Commit | Line | Data |
---|---|---|
6aa8b732 AK |
1 | /* |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * | |
4 | * AMD SVM support | |
5 | * | |
6 | * Copyright (C) 2006 Qumranet, Inc. | |
7 | * | |
8 | * Authors: | |
9 | * Yaniv Kamay <yaniv@qumranet.com> | |
10 | * Avi Kivity <avi@qumranet.com> | |
11 | * | |
12 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
13 | * the COPYING file in the top-level directory. | |
14 | * | |
15 | */ | |
edf88417 AK |
16 | #include <linux/kvm_host.h> |
17 | ||
85f455f7 | 18 | #include "irq.h" |
1d737c8a | 19 | #include "mmu.h" |
5fdbf976 | 20 | #include "kvm_cache_regs.h" |
fe4c7b19 | 21 | #include "x86.h" |
e495606d | 22 | |
6aa8b732 | 23 | #include <linux/module.h> |
9d8f549d | 24 | #include <linux/kernel.h> |
6aa8b732 AK |
25 | #include <linux/vmalloc.h> |
26 | #include <linux/highmem.h> | |
e8edc6e0 | 27 | #include <linux/sched.h> |
229456fc | 28 | #include <linux/ftrace_event.h> |
5a0e3ad6 | 29 | #include <linux/slab.h> |
6aa8b732 | 30 | |
e495606d | 31 | #include <asm/desc.h> |
6aa8b732 | 32 | |
63d1142f | 33 | #include <asm/virtext.h> |
229456fc | 34 | #include "trace.h" |
63d1142f | 35 | |
4ecac3fd AK |
36 | #define __ex(x) __kvm_handle_fault_on_reboot(x) |
37 | ||
6aa8b732 AK |
38 | MODULE_AUTHOR("Qumranet"); |
39 | MODULE_LICENSE("GPL"); | |
40 | ||
41 | #define IOPM_ALLOC_ORDER 2 | |
42 | #define MSRPM_ALLOC_ORDER 1 | |
43 | ||
6aa8b732 AK |
44 | #define SEG_TYPE_LDT 2 |
45 | #define SEG_TYPE_BUSY_TSS16 3 | |
46 | ||
6bc31bdc AP |
47 | #define SVM_FEATURE_NPT (1 << 0) |
48 | #define SVM_FEATURE_LBRV (1 << 1) | |
49 | #define SVM_FEATURE_SVML (1 << 2) | |
50 | #define SVM_FEATURE_NRIP (1 << 3) | |
51 | #define SVM_FEATURE_PAUSE_FILTER (1 << 10) | |
80b7706e | 52 | |
410e4d57 JR |
53 | #define NESTED_EXIT_HOST 0 /* Exit handled on host level */ |
54 | #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */ | |
55 | #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */ | |
56 | ||
24e09cbf JR |
57 | #define DEBUGCTL_RESERVED_BITS (~(0x3fULL)) |
58 | ||
6c8166a7 AK |
59 | static const u32 host_save_user_msrs[] = { |
60 | #ifdef CONFIG_X86_64 | |
61 | MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE, | |
62 | MSR_FS_BASE, | |
63 | #endif | |
64 | MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, | |
65 | }; | |
66 | ||
67 | #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs) | |
68 | ||
69 | struct kvm_vcpu; | |
70 | ||
e6aa9abd JR |
71 | struct nested_state { |
72 | struct vmcb *hsave; | |
73 | u64 hsave_msr; | |
4a810181 | 74 | u64 vm_cr_msr; |
e6aa9abd JR |
75 | u64 vmcb; |
76 | ||
77 | /* These are the merged vectors */ | |
78 | u32 *msrpm; | |
79 | ||
80 | /* gpa pointers to the real vectors */ | |
81 | u64 vmcb_msrpm; | |
ce2ac085 | 82 | u64 vmcb_iopm; |
aad42c64 | 83 | |
cd3ff653 JR |
84 | /* A VMEXIT is required but not yet emulated */ |
85 | bool exit_required; | |
86 | ||
aad42c64 JR |
87 | /* cache for intercepts of the guest */ |
88 | u16 intercept_cr_read; | |
89 | u16 intercept_cr_write; | |
90 | u16 intercept_dr_read; | |
91 | u16 intercept_dr_write; | |
92 | u32 intercept_exceptions; | |
93 | u64 intercept; | |
94 | ||
e6aa9abd JR |
95 | }; |
96 | ||
323c3d80 JR |
97 | #define MSRPM_OFFSETS 16 |
98 | static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly; | |
99 | ||
6c8166a7 AK |
100 | struct vcpu_svm { |
101 | struct kvm_vcpu vcpu; | |
102 | struct vmcb *vmcb; | |
103 | unsigned long vmcb_pa; | |
104 | struct svm_cpu_data *svm_data; | |
105 | uint64_t asid_generation; | |
106 | uint64_t sysenter_esp; | |
107 | uint64_t sysenter_eip; | |
108 | ||
109 | u64 next_rip; | |
110 | ||
111 | u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS]; | |
