8774 | ||
51aa01d1 AK |
8775 | vmx_complete_atomic_exit(vmx); |
8776 | vmx_recover_nmi_blocking(vmx); | |
cf393f75 | 8777 | vmx_complete_interrupts(vmx); |
6aa8b732 AK |
8778 | } |
8779 | ||
4fa7734c PB |
8780 | static void vmx_load_vmcs01(struct kvm_vcpu *vcpu) |
8781 | { | |
8782 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
8783 | int cpu; | |
8784 | ||
8785 | if (vmx->loaded_vmcs == &vmx->vmcs01) | |
8786 | return; | |
8787 | ||
8788 | cpu = get_cpu(); | |
8789 | vmx->loaded_vmcs = &vmx->vmcs01; | |
8790 | vmx_vcpu_put(vcpu); | |
8791 | vmx_vcpu_load(vcpu, cpu); | |
8792 | vcpu->cpu = cpu; | |
8793 | put_cpu(); | |
8794 | } | |
8795 | ||
6aa8b732 AK |
8796 | static void vmx_free_vcpu(struct kvm_vcpu *vcpu) |
8797 | { | |
fb3f0f51 RR |
8798 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
8799 | ||
843e4330 | 8800 | if (enable_pml) |
a3eaa864 | 8801 | vmx_destroy_pml_buffer(vmx); |
991e7a0e | 8802 | free_vpid(vmx->vpid); |
4fa7734c PB |
8803 | leave_guest_mode(vcpu); |
8804 | vmx_load_vmcs01(vcpu); | |
26a865f4 | 8805 | free_nested(vmx); |
4fa7734c | 8806 | free_loaded_vmcs(vmx->loaded_vmcs); |
fb3f0f51 RR |
8807 | kfree(vmx->guest_msrs); |
8808 | kvm_vcpu_uninit(vcpu); | |
a4770347 | 8809 | kmem_cache_free(kvm_vcpu_cache, vmx); |
6aa8b732 AK |
8810 | } |
8811 | ||
fb3f0f51 | 8812 | static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id) |
6aa8b732 | 8813 | { |
fb3f0f51 | 8814 | int err; |
c16f862d | 8815 | struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
15ad7146 | 8816 | int cpu; |
6aa8b732 | 8817 | |
a2fa3e9f | 8818 | if (!vmx) |
fb3f0f51 RR |
8819 | return ERR_PTR(-ENOMEM); |
8820 | ||
991e7a0e | 8821 | vmx->vpid = allocate_vpid(); |
2384d2b3 | 8822 | |
fb3f0f51 RR |
8823 | err = kvm_vcpu_init(&vmx->vcpu, kvm, id); |
8824 | if (err) | |
8825 | goto free_vcpu; | |
965b58a5 | 8826 | |
a2fa3e9f | 8827 | vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL); |
03916db9 PB |
8828 | BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0]) |
8829 | > PAGE_SIZE); | |
0123be42 | 8830 | |
be6d05cf | 8831 | err = -ENOMEM; |
fb3f0f51 | 8832 | if (!vmx->guest_msrs) { |
fb3f0f51 RR |
8833 | goto uninit_vcpu; |
8834 | } | |
965b58a5 | 8835 | |
d462b819 NHE |
8836 | vmx->loaded_vmcs = &vmx->vmcs01; |
8837 | vmx->loaded_vmcs->vmcs = alloc_vmcs(); | |
8838 | if (!vmx->loaded_vmcs->vmcs) | |
fb3f0f51 | 8839 | goto free_msrs; |
d462b819 NHE |
8840 | if (!vmm_exclusive) |
8841 | kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id()))); | |
8842 | loaded_vmcs_init(vmx->loaded_vmcs); | |
8843 | if (!vmm_exclusive) | |
8844 | kvm_cpu_vmxoff(); | |
a2fa3e9f | 8845 | |
15ad7146 AK |
8846 | cpu = get_cpu(); |
8847 | vmx_vcpu_load(&vmx->vcpu, cpu); | |
e48672fa | 8848 | vmx->vcpu.cpu = cpu; |
8b9cf98c | 8849 | err = vmx_vcpu_setup(vmx); |
fb3f0f51 | 8850 | vmx_vcpu_put(&vmx->vcpu); |
15ad7146 | 8851 | put_cpu(); |
fb3f0f51 RR |
8852 | if (err) |
8853 | goto free_vmcs; | |
35754c98 | 8854 | if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) { |
be6d05cf JK |
8855 | err = alloc_apic_access_page(kvm); |
8856 | if (err) | |
5e4a0b3c | 8857 | goto free_vmcs; |
a63cb560 | 8858 | } |
fb3f0f51 | 8859 | |
b927a3ce SY |
8860 | if (enable_ept) { |
8861 | if (!kvm->arch.ept_identity_map_addr) | |
8862 | kvm->arch.ept_identity_map_addr = | |
8863 | VMX_EPT_IDENTITY_PAGETABLE_ADDR; | |
f51770ed TC |
8864 | err = init_rmode_identity_map(kvm); |
8865 | if (err) | |
93ea5388 | 8866 | goto free_vmcs; |
b927a3ce | 8867 | } |
b7ebfb05 | 8868 | |
5c614b35 | 8869 | if (nested) { |
b9c237bb | 8870 | nested_vmx_setup_ctls_msrs(vmx); |
5c614b35 WL |
8871 | vmx->nested.vpid02 = allocate_vpid(); |
8872 | } | |
b9c237bb | 8873 | |
705699a1 | 8874 | vmx->nested.posted_intr_nv = -1; |
a9d30f33 NHE |
8875 | vmx->nested.current_vmptr = -1ull; |
8876 | vmx->nested.current_vmcs12 = NULL; | |
8877 | ||
843e4330 KH |
8878 | /* |
8879 | * If PML is turned on, failure on enabling PML just results in failure | |
8880 | * of creating the vcpu, therefore we can simplify PML logic (by | |
8881 | * avoiding dealing with cases, such as enabling PML partially on vcpus | |
8882 | * for the guest, etc. | |
8883 | */ | |
8884 | if (enable_pml) { | |
a3eaa864 | 8885 | err = vmx_create_pml_buffer(vmx); |
843e4330 KH |
8886 | if (err) |
8887 | goto free_vmcs; | |
8888 | } | |
8889 | ||
fb3f0f51 RR |
8890 | return &vmx->vcpu; |
8891 | ||
8892 | free_vmcs: | |
5c614b35 | 8893 | free_vpid(vmx->nested.vpid02); |
5f3fbc34 | 8894 | free_loaded_vmcs(vmx->loaded_vmcs); |
fb3f0f51 | 8895 | free_msrs: |
fb3f0f51 RR |
8896 | kfree(vmx->guest_msrs); |
8897 | uninit_vcpu: | |
8898 | kvm_vcpu_uninit(&vmx->vcpu); | |
8899 | free_vcpu: | |
991e7a0e | 8900 | free_vpid(vmx->vpid); |
a4770347 | 8901 | kmem_cache_free(kvm_vcpu_cache, vmx); |
fb3f0f51 | 8902 | return ERR_PTR(err); |
6aa8b732 AK |
8903 | } |
8904 | ||
002c7f7c YS |
8905 | static void __init vmx_check_processor_compat(void *rtn) |
8906 | { | |
8907 | struct vmcs_config vmcs_conf; | |
8908 | ||
8909 | *(int *)rtn = 0; | |
8910 | if (setup_vmcs_config(&vmcs_conf) < 0) | |
8911 | *(int *)rtn = -EIO; | |
8912 | if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) { | |
8913 | printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n", | |
8914 | smp_processor_id()); | |
8915 | *(int *)rtn = -EIO; | |
8916 | } | |
8917 | } | |
8918 | ||
67253af5 SY |
8919 | static int get_ept_level(void) |
8920 | { | |
8921 | return VMX_EPT_DEFAULT_GAW + 1; | |
8922 | } | |
8923 | ||
4b12f0de | 8924 | static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio) |
64d4d521 | 8925 | { |
b18d5431 XG |
8926 | u8 cache; |
8927 | u64 ipat = 0; | |
4b12f0de | 8928 | |
522c68c4 | 8929 | /* For VT-d and EPT combination |
606decd6 | 8930 | * 1. MMIO: always map as UC |
522c68c4 SY |
8931 | * 2. EPT with VT-d: |
8932 | * a. VT-d without snooping control feature: can't guarantee the | |
606decd6 | 8933 | * result, try to trust guest. |
522c68c4 SY |
8934 | * b. VT-d with snooping control feature: snooping control feature of |
8935 | * VT-d engine can guarantee the cache correctness. Just set it | |
8936 | * to WB to keep consistent with host. So the same as item 3. | |
a19a6d11 | 8937 | * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep |
522c68c4 SY |
8938 | * consistent with host MTRR |
8939 | */ | |
606decd6 PB |
8940 | if (is_mmio) { |
8941 | cache = MTRR_TYPE_UNCACHABLE; | |
8942 | goto exit; | |
8943 | } | |
8944 | ||
8945 | if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) { | |
b18d5431 XG |
8946 | ipat = VMX_EPT_IPAT_BIT; |
8947 | cache = MTRR_TYPE_WRBACK; | |
8948 | goto exit; | |
8949 | } | |
8950 | ||
8951 | if (kvm_read_cr0(vcpu) & X86_CR0_CD) { | |
8952 | ipat = VMX_EPT_IPAT_BIT; | |
0da029ed | 8953 | if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED)) |
fb279950 XG |
8954 | cache = MTRR_TYPE_WRBACK; |
8955 | else | |
8956 | cache = MTRR_TYPE_UNCACHABLE; | |
b18d5431 XG |
8957 | goto exit; |
8958 | } | |
8959 | ||
ff53604b | 8960 | cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn); |
b18d5431 XG |
8961 | |
8962 | exit: | |
8963 | return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat; | |
64d4d521 SY |
8964 | } |
8965 | ||
17cc3935 | 8966 | static int vmx_get_lpage_level(void) |
344f414f | 8967 | { |
878403b7 SY |
8968 | if (enable_ept && !cpu_has_vmx_ept_1g_page()) |
8969 | return PT_DIRECTORY_LEVEL; | |
8970 | else | |
8971 | /* For shadow and EPT supported 1GB page */ | |
8972 | return PT_PDPE_LEVEL; | |
344f414f JR |
8973 | } |
8974 | ||
feda805f XG |
8975 | static void vmcs_set_secondary_exec_control(u32 new_ctl) |
8976 | { | |
8977 | /* | |
8978 | * These bits in the secondary execution controls field | |
8979 | * are dynamic, the others are mostly based on the hypervisor | |
8980 | * architecture and the guest's CPUID. Do not touch the | |
8981 | * dynamic bits. | |
8982 | */ | |
8983 | u32 mask = | |
8984 | SECONDARY_EXEC_SHADOW_VMCS | | |
8985 | SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE | | |
8986 | SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES; | |
8987 | ||
8988 | u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL); | |
8989 | ||
8990 | vmcs_write32(SECONDARY_VM_EXEC_CONTROL, | |
8991 | (new_ctl & ~mask) | (cur_ctl & mask)); | |
8992 | } | |
8993 | ||
0e851880 SY |
8994 | static void vmx_cpuid_update(struct kvm_vcpu *vcpu) |
8995 | { | |
4e47c7a6 SY |
8996 | struct kvm_cpuid_entry2 *best; |
8997 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
feda805f | 8998 | u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx); |
4e47c7a6 | 8999 | |
4e47c7a6 | 9000 | if (vmx_rdtscp_supported()) { |
1cea0ce6 XG |
9001 | bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu); |
9002 | if (!rdtscp_enabled) | |
feda805f | 9003 | secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP; |
f36201e5 | 9004 | |
8b97265a | 9005 | if (nested) { |
1cea0ce6 | 9006 | if (rdtscp_enabled) |
8b97265a PB |
9007 | vmx->nested.nested_vmx_secondary_ctls_high |= |
9008 | SECONDARY_EXEC_RDTSCP; | |
9009 | else | |
9010 | vmx->nested.nested_vmx_secondary_ctls_high &= | |
9011 | ~SECONDARY_EXEC_RDTSCP; | |
9012 | } | |
4e47c7a6 | 9013 | } |
ad756a16 | 9014 | |
ad756a16 MJ |
9015 | /* Exposing INVPCID only when PCID is exposed */ |
9016 | best = kvm_find_cpuid_entry(vcpu, 0x7, 0); | |
9017 | if (vmx_invpcid_supported() && | |
29541bb8 XG |
9018 | (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) || |
9019 | !guest_cpuid_has_pcid(vcpu))) { | |
feda805f | 9020 | secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID; |
29541bb8 | 9021 | |
ad756a16 | 9022 | if (best) |
4f977045 | 9023 | best->ebx &= ~bit(X86_FEATURE_INVPCID); |
ad756a16 | 9024 | } |
8b3e34e4 | 9025 | |
45bdbcfd HH |
9026 | if (cpu_has_secondary_exec_ctrls()) |
9027 | vmcs_set_secondary_exec_control(secondary_exec_ctl); | |
feda805f | 9028 | |
8b3e34e4 XG |
9029 | if (static_cpu_has(X86_FEATURE_PCOMMIT) && nested) { |
9030 | if (guest_cpuid_has_pcommit(vcpu)) | |
9031 | vmx->nested.nested_vmx_secondary_ctls_high |= | |
9032 | SECONDARY_EXEC_PCOMMIT; | |
9033 | else | |
9034 | vmx->nested.nested_vmx_secondary_ctls_high &= | |
9035 | ~SECONDARY_EXEC_PCOMMIT; | |
9036 | } | |
0e851880 SY |
9037 | } |
9038 | ||
d4330ef2 JR |
9039 | static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry) |
9040 | { | |
7b8050f5 NHE |
9041 | if (func == 1 && nested) |
9042 | entry->ecx |= bit(X86_FEATURE_VMX); | |
d4330ef2 JR |
9043 | } |
9044 | ||
25d92081 YZ |
9045 | static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu, |
9046 | struct x86_exception *fault) | |
9047 | { | |
533558bc JK |
9048 | struct vmcs12 *vmcs12 = get_vmcs12(vcpu); |
9049 | u32 exit_reason; | |
25d92081 YZ |
9050 | |
9051 | if (fault->error_code & PFERR_RSVD_MASK) | |
533558bc | 9052 | exit_reason = EXIT_REASON_EPT_MISCONFIG; |
25d92081 | 9053 | else |
533558bc JK |
9054 | exit_reason = EXIT_REASON_EPT_VIOLATION; |
9055 | nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification); | |
25d92081 YZ |
9056 | vmcs12->guest_physical_address = fault->address; |
9057 | } | |
9058 | ||
155a97a3 NHE |
9059 | /* Callbacks for nested_ept_init_mmu_context: */ |
9060 | ||
9061 | static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu) | |
9062 | { | |
9063 | /* return the page table to be shadowed - in our case, EPT12 */ | |
9064 | return get_vmcs12(vcpu)->ept_pointer; | |
9065 | } | |
9066 | ||
8a3c1a33 | 9067 | static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu) |
155a97a3 | 9068 | { |
ad896af0 PB |
9069 | WARN_ON(mmu_is_nested(vcpu)); |
9070 | kvm_init_shadow_ept_mmu(vcpu, | |
b9c237bb WV |
9071 | to_vmx(vcpu)->nested.nested_vmx_ept_caps & |
9072 | VMX_EPT_EXECUTE_ONLY_BIT); | |
155a97a3 NHE |
9073 | vcpu->arch.mmu.set_cr3 = vmx_set_cr3; |
9074 | vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3; | |
9075 | vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault; | |
9076 | ||
9077 | vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu; | |
155a97a3 NHE |
9078 | } |
9079 | ||
9080 | static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu) | |
9081 | { | |
9082 | vcpu->arch.walk_mmu = &vcpu->arch.mmu; | |
9083 | } | |
9084 | ||
19d5f10b EK |
