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. | |
8 | * | |
9 | * Authors: | |
10 | * Avi Kivity <avi@qumranet.com> | |
11 | * Yaniv Kamay <yaniv@qumranet.com> | |
12 | * | |
13 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
14 | * the COPYING file in the top-level directory. | |
15 | * | |
16 | */ | |
17 | ||
85f455f7 | 18 | #include "irq.h" |
1d737c8a | 19 | #include "mmu.h" |
e495606d | 20 | |
edf88417 | 21 | #include <linux/kvm_host.h> |
6aa8b732 | 22 | #include <linux/module.h> |
9d8f549d | 23 | #include <linux/kernel.h> |
6aa8b732 AK |
24 | #include <linux/mm.h> |
25 | #include <linux/highmem.h> | |
e8edc6e0 | 26 | #include <linux/sched.h> |
c7addb90 | 27 | #include <linux/moduleparam.h> |
229456fc | 28 | #include <linux/ftrace_event.h> |
5fdbf976 | 29 | #include "kvm_cache_regs.h" |
35920a35 | 30 | #include "x86.h" |
e495606d | 31 | |
6aa8b732 | 32 | #include <asm/io.h> |
3b3be0d1 | 33 | #include <asm/desc.h> |
13673a90 | 34 | #include <asm/vmx.h> |
6210e37b | 35 | #include <asm/virtext.h> |
a0861c02 | 36 | #include <asm/mce.h> |
6aa8b732 | 37 | |
229456fc MT |
38 | #include "trace.h" |
39 | ||
4ecac3fd AK |
40 | #define __ex(x) __kvm_handle_fault_on_reboot(x) |
41 | ||
6aa8b732 AK |
42 | MODULE_AUTHOR("Qumranet"); |
43 | MODULE_LICENSE("GPL"); | |
44 | ||
4462d21a | 45 | static int __read_mostly bypass_guest_pf = 1; |
c1f8bc04 | 46 | module_param(bypass_guest_pf, bool, S_IRUGO); |
c7addb90 | 47 | |
4462d21a | 48 | static int __read_mostly enable_vpid = 1; |
736caefe | 49 | module_param_named(vpid, enable_vpid, bool, 0444); |
2384d2b3 | 50 | |
4462d21a | 51 | static int __read_mostly flexpriority_enabled = 1; |
736caefe | 52 | module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO); |
4c9fc8ef | 53 | |
4462d21a | 54 | static int __read_mostly enable_ept = 1; |
736caefe | 55 | module_param_named(ept, enable_ept, bool, S_IRUGO); |
d56f546d | 56 | |
3a624e29 NK |
57 | static int __read_mostly enable_unrestricted_guest = 1; |
58 | module_param_named(unrestricted_guest, | |
59 | enable_unrestricted_guest, bool, S_IRUGO); | |
60 | ||
4462d21a | 61 | static int __read_mostly emulate_invalid_guest_state = 0; |
c1f8bc04 | 62 | module_param(emulate_invalid_guest_state, bool, S_IRUGO); |
04fa4d32 | 63 | |
4b8d54f9 ZE |
64 | /* |
65 | * These 2 parameters are used to config the controls for Pause-Loop Exiting: | |
66 | * ple_gap: upper bound on the amount of time between two successive | |
67 | * executions of PAUSE in a loop. Also indicate if ple enabled. | |
68 | * According to test, this time is usually small than 41 cycles. | |
69 | * ple_window: upper bound on the amount of time a guest is allowed to execute | |
70 | * in a PAUSE loop. Tests indicate that most spinlocks are held for | |
71 | * less than 2^12 cycles | |
72 | * Time is measured based on a counter that runs at the same rate as the TSC, | |
73 | * refer SDM volume 3b section 21.6.13 & 22.1.3. | |
74 | */ | |
75 | #define KVM_VMX_DEFAULT_PLE_GAP 41 | |
76 | #define KVM_VMX_DEFAULT_PLE_WINDOW 4096 | |
77 | static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP; | |
78 | module_param(ple_gap, int, S_IRUGO); | |
79 | ||
80 | static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW; | |
81 | module_param(ple_window, int, S_IRUGO); | |
82 | ||
a2fa3e9f GH |
83 | struct vmcs { |
84 | u32 revision_id; | |
85 | u32 abort; | |
86 | char data[0]; | |
87 | }; | |
88 | ||
89 | struct vcpu_vmx { | |
fb3f0f51 | 90 | struct kvm_vcpu vcpu; |
543e4243 | 91 | struct list_head local_vcpus_link; |
313dbd49 | 92 | unsigned long host_rsp; |
a2fa3e9f | 93 | int launched; |
29bd8a78 | 94 | u8 fail; |
1155f76a | 95 | u32 idt_vectoring_info; |
a2fa3e9f GH |
96 | struct kvm_msr_entry *guest_msrs; |
97 | struct kvm_msr_entry *host_msrs; | |
98 | int nmsrs; | |
99 | int save_nmsrs; | |
100 | int msr_offset_efer; | |
101 | #ifdef CONFIG_X86_64 | |
102 | int msr_offset_kernel_gs_base; | |
103 | #endif | |
104 | struct vmcs *vmcs; | |
105 | struct { | |
106 | int loaded; | |
107 | u16 fs_sel, gs_sel, ldt_sel; | |
152d3f2f LV |
108 | int gs_ldt_reload_needed; |
109 | int fs_reload_needed; | |
51c6cf66 | 110 | int guest_efer_loaded; |
d77c26fc | 111 | } host_state; |
9c8cba37 | 112 | struct { |
7ffd92c5 AK |
113 | int vm86_active; |
114 | u8 save_iopl; | |
115 | struct kvm_save_segment { | |
116 | u16 selector; | |
117 | unsigned long base; | |
118 | u32 limit; | |
119 | u32 ar; | |
120 | } tr, es, ds, fs, gs; | |
9c8cba37 AK |
121 | struct { |
122 | bool pending; | |
123 | u8 vector; | |
124 | unsigned rip; | |
125 | } irq; | |
126 | } rmode; | |
2384d2b3 | 127 | int vpid; |
04fa4d32 | 128 | bool emulation_required; |
3b86cd99 JK |
129 | |
130 | /* Support for vnmi-less CPUs */ | |
131 | int soft_vnmi_blocked; | |
132 | ktime_t entry_time; | |
133 | s64 vnmi_blocked_time; | |
a0861c02 | 134 | u32 exit_reason; |
a2fa3e9f GH |
135 | }; |
136 | ||
137 | static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu) | |
138 | { | |
fb3f0f51 | 139 | return container_of(vcpu, struct vcpu_vmx, vcpu); |
a2fa3e9f GH |
140 | } |
141 | ||
b7ebfb05 | 142 | static int init_rmode(struct kvm *kvm); |
4e1096d2 | 143 | static u64 construct_eptp(unsigned long root_hpa); |
75880a01 | 144 | |
6aa8b732 AK |
145 | static DEFINE_PER_CPU(struct vmcs *, vmxarea); |
146 | static DEFINE_PER_CPU(struct vmcs *, current_vmcs); | |
543e4243 | 147 | static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu); |
6aa8b732 | 148 | |
3e7c73e9 AK |
149 | static unsigned long *vmx_io_bitmap_a; |
150 | static unsigned long *vmx_io_bitmap_b; | |
5897297b AK |
151 | static unsigned long *vmx_msr_bitmap_legacy; |
152 | static unsigned long *vmx_msr_bitmap_longmode; | |
fdef3ad1 | 153 | |
2384d2b3 SY |
154 | static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS); |
155 | static DEFINE_SPINLOCK(vmx_vpid_lock); | |
156 | ||
1c3d14fe | 157 | static struct vmcs_config { |
6aa8b732 AK |
158 | int size; |
159 | int order; | |
160 | u32 revision_id; | |
1c3d14fe YS |
161 | u32 pin_based_exec_ctrl; |
162 | u32 cpu_based_exec_ctrl; | |
f78e0e2e | 163 | u32 cpu_based_2nd_exec_ctrl; |
1c3d14fe YS |
164 | u32 vmexit_ctrl; |
165 | u32 vmentry_ctrl; | |
166 | } vmcs_config; | |
6aa8b732 | 167 | |
efff9e53 | 168 | static struct vmx_capability { |
d56f546d SY |
169 | u32 ept; |
170 | u32 vpid; | |
171 | } vmx_capability; | |
172 | ||
6aa8b732 AK |
173 | #define VMX_SEGMENT_FIELD(seg) \ |
174 | [VCPU_SREG_##seg] = { \ | |
175 | .selector = GUEST_##seg##_SELECTOR, \ | |
176 | .base = GUEST_##seg##_BASE, \ | |
177 | .limit = GUEST_##seg##_LIMIT, \ | |
178 | .ar_bytes = GUEST_##seg##_AR_BYTES, \ | |
179 | } | |
180 | ||
181 | static struct kvm_vmx_segment_field { | |
182 | unsigned selector; | |
183 | unsigned base; | |
184 | unsigned limit; | |
185 | unsigned ar_bytes; | |
186 | } kvm_vmx_segment_fields[] = { | |
187 | VMX_SEGMENT_FIELD(CS), | |
188 | VMX_SEGMENT_FIELD(DS), | |
189 | VMX_SEGMENT_FIELD(ES), | |
190 | VMX_SEGMENT_FIELD(FS), | |
191 | VMX_SEGMENT_FIELD(GS), | |
192 | VMX_SEGMENT_FIELD(SS), | |
193 | VMX_SEGMENT_FIELD(TR), | |
194 | VMX_SEGMENT_FIELD(LDTR), | |
195 | }; | |
196 | ||
6de4f3ad AK |
197 | static void ept_save_pdptrs(struct kvm_vcpu *vcpu); |
198 | ||
4d56c8a7 AK |
199 | /* |
200 | * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it | |
201 | * away by decrementing the array size. | |
202 | */ | |
6aa8b732 | 203 | static const u32 vmx_msr_index[] = { |
05b3e0c2 | 204 | #ifdef CONFIG_X86_64 |
6aa8b732 AK |
205 | MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE, |
206 | #endif | |
207 | MSR_EFER, MSR_K6_STAR, | |
208 | }; | |
9d8f549d | 209 | #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index) |
6aa8b732 | 210 | |
a2fa3e9f GH |
211 | static void load_msrs(struct kvm_msr_entry *e, int n) |
212 | { | |
213 | int i; | |
214 | ||
215 | for (i = 0; i < n; ++i) | |
216 | wrmsrl(e[i].index, e[i].data); | |
217 | } | |
218 | ||
219 | static void save_msrs(struct kvm_msr_entry *e, int n) | |
220 | { | |
221 | int i; | |
222 | ||
223 | for (i = 0; i < n; ++i) | |
224 | rdmsrl(e[i].index, e[i].data); | |
225 | } | |
226 | ||
6aa8b732 AK |
227 | static inline int is_page_fault(u32 intr_info) |
228 | { | |
229 | return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK | | |
230 | INTR_INFO_VALID_MASK)) == | |
8ab2d2e2 | 231 | (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK); |
6aa8b732 AK |
232 | } |
233 | ||
2ab455cc AL |
234 | static inline int is_no_device(u32 intr_info) |
235 | { | |
236 | return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK | | |
237 | INTR_INFO_VALID_MASK)) == | |
8ab2d2e2 | 238 | (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK); |
2ab455cc AL |
239 | } |
240 | ||
7aa81cc0 AL |
241 | static inline int is_invalid_opcode(u32 intr_info) |
242 | { | |
243 | return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK | | |
244 | INTR_INFO_VALID_MASK)) == | |
8ab2d2e2 | 245 | (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK); |
7aa81cc0 AL |
246 | } |
247 | ||
6aa8b732 AK |
248 | static inline int is_external_interrupt(u32 intr_info) |
249 | { | |
250 | return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK)) | |
251 | == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK); | |
252 | } | |
253 | ||
a0861c02 AK |
254 | static inline int is_machine_check(u32 intr_info) |
255 | { | |
256 | return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK | | |
257 | INTR_INFO_VALID_MASK)) == | |
258 | (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK); | |
259 | } | |
260 | ||
25c5f225 SY |
261 | static inline int cpu_has_vmx_msr_bitmap(void) |
262 | { | |
04547156 | 263 | return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS; |
25c5f225 SY |
264 | } |
265 | ||
6e5d865c YS |
266 | static inline int cpu_has_vmx_tpr_shadow(void) |
267 | { | |
04547156 | 268 | return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW; |
6e5d865c YS |
269 | } |
270 | ||
271 | static inline int vm_need_tpr_shadow(struct kvm *kvm) | |
272 | { | |
04547156 | 273 | return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm)); |
6e5d865c YS |
274 | } |
275 | ||
f78e0e2e SY |
276 | static inline int cpu_has_secondary_exec_ctrls(void) |
277 | { | |
04547156 SY |
278 | return vmcs_config.cpu_based_exec_ctrl & |
279 | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS; | |
f78e0e2e SY |
280 | } |
281 | ||
774ead3a | 282 | static inline bool cpu_has_vmx_virtualize_apic_accesses(void) |
f78e0e2e | 283 | { |
04547156 SY |
284 | return vmcs_config.cpu_based_2nd_exec_ctrl & |
285 | SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES; | |
286 | } | |
287 | ||
288 | static inline bool cpu_has_vmx_flexpriority(void) | |
289 | { | |
290 | return cpu_has_vmx_tpr_shadow() && | |
291 | cpu_has_vmx_virtualize_apic_accesses(); | |
f78e0e2e SY |
292 | } |
293 | ||
e799794e MT |
294 | static inline bool cpu_has_vmx_ept_execute_only(void) |
295 | { | |
296 | return !!(vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT); | |
297 | } | |
298 | ||
299 | static inline bool cpu_has_vmx_eptp_uncacheable(void) | |
300 | { | |
301 | return !!(vmx_capability.ept & VMX_EPTP_UC_BIT); | |
302 | } | |
303 | ||
304 | static inline bool cpu_has_vmx_eptp_writeback(void) | |
305 | { | |
306 | return !!(vmx_capability.ept & VMX_EPTP_WB_BIT); | |
307 | } | |
308 | ||
309 | static inline bool cpu_has_vmx_ept_2m_page(void) | |
310 | { | |
311 | return !!(vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT); | |
312 | } | |
313 | ||
d56f546d SY |
314 | static inline int cpu_has_vmx_invept_individual_addr(void) |
315 | { | |
04547156 | 316 | return !!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT); |
d56f546d SY |
317 | } |
318 | ||
319 | static inline int cpu_has_vmx_invept_context(void) | |
320 | { | |
04547156 | 321 | return !!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT); |
d56f546d SY |
322 | } |
323 | ||
324 | static inline int cpu_has_vmx_invept_global(void) | |
325 | { | |
04547156 | 326 | return !!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT); |
d56f546d SY |
327 | } |
328 | ||
329 | static inline int cpu_has_vmx_ept(void) | |
330 | { | |
04547156 SY |
331 | return vmcs_config.cpu_based_2nd_exec_ctrl & |
332 | SECONDARY_EXEC_ENABLE_EPT; | |
d56f546d SY |
333 | } |
334 | ||
3a624e29 NK |
335 | static inline int cpu_has_vmx_unrestricted_guest(void) |
336 | { | |
337 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
338 | SECONDARY_EXEC_UNRESTRICTED_GUEST; | |
339 | } | |
340 | ||
4b8d54f9 ZE |
341 | static inline int cpu_has_vmx_ple(void) |
342 | { | |
343 | return vmcs_config.cpu_based_2nd_exec_ctrl & | |
344 | SECONDARY_EXEC_PAUSE_LOOP_EXITING; | |
345 | } | |
346 | ||
f78e0e2e SY |
347 | static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm) |
348 | { | |
04547156 SY |
349 | return flexpriority_enabled && |
350 | (cpu_has_vmx_virtualize_apic_accesses()) && | |
351 | (irqchip_in_kernel(kvm)); | |
f78e0e2e SY |
352 | } |
353 | ||
2384d2b3 SY |
354 | static inline int cpu_has_vmx_vpid(void) |
355 | { | |
04547156 SY |
356 | return vmcs_config.cpu_based_2nd_exec_ctrl & |
357 | SECONDARY_EXEC_ENABLE_VPID; | |
2384d2b3 SY |
358 | } |
359 | ||
f08864b4 SY |
360 | static inline int cpu_has_virtual_nmis(void) |
361 | { | |
362 | return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS; | |
363 | } | |
364 | ||
04547156 SY |
365 | static inline bool report_flexpriority(void) |
366 | { | |
367 | return flexpriority_enabled; | |
368 | } | |
369 | ||
8b9cf98c | 370 | static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr) |
7725f0ba AK |
371 | { |
372 | int i; | |
373 | ||
a2fa3e9f GH |
374 | for (i = 0; i < vmx->nmsrs; ++i) |
375 | if (vmx->guest_msrs[i].index == msr) | |
a75beee6 ED |
376 | return i; |
377 | return -1; | |
378 | } | |
379 | ||
2384d2b3 SY |
380 | static inline void __invvpid(int ext, u16 vpid, gva_t gva) |
381 | { | |
382 | struct { | |
383 | u64 vpid : 16; | |
384 | u64 rsvd : 48; | |
385 | u64 gva; | |
386 | } operand = { vpid, 0, gva }; | |
387 | ||
4ecac3fd | 388 | asm volatile (__ex(ASM_VMX_INVVPID) |
2384d2b3 SY |
389 | /* CF==1 or ZF==1 --> rc = -1 */ |
390 | "; ja 1f ; ud2 ; 1:" | |
391 | : : "a"(&operand), "c"(ext) : "cc", "memory"); | |
392 | } | |
393 | ||
1439442c SY |
394 | static inline void __invept(int ext, u64 eptp, gpa_t gpa) |
395 | { | |
396 | struct { | |
397 | u64 eptp, gpa; | |
398 | } operand = {eptp, gpa}; | |
399 | ||
4ecac3fd | 400 | asm volatile (__ex(ASM_VMX_INVEPT) |
1439442c SY |
401 | /* CF==1 or ZF==1 --> rc = -1 */ |
402 | "; ja 1f ; ud2 ; 1:\n" | |
403 | : : "a" (&operand), "c" (ext) : "cc", "memory"); | |
404 | } | |
405 | ||
8b9cf98c | 406 | static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr) |
a75beee6 ED |
407 | { |
408 | int i; | |
409 | ||
8b9cf98c | 410 | i = __find_msr_index(vmx, msr); |
a75beee6 | 411 | if (i >= 0) |
a2fa3e9f | 412 | return &vmx->guest_msrs[i]; |
8b6d44c7 | 413 | return NULL; |
7725f0ba AK |
414 | } |
415 | ||
6aa8b732 AK |
416 | static void vmcs_clear(struct vmcs *vmcs) |
417 | { | |
418 | u64 phys_addr = __pa(vmcs); | |
419 | u8 error; | |
420 | ||
4ecac3fd | 421 | asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0" |
6aa8b732 AK |
422 | : "=g"(error) : "a"(&phys_addr), "m"(phys_addr) |
423 | : "cc", "memory"); | |
424 | if (error) | |
425 | printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n", | |
426 | vmcs, phys_addr); | |
427 | } | |
428 | ||
429 | static void __vcpu_clear(void *arg) | |
430 | { | |
8b9cf98c | 431 | struct vcpu_vmx *vmx = arg; |
d3b2c338 | 432 | int cpu = raw_smp_processor_id(); |
6aa8b732 | 433 | |
8b9cf98c | 434 | if (vmx->vcpu.cpu == cpu) |
a2fa3e9f GH |
435 | vmcs_clear(vmx->vmcs); |
436 | if (per_cpu(current_vmcs, cpu) == vmx->vmcs) | |
6aa8b732 | 437 | per_cpu(current_vmcs, cpu) = NULL; |
ad312c7c | 438 | rdtscll(vmx->vcpu.arch.host_tsc); |
543e4243 AK |
439 | list_del(&vmx->local_vcpus_link); |
440 | vmx->vcpu.cpu = -1; | |
441 | vmx->launched = 0; | |
6aa8b732 AK |
442 | } |
443 | ||
8b9cf98c | 444 | static void vcpu_clear(struct vcpu_vmx *vmx) |
8d0be2b3 | 445 | { |
eae5ecb5 AK |
446 | if (vmx->vcpu.cpu == -1) |
447 | return; | |
8691e5a8 | 448 | smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1); |
8d0be2b3 AK |
449 | } |
450 | ||
2384d2b3 SY |
451 | static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx) |
452 | { | |
453 | if (vmx->vpid == 0) | |
454 | return; | |
455 | ||
456 | __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0); | |
457 | } | |
458 | ||
1439442c SY |
459 | static inline void ept_sync_global(void) |
460 | { | |
461 | if (cpu_has_vmx_invept_global()) | |
462 | __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0); | |
463 | } | |
464 | ||
465 | static inline void ept_sync_context(u64 eptp) | |
466 | { | |
089d034e | 467 | if (enable_ept) { |
1439442c SY |
468 | if (cpu_has_vmx_invept_context()) |
469 | __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0); | |
470 | else | |
471 | ept_sync_global(); | |
472 | } | |
473 | } | |
474 | ||
475 | static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa) | |
476 | { | |
089d034e | 477 | if (enable_ept) { |
1439442c SY |
478 | if (cpu_has_vmx_invept_individual_addr()) |
479 | __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR, | |
480 | eptp, gpa); | |
481 | else | |
482 | ept_sync_context(eptp); | |
483 | } | |
484 | } | |
485 | ||
6aa8b732 AK |
486 | static unsigned long vmcs_readl(unsigned long field) |
487 | { | |
488 | unsigned long value; | |
489 | ||
4ecac3fd | 490 | asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX) |
6aa8b732 AK |
491 | : "=a"(value) : "d"(field) : "cc"); |
492 | return value; | |
493 | } | |
494 | ||
495 | static u16 vmcs_read16(unsigned long field) | |
496 | { | |
497 | return vmcs_readl(field); | |
498 | } | |
499 | ||
500 | static u32 vmcs_read32(unsigned long field) | |
501 | { | |
502 | return vmcs_readl(field); | |
503 | } | |
504 | ||
505 | static u64 vmcs_read64(unsigned long field) | |
506 | { | |
05b3e0c2 | 507 | #ifdef CONFIG_X86_64 |
6aa8b732 AK |
508 | return vmcs_readl(field); |
509 | #else | |
510 | return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32); | |
511 | #endif | |
512 | } | |
513 | ||
e52de1b8 AK |
514 | static noinline void vmwrite_error(unsigned long field, unsigned long value) |
515 | { | |
516 | printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n", | |
517 | field, value, vmcs_read32(VM_INSTRUCTION_ERROR)); | |
518 | dump_stack(); | |
519 | } | |
520 | ||
6aa8b732 AK |
521 | static void vmcs_writel(unsigned long field, unsigned long value) |
522 | { | |
523 | u8 error; | |
524 | ||
4ecac3fd | 525 | asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0" |
d77c26fc | 526 | : "=q"(error) : "a"(value), "d"(field) : "cc"); |
e52de1b8 AK |
527 | if (unlikely(error)) |
528 | vmwrite_error(field, value); | |
6aa8b732 AK |
529 | } |
530 | ||
531 | static void vmcs_write16(unsigned long field, u16 value) | |
532 | { | |
533 | vmcs_writel(field, value); | |
534 | } | |
535 | ||
536 | static void vmcs_write32(unsigned long field, u32 value) | |
537 | { | |
538 | vmcs_writel(field, value); | |
539 | } | |
540 | ||
541 | static void vmcs_write64(unsigned long field, u64 value) | |
542 | { | |
6aa8b732 | 543 | vmcs_writel(field, value); |
7682f2d0 | 544 | #ifndef CONFIG_X86_64 |
6aa8b732 AK |
545 | asm volatile (""); |
546 | vmcs_writel(field+1, value >> 32); | |
547 | #endif | |
548 | } | |
549 | ||
2ab455cc AL |
550 | static void vmcs_clear_bits(unsigned long field, u32 mask) |
551 | { | |
552 | vmcs_writel(field, vmcs_readl(field) & ~mask); | |
553 | } | |
554 | ||
555 | static void vmcs_set_bits(unsigned long field, u32 mask) | |
556 | { | |
557 | vmcs_writel(field, vmcs_readl(field) | mask); | |
558 | } | |
559 | ||
abd3f2d6 AK |
560 | static void update_exception_bitmap(struct kvm_vcpu *vcpu) |
561 | { | |
562 | u32 eb; | |
563 | ||
a0861c02 | 564 | eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR); |
abd3f2d6 AK |
565 | if (!vcpu->fpu_active) |
566 | eb |= 1u << NM_VECTOR; | |
e8a48342 AK |
567 | /* |
568 | * Unconditionally intercept #DB so we can maintain dr6 without | |
569 | * reading it every exit. | |