112 | u64 host_gs_base; | |
6c8166a7 AK |
113 | |
114 | u32 *msrpm; | |
6c8166a7 | 115 | |
e6aa9abd | 116 | struct nested_state nested; |
6be7d306 JK |
117 | |
118 | bool nmi_singlestep; | |
66b7138f JK |
119 | |
120 | unsigned int3_injected; | |
121 | unsigned long int3_rip; | |
6c8166a7 AK |
122 | }; |
123 | ||
455716fa JR |
124 | #define MSR_INVALID 0xffffffffU |
125 | ||
ac72a9b7 JR |
126 | static struct svm_direct_access_msrs { |
127 | u32 index; /* Index of the MSR */ | |
128 | bool always; /* True if intercept is always on */ | |
129 | } direct_access_msrs[] = { | |
130 | { .index = MSR_K6_STAR, .always = true }, | |
131 | { .index = MSR_IA32_SYSENTER_CS, .always = true }, | |
132 | #ifdef CONFIG_X86_64 | |
133 | { .index = MSR_GS_BASE, .always = true }, | |
134 | { .index = MSR_FS_BASE, .always = true }, | |
135 | { .index = MSR_KERNEL_GS_BASE, .always = true }, | |
136 | { .index = MSR_LSTAR, .always = true }, | |
137 | { .index = MSR_CSTAR, .always = true }, | |
138 | { .index = MSR_SYSCALL_MASK, .always = true }, | |
139 | #endif | |
140 | { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false }, | |
141 | { .index = MSR_IA32_LASTBRANCHTOIP, .always = false }, | |
142 | { .index = MSR_IA32_LASTINTFROMIP, .always = false }, | |
143 | { .index = MSR_IA32_LASTINTTOIP, .always = false }, | |
144 | { .index = MSR_INVALID, .always = false }, | |
145 | }; | |
146 | ||
709ddebf JR |
147 | /* enable NPT for AMD64 and X86 with PAE */ |
148 | #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE) | |
149 | static bool npt_enabled = true; | |
150 | #else | |
e0231715 | 151 | static bool npt_enabled; |
709ddebf | 152 | #endif |
6c7dac72 JR |
153 | static int npt = 1; |
154 | ||
155 | module_param(npt, int, S_IRUGO); | |
e3da3acd | 156 | |
4b6e4dca | 157 | static int nested = 1; |
236de055 AG |
158 | module_param(nested, int, S_IRUGO); |
159 | ||
44874f84 | 160 | static void svm_flush_tlb(struct kvm_vcpu *vcpu); |
a5c3832d | 161 | static void svm_complete_interrupts(struct vcpu_svm *svm); |
04d2cc77 | 162 | |
410e4d57 | 163 | static int nested_svm_exit_handled(struct vcpu_svm *svm); |
b8e88bc8 | 164 | static int nested_svm_intercept(struct vcpu_svm *svm); |
cf74a78b | 165 | static int nested_svm_vmexit(struct vcpu_svm *svm); |
cf74a78b AG |
166 | static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr, |
167 | bool has_error_code, u32 error_code); | |
168 | ||
a2fa3e9f GH |
169 | static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu) |
170 | { | |
fb3f0f51 | 171 | return container_of(vcpu, struct vcpu_svm, vcpu); |
a2fa3e9f GH |
172 | } |
173 | ||
3d6368ef AG |
174 | static inline bool is_nested(struct vcpu_svm *svm) |
175 | { | |
e6aa9abd | 176 | return svm->nested.vmcb; |
3d6368ef AG |
177 | } |
178 | ||
2af9194d JR |
179 | static inline void enable_gif(struct vcpu_svm *svm) |
180 | { | |
181 | svm->vcpu.arch.hflags |= HF_GIF_MASK; | |
182 | } | |
183 | ||
184 | static inline void disable_gif(struct vcpu_svm *svm) | |
185 | { | |
186 | svm->vcpu.arch.hflags &= ~HF_GIF_MASK; | |
187 | } | |
188 | ||
189 | static inline bool gif_set(struct vcpu_svm *svm) | |
190 | { | |
191 | return !!(svm->vcpu.arch.hflags & HF_GIF_MASK); | |
192 | } | |
193 | ||
4866d5e3 | 194 | static unsigned long iopm_base; |
6aa8b732 AK |
195 | |
196 | struct kvm_ldttss_desc { | |
197 | u16 limit0; | |
198 | u16 base0; | |
e0231715 JR |
199 | unsigned base1:8, type:5, dpl:2, p:1; |
200 | unsigned limit1:4, zero0:3, g:1, base2:8; | |
6aa8b732 AK |
201 | u32 base3; |
202 | u32 zero1; | |
203 | } __attribute__((packed)); | |
204 | ||
205 | struct svm_cpu_data { | |
206 | int cpu; | |
207 | ||