9085 | static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12, |
9086 | u16 error_code) | |
9087 | { | |
9088 | bool inequality, bit; | |
9089 | ||
9090 | bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0; | |
9091 | inequality = | |
9092 | (error_code & vmcs12->page_fault_error_code_mask) != | |
9093 | vmcs12->page_fault_error_code_match; | |
9094 | return inequality ^ bit; | |
9095 | } | |
9096 | ||
feaf0c7d GN |
9097 | static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu, |
9098 | struct x86_exception *fault) | |
9099 | { | |
9100 | struct vmcs12 *vmcs12 = get_vmcs12(vcpu); | |
9101 | ||
9102 | WARN_ON(!is_guest_mode(vcpu)); | |
9103 | ||
19d5f10b | 9104 | if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code)) |
533558bc JK |
9105 | nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason, |
9106 | vmcs_read32(VM_EXIT_INTR_INFO), | |
9107 | vmcs_readl(EXIT_QUALIFICATION)); | |
feaf0c7d GN |
9108 | else |
9109 | kvm_inject_page_fault(vcpu, fault); | |
9110 | } | |
9111 | ||
a2bcba50 WL |
9112 | static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu, |
9113 | struct vmcs12 *vmcs12) | |
9114 | { | |
9115 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
9090422f | 9116 | int maxphyaddr = cpuid_maxphyaddr(vcpu); |
a2bcba50 WL |
9117 | |
9118 | if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) { | |
9090422f EK |
9119 | if (!PAGE_ALIGNED(vmcs12->apic_access_addr) || |
9120 | vmcs12->apic_access_addr >> maxphyaddr) | |
a2bcba50 WL |
9121 | return false; |
9122 | ||
9123 | /* | |
9124 | * Translate L1 physical address to host physical | |
9125 | * address for vmcs02. Keep the page pinned, so this | |
9126 | * physical address remains valid. We keep a reference | |
9127 | * to it so we can release it later. | |
9128 | */ | |
9129 | if (vmx->nested.apic_access_page) /* shouldn't happen */ | |
9130 | nested_release_page(vmx->nested.apic_access_page); | |
9131 | vmx->nested.apic_access_page = | |
9132 | nested_get_page(vcpu, vmcs12->apic_access_addr); | |
9133 | } | |
a7c0b07d WL |
9134 | |
9135 | if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) { | |
9090422f EK |
9136 | if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr) || |
9137 | vmcs12->virtual_apic_page_addr >> maxphyaddr) | |
a7c0b07d WL |
9138 | return false; |
9139 | ||
9140 | if (vmx->nested.virtual_apic_page) /* shouldn't happen */ | |
9141 | nested_release_page(vmx->nested.virtual_apic_page); | |
9142 | vmx->nested.virtual_apic_page = | |
9143 | nested_get_page(vcpu, vmcs12->virtual_apic_page_addr); | |
9144 | ||
9145 | /* | |
9146 | * Failing the vm entry is _not_ what the processor does | |
9147 | * but it's basically the only possibility we have. | |
9148 | * We could still enter the guest if CR8 load exits are | |
9149 | * enabled, CR8 store exits are enabled, and virtualize APIC | |
9150 | * access is disabled; in this case the processor would never | |
9151 | * use the TPR shadow and we could simply clear the bit from | |
9152 | * the execution control. But such a configuration is useless, | |
9153 | * so let's keep the code simple. | |
9154 | */ | |
9155 | if (!vmx->nested.virtual_apic_page) | |
9156 | return false; | |
9157 | } | |
9158 | ||
705699a1 | 9159 | if (nested_cpu_has_posted_intr(vmcs12)) { |
9090422f EK |
9160 | if (!IS_ALIGNED(vmcs12->posted_intr_desc_addr, 64) || |
9161 | vmcs12->posted_intr_desc_addr >> maxphyaddr) | |
705699a1 WV |
9162 | return false; |
9163 | ||
9164 | if (vmx->nested.pi_desc_page) { /* shouldn't happen */ | |
9165 | kunmap(vmx->nested.pi_desc_page); | |
9166 | nested_release_page(vmx->nested.pi_desc_page); | |
9167 | } | |
9168 | vmx->nested.pi_desc_page = | |
9169 | nested_get_page(vcpu, vmcs12->posted_intr_desc_addr); | |
9170 | if (!vmx->nested.pi_desc_page) | |
9171 | return false; | |
9172 | ||
9173 | vmx->nested.pi_desc = | |
9174 | (struct pi_desc *)kmap(vmx->nested.pi_desc_page); | |
9175 | if (!vmx->nested.pi_desc) { | |
9176 | nested_release_page_clean(vmx->nested.pi_desc_page); | |
9177 | return false; | |
9178 | } | |
9179 | vmx->nested.pi_desc = | |
9180 | (struct pi_desc *)((void *)vmx->nested.pi_desc + | |
9181 | (unsigned long)(vmcs12->posted_intr_desc_addr & | |
9182 | (PAGE_SIZE - 1))); | |
9183 | } | |
9184 | ||
a2bcba50 WL |
9185 | return true; |
9186 | } | |
9187 | ||
f4124500 JK |
9188 | static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu) |
9189 | { | |
9190 | u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value; | |
9191 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
9192 | ||
9193 | if (vcpu->arch.virtual_tsc_khz == 0) | |
9194 | return; | |
9195 | ||
9196 | /* Make sure short timeouts reliably trigger an immediate vmexit. | |
9197 | * hrtimer_start does not guarantee this. */ | |
9198 | if (preemption_timeout <= 1) { | |
9199 | vmx_preemption_timer_fn(&vmx->nested.preemption_timer); | |
9200 | return; | |
9201 | } | |
9202 | ||
9203 | preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE; | |
9204 | preemption_timeout *= 1000000; | |
9205 | do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz); | |
9206 | hrtimer_start(&vmx->nested.preemption_timer, | |
9207 | ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL); | |
9208 | } | |
9209 | ||
3af18d9c WV |
9210 | static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu, |
9211 | struct vmcs12 *vmcs12) | |
9212 | { | |
9213 | int maxphyaddr; | |
9214 | u64 addr; | |
9215 | ||
9216 | if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS)) | |
9217 | return 0; | |
9218 | ||
9219 | if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) { | |
9220 | WARN_ON(1); | |
9221 | return -EINVAL; | |
9222 | } | |
9223 | maxphyaddr = cpuid_maxphyaddr(vcpu); | |
9224 | ||
9225 | if (!PAGE_ALIGNED(vmcs12->msr_bitmap) || | |
9226 | ((addr + PAGE_SIZE) >> maxphyaddr)) | |
9227 | return -EINVAL; | |
9228 | ||
9229 | return 0; | |
9230 | } | |
9231 | ||
9232 | /* | |
9233 | * Merge L0's and L1's MSR bitmap, return false to indicate that | |
9234 | * we do not use the hardware. | |
9235 | */ | |
9236 | static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu, | |
9237 | struct vmcs12 *vmcs12) | |
9238 | { | |
82f0dd4b | 9239 | int msr; |
f2b93280 WV |
9240 | struct page *page; |
9241 | unsigned long *msr_bitmap; | |
9242 | ||
9243 | if (!nested_cpu_has_virt_x2apic_mode(vmcs12)) | |
9244 | return false; | |
9245 | ||
9246 | page = nested_get_page(vcpu, vmcs12->msr_bitmap); | |
9247 | if (!page) { | |
9248 | WARN_ON(1); | |
9249 | return false; | |
9250 | } | |
9251 | msr_bitmap = (unsigned long *)kmap(page); | |
9252 | if (!msr_bitmap) { | |
9253 | nested_release_page_clean(page); | |
9254 | WARN_ON(1); | |
9255 | return false; | |
9256 | } | |
9257 | ||
9258 | if (nested_cpu_has_virt_x2apic_mode(vmcs12)) { | |
82f0dd4b WV |
9259 | if (nested_cpu_has_apic_reg_virt(vmcs12)) |
9260 | for (msr = 0x800; msr <= 0x8ff; msr++) | |
9261 | nested_vmx_disable_intercept_for_msr( | |
9262 | msr_bitmap, | |
9263 | vmx_msr_bitmap_nested, | |
9264 | msr, MSR_TYPE_R); | |
f2b93280 WV |
9265 | /* TPR is allowed */ |
9266 | nested_vmx_disable_intercept_for_msr(msr_bitmap, | |
9267 | vmx_msr_bitmap_nested, | |
9268 | APIC_BASE_MSR + (APIC_TASKPRI >> 4), | |
9269 | MSR_TYPE_R | MSR_TYPE_W); | |
608406e2 WV |
9270 | if (nested_cpu_has_vid(vmcs12)) { |
9271 | /* EOI and self-IPI are allowed */ | |
9272 | nested_vmx_disable_intercept_for_msr( | |
9273 | msr_bitmap, | |
9274 | vmx_msr_bitmap_nested, | |
9275 | APIC_BASE_MSR + (APIC_EOI >> 4), | |
9276 | MSR_TYPE_W); | |
9277 | nested_vmx_disable_intercept_for_msr( | |
9278 | msr_bitmap, | |
9279 | vmx_msr_bitmap_nested, | |
9280 | APIC_BASE_MSR + (APIC_SELF_IPI >> 4), | |
9281 | MSR_TYPE_W); | |
9282 | } | |
82f0dd4b WV |
9283 | } else { |
9284 | /* | |
9285 | * Enable reading intercept of all the x2apic | |
9286 | * MSRs. We should not rely on vmcs12 to do any | |
9287 | * optimizations here, it may have been modified | |
9288 | * by L1. | |
9289 | */ | |
9290 | for (msr = 0x800; msr <= 0x8ff; msr++) | |
9291 | __vmx_enable_intercept_for_msr( | |
9292 | vmx_msr_bitmap_nested, | |
9293 | msr, | |
9294 | MSR_TYPE_R); | |
9295 | ||
f2b93280 WV |
9296 | __vmx_enable_intercept_for_msr( |
9297 | vmx_msr_bitmap_nested, | |
9298 | APIC_BASE_MSR + (APIC_TASKPRI >> 4), | |
82f0dd4b | 9299 | MSR_TYPE_W); |
608406e2 WV |
9300 | __vmx_enable_intercept_for_msr( |
9301 | vmx_msr_bitmap_nested, | |
9302 | APIC_BASE_MSR + (APIC_EOI >> 4), | |
9303 | MSR_TYPE_W); | |
9304 | __vmx_enable_intercept_for_msr( | |
9305 | vmx_msr_bitmap_nested, | |
9306 | APIC_BASE_MSR + (APIC_SELF_IPI >> 4), | |
9307 | MSR_TYPE_W); | |
82f0dd4b | 9308 | } |
f2b93280 WV |
9309 | kunmap(page); |
9310 | nested_release_page_clean(page); | |
9311 | ||
9312 | return true; | |
9313 | } | |
9314 | ||
9315 | static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu, | |
9316 | struct vmcs12 *vmcs12) | |
9317 | { | |
82f0dd4b | 9318 | if (!nested_cpu_has_virt_x2apic_mode(vmcs12) && |
608406e2 | 9319 | !nested_cpu_has_apic_reg_virt(vmcs12) && |
705699a1 WV |
9320 | !nested_cpu_has_vid(vmcs12) && |
9321 | !nested_cpu_has_posted_intr(vmcs12)) | |
f2b93280 WV |
9322 | return 0; |
9323 | ||
9324 | /* | |
9325 | * If virtualize x2apic mode is enabled, | |
9326 | * virtualize apic access must be disabled. | |
9327 | */ | |
82f0dd4b WV |
9328 | if (nested_cpu_has_virt_x2apic_mode(vmcs12) && |
9329 | nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) | |
f2b93280 WV |
9330 | return -EINVAL; |
9331 | ||
608406e2 WV |
9332 | /* |
9333 | * If virtual interrupt delivery is enabled, | |
9334 | * we must exit on external interrupts. | |
9335 | */ | |
9336 | if (nested_cpu_has_vid(vmcs12) && | |
9337 | !nested_exit_on_intr(vcpu)) | |
9338 | return -EINVAL; | |
9339 | ||
705699a1 WV |
9340 | /* |
9341 | * bits 15:8 should be zero in posted_intr_nv, | |
9342 | * the descriptor address has been already checked | |
9343 | * in nested_get_vmcs12_pages. | |
9344 | */ | |
9345 | if (nested_cpu_has_posted_intr(vmcs12) && | |
9346 | (!nested_cpu_has_vid(vmcs12) || | |
9347 | !nested_exit_intr_ack_set(vcpu) || | |
9348 | vmcs12->posted_intr_nv & 0xff00)) | |
9349 | return -EINVAL; | |
9350 | ||
f2b93280 WV |
9351 | /* tpr shadow is needed by all apicv features. */ |
9352 | if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) | |
9353 | return -EINVAL; | |
9354 | ||
9355 | return 0; | |
3af18d9c WV |
9356 | } |
9357 | ||
e9ac033e EK |
9358 | static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu, |
9359 | unsigned long count_field, | |
92d71bc6 | 9360 | unsigned long addr_field) |
ff651cb6 | 9361 | { |
92d71bc6 | 9362 | int maxphyaddr; |
e9ac033e EK |
9363 | u64 count, addr; |
9364 | ||
9365 | if (vmcs12_read_any(vcpu, count_field, &count) || | |
9366 | vmcs12_read_any(vcpu, addr_field, &addr)) { | |
9367 | WARN_ON(1); | |
9368 | return -EINVAL; | |
9369 | } | |
9370 | if (count == 0) | |
9371 | return 0; | |
92d71bc6 | 9372 | maxphyaddr = cpuid_maxphyaddr(vcpu); |
e9ac033e EK |
9373 | if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr || |
9374 | (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) { | |
9375 | pr_warn_ratelimited( | |
9376 | "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)", | |
9377 | addr_field, maxphyaddr, count, addr); | |
9378 | return -EINVAL; | |
9379 | } | |
9380 | return 0; | |
9381 | } | |
9382 | ||
9383 | static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu, | |
9384 | struct vmcs12 *vmcs12) | |
9385 | { | |
e9ac033e EK |
9386 | if (vmcs12->vm_exit_msr_load_count == 0 && |
9387 | vmcs12->vm_exit_msr_store_count == 0 && | |
9388 | vmcs12->vm_entry_msr_load_count == 0) | |
9389 | return 0; /* Fast path */ | |
e9ac033e | 9390 | if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT, |
92d71bc6 | 9391 | VM_EXIT_MSR_LOAD_ADDR) || |
e9ac033e | 9392 | nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT, |
92d71bc6 | 9393 | VM_EXIT_MSR_STORE_ADDR) || |
e9ac033e | 9394 | nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT, |
92d71bc6 | 9395 | VM_ENTRY_MSR_LOAD_ADDR)) |
e9ac033e EK |
9396 | return -EINVAL; |
9397 | return 0; | |
9398 | } | |
9399 | ||