570 | */ | |
571 | eb |= 1u << DB_VECTOR; | |
d0bfb940 | 572 | if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) { |
d0bfb940 JK |
573 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) |
574 | eb |= 1u << BP_VECTOR; | |
575 | } | |
7ffd92c5 | 576 | if (to_vmx(vcpu)->rmode.vm86_active) |
abd3f2d6 | 577 | eb = ~0; |
089d034e | 578 | if (enable_ept) |
1439442c | 579 | eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */ |
abd3f2d6 AK |
580 | vmcs_write32(EXCEPTION_BITMAP, eb); |
581 | } | |
582 | ||
33ed6329 AK |
583 | static void reload_tss(void) |
584 | { | |
33ed6329 AK |
585 | /* |
586 | * VT restores TR but not its size. Useless. | |
587 | */ | |
588 | struct descriptor_table gdt; | |
a5f61300 | 589 | struct desc_struct *descs; |
33ed6329 | 590 | |
d6e88aec | 591 | kvm_get_gdt(&gdt); |
33ed6329 AK |
592 | descs = (void *)gdt.base; |
593 | descs[GDT_ENTRY_TSS].type = 9; /* available TSS */ | |
594 | load_TR_desc(); | |
33ed6329 AK |
595 | } |
596 | ||
8b9cf98c | 597 | static void load_transition_efer(struct vcpu_vmx *vmx) |
2cc51560 | 598 | { |
a2fa3e9f | 599 | int efer_offset = vmx->msr_offset_efer; |
3a34a881 RK |
600 | u64 host_efer; |
601 | u64 guest_efer; | |
51c6cf66 AK |
602 | u64 ignore_bits; |
603 | ||
604 | if (efer_offset < 0) | |
605 | return; | |
3a34a881 RK |
606 | host_efer = vmx->host_msrs[efer_offset].data; |
607 | guest_efer = vmx->guest_msrs[efer_offset].data; | |
608 | ||
51c6cf66 AK |
609 | /* |
610 | * NX is emulated; LMA and LME handled by hardware; SCE meaninless | |
611 | * outside long mode | |
612 | */ | |
613 | ignore_bits = EFER_NX | EFER_SCE; | |
614 | #ifdef CONFIG_X86_64 | |
615 | ignore_bits |= EFER_LMA | EFER_LME; | |
616 | /* SCE is meaningful only in long mode on Intel */ | |
617 | if (guest_efer & EFER_LMA) | |
618 | ignore_bits &= ~(u64)EFER_SCE; | |
619 | #endif | |
620 | if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits)) | |
621 | return; | |
2cc51560 | 622 | |
51c6cf66 AK |
623 | vmx->host_state.guest_efer_loaded = 1; |
624 | guest_efer &= ~ignore_bits; | |
625 | guest_efer |= host_efer & ignore_bits; | |
626 | wrmsrl(MSR_EFER, guest_efer); | |
8b9cf98c | 627 | vmx->vcpu.stat.efer_reload++; |
2cc51560 ED |
628 | } |
629 | ||
51c6cf66 AK |
630 | static void reload_host_efer(struct vcpu_vmx *vmx) |
631 | { | |
632 | if (vmx->host_state.guest_efer_loaded) { | |
633 | vmx->host_state.guest_efer_loaded = 0; | |
634 | load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1); | |
635 | } | |
636 | } | |
637 | ||
04d2cc77 | 638 | static void vmx_save_host_state(struct kvm_vcpu *vcpu) |
33ed6329 | 639 | { |
04d2cc77 AK |
640 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
641 | ||
a2fa3e9f | 642 | if (vmx->host_state.loaded) |
33ed6329 AK |
643 | return; |
644 | ||
a2fa3e9f | 645 | vmx->host_state.loaded = 1; |
33ed6329 AK |
646 | /* |
647 | * Set host fs and gs selectors. Unfortunately, 22.2.3 does not | |
648 | * allow segment selectors with cpl > 0 or ti == 1. | |
649 | */ | |
d6e88aec | 650 | vmx->host_state.ldt_sel = kvm_read_ldt(); |
152d3f2f | 651 | vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel; |
d6e88aec | 652 | vmx->host_state.fs_sel = kvm_read_fs(); |
152d3f2f | 653 | if (!(vmx->host_state.fs_sel & 7)) { |
a2fa3e9f | 654 | vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel); |
152d3f2f LV |
655 | vmx->host_state.fs_reload_needed = 0; |
656 | } else { | |
33ed6329 | 657 | vmcs_write16(HOST_FS_SELECTOR, 0); |
152d3f2f | 658 | vmx->host_state.fs_reload_needed = 1; |
33ed6329 | 659 | } |
d6e88aec | 660 | vmx->host_state.gs_sel = kvm_read_gs(); |
a2fa3e9f GH |
661 | if (!(vmx->host_state.gs_sel & 7)) |
662 | vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel); | |
33ed6329 AK |
663 | else { |
664 | vmcs_write16(HOST_GS_SELECTOR, 0); | |
152d3f2f | 665 | vmx->host_state.gs_ldt_reload_needed = 1; |
33ed6329 AK |
666 | } |
667 | ||
668 | #ifdef CONFIG_X86_64 | |
669 | vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE)); | |
670 | vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE)); | |
671 | #else | |
a2fa3e9f GH |
672 | vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel)); |
673 | vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel)); | |
33ed6329 | 674 | #endif |
707c0874 AK |
675 | |
676 | #ifdef CONFIG_X86_64 | |
d77c26fc | 677 | if (is_long_mode(&vmx->vcpu)) |
a2fa3e9f GH |
678 | save_msrs(vmx->host_msrs + |
679 | vmx->msr_offset_kernel_gs_base, 1); | |
d77c26fc | 680 | |
707c0874 | 681 | #endif |
a2fa3e9f | 682 | load_msrs(vmx->guest_msrs, vmx->save_nmsrs); |
51c6cf66 | 683 | load_transition_efer(vmx); |
33ed6329 AK |
684 | } |
685 | ||
a9b21b62 | 686 | static void __vmx_load_host_state(struct vcpu_vmx *vmx) |
33ed6329 | 687 | { |
15ad7146 | 688 | unsigned long flags; |
33ed6329 | 689 | |
a2fa3e9f | 690 | if (!vmx->host_state.loaded) |
33ed6329 AK |
691 | return; |
692 | ||
e1beb1d3 | 693 | ++vmx->vcpu.stat.host_state_reload; |
a2fa3e9f | 694 | vmx->host_state.loaded = 0; |
152d3f2f | 695 | if (vmx->host_state.fs_reload_needed) |
d6e88aec | 696 | kvm_load_fs(vmx->host_state.fs_sel); |
152d3f2f | 697 | if (vmx->host_state.gs_ldt_reload_needed) { |
d6e88aec | 698 | kvm_load_ldt(vmx->host_state.ldt_sel); |
33ed6329 AK |
699 | /* |
700 | * If we have to reload gs, we must take care to | |
701 | * preserve our gs base. | |
702 | */ | |
15ad7146 | 703 | local_irq_save(flags); |
d6e88aec | 704 | kvm_load_gs(vmx->host_state.gs_sel); |
33ed6329 AK |
705 | #ifdef CONFIG_X86_64 |
706 | wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE)); | |
707 | #endif | |
15ad7146 | 708 | local_irq_restore(flags); |
33ed6329 | 709 | } |
152d3f2f | 710 | reload_tss(); |
a2fa3e9f GH |
711 | save_msrs(vmx->guest_msrs, vmx->save_nmsrs); |
712 | load_msrs(vmx->host_msrs, vmx->save_nmsrs); | |
51c6cf66 | 713 | reload_host_efer(vmx); |
33ed6329 AK |
714 | } |
715 | ||
a9b21b62 AK |
716 | static void vmx_load_host_state(struct vcpu_vmx *vmx) |
717 | { | |
718 | preempt_disable(); | |
719 | __vmx_load_host_state(vmx); | |
720 | preempt_enable(); | |
721 | } | |
722 | ||
6aa8b732 AK |
723 | /* |
724 | * Switches to specified vcpu, until a matching vcpu_put(), but assumes | |
725 | * vcpu mutex is already taken. | |
726 | */ | |
15ad7146 | 727 | static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
6aa8b732 | 728 | { |
a2fa3e9f GH |
729 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
730 | u64 phys_addr = __pa(vmx->vmcs); | |
019960ae | 731 | u64 tsc_this, delta, new_offset; |
6aa8b732 | 732 | |
a3d7f85f | 733 | if (vcpu->cpu != cpu) { |
8b9cf98c | 734 | vcpu_clear(vmx); |
2f599714 | 735 | kvm_migrate_timers(vcpu); |
eb5109e3 | 736 | set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests); |
543e4243 AK |
737 | local_irq_disable(); |
738 | list_add(&vmx->local_vcpus_link, | |
739 | &per_cpu(vcpus_on_cpu, cpu)); | |
740 | local_irq_enable(); | |
a3d7f85f | 741 | } |
6aa8b732 | 742 | |
a2fa3e9f | 743 | if (per_cpu(current_vmcs, cpu) != vmx->vmcs) { |
6aa8b732 AK |
744 | u8 error; |
745 | ||
a2fa3e9f | 746 | per_cpu(current_vmcs, cpu) = vmx->vmcs; |
4ecac3fd | 747 | asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0" |
6aa8b732 AK |
748 | : "=g"(error) : "a"(&phys_addr), "m"(phys_addr) |
749 | : "cc"); | |
750 | if (error) | |
751 | printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n", | |
a2fa3e9f | 752 | vmx->vmcs, phys_addr); |
6aa8b732 AK |
753 | } |
754 | ||
755 | if (vcpu->cpu != cpu) { | |
756 | struct descriptor_table dt; | |
757 | unsigned long sysenter_esp; | |
758 | ||
759 | vcpu->cpu = cpu; | |
760 | /* | |
761 | * Linux uses per-cpu TSS and GDT, so set these when switching | |
762 | * processors. | |
763 | */ | |
d6e88aec AK |
764 | vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */ |
765 | kvm_get_gdt(&dt); | |
6aa8b732 AK |
766 | vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */ |
767 | ||
768 | rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp); | |
769 | vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */ | |
7700270e AK |
770 | |
771 | /* | |
772 | * Make sure the time stamp counter is monotonous. | |
773 | */ | |
774 | rdtscll(tsc_this); | |
019960ae AK |
775 | if (tsc_this < vcpu->arch.host_tsc) { |
776 | delta = vcpu->arch.host_tsc - tsc_this; | |
777 | new_offset = vmcs_read64(TSC_OFFSET) + delta; | |
778 | vmcs_write64(TSC_OFFSET, new_offset); | |
779 | } | |
6aa8b732 | 780 | } |
6aa8b732 AK |
781 | } |
782 | ||
783 | static void vmx_vcpu_put(struct kvm_vcpu *vcpu) | |
784 | { | |
a9b21b62 | 785 | __vmx_load_host_state(to_vmx(vcpu)); |
6aa8b732 AK |
786 | } |
787 | ||
5fd86fcf AK |
788 | static void vmx_fpu_activate(struct kvm_vcpu *vcpu) |
789 | { | |
790 | if (vcpu->fpu_active) | |
791 | return; | |
792 | vcpu->fpu_active = 1; | |
707d92fa | 793 | vmcs_clear_bits(GUEST_CR0, X86_CR0_TS); |
ad312c7c | 794 | if (vcpu->arch.cr0 & X86_CR0_TS) |
707d92fa | 795 | vmcs_set_bits(GUEST_CR0, X86_CR0_TS); |
5fd86fcf AK |
796 | update_exception_bitmap(vcpu); |
797 | } | |
798 | ||
799 | static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu) | |
800 | { | |
801 | if (!vcpu->fpu_active) | |
802 | return; | |
803 | vcpu->fpu_active = 0; | |
707d92fa | 804 | vmcs_set_bits(GUEST_CR0, X86_CR0_TS); |
5fd86fcf AK |
805 | update_exception_bitmap(vcpu); |
806 | } | |
807 | ||
6aa8b732 AK |
808 | static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu) |
809 | { | |
345dcaa8 AK |
810 | unsigned long rflags; |
811 | ||
812 | rflags = vmcs_readl(GUEST_RFLAGS); | |
813 | if (to_vmx(vcpu)->rmode.vm86_active) | |
814 | rflags &= ~(unsigned long)(X86_EFLAGS_IOPL | X86_EFLAGS_VM); | |
815 | return rflags; | |
6aa8b732 AK |
816 | } |
817 | ||
818 | static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
819 | { | |
7ffd92c5 | 820 | if (to_vmx(vcpu)->rmode.vm86_active) |
053de044 | 821 | rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM; |
6aa8b732 AK |
822 | vmcs_writel(GUEST_RFLAGS, rflags); |
823 | } | |
824 | ||
2809f5d2 GC |
825 | static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask) |
826 | { | |
827 | u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO); | |
828 | int ret = 0; | |
829 | ||
830 | if (interruptibility & GUEST_INTR_STATE_STI) | |
831 | ret |= X86_SHADOW_INT_STI; | |
832 | if (interruptibility & GUEST_INTR_STATE_MOV_SS) | |
833 | ret |= X86_SHADOW_INT_MOV_SS; | |
834 | ||
835 | return ret & mask; | |
836 | } | |
837 | ||
838 | static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask) | |
839 | { | |
840 | u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO); | |
841 | u32 interruptibility = interruptibility_old; | |
842 | ||
843 | interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS); | |
844 | ||
845 | if (mask & X86_SHADOW_INT_MOV_SS) | |
846 | interruptibility |= GUEST_INTR_STATE_MOV_SS; | |
847 | if (mask & X86_SHADOW_INT_STI) | |
848 | interruptibility |= GUEST_INTR_STATE_STI; | |
849 | ||
850 | if ((interruptibility != interruptibility_old)) | |
851 | vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility); | |
852 | } | |
853 | ||
6aa8b732 AK |
854 | static void skip_emulated_instruction(struct kvm_vcpu *vcpu) |
855 | { | |
856 | unsigned long rip; | |
6aa8b732 | 857 | |
5fdbf976 | 858 | rip = kvm_rip_read(vcpu); |
6aa8b732 | 859 | rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN); |
5fdbf976 | 860 | kvm_rip_write(vcpu, rip); |
6aa8b732 | 861 | |
2809f5d2 GC |
862 | /* skipping an emulated instruction also counts */ |
863 | vmx_set_interrupt_shadow(vcpu, 0); | |
6aa8b732 AK |
864 | } |
865 | ||
298101da AK |
866 | static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr, |
867 | bool has_error_code, u32 error_code) | |
868 | { | |
77ab6db0 | 869 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
8ab2d2e2 | 870 | u32 intr_info = nr | INTR_INFO_VALID_MASK; |
77ab6db0 | 871 | |
8ab2d2e2 | 872 | if (has_error_code) { |
77ab6db0 | 873 | vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code); |
8ab2d2e2 JK |
874 | intr_info |= INTR_INFO_DELIVER_CODE_MASK; |
875 | } | |
77ab6db0 | 876 | |
7ffd92c5 | 877 | if (vmx->rmode.vm86_active) { |
77ab6db0 JK |
878 | vmx->rmode.irq.pending = true; |
879 | vmx->rmode.irq.vector = nr; | |
880 | vmx->rmode.irq.rip = kvm_rip_read(vcpu); | |
ae0bb3e0 GN |
881 | if (kvm_exception_is_soft(nr)) |
882 | vmx->rmode.irq.rip += | |
883 | vmx->vcpu.arch.event_exit_inst_len; | |
8ab2d2e2 JK |
884 | intr_info |= INTR_TYPE_SOFT_INTR; |
885 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info); | |
77ab6db0 JK |
886 | vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1); |
887 | kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1); | |
888 | return; | |
889 | } | |
890 | ||
66fd3f7f GN |
891 | if (kvm_exception_is_soft(nr)) { |
892 | vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, | |
893 | vmx->vcpu.arch.event_exit_inst_len); | |
8ab2d2e2 JK |
894 | intr_info |= INTR_TYPE_SOFT_EXCEPTION; |
895 | } else | |
896 | intr_info |= INTR_TYPE_HARD_EXCEPTION; | |
897 | ||
898 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info); | |
298101da AK |
899 | } |
900 | ||
a75beee6 ED |
901 | /* |
902 | * Swap MSR entry in host/guest MSR entry array. | |
903 | */ | |
54e11fa1 | 904 | #ifdef CONFIG_X86_64 |
8b9cf98c | 905 | static void move_msr_up(struct vcpu_vmx *vmx, int from, int to) |
a75beee6 | 906 | { |
a2fa3e9f GH |
907 | struct kvm_msr_entry tmp; |
908 | ||
909 | tmp = vmx->guest_msrs[to]; | |
910 | vmx->guest_msrs[to] = vmx->guest_msrs[from]; | |
911 | vmx->guest_msrs[from] = tmp; | |
912 | tmp = vmx->host_msrs[to]; | |
913 | vmx->host_msrs[to] = vmx->host_msrs[from]; | |
914 | vmx->host_msrs[from] = tmp; | |
a75beee6 | 915 | } |
54e11fa1 | 916 | #endif |
a75beee6 | 917 | |
e38aea3e AK |
918 | /* |
919 | * Set up the vmcs to automatically save and restore system | |
920 | * msrs. Don't touch the 64-bit msrs if the guest is in legacy | |
921 | * mode, as fiddling with msrs is very expensive. | |
922 | */ | |
8b9cf98c | 923 | static void setup_msrs(struct vcpu_vmx *vmx) |
e38aea3e | 924 | { |
2cc51560 | 925 | int save_nmsrs; |
5897297b | 926 | unsigned long *msr_bitmap; |
e38aea3e | 927 | |
33f9c505 | 928 | vmx_load_host_state(vmx); |
a75beee6 ED |
929 | save_nmsrs = 0; |
930 | #ifdef CONFIG_X86_64 | |
8b9cf98c | 931 | if (is_long_mode(&vmx->vcpu)) { |
2cc51560 ED |
932 | int index; |
933 | ||
8b9cf98c | 934 | index = __find_msr_index(vmx, MSR_SYSCALL_MASK); |
a75beee6 | 935 | if (index >= 0) |
8b9cf98c RR |
936 | move_msr_up(vmx, index, save_nmsrs++); |
937 | index = __find_msr_index(vmx, MSR_LSTAR); | |
a75beee6 | 938 | if (index >= 0) |
8b9cf98c RR |
939 | move_msr_up(vmx, index, save_nmsrs++); |
940 | index = __find_msr_index(vmx, MSR_CSTAR); | |
a75beee6 | 941 | if (index >= 0) |
8b9cf98c RR |
942 | move_msr_up(vmx, index, save_nmsrs++); |
943 | index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE); | |
a75beee6 | 944 | if (index >= 0) |
8b9cf98c | 945 | move_msr_up(vmx, index, save_nmsrs++); |
a75beee6 ED |
946 | /* |
947 | * MSR_K6_STAR is only needed on long mode guests, and only | |
948 | * if efer.sce is enabled. | |
949 | */ | |
8b9cf98c | 950 | index = __find_msr_index(vmx, MSR_K6_STAR); |
ad312c7c | 951 | if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE)) |
8b9cf98c | 952 | move_msr_up(vmx, index, save_nmsrs++); |
a75beee6 ED |
953 | } |
954 | #endif | |
a2fa3e9f | 955 | vmx->save_nmsrs = save_nmsrs; |
e38aea3e | 956 | |
4d56c8a7 | 957 | #ifdef CONFIG_X86_64 |
a2fa3e9f | 958 | vmx->msr_offset_kernel_gs_base = |
8b9cf98c | 959 | __find_msr_index(vmx, MSR_KERNEL_GS_BASE); |
4d56c8a7 | 960 | #endif |
8b9cf98c | 961 | vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER); |
5897297b AK |
962 | |
963 | if (cpu_has_vmx_msr_bitmap()) { | |
964 | if (is_long_mode(&vmx->vcpu)) | |
965 | msr_bitmap = vmx_msr_bitmap_longmode; | |
966 | else | |
967 | msr_bitmap = vmx_msr_bitmap_legacy; | |
968 | ||
969 | vmcs_write64(MSR_BITMAP, __pa(msr_bitmap)); | |
970 | } | |
e38aea3e AK |
971 | } |
972 | ||
6aa8b732 AK |
973 | /* |
974 | * reads and returns guest's timestamp counter "register" | |
975 | * guest_tsc = host_tsc + tsc_offset -- 21.3 | |
976 | */ | |
977 | static u64 guest_read_tsc(void) | |
978 | { | |
979 | u64 host_tsc, tsc_offset; | |
980 | ||
981 | rdtscll(host_tsc); | |
982 | tsc_offset = vmcs_read64(TSC_OFFSET); | |
983 | return host_tsc + tsc_offset; | |
984 | } | |
985 | ||
986 | /* | |
987 | * writes 'guest_tsc' into guest's timestamp counter "register" | |
988 | * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc | |
989 | */ | |
53f658b3 | 990 | static void guest_write_tsc(u64 guest_tsc, u64 host_tsc) |
6aa8b732 | 991 | { |
6aa8b732 AK |
992 | vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc); |
993 | } | |
994 | ||
6aa8b732 AK |
995 | /* |
996 | * Reads an msr value (of 'msr_index') into 'pdata'. | |
997 | * Returns 0 on success, non-0 otherwise. | |
998 | * Assumes vcpu_load() was already called. | |
999 | */ | |
1000 | static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata) | |
1001 | { | |
1002 | u64 data; | |
a2fa3e9f | 1003 | struct kvm_msr_entry *msr; |
6aa8b732 AK |
1004 | |
1005 | if (!pdata) { | |
1006 | printk(KERN_ERR "BUG: get_msr called with NULL pdata\n"); | |
1007 | return -EINVAL; | |
1008 | } | |
1009 | ||
1010 | switch (msr_index) { | |
05b3e0c2 | 1011 | #ifdef CONFIG_X86_64 |
6aa8b732 AK |
1012 | case MSR_FS_BASE: |
1013 | data = vmcs_readl(GUEST_FS_BASE); | |
1014 | break; | |
1015 | case MSR_GS_BASE: | |
1016 | data = vmcs_readl(GUEST_GS_BASE); | |
1017 | break; | |
1018 | case MSR_EFER: | |
3bab1f5d | 1019 | return kvm_get_msr_common(vcpu, msr_index, pdata); |
6aa8b732 | 1020 | #endif |
af24a4e4 | 1021 | case MSR_IA32_TSC: |
6aa8b732 AK |
1022 | data = guest_read_tsc(); |
1023 | break; | |
1024 | case MSR_IA32_SYSENTER_CS: | |
1025 | data = vmcs_read32(GUEST_SYSENTER_CS); | |
1026 | break; | |
1027 | case MSR_IA32_SYSENTER_EIP: | |
f5b42c33 | 1028 | data = vmcs_readl(GUEST_SYSENTER_EIP); |
6aa8b732 AK |
1029 | break; |
1030 | case MSR_IA32_SYSENTER_ESP: | |
f5b42c33 | 1031 | data = vmcs_readl(GUEST_SYSENTER_ESP); |
6aa8b732 | 1032 | break; |
6aa8b732 | 1033 | default: |
8b9cf98c | 1034 | msr = find_msr_entry(to_vmx(vcpu), msr_index); |
3bab1f5d | 1035 | if (msr) { |
542423b0 | 1036 | vmx_load_host_state(to_vmx(vcpu)); |
3bab1f5d AK |
1037 | data = msr->data; |
1038 | break; | |
6aa8b732 | 1039 | } |
3bab1f5d | 1040 | return kvm_get_msr_common(vcpu, msr_index, pdata); |
6aa8b732 AK |
1041 | } |
1042 | ||
1043 | *pdata = data; | |
1044 | return 0; | |
1045 | } | |
1046 | ||
1047 | /* | |
1048 | * Writes msr value into into the appropriate "register". | |
1049 | * Returns 0 on success, non-0 otherwise. | |
1050 | * Assumes vcpu_load() was already called. | |
1051 | */ | |
1052 | static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data) | |
1053 | { | |
a2fa3e9f GH |
1054 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
1055 | struct kvm_msr_entry *msr; | |
53f658b3 | 1056 | u64 host_tsc; |
2cc51560 ED |
1057 | int ret = 0; |
1058 | ||
6aa8b732 | 1059 | switch (msr_index) { |
3bab1f5d | 1060 | case MSR_EFER: |
a9b21b62 | 1061 | vmx_load_host_state(vmx); |
2cc51560 | 1062 | ret = kvm_set_msr_common(vcpu, msr_index, data); |
2cc51560 | 1063 | break; |
16175a79 | 1064 | #ifdef CONFIG_X86_64 |
6aa8b732 AK |
1065 | case MSR_FS_BASE: |
1066 | vmcs_writel(GUEST_FS_BASE, data); | |
1067 | break; | |
1068 | case MSR_GS_BASE: | |
1069 | vmcs_writel(GUEST_GS_BASE, data); | |
1070 | break; | |
1071 | #endif | |
1072 | case MSR_IA32_SYSENTER_CS: | |
1073 | vmcs_write32(GUEST_SYSENTER_CS, data); | |
1074 | break; | |
1075 | case MSR_IA32_SYSENTER_EIP: | |
f5b42c33 | 1076 | vmcs_writel(GUEST_SYSENTER_EIP, data); |
6aa8b732 AK |
1077 | break; |
1078 | case MSR_IA32_SYSENTER_ESP: | |
f5b42c33 | 1079 | vmcs_writel(GUEST_SYSENTER_ESP, data); |
6aa8b732 | 1080 | break; |
af24a4e4 | 1081 | case MSR_IA32_TSC: |
53f658b3 MT |
1082 | rdtscll(host_tsc); |
1083 | guest_write_tsc(data, host_tsc); | |
6aa8b732 | 1084 | break; |
468d472f SY |
1085 | case MSR_IA32_CR_PAT: |
1086 | if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) { | |
1087 | vmcs_write64(GUEST_IA32_PAT, data); | |
1088 | vcpu->arch.pat = data; | |