9400 | static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu, | |
9401 | struct vmx_msr_entry *e) | |
9402 | { | |
9403 | /* x2APIC MSR accesses are not allowed */ | |
8a9781f7 | 9404 | if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8) |
e9ac033e EK |
9405 | return -EINVAL; |
9406 | if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */ | |
9407 | e->index == MSR_IA32_UCODE_REV) | |
9408 | return -EINVAL; | |
9409 | if (e->reserved != 0) | |
ff651cb6 WV |
9410 | return -EINVAL; |
9411 | return 0; | |
9412 | } | |
9413 | ||
e9ac033e EK |
9414 | static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu, |
9415 | struct vmx_msr_entry *e) | |
ff651cb6 WV |
9416 | { |
9417 | if (e->index == MSR_FS_BASE || | |
9418 | e->index == MSR_GS_BASE || | |
e9ac033e EK |
9419 | e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */ |
9420 | nested_vmx_msr_check_common(vcpu, e)) | |
9421 | return -EINVAL; | |
9422 | return 0; | |
9423 | } | |
9424 | ||
9425 | static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu, | |
9426 | struct vmx_msr_entry *e) | |
9427 | { | |
9428 | if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */ | |
9429 | nested_vmx_msr_check_common(vcpu, e)) | |
ff651cb6 WV |
9430 | return -EINVAL; |
9431 | return 0; | |
9432 | } | |
9433 | ||
9434 | /* | |
9435 | * Load guest's/host's msr at nested entry/exit. | |
9436 | * return 0 for success, entry index for failure. | |
9437 | */ | |
9438 | static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count) | |
9439 | { | |
9440 | u32 i; | |
9441 | struct vmx_msr_entry e; | |
9442 | struct msr_data msr; | |
9443 | ||
9444 | msr.host_initiated = false; | |
9445 | for (i = 0; i < count; i++) { | |
54bf36aa PB |
9446 | if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e), |
9447 | &e, sizeof(e))) { | |
e9ac033e EK |
9448 | pr_warn_ratelimited( |
9449 | "%s cannot read MSR entry (%u, 0x%08llx)\n", | |
9450 | __func__, i, gpa + i * sizeof(e)); | |
ff651cb6 | 9451 | goto fail; |
e9ac033e EK |
9452 | } |
9453 | if (nested_vmx_load_msr_check(vcpu, &e)) { | |
9454 | pr_warn_ratelimited( | |
9455 | "%s check failed (%u, 0x%x, 0x%x)\n", | |
9456 | __func__, i, e.index, e.reserved); | |
9457 | goto fail; | |
9458 | } | |
ff651cb6 WV |
9459 | msr.index = e.index; |
9460 | msr.data = e.value; | |
e9ac033e EK |
9461 | if (kvm_set_msr(vcpu, &msr)) { |
9462 | pr_warn_ratelimited( | |
9463 | "%s cannot write MSR (%u, 0x%x, 0x%llx)\n", | |
9464 | __func__, i, e.index, e.value); | |
ff651cb6 | 9465 | goto fail; |
e9ac033e | 9466 | } |
ff651cb6 WV |
9467 | } |
9468 | return 0; | |
9469 | fail: | |
9470 | return i + 1; | |
9471 | } | |
9472 | ||
9473 | static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count) | |
9474 | { | |
9475 | u32 i; | |
9476 | struct vmx_msr_entry e; | |
9477 | ||
9478 | for (i = 0; i < count; i++) { | |
609e36d3 | 9479 | struct msr_data msr_info; |
54bf36aa PB |
9480 | if (kvm_vcpu_read_guest(vcpu, |
9481 | gpa + i * sizeof(e), | |
9482 | &e, 2 * sizeof(u32))) { | |
e9ac033e EK |
9483 | pr_warn_ratelimited( |
9484 | "%s cannot read MSR entry (%u, 0x%08llx)\n", | |
9485 | __func__, i, gpa + i * sizeof(e)); | |
ff651cb6 | 9486 | return -EINVAL; |
e9ac033e EK |
9487 | } |
9488 | if (nested_vmx_store_msr_check(vcpu, &e)) { | |
9489 | pr_warn_ratelimited( | |
9490 | "%s check failed (%u, 0x%x, 0x%x)\n", | |
9491 | __func__, i, e.index, e.reserved); | |
ff651cb6 | 9492 | return -EINVAL; |
e9ac033e | 9493 | } |
609e36d3 PB |
9494 | msr_info.host_initiated = false; |
9495 | msr_info.index = e.index; | |
9496 | if (kvm_get_msr(vcpu, &msr_info)) { | |
e9ac033e EK |
9497 | pr_warn_ratelimited( |
9498 | "%s cannot read MSR (%u, 0x%x)\n", | |
9499 | __func__, i, e.index); | |
9500 | return -EINVAL; | |
9501 | } | |
54bf36aa PB |
9502 | if (kvm_vcpu_write_guest(vcpu, |
9503 | gpa + i * sizeof(e) + | |
9504 | offsetof(struct vmx_msr_entry, value), | |
9505 | &msr_info.data, sizeof(msr_info.data))) { | |
e9ac033e EK |
9506 | pr_warn_ratelimited( |
9507 | "%s cannot write MSR (%u, 0x%x, 0x%llx)\n", | |
609e36d3 | 9508 | __func__, i, e.index, msr_info.data); |
e9ac033e EK |
9509 | return -EINVAL; |
9510 | } | |
ff651cb6 WV |
9511 | } |
9512 | return 0; | |
9513 | } | |
9514 | ||
fe3ef05c NHE |
9515 | /* |
9516 | * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested | |
9517 | * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it | |
b4619660 | 9518 | * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2 |
fe3ef05c NHE |
9519 | * guest in a way that will both be appropriate to L1's requests, and our |
9520 | * needs. In addition to modifying the active vmcs (which is vmcs02), this | |
9521 | * function also has additional necessary side-effects, like setting various | |
9522 | * vcpu->arch fields. | |
9523 | */ | |
9524 | static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12) | |
9525 | { | |
9526 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
9527 | u32 exec_control; | |
9528 | ||
9529 | vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector); | |
9530 | vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector); | |
9531 | vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector); | |
9532 | vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector); | |
9533 | vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector); | |
9534 | vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector); | |
9535 | vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector); | |
9536 | vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector); | |
9537 | vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit); | |
9538 | vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit); | |
9539 | vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit); | |
9540 | vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit); | |
9541 | vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit); | |
9542 | vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit); | |
9543 | vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit); | |
9544 | vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit); | |
9545 | vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit); | |
9546 | vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit); | |
9547 | vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes); | |
9548 | vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes); | |
9549 | vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes); | |
9550 | vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes); | |
9551 | vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes); | |
9552 | vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes); | |
9553 | vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes); | |
9554 | vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes); | |
9555 | vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base); | |
9556 | vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base); | |
9557 | vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base); | |
9558 | vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base); | |
9559 | vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base); | |
9560 | vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base); | |
9561 | vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base); | |
9562 | vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base); | |
9563 | vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base); | |
9564 | vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base); | |
9565 | ||
2996fca0 JK |
9566 | if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) { |
9567 | kvm_set_dr(vcpu, 7, vmcs12->guest_dr7); | |
9568 | vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl); | |
9569 | } else { | |
9570 | kvm_set_dr(vcpu, 7, vcpu->arch.dr7); | |
9571 | vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl); | |
9572 | } | |
fe3ef05c NHE |
9573 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, |
9574 | vmcs12->vm_entry_intr_info_field); | |
9575 | vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, | |
9576 | vmcs12->vm_entry_exception_error_code); | |
9577 | vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, | |
9578 | vmcs12->vm_entry_instruction_len); | |
9579 | vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, | |
9580 | vmcs12->guest_interruptibility_info); | |
fe3ef05c | 9581 | vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs); |
63fbf59f | 9582 | vmx_set_rflags(vcpu, vmcs12->guest_rflags); |
fe3ef05c NHE |
9583 | vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, |
9584 | vmcs12->guest_pending_dbg_exceptions); | |
9585 | vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp); | |
9586 | vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip); | |
9587 | ||
81dc01f7 WL |
9588 | if (nested_cpu_has_xsaves(vmcs12)) |
9589 | vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap); | |
fe3ef05c NHE |
9590 | vmcs_write64(VMCS_LINK_POINTER, -1ull); |
9591 | ||
f4124500 JK |
9592 | exec_control = vmcs12->pin_based_vm_exec_control; |
9593 | exec_control |= vmcs_config.pin_based_exec_ctrl; | |
705699a1 WV |
9594 | exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER; |
9595 | ||
9596 | if (nested_cpu_has_posted_intr(vmcs12)) { | |
9597 | /* | |
9598 | * Note that we use L0's vector here and in | |
9599 | * vmx_deliver_nested_posted_interrupt. | |
9600 | */ | |
9601 | vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv; | |
9602 | vmx->nested.pi_pending = false; | |
0bcf261c | 9603 | vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR); |
705699a1 WV |
9604 | vmcs_write64(POSTED_INTR_DESC_ADDR, |
9605 | page_to_phys(vmx->nested.pi_desc_page) + | |
9606 | (unsigned long)(vmcs12->posted_intr_desc_addr & | |
9607 | (PAGE_SIZE - 1))); | |
9608 | } else | |
9609 | exec_control &= ~PIN_BASED_POSTED_INTR; | |
9610 | ||
f4124500 | 9611 | vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control); |
fe3ef05c | 9612 | |
f4124500 JK |
9613 | vmx->nested.preemption_timer_expired = false; |
9614 | if (nested_cpu_has_preemption_timer(vmcs12)) | |
9615 | vmx_start_preemption_timer(vcpu); | |
0238ea91 | 9616 | |
fe3ef05c NHE |
9617 | /* |
9618 | * Whether page-faults are trapped is determined by a combination of | |
9619 | * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF. | |
9620 | * If enable_ept, L0 doesn't care about page faults and we should | |
9621 | * set all of these to L1's desires. However, if !enable_ept, L0 does | |
9622 | * care about (at least some) page faults, and because it is not easy | |
9623 | * (if at all possible?) to merge L0 and L1's desires, we simply ask | |
9624 | * to exit on each and every L2 page fault. This is done by setting | |
9625 | * MASK=MATCH=0 and (see below) EB.PF=1. | |
9626 | * Note that below we don't need special code to set EB.PF beyond the | |
9627 | * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept, | |
9628 | * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when | |
9629 | * !enable_ept, EB.PF is 1, so the "or" will always be 1. | |
9630 | * | |
9631 | * A problem with this approach (when !enable_ept) is that L1 may be | |
9632 | * injected with more page faults than it asked for. This could have | |
9633 | * caused problems, but in practice existing hypervisors don't care. | |
9634 | * To fix this, we will need to emulate the PFEC checking (on the L1 | |
9635 | * page tables), using walk_addr(), when injecting PFs to L1. | |
9636 | */ | |
9637 | vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, | |
9638 | enable_ept ? vmcs12->page_fault_error_code_mask : 0); | |
9639 | vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, | |
9640 | enable_ept ? vmcs12->page_fault_error_code_match : 0); | |
9641 | ||
9642 | if (cpu_has_secondary_exec_ctrls()) { | |
f4124500 | 9643 | exec_control = vmx_secondary_exec_control(vmx); |
e2821620 | 9644 | |
fe3ef05c | 9645 | /* Take the following fields only from vmcs12 */ |
696dfd95 | 9646 | exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES | |
b3a2a907 | 9647 | SECONDARY_EXEC_RDTSCP | |
696dfd95 | 9648 | SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY | |
8b3e34e4 XG |
9649 | SECONDARY_EXEC_APIC_REGISTER_VIRT | |
9650 | SECONDARY_EXEC_PCOMMIT); | |
fe3ef05c NHE |
9651 | if (nested_cpu_has(vmcs12, |
9652 | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) | |
9653 | exec_control |= vmcs12->secondary_vm_exec_control; | |
9654 | ||
9655 | if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) { | |
fe3ef05c NHE |
9656 | /* |
9657 | * If translation failed, no matter: This feature asks | |
9658 | * to exit when accessing the given address, and if it | |
9659 | * can never be accessed, this feature won't do | |
9660 | * anything anyway. | |
9661 | */ | |
9662 | if (!vmx->nested.apic_access_page) | |
9663 | exec_control &= | |
9664 | ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES; | |
9665 | else | |
9666 | vmcs_write64(APIC_ACCESS_ADDR, | |
9667 | page_to_phys(vmx->nested.apic_access_page)); | |
f2b93280 | 9668 | } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) && |
35754c98 | 9669 | cpu_need_virtualize_apic_accesses(&vmx->vcpu)) { |
ca3f257a JK |
9670 | exec_control |= |
9671 | SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES; | |
38b99173 | 9672 | kvm_vcpu_reload_apic_access_page(vcpu); |
fe3ef05c NHE |
9673 | } |
9674 | ||
608406e2 WV |
9675 | if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) { |
9676 | vmcs_write64(EOI_EXIT_BITMAP0, | |
9677 | vmcs12->eoi_exit_bitmap0); | |
9678 | vmcs_write64(EOI_EXIT_BITMAP1, | |
9679 | vmcs12->eoi_exit_bitmap1); | |
9680 | vmcs_write64(EOI_EXIT_BITMAP2, | |
9681 | vmcs12->eoi_exit_bitmap2); | |
9682 | vmcs_write64(EOI_EXIT_BITMAP3, | |
9683 | vmcs12->eoi_exit_bitmap3); | |
9684 | vmcs_write16(GUEST_INTR_STATUS, | |
9685 | vmcs12->guest_intr_status); | |
9686 | } | |
9687 | ||
fe3ef05c NHE |
9688 | vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control); |
9689 | } | |
9690 | ||
9691 | ||
9692 | /* | |
9693 | * Set host-state according to L0's settings (vmcs12 is irrelevant here) | |
9694 | * Some constant fields are set here by vmx_set_constant_host_state(). | |
9695 | * Other fields are different per CPU, and will be set later when | |
9696 | * vmx_vcpu_load() is called, and when vmx_save_host_state() is called. | |
9697 | */ | |
a547c6db | 9698 | vmx_set_constant_host_state(vmx); |
fe3ef05c NHE |
9699 | |
9700 | /* | |
9701 | * HOST_RSP is normally set correctly in vmx_vcpu_run() just before | |
9702 | * entry, but only if the current (host) sp changed from the value | |
9703 | * we wrote last (vmx->host_rsp). This cache is no longer relevant | |
9704 | * if we switch vmcs, and rather than hold a separate cache per vmcs, | |
9705 | * here we just force the write to happen on entry. | |
9706 | */ | |
9707 | vmx->host_rsp = 0; | |
9708 | ||
9709 | exec_control = vmx_exec_control(vmx); /* L0's desires */ | |
9710 | exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING; | |
9711 | exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING; | |
9712 | exec_control &= ~CPU_BASED_TPR_SHADOW; | |
9713 | exec_control |= vmcs12->cpu_based_vm_exec_control; | |
a7c0b07d WL |
9714 | |
9715 | if (exec_control & CPU_BASED_TPR_SHADOW) { | |
9716 | vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, | |
9717 | page_to_phys(vmx->nested.virtual_apic_page)); | |
9718 | vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold); | |
9719 | } | |
9720 | ||
3af18d9c | 9721 | if (cpu_has_vmx_msr_bitmap() && |
670125bd WV |
9722 | exec_control & CPU_BASED_USE_MSR_BITMAPS) { |
9723 | nested_vmx_merge_msr_bitmap(vcpu, vmcs12); | |
9724 | /* MSR_BITMAP will be set by following vmx_set_efer. */ | |
3af18d9c WV |
9725 | } else |
9726 | exec_control &= ~CPU_BASED_USE_MSR_BITMAPS; | |
9727 | ||
fe3ef05c | 9728 | /* |
3af18d9c | 9729 | * Merging of IO bitmap not currently supported. |
fe3ef05c NHE |
9730 | * Rather, exit every time. |
9731 | */ | |
fe3ef05c NHE |
9732 | exec_control &= ~CPU_BASED_USE_IO_BITMAPS; |
9733 | exec_control |= CPU_BASED_UNCOND_IO_EXITING; | |
9734 | ||
9735 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control); | |
9736 | ||
9737 | /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the | |
9738 | * bitwise-or of what L1 wants to trap for L2, and what we want to | |
9739 | * trap. Note that CR0.TS also needs updating - we do this later. | |
9740 | */ | |
9741 | update_exception_bitmap(vcpu); | |
9742 | vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask; | |
9743 | vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits); | |
9744 | ||
8049d651 NHE |
9745 | /* L2->L1 exit controls are emulated - the hardware exit is to L0 so |
9746 | * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER | |
9747 | * bits are further modified by vmx_set_efer() below. | |
9748 | */ | |
f4124500 | 9749 | vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl); |
8049d651 NHE |
9750 | |
9751 | /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are | |
9752 | * emulated by vmx_set_efer(), below. | |
9753 | */ | |
2961e876 | 9754 | vm_entry_controls_init(vmx, |
8049d651 NHE |
9755 | (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER & |
9756 | ~VM_ENTRY_IA32E_MODE) | | |
fe3ef05c NHE |
9757 | (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE)); |
9758 | ||
44811c02 | 9759 | if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) { |
fe3ef05c | 9760 | vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat); |
44811c02 JK |
9761 | vcpu->arch.pat = vmcs12->guest_ia32_pat; |
9762 | } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) | |
fe3ef05c NHE |
9763 | vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat); |
9764 | ||
9765 | ||
9766 | set_cr4_guest_host_mask(vmx); | |
9767 | ||
36be0b9d PB |
9768 | if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) |
9769 | vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs); | |
9770 | ||
27fc51b2 NHE |
9771 | if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING) |
9772 | vmcs_write64(TSC_OFFSET, | |
9773 | vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset); | |
9774 | else | |
9775 | vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset); | |
fe3ef05c NHE |
9776 | |
9777 | if (enable_vpid) { | |
9778 | /* | |
5c614b35 WL |
9779 | * There is no direct mapping between vpid02 and vpid12, the |
9780 | * vpid02 is per-vCPU for L0 and reused while the value of | |
9781 | * vpid12 is changed w/ one invvpid during nested vmentry. | |
9782 | * The vpid12 is allocated by L1 for L2, so it will not | |
9783 | * influence global bitmap(for vpid01 and vpid02 allocation) | |
9784 | * even if spawn a lot of nested vCPUs. | |
fe3ef05c | 9785 | */ |
5c614b35 WL |
9786 | if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) { |
9787 | vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02); | |
9788 | if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) { | |
9789 | vmx->nested.last_vpid = vmcs12->virtual_processor_id; | |
9790 | __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02); | |
9791 | } | |
9792 | } else { | |
9793 | vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid); | |
9794 | vmx_flush_tlb(vcpu); | |
9795 | } | |
9796 | ||
fe3ef05c NHE |
9797 | } |
9798 | ||
155a97a3 NHE |
9799 | if (nested_cpu_has_ept(vmcs12)) { |
9800 | kvm_mmu_unload(vcpu); | |
9801 | nested_ept_init_mmu_context(vcpu); | |
9802 | } | |
9803 | ||
fe3ef05c NHE |
9804 | if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) |
9805 | vcpu->arch.efer = vmcs12->guest_ia32_efer; | |
d1fa0352 | 9806 | else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) |
fe3ef05c NHE |
9807 | vcpu->arch.efer |= (EFER_LMA | EFER_LME); |
9808 | else | |
9809 | vcpu->arch.efer &= ~(EFER_LMA | EFER_LME); | |
9810 | /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */ | |
9811 | vmx_set_efer(vcpu, vcpu->arch.efer); | |
9812 | ||
9813 | /* | |
9814 | * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified | |
9815 | * TS bit (for lazy fpu) and bits which we consider mandatory enabled. | |
9816 | * The CR0_READ_SHADOW is what L2 should have expected to read given | |
9817 | * the specifications by L1; It's not enough to take | |
9818 | * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we | |
9819 | * have more bits than L1 expected. | |
9820 | */ | |
9821 | vmx_set_cr0(vcpu, vmcs12->guest_cr0); | |
9822 | vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12)); | |
9823 | ||
9824 | vmx_set_cr4(vcpu, vmcs12->guest_cr4); | |
9825 | vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12)); | |
9826 | ||
9827 | /* shadow page tables on either EPT or shadow page tables */ | |
9828 | kvm_set_cr3(vcpu, vmcs12->guest_cr3); | |
9829 | kvm_mmu_reset_context(vcpu); | |
9830 | ||
feaf0c7d GN |
9831 | if (!enable_ept) |
9832 | vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested; | |
9833 | ||
3633cfc3 NHE |
9834 | /* |
9835 | * L1 may access the L2's PDPTR, so save them to construct vmcs12 | |
9836 | */ | |
9837 | if (enable_ept) { | |
9838 | vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0); | |
9839 | vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1); | |
9840 | vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2); | |
9841 | vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3); | |
9842 | } | |
9843 | ||
fe3ef05c NHE |
9844 | kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp); |
9845 | kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip); | |
9846 | } | |
9847 | ||
cd232ad0 NHE |
9848 | /* |
9849 | * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1 | |
9850 | * for running an L2 nested guest. | |
9851 | */ | |
9852 | static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch) | |
9853 | { | |
9854 | struct vmcs12 *vmcs12; | |
9855 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
9856 | int cpu; | |
9857 | struct loaded_vmcs *vmcs02; | |
384bb783 | 9858 | bool ia32e; |
ff651cb6 | 9859 | u32 msr_entry_idx; |
cd232ad0 NHE |
9860 | |
9861 | if (!nested_vmx_check_permission(vcpu) || | |
9862 | !nested_vmx_check_vmcs12(vcpu)) | |
9863 | return 1; | |
9864 | ||
9865 | skip_emulated_instruction(vcpu); | |
9866 | vmcs12 = get_vmcs12(vcpu); | |
9867 | ||
012f83cb AG |
9868 | if (enable_shadow_vmcs) |
9869 | copy_shadow_to_vmcs12(vmx); | |
9870 | ||
7c177938 NHE |
9871 | /* |
9872 | * The nested entry process starts with enforcing various prerequisites | |
9873 | * on vmcs12 as required by the Intel SDM, and act appropriately when | |
9874 | * they fail: As the SDM explains, some conditions should cause the | |
9875 | * instruction to fail, while others will cause the instruction to seem | |
9876 | * to succeed, but return an EXIT_REASON_INVALID_STATE. | |
9877 | * To speed up the normal (success) code path, we should avoid checking | |
9878 | * for misconfigurations which will anyway be caught by the processor | |
9879 | * when using the merged vmcs02. | |
9880 | */ | |
9881 | if (vmcs12->launch_state == launch) { | |
9882 | nested_vmx_failValid(vcpu, | |
9883 | launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS | |
9884 | : VMXERR_VMRESUME_NONLAUNCHED_VMCS); | |
9885 | return 1; | |
9886 | } | |
9887 | ||
6dfacadd JK |
9888 | if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE && |
9889 | vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) { | |
26539bd0 PB |
9890 | nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD); |
9891 | return 1; | |
9892 | } | |
9893 | ||
3af18d9c | 9894 | if (!nested_get_vmcs12_pages(vcpu, vmcs12)) { |
7c177938 NHE |
9895 | nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD); |
9896 | return 1; | |
9897 | } | |
9898 | ||
3af18d9c | 9899 | if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12)) { |
7c177938 NHE |
9900 | nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD); |
9901 | return 1; | |
9902 | } | |
9903 | ||
f2b93280 WV |
9904 | if (nested_vmx_check_apicv_controls(vcpu, vmcs12)) { |
9905 | nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD); | |
9906 | return 1; | |
9907 | } | |
9908 | ||
e9ac033e EK |
9909 | if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12)) { |
9910 | nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD); | |
9911 | return 1; | |
9912 | } | |
9913 | ||
7c177938 | 9914 | if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control, |
b9c237bb WV |