1089 | break; | |
1090 | } | |
1091 | /* Otherwise falls through to kvm_set_msr_common */ | |
6aa8b732 | 1092 | default: |
8b9cf98c | 1093 | msr = find_msr_entry(vmx, msr_index); |
3bab1f5d | 1094 | if (msr) { |
542423b0 | 1095 | vmx_load_host_state(vmx); |
3bab1f5d AK |
1096 | msr->data = data; |
1097 | break; | |
6aa8b732 | 1098 | } |
2cc51560 | 1099 | ret = kvm_set_msr_common(vcpu, msr_index, data); |
6aa8b732 AK |
1100 | } |
1101 | ||
2cc51560 | 1102 | return ret; |
6aa8b732 AK |
1103 | } |
1104 | ||
5fdbf976 | 1105 | static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg) |
6aa8b732 | 1106 | { |
5fdbf976 MT |
1107 | __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail); |
1108 | switch (reg) { | |
1109 | case VCPU_REGS_RSP: | |
1110 | vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP); | |
1111 | break; | |
1112 | case VCPU_REGS_RIP: | |
1113 | vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP); | |
1114 | break; | |
6de4f3ad AK |
1115 | case VCPU_EXREG_PDPTR: |
1116 | if (enable_ept) | |
1117 | ept_save_pdptrs(vcpu); | |
1118 | break; | |
5fdbf976 MT |
1119 | default: |
1120 | break; | |
1121 | } | |
6aa8b732 AK |
1122 | } |
1123 | ||
355be0b9 | 1124 | static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg) |
6aa8b732 | 1125 | { |
ae675ef0 JK |
1126 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) |
1127 | vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]); | |
1128 | else | |
1129 | vmcs_writel(GUEST_DR7, vcpu->arch.dr7); | |
1130 | ||
abd3f2d6 | 1131 | update_exception_bitmap(vcpu); |
6aa8b732 AK |
1132 | } |
1133 | ||
1134 | static __init int cpu_has_kvm_support(void) | |
1135 | { | |
6210e37b | 1136 | return cpu_has_vmx(); |
6aa8b732 AK |
1137 | } |
1138 | ||
1139 | static __init int vmx_disabled_by_bios(void) | |
1140 | { | |
1141 | u64 msr; | |
1142 | ||
1143 | rdmsrl(MSR_IA32_FEATURE_CONTROL, msr); | |
9ea542fa SY |
1144 | return (msr & (FEATURE_CONTROL_LOCKED | |
1145 | FEATURE_CONTROL_VMXON_ENABLED)) | |
1146 | == FEATURE_CONTROL_LOCKED; | |
62b3ffb8 | 1147 | /* locked but not enabled */ |
6aa8b732 AK |
1148 | } |
1149 | ||
10474ae8 | 1150 | static int hardware_enable(void *garbage) |
6aa8b732 AK |
1151 | { |
1152 | int cpu = raw_smp_processor_id(); | |
1153 | u64 phys_addr = __pa(per_cpu(vmxarea, cpu)); | |
1154 | u64 old; | |
1155 | ||
10474ae8 AG |
1156 | if (read_cr4() & X86_CR4_VMXE) |
1157 | return -EBUSY; | |
1158 | ||
543e4243 | 1159 | INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu)); |
6aa8b732 | 1160 | rdmsrl(MSR_IA32_FEATURE_CONTROL, old); |
9ea542fa SY |
1161 | if ((old & (FEATURE_CONTROL_LOCKED | |
1162 | FEATURE_CONTROL_VMXON_ENABLED)) | |
1163 | != (FEATURE_CONTROL_LOCKED | | |
1164 | FEATURE_CONTROL_VMXON_ENABLED)) | |
6aa8b732 | 1165 | /* enable and lock */ |
62b3ffb8 | 1166 | wrmsrl(MSR_IA32_FEATURE_CONTROL, old | |
9ea542fa SY |
1167 | FEATURE_CONTROL_LOCKED | |
1168 | FEATURE_CONTROL_VMXON_ENABLED); | |
66aee91a | 1169 | write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */ |
4ecac3fd AK |
1170 | asm volatile (ASM_VMX_VMXON_RAX |
1171 | : : "a"(&phys_addr), "m"(phys_addr) | |
6aa8b732 | 1172 | : "memory", "cc"); |
10474ae8 AG |
1173 | |
1174 | ept_sync_global(); | |
1175 | ||
1176 | return 0; | |
6aa8b732 AK |
1177 | } |
1178 | ||
543e4243 AK |
1179 | static void vmclear_local_vcpus(void) |
1180 | { | |
1181 | int cpu = raw_smp_processor_id(); | |
1182 | struct vcpu_vmx *vmx, *n; | |
1183 | ||
1184 | list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu), | |
1185 | local_vcpus_link) | |
1186 | __vcpu_clear(vmx); | |
1187 | } | |
1188 | ||
710ff4a8 EH |
1189 | |
1190 | /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot() | |
1191 | * tricks. | |
1192 | */ | |
1193 | static void kvm_cpu_vmxoff(void) | |
6aa8b732 | 1194 | { |
4ecac3fd | 1195 | asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc"); |
e693d71b | 1196 | write_cr4(read_cr4() & ~X86_CR4_VMXE); |
6aa8b732 AK |
1197 | } |
1198 | ||
710ff4a8 EH |
1199 | static void hardware_disable(void *garbage) |
1200 | { | |
1201 | vmclear_local_vcpus(); | |
1202 | kvm_cpu_vmxoff(); | |
1203 | } | |
1204 | ||
1c3d14fe | 1205 | static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt, |
d77c26fc | 1206 | u32 msr, u32 *result) |
1c3d14fe YS |
1207 | { |
1208 | u32 vmx_msr_low, vmx_msr_high; | |
1209 | u32 ctl = ctl_min | ctl_opt; | |
1210 | ||
1211 | rdmsr(msr, vmx_msr_low, vmx_msr_high); | |
1212 | ||
1213 | ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */ | |
1214 | ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */ | |
1215 | ||
1216 | /* Ensure minimum (required) set of control bits are supported. */ | |
1217 | if (ctl_min & ~ctl) | |
002c7f7c | 1218 | return -EIO; |
1c3d14fe YS |
1219 | |
1220 | *result = ctl; | |
1221 | return 0; | |
1222 | } | |
1223 | ||
002c7f7c | 1224 | static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf) |
6aa8b732 AK |
1225 | { |
1226 | u32 vmx_msr_low, vmx_msr_high; | |
d56f546d | 1227 | u32 min, opt, min2, opt2; |
1c3d14fe YS |
1228 | u32 _pin_based_exec_control = 0; |
1229 | u32 _cpu_based_exec_control = 0; | |
f78e0e2e | 1230 | u32 _cpu_based_2nd_exec_control = 0; |
1c3d14fe YS |
1231 | u32 _vmexit_control = 0; |
1232 | u32 _vmentry_control = 0; | |
1233 | ||
1234 | min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING; | |
f08864b4 | 1235 | opt = PIN_BASED_VIRTUAL_NMIS; |
1c3d14fe YS |
1236 | if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS, |
1237 | &_pin_based_exec_control) < 0) | |
002c7f7c | 1238 | return -EIO; |
1c3d14fe YS |
1239 | |
1240 | min = CPU_BASED_HLT_EXITING | | |
1241 | #ifdef CONFIG_X86_64 | |
1242 | CPU_BASED_CR8_LOAD_EXITING | | |
1243 | CPU_BASED_CR8_STORE_EXITING | | |
1244 | #endif | |
d56f546d SY |
1245 | CPU_BASED_CR3_LOAD_EXITING | |
1246 | CPU_BASED_CR3_STORE_EXITING | | |
1c3d14fe YS |
1247 | CPU_BASED_USE_IO_BITMAPS | |
1248 | CPU_BASED_MOV_DR_EXITING | | |
a7052897 MT |
1249 | CPU_BASED_USE_TSC_OFFSETING | |
1250 | CPU_BASED_INVLPG_EXITING; | |
f78e0e2e | 1251 | opt = CPU_BASED_TPR_SHADOW | |
25c5f225 | 1252 | CPU_BASED_USE_MSR_BITMAPS | |
f78e0e2e | 1253 | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS; |
1c3d14fe YS |
1254 | if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS, |
1255 | &_cpu_based_exec_control) < 0) | |
002c7f7c | 1256 | return -EIO; |
6e5d865c YS |
1257 | #ifdef CONFIG_X86_64 |
1258 | if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW)) | |
1259 | _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING & | |
1260 | ~CPU_BASED_CR8_STORE_EXITING; | |
1261 | #endif | |
f78e0e2e | 1262 | if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) { |
d56f546d SY |
1263 | min2 = 0; |
1264 | opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES | | |
2384d2b3 | 1265 | SECONDARY_EXEC_WBINVD_EXITING | |
d56f546d | 1266 | SECONDARY_EXEC_ENABLE_VPID | |
3a624e29 | 1267 | SECONDARY_EXEC_ENABLE_EPT | |
4b8d54f9 ZE |
1268 | SECONDARY_EXEC_UNRESTRICTED_GUEST | |
1269 | SECONDARY_EXEC_PAUSE_LOOP_EXITING; | |
d56f546d SY |
1270 | if (adjust_vmx_controls(min2, opt2, |
1271 | MSR_IA32_VMX_PROCBASED_CTLS2, | |
f78e0e2e SY |
1272 | &_cpu_based_2nd_exec_control) < 0) |
1273 | return -EIO; | |
1274 | } | |
1275 | #ifndef CONFIG_X86_64 | |
1276 | if (!(_cpu_based_2nd_exec_control & | |
1277 | SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) | |
1278 | _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW; | |
1279 | #endif | |
d56f546d | 1280 | if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) { |
a7052897 MT |
1281 | /* CR3 accesses and invlpg don't need to cause VM Exits when EPT |
1282 | enabled */ | |
5fff7d27 GN |
1283 | _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING | |
1284 | CPU_BASED_CR3_STORE_EXITING | | |
1285 | CPU_BASED_INVLPG_EXITING); | |
d56f546d SY |
1286 | rdmsr(MSR_IA32_VMX_EPT_VPID_CAP, |
1287 | vmx_capability.ept, vmx_capability.vpid); | |
1288 | } | |
1c3d14fe YS |
1289 | |
1290 | min = 0; | |
1291 | #ifdef CONFIG_X86_64 | |
1292 | min |= VM_EXIT_HOST_ADDR_SPACE_SIZE; | |
1293 | #endif | |
468d472f | 1294 | opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT; |
1c3d14fe YS |
1295 | if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS, |
1296 | &_vmexit_control) < 0) | |
002c7f7c | 1297 | return -EIO; |
1c3d14fe | 1298 | |
468d472f SY |
1299 | min = 0; |
1300 | opt = VM_ENTRY_LOAD_IA32_PAT; | |
1c3d14fe YS |
1301 | if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS, |
1302 | &_vmentry_control) < 0) | |
002c7f7c | 1303 | return -EIO; |
6aa8b732 | 1304 | |
c68876fd | 1305 | rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high); |
1c3d14fe YS |
1306 | |
1307 | /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */ | |
1308 | if ((vmx_msr_high & 0x1fff) > PAGE_SIZE) | |
002c7f7c | 1309 | return -EIO; |
1c3d14fe YS |
1310 | |
1311 | #ifdef CONFIG_X86_64 | |
1312 | /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */ | |
1313 | if (vmx_msr_high & (1u<<16)) | |
002c7f7c | 1314 | return -EIO; |
1c3d14fe YS |
1315 | #endif |
1316 | ||
1317 | /* Require Write-Back (WB) memory type for VMCS accesses. */ | |
1318 | if (((vmx_msr_high >> 18) & 15) != 6) | |
002c7f7c | 1319 | return -EIO; |
1c3d14fe | 1320 | |
002c7f7c YS |
1321 | vmcs_conf->size = vmx_msr_high & 0x1fff; |
1322 | vmcs_conf->order = get_order(vmcs_config.size); | |
1323 | vmcs_conf->revision_id = vmx_msr_low; | |
1c3d14fe | 1324 | |
002c7f7c YS |
1325 | vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control; |
1326 | vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control; | |
f78e0e2e | 1327 | vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control; |
002c7f7c YS |
1328 | vmcs_conf->vmexit_ctrl = _vmexit_control; |
1329 | vmcs_conf->vmentry_ctrl = _vmentry_control; | |
1c3d14fe YS |
1330 | |
1331 | return 0; | |
c68876fd | 1332 | } |
6aa8b732 AK |
1333 | |
1334 | static struct vmcs *alloc_vmcs_cpu(int cpu) | |
1335 | { | |
1336 | int node = cpu_to_node(cpu); | |
1337 | struct page *pages; | |
1338 | struct vmcs *vmcs; | |
1339 | ||
6484eb3e | 1340 | pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order); |
6aa8b732 AK |
1341 | if (!pages) |
1342 | return NULL; | |
1343 | vmcs = page_address(pages); | |
1c3d14fe YS |
1344 | memset(vmcs, 0, vmcs_config.size); |
1345 | vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */ | |
6aa8b732 AK |
1346 | return vmcs; |
1347 | } | |
1348 | ||
1349 | static struct vmcs *alloc_vmcs(void) | |
1350 | { | |
d3b2c338 | 1351 | return alloc_vmcs_cpu(raw_smp_processor_id()); |
6aa8b732 AK |
1352 | } |
1353 | ||
1354 | static void free_vmcs(struct vmcs *vmcs) | |
1355 | { | |
1c3d14fe | 1356 | free_pages((unsigned long)vmcs, vmcs_config.order); |
6aa8b732 AK |
1357 | } |
1358 | ||
39959588 | 1359 | static void free_kvm_area(void) |
6aa8b732 AK |
1360 | { |
1361 | int cpu; | |
1362 | ||
3230bb47 | 1363 | for_each_possible_cpu(cpu) { |
6aa8b732 | 1364 | free_vmcs(per_cpu(vmxarea, cpu)); |
3230bb47 ZA |
1365 | per_cpu(vmxarea, cpu) = NULL; |
1366 | } | |
6aa8b732 AK |
1367 | } |
1368 | ||
6aa8b732 AK |
1369 | static __init int alloc_kvm_area(void) |
1370 | { | |
1371 | int cpu; | |
1372 | ||
3230bb47 | 1373 | for_each_possible_cpu(cpu) { |
6aa8b732 AK |
1374 | struct vmcs *vmcs; |
1375 | ||
1376 | vmcs = alloc_vmcs_cpu(cpu); | |
1377 | if (!vmcs) { | |
1378 | free_kvm_area(); | |
1379 | return -ENOMEM; | |
1380 | } | |
1381 | ||
1382 | per_cpu(vmxarea, cpu) = vmcs; | |
1383 | } | |
1384 | return 0; | |
1385 | } | |
1386 | ||
1387 | static __init int hardware_setup(void) | |
1388 | { | |
002c7f7c YS |
1389 | if (setup_vmcs_config(&vmcs_config) < 0) |
1390 | return -EIO; | |
50a37eb4 JR |
1391 | |
1392 | if (boot_cpu_has(X86_FEATURE_NX)) | |
1393 | kvm_enable_efer_bits(EFER_NX); | |
1394 | ||
93ba03c2 SY |
1395 | if (!cpu_has_vmx_vpid()) |
1396 | enable_vpid = 0; | |
1397 | ||
3a624e29 | 1398 | if (!cpu_has_vmx_ept()) { |
93ba03c2 | 1399 | enable_ept = 0; |
3a624e29 NK |
1400 | enable_unrestricted_guest = 0; |
1401 | } | |
1402 | ||
1403 | if (!cpu_has_vmx_unrestricted_guest()) | |
1404 | enable_unrestricted_guest = 0; | |
93ba03c2 SY |
1405 | |
1406 | if (!cpu_has_vmx_flexpriority()) | |
1407 | flexpriority_enabled = 0; | |
1408 | ||
95ba8273 GN |
1409 | if (!cpu_has_vmx_tpr_shadow()) |
1410 | kvm_x86_ops->update_cr8_intercept = NULL; | |
1411 | ||
54dee993 MT |
1412 | if (enable_ept && !cpu_has_vmx_ept_2m_page()) |
1413 | kvm_disable_largepages(); | |
1414 | ||
4b8d54f9 ZE |
1415 | if (!cpu_has_vmx_ple()) |
1416 | ple_gap = 0; | |
1417 | ||
6aa8b732 AK |
1418 | return alloc_kvm_area(); |
1419 | } | |
1420 | ||
1421 | static __exit void hardware_unsetup(void) | |
1422 | { | |
1423 | free_kvm_area(); | |
1424 | } | |
1425 | ||
6aa8b732 AK |
1426 | static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save) |
1427 | { | |
1428 | struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg]; | |
1429 | ||
6af11b9e | 1430 | if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) { |
6aa8b732 AK |
1431 | vmcs_write16(sf->selector, save->selector); |
1432 | vmcs_writel(sf->base, save->base); | |
1433 | vmcs_write32(sf->limit, save->limit); | |
1434 | vmcs_write32(sf->ar_bytes, save->ar); | |
1435 | } else { | |
1436 | u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK) | |
1437 | << AR_DPL_SHIFT; | |
1438 | vmcs_write32(sf->ar_bytes, 0x93 | dpl); | |
1439 | } | |
1440 | } | |
1441 | ||
1442 | static void enter_pmode(struct kvm_vcpu *vcpu) | |
1443 | { | |
1444 | unsigned long flags; | |
a89a8fb9 | 1445 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
6aa8b732 | 1446 | |
a89a8fb9 | 1447 | vmx->emulation_required = 1; |
7ffd92c5 | 1448 | vmx->rmode.vm86_active = 0; |
6aa8b732 | 1449 | |
7ffd92c5 AK |
1450 | vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base); |
1451 | vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit); | |
1452 | vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar); | |
6aa8b732 AK |
1453 | |
1454 | flags = vmcs_readl(GUEST_RFLAGS); | |
053de044 | 1455 | flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM); |
7ffd92c5 | 1456 | flags |= (vmx->rmode.save_iopl << IOPL_SHIFT); |
6aa8b732 AK |
1457 | vmcs_writel(GUEST_RFLAGS, flags); |
1458 | ||
66aee91a RR |
1459 | vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) | |
1460 | (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME)); | |
6aa8b732 AK |
1461 | |
1462 | update_exception_bitmap(vcpu); | |
1463 | ||
a89a8fb9 MG |
1464 | if (emulate_invalid_guest_state) |
1465 | return; | |
1466 | ||
7ffd92c5 AK |
1467 | fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es); |
1468 | fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds); | |
1469 | fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs); | |
1470 | fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs); | |
6aa8b732 AK |
1471 | |
1472 | vmcs_write16(GUEST_SS_SELECTOR, 0); | |
1473 | vmcs_write32(GUEST_SS_AR_BYTES, 0x93); | |
1474 | ||
1475 | vmcs_write16(GUEST_CS_SELECTOR, | |
1476 | vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK); | |
1477 | vmcs_write32(GUEST_CS_AR_BYTES, 0x9b); | |
1478 | } | |
1479 | ||
d77c26fc | 1480 | static gva_t rmode_tss_base(struct kvm *kvm) |
6aa8b732 | 1481 | { |
bfc6d222 | 1482 | if (!kvm->arch.tss_addr) { |
cbc94022 IE |
1483 | gfn_t base_gfn = kvm->memslots[0].base_gfn + |
1484 | kvm->memslots[0].npages - 3; | |
1485 | return base_gfn << PAGE_SHIFT; | |
1486 | } | |
bfc6d222 | 1487 | return kvm->arch.tss_addr; |
6aa8b732 AK |
1488 | } |
1489 | ||
1490 | static void fix_rmode_seg(int seg, struct kvm_save_segment *save) | |
1491 | { | |
1492 | struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg]; | |
1493 | ||
1494 | save->selector = vmcs_read16(sf->selector); | |
1495 | save->base = vmcs_readl(sf->base); | |
1496 | save->limit = vmcs_read32(sf->limit); | |
1497 | save->ar = vmcs_read32(sf->ar_bytes); | |
15b00f32 JK |
1498 | vmcs_write16(sf->selector, save->base >> 4); |
1499 | vmcs_write32(sf->base, save->base & 0xfffff); | |
6aa8b732 AK |
1500 | vmcs_write32(sf->limit, 0xffff); |
1501 | vmcs_write32(sf->ar_bytes, 0xf3); | |
1502 | } | |
1503 | ||
1504 | static void enter_rmode(struct kvm_vcpu *vcpu) | |
1505 | { | |
1506 | unsigned long flags; | |
a89a8fb9 | 1507 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
6aa8b732 | 1508 | |
3a624e29 NK |
1509 | if (enable_unrestricted_guest) |
1510 | return; | |
1511 | ||
a89a8fb9 | 1512 | vmx->emulation_required = 1; |
7ffd92c5 | 1513 | vmx->rmode.vm86_active = 1; |
6aa8b732 | 1514 | |
7ffd92c5 | 1515 | vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE); |
6aa8b732 AK |
1516 | vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm)); |
1517 | ||
7ffd92c5 | 1518 | vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT); |
6aa8b732 AK |
1519 | vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1); |
1520 | ||
7ffd92c5 | 1521 | vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES); |
6aa8b732 AK |
1522 | vmcs_write32(GUEST_TR_AR_BYTES, 0x008b); |
1523 | ||
1524 | flags = vmcs_readl(GUEST_RFLAGS); | |
7ffd92c5 | 1525 | vmx->rmode.save_iopl |
ad312c7c | 1526 | = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT; |
6aa8b732 | 1527 | |
053de044 | 1528 | flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM; |
6aa8b732 AK |
1529 | |
1530 | vmcs_writel(GUEST_RFLAGS, flags); | |
66aee91a | 1531 | vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME); |
6aa8b732 AK |
1532 | update_exception_bitmap(vcpu); |
1533 | ||
a89a8fb9 MG |
1534 | if (emulate_invalid_guest_state) |
1535 | goto continue_rmode; | |
1536 | ||
6aa8b732 AK |
1537 | vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4); |
1538 | vmcs_write32(GUEST_SS_LIMIT, 0xffff); | |
1539 | vmcs_write32(GUEST_SS_AR_BYTES, 0xf3); | |
1540 | ||
1541 | vmcs_write32(GUEST_CS_AR_BYTES, 0xf3); | |
abacf8df | 1542 | vmcs_write32(GUEST_CS_LIMIT, 0xffff); |
8cb5b033 AK |
1543 | if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000) |
1544 | vmcs_writel(GUEST_CS_BASE, 0xf0000); | |
6aa8b732 AK |
1545 | vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4); |
1546 | ||
7ffd92c5 AK |
1547 | fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es); |
1548 | fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds); | |
1549 | fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs); | |
1550 | fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs); | |
75880a01 | 1551 | |
a89a8fb9 | 1552 | continue_rmode: |
8668a3c4 | 1553 | kvm_mmu_reset_context(vcpu); |
b7ebfb05 | 1554 | init_rmode(vcpu->kvm); |
6aa8b732 AK |
1555 | } |
1556 | ||
401d10de AS |
1557 | static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer) |
1558 | { | |
1559 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
1560 | struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER); | |
1561 | ||
1562 | vcpu->arch.shadow_efer = efer; | |
1563 | if (!msr) | |
1564 | return; | |
1565 | if (efer & EFER_LMA) { | |
1566 | vmcs_write32(VM_ENTRY_CONTROLS, | |
1567 | vmcs_read32(VM_ENTRY_CONTROLS) | | |
1568 | VM_ENTRY_IA32E_MODE); | |
1569 | msr->data = efer; | |
1570 | } else { | |
1571 | vmcs_write32(VM_ENTRY_CONTROLS, | |
1572 | vmcs_read32(VM_ENTRY_CONTROLS) & | |
1573 | ~VM_ENTRY_IA32E_MODE); | |
1574 | ||
1575 | msr->data = efer & ~EFER_LME; | |
1576 | } | |
1577 | setup_msrs(vmx); | |
1578 | } | |
1579 | ||
05b3e0c2 | 1580 | #ifdef CONFIG_X86_64 |
6aa8b732 AK |
1581 | |
1582 | static void enter_lmode(struct kvm_vcpu *vcpu) | |
1583 | { | |
1584 | u32 guest_tr_ar; | |
1585 | ||
1586 | guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES); | |
1587 | if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) { | |
1588 | printk(KERN_DEBUG "%s: tss fixup for long mode. \n", | |
b8688d51 | 1589 | __func__); |
6aa8b732 AK |
1590 | vmcs_write32(GUEST_TR_AR_BYTES, |
1591 | (guest_tr_ar & ~AR_TYPE_MASK) | |
1592 | | AR_TYPE_BUSY_64_TSS); | |
1593 | } | |
ad312c7c | 1594 | vcpu->arch.shadow_efer |= EFER_LMA; |
401d10de | 1595 | vmx_set_efer(vcpu, vcpu->arch.shadow_efer); |
6aa8b732 AK |
1596 | } |
1597 | ||
1598 | static void exit_lmode(struct kvm_vcpu *vcpu) | |
1599 | { | |
ad312c7c | 1600 | vcpu->arch.shadow_efer &= ~EFER_LMA; |
6aa8b732 AK |
1601 | |
1602 | vmcs_write32(VM_ENTRY_CONTROLS, | |
1603 | vmcs_read32(VM_ENTRY_CONTROLS) | |
1e4e6e00 | 1604 | & ~VM_ENTRY_IA32E_MODE); |
6aa8b732 AK |
1605 | } |
1606 | ||
1607 | #endif | |
1608 | ||
2384d2b3 SY |
1609 | static void vmx_flush_tlb(struct kvm_vcpu *vcpu) |
1610 | { | |
1611 | vpid_sync_vcpu_all(to_vmx(vcpu)); | |
089d034e | 1612 | if (enable_ept) |
4e1096d2 | 1613 | ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa)); |
2384d2b3 SY |
1614 | } |
1615 | ||
25c4c276 | 1616 | static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu) |
399badf3 | 1617 | { |
ad312c7c ZX |
1618 | vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK; |
1619 | vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK; | |
399badf3 AK |
1620 | } |
1621 | ||
1439442c SY |
1622 | static void ept_load_pdptrs(struct kvm_vcpu *vcpu) |
1623 | { | |
6de4f3ad AK |
1624 | if (!test_bit(VCPU_EXREG_PDPTR, |
1625 | (unsigned long *)&vcpu->arch.regs_dirty)) | |
1626 | return; | |
1627 | ||
1439442c | 1628 | if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) { |
1439442c SY |
1629 | vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]); |
1630 | vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]); | |
1631 | vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]); | |
1632 | vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]); | |
1633 | } | |
1634 | } | |
1635 | ||
8f5d549f AK |
1636 | static void ept_save_pdptrs(struct kvm_vcpu *vcpu) |
1637 | { | |
1638 | if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) { | |
1639 | vcpu->arch.pdptrs[0] = vmcs_read64(GUEST_PDPTR0); | |
1640 | vcpu->arch.pdptrs[1] = vmcs_read64(GUEST_PDPTR1); | |
1641 | vcpu->arch.pdptrs[2] = vmcs_read64(GUEST_PDPTR2); | |
1642 | vcpu->arch.pdptrs[3] = vmcs_read64(GUEST_PDPTR3); | |
1643 | } | |
6de4f3ad AK |
1644 | |
1645 | __set_bit(VCPU_EXREG_PDPTR, | |
1646 | (unsigned long *)&vcpu->arch.regs_avail); | |
1647 | __set_bit(VCPU_EXREG_PDPTR, | |
1648 | (unsigned long *)&vcpu->arch.regs_dirty); | |
8f5d549f AK |
1649 | } |
1650 | ||
1439442c SY |
1651 | static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4); |
1652 | ||
1653 | static void ept_update_paging_mode_cr0(unsigned long *hw_cr0, | |
1654 | unsigned long cr0, | |
1655 | struct kvm_vcpu *vcpu) | |
1656 | { | |
1657 | if (!(cr0 & X86_CR0_PG)) { | |
1658 | /* From paging/starting to nonpaging */ | |
1659 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, | |
65267ea1 | 1660 | vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) | |
1439442c SY |
1661 | (CPU_BASED_CR3_LOAD_EXITING | |
1662 | CPU_BASED_CR3_STORE_EXITING)); | |
1663 | vcpu->arch.cr0 = cr0; | |
1664 | vmx_set_cr4(vcpu, vcpu->arch.cr4); | |
1439442c SY |
1665 | } else if (!is_paging(vcpu)) { |
1666 | /* From nonpaging to paging */ | |
1667 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, | |
65267ea1 | 1668 | vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) & |
1439442c SY |
1669 | ~(CPU_BASED_CR3_LOAD_EXITING | |
1670 | CPU_BASED_CR3_STORE_EXITING)); | |
1671 | vcpu->arch.cr0 = cr0; | |
1672 | vmx_set_cr4(vcpu, vcpu->arch.cr4); | |
1439442c | 1673 | } |
95eb84a7 SY |
1674 | |
1675 | if (!(cr0 & X86_CR0_WP)) | |
1676 | *hw_cr0 &= ~X86_CR0_WP; | |
1439442c SY |
1677 | } |
1678 | ||
1679 | static void ept_update_paging_mode_cr4(unsigned long *hw_cr4, | |
1680 | struct kvm_vcpu *vcpu) | |
1681 | { | |
1682 | if (!is_paging(vcpu)) { | |
1683 | *hw_cr4 &= ~X86_CR4_PAE; | |
1684 | *hw_cr4 |= X86_CR4_PSE; | |
1685 | } else if (!(vcpu->arch.cr4 & X86_CR4_PAE)) | |
1686 | *hw_cr4 &= ~X86_CR4_PAE; | |
1687 | } | |
1688 | ||
6aa8b732 AK |
1689 | static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
1690 | { | |
7ffd92c5 | 1691 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
3a624e29 NK |
1692 | unsigned long hw_cr0; |
1693 | ||
1694 | if (enable_unrestricted_guest) | |
1695 | hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST) | |
1696 | | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST; | |
1697 | else | |
1698 | hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON; | |
1439442c | 1699 | |
5fd86fcf AK |
1700 | vmx_fpu_deactivate(vcpu); |
1701 | ||
7ffd92c5 | 1702 | if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE)) |
6aa8b732 AK |
1703 | enter_pmode(vcpu); |
1704 | ||
7ffd92c5 | 1705 | if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE)) |
6aa8b732 AK |
1706 | enter_rmode(vcpu); |
1707 | ||
05b3e0c2 | 1708 | #ifdef CONFIG_X86_64 |
ad312c7c | 1709 | if (vcpu->arch.shadow_efer & EFER_LME) { |
707d92fa | 1710 | if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) |
6aa8b732 | 1711 | enter_lmode(vcpu); |
707d92fa | 1712 | if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) |
6aa8b732 AK |
1713 | exit_lmode(vcpu); |
1714 | } | |
1715 | #endif | |
1716 | ||
089d034e | 1717 | if (enable_ept) |
1439442c SY |
1718 | ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu); |
1719 | ||
6aa8b732 | 1720 | vmcs_writel(CR0_READ_SHADOW, cr0); |
1439442c | 1721 | vmcs_writel(GUEST_CR0, hw_cr0); |
ad312c7c | 1722 | vcpu->arch.cr0 = cr0; |
5fd86fcf | 1723 | |
707d92fa | 1724 | if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE)) |
5fd86fcf | 1725 | vmx_fpu_activate(vcpu); |
6aa8b732 AK |
1726 | } |
1727 | ||
1439442c SY |
1728 | static u64 construct_eptp(unsigned long root_hpa) |
1729 | { | |
1730 | u64 eptp; | |
1731 | ||
1732 | /* TODO write the value reading from MSR */ | |
1733 | eptp = VMX_EPT_DEFAULT_MT | | |
1734 | VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT; | |
1735 | eptp |= (root_hpa & PAGE_MASK); | |
1736 | ||
1737 | return eptp; | |
1738 | } | |
1739 | ||
6aa8b732 AK |
1740 | static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3) |
1741 | { | |
1439442c SY |
1742 | unsigned long guest_cr3; |
1743 | u64 eptp; | |
1744 | ||
1745 | guest_cr3 = cr3; | |
089d034e | 1746 | if (enable_ept) { |
1439442c SY |
1747 | eptp = construct_eptp(cr3); |
1748 | vmcs_write64(EPT_POINTER, eptp); | |
1439442c | 1749 | guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 : |
b927a3ce | 1750 | vcpu->kvm->arch.ept_identity_map_addr; |
1439442c SY |
1751 | } |
1752 | ||
2384d2b3 | 1753 | vmx_flush_tlb(vcpu); |
1439442c | 1754 | vmcs_writel(GUEST_CR3, guest_cr3); |
ad312c7c | 1755 | if (vcpu->arch.cr0 & X86_CR0_PE) |
5fd86fcf | 1756 | vmx_fpu_deactivate(vcpu); |
6aa8b732 AK |
1757 | } |
1758 | ||
1759 | static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) | |
1760 | { | |
7ffd92c5 | 1761 | unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ? |
1439442c SY |
1762 | KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON); |
1763 | ||
ad312c7c | 1764 | vcpu->arch.cr4 = cr4; |
089d034e | 1765 | if (enable_ept) |
1439442c SY |
1766 | ept_update_paging_mode_cr4(&hw_cr4, vcpu); |
1767 | ||
1768 | vmcs_writel(CR4_READ_SHADOW, cr4); | |
1769 | vmcs_writel(GUEST_CR4, hw_cr4); | |
6aa8b732 AK |
1770 | } |
1771 | ||
6aa8b732 AK |
1772 | static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg) |
1773 | { | |
1774 | struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg]; | |
1775 | ||
1776 | return vmcs_readl(sf->base); | |
1777 | } | |
1778 | ||
1779 | static void vmx_get_segment(struct kvm_vcpu *vcpu, | |
1780 | struct kvm_segment *var, int seg) | |
1781 | { | |
1782 | struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg]; | |
1783 | u32 ar; | |
1784 | ||
1785 | var->base = vmcs_readl(sf->base); | |
1786 | var->limit = vmcs_read32(sf->limit); | |
1787 | var->selector = vmcs_read16(sf->selector); | |
1788 | ar = vmcs_read32(sf->ar_bytes); | |
9fd4a3b7 | 1789 | if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state) |
6aa8b732 AK |
1790 | ar = 0; |
1791 | var->type = ar & 15; | |
1792 | var->s = (ar >> 4) & 1; | |
1793 | var->dpl = (ar >> 5) & 3; | |
1794 | var->present = (ar >> 7) & 1; | |
1795 | var->avl = (ar >> 12) & 1; | |
1796 | var->l = (ar >> 13) & 1; | |
1797 | var->db = (ar >> 14) & 1; | |
1798 | var->g = (ar >> 15) & 1; | |
1799 | var->unusable = (ar >> 16) & 1; | |
1800 | } | |
1801 | ||
2e4d2653 IE |
1802 | static int vmx_get_cpl(struct kvm_vcpu *vcpu) |
1803 | { | |
2e4d2653 IE |
1804 | if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */ |
1805 | return 0; | |
1806 | ||
1807 | if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */ | |
1808 | return 3; | |
1809 | ||
eab4b8aa | 1810 | return vmcs_read16(GUEST_CS_SELECTOR) & 3; |
2e4d2653 IE |
1811 | } |
1812 | ||
653e3108 | 1813 | static u32 vmx_segment_access_rights(struct kvm_segment *var) |
6aa8b732 | 1814 | { |
6aa8b732 AK |
1815 | u32 ar; |
1816 | ||
653e3108 | 1817 | if (var->unusable) |
6aa8b732 AK |
1818 | ar = 1 << 16; |
1819 | else { | |
1820 | ar = var->type & 15; | |
1821 | ar |= (var->s & 1) << 4; | |
1822 | ar |= (var->dpl & 3) << 5; | |
1823 | ar |= (var->present & 1) << 7; | |
1824 | ar |= (var->avl & 1) << 12; | |
1825 | ar |= (var->l & 1) << 13; | |
1826 | ar |= (var->db & 1) << 14; | |
1827 | ar |= (var->g & 1) << 15; | |
1828 | } | |
f7fbf1fd UL |
1829 | if (ar == 0) /* a 0 value means unusable */ |
1830 | ar = AR_UNUSABLE_MASK; | |
653e3108 AK |
1831 | |
1832 | return ar; | |
1833 | } | |
1834 | ||
1835 | static void vmx_set_segment(struct kvm_vcpu *vcpu, | |
1836 | struct kvm_segment *var, int seg) | |
1837 | { | |
7ffd92c5 | 1838 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
653e3108 AK |
1839 | struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg]; |
1840 | u32 ar; | |
1841 | ||
7ffd92c5 AK |
1842 | if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) { |
1843 | vmx->rmode.tr.selector = var->selector; | |
1844 | vmx->rmode.tr.base = var->base; | |
1845 | vmx->rmode.tr.limit = var->limit; | |
1846 | vmx->rmode.tr.ar = vmx_segment_access_rights(var); | |
653e3108 AK |
1847 | return; |
1848 | } | |
1849 | vmcs_writel(sf->base, var->base); | |
1850 | vmcs_write32(sf->limit, var->limit); | |
1851 | vmcs_write16(sf->selector, var->selector); | |
7ffd92c5 | 1852 | if (vmx->rmode.vm86_active && var->s) { |
653e3108 AK |
1853 | /* |
1854 | * Hack real-mode segments into vm86 compatibility. | |
1855 | */ | |
1856 | if (var->base == 0xffff0000 && var->selector == 0xf000) | |
1857 | vmcs_writel(sf->base, 0xf0000); | |
1858 | ar = 0xf3; | |
1859 | } else | |
1860 | ar = vmx_segment_access_rights(var); | |
3a624e29 NK |
1861 | |
1862 | /* | |
1863 | * Fix the "Accessed" bit in AR field of segment registers for older | |
1864 | * qemu binaries. | |
1865 | * IA32 arch specifies that at the time of processor reset the | |
1866 | * "Accessed" bit in the AR field of segment registers is 1. And qemu | |
1867 | * is setting it to 0 in the usedland code. This causes invalid guest | |
1868 | * state vmexit when "unrestricted guest" mode is turned on. | |
1869 | * Fix for this setup issue in cpu_reset is being pushed in the qemu | |
1870 | * tree. Newer qemu binaries with that qemu fix would not need this | |
1871 | * kvm hack. | |
1872 | */ | |
1873 | if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR)) | |
1874 | ar |= 0x1; /* Accessed */ | |
1875 | ||
6aa8b732 AK |
1876 | vmcs_write32(sf->ar_bytes, ar); |
1877 | } | |
1878 | ||
6aa8b732 AK |
1879 | static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l) |
1880 | { | |
1881 | u32 ar = vmcs_read32(GUEST_CS_AR_BYTES); | |
1882 | ||
1883 | *db = (ar >> 14) & 1; | |
1884 | *l = (ar >> 13) & 1; | |
1885 | } | |
1886 | ||
1887 | static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt) | |
1888 | { | |
1889 | dt->limit = vmcs_read32(GUEST_IDTR_LIMIT); | |
1890 | dt->base = vmcs_readl(GUEST_IDTR_BASE); | |
1891 | } | |
1892 | ||
1893 | static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt) | |
1894 | { | |
1895 | vmcs_write32(GUEST_IDTR_LIMIT, dt->limit); | |
1896 | vmcs_writel(GUEST_IDTR_BASE, dt->base); | |
1897 | } | |
1898 | ||
1899 | static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt) | |
1900 | { | |
1901 | dt->limit = vmcs_read32(GUEST_GDTR_LIMIT); | |
1902 | dt->base = vmcs_readl(GUEST_GDTR_BASE); | |
1903 | } | |
1904 | ||
1905 | static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt) | |
1906 | { | |
1907 | vmcs_write32(GUEST_GDTR_LIMIT, dt->limit); | |
1908 | vmcs_writel(GUEST_GDTR_BASE, dt->base); | |
1909 | } | |
1910 | ||
648dfaa7 MG |
1911 | static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg) |
1912 | { | |
1913 | struct kvm_segment var; | |
1914 | u32 ar; | |
1915 | ||
1916 | vmx_get_segment(vcpu, &var, seg); | |
1917 | ar = vmx_segment_access_rights(&var); | |
1918 | ||
1919 | if (var.base != (var.selector << 4)) | |
1920 | return false; | |
1921 | if (var.limit != 0xffff) | |
1922 | return false; | |
1923 | if (ar != 0xf3) | |
1924 | return false; | |
1925 | ||
1926 | return true; | |
1927 | } | |
1928 | ||
1929 | static bool code_segment_valid(struct kvm_vcpu *vcpu) | |
1930 | { | |
1931 | struct kvm_segment cs; | |
1932 | unsigned int cs_rpl; | |
1933 | ||
1934 | vmx_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
1935 | cs_rpl = cs.selector & SELECTOR_RPL_MASK; | |
1936 | ||
1872a3f4 AK |
1937 | if (cs.unusable) |
1938 | return false; | |
648dfaa7 MG |
1939 | if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK)) |
1940 | return false; | |
1941 | if (!cs.s) | |
1942 | return false; | |
1872a3f4 | 1943 | if (cs.type & AR_TYPE_WRITEABLE_MASK) { |
648dfaa7 MG |
1944 | if (cs.dpl > cs_rpl) |
1945 | return false; | |
1872a3f4 | 1946 | } else { |
648dfaa7 MG |
1947 | if (cs.dpl != cs_rpl) |
1948 | return false; | |
1949 | } | |
1950 | if (!cs.present) | |
1951 | return false; | |
1952 | ||
1953 | /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */ | |
1954 | return true; | |
1955 | } | |
1956 | ||
1957 | static bool stack_segment_valid(struct kvm_vcpu *vcpu) | |
1958 | { | |
1959 | struct kvm_segment ss; | |
1960 | unsigned int ss_rpl; | |
1961 | ||
1962 | vmx_get_segment(vcpu, &ss, VCPU_SREG_SS); | |
1963 | ss_rpl = ss.selector & SELECTOR_RPL_MASK; | |
1964 | ||
1872a3f4 AK |
1965 | if (ss.unusable) |
1966 | return true; | |
1967 | if (ss.type != 3 && ss.type != 7) | |
648dfaa7 MG |
1968 | return false; |
1969 | if (!ss.s) | |
1970 | return false; | |
1971 | if (ss.dpl != ss_rpl) /* DPL != RPL */ | |
1972 | return false; | |
1973 | if (!ss.present) | |
1974 | return false; | |
1975 | ||
1976 | return true; | |
1977 | } | |
1978 | ||
1979 | static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg) | |
1980 | { | |
1981 | struct kvm_segment var; | |
1982 | unsigned int rpl; | |
1983 | ||
1984 | vmx_get_segment(vcpu, &var, seg); | |
1985 | rpl = var.selector & SELECTOR_RPL_MASK; | |
1986 | ||
1872a3f4 AK |
1987 | if (var.unusable) |
1988 | return true; | |
648dfaa7 MG |
1989 | if (!var.s) |
1990 | return false; | |
1991 | if (!var.present) | |
1992 | return false; | |
1993 | if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) { | |
1994 | if (var.dpl < rpl) /* DPL < RPL */ | |
1995 | return false; | |
1996 | } | |
1997 | ||
1998 | /* TODO: Add other members to kvm_segment_field to allow checking for other access | |
1999 | * rights flags | |
2000 | */ | |
2001 | return true; | |
2002 | } | |
2003 | ||
2004 | static bool tr_valid(struct kvm_vcpu *vcpu) | |
2005 | { | |
2006 | struct kvm_segment tr; | |
2007 | ||
2008 | vmx_get_segment(vcpu, &tr, VCPU_SREG_TR); | |
2009 | ||
1872a3f4 AK |
2010 | if (tr.unusable) |
2011 | return false; | |
648dfaa7 MG |
2012 | if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */ |
2013 | return false; | |
1872a3f4 | 2014 | if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */ |
648dfaa7 MG |
2015 | return false; |
2016 | if (!tr.present) | |
2017 | return false; | |
2018 | ||
2019 | return true; | |
2020 | } | |
2021 | ||
2022 | static bool ldtr_valid(struct kvm_vcpu *vcpu) | |
2023 | { | |
2024 | struct kvm_segment ldtr; | |
2025 | ||
2026 | vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR); | |
2027 | ||
1872a3f4 AK |
2028 | if (ldtr.unusable) |
2029 | return true; | |
648dfaa7 MG |
2030 | if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */ |
2031 | return false; | |
2032 | if (ldtr.type != 2) | |
2033 | return false; | |
2034 | if (!ldtr.present) | |
2035 | return false; | |
2036 | ||
2037 | return true; | |
2038 | } | |
2039 | ||
2040 | static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu) | |
2041 | { | |
2042 | struct kvm_segment cs, ss; | |
2043 | ||
2044 | vmx_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
2045 | vmx_get_segment(vcpu, &ss, VCPU_SREG_SS); | |
2046 | ||
2047 | return ((cs.selector & SELECTOR_RPL_MASK) == | |
2048 | (ss.selector & SELECTOR_RPL_MASK)); | |
2049 | } | |
2050 | ||
2051 | /* | |
2052 | * Check if guest state is valid. Returns true if valid, false if | |
2053 | * not. | |
2054 | * We assume that registers are always usable | |
2055 | */ | |
2056 | static bool guest_state_valid(struct kvm_vcpu *vcpu) | |
2057 | { | |
2058 | /* real mode guest state checks */ | |
2059 | if (!(vcpu->arch.cr0 & X86_CR0_PE)) { | |
2060 | if (!rmode_segment_valid(vcpu, VCPU_SREG_CS)) | |
2061 | return false; | |
2062 | if (!rmode_segment_valid(vcpu, VCPU_SREG_SS)) | |
2063 | return false; | |
2064 | if (!rmode_segment_valid(vcpu, VCPU_SREG_DS)) | |
2065 | return false; | |
2066 | if (!rmode_segment_valid(vcpu, VCPU_SREG_ES)) | |
2067 | return false; | |
2068 | if (!rmode_segment_valid(vcpu, VCPU_SREG_FS)) | |
2069 | return false; | |
2070 | if (!rmode_segment_valid(vcpu, VCPU_SREG_GS)) | |
2071 | return false; | |
2072 | } else { | |
2073 | /* protected mode guest state checks */ | |
2074 | if (!cs_ss_rpl_check(vcpu)) | |
2075 | return false; | |
2076 | if (!code_segment_valid(vcpu)) | |
2077 | return false; | |
2078 | if (!stack_segment_valid(vcpu)) | |
2079 | return false; | |
2080 | if (!data_segment_valid(vcpu, VCPU_SREG_DS)) | |
2081 | return false; | |
2082 | if (!data_segment_valid(vcpu, VCPU_SREG_ES)) | |
2083 | return false; | |
2084 | if (!data_segment_valid(vcpu, VCPU_SREG_FS)) | |
2085 | return false; | |
2086 | if (!data_segment_valid(vcpu, VCPU_SREG_GS)) | |
2087 | return false; | |
2088 | if (!tr_valid(vcpu)) | |
2089 | return false; | |
2090 | if (!ldtr_valid(vcpu)) | |
2091 | return false; | |
2092 | } | |
2093 | /* TODO: | |
2094 | * - Add checks on RIP | |
2095 | * - Add checks on RFLAGS | |
2096 | */ | |
2097 | ||
2098 | return true; | |
2099 | } | |
2100 | ||
d77c26fc | 2101 | static int init_rmode_tss(struct kvm *kvm) |
6aa8b732 | 2102 | { |
6aa8b732 | 2103 | gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT; |
195aefde | 2104 | u16 data = 0; |
10589a46 | 2105 | int ret = 0; |
195aefde | 2106 | int r; |
6aa8b732 | 2107 | |
195aefde IE |
2108 | r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE); |
2109 | if (r < 0) | |
10589a46 | 2110 | goto out; |
195aefde | 2111 | data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE; |
464d17c8 SY |
2112 | r = kvm_write_guest_page(kvm, fn++, &data, |
2113 | TSS_IOPB_BASE_OFFSET, sizeof(u16)); | |
195aefde | 2114 | if (r < 0) |
10589a46 | 2115 | goto out; |
195aefde IE |
2116 | r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE); |
2117 | if (r < 0) | |
10589a46 | 2118 | goto out; |
195aefde IE |
2119 | r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE); |
2120 | if (r < 0) | |
10589a46 | 2121 | goto out; |
195aefde | 2122 | data = ~0; |
10589a46 MT |
2123 | r = kvm_write_guest_page(kvm, fn, &data, |
2124 | RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1, | |
2125 | sizeof(u8)); | |
195aefde | 2126 | if (r < 0) |
10589a46 MT |
2127 | goto out; |
2128 | ||
2129 | ret = 1; | |
2130 | out: | |
10589a46 | 2131 | return ret; |
6aa8b732 AK |
2132 | } |
2133 | ||
b7ebfb05 SY |
2134 | static int init_rmode_identity_map(struct kvm *kvm) |
2135 | { | |
2136 | int i, r, ret; | |
2137 | pfn_t identity_map_pfn; | |
2138 | u32 tmp; | |
2139 | ||
089d034e | 2140 | if (!enable_ept) |
b7ebfb05 SY |
2141 | return 1; |
2142 | if (unlikely(!kvm->arch.ept_identity_pagetable)) { | |
2143 | printk(KERN_ERR "EPT: identity-mapping pagetable " | |
2144 | "haven't been allocated!\n"); | |
2145 | return 0; | |
2146 | } | |
2147 | if (likely(kvm->arch.ept_identity_pagetable_done)) | |
2148 | return 1; | |
2149 | ret = 0; | |
b927a3ce | 2150 | identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT; |