9915 | vmx->nested.nested_vmx_true_procbased_ctls_low, |
9916 | vmx->nested.nested_vmx_procbased_ctls_high) || | |
7c177938 | 9917 | !vmx_control_verify(vmcs12->secondary_vm_exec_control, |
b9c237bb WV |
9918 | vmx->nested.nested_vmx_secondary_ctls_low, |
9919 | vmx->nested.nested_vmx_secondary_ctls_high) || | |
7c177938 | 9920 | !vmx_control_verify(vmcs12->pin_based_vm_exec_control, |
b9c237bb WV |
9921 | vmx->nested.nested_vmx_pinbased_ctls_low, |
9922 | vmx->nested.nested_vmx_pinbased_ctls_high) || | |
7c177938 | 9923 | !vmx_control_verify(vmcs12->vm_exit_controls, |
b9c237bb WV |
9924 | vmx->nested.nested_vmx_true_exit_ctls_low, |
9925 | vmx->nested.nested_vmx_exit_ctls_high) || | |
7c177938 | 9926 | !vmx_control_verify(vmcs12->vm_entry_controls, |
b9c237bb WV |
9927 | vmx->nested.nested_vmx_true_entry_ctls_low, |
9928 | vmx->nested.nested_vmx_entry_ctls_high)) | |
7c177938 NHE |
9929 | { |
9930 | nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD); | |
9931 | return 1; | |
9932 | } | |
9933 | ||
9934 | if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) || | |
9935 | ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) { | |
9936 | nested_vmx_failValid(vcpu, | |
9937 | VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); | |
9938 | return 1; | |
9939 | } | |
9940 | ||
b9c237bb | 9941 | if (!nested_cr0_valid(vcpu, vmcs12->guest_cr0) || |
7c177938 NHE |
9942 | ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) { |
9943 | nested_vmx_entry_failure(vcpu, vmcs12, | |
9944 | EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT); | |
9945 | return 1; | |
9946 | } | |
9947 | if (vmcs12->vmcs_link_pointer != -1ull) { | |
9948 | nested_vmx_entry_failure(vcpu, vmcs12, | |
9949 | EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR); | |
9950 | return 1; | |
9951 | } | |
9952 | ||
384bb783 | 9953 | /* |
cb0c8cda | 9954 | * If the load IA32_EFER VM-entry control is 1, the following checks |
384bb783 JK |
9955 | * are performed on the field for the IA32_EFER MSR: |
9956 | * - Bits reserved in the IA32_EFER MSR must be 0. | |
9957 | * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of | |
9958 | * the IA-32e mode guest VM-exit control. It must also be identical | |
9959 | * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to | |
9960 | * CR0.PG) is 1. | |
9961 | */ | |
9962 | if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) { | |
9963 | ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0; | |
9964 | if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) || | |
9965 | ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) || | |
9966 | ((vmcs12->guest_cr0 & X86_CR0_PG) && | |
9967 | ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) { | |
9968 | nested_vmx_entry_failure(vcpu, vmcs12, | |
9969 | EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT); | |
9970 | return 1; | |
9971 | } | |
9972 | } | |
9973 | ||
9974 | /* | |
9975 | * If the load IA32_EFER VM-exit control is 1, bits reserved in the | |
9976 | * IA32_EFER MSR must be 0 in the field for that register. In addition, | |
9977 | * the values of the LMA and LME bits in the field must each be that of | |
9978 | * the host address-space size VM-exit control. | |
9979 | */ | |
9980 | if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) { | |
9981 | ia32e = (vmcs12->vm_exit_controls & | |
9982 | VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0; | |
9983 | if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) || | |
9984 | ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) || | |
9985 | ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) { | |
9986 | nested_vmx_entry_failure(vcpu, vmcs12, | |
9987 | EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT); | |
9988 | return 1; | |
9989 | } | |
9990 | } | |
9991 | ||
7c177938 NHE |
9992 | /* |
9993 | * We're finally done with prerequisite checking, and can start with | |
9994 | * the nested entry. | |
9995 | */ | |
9996 | ||
cd232ad0 NHE |
9997 | vmcs02 = nested_get_current_vmcs02(vmx); |
9998 | if (!vmcs02) | |
9999 | return -ENOMEM; | |
10000 | ||
10001 | enter_guest_mode(vcpu); | |
10002 | ||
10003 | vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET); | |
10004 | ||
2996fca0 JK |
10005 | if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) |
10006 | vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL); | |
10007 | ||
cd232ad0 NHE |
10008 | cpu = get_cpu(); |
10009 | vmx->loaded_vmcs = vmcs02; | |
10010 | vmx_vcpu_put(vcpu); | |
10011 | vmx_vcpu_load(vcpu, cpu); | |
10012 | vcpu->cpu = cpu; | |
10013 | put_cpu(); | |
10014 | ||
36c3cc42 JK |
10015 | vmx_segment_cache_clear(vmx); |
10016 | ||
cd232ad0 NHE |
10017 | prepare_vmcs02(vcpu, vmcs12); |
10018 | ||
ff651cb6 WV |
10019 | msr_entry_idx = nested_vmx_load_msr(vcpu, |
10020 | vmcs12->vm_entry_msr_load_addr, | |
10021 | vmcs12->vm_entry_msr_load_count); | |
10022 | if (msr_entry_idx) { | |
10023 | leave_guest_mode(vcpu); | |
10024 | vmx_load_vmcs01(vcpu); | |
10025 | nested_vmx_entry_failure(vcpu, vmcs12, | |
10026 | EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx); | |
10027 | return 1; | |
10028 | } | |
10029 | ||
10030 | vmcs12->launch_state = 1; | |
10031 | ||
6dfacadd | 10032 | if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) |
5cb56059 | 10033 | return kvm_vcpu_halt(vcpu); |
6dfacadd | 10034 | |
7af40ad3 JK |
10035 | vmx->nested.nested_run_pending = 1; |
10036 | ||
cd232ad0 NHE |
10037 | /* |
10038 | * Note no nested_vmx_succeed or nested_vmx_fail here. At this point | |
10039 | * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet | |
10040 | * returned as far as L1 is concerned. It will only return (and set | |
10041 | * the success flag) when L2 exits (see nested_vmx_vmexit()). | |
10042 | */ | |
10043 | return 1; | |
10044 | } | |
10045 | ||
4704d0be NHE |
10046 | /* |
10047 | * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date | |
10048 | * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK). | |
10049 | * This function returns the new value we should put in vmcs12.guest_cr0. | |
10050 | * It's not enough to just return the vmcs02 GUEST_CR0. Rather, | |
10051 | * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now | |
10052 | * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0 | |
10053 | * didn't trap the bit, because if L1 did, so would L0). | |
10054 | * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have | |
10055 | * been modified by L2, and L1 knows it. So just leave the old value of | |
10056 | * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0 | |
10057 | * isn't relevant, because if L0 traps this bit it can set it to anything. | |
10058 | * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have | |
10059 | * changed these bits, and therefore they need to be updated, but L0 | |
10060 | * didn't necessarily allow them to be changed in GUEST_CR0 - and rather | |
10061 | * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there. | |
10062 | */ | |
10063 | static inline unsigned long | |
10064 | vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12) | |
10065 | { | |
10066 | return | |
10067 | /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) | | |
10068 | /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) | | |
10069 | /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask | | |
10070 | vcpu->arch.cr0_guest_owned_bits)); | |
10071 | } | |
10072 | ||
10073 | static inline unsigned long | |
10074 | vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12) | |
10075 | { | |
10076 | return | |
10077 | /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) | | |
10078 | /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) | | |
10079 | /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask | | |
10080 | vcpu->arch.cr4_guest_owned_bits)); | |
10081 | } | |
10082 | ||
5f3d5799 JK |
10083 | static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu, |
10084 | struct vmcs12 *vmcs12) | |
10085 | { | |
10086 | u32 idt_vectoring; | |
10087 | unsigned int nr; | |
10088 | ||
851eb667 | 10089 | if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) { |
5f3d5799 JK |
10090 | nr = vcpu->arch.exception.nr; |
10091 | idt_vectoring = nr | VECTORING_INFO_VALID_MASK; | |
10092 | ||
10093 | if (kvm_exception_is_soft(nr)) { | |
10094 | vmcs12->vm_exit_instruction_len = | |
10095 | vcpu->arch.event_exit_inst_len; | |
10096 | idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION; | |
10097 | } else | |
10098 | idt_vectoring |= INTR_TYPE_HARD_EXCEPTION; | |
10099 | ||
10100 | if (vcpu->arch.exception.has_error_code) { | |
10101 | idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK; | |
10102 | vmcs12->idt_vectoring_error_code = | |
10103 | vcpu->arch.exception.error_code; | |
10104 | } | |
10105 | ||
10106 | vmcs12->idt_vectoring_info_field = idt_vectoring; | |
cd2633c5 | 10107 | } else if (vcpu->arch.nmi_injected) { |
5f3d5799 JK |
10108 | vmcs12->idt_vectoring_info_field = |
10109 | INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR; | |
10110 | } else if (vcpu->arch.interrupt.pending) { | |
10111 | nr = vcpu->arch.interrupt.nr; | |
10112 | idt_vectoring = nr | VECTORING_INFO_VALID_MASK; | |
10113 | ||
10114 | if (vcpu->arch.interrupt.soft) { | |
10115 | idt_vectoring |= INTR_TYPE_SOFT_INTR; | |
10116 | vmcs12->vm_entry_instruction_len = | |
10117 | vcpu->arch.event_exit_inst_len; | |
10118 | } else | |
10119 | idt_vectoring |= INTR_TYPE_EXT_INTR; | |
10120 | ||
10121 | vmcs12->idt_vectoring_info_field = idt_vectoring; | |
10122 | } | |
10123 | } | |
10124 | ||
b6b8a145 JK |
10125 | static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr) |
10126 | { | |
10127 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
10128 | ||
f4124500 JK |
10129 | if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) && |
10130 | vmx->nested.preemption_timer_expired) { | |
10131 | if (vmx->nested.nested_run_pending) | |
10132 | return -EBUSY; | |
10133 | nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0); | |
10134 | return 0; | |
10135 | } | |
10136 | ||
b6b8a145 | 10137 | if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) { |
220c5672 JK |
10138 | if (vmx->nested.nested_run_pending || |
10139 | vcpu->arch.interrupt.pending) | |
b6b8a145 JK |
10140 | return -EBUSY; |
10141 | nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, | |
10142 | NMI_VECTOR | INTR_TYPE_NMI_INTR | | |
10143 | INTR_INFO_VALID_MASK, 0); | |
10144 | /* | |
10145 | * The NMI-triggered VM exit counts as injection: | |
10146 | * clear this one and block further NMIs. | |
10147 | */ | |
10148 | vcpu->arch.nmi_pending = 0; | |
10149 | vmx_set_nmi_mask(vcpu, true); | |
10150 | return 0; | |
10151 | } | |
10152 | ||
10153 | if ((kvm_cpu_has_interrupt(vcpu) || external_intr) && | |
10154 | nested_exit_on_intr(vcpu)) { | |
10155 | if (vmx->nested.nested_run_pending) | |
10156 | return -EBUSY; | |
10157 | nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0); | |
705699a1 | 10158 | return 0; |
b6b8a145 JK |
10159 | } |
10160 | ||
705699a1 | 10161 | return vmx_complete_nested_posted_interrupt(vcpu); |
b6b8a145 JK |
10162 | } |
10163 | ||
f4124500 JK |
10164 | static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu) |
10165 | { | |
10166 | ktime_t remaining = | |
10167 | hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer); | |
10168 | u64 value; | |
10169 | ||
10170 | if (ktime_to_ns(remaining) <= 0) | |
10171 | return 0; | |
10172 | ||
10173 | value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz; | |
10174 | do_div(value, 1000000); | |
10175 | return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE; | |
10176 | } | |
10177 | ||
4704d0be NHE |
10178 | /* |
10179 | * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits | |
10180 | * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12), | |
10181 | * and this function updates it to reflect the changes to the guest state while | |
10182 | * L2 was running (and perhaps made some exits which were handled directly by L0 | |
10183 | * without going back to L1), and to reflect the exit reason. | |
10184 | * Note that we do not have to copy here all VMCS fields, just those that | |
10185 | * could have changed by the L2 guest or the exit - i.e., the guest-state and | |
10186 | * exit-information fields only. Other fields are modified by L1 with VMWRITE, | |
10187 | * which already writes to vmcs12 directly. | |
10188 | */ | |
533558bc JK |
10189 | static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12, |
10190 | u32 exit_reason, u32 exit_intr_info, | |
10191 | unsigned long exit_qualification) | |
4704d0be NHE |
10192 | { |
10193 | /* update guest state fields: */ | |
10194 | vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12); | |
10195 | vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12); | |
10196 | ||
4704d0be NHE |
10197 | vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP); |
10198 | vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP); | |
10199 | vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS); | |
10200 | ||
10201 | vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR); | |
10202 | vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR); | |
10203 | vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR); | |
10204 | vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR); | |
10205 | vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR); | |
10206 | vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR); | |
10207 | vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR); | |
10208 | vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR); | |
10209 | vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT); | |
10210 | vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT); | |
10211 | vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT); | |
10212 | vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT); | |
10213 | vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT); | |
10214 | vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT); | |
10215 | vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT); | |
10216 | vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT); | |
10217 | vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT); | |
10218 | vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT); | |
10219 | vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES); | |
10220 | vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES); | |
10221 | vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES); | |
10222 | vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES); | |
10223 | vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES); | |
10224 | vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES); | |
10225 | vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES); | |
10226 | vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES); | |
10227 | vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE); | |
10228 | vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE); | |
10229 | vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE); | |
10230 | vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE); | |
10231 | vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE); | |
10232 | vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE); | |
10233 | vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE); | |
10234 | vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE); | |
10235 | vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE); | |
10236 | vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE); | |
10237 | ||
4704d0be NHE |
10238 | vmcs12->guest_interruptibility_info = |
10239 | vmcs_read32(GUEST_INTERRUPTIBILITY_INFO); | |
10240 | vmcs12->guest_pending_dbg_exceptions = | |
10241 | vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS); | |
3edf1e69 JK |
10242 | if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED) |
10243 | vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT; | |
10244 | else | |
10245 | vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE; | |
4704d0be | 10246 | |
f4124500 JK |
10247 | if (nested_cpu_has_preemption_timer(vmcs12)) { |
10248 | if (vmcs12->vm_exit_controls & | |
10249 | VM_EXIT_SAVE_VMX_PREEMPTION_TIMER) | |
10250 | vmcs12->vmx_preemption_timer_value = | |
10251 | vmx_get_preemption_timer_value(vcpu); | |
10252 | hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer); | |
10253 | } | |
7854cbca | 10254 | |
3633cfc3 NHE |
10255 | /* |
10256 | * In some cases (usually, nested EPT), L2 is allowed to change its | |
10257 | * own CR3 without exiting. If it has changed it, we must keep it. | |
10258 | * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined | |
10259 | * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12. | |
10260 | * | |
10261 | * Additionally, restore L2's PDPTR to vmcs12. | |
10262 | */ | |
10263 | if (enable_ept) { | |
f3531054 | 10264 | vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3); |
3633cfc3 NHE |
10265 | vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0); |
10266 | vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1); | |
10267 | vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2); | |
10268 | vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3); | |
10269 | } | |
10270 | ||
608406e2 WV |
10271 | if (nested_cpu_has_vid(vmcs12)) |
10272 | vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS); | |
10273 | ||
c18911a2 JK |
10274 | vmcs12->vm_entry_controls = |
10275 | (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) | | |
2961e876 | 10276 | (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE); |
c18911a2 | 10277 | |
2996fca0 JK |
10278 | if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) { |
10279 | kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7); | |
10280 | vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL); | |
10281 | } | |
10282 | ||
4704d0be NHE |
10283 | /* TODO: These cannot have changed unless we have MSR bitmaps and |
10284 | * the relevant bit asks not to trap the change */ | |
b8c07d55 | 10285 | if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT) |
4704d0be | 10286 | vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT); |
10ba54a5 JK |
10287 | if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER) |
10288 | vmcs12->guest_ia32_efer = vcpu->arch.efer; | |
4704d0be NHE |
10289 | vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS); |
10290 | vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP); | |
10291 | vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP); | |
a87036ad | 10292 | if (kvm_mpx_supported()) |
36be0b9d | 10293 | vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS); |
81dc01f7 WL |
10294 | if (nested_cpu_has_xsaves(vmcs12)) |
10295 | vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP); | |
4704d0be NHE |
10296 | |
10297 | /* update exit information fields: */ | |
10298 | ||
533558bc JK |
10299 | vmcs12->vm_exit_reason = exit_reason; |
10300 | vmcs12->exit_qualification = exit_qualification; | |
4704d0be | 10301 | |
533558bc | 10302 | vmcs12->vm_exit_intr_info = exit_intr_info; |
c0d1c770 JK |
10303 | if ((vmcs12->vm_exit_intr_info & |
10304 | (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) == | |
10305 | (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) | |
10306 | vmcs12->vm_exit_intr_error_code = | |
10307 | vmcs_read32(VM_EXIT_INTR_ERROR_CODE); | |
5f3d5799 | 10308 | vmcs12->idt_vectoring_info_field = 0; |
4704d0be NHE |
10309 | vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN); |
10310 | vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO); | |
10311 | ||
5f3d5799 JK |
10312 | if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) { |
10313 | /* vm_entry_intr_info_field is cleared on exit. Emulate this | |
10314 | * instead of reading the real value. */ | |
4704d0be | 10315 | vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK; |
5f3d5799 JK |
10316 | |
10317 | /* | |
10318 | * Transfer the event that L0 or L1 may wanted to inject into | |
10319 | * L2 to IDT_VECTORING_INFO_FIELD. | |
10320 | */ | |
10321 | vmcs12_save_pending_event(vcpu, vmcs12); | |
10322 | } | |
10323 | ||
10324 | /* | |
10325 | * Drop what we picked up for L2 via vmx_complete_interrupts. It is | |
10326 | * preserved above and would only end up incorrectly in L1. | |
10327 | */ | |
10328 | vcpu->arch.nmi_injected = false; | |
10329 | kvm_clear_exception_queue(vcpu); | |
10330 | kvm_clear_interrupt_queue(vcpu); | |
4704d0be NHE |
10331 | } |
10332 | ||
10333 | /* | |
10334 | * A part of what we need to when the nested L2 guest exits and we want to | |
10335 | * run its L1 parent, is to reset L1's guest state to the host state specified | |
10336 | * in vmcs12. | |
10337 | * This function is to be called not only on normal nested exit, but also on | |
10338 | * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry | |
10339 | * Failures During or After Loading Guest State"). | |
10340 | * This function should be called when the active VMCS is L1's (vmcs01). | |
10341 | */ | |
733568f9 JK |
10342 | static void load_vmcs12_host_state(struct kvm_vcpu *vcpu, |
10343 | struct vmcs12 *vmcs12) | |
4704d0be | 10344 | { |
21feb4eb ACL |
10345 | struct kvm_segment seg; |
10346 | ||
4704d0be NHE |
10347 | if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) |
10348 | vcpu->arch.efer = vmcs12->host_ia32_efer; | |
d1fa0352 | 10349 | else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) |
4704d0be NHE |
10350 | vcpu->arch.efer |= (EFER_LMA | EFER_LME); |
10351 | else | |
10352 | vcpu->arch.efer &= ~(EFER_LMA | EFER_LME); | |
10353 | vmx_set_efer(vcpu, vcpu->arch.efer); | |
10354 | ||
10355 | kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp); | |
10356 | kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip); | |
1adfa76a | 10357 | vmx_set_rflags(vcpu, X86_EFLAGS_FIXED); |
4704d0be NHE |
10358 | /* |
10359 | * Note that calling vmx_set_cr0 is important, even if cr0 hasn't | |
10360 | * actually changed, because it depends on the current state of | |
10361 | * fpu_active (which may have changed). | |
10362 | * Note that vmx_set_cr0 refers to efer set above. | |
10363 | */ | |
9e3e4dbf | 10364 | vmx_set_cr0(vcpu, vmcs12->host_cr0); |
4704d0be NHE |
10365 | /* |
10366 | * If we did fpu_activate()/fpu_deactivate() during L2's run, we need | |
10367 | * to apply the same changes to L1's vmcs. We just set cr0 correctly, | |
10368 | * but we also need to update cr0_guest_host_mask and exception_bitmap. | |
10369 | */ | |
10370 | update_exception_bitmap(vcpu); | |
10371 | vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0); | |
10372 | vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits); | |
10373 | ||
10374 | /* | |
10375 | * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01 | |
10376 | * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask(); | |
10377 | */ | |
10378 | vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK); | |
10379 | kvm_set_cr4(vcpu, vmcs12->host_cr4); | |
10380 | ||
29bf08f1 | 10381 | nested_ept_uninit_mmu_context(vcpu); |
155a97a3 | 10382 | |
4704d0be NHE |
10383 | kvm_set_cr3(vcpu, vmcs12->host_cr3); |
10384 | kvm_mmu_reset_context(vcpu); | |
10385 | ||
feaf0c7d GN |
10386 | if (!enable_ept) |
10387 | vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault; | |
10388 | ||
4704d0be NHE |
10389 | if (enable_vpid) { |
10390 | /* | |
10391 | * Trivially support vpid by letting L2s share their parent | |
10392 | * L1's vpid. TODO: move to a more elaborate solution, giving | |
10393 | * each L2 its own vpid and exposing the vpid feature to L1. | |
10394 | */ | |
10395 | vmx_flush_tlb(vcpu); | |
10396 | } | |
10397 | ||
10398 | ||
10399 | vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs); | |
10400 | vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp); | |
10401 | vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip); | |
10402 | vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base); | |
10403 | vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base); | |
4704d0be | 10404 | |
36be0b9d PB |
10405 | /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */ |
10406 | if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS) | |
10407 | vmcs_write64(GUEST_BNDCFGS, 0); | |
10408 | ||
44811c02 | 10409 | if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) { |