b7ebfb05 SY |
2151 | r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE); |
2152 | if (r < 0) | |
2153 | goto out; | |
2154 | /* Set up identity-mapping pagetable for EPT in real mode */ | |
2155 | for (i = 0; i < PT32_ENT_PER_PAGE; i++) { | |
2156 | tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | | |
2157 | _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE); | |
2158 | r = kvm_write_guest_page(kvm, identity_map_pfn, | |
2159 | &tmp, i * sizeof(tmp), sizeof(tmp)); | |
2160 | if (r < 0) | |
2161 | goto out; | |
2162 | } | |
2163 | kvm->arch.ept_identity_pagetable_done = true; | |
2164 | ret = 1; | |
2165 | out: | |
2166 | return ret; | |
2167 | } | |
2168 | ||
6aa8b732 AK |
2169 | static void seg_setup(int seg) |
2170 | { | |
2171 | struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg]; | |
3a624e29 | 2172 | unsigned int ar; |
6aa8b732 AK |
2173 | |
2174 | vmcs_write16(sf->selector, 0); | |
2175 | vmcs_writel(sf->base, 0); | |
2176 | vmcs_write32(sf->limit, 0xffff); | |
3a624e29 NK |
2177 | if (enable_unrestricted_guest) { |
2178 | ar = 0x93; | |
2179 | if (seg == VCPU_SREG_CS) | |
2180 | ar |= 0x08; /* code segment */ | |
2181 | } else | |
2182 | ar = 0xf3; | |
2183 | ||
2184 | vmcs_write32(sf->ar_bytes, ar); | |
6aa8b732 AK |
2185 | } |
2186 | ||
f78e0e2e SY |
2187 | static int alloc_apic_access_page(struct kvm *kvm) |
2188 | { | |
2189 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2190 | int r = 0; | |
2191 | ||
72dc67a6 | 2192 | down_write(&kvm->slots_lock); |
bfc6d222 | 2193 | if (kvm->arch.apic_access_page) |
f78e0e2e SY |
2194 | goto out; |
2195 | kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT; | |
2196 | kvm_userspace_mem.flags = 0; | |
2197 | kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL; | |
2198 | kvm_userspace_mem.memory_size = PAGE_SIZE; | |
2199 | r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0); | |
2200 | if (r) | |
2201 | goto out; | |
72dc67a6 | 2202 | |
bfc6d222 | 2203 | kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00); |
f78e0e2e | 2204 | out: |
72dc67a6 | 2205 | up_write(&kvm->slots_lock); |
f78e0e2e SY |
2206 | return r; |
2207 | } | |
2208 | ||
b7ebfb05 SY |
2209 | static int alloc_identity_pagetable(struct kvm *kvm) |
2210 | { | |
2211 | struct kvm_userspace_memory_region kvm_userspace_mem; | |
2212 | int r = 0; | |
2213 | ||
2214 | down_write(&kvm->slots_lock); | |
2215 | if (kvm->arch.ept_identity_pagetable) | |
2216 | goto out; | |
2217 | kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT; | |
2218 | kvm_userspace_mem.flags = 0; | |
b927a3ce SY |
2219 | kvm_userspace_mem.guest_phys_addr = |
2220 | kvm->arch.ept_identity_map_addr; | |
b7ebfb05 SY |
2221 | kvm_userspace_mem.memory_size = PAGE_SIZE; |
2222 | r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0); | |
2223 | if (r) | |
2224 | goto out; | |
2225 | ||
b7ebfb05 | 2226 | kvm->arch.ept_identity_pagetable = gfn_to_page(kvm, |
b927a3ce | 2227 | kvm->arch.ept_identity_map_addr >> PAGE_SHIFT); |
b7ebfb05 SY |
2228 | out: |
2229 | up_write(&kvm->slots_lock); | |
2230 | return r; | |
2231 | } | |
2232 | ||
2384d2b3 SY |
2233 | static void allocate_vpid(struct vcpu_vmx *vmx) |
2234 | { | |
2235 | int vpid; | |
2236 | ||
2237 | vmx->vpid = 0; | |
919818ab | 2238 | if (!enable_vpid) |
2384d2b3 SY |
2239 | return; |
2240 | spin_lock(&vmx_vpid_lock); | |
2241 | vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS); | |
2242 | if (vpid < VMX_NR_VPIDS) { | |
2243 | vmx->vpid = vpid; | |
2244 | __set_bit(vpid, vmx_vpid_bitmap); | |
2245 | } | |
2246 | spin_unlock(&vmx_vpid_lock); | |
2247 | } | |
2248 | ||
5897297b | 2249 | static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr) |
25c5f225 | 2250 | { |
3e7c73e9 | 2251 | int f = sizeof(unsigned long); |
25c5f225 SY |
2252 | |
2253 | if (!cpu_has_vmx_msr_bitmap()) | |
2254 | return; | |
2255 | ||
2256 | /* | |
2257 | * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals | |
2258 | * have the write-low and read-high bitmap offsets the wrong way round. | |
2259 | * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff. | |
2260 | */ | |
25c5f225 | 2261 | if (msr <= 0x1fff) { |
3e7c73e9 AK |
2262 | __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */ |
2263 | __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */ | |
25c5f225 SY |
2264 | } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) { |
2265 | msr &= 0x1fff; | |
3e7c73e9 AK |
2266 | __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */ |
2267 | __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */ | |
25c5f225 | 2268 | } |
25c5f225 SY |
2269 | } |
2270 | ||
5897297b AK |
2271 | static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only) |
2272 | { | |
2273 | if (!longmode_only) | |
2274 | __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr); | |
2275 | __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr); | |
2276 | } | |
2277 | ||
6aa8b732 AK |
2278 | /* |
2279 | * Sets up the vmcs for emulated real mode. | |
2280 | */ | |
8b9cf98c | 2281 | static int vmx_vcpu_setup(struct vcpu_vmx *vmx) |
6aa8b732 | 2282 | { |
468d472f | 2283 | u32 host_sysenter_cs, msr_low, msr_high; |
6aa8b732 | 2284 | u32 junk; |
53f658b3 | 2285 | u64 host_pat, tsc_this, tsc_base; |
6aa8b732 AK |
2286 | unsigned long a; |
2287 | struct descriptor_table dt; | |
2288 | int i; | |
cd2276a7 | 2289 | unsigned long kvm_vmx_return; |
6e5d865c | 2290 | u32 exec_control; |
6aa8b732 | 2291 | |
6aa8b732 | 2292 | /* I/O */ |
3e7c73e9 AK |
2293 | vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a)); |
2294 | vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b)); | |
6aa8b732 | 2295 | |
25c5f225 | 2296 | if (cpu_has_vmx_msr_bitmap()) |
5897297b | 2297 | vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy)); |
25c5f225 | 2298 | |
6aa8b732 AK |
2299 | vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */ |
2300 | ||
6aa8b732 | 2301 | /* Control */ |
1c3d14fe YS |
2302 | vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, |
2303 | vmcs_config.pin_based_exec_ctrl); | |
6e5d865c YS |
2304 | |
2305 | exec_control = vmcs_config.cpu_based_exec_ctrl; | |
2306 | if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) { | |
2307 | exec_control &= ~CPU_BASED_TPR_SHADOW; | |
2308 | #ifdef CONFIG_X86_64 | |
2309 | exec_control |= CPU_BASED_CR8_STORE_EXITING | | |
2310 | CPU_BASED_CR8_LOAD_EXITING; | |
2311 | #endif | |
2312 | } | |
089d034e | 2313 | if (!enable_ept) |
d56f546d | 2314 | exec_control |= CPU_BASED_CR3_STORE_EXITING | |
83dbc83a MT |
2315 | CPU_BASED_CR3_LOAD_EXITING | |
2316 | CPU_BASED_INVLPG_EXITING; | |
6e5d865c | 2317 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control); |
6aa8b732 | 2318 | |
83ff3b9d SY |
2319 | if (cpu_has_secondary_exec_ctrls()) { |
2320 | exec_control = vmcs_config.cpu_based_2nd_exec_ctrl; | |
2321 | if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm)) | |
2322 | exec_control &= | |
2323 | ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES; | |
2384d2b3 SY |
2324 | if (vmx->vpid == 0) |
2325 | exec_control &= ~SECONDARY_EXEC_ENABLE_VPID; | |
089d034e | 2326 | if (!enable_ept) |
d56f546d | 2327 | exec_control &= ~SECONDARY_EXEC_ENABLE_EPT; |
3a624e29 NK |
2328 | if (!enable_unrestricted_guest) |
2329 | exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST; | |
4b8d54f9 ZE |
2330 | if (!ple_gap) |
2331 | exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING; | |
83ff3b9d SY |
2332 | vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control); |
2333 | } | |
f78e0e2e | 2334 | |
4b8d54f9 ZE |
2335 | if (ple_gap) { |
2336 | vmcs_write32(PLE_GAP, ple_gap); | |
2337 | vmcs_write32(PLE_WINDOW, ple_window); | |
2338 | } | |
2339 | ||
c7addb90 AK |
2340 | vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf); |
2341 | vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf); | |
6aa8b732 AK |
2342 | vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */ |
2343 | ||
2344 | vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */ | |
2345 | vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */ | |
2346 | vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */ | |
2347 | ||
2348 | vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */ | |
2349 | vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */ | |
2350 | vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */ | |
d6e88aec AK |
2351 | vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs()); /* 22.2.4 */ |
2352 | vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs()); /* 22.2.4 */ | |
6aa8b732 | 2353 | vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */ |
05b3e0c2 | 2354 | #ifdef CONFIG_X86_64 |
6aa8b732 AK |
2355 | rdmsrl(MSR_FS_BASE, a); |
2356 | vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */ | |
2357 | rdmsrl(MSR_GS_BASE, a); | |
2358 | vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */ | |
2359 | #else | |
2360 | vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */ | |
2361 | vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */ | |
2362 | #endif | |
2363 | ||
2364 | vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */ | |
2365 | ||
d6e88aec | 2366 | kvm_get_idt(&dt); |
6aa8b732 AK |
2367 | vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */ |
2368 | ||
d77c26fc | 2369 | asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return)); |
cd2276a7 | 2370 | vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */ |
2cc51560 ED |
2371 | vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0); |
2372 | vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0); | |
2373 | vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0); | |
6aa8b732 AK |
2374 | |
2375 | rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk); | |
2376 | vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs); | |
2377 | rdmsrl(MSR_IA32_SYSENTER_ESP, a); | |
2378 | vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */ | |
2379 | rdmsrl(MSR_IA32_SYSENTER_EIP, a); | |
2380 | vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */ | |
2381 | ||
468d472f SY |
2382 | if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) { |
2383 | rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high); | |
2384 | host_pat = msr_low | ((u64) msr_high << 32); | |
2385 | vmcs_write64(HOST_IA32_PAT, host_pat); | |
2386 | } | |
2387 | if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) { | |
2388 | rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high); | |
2389 | host_pat = msr_low | ((u64) msr_high << 32); | |
2390 | /* Write the default value follow host pat */ | |
2391 | vmcs_write64(GUEST_IA32_PAT, host_pat); | |
2392 | /* Keep arch.pat sync with GUEST_IA32_PAT */ | |
2393 | vmx->vcpu.arch.pat = host_pat; | |
2394 | } | |
2395 | ||
6aa8b732 AK |
2396 | for (i = 0; i < NR_VMX_MSR; ++i) { |
2397 | u32 index = vmx_msr_index[i]; | |
2398 | u32 data_low, data_high; | |
2399 | u64 data; | |
a2fa3e9f | 2400 | int j = vmx->nmsrs; |
6aa8b732 AK |
2401 | |
2402 | if (rdmsr_safe(index, &data_low, &data_high) < 0) | |
2403 | continue; | |
432bd6cb AK |
2404 | if (wrmsr_safe(index, data_low, data_high) < 0) |
2405 | continue; | |
6aa8b732 | 2406 | data = data_low | ((u64)data_high << 32); |
a2fa3e9f GH |
2407 | vmx->host_msrs[j].index = index; |
2408 | vmx->host_msrs[j].reserved = 0; | |
2409 | vmx->host_msrs[j].data = data; | |
2410 | vmx->guest_msrs[j] = vmx->host_msrs[j]; | |
2411 | ++vmx->nmsrs; | |
6aa8b732 | 2412 | } |
6aa8b732 | 2413 | |
1c3d14fe | 2414 | vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl); |
6aa8b732 AK |
2415 | |
2416 | /* 22.2.1, 20.8.1 */ | |
1c3d14fe YS |
2417 | vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl); |
2418 | ||
e00c8cf2 AK |
2419 | vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL); |
2420 | vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK); | |
2421 | ||
53f658b3 MT |
2422 | tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc; |
2423 | rdtscll(tsc_this); | |
2424 | if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc) | |
2425 | tsc_base = tsc_this; | |
2426 | ||
2427 | guest_write_tsc(0, tsc_base); | |
f78e0e2e | 2428 | |
e00c8cf2 AK |
2429 | return 0; |
2430 | } | |
2431 | ||
b7ebfb05 SY |
2432 | static int init_rmode(struct kvm *kvm) |
2433 | { | |
2434 | if (!init_rmode_tss(kvm)) | |
2435 | return 0; | |
2436 | if (!init_rmode_identity_map(kvm)) | |
2437 | return 0; | |
2438 | return 1; | |
2439 | } | |
2440 | ||
e00c8cf2 AK |
2441 | static int vmx_vcpu_reset(struct kvm_vcpu *vcpu) |
2442 | { | |
2443 | struct vcpu_vmx *vmx = to_vmx(vcpu); | |
2444 | u64 msr; | |
2445 | int ret; | |
2446 | ||
5fdbf976 | 2447 | vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)); |
3200f405 | 2448 | down_read(&vcpu->kvm->slots_lock); |
b7ebfb05 | 2449 | if (!init_rmode(vmx->vcpu.kvm)) { |
e00c8cf2 AK |
2450 | ret = -ENOMEM; |
2451 | goto out; | |
2452 | } | |
2453 | ||
7ffd92c5 | 2454 | vmx->rmode.vm86_active = 0; |
e00c8cf2 | 2455 | |
3b86cd99 JK |
2456 | vmx->soft_vnmi_blocked = 0; |
2457 | ||
ad312c7c | 2458 | vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val(); |
2d3ad1f4 | 2459 | kvm_set_cr8(&vmx->vcpu, 0); |
e00c8cf2 | 2460 | msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE; |
c5af89b6 | 2461 | if (kvm_vcpu_is_bsp(&vmx->vcpu)) |
e00c8cf2 AK |
2462 | msr |= MSR_IA32_APICBASE_BSP; |
2463 | kvm_set_apic_base(&vmx->vcpu, msr); | |
2464 | ||
2465 | fx_init(&vmx->vcpu); | |
2466 | ||
5706be0d | 2467 | seg_setup(VCPU_SREG_CS); |
e00c8cf2 AK |
2468 | /* |
2469 | * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode | |
2470 | * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh. | |
2471 | */ | |
c5af89b6 | 2472 | if (kvm_vcpu_is_bsp(&vmx->vcpu)) { |
e00c8cf2 AK |
2473 | vmcs_write16(GUEST_CS_SELECTOR, 0xf000); |
2474 | vmcs_writel(GUEST_CS_BASE, 0x000f0000); | |
2475 | } else { | |
ad312c7c ZX |
2476 | vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8); |
2477 | vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12); | |
e00c8cf2 | 2478 | } |
e00c8cf2 AK |
2479 | |
2480 | seg_setup(VCPU_SREG_DS); | |
2481 | seg_setup(VCPU_SREG_ES); | |
2482 | seg_setup(VCPU_SREG_FS); | |
2483 | seg_setup(VCPU_SREG_GS); | |
2484 | seg_setup(VCPU_SREG_SS); | |
2485 | ||
2486 | vmcs_write16(GUEST_TR_SELECTOR, 0); | |
2487 | vmcs_writel(GUEST_TR_BASE, 0); | |
2488 | vmcs_write32(GUEST_TR_LIMIT, 0xffff); | |
2489 | vmcs_write32(GUEST_TR_AR_BYTES, 0x008b); | |
2490 | ||
2491 | vmcs_write16(GUEST_LDTR_SELECTOR, 0); | |
2492 | vmcs_writel(GUEST_LDTR_BASE, 0); | |
2493 | vmcs_write32(GUEST_LDTR_LIMIT, 0xffff); | |
2494 | vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082); | |
2495 | ||
2496 | vmcs_write32(GUEST_SYSENTER_CS, 0); | |
2497 | vmcs_writel(GUEST_SYSENTER_ESP, 0); | |
2498 | vmcs_writel(GUEST_SYSENTER_EIP, 0); | |
2499 | ||
2500 | vmcs_writel(GUEST_RFLAGS, 0x02); | |
c5af89b6 | 2501 | if (kvm_vcpu_is_bsp(&vmx->vcpu)) |
5fdbf976 | 2502 | kvm_rip_write(vcpu, 0xfff0); |
e00c8cf2 | 2503 | else |
5fdbf976 MT |
2504 | kvm_rip_write(vcpu, 0); |
2505 | kvm_register_write(vcpu, VCPU_REGS_RSP, 0); | |
e00c8cf2 | 2506 | |
e00c8cf2 AK |
2507 | vmcs_writel(GUEST_DR7, 0x400); |
2508 | ||
2509 | vmcs_writel(GUEST_GDTR_BASE, 0); | |
2510 | vmcs_write32(GUEST_GDTR_LIMIT, 0xffff); | |
2511 | ||
2512 | vmcs_writel(GUEST_IDTR_BASE, 0); | |
2513 | vmcs_write32(GUEST_IDTR_LIMIT, 0xffff); | |
2514 | ||
2515 | vmcs_write32(GUEST_ACTIVITY_STATE, 0); | |
2516 | vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0); | |
2517 | vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0); | |
2518 | ||
e00c8cf2 AK |
2519 | /* Special registers */ |
2520 | vmcs_write64(GUEST_IA32_DEBUGCTL, 0); | |
2521 | ||
2522 | setup_msrs(vmx); | |
2523 | ||
6aa8b732 AK |
2524 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */ |
2525 | ||
f78e0e2e SY |
2526 | if (cpu_has_vmx_tpr_shadow()) { |
2527 | vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0); | |
2528 | if (vm_need_tpr_shadow(vmx->vcpu.kvm)) | |
2529 | vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, | |
ad312c7c | 2530 | page_to_phys(vmx->vcpu.arch.apic->regs_page)); |
f78e0e2e SY |
2531 | vmcs_write32(TPR_THRESHOLD, 0); |
2532 | } | |
2533 | ||
2534 | if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm)) | |
2535 | vmcs_write64(APIC_ACCESS_ADDR, | |
bfc6d222 | 2536 | page_to_phys(vmx->vcpu.kvm->arch.apic_access_page)); |
6aa8b732 | 2537 | |
2384d2b3 SY |
2538 | if (vmx->vpid != 0) |
2539 | vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid); | |
2540 | ||
ad312c7c ZX |
2541 | vmx->vcpu.arch.cr0 = 0x60000010; |
2542 | vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */ | |
8b9cf98c | 2543 | vmx_set_cr4(&vmx->vcpu, 0); |
8b9cf98c | 2544 | vmx_set_efer(&vmx->vcpu, 0); |
8b9cf98c RR |
2545 | vmx_fpu_activate(&vmx->vcpu); |
2546 | update_exception_bitmap(&vmx->vcpu); | |
6aa8b732 | 2547 | |
2384d2b3 SY |
2548 | vpid_sync_vcpu_all(vmx); |
2549 | ||
3200f405 | 2550 | ret = 0; |
6aa8b732 | 2551 | |
a89a8fb9 MG |
2552 | /* HACK: Don't enable emulation on guest boot/reset */ |
2553 | vmx->emulation_required = 0; | |
2554 | ||
6aa8b732 | 2555 | out: |
3200f405 | 2556 | up_read(&vcpu->kvm->slots_lock); |
6aa8b732 AK |
2557 | return ret; |
2558 | } | |
2559 | ||
3b86cd99 JK |
2560 | static void enable_irq_window(struct kvm_vcpu *vcpu) |
2561 | { | |
2562 | u32 cpu_based_vm_exec_control; | |
2563 | ||
2564 | cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); | |
2565 | cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING; | |
2566 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); | |
2567 | } | |
2568 | ||
2569 | static void enable_nmi_window(struct kvm_vcpu *vcpu) | |
2570 | { | |
2571 | u32 cpu_based_vm_exec_control; | |
2572 | ||
2573 | if (!cpu_has_virtual_nmis()) { | |
2574 | enable_irq_window(vcpu); | |
2575 | return; | |
2576 | } | |
2577 | ||
2578 | cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); | |
2579 | cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING; | |
2580 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); | |
2581 | } | |
2582 | ||
66fd3f7f | 2583 | static void vmx_inject_irq(struct kvm_vcpu *vcpu) |
85f455f7 | 2584 | { |
9c8cba37 | 2585 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
66fd3f7f GN |
2586 | uint32_t intr; |
2587 | int irq = vcpu->arch.interrupt.nr; | |
9c8cba37 | 2588 | |
229456fc | 2589 | trace_kvm_inj_virq(irq); |
2714d1d3 | 2590 | |
fa89a817 | 2591 | ++vcpu->stat.irq_injections; |
7ffd92c5 | 2592 | if (vmx->rmode.vm86_active) { |
9c8cba37 AK |
2593 | vmx->rmode.irq.pending = true; |
2594 | vmx->rmode.irq.vector = irq; | |
5fdbf976 | 2595 | vmx->rmode.irq.rip = kvm_rip_read(vcpu); |
ae0bb3e0 GN |
2596 | if (vcpu->arch.interrupt.soft) |
2597 | vmx->rmode.irq.rip += | |
2598 | vmx->vcpu.arch.event_exit_inst_len; | |
9c5623e3 AK |
2599 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, |
2600 | irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK); | |
2601 | vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1); | |
5fdbf976 | 2602 | kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1); |
85f455f7 ED |
2603 | return; |
2604 | } | |
66fd3f7f GN |
2605 | intr = irq | INTR_INFO_VALID_MASK; |
2606 | if (vcpu->arch.interrupt.soft) { | |
2607 | intr |= INTR_TYPE_SOFT_INTR; | |
2608 | vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, | |
2609 | vmx->vcpu.arch.event_exit_inst_len); | |
2610 | } else | |
2611 | intr |= INTR_TYPE_EXT_INTR; | |
2612 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr); | |
85f455f7 ED |
2613 | } |
2614 | ||
f08864b4 SY |
2615 | static void vmx_inject_nmi(struct kvm_vcpu *vcpu) |
2616 | { | |
66a5a347 JK |
2617 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
2618 | ||
3b86cd99 JK |
2619 | if (!cpu_has_virtual_nmis()) { |
2620 | /* | |
2621 | * Tracking the NMI-blocked state in software is built upon | |
2622 | * finding the next open IRQ window. This, in turn, depends on | |
2623 | * well-behaving guests: They have to keep IRQs disabled at | |
2624 | * least as long as the NMI handler runs. Otherwise we may | |
2625 | * cause NMI nesting, maybe breaking the guest. But as this is | |
2626 | * highly unlikely, we can live with the residual risk. | |
2627 | */ | |
2628 | vmx->soft_vnmi_blocked = 1; | |