4704d0be | 10410 | vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat); |
44811c02 JK |
10411 | vcpu->arch.pat = vmcs12->host_ia32_pat; |
10412 | } | |
4704d0be NHE |
10413 | if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) |
10414 | vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL, | |
10415 | vmcs12->host_ia32_perf_global_ctrl); | |
503cd0c5 | 10416 | |
21feb4eb ACL |
10417 | /* Set L1 segment info according to Intel SDM |
10418 | 27.5.2 Loading Host Segment and Descriptor-Table Registers */ | |
10419 | seg = (struct kvm_segment) { | |
10420 | .base = 0, | |
10421 | .limit = 0xFFFFFFFF, | |
10422 | .selector = vmcs12->host_cs_selector, | |
10423 | .type = 11, | |
10424 | .present = 1, | |
10425 | .s = 1, | |
10426 | .g = 1 | |
10427 | }; | |
10428 | if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) | |
10429 | seg.l = 1; | |
10430 | else | |
10431 | seg.db = 1; | |
10432 | vmx_set_segment(vcpu, &seg, VCPU_SREG_CS); | |
10433 | seg = (struct kvm_segment) { | |
10434 | .base = 0, | |
10435 | .limit = 0xFFFFFFFF, | |
10436 | .type = 3, | |
10437 | .present = 1, | |
10438 | .s = 1, | |
10439 | .db = 1, | |
10440 | .g = 1 | |
10441 | }; | |
10442 | seg.selector = vmcs12->host_ds_selector; | |
10443 | vmx_set_segment(vcpu, &seg, VCPU_SREG_DS); | |
10444 | seg.selector = vmcs12->host_es_selector; | |
10445 | vmx_set_segment(vcpu, &seg, VCPU_SREG_ES); | |
10446 | seg.selector = vmcs12->host_ss_selector; | |
10447 | vmx_set_segment(vcpu, &seg, VCPU_SREG_SS); | |
10448 | seg.selector = vmcs12->host_fs_selector; | |
10449 | seg.base = vmcs12->host_fs_base; | |
10450 | vmx_set_segment(vcpu, &seg, VCPU_SREG_FS); | |
10451 | seg.selector = vmcs12->host_gs_selector; | |
10452 | seg.base = vmcs12->host_gs_base; | |
10453 | vmx_set_segment(vcpu, &seg, VCPU_SREG_GS); | |
10454 | seg = (struct kvm_segment) { | |
205befd9 | 10455 | .base = vmcs12->host_tr_base, |
21feb4eb ACL |
10456 | .limit = 0x67, |
10457 | .selector = vmcs12->host_tr_selector, | |
10458 | .type = 11, | |
10459 | .present = 1 | |
10460 | }; | |
10461 | vmx_set_segment(vcpu, &seg, VCPU_SREG_TR); | |
10462 | ||
503cd0c5 JK |
10463 | kvm_set_dr(vcpu, 7, 0x400); |
10464 | vmcs_write64(GUEST_IA32_DEBUGCTL, 0); | |
ff651cb6 | 10465 | |
3af18d9c WV |
10466 | if (cpu_has_vmx_msr_bitmap()) |
10467 | vmx_set_msr_bitmap(vcpu); | |
10468 | ||
ff651cb6 WV |
10469 | if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr, |
10470 | vmcs12->vm_exit_msr_load_count)) | |
10471 | nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL); | |
4704d0be NHE |
10472 | } |
10473 | ||
10474 | /* | |
10475 | * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1 | |
10476 | * and modify vmcs12 to make it see what it would expect to see there if | |
10477 | * L2 was its real guest. Must only be called when in L2 (is_guest_mode()) | |
10478 | */ | |
533558bc JK |
10479 | static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason, |
10480 | u32 exit_intr_info, | |
10481 | unsigned long exit_qualification) | |
4704d0be NHE |
10482 | { |
10483 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
4704d0be NHE |
10484 | struct vmcs12 *vmcs12 = get_vmcs12(vcpu); |
10485 | ||
5f3d5799 JK |
10486 | /* trying to cancel vmlaunch/vmresume is a bug */ |
10487 | WARN_ON_ONCE(vmx->nested.nested_run_pending); | |
10488 | ||
4704d0be | 10489 | leave_guest_mode(vcpu); |
533558bc JK |
10490 | prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info, |
10491 | exit_qualification); | |
4704d0be | 10492 | |
ff651cb6 WV |
10493 | if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr, |
10494 | vmcs12->vm_exit_msr_store_count)) | |
10495 | nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL); | |
10496 | ||
f3380ca5 WL |
10497 | vmx_load_vmcs01(vcpu); |
10498 | ||
77b0f5d6 BD |
10499 | if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT) |
10500 | && nested_exit_intr_ack_set(vcpu)) { | |
10501 | int irq = kvm_cpu_get_interrupt(vcpu); | |
10502 | WARN_ON(irq < 0); | |
10503 | vmcs12->vm_exit_intr_info = irq | | |
10504 | INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR; | |
10505 | } | |
10506 | ||
542060ea JK |
10507 | trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason, |
10508 | vmcs12->exit_qualification, | |
10509 | vmcs12->idt_vectoring_info_field, | |
10510 | vmcs12->vm_exit_intr_info, | |
10511 | vmcs12->vm_exit_intr_error_code, | |
10512 | KVM_ISA_VMX); | |
4704d0be | 10513 | |
2961e876 GN |
10514 | vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS)); |
10515 | vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS)); | |
36c3cc42 JK |
10516 | vmx_segment_cache_clear(vmx); |
10517 | ||
4704d0be NHE |
10518 | /* if no vmcs02 cache requested, remove the one we used */ |
10519 | if (VMCS02_POOL_SIZE == 0) | |
10520 | nested_free_vmcs02(vmx, vmx->nested.current_vmptr); | |
10521 | ||
10522 | load_vmcs12_host_state(vcpu, vmcs12); | |
10523 | ||
27fc51b2 | 10524 | /* Update TSC_OFFSET if TSC was changed while L2 ran */ |
4704d0be NHE |
10525 | vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset); |
10526 | ||
10527 | /* This is needed for same reason as it was needed in prepare_vmcs02 */ | |
10528 | vmx->host_rsp = 0; | |
10529 | ||
10530 | /* Unpin physical memory we referred to in vmcs02 */ | |
10531 | if (vmx->nested.apic_access_page) { | |
10532 | nested_release_page(vmx->nested.apic_access_page); | |
48d89b92 | 10533 | vmx->nested.apic_access_page = NULL; |
4704d0be | 10534 | } |
a7c0b07d WL |
10535 | if (vmx->nested.virtual_apic_page) { |
10536 | nested_release_page(vmx->nested.virtual_apic_page); | |
48d89b92 | 10537 | vmx->nested.virtual_apic_page = NULL; |
a7c0b07d | 10538 | } |
705699a1 WV |
10539 | if (vmx->nested.pi_desc_page) { |
10540 | kunmap(vmx->nested.pi_desc_page); | |
10541 | nested_release_page(vmx->nested.pi_desc_page); | |
10542 | vmx->nested.pi_desc_page = NULL; | |
10543 | vmx->nested.pi_desc = NULL; | |
10544 | } | |
4704d0be | 10545 | |
38b99173 TC |
10546 | /* |
10547 | * We are now running in L2, mmu_notifier will force to reload the | |
10548 | * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1. | |
10549 | */ | |
10550 | kvm_vcpu_reload_apic_access_page(vcpu); | |
10551 | ||
4704d0be NHE |
10552 | /* |
10553 | * Exiting from L2 to L1, we're now back to L1 which thinks it just | |
10554 | * finished a VMLAUNCH or VMRESUME instruction, so we need to set the | |
10555 | * success or failure flag accordingly. | |
10556 | */ | |
10557 | if (unlikely(vmx->fail)) { | |
10558 | vmx->fail = 0; | |
10559 | nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR)); | |
10560 | } else | |
10561 | nested_vmx_succeed(vcpu); | |
012f83cb AG |
10562 | if (enable_shadow_vmcs) |
10563 | vmx->nested.sync_shadow_vmcs = true; | |
b6b8a145 JK |
10564 | |
10565 | /* in case we halted in L2 */ | |
10566 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; | |
4704d0be NHE |
10567 | } |
10568 | ||
42124925 JK |
10569 | /* |
10570 | * Forcibly leave nested mode in order to be able to reset the VCPU later on. | |
10571 | */ | |
10572 | static void vmx_leave_nested(struct kvm_vcpu *vcpu) | |
10573 | { | |
10574 | if (is_guest_mode(vcpu)) | |
533558bc | 10575 | nested_vmx_vmexit(vcpu, -1, 0, 0); |
42124925 JK |
10576 | free_nested(to_vmx(vcpu)); |
10577 | } | |
10578 | ||
7c177938 NHE |
10579 | /* |
10580 | * L1's failure to enter L2 is a subset of a normal exit, as explained in | |
10581 | * 23.7 "VM-entry failures during or after loading guest state" (this also | |
10582 | * lists the acceptable exit-reason and exit-qualification parameters). | |
10583 | * It should only be called before L2 actually succeeded to run, and when | |
10584 | * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss). | |
10585 | */ | |
10586 | static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu, | |
10587 | struct vmcs12 *vmcs12, | |
10588 | u32 reason, unsigned long qualification) | |
10589 | { | |
10590 | load_vmcs12_host_state(vcpu, vmcs12); | |
10591 | vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY; | |
10592 | vmcs12->exit_qualification = qualification; | |
10593 | nested_vmx_succeed(vcpu); | |
012f83cb AG |
10594 | if (enable_shadow_vmcs) |
10595 | to_vmx(vcpu)->nested.sync_shadow_vmcs = true; | |
7c177938 NHE |
10596 | } |
10597 | ||
8a76d7f2 JR |
10598 | static int vmx_check_intercept(struct kvm_vcpu *vcpu, |
10599 | struct x86_instruction_info *info, | |
10600 | enum x86_intercept_stage stage) | |
10601 | { | |
10602 | return X86EMUL_CONTINUE; | |
10603 | } | |
10604 | ||
48d89b92 | 10605 | static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu) |
ae97a3b8 | 10606 | { |
b4a2d31d RK |
10607 | if (ple_gap) |
10608 | shrink_ple_window(vcpu); | |
ae97a3b8 RK |
10609 | } |
10610 | ||
843e4330 KH |
10611 | static void vmx_slot_enable_log_dirty(struct kvm *kvm, |
10612 | struct kvm_memory_slot *slot) | |
10613 | { | |
10614 | kvm_mmu_slot_leaf_clear_dirty(kvm, slot); | |
10615 | kvm_mmu_slot_largepage_remove_write_access(kvm, slot); | |
10616 | } | |
10617 | ||
10618 | static void vmx_slot_disable_log_dirty(struct kvm *kvm, | |
10619 | struct kvm_memory_slot *slot) | |
10620 | { | |
10621 | kvm_mmu_slot_set_dirty(kvm, slot); | |
10622 | } | |
10623 | ||
10624 | static void vmx_flush_log_dirty(struct kvm *kvm) | |
10625 | { | |
10626 | kvm_flush_pml_buffers(kvm); | |
10627 | } | |
10628 | ||
10629 | static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm, | |
10630 | struct kvm_memory_slot *memslot, | |
10631 | gfn_t offset, unsigned long mask) | |
10632 | { | |
10633 | kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask); | |
10634 | } | |
10635 | ||
bf9f6ac8 FW |
10636 | /* |
10637 | * This routine does the following things for vCPU which is going | |
10638 | * to be blocked if VT-d PI is enabled. | |
10639 | * - Store the vCPU to the wakeup list, so when interrupts happen | |
10640 | * we can find the right vCPU to wake up. | |
10641 | * - Change the Posted-interrupt descriptor as below: | |
10642 | * 'NDST' <-- vcpu->pre_pcpu | |
10643 | * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR | |
10644 | * - If 'ON' is set during this process, which means at least one | |
10645 | * interrupt is posted for this vCPU, we cannot block it, in | |
10646 | * this case, return 1, otherwise, return 0. | |
10647 | * | |
10648 | */ | |
10649 | static int vmx_pre_block(struct kvm_vcpu *vcpu) | |
10650 | { | |
10651 | unsigned long flags; | |
10652 | unsigned int dest; | |
10653 | struct pi_desc old, new; | |
10654 | struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu); | |
10655 | ||
10656 | if (!kvm_arch_has_assigned_device(vcpu->kvm) || | |
10657 | !irq_remapping_cap(IRQ_POSTING_CAP)) | |
10658 | return 0; | |
10659 | ||
10660 | vcpu->pre_pcpu = vcpu->cpu; | |
10661 | spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock, | |
10662 | vcpu->pre_pcpu), flags); | |
10663 | list_add_tail(&vcpu->blocked_vcpu_list, | |
10664 | &per_cpu(blocked_vcpu_on_cpu, | |
10665 | vcpu->pre_pcpu)); | |
10666 | spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock, | |
10667 | vcpu->pre_pcpu), flags); | |
10668 | ||
10669 | do { | |
10670 | old.control = new.control = pi_desc->control; | |
10671 | ||
10672 | /* | |
10673 | * We should not block the vCPU if | |
10674 | * an interrupt is posted for it. | |
10675 | */ | |
10676 | if (pi_test_on(pi_desc) == 1) { | |
10677 | spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock, | |
10678 | vcpu->pre_pcpu), flags); | |
10679 | list_del(&vcpu->blocked_vcpu_list); | |
10680 | spin_unlock_irqrestore( | |
10681 | &per_cpu(blocked_vcpu_on_cpu_lock, | |
10682 | vcpu->pre_pcpu), flags); | |
10683 | vcpu->pre_pcpu = -1; | |
10684 | ||
10685 | return 1; | |
10686 | } | |
10687 | ||
10688 | WARN((pi_desc->sn == 1), | |
10689 | "Warning: SN field of posted-interrupts " | |
10690 | "is set before blocking\n"); | |
10691 | ||
10692 | /* | |
10693 | * Since vCPU can be preempted during this process, | |
10694 | * vcpu->cpu could be different with pre_pcpu, we | |
10695 | * need to set pre_pcpu as the destination of wakeup | |
10696 | * notification event, then we can find the right vCPU | |
10697 | * to wakeup in wakeup handler if interrupts happen | |