2629 | vmx->vnmi_blocked_time = 0; | |
2630 | } | |
2631 | ||
487b391d | 2632 | ++vcpu->stat.nmi_injections; |
7ffd92c5 | 2633 | if (vmx->rmode.vm86_active) { |
66a5a347 JK |
2634 | vmx->rmode.irq.pending = true; |
2635 | vmx->rmode.irq.vector = NMI_VECTOR; | |
2636 | vmx->rmode.irq.rip = kvm_rip_read(vcpu); | |
2637 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, | |
2638 | NMI_VECTOR | INTR_TYPE_SOFT_INTR | | |
2639 | INTR_INFO_VALID_MASK); | |
2640 | vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1); | |
2641 | kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1); | |
2642 | return; | |
2643 | } | |
f08864b4 SY |
2644 | vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, |
2645 | INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR); | |
f08864b4 SY |
2646 | } |
2647 | ||
c4282df9 | 2648 | static int vmx_nmi_allowed(struct kvm_vcpu *vcpu) |
33f089ca | 2649 | { |
3b86cd99 | 2650 | if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked) |
c4282df9 | 2651 | return 0; |
33f089ca | 2652 | |
c4282df9 GN |
2653 | return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & |
2654 | (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS | | |
2655 | GUEST_INTR_STATE_NMI)); | |
33f089ca JK |
2656 | } |
2657 | ||
78646121 GN |
2658 | static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu) |
2659 | { | |
c4282df9 GN |
2660 | return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) && |
2661 | !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & | |
2662 | (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS)); | |
78646121 GN |
2663 | } |
2664 | ||
cbc94022 IE |
2665 | static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr) |
2666 | { | |
2667 | int ret; | |
2668 | struct kvm_userspace_memory_region tss_mem = { | |
6fe63979 | 2669 | .slot = TSS_PRIVATE_MEMSLOT, |
cbc94022 IE |
2670 | .guest_phys_addr = addr, |
2671 | .memory_size = PAGE_SIZE * 3, | |
2672 | .flags = 0, | |
2673 | }; | |
2674 | ||
2675 | ret = kvm_set_memory_region(kvm, &tss_mem, 0); | |
2676 | if (ret) | |
2677 | return ret; | |
bfc6d222 | 2678 | kvm->arch.tss_addr = addr; |
cbc94022 IE |
2679 | return 0; |
2680 | } | |
2681 | ||
6aa8b732 AK |
2682 | static int handle_rmode_exception(struct kvm_vcpu *vcpu, |
2683 | int vec, u32 err_code) | |
2684 | { | |
b3f37707 NK |
2685 | /* |
2686 | * Instruction with address size override prefix opcode 0x67 | |
2687 | * Cause the #SS fault with 0 error code in VM86 mode. | |
2688 | */ | |
2689 | if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) | |
851ba692 | 2690 | if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE) |
6aa8b732 | 2691 | return 1; |
77ab6db0 JK |
2692 | /* |
2693 | * Forward all other exceptions that are valid in real mode. | |
2694 | * FIXME: Breaks guest debugging in real mode, needs to be fixed with | |
2695 | * the required debugging infrastructure rework. | |
2696 | */ | |
2697 | switch (vec) { | |
77ab6db0 | 2698 | case DB_VECTOR: |
d0bfb940 JK |
2699 | if (vcpu->guest_debug & |
2700 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) | |
2701 | return 0; | |
2702 | kvm_queue_exception(vcpu, vec); | |
2703 | return 1; | |
77ab6db0 | 2704 | case BP_VECTOR: |
d0bfb940 JK |
2705 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) |
2706 | return 0; | |
2707 | /* fall through */ | |
2708 | case DE_VECTOR: | |
77ab6db0 JK |
2709 | case OF_VECTOR: |
2710 | case BR_VECTOR: | |
2711 | case UD_VECTOR: | |
2712 | case DF_VECTOR: | |
2713 | case SS_VECTOR: | |
2714 | case GP_VECTOR: | |
2715 | case MF_VECTOR: | |
2716 | kvm_queue_exception(vcpu, vec); | |
2717 | return 1; | |
2718 | } | |
6aa8b732 AK |
2719 | return 0; |
2720 | } | |
2721 | ||
a0861c02 AK |
2722 | /* |
2723 | * Trigger machine check on the host. We assume all the MSRs are already set up | |
2724 | * by the CPU and that we still run on the same CPU as the MCE occurred on. | |
2725 | * We pass a fake environment to the machine check handler because we want | |
2726 | * the guest to be always treated like user space, no matter what context | |
2727 | * it used internally. | |
2728 | */ | |
2729 | static void kvm_machine_check(void) | |
2730 | { | |
2731 | #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64) | |
2732 | struct pt_regs regs = { | |
2733 | .cs = 3, /* Fake ring 3 no matter what the guest ran on */ | |
2734 | .flags = X86_EFLAGS_IF, | |
2735 | }; | |
2736 | ||
2737 | do_machine_check(®s, 0); | |
2738 | #endif | |
2739 | } | |
2740 | ||
851ba692 | 2741 | static int handle_machine_check(struct kvm_vcpu *vcpu) |
a0861c02 AK |
2742 | { |
2743 | /* already handled by vcpu_run */ | |
2744 | return 1; | |
2745 | } | |
2746 | ||
851ba692 | 2747 | static int handle_exception(struct kvm_vcpu *vcpu) |
6aa8b732 | 2748 | { |
1155f76a | 2749 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
851ba692 | 2750 | struct kvm_run *kvm_run = vcpu->run; |
d0bfb940 | 2751 | u32 intr_info, ex_no, error_code; |
42dbaa5a | 2752 | unsigned long cr2, rip, dr6; |
6aa8b732 AK |
2753 | u32 vect_info; |
2754 | enum emulation_result er; | |
2755 | ||
1155f76a | 2756 | vect_info = vmx->idt_vectoring_info; |
6aa8b732 AK |
2757 | intr_info = vmcs_read32(VM_EXIT_INTR_INFO); |
2758 | ||
a0861c02 | 2759 | if (is_machine_check(intr_info)) |
851ba692 | 2760 | return handle_machine_check(vcpu); |
a0861c02 | 2761 | |
6aa8b732 | 2762 | if ((vect_info & VECTORING_INFO_VALID_MASK) && |
d77c26fc | 2763 | !is_page_fault(intr_info)) |
6aa8b732 | 2764 | printk(KERN_ERR "%s: unexpected, vectoring info 0x%x " |
b8688d51 | 2765 | "intr info 0x%x\n", __func__, vect_info, intr_info); |
6aa8b732 | 2766 | |
e4a41889 | 2767 | if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR) |
1b6269db | 2768 | return 1; /* already handled by vmx_vcpu_run() */ |
2ab455cc AL |
2769 | |
2770 | if (is_no_device(intr_info)) { | |
5fd86fcf | 2771 | vmx_fpu_activate(vcpu); |
2ab455cc AL |
2772 | return 1; |
2773 | } | |
2774 | ||
7aa81cc0 | 2775 | if (is_invalid_opcode(intr_info)) { |
851ba692 | 2776 | er = emulate_instruction(vcpu, 0, 0, EMULTYPE_TRAP_UD); |
7aa81cc0 | 2777 | if (er != EMULATE_DONE) |
7ee5d940 | 2778 | kvm_queue_exception(vcpu, UD_VECTOR); |
7aa81cc0 AL |
2779 | return 1; |
2780 | } | |
2781 | ||
6aa8b732 | 2782 | error_code = 0; |
5fdbf976 | 2783 | rip = kvm_rip_read(vcpu); |
2e11384c | 2784 | if (intr_info & INTR_INFO_DELIVER_CODE_MASK) |
6aa8b732 AK |
2785 | error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE); |
2786 | if (is_page_fault(intr_info)) { | |
1439442c | 2787 | /* EPT won't cause page fault directly */ |
089d034e | 2788 | if (enable_ept) |
1439442c | 2789 | BUG(); |
6aa8b732 | 2790 | cr2 = vmcs_readl(EXIT_QUALIFICATION); |
229456fc MT |
2791 | trace_kvm_page_fault(cr2, error_code); |
2792 | ||
3298b75c | 2793 | if (kvm_event_needs_reinjection(vcpu)) |
577bdc49 | 2794 | kvm_mmu_unprotect_page_virt(vcpu, cr2); |
3067714c | 2795 | return kvm_mmu_page_fault(vcpu, cr2, error_code); |
6aa8b732 AK |
2796 | } |
2797 | ||
7ffd92c5 | 2798 | if (vmx->rmode.vm86_active && |
6aa8b732 | 2799 | handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK, |
72d6e5a0 | 2800 | error_code)) { |
ad312c7c ZX |
2801 | if (vcpu->arch.halt_request) { |
2802 | vcpu->arch.halt_request = 0; | |
72d6e5a0 AK |
2803 | return kvm_emulate_halt(vcpu); |
2804 | } | |
6aa8b732 | 2805 | return 1; |
72d6e5a0 | 2806 | } |
6aa8b732 | 2807 | |
d0bfb940 | 2808 | ex_no = intr_info & INTR_INFO_VECTOR_MASK; |
42dbaa5a JK |
2809 | switch (ex_no) { |
2810 | case DB_VECTOR: | |
2811 | dr6 = vmcs_readl(EXIT_QUALIFICATION); | |
2812 | if (!(vcpu->guest_debug & | |
2813 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) { | |
2814 | vcpu->arch.dr6 = dr6 | DR6_FIXED_1; | |
2815 | kvm_queue_exception(vcpu, DB_VECTOR); | |
2816 | return 1; | |
2817 | } | |
2818 | kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1; | |
2819 | kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7); | |
2820 | /* fall through */ | |
2821 | case BP_VECTOR: | |
6aa8b732 | 2822 | kvm_run->exit_reason = KVM_EXIT_DEBUG; |
d0bfb940 JK |
2823 | kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip; |
2824 | kvm_run->debug.arch.exception = ex_no; | |
42dbaa5a JK |
2825 | break; |
2826 | default: | |
d0bfb940 JK |
2827 | kvm_run->exit_reason = KVM_EXIT_EXCEPTION; |
2828 | kvm_run->ex.exception = ex_no; | |
2829 | kvm_run->ex.error_code = error_code; | |
42dbaa5a | 2830 | break; |
6aa8b732 | 2831 | } |
6aa8b732 AK |
2832 | return 0; |
2833 | } | |
2834 | ||
851ba692 | 2835 | static int handle_external_interrupt(struct kvm_vcpu *vcpu) |
6aa8b732 | 2836 | { |
1165f5fe | 2837 | ++vcpu->stat.irq_exits; |
6aa8b732 AK |
2838 | return 1; |
2839 | } | |
2840 | ||
851ba692 | 2841 | static int handle_triple_fault(struct kvm_vcpu *vcpu) |
988ad74f | 2842 | { |
851ba692 | 2843 | vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN; |
988ad74f AK |
2844 | return 0; |
2845 | } | |
6aa8b732 | 2846 | |
851ba692 | 2847 | static int handle_io(struct kvm_vcpu *vcpu) |
6aa8b732 | 2848 | { |
bfdaab09 | 2849 | unsigned long exit_qualification; |
34c33d16 | 2850 | int size, in, string; |
039576c0 | 2851 | unsigned port; |
6aa8b732 | 2852 | |
1165f5fe | 2853 | ++vcpu->stat.io_exits; |
bfdaab09 | 2854 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
039576c0 | 2855 | string = (exit_qualification & 16) != 0; |
e70669ab LV |
2856 | |
2857 | if (string) { | |
851ba692 | 2858 | if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DO_MMIO) |
e70669ab LV |
2859 | return 0; |
2860 | return 1; | |
2861 | } | |
2862 | ||
2863 | size = (exit_qualification & 7) + 1; | |
2864 | in = (exit_qualification & 8) != 0; | |
039576c0 | 2865 | port = exit_qualification >> 16; |
e70669ab | 2866 | |
e93f36bc | 2867 | skip_emulated_instruction(vcpu); |
851ba692 | 2868 | return kvm_emulate_pio(vcpu, in, size, port); |
6aa8b732 AK |
2869 | } |
2870 | ||
102d8325 IM |
2871 | static void |
2872 | vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall) | |
2873 | { | |
2874 | /* | |
2875 | * Patch in the VMCALL instruction: | |
2876 | */ | |
2877 | hypercall[0] = 0x0f; | |
2878 | hypercall[1] = 0x01; | |
2879 | hypercall[2] = 0xc1; | |
102d8325 IM |
2880 | } |
2881 | ||
851ba692 | 2882 | static int handle_cr(struct kvm_vcpu *vcpu) |
6aa8b732 | 2883 | { |
229456fc | 2884 | unsigned long exit_qualification, val; |
6aa8b732 AK |
2885 | int cr; |
2886 | int reg; | |
2887 | ||
bfdaab09 | 2888 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
6aa8b732 AK |
2889 | cr = exit_qualification & 15; |
2890 | reg = (exit_qualification >> 8) & 15; | |
2891 | switch ((exit_qualification >> 4) & 3) { | |
2892 | case 0: /* mov to cr */ | |
229456fc MT |
2893 | val = kvm_register_read(vcpu, reg); |
2894 | trace_kvm_cr_write(cr, val); | |
6aa8b732 AK |
2895 | switch (cr) { |
2896 | case 0: | |
229456fc | 2897 | kvm_set_cr0(vcpu, val); |
6aa8b732 AK |
2898 | skip_emulated_instruction(vcpu); |
2899 | return 1; | |
2900 | case 3: | |
229456fc | 2901 | kvm_set_cr3(vcpu, val); |
6aa8b732 AK |
2902 | skip_emulated_instruction(vcpu); |
2903 | return 1; | |
2904 | case 4: | |
229456fc | 2905 | kvm_set_cr4(vcpu, val); |
6aa8b732 AK |
2906 | skip_emulated_instruction(vcpu); |
2907 | return 1; | |
0a5fff19 GN |
2908 | case 8: { |
2909 | u8 cr8_prev = kvm_get_cr8(vcpu); | |
2910 | u8 cr8 = kvm_register_read(vcpu, reg); | |
2911 | kvm_set_cr8(vcpu, cr8); | |
2912 | skip_emulated_instruction(vcpu); | |
2913 | if (irqchip_in_kernel(vcpu->kvm)) | |
2914 | return 1; | |
2915 | if (cr8_prev <= cr8) | |
2916 | return 1; | |
851ba692 | 2917 | vcpu->run->exit_reason = KVM_EXIT_SET_TPR; |
0a5fff19 GN |
2918 | return 0; |
2919 | } | |
6aa8b732 AK |
2920 | }; |
2921 | break; | |
25c4c276 | 2922 | case 2: /* clts */ |
5fd86fcf | 2923 | vmx_fpu_deactivate(vcpu); |
ad312c7c ZX |
2924 | vcpu->arch.cr0 &= ~X86_CR0_TS; |
2925 | vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0); | |
5fd86fcf | 2926 | vmx_fpu_activate(vcpu); |
25c4c276 AL |
2927 | skip_emulated_instruction(vcpu); |
2928 | return 1; | |
6aa8b732 AK |
2929 | case 1: /*mov from cr*/ |
2930 | switch (cr) { | |
2931 | case 3: | |
5fdbf976 | 2932 | kvm_register_write(vcpu, reg, vcpu->arch.cr3); |
229456fc | 2933 | trace_kvm_cr_read(cr, vcpu->arch.cr3); |
6aa8b732 AK |
2934 | skip_emulated_instruction(vcpu); |
2935 | return 1; | |
2936 | case 8: | |
229456fc MT |
2937 | val = kvm_get_cr8(vcpu); |
2938 | kvm_register_write(vcpu, reg, val); | |
2939 | trace_kvm_cr_read(cr, val); | |
6aa8b732 AK |
2940 | skip_emulated_instruction(vcpu); |
2941 | return 1; | |
2942 | } | |
2943 | break; | |
2944 | case 3: /* lmsw */ | |
2d3ad1f4 | 2945 | kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f); |
6aa8b732 AK |
2946 | |
2947 | skip_emulated_instruction(vcpu); | |
2948 | return 1; | |
2949 | default: | |
2950 | break; | |
2951 | } | |
851ba692 | 2952 | vcpu->run->exit_reason = 0; |
f0242478 | 2953 | pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n", |
6aa8b732 AK |
2954 | (int)(exit_qualification >> 4) & 3, cr); |
2955 | return 0; | |
2956 | } | |
2957 | ||
851ba692 | 2958 | static int handle_dr(struct kvm_vcpu *vcpu) |
6aa8b732 | 2959 | { |
bfdaab09 | 2960 | unsigned long exit_qualification; |
6aa8b732 AK |
2961 | unsigned long val; |
2962 | int dr, reg; | |
2963 | ||
0a79b009 AK |
2964 | if (!kvm_require_cpl(vcpu, 0)) |
2965 | return 1; | |
42dbaa5a JK |
2966 | dr = vmcs_readl(GUEST_DR7); |
2967 | if (dr & DR7_GD) { | |
2968 | /* | |
2969 | * As the vm-exit takes precedence over the debug trap, we | |
2970 | * need to emulate the latter, either for the host or the | |
2971 | * guest debugging itself. | |
2972 | */ | |
2973 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) { | |
851ba692 AK |
2974 | vcpu->run->debug.arch.dr6 = vcpu->arch.dr6; |
2975 | vcpu->run->debug.arch.dr7 = dr; | |
2976 | vcpu->run->debug.arch.pc = | |
42dbaa5a JK |
2977 | vmcs_readl(GUEST_CS_BASE) + |
2978 | vmcs_readl(GUEST_RIP); | |
851ba692 AK |
2979 | vcpu->run->debug.arch.exception = DB_VECTOR; |
2980 | vcpu->run->exit_reason = KVM_EXIT_DEBUG; | |
42dbaa5a JK |
2981 | return 0; |
2982 | } else { | |
2983 | vcpu->arch.dr7 &= ~DR7_GD; | |
2984 | vcpu->arch.dr6 |= DR6_BD; | |
2985 | vmcs_writel(GUEST_DR7, vcpu->arch.dr7); | |
2986 | kvm_queue_exception(vcpu, DB_VECTOR); | |
2987 | return 1; | |
2988 | } | |
2989 | } | |
2990 | ||
bfdaab09 | 2991 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
42dbaa5a JK |
2992 | dr = exit_qualification & DEBUG_REG_ACCESS_NUM; |
2993 | reg = DEBUG_REG_ACCESS_REG(exit_qualification); | |
2994 | if (exit_qualification & TYPE_MOV_FROM_DR) { | |
6aa8b732 | 2995 | switch (dr) { |
42dbaa5a JK |
2996 | case 0 ... 3: |
2997 | val = vcpu->arch.db[dr]; | |
2998 | break; | |
6aa8b732 | 2999 | case 6: |
42dbaa5a | 3000 | val = vcpu->arch.dr6; |
6aa8b732 AK |
3001 | break; |
3002 | case 7: | |
42dbaa5a | 3003 | val = vcpu->arch.dr7; |
6aa8b732 AK |
3004 | break; |
3005 | default: | |
3006 | val = 0; | |
3007 | } | |
5fdbf976 | 3008 | kvm_register_write(vcpu, reg, val); |
6aa8b732 | 3009 | } else { |
42dbaa5a JK |
3010 | val = vcpu->arch.regs[reg]; |
3011 | switch (dr) { | |
3012 | case 0 ... 3: | |
3013 | vcpu->arch.db[dr] = val; | |
3014 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) | |
3015 | vcpu->arch.eff_db[dr] = val; | |
3016 | break; | |
3017 | case 4 ... 5: | |
3018 | if (vcpu->arch.cr4 & X86_CR4_DE) | |
3019 | kvm_queue_exception(vcpu, UD_VECTOR); | |
3020 | break; | |
3021 | case 6: | |
3022 | if (val & 0xffffffff00000000ULL) { | |
3023 | kvm_queue_exception(vcpu, GP_VECTOR); | |
3024 | break; | |
3025 | } | |
3026 | vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1; | |
3027 | break; | |
3028 | case 7: | |
3029 | if (val & 0xffffffff00000000ULL) { | |
3030 | kvm_queue_exception(vcpu, GP_VECTOR); | |
3031 | break; | |
3032 | } | |
3033 | vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1; | |
3034 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) { | |
3035 | vmcs_writel(GUEST_DR7, vcpu->arch.dr7); | |
3036 | vcpu->arch.switch_db_regs = | |
3037 | (val & DR7_BP_EN_MASK); | |
3038 | } | |
3039 | break; | |
3040 | } | |
6aa8b732 | 3041 | } |
6aa8b732 AK |
3042 | skip_emulated_instruction(vcpu); |
3043 | return 1; | |
3044 | } | |
3045 | ||
851ba692 | 3046 | static int handle_cpuid(struct kvm_vcpu *vcpu) |
6aa8b732 | 3047 | { |
06465c5a AK |
3048 | kvm_emulate_cpuid(vcpu); |
3049 | return 1; | |
6aa8b732 AK |
3050 | } |
3051 | ||
851ba692 | 3052 | static int handle_rdmsr(struct kvm_vcpu *vcpu) |
6aa8b732 | 3053 | { |
ad312c7c | 3054 | u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX]; |
6aa8b732 AK |
3055 | u64 data; |
3056 | ||
3057 | if (vmx_get_msr(vcpu, ecx, &data)) { | |
c1a5d4f9 | 3058 | kvm_inject_gp(vcpu, 0); |
6aa8b732 AK |
3059 | return 1; |
3060 | } | |
3061 | ||
229456fc | 3062 | trace_kvm_msr_read(ecx, data); |
2714d1d3 | 3063 | |
6aa8b732 | 3064 | /* FIXME: handling of bits 32:63 of rax, rdx */ |
ad312c7c ZX |
3065 | vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u; |
3066 | vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u; | |
6aa8b732 AK |
3067 | skip_emulated_instruction(vcpu); |
3068 | return 1; | |
3069 | } | |
3070 | ||
851ba692 | 3071 | static int handle_wrmsr(struct kvm_vcpu *vcpu) |
6aa8b732 | 3072 | { |
ad312c7c ZX |
3073 | u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX]; |
3074 | u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u) | |
3075 | | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32); | |
6aa8b732 | 3076 | |
229456fc | 3077 | trace_kvm_msr_write(ecx, data); |
2714d1d3 | 3078 | |
6aa8b732 | 3079 | if (vmx_set_msr(vcpu, ecx, data) != 0) { |
c1a5d4f9 | 3080 | kvm_inject_gp(vcpu, 0); |
6aa8b732 AK |
3081 | return 1; |
3082 | } | |
3083 | ||
3084 | skip_emulated_instruction(vcpu); | |
3085 | return 1; | |
3086 | } | |
3087 | ||
851ba692 | 3088 | static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu) |
6e5d865c YS |
3089 | { |
3090 | return 1; | |
3091 | } | |
3092 | ||
851ba692 | 3093 | static int handle_interrupt_window(struct kvm_vcpu *vcpu) |
6aa8b732 | 3094 | { |
85f455f7 ED |
3095 | u32 cpu_based_vm_exec_control; |
3096 | ||
3097 | /* clear pending irq */ | |
3098 | cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); | |
3099 | cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING; | |
3100 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); | |
2714d1d3 | 3101 | |
a26bf12a | 3102 | ++vcpu->stat.irq_window_exits; |
2714d1d3 | 3103 | |
c1150d8c DL |
3104 | /* |
3105 | * If the user space waits to inject interrupts, exit as soon as | |
3106 | * possible | |
3107 | */ | |
8061823a | 3108 | if (!irqchip_in_kernel(vcpu->kvm) && |
851ba692 | 3109 | vcpu->run->request_interrupt_window && |
8061823a | 3110 | !kvm_cpu_has_interrupt(vcpu)) { |
851ba692 | 3111 | vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN; |
c1150d8c DL |
3112 | return 0; |
3113 | } | |
6aa8b732 AK |
3114 | return 1; |
3115 | } | |
3116 | ||
851ba692 | 3117 | static int handle_halt(struct kvm_vcpu *vcpu) |
6aa8b732 AK |
3118 | { |
3119 | skip_emulated_instruction(vcpu); | |
d3bef15f | 3120 | return kvm_emulate_halt(vcpu); |
6aa8b732 AK |
3121 | } |
3122 | ||
851ba692 | 3123 | static int handle_vmcall(struct kvm_vcpu *vcpu) |
c21415e8 | 3124 | { |
510043da | 3125 | skip_emulated_instruction(vcpu); |
7aa81cc0 AL |
3126 | kvm_emulate_hypercall(vcpu); |
3127 | return 1; | |
c21415e8 IM |
3128 | } |
3129 | ||
851ba692 | 3130 | static int handle_vmx_insn(struct kvm_vcpu *vcpu) |
e3c7cb6a AK |
3131 | { |
3132 | kvm_queue_exception(vcpu, UD_VECTOR); | |
3133 | return 1; | |
3134 | } | |
3135 | ||
851ba692 | 3136 | static int handle_invlpg(struct kvm_vcpu *vcpu) |
a7052897 | 3137 | { |
f9c617f6 | 3138 | unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
a7052897 MT |
3139 | |
3140 | kvm_mmu_invlpg(vcpu, exit_qualification); | |
3141 | skip_emulated_instruction(vcpu); | |
3142 | return 1; | |
3143 | } | |
3144 | ||
851ba692 | 3145 | static int handle_wbinvd(struct kvm_vcpu *vcpu) |
e5edaa01 ED |
3146 | { |
3147 | skip_emulated_instruction(vcpu); | |
3148 | /* TODO: Add support for VT-d/pass-through device */ | |
3149 | return 1; | |
3150 | } | |
3151 | ||
851ba692 | 3152 | static int handle_apic_access(struct kvm_vcpu *vcpu) |
f78e0e2e | 3153 | { |
f9c617f6 | 3154 | unsigned long exit_qualification; |
f78e0e2e SY |
3155 | enum emulation_result er; |
3156 | unsigned long offset; | |
3157 | ||
f9c617f6 | 3158 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
f78e0e2e SY |
3159 | offset = exit_qualification & 0xffful; |
3160 | ||
851ba692 | 3161 | er = emulate_instruction(vcpu, 0, 0, 0); |
f78e0e2e SY |
3162 | |
3163 | if (er != EMULATE_DONE) { | |
3164 | printk(KERN_ERR | |
3165 | "Fail to handle apic access vmexit! Offset is 0x%lx\n", | |
3166 | offset); | |
7f582ab6 | 3167 | return -ENOEXEC; |
f78e0e2e SY |
3168 | } |
3169 | return 1; | |
3170 | } | |
3171 | ||
851ba692 | 3172 | static int handle_task_switch(struct kvm_vcpu *vcpu) |
37817f29 | 3173 | { |
60637aac | 3174 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
37817f29 IE |
3175 | unsigned long exit_qualification; |
3176 | u16 tss_selector; | |
64a7ec06 GN |
3177 | int reason, type, idt_v; |
3178 | ||
3179 | idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK); | |
3180 | type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK); | |
37817f29 IE |
3181 | |
3182 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); | |
3183 | ||
3184 | reason = (u32)exit_qualification >> 30; | |
64a7ec06 GN |
3185 | if (reason == TASK_SWITCH_GATE && idt_v) { |
3186 | switch (type) { | |
3187 | case INTR_TYPE_NMI_INTR: | |
3188 | vcpu->arch.nmi_injected = false; | |
3189 | if (cpu_has_virtual_nmis()) | |
3190 | vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, | |
3191 | GUEST_INTR_STATE_NMI); | |
3192 | break; | |
3193 | case INTR_TYPE_EXT_INTR: | |
66fd3f7f | 3194 | case INTR_TYPE_SOFT_INTR: |
64a7ec06 GN |
3195 | kvm_clear_interrupt_queue(vcpu); |
3196 | break; | |
3197 | case INTR_TYPE_HARD_EXCEPTION: | |
3198 | case INTR_TYPE_SOFT_EXCEPTION: | |
3199 | kvm_clear_exception_queue(vcpu); | |
3200 | break; | |
3201 | default: | |
3202 | break; | |
3203 | } | |
60637aac | 3204 | } |
37817f29 IE |
3205 | tss_selector = exit_qualification; |
3206 | ||
64a7ec06 GN |
3207 | if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION && |
3208 | type != INTR_TYPE_EXT_INTR && | |
3209 | type != INTR_TYPE_NMI_INTR)) | |
3210 | skip_emulated_instruction(vcpu); | |
3211 | ||
42dbaa5a JK |
3212 | if (!kvm_task_switch(vcpu, tss_selector, reason)) |
3213 | return 0; | |
3214 | ||
3215 | /* clear all local breakpoint enable flags */ | |
3216 | vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55); | |
3217 | ||
3218 | /* | |
3219 | * TODO: What about debug traps on tss switch? | |
3220 | * Are we supposed to inject them and update dr6? | |
3221 | */ | |
3222 | ||
3223 | return 1; | |
37817f29 IE |
3224 | } |
3225 | ||
851ba692 | 3226 | static int handle_ept_violation(struct kvm_vcpu *vcpu) |
1439442c | 3227 | { |
f9c617f6 | 3228 | unsigned long exit_qualification; |
1439442c | 3229 | gpa_t gpa; |
1439442c | 3230 | int gla_validity; |
1439442c | 3231 | |
f9c617f6 | 3232 | exit_qualification = vmcs_readl(EXIT_QUALIFICATION); |
1439442c SY |
3233 | |
3234 | if (exit_qualification & (1 << 6)) { | |
3235 | printk(KERN_ERR "EPT: GPA exceeds GAW!\n"); | |
7f582ab6 | 3236 | return -EINVAL; |
1439442c SY |
3237 | } |
3238 | ||
3239 | gla_validity = (exit_qualification >> 7) & 0x3; | |
3240 | if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) { | |
3241 | printk(KERN_ERR "EPT: Handling EPT violation failed!\n"); | |
3242 | printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n", | |
3243 | (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS), | |
f9c617f6 | 3244 | vmcs_readl(GUEST_LINEAR_ADDRESS)); |
1439442c SY |
3245 | printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n", |
3246 | (long unsigned int)exit_qualification); | |
851ba692 AK |
3247 | vcpu->run->exit_reason = KVM_EXIT_UNKNOWN; |
3248 | vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION; | |
596ae895 | 3249 | return 0; |
1439442c SY |
3250 | } |
3251 | ||
3252 | gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS); | |
229456fc | 3253 | trace_kvm_page_fault(gpa, exit_qualification); |
49cd7d22 | 3254 | return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0); |
1439442c SY |
3255 | } |
3256 | ||
68f89400 MT |
3257 | static u64 ept_rsvd_mask(u64 spte, int level) |
3258 | { | |
3259 | int i; | |
3260 | u64 mask = 0; | |
3261 | ||
3262 | for (i = 51; i > boot_cpu_data.x86_phys_bits; i--) | |
3263 | mask |= (1ULL << i); | |
3264 | ||
3265 | if (level > 2) | |
3266 | /* bits 7:3 reserved */ | |
3267 | mask |= 0xf8; | |
3268 | else if (level == 2) { | |
3269 | if (spte & (1ULL << 7)) | |
3270 | /* 2MB ref, bits 20:12 reserved */ | |
3271 | mask |= 0x1ff000; | |
3272 | else | |
3273 | /* bits 6:3 reserved */ | |
3274 | mask |= 0x78; | |
3275 | } | |
3276 | ||
3277 | return mask; | |
3278 | } | |
3279 | ||
3280 | static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte, | |
3281 | int level) | |
3282 | { | |
3283 | printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level); | |
3284 | ||
3285 | /* 010b (write-only) */ | |
3286 | WARN_ON((spte & 0x7) == 0x2); | |
3287 | ||
3288 | /* 110b (write/execute) */ | |
3289 | WARN_ON((spte & 0x7) == 0x6); | |
3290 | ||
3291 | /* 100b (execute-only) and value not supported by logical processor */ | |
3292 | if (!cpu_has_vmx_ept_execute_only()) | |
3293 | WARN_ON((spte & 0x7) == 0x4); | |
3294 | ||
3295 | /* not 000b */ | |
3296 | if ((spte & 0x7)) { | |
3297 | u64 rsvd_bits = spte & ept_rsvd_mask(spte, level); | |
3298 | ||
3299 | if (rsvd_bits != 0) { | |
3300 | printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n", | |
3301 | __func__, rsvd_bits); | |
3302 | WARN_ON(1); | |
3303 | } | |
3304 | ||
3305 | if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) { | |
3306 | u64 ept_mem_type = (spte & 0x38) >> 3; | |
3307 | ||
3308 | if (ept_mem_type == 2 || ept_mem_type == 3 || | |
3309 | ept_mem_type == 7) { | |
3310 | printk(KERN_ERR "%s: ept_mem_type=0x%llx\n", | |
3311 | __func__, ept_mem_type); | |
3312 | WARN_ON(1); | |
3313 | } | |
3314 | } | |
3315 | } | |
3316 | } | |
3317 | ||
851ba692 | 3318 | static int handle_ept_misconfig(struct kvm_vcpu *vcpu) |
68f89400 MT |
3319 | { |
3320 | u64 sptes[4]; | |
3321 | int nr_sptes, i; | |
3322 | gpa_t gpa; | |
3323 | ||
3324 | gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS); | |
3325 | ||
3326 | printk(KERN_ERR "EPT: Misconfiguration.\n"); | |
3327 | printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa); | |
3328 | ||
3329 | nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes); | |
3330 | ||
3331 | for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i) | |
3332 | ept_misconfig_inspect_spte(vcpu, sptes[i-1], i); | |
3333 | ||
851ba692 AK |
3334 | vcpu->run->exit_reason = KVM_EXIT_UNKNOWN; |
3335 | vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG; | |
68f89400 MT |
3336 | |
3337 | return 0; | |
3338 | } | |
3339 | ||
851ba692 | 3340 | static int handle_nmi_window(struct kvm_vcpu *vcpu) |
f08864b4 SY |
3341 | { |
3342 | u32 cpu_based_vm_exec_control; | |
3343 | ||
3344 | /* clear pending NMI */ | |
3345 | cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); | |
3346 | cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING; | |
3347 | vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); | |
3348 | ++vcpu->stat.nmi_window_exits; | |
3349 | ||
3350 | return 1; | |
3351 | } | |
3352 | ||
80ced186 | 3353 | static int handle_invalid_guest_state(struct kvm_vcpu *vcpu) |
ea953ef0 | 3354 | { |
8b3079a5 AK |
3355 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
3356 | enum emulation_result err = EMULATE_DONE; | |
80ced186 | 3357 | int ret = 1; |
ea953ef0 MG |
3358 | |
3359 | while (!guest_state_valid(vcpu)) { | |
851ba692 | 3360 | err = emulate_instruction(vcpu, 0, 0, 0); |
ea953ef0 | 3361 | |
80ced186 MG |
3362 | if (err == EMULATE_DO_MMIO) { |
3363 | ret = 0; | |
3364 | goto out; | |
3365 | } | |
1d5a4d9b GT |
3366 | |
3367 | if (err != EMULATE_DONE) { | |
3368 | kvm_report_emulation_failure(vcpu, "emulation failure"); | |
80ced186 MG |
3369 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; |
3370 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
3371 | ret = 0; | |
3372 | goto out; | |
ea953ef0 MG |
3373 | } |
3374 | ||
3375 | if (signal_pending(current)) | |
80ced186 | 3376 | goto out; |
ea953ef0 MG |
3377 | if (need_resched()) |
3378 | schedule(); | |
3379 | } | |
3380 | ||
80ced186 MG |
3381 | vmx->emulation_required = 0; |
3382 | out: | |
3383 | return ret; | |
ea953ef0 MG |
3384 | } |
3385 | ||
4b8d54f9 ZE |
3386 | /* |
3387 | * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE | |
3388 | * exiting, so only get here on cpu with PAUSE-Loop-Exiting. | |
3389 | */ | |
3390 | static int handle_pause(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) | |
3391 | { | |
3392 | skip_emulated_instruction(vcpu); | |
3393 | kvm_vcpu_on_spin(vcpu); | |
3394 | ||
3395 | return 1; | |
3396 | } | |
3397 | ||
6aa8b732 AK |
3398 | /* |
3399 | * The exit handlers return 1 if the exit was handled fully and guest execution | |
3400 | * may resume. Otherwise they set the kvm_run parameter to indicate what needs | |
3401 | * to be done to userspace and return 0. | |
3402 | */ | |
851ba692 | 3403 | static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = { |
6aa8b732 AK |
3404 | [EXIT_REASON_EXCEPTION_NMI] = handle_exception, |
3405 | [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt, | |
988ad74f | 3406 | [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault, |
f08864b4 | 3407 | [EXIT_REASON_NMI_WINDOW] = handle_nmi_window, |
6aa8b732 | 3408 | [EXIT_REASON_IO_INSTRUCTION] = handle_io, |
6aa8b732 AK |
3409 | [EXIT_REASON_CR_ACCESS] = handle_cr, |
3410 | [EXIT_REASON_DR_ACCESS] = handle_dr, | |
3411 | [EXIT_REASON_CPUID] = handle_cpuid, | |
3412 | [EXIT_REASON_MSR_READ] = handle_rdmsr, | |
3413 | [EXIT_REASON_MSR_WRITE] = handle_wrmsr, | |
3414 | [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window, | |
3415 | [EXIT_REASON_HLT] = handle_halt, | |
a7052897 | 3416 | [EXIT_REASON_INVLPG] = handle_invlpg, |
c21415e8 | 3417 | [EXIT_REASON_VMCALL] = handle_vmcall, |
e3c7cb6a AK |
3418 | [EXIT_REASON_VMCLEAR] = handle_vmx_insn, |
3419 | [EXIT_REASON_VMLAUNCH] = handle_vmx_insn, | |
3420 | [EXIT_REASON_VMPTRLD] = handle_vmx_insn, | |
3421 | [EXIT_REASON_VMPTRST] = handle_vmx_insn, | |
3422 | [EXIT_REASON_VMREAD] = handle_vmx_insn, | |
3423 | [EXIT_REASON_VMRESUME] = handle_vmx_insn, | |
3424 | [EXIT_REASON_VMWRITE] = handle_vmx_insn, | |
3425 | [EXIT_REASON_VMOFF] = handle_vmx_insn, | |
3426 | [EXIT_REASON_VMON] = handle_vmx_insn, | |
f78e0e2e SY |
3427 | [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold, |
3428 | [EXIT_REASON_APIC_ACCESS] = handle_apic_access, | |
e5edaa01 | 3429 | [EXIT_REASON_WBINVD] = handle_wbinvd, |
37817f29 | 3430 | [EXIT_REASON_TASK_SWITCH] = handle_task_switch, |
a0861c02 | 3431 | [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check, |
68f89400 MT |
3432 | [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation, |
3433 | [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig, | |
4b8d54f9 | 3434 | [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause, |
6aa8b732 AK |
3435 | }; |
3436 | ||
3437 | static const int kvm_vmx_max_exit_handlers = | |
50a3485c | 3438 | ARRAY_SIZE(kvm_vmx_exit_handlers); |
6aa8b732 AK |
3439 | |
3440 | /* | |
3441 | * The guest has exited. See if we can fix it or if we need userspace | |
3442 | * assistance. | |
3443 | */ | |
851ba692 | 3444 | static int vmx_handle_exit(struct kvm_vcpu *vcpu) |
6aa8b732 | 3445 | { |
29bd8a78 | 3446 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
a0861c02 | 3447 | u32 exit_reason = vmx->exit_reason; |
1155f76a | 3448 | u32 vectoring_info = vmx->idt_vectoring_info; |
29bd8a78 | 3449 | |
229456fc | 3450 | trace_kvm_exit(exit_reason, kvm_rip_read(vcpu)); |
2714d1d3 | 3451 | |
80ced186 MG |
3452 | /* If guest state is invalid, start emulating */ |
3453 | if (vmx->emulation_required && emulate_invalid_guest_state) | |
3454 | return handle_invalid_guest_state(vcpu); | |
1d5a4d9b | 3455 | |
1439442c SY |
3456 | /* Access CR3 don't cause VMExit in paging mode, so we need |
3457 | * to sync with guest real CR3. */ | |
6de4f3ad | 3458 | if (enable_ept && is_paging(vcpu)) |
1439442c | 3459 | vcpu->arch.cr3 = vmcs_readl(GUEST_CR3); |
1439442c | 3460 | |
29bd8a78 | 3461 | if (unlikely(vmx->fail)) { |
851ba692 AK |
3462 | vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY; |
3463 | vcpu->run->fail_entry.hardware_entry_failure_reason | |
29bd8a78 AK |
3464 | = vmcs_read32(VM_INSTRUCTION_ERROR); |
3465 | return 0; | |
3466 | } | |
6aa8b732 | 3467 | |
d77c26fc | 3468 | if ((vectoring_info & VECTORING_INFO_VALID_MASK) && |
1439442c | 3469 | (exit_reason != EXIT_REASON_EXCEPTION_NMI && |
60637aac JK |
3470 | exit_reason != EXIT_REASON_EPT_VIOLATION && |
3471 | exit_reason != EXIT_REASON_TASK_SWITCH)) | |
3472 | printk(KERN_WARNING "%s: unexpected, valid vectoring info " | |
3473 | "(0x%x) and exit reason is 0x%x\n", | |
3474 | __func__, vectoring_info, exit_reason); | |
3b86cd99 JK |
3475 | |
3476 | if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) { | |
c4282df9 | 3477 | if (vmx_interrupt_allowed(vcpu)) { |
3b86cd99 | 3478 | vmx->soft_vnmi_blocked = 0; |
3b86cd99 | 3479 | } else if (vmx->vnmi_blocked_time > 1000000000LL && |
4531220b | 3480 | vcpu->arch.nmi_pending) { |
3b86cd99 JK |
3481 | /* |
3482 | * This CPU don't support us in finding the end of an | |
3483 | * NMI-blocked window if the guest runs with IRQs | |
3484 | * disabled. So we pull the trigger after 1 s of | |
3485 | * futile waiting, but inform the user about this. | |
3486 | */ | |
3487 | printk(KERN_WARNING "%s: Breaking out of NMI-blocked " | |
3488 | "state on VCPU %d after 1 s timeout\n", | |
3489 | __func__, vcpu->vcpu_id); | |
3490 | vmx->soft_vnmi_blocked = 0; | |
3b86cd99 | 3491 | } |
3b86cd99 JK |
3492 | } |
3493 | ||
6aa8b732 AK |
3494 | if (exit_reason < kvm_vmx_max_exit_handlers |
3495 | && kvm_vmx_exit_handlers[exit_reason]) | |
851ba692 | 3496 | return kvm_vmx_exit_handlers[exit_reason](vcpu); |
6aa8b732 | 3497 | else { |
851ba692 AK |
3498 | vcpu->run->exit_reason = KVM_EXIT_UNKNOWN; |
3499 | vcpu->run->hw.hardware_exit_reason = exit_reason; | |
6aa8b732 AK |
3500 | } |
3501 | return 0; | |
3502 | } | |
3503 | ||
95ba8273 | 3504 | static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr) |
6e5d865c | 3505 | { |
95ba8273 | 3506 | if (irr == -1 || tpr < irr) { |
6e5d865c YS |
3507 | vmcs_write32(TPR_THRESHOLD, 0); |
3508 | return; | |
3509 | } | |
3510 | ||
95ba8273 | 3511 | vmcs_write32(TPR_THRESHOLD, irr); |
6e5d865c YS |
3512 | } |
3513 | ||
cf393f75 AK |
3514 | static void vmx_complete_interrupts(struct vcpu_vmx *vmx) |
3515 | { | |
3516 | u32 exit_intr_info; | |
7b4a25cb | 3517 | u32 idt_vectoring_info = vmx->idt_vectoring_info; |
cf393f75 AK |
3518 | bool unblock_nmi; |
3519 | u8 vector; | |
668f612f AK |
3520 | int type; |
3521 | bool idtv_info_valid; | |
cf393f75 AK |
3522 | |
3523 | exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO); | |
20f65983 | 3524 | |
a0861c02 AK |
3525 | vmx->exit_reason = vmcs_read32(VM_EXIT_REASON); |
3526 | ||
3527 | /* Handle machine checks before interrupts are enabled */ | |
3528 | if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY) | |
3529 | || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI | |
3530 | && is_machine_check(exit_intr_info))) | |
3531 | kvm_machine_check(); | |
3532 | ||
20f65983 GN |
3533 | /* We need to handle NMIs before interrupts are enabled */ |
3534 | if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR && | |
229456fc | 3535 | (exit_intr_info & INTR_INFO_VALID_MASK)) |
20f65983 | 3536 | asm("int $2"); |
20f65983 GN |
3537 | |
3538 | idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK; | |
3539 | ||
cf393f75 AK |
3540 | if (cpu_has_virtual_nmis()) { |
3541 | unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0; | |
3542 | vector = exit_intr_info & INTR_INFO_VECTOR_MASK; | |
3543 | /* | |
7b4a25cb | 3544 | * SDM 3: 27.7.1.2 (September 2008) |
cf393f75 AK |
3545 | * Re-set bit "block by NMI" before VM entry if vmexit caused by |
3546 | * a guest IRET fault. | |
7b4a25cb GN |
3547 | * SDM 3: 23.2.2 (September 2008) |
3548 | * Bit 12 is undefined in any of the following cases: | |
3549 | * If the VM exit sets the valid bit in the IDT-vectoring | |
3550 | * information field. | |
3551 | * If the VM exit is due to a double fault. | |
cf393f75 | 3552 | */ |
7b4a25cb GN |
3553 | if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi && |
3554 | vector != DF_VECTOR && !idtv_info_valid) | |
cf393f75 AK |
3555 | vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, |
3556 | GUEST_INTR_STATE_NMI); | |
3b86cd99 JK |
3557 | } else if (unlikely(vmx->soft_vnmi_blocked)) |
3558 | vmx->vnmi_blocked_time += | |
3559 | ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time)); | |
668f612f | 3560 | |
37b96e98 GN |
3561 | vmx->vcpu.arch.nmi_injected = false; |
3562 | kvm_clear_exception_queue(&vmx->vcpu); | |
3563 | kvm_clear_interrupt_queue(&vmx->vcpu); | |
3564 | ||
3565 | if (!idtv_info_valid) | |
3566 | return; | |
3567 | ||
668f612f AK |
3568 | vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK; |
3569 | type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK; | |
37b96e98 | 3570 | |
64a7ec06 | 3571 | switch (type) { |
37b96e98 GN |
3572 | case INTR_TYPE_NMI_INTR: |
3573 | vmx->vcpu.arch.nmi_injected = true; | |
668f612f | 3574 | /* |
7b4a25cb | 3575 | * SDM 3: 27.7.1.2 (September 2008) |
37b96e98 GN |
3576 | * Clear bit "block by NMI" before VM entry if a NMI |
3577 | * delivery faulted. | |
668f612f | 3578 | */ |
37b96e98 GN |
3579 | vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO, |
3580 | GUEST_INTR_STATE_NMI); | |
3581 | break; | |
37b96e98 | 3582 | case INTR_TYPE_SOFT_EXCEPTION: |
66fd3f7f GN |
3583 | vmx->vcpu.arch.event_exit_inst_len = |
3584 | vmcs_read32(VM_EXIT_INSTRUCTION_LEN); | |
3585 | /* fall through */ | |
3586 | case INTR_TYPE_HARD_EXCEPTION: | |
35920a35 | 3587 | if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) { |
37b96e98 GN |
3588 | u32 err = vmcs_read32(IDT_VECTORING_ERROR_CODE); |
3589 | kvm_queue_exception_e(&vmx->vcpu, vector, err); | |
35920a35 AK |
3590 | } else |
3591 | kvm_queue_exception(&vmx->vcpu, vector); | |
37b96e98 | 3592 | break; |
66fd3f7f GN |
3593 | case INTR_TYPE_SOFT_INTR: |
3594 | vmx->vcpu.arch.event_exit_inst_len = | |
3595 | vmcs_read32(VM_EXIT_INSTRUCTION_LEN); | |
3596 | /* fall through */ | |
37b96e98 | 3597 | case INTR_TYPE_EXT_INTR: |
66fd3f7f GN |
3598 | kvm_queue_interrupt(&vmx->vcpu, vector, |
3599 | type == INTR_TYPE_SOFT_INTR); | |
37b96e98 GN |
3600 | break; |
3601 | default: | |
3602 | break; | |
f7d9238f | 3603 | } |
cf393f75 AK |
3604 | } |
3605 | ||
9c8cba37 AK |
3606 | /* |
3607 | * Failure to inject an interrupt should give us the information | |
3608 | * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs | |
3609 | * when fetching the interrupt redirection bitmap in the real-mode | |
3610 | * tss, this doesn't happen. So we do it ourselves. | |
3611 | */ | |
3612 | static void fixup_rmode_irq(struct vcpu_vmx *vmx) | |
3613 | { | |
3614 | vmx->rmode.irq.pending = 0; | |
5fdbf976 | 3615 | if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip) |
9c8cba37 | 3616 | return; |
5fdbf976 | 3617 | kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip); |
9c8cba37 AK |
3618 | if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) { |
3619 | vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK; | |
3620 | vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR; | |
3621 | return; | |
3622 | } | |
3623 | vmx->idt_vectoring_info = | |
3624 | VECTORING_INFO_VALID_MASK | |
3625 | | INTR_TYPE_EXT_INTR | |
3626 | | vmx->rmode.irq.vector; | |
3627 | } | |
3628 | ||
c801949d AK |
3629 | #ifdef CONFIG_X86_64 |
3630 | #define R "r" | |
3631 | #define Q "q" | |
3632 | #else | |
3633 | #define R "e" | |
3634 | #define Q "l" | |
3635 | #endif | |
3636 | ||
851ba692 | 3637 | static void vmx_vcpu_run(struct kvm_vcpu *vcpu) |
6aa8b732 | 3638 | { |