10698 | * when the vCPU is in blocked state. | |
10699 | */ | |
10700 | dest = cpu_physical_id(vcpu->pre_pcpu); | |
10701 | ||
10702 | if (x2apic_enabled()) | |
10703 | new.ndst = dest; | |
10704 | else | |
10705 | new.ndst = (dest << 8) & 0xFF00; | |
10706 | ||
10707 | /* set 'NV' to 'wakeup vector' */ | |
10708 | new.nv = POSTED_INTR_WAKEUP_VECTOR; | |
10709 | } while (cmpxchg(&pi_desc->control, old.control, | |
10710 | new.control) != old.control); | |
10711 | ||
10712 | return 0; | |
10713 | } | |
10714 | ||
10715 | static void vmx_post_block(struct kvm_vcpu *vcpu) | |
10716 | { | |
10717 | struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu); | |
10718 | struct pi_desc old, new; | |
10719 | unsigned int dest; | |
10720 | unsigned long flags; | |
10721 | ||
10722 | if (!kvm_arch_has_assigned_device(vcpu->kvm) || | |
10723 | !irq_remapping_cap(IRQ_POSTING_CAP)) | |
10724 | return; | |
10725 | ||
10726 | do { | |
10727 | old.control = new.control = pi_desc->control; | |
10728 | ||
10729 | dest = cpu_physical_id(vcpu->cpu); | |
10730 | ||
10731 | if (x2apic_enabled()) | |
10732 | new.ndst = dest; | |
10733 | else | |
10734 | new.ndst = (dest << 8) & 0xFF00; | |
10735 | ||
10736 | /* Allow posting non-urgent interrupts */ | |
10737 | new.sn = 0; | |
10738 | ||
10739 | /* set 'NV' to 'notification vector' */ | |
10740 | new.nv = POSTED_INTR_VECTOR; | |
10741 | } while (cmpxchg(&pi_desc->control, old.control, | |
10742 | new.control) != old.control); | |
10743 | ||
10744 | if(vcpu->pre_pcpu != -1) { | |
10745 | spin_lock_irqsave( | |
10746 | &per_cpu(blocked_vcpu_on_cpu_lock, | |
10747 | vcpu->pre_pcpu), flags); | |
10748 | list_del(&vcpu->blocked_vcpu_list); | |
10749 | spin_unlock_irqrestore( | |
10750 | &per_cpu(blocked_vcpu_on_cpu_lock, | |
10751 | vcpu->pre_pcpu), flags); | |
10752 | vcpu->pre_pcpu = -1; | |
10753 | } | |
10754 | } | |
10755 | ||
efc64404 FW |
10756 | /* |
10757 | * vmx_update_pi_irte - set IRTE for Posted-Interrupts | |
10758 | * | |
10759 | * @kvm: kvm | |
10760 | * @host_irq: host irq of the interrupt | |
10761 | * @guest_irq: gsi of the interrupt | |
10762 | * @set: set or unset PI | |
10763 | * returns 0 on success, < 0 on failure | |
10764 | */ | |
10765 | static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq, | |
10766 | uint32_t guest_irq, bool set) | |
10767 | { | |
10768 | struct kvm_kernel_irq_routing_entry *e; | |
10769 | struct kvm_irq_routing_table *irq_rt; | |
10770 | struct kvm_lapic_irq irq; | |
10771 | struct kvm_vcpu *vcpu; | |
10772 | struct vcpu_data vcpu_info; | |
10773 | int idx, ret = -EINVAL; | |
10774 | ||
10775 | if (!kvm_arch_has_assigned_device(kvm) || | |
10776 | !irq_remapping_cap(IRQ_POSTING_CAP)) | |
10777 | return 0; | |
10778 | ||
10779 | idx = srcu_read_lock(&kvm->irq_srcu); | |
10780 | irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu); | |
10781 | BUG_ON(guest_irq >= irq_rt->nr_rt_entries); | |
10782 | ||
10783 | hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) { | |
10784 | if (e->type != KVM_IRQ_ROUTING_MSI) | |
10785 | continue; | |
10786 | /* | |
10787 | * VT-d PI cannot support posting multicast/broadcast | |
10788 | * interrupts to a vCPU, we still use interrupt remapping | |
10789 | * for these kind of interrupts. | |
10790 | * | |
10791 | * For lowest-priority interrupts, we only support | |
10792 | * those with single CPU as the destination, e.g. user | |
10793 | * configures the interrupts via /proc/irq or uses | |
10794 | * irqbalance to make the interrupts single-CPU. | |
10795 | * | |
10796 | * We will support full lowest-priority interrupt later. | |
10797 | */ | |
10798 | ||
10799 | kvm_set_msi_irq(e, &irq); | |
23a1c257 FW |
10800 | if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) { |
10801 | /* | |
10802 | * Make sure the IRTE is in remapped mode if | |
10803 | * we don't handle it in posted mode. | |
10804 | */ | |
10805 | ret = irq_set_vcpu_affinity(host_irq, NULL); | |
10806 | if (ret < 0) { | |
10807 | printk(KERN_INFO | |
10808 | "failed to back to remapped mode, irq: %u\n", | |
10809 | host_irq); | |
10810 | goto out; | |
10811 | } | |
10812 | ||
efc64404 | 10813 | continue; |
23a1c257 | 10814 | } |
efc64404 FW |
10815 | |
10816 | vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu)); | |
10817 | vcpu_info.vector = irq.vector; | |
10818 | ||
b6ce9780 | 10819 | trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi, |
efc64404 FW |
10820 | vcpu_info.vector, vcpu_info.pi_desc_addr, set); |
10821 | ||
10822 | if (set) | |
10823 | ret = irq_set_vcpu_affinity(host_irq, &vcpu_info); | |
10824 | else { | |
10825 | /* suppress notification event before unposting */ | |
10826 | pi_set_sn(vcpu_to_pi_desc(vcpu)); | |
10827 | ret = irq_set_vcpu_affinity(host_irq, NULL); | |
10828 | pi_clear_sn(vcpu_to_pi_desc(vcpu)); | |
10829 | } | |
10830 | ||
10831 | if (ret < 0) { | |
10832 | printk(KERN_INFO "%s: failed to update PI IRTE\n", | |
10833 | __func__); | |
10834 | goto out; | |
10835 | } | |
10836 | } | |
10837 | ||
10838 | ret = 0; | |
10839 | out: | |
10840 | srcu_read_unlock(&kvm->irq_srcu, idx); | |
10841 | return ret; | |
10842 | } | |
10843 | ||
cbdd1bea | 10844 | static struct kvm_x86_ops vmx_x86_ops = { |
6aa8b732 AK |
10845 | .cpu_has_kvm_support = cpu_has_kvm_support, |
10846 | .disabled_by_bios = vmx_disabled_by_bios, | |
10847 | .hardware_setup = hardware_setup, | |
10848 | .hardware_unsetup = hardware_unsetup, | |
002c7f7c | 10849 | .check_processor_compatibility = vmx_check_processor_compat, |
6aa8b732 AK |
10850 | .hardware_enable = hardware_enable, |
10851 | .hardware_disable = hardware_disable, | |
04547156 | 10852 | .cpu_has_accelerated_tpr = report_flexpriority, |
6d396b55 | 10853 | .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase, |
6aa8b732 AK |
10854 | |
10855 | .vcpu_create = vmx_create_vcpu, | |
10856 | .vcpu_free = vmx_free_vcpu, | |
04d2cc77 | 10857 | .vcpu_reset = vmx_vcpu_reset, |
6aa8b732 | 10858 | |
04d2cc77 | 10859 | .prepare_guest_switch = vmx_save_host_state, |
6aa8b732 AK |
10860 | .vcpu_load = vmx_vcpu_load, |
10861 | .vcpu_put = vmx_vcpu_put, | |
10862 | ||
a96036b8 | 10863 | .update_bp_intercept = update_exception_bitmap, |
6aa8b732 AK |
10864 | .get_msr = vmx_get_msr, |
10865 | .set_msr = vmx_set_msr, | |
10866 | .get_segment_base = vmx_get_segment_base, | |
10867 | .get_segment = vmx_get_segment, | |
10868 | .set_segment = vmx_set_segment, | |
2e4d2653 | 10869 | .get_cpl = vmx_get_cpl, |
6aa8b732 | 10870 | .get_cs_db_l_bits = vmx_get_cs_db_l_bits, |
e8467fda | 10871 | .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits, |
aff48baa | 10872 | .decache_cr3 = vmx_decache_cr3, |
25c4c276 | 10873 | .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits, |
6aa8b732 | 10874 | .set_cr0 = vmx_set_cr0, |
6aa8b732 AK |
10875 | .set_cr3 = vmx_set_cr3, |
10876 | .set_cr4 = vmx_set_cr4, | |
6aa8b732 | 10877 | .set_efer = vmx_set_efer, |
6aa8b732 AK |
10878 | .get_idt = vmx_get_idt, |
10879 | .set_idt = vmx_set_idt, | |
10880 | .get_gdt = vmx_get_gdt, | |
10881 | .set_gdt = vmx_set_gdt, | |
73aaf249 JK |
10882 | .get_dr6 = vmx_get_dr6, |
10883 | .set_dr6 = vmx_set_dr6, | |
020df079 | 10884 | .set_dr7 = vmx_set_dr7, |
81908bf4 | 10885 | .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs, |
5fdbf976 | 10886 | .cache_reg = vmx_cache_reg, |
6aa8b732 AK |
10887 | .get_rflags = vmx_get_rflags, |
10888 | .set_rflags = vmx_set_rflags, | |
0fdd74f7 | 10889 | .fpu_activate = vmx_fpu_activate, |
02daab21 | 10890 | .fpu_deactivate = vmx_fpu_deactivate, |
6aa8b732 AK |
10891 | |
10892 | .tlb_flush = vmx_flush_tlb, | |
6aa8b732 | 10893 | |
6aa8b732 | 10894 | .run = vmx_vcpu_run, |
6062d012 | 10895 | .handle_exit = vmx_handle_exit, |
6aa8b732 | 10896 | .skip_emulated_instruction = skip_emulated_instruction, |
2809f5d2 GC |
10897 | .set_interrupt_shadow = vmx_set_interrupt_shadow, |
10898 | .get_interrupt_shadow = vmx_get_interrupt_shadow, | |
102d8325 | 10899 | .patch_hypercall = vmx_patch_hypercall, |
2a8067f1 | 10900 | .set_irq = vmx_inject_irq, |
95ba8273 | 10901 | .set_nmi = vmx_inject_nmi, |
298101da | 10902 | .queue_exception = vmx_queue_exception, |
b463a6f7 | 10903 | .cancel_injection = vmx_cancel_injection, |
78646121 | 10904 | .interrupt_allowed = vmx_interrupt_allowed, |
95ba8273 | 10905 | .nmi_allowed = vmx_nmi_allowed, |
3cfc3092 JK |
10906 | .get_nmi_mask = vmx_get_nmi_mask, |
10907 | .set_nmi_mask = vmx_set_nmi_mask, | |
95ba8273 GN |
10908 | .enable_nmi_window = enable_nmi_window, |
10909 | .enable_irq_window = enable_irq_window, | |
10910 | .update_cr8_intercept = update_cr8_intercept, | |
8d14695f | 10911 | .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode, |
38b99173 | 10912 | .set_apic_access_page_addr = vmx_set_apic_access_page_addr, |
d62caabb AS |
10913 | .get_enable_apicv = vmx_get_enable_apicv, |
10914 | .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl, | |
c7c9c56c YZ |
10915 | .load_eoi_exitmap = vmx_load_eoi_exitmap, |
10916 | .hwapic_irr_update = vmx_hwapic_irr_update, | |
10917 | .hwapic_isr_update = vmx_hwapic_isr_update, | |
a20ed54d YZ |
10918 | .sync_pir_to_irr = vmx_sync_pir_to_irr, |
10919 | .deliver_posted_interrupt = vmx_deliver_posted_interrupt, | |
95ba8273 | 10920 | |
cbc94022 | 10921 | .set_tss_addr = vmx_set_tss_addr, |
67253af5 | 10922 | .get_tdp_level = get_ept_level, |
4b12f0de | 10923 | .get_mt_mask = vmx_get_mt_mask, |
229456fc | 10924 | |
586f9607 | 10925 | .get_exit_info = vmx_get_exit_info, |
586f9607 | 10926 | |
17cc3935 | 10927 | .get_lpage_level = vmx_get_lpage_level, |
0e851880 SY |
10928 | |
10929 | .cpuid_update = vmx_cpuid_update, | |
4e47c7a6 SY |
10930 | |
10931 | .rdtscp_supported = vmx_rdtscp_supported, | |
ad756a16 | 10932 | .invpcid_supported = vmx_invpcid_supported, |
d4330ef2 JR |
10933 | |
10934 | .set_supported_cpuid = vmx_set_supported_cpuid, | |
f5f48ee1 SY |
10935 | |
10936 | .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit, | |
99e3e30a | 10937 | |
ba904635 | 10938 | .read_tsc_offset = vmx_read_tsc_offset, |
99e3e30a | 10939 | .write_tsc_offset = vmx_write_tsc_offset, |
58ea6767 | 10940 | .adjust_tsc_offset_guest = vmx_adjust_tsc_offset_guest, |
d5c1785d | 10941 | .read_l1_tsc = vmx_read_l1_tsc, |
1c97f0a0 JR |
10942 | |
10943 | .set_tdp_cr3 = vmx_set_cr3, | |
8a76d7f2 JR |
10944 | |
10945 | .check_intercept = vmx_check_intercept, | |
a547c6db | 10946 | .handle_external_intr = vmx_handle_external_intr, |
da8999d3 | 10947 | .mpx_supported = vmx_mpx_supported, |
55412b2e | 10948 | .xsaves_supported = vmx_xsaves_supported, |
b6b8a145 JK |
10949 | |
10950 | .check_nested_events = vmx_check_nested_events, | |
ae97a3b8 RK |
10951 | |
10952 | .sched_in = vmx_sched_in, | |
843e4330 KH |
10953 | |
10954 | .slot_enable_log_dirty = vmx_slot_enable_log_dirty, | |
10955 | .slot_disable_log_dirty = vmx_slot_disable_log_dirty, | |
10956 | .flush_log_dirty = vmx_flush_log_dirty, | |
10957 | .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked, | |
25462f7f | 10958 | |
bf9f6ac8 FW |
10959 | .pre_block = vmx_pre_block, |
10960 | .post_block = vmx_post_block, | |
10961 | ||
25462f7f | 10962 | .pmu_ops = &intel_pmu_ops, |
efc64404 FW |
10963 | |
10964 | .update_pi_irte = vmx_update_pi_irte, | |
6aa8b732 AK |
10965 | }; |
10966 | ||
10967 | static int __init vmx_init(void) | |
10968 | { | |
34a1cd60 TC |
10969 | int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), |
10970 | __alignof__(struct vcpu_vmx), THIS_MODULE); | |
fdef3ad1 | 10971 | if (r) |
34a1cd60 | 10972 | return r; |
25c5f225 | 10973 | |
2965faa5 | 10974 | #ifdef CONFIG_KEXEC_CORE |
8f536b76 ZY |
10975 | rcu_assign_pointer(crash_vmclear_loaded_vmcss, |
10976 | crash_vmclear_local_loaded_vmcss); | |
10977 | #endif | |
10978 | ||
fdef3ad1 | 10979 | return 0; |
6aa8b732 AK |
10980 | } |
10981 | ||
10982 | static void __exit vmx_exit(void) | |
10983 | { | |
2965faa5 | 10984 | #ifdef CONFIG_KEXEC_CORE |
3b63a43f | 10985 | RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL); |
8f536b76 ZY |
10986 | synchronize_rcu(); |
10987 | #endif | |
10988 | ||
cb498ea2 | 10989 | kvm_exit(); |
6aa8b732 AK |
10990 | } |
10991 | ||
10992 | module_init(vmx_init) | |
10993 | module_exit(vmx_exit) |