a2fa3e9f | 3639 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
e6adf283 | 3640 | |
8f5d549f AK |
3641 | if (enable_ept && is_paging(vcpu)) { |
3642 | vmcs_writel(GUEST_CR3, vcpu->arch.cr3); | |
3643 | ept_load_pdptrs(vcpu); | |
3644 | } | |
3b86cd99 JK |
3645 | /* Record the guest's net vcpu time for enforced NMI injections. */ |
3646 | if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) | |
3647 | vmx->entry_time = ktime_get(); | |
3648 | ||
80ced186 MG |
3649 | /* Don't enter VMX if guest state is invalid, let the exit handler |
3650 | start emulation until we arrive back to a valid state */ | |
3651 | if (vmx->emulation_required && emulate_invalid_guest_state) | |
a89a8fb9 | 3652 | return; |
a89a8fb9 | 3653 | |
5fdbf976 MT |
3654 | if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty)) |
3655 | vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]); | |
3656 | if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty)) | |
3657 | vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]); | |
3658 | ||
787ff736 GN |
3659 | /* When single-stepping over STI and MOV SS, we must clear the |
3660 | * corresponding interruptibility bits in the guest state. Otherwise | |
3661 | * vmentry fails as it then expects bit 14 (BS) in pending debug | |
3662 | * exceptions being set, but that's not correct for the guest debugging | |
3663 | * case. */ | |
3664 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) | |
3665 | vmx_set_interrupt_shadow(vcpu, 0); | |
3666 | ||
e6adf283 AK |
3667 | /* |
3668 | * Loading guest fpu may have cleared host cr0.ts | |
3669 | */ | |
3670 | vmcs_writel(HOST_CR0, read_cr0()); | |
3671 | ||
e8a48342 AK |
3672 | if (vcpu->arch.switch_db_regs) |
3673 | set_debugreg(vcpu->arch.dr6, 6); | |
42dbaa5a | 3674 | |
d77c26fc | 3675 | asm( |
6aa8b732 | 3676 | /* Store host registers */ |
c801949d AK |
3677 | "push %%"R"dx; push %%"R"bp;" |
3678 | "push %%"R"cx \n\t" | |
313dbd49 AK |
3679 | "cmp %%"R"sp, %c[host_rsp](%0) \n\t" |
3680 | "je 1f \n\t" | |
3681 | "mov %%"R"sp, %c[host_rsp](%0) \n\t" | |
4ecac3fd | 3682 | __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t" |
313dbd49 | 3683 | "1: \n\t" |
d3edefc0 AK |
3684 | /* Reload cr2 if changed */ |
3685 | "mov %c[cr2](%0), %%"R"ax \n\t" | |
3686 | "mov %%cr2, %%"R"dx \n\t" | |
3687 | "cmp %%"R"ax, %%"R"dx \n\t" | |
3688 | "je 2f \n\t" | |
3689 | "mov %%"R"ax, %%cr2 \n\t" | |
3690 | "2: \n\t" | |
6aa8b732 | 3691 | /* Check if vmlaunch of vmresume is needed */ |
e08aa78a | 3692 | "cmpl $0, %c[launched](%0) \n\t" |
6aa8b732 | 3693 | /* Load guest registers. Don't clobber flags. */ |
c801949d AK |
3694 | "mov %c[rax](%0), %%"R"ax \n\t" |
3695 | "mov %c[rbx](%0), %%"R"bx \n\t" | |
3696 | "mov %c[rdx](%0), %%"R"dx \n\t" | |
3697 | "mov %c[rsi](%0), %%"R"si \n\t" | |
3698 | "mov %c[rdi](%0), %%"R"di \n\t" | |
3699 | "mov %c[rbp](%0), %%"R"bp \n\t" | |
05b3e0c2 | 3700 | #ifdef CONFIG_X86_64 |
e08aa78a AK |
3701 | "mov %c[r8](%0), %%r8 \n\t" |
3702 | "mov %c[r9](%0), %%r9 \n\t" | |
3703 | "mov %c[r10](%0), %%r10 \n\t" | |
3704 | "mov %c[r11](%0), %%r11 \n\t" | |
3705 | "mov %c[r12](%0), %%r12 \n\t" | |
3706 | "mov %c[r13](%0), %%r13 \n\t" | |
3707 | "mov %c[r14](%0), %%r14 \n\t" | |
3708 | "mov %c[r15](%0), %%r15 \n\t" | |
6aa8b732 | 3709 | #endif |
c801949d AK |
3710 | "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */ |
3711 | ||
6aa8b732 | 3712 | /* Enter guest mode */ |
cd2276a7 | 3713 | "jne .Llaunched \n\t" |
4ecac3fd | 3714 | __ex(ASM_VMX_VMLAUNCH) "\n\t" |
cd2276a7 | 3715 | "jmp .Lkvm_vmx_return \n\t" |
4ecac3fd | 3716 | ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t" |
cd2276a7 | 3717 | ".Lkvm_vmx_return: " |
6aa8b732 | 3718 | /* Save guest registers, load host registers, keep flags */ |
c801949d AK |
3719 | "xchg %0, (%%"R"sp) \n\t" |
3720 | "mov %%"R"ax, %c[rax](%0) \n\t" | |
3721 | "mov %%"R"bx, %c[rbx](%0) \n\t" | |
3722 | "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t" | |
3723 | "mov %%"R"dx, %c[rdx](%0) \n\t" | |
3724 | "mov %%"R"si, %c[rsi](%0) \n\t" | |
3725 | "mov %%"R"di, %c[rdi](%0) \n\t" | |
3726 | "mov %%"R"bp, %c[rbp](%0) \n\t" | |
05b3e0c2 | 3727 | #ifdef CONFIG_X86_64 |
e08aa78a AK |
3728 | "mov %%r8, %c[r8](%0) \n\t" |
3729 | "mov %%r9, %c[r9](%0) \n\t" | |
3730 | "mov %%r10, %c[r10](%0) \n\t" | |
3731 | "mov %%r11, %c[r11](%0) \n\t" | |
3732 | "mov %%r12, %c[r12](%0) \n\t" | |
3733 | "mov %%r13, %c[r13](%0) \n\t" | |
3734 | "mov %%r14, %c[r14](%0) \n\t" | |
3735 | "mov %%r15, %c[r15](%0) \n\t" | |
6aa8b732 | 3736 | #endif |
c801949d AK |
3737 | "mov %%cr2, %%"R"ax \n\t" |
3738 | "mov %%"R"ax, %c[cr2](%0) \n\t" | |
3739 | ||
3740 | "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t" | |
e08aa78a AK |
3741 | "setbe %c[fail](%0) \n\t" |
3742 | : : "c"(vmx), "d"((unsigned long)HOST_RSP), | |
3743 | [launched]"i"(offsetof(struct vcpu_vmx, launched)), | |
3744 | [fail]"i"(offsetof(struct vcpu_vmx, fail)), | |
313dbd49 | 3745 | [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)), |
ad312c7c ZX |
3746 | [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])), |
3747 | [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])), | |
3748 | [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])), | |
3749 | [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])), | |
3750 | [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])), | |
3751 | [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])), | |
3752 | [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])), | |
05b3e0c2 | 3753 | #ifdef CONFIG_X86_64 |
ad312c7c ZX |
3754 | [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])), |
3755 | [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])), | |
3756 | [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])), | |
3757 | [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])), | |
3758 | [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])), | |
3759 | [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])), | |
3760 | [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])), | |
3761 | [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])), | |
6aa8b732 | 3762 | #endif |
ad312c7c | 3763 | [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)) |
c2036300 | 3764 | : "cc", "memory" |
c801949d | 3765 | , R"bx", R"di", R"si" |
c2036300 | 3766 | #ifdef CONFIG_X86_64 |
c2036300 LV |
3767 | , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15" |
3768 | #endif | |
3769 | ); | |
6aa8b732 | 3770 | |
6de4f3ad AK |
3771 | vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP) |
3772 | | (1 << VCPU_EXREG_PDPTR)); | |
5fdbf976 MT |
3773 | vcpu->arch.regs_dirty = 0; |
3774 | ||
e8a48342 AK |
3775 | if (vcpu->arch.switch_db_regs) |
3776 | get_debugreg(vcpu->arch.dr6, 6); | |
42dbaa5a | 3777 | |
1155f76a | 3778 | vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD); |
9c8cba37 AK |
3779 | if (vmx->rmode.irq.pending) |
3780 | fixup_rmode_irq(vmx); | |
1155f76a | 3781 | |
d77c26fc | 3782 | asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS)); |
15ad7146 | 3783 | vmx->launched = 1; |
1b6269db | 3784 | |
cf393f75 | 3785 | vmx_complete_interrupts(vmx); |
6aa8b732 AK |
3786 | } |
3787 | ||
c801949d AK |
3788 | #undef R |
3789 | #undef Q | |
3790 | ||
6aa8b732 AK |
3791 | static void vmx_free_vmcs(struct kvm_vcpu *vcpu) |
3792 | { | |
a2fa3e9f GH |
3793 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
3794 | ||
3795 | if (vmx->vmcs) { | |
543e4243 | 3796 | vcpu_clear(vmx); |
a2fa3e9f GH |
3797 | free_vmcs(vmx->vmcs); |
3798 | vmx->vmcs = NULL; | |
6aa8b732 AK |
3799 | } |
3800 | } | |
3801 | ||
3802 | static void vmx_free_vcpu(struct kvm_vcpu *vcpu) | |
3803 | { | |
fb3f0f51 RR |
3804 | struct vcpu_vmx *vmx = to_vmx(vcpu); |
3805 | ||
2384d2b3 SY |
3806 | spin_lock(&vmx_vpid_lock); |
3807 | if (vmx->vpid != 0) | |
3808 | __clear_bit(vmx->vpid, vmx_vpid_bitmap); | |
3809 | spin_unlock(&vmx_vpid_lock); | |
6aa8b732 | 3810 | vmx_free_vmcs(vcpu); |
fb3f0f51 RR |
3811 | kfree(vmx->host_msrs); |
3812 | kfree(vmx->guest_msrs); | |
3813 | kvm_vcpu_uninit(vcpu); | |
a4770347 | 3814 | kmem_cache_free(kvm_vcpu_cache, vmx); |
6aa8b732 AK |
3815 | } |
3816 | ||
fb3f0f51 | 3817 | static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id) |
6aa8b732 | 3818 | { |
fb3f0f51 | 3819 | int err; |
c16f862d | 3820 | struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
15ad7146 | 3821 | int cpu; |
6aa8b732 | 3822 | |
a2fa3e9f | 3823 | if (!vmx) |
fb3f0f51 RR |
3824 | return ERR_PTR(-ENOMEM); |
3825 | ||
2384d2b3 SY |
3826 | allocate_vpid(vmx); |
3827 | ||
fb3f0f51 RR |
3828 | err = kvm_vcpu_init(&vmx->vcpu, kvm, id); |
3829 | if (err) | |
3830 | goto free_vcpu; | |
965b58a5 | 3831 | |
a2fa3e9f | 3832 | vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL); |
fb3f0f51 RR |
3833 | if (!vmx->guest_msrs) { |
3834 | err = -ENOMEM; | |
3835 | goto uninit_vcpu; | |
3836 | } | |
965b58a5 | 3837 | |
a2fa3e9f GH |
3838 | vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3839 | if (!vmx->host_msrs) | |
fb3f0f51 | 3840 | goto free_guest_msrs; |
965b58a5 | 3841 | |
a2fa3e9f GH |
3842 | vmx->vmcs = alloc_vmcs(); |
3843 | if (!vmx->vmcs) | |
fb3f0f51 | 3844 | goto free_msrs; |
a2fa3e9f GH |
3845 | |
3846 | vmcs_clear(vmx->vmcs); | |
3847 | ||
15ad7146 AK |
3848 | cpu = get_cpu(); |
3849 | vmx_vcpu_load(&vmx->vcpu, cpu); | |
8b9cf98c | 3850 | err = vmx_vcpu_setup(vmx); |
fb3f0f51 | 3851 | vmx_vcpu_put(&vmx->vcpu); |
15ad7146 | 3852 | put_cpu(); |
fb3f0f51 RR |
3853 | if (err) |
3854 | goto free_vmcs; | |
5e4a0b3c MT |
3855 | if (vm_need_virtualize_apic_accesses(kvm)) |
3856 | if (alloc_apic_access_page(kvm) != 0) | |
3857 | goto free_vmcs; | |
fb3f0f51 | 3858 | |
b927a3ce SY |
3859 | if (enable_ept) { |
3860 | if (!kvm->arch.ept_identity_map_addr) | |
3861 | kvm->arch.ept_identity_map_addr = | |
3862 | VMX_EPT_IDENTITY_PAGETABLE_ADDR; | |
b7ebfb05 SY |
3863 | if (alloc_identity_pagetable(kvm) != 0) |
3864 | goto free_vmcs; | |
b927a3ce | 3865 | } |
b7ebfb05 | 3866 | |
fb3f0f51 RR |
3867 | return &vmx->vcpu; |
3868 | ||
3869 | free_vmcs: | |
3870 | free_vmcs(vmx->vmcs); | |
3871 | free_msrs: | |
3872 | kfree(vmx->host_msrs); | |
3873 | free_guest_msrs: | |
3874 | kfree(vmx->guest_msrs); | |
3875 | uninit_vcpu: | |
3876 | kvm_vcpu_uninit(&vmx->vcpu); | |
3877 | free_vcpu: | |
a4770347 | 3878 | kmem_cache_free(kvm_vcpu_cache, vmx); |
fb3f0f51 | 3879 | return ERR_PTR(err); |
6aa8b732 AK |
3880 | } |
3881 | ||
002c7f7c YS |
3882 | static void __init vmx_check_processor_compat(void *rtn) |
3883 | { | |
3884 | struct vmcs_config vmcs_conf; | |
3885 | ||
3886 | *(int *)rtn = 0; | |
3887 | if (setup_vmcs_config(&vmcs_conf) < 0) | |
3888 | *(int *)rtn = -EIO; | |
3889 | if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) { | |
3890 | printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n", | |
3891 | smp_processor_id()); | |
3892 | *(int *)rtn = -EIO; | |
3893 | } | |
3894 | } | |
3895 | ||
67253af5 SY |
3896 | static int get_ept_level(void) |
3897 | { | |
3898 | return VMX_EPT_DEFAULT_GAW + 1; | |
3899 | } | |
3900 | ||
4b12f0de | 3901 | static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio) |
64d4d521 | 3902 | { |
4b12f0de SY |
3903 | u64 ret; |
3904 | ||
522c68c4 SY |
3905 | /* For VT-d and EPT combination |
3906 | * 1. MMIO: always map as UC | |
3907 | * 2. EPT with VT-d: | |
3908 | * a. VT-d without snooping control feature: can't guarantee the | |
3909 | * result, try to trust guest. | |
3910 | * b. VT-d with snooping control feature: snooping control feature of | |
3911 | * VT-d engine can guarantee the cache correctness. Just set it | |
3912 | * to WB to keep consistent with host. So the same as item 3. | |
3913 | * 3. EPT without VT-d: always map as WB and set IGMT=1 to keep | |
3914 | * consistent with host MTRR | |
3915 | */ | |
4b12f0de SY |
3916 | if (is_mmio) |
3917 | ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT; | |
522c68c4 SY |
3918 | else if (vcpu->kvm->arch.iommu_domain && |
3919 | !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY)) | |
3920 | ret = kvm_get_guest_memory_type(vcpu, gfn) << | |
3921 | VMX_EPT_MT_EPTE_SHIFT; | |
4b12f0de | 3922 | else |
522c68c4 SY |
3923 | ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT) |
3924 | | VMX_EPT_IGMT_BIT; | |
4b12f0de SY |
3925 | |
3926 | return ret; | |
64d4d521 SY |
3927 | } |
3928 | ||
229456fc MT |
3929 | static const struct trace_print_flags vmx_exit_reasons_str[] = { |
3930 | { EXIT_REASON_EXCEPTION_NMI, "exception" }, | |
3931 | { EXIT_REASON_EXTERNAL_INTERRUPT, "ext_irq" }, | |
3932 | { EXIT_REASON_TRIPLE_FAULT, "triple_fault" }, | |
3933 | { EXIT_REASON_NMI_WINDOW, "nmi_window" }, | |
3934 | { EXIT_REASON_IO_INSTRUCTION, "io_instruction" }, | |
3935 | { EXIT_REASON_CR_ACCESS, "cr_access" }, | |
3936 | { EXIT_REASON_DR_ACCESS, "dr_access" }, | |
3937 | { EXIT_REASON_CPUID, "cpuid" }, | |
3938 | { EXIT_REASON_MSR_READ, "rdmsr" }, | |
3939 | { EXIT_REASON_MSR_WRITE, "wrmsr" }, | |
3940 | { EXIT_REASON_PENDING_INTERRUPT, "interrupt_window" }, | |
3941 | { EXIT_REASON_HLT, "halt" }, | |
3942 | { EXIT_REASON_INVLPG, "invlpg" }, | |
3943 | { EXIT_REASON_VMCALL, "hypercall" }, | |
3944 | { EXIT_REASON_TPR_BELOW_THRESHOLD, "tpr_below_thres" }, | |
3945 | { EXIT_REASON_APIC_ACCESS, "apic_access" }, | |
3946 | { EXIT_REASON_WBINVD, "wbinvd" }, | |
3947 | { EXIT_REASON_TASK_SWITCH, "task_switch" }, | |
3948 | { EXIT_REASON_EPT_VIOLATION, "ept_violation" }, | |
3949 | { -1, NULL } | |
3950 | }; | |
3951 | ||
344f414f JR |
3952 | static bool vmx_gb_page_enable(void) |
3953 | { | |
3954 | return false; | |
3955 | } | |
3956 | ||
cbdd1bea | 3957 | static struct kvm_x86_ops vmx_x86_ops = { |
6aa8b732 AK |
3958 | .cpu_has_kvm_support = cpu_has_kvm_support, |
3959 | .disabled_by_bios = vmx_disabled_by_bios, | |
3960 | .hardware_setup = hardware_setup, | |
3961 | .hardware_unsetup = hardware_unsetup, | |
002c7f7c | 3962 | .check_processor_compatibility = vmx_check_processor_compat, |
6aa8b732 AK |
3963 | .hardware_enable = hardware_enable, |
3964 | .hardware_disable = hardware_disable, | |
04547156 | 3965 | .cpu_has_accelerated_tpr = report_flexpriority, |
6aa8b732 AK |
3966 | |
3967 | .vcpu_create = vmx_create_vcpu, | |
3968 | .vcpu_free = vmx_free_vcpu, | |
04d2cc77 | 3969 | .vcpu_reset = vmx_vcpu_reset, |
6aa8b732 | 3970 | |
04d2cc77 | 3971 | .prepare_guest_switch = vmx_save_host_state, |
6aa8b732 AK |
3972 | .vcpu_load = vmx_vcpu_load, |
3973 | .vcpu_put = vmx_vcpu_put, | |
3974 | ||
3975 | .set_guest_debug = set_guest_debug, | |
3976 | .get_msr = vmx_get_msr, | |
3977 | .set_msr = vmx_set_msr, | |
3978 | .get_segment_base = vmx_get_segment_base, | |
3979 | .get_segment = vmx_get_segment, | |
3980 | .set_segment = vmx_set_segment, | |
2e4d2653 | 3981 | .get_cpl = vmx_get_cpl, |
6aa8b732 | 3982 | .get_cs_db_l_bits = vmx_get_cs_db_l_bits, |
25c4c276 | 3983 | .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits, |
6aa8b732 | 3984 | .set_cr0 = vmx_set_cr0, |
6aa8b732 AK |
3985 | .set_cr3 = vmx_set_cr3, |
3986 | .set_cr4 = vmx_set_cr4, | |
6aa8b732 | 3987 | .set_efer = vmx_set_efer, |
6aa8b732 AK |
3988 | .get_idt = vmx_get_idt, |
3989 | .set_idt = vmx_set_idt, | |
3990 | .get_gdt = vmx_get_gdt, | |
3991 | .set_gdt = vmx_set_gdt, | |
5fdbf976 | 3992 | .cache_reg = vmx_cache_reg, |
6aa8b732 AK |
3993 | .get_rflags = vmx_get_rflags, |
3994 | .set_rflags = vmx_set_rflags, | |
3995 | ||
3996 | .tlb_flush = vmx_flush_tlb, | |
6aa8b732 | 3997 | |
6aa8b732 | 3998 | .run = vmx_vcpu_run, |
6062d012 | 3999 | .handle_exit = vmx_handle_exit, |
6aa8b732 | 4000 | .skip_emulated_instruction = skip_emulated_instruction, |
2809f5d2 GC |
4001 | .set_interrupt_shadow = vmx_set_interrupt_shadow, |
4002 | .get_interrupt_shadow = vmx_get_interrupt_shadow, | |
102d8325 | 4003 | .patch_hypercall = vmx_patch_hypercall, |
2a8067f1 | 4004 | .set_irq = vmx_inject_irq, |
95ba8273 | 4005 | .set_nmi = vmx_inject_nmi, |
298101da | 4006 | .queue_exception = vmx_queue_exception, |
78646121 | 4007 | .interrupt_allowed = vmx_interrupt_allowed, |
95ba8273 GN |
4008 | .nmi_allowed = vmx_nmi_allowed, |
4009 | .enable_nmi_window = enable_nmi_window, | |
4010 | .enable_irq_window = enable_irq_window, | |
4011 | .update_cr8_intercept = update_cr8_intercept, | |
95ba8273 | 4012 | |
cbc94022 | 4013 | .set_tss_addr = vmx_set_tss_addr, |
67253af5 | 4014 | .get_tdp_level = get_ept_level, |
4b12f0de | 4015 | .get_mt_mask = vmx_get_mt_mask, |
229456fc MT |
4016 | |
4017 | .exit_reasons_str = vmx_exit_reasons_str, | |
344f414f | 4018 | .gb_page_enable = vmx_gb_page_enable, |
6aa8b732 AK |
4019 | }; |
4020 | ||
4021 | static int __init vmx_init(void) | |
4022 | { | |
fdef3ad1 HQ |
4023 | int r; |
4024 | ||
3e7c73e9 | 4025 | vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL); |
fdef3ad1 HQ |
4026 | if (!vmx_io_bitmap_a) |
4027 | return -ENOMEM; | |
4028 | ||
3e7c73e9 | 4029 | vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL); |
fdef3ad1 HQ |
4030 | if (!vmx_io_bitmap_b) { |
4031 | r = -ENOMEM; | |
4032 | goto out; | |
4033 | } | |
4034 | ||
5897297b AK |
4035 | vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL); |
4036 | if (!vmx_msr_bitmap_legacy) { | |
25c5f225 SY |
4037 | r = -ENOMEM; |
4038 | goto out1; | |
4039 | } | |
4040 | ||
5897297b AK |
4041 | vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL); |
4042 | if (!vmx_msr_bitmap_longmode) { | |
4043 | r = -ENOMEM; | |
4044 | goto out2; | |
4045 | } | |
4046 | ||
fdef3ad1 HQ |
4047 | /* |
4048 | * Allow direct access to the PC debug port (it is often used for I/O | |
4049 | * delays, but the vmexits simply slow things down). | |
4050 | */ | |
3e7c73e9 AK |
4051 | memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE); |
4052 | clear_bit(0x80, vmx_io_bitmap_a); | |
fdef3ad1 | 4053 | |
3e7c73e9 | 4054 | memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE); |
fdef3ad1 | 4055 | |
5897297b AK |
4056 | memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE); |
4057 | memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE); | |
25c5f225 | 4058 | |
2384d2b3 SY |
4059 | set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */ |
4060 | ||
cb498ea2 | 4061 | r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE); |
fdef3ad1 | 4062 | if (r) |
5897297b | 4063 | goto out3; |
25c5f225 | 4064 | |
5897297b AK |
4065 | vmx_disable_intercept_for_msr(MSR_FS_BASE, false); |
4066 | vmx_disable_intercept_for_msr(MSR_GS_BASE, false); | |
4067 | vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true); | |
4068 | vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false); | |
4069 | vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false); | |
4070 | vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false); | |
fdef3ad1 | 4071 | |
089d034e | 4072 | if (enable_ept) { |
1439442c | 4073 | bypass_guest_pf = 0; |
5fdbcb9d | 4074 | kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK | |
2aaf69dc | 4075 | VMX_EPT_WRITABLE_MASK); |
534e38b4 | 4076 | kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull, |
4b12f0de | 4077 | VMX_EPT_EXECUTABLE_MASK); |
5fdbcb9d SY |
4078 | kvm_enable_tdp(); |
4079 | } else | |
4080 | kvm_disable_tdp(); | |
1439442c | 4081 | |
c7addb90 AK |
4082 | if (bypass_guest_pf) |
4083 | kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull); | |
4084 | ||
fdef3ad1 HQ |
4085 | return 0; |
4086 | ||
5897297b AK |
4087 | out3: |
4088 | free_page((unsigned long)vmx_msr_bitmap_longmode); | |
25c5f225 | 4089 | out2: |
5897297b | 4090 | free_page((unsigned long)vmx_msr_bitmap_legacy); |
fdef3ad1 | 4091 | out1: |
3e7c73e9 | 4092 | free_page((unsigned long)vmx_io_bitmap_b); |
fdef3ad1 | 4093 | out: |
3e7c73e9 | 4094 | free_page((unsigned long)vmx_io_bitmap_a); |
fdef3ad1 | 4095 | return r; |
6aa8b732 AK |
4096 | } |
4097 | ||
4098 | static void __exit vmx_exit(void) | |
4099 | { | |
5897297b AK |
4100 | free_page((unsigned long)vmx_msr_bitmap_legacy); |
4101 | free_page((unsigned long)vmx_msr_bitmap_longmode); | |
3e7c73e9 AK |
4102 | free_page((unsigned long)vmx_io_bitmap_b); |
4103 | free_page((unsigned long)vmx_io_bitmap_a); | |
fdef3ad1 | 4104 | |
cb498ea2 | 4105 | kvm_exit(); |
6aa8b732 AK |
4106 | } |
4107 | ||
4108 | module_init(vmx_init) | |
4109 | module_exit(vmx_exit) |