5008fdf5 AK |
208 | u64 asid_generation; |
209 | u32 max_asid; | |
210 | u32 next_asid; | |
6aa8b732 AK |
211 | struct kvm_ldttss_desc *tss_desc; |
212 | ||
213 | struct page *save_area; | |
214 | }; | |
215 | ||
216 | static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data); | |
80b7706e | 217 | static uint32_t svm_features; |
6aa8b732 AK |
218 | |
219 | struct svm_init_data { | |
220 | int cpu; | |
221 | int r; | |
222 | }; | |
223 | ||
224 | static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000}; | |
225 | ||
9d8f549d | 226 | #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges) |
6aa8b732 AK |
227 | #define MSRS_RANGE_SIZE 2048 |
228 | #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2) | |
229 | ||
455716fa JR |
230 | static u32 svm_msrpm_offset(u32 msr) |
231 | { | |
232 | u32 offset; | |
233 | int i; | |
234 | ||
235 | for (i = 0; i < NUM_MSR_MAPS; i++) { | |
236 | if (msr < msrpm_ranges[i] || | |
237 | msr >= msrpm_ranges[i] + MSRS_IN_RANGE) | |
238 | continue; | |
239 | ||
240 | offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */ | |
241 | offset += (i * MSRS_RANGE_SIZE); /* add range offset */ | |
242 | ||
243 | /* Now we have the u8 offset - but need the u32 offset */ | |
244 | return offset / 4; | |
245 | } | |
246 | ||
247 | /* MSR not in any range */ | |
248 | return MSR_INVALID; | |
249 | } | |
250 | ||
6aa8b732 AK |
251 | #define MAX_INST_SIZE 15 |
252 | ||
80b7706e JR |
253 | static inline u32 svm_has(u32 feat) |
254 | { | |
255 | return svm_features & feat; | |
256 | } | |
257 | ||
6aa8b732 AK |
258 | static inline void clgi(void) |
259 | { | |
4ecac3fd | 260 | asm volatile (__ex(SVM_CLGI)); |
6aa8b732 AK |
261 | } |
262 | ||
263 | static inline void stgi(void) | |
264 | { | |
4ecac3fd | 265 | asm volatile (__ex(SVM_STGI)); |
6aa8b732 AK |
266 | } |
267 | ||
268 | static inline void invlpga(unsigned long addr, u32 asid) | |
269 | { | |
e0231715 | 270 | asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid)); |
6aa8b732 AK |
271 | } |
272 | ||
6aa8b732 AK |
273 | static inline void force_new_asid(struct kvm_vcpu *vcpu) |
274 | { | |
a2fa3e9f | 275 | to_svm(vcpu)->asid_generation--; |
6aa8b732 AK |
276 | } |
277 | ||
278 | static inline void flush_guest_tlb(struct kvm_vcpu *vcpu) | |
279 | { | |
280 | force_new_asid(vcpu); | |
281 | } | |
282 | ||
283 | static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer) | |
284 | { | |
709ddebf | 285 | if (!npt_enabled && !(efer & EFER_LMA)) |
2b5203ee | 286 | efer &= ~EFER_LME; |
6aa8b732 | 287 | |
9962d032 | 288 | to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME; |
f6801dff | 289 | vcpu->arch.efer = efer; |
6aa8b732 AK |
290 | } |
291 | ||
6aa8b732 AK |
292 | static int is_external_interrupt(u32 info) |
293 | { | |
294 | info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID; | |
295 | return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR); | |
296 | } | |
297 | ||
2809f5d2 GC |
298 | static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask) |
299 | { | |
300 | struct vcpu_svm *svm = to_svm(vcpu); | |
301 | u32 ret = 0; | |
302 | ||
303 | if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) | |
48005f64 | 304 | ret |= KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS; |
2809f5d2 GC |
305 | return ret & mask; |
306 | } | |
307 | ||
308 | static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask) | |
309 | { | |
310 | struct vcpu_svm *svm = to_svm(vcpu); | |
311 | ||
312 | if (mask == 0) | |
313 | svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK; | |
314 | else | |
315 | svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK; | |
316 | ||
317 | } | |
318 | ||
6aa8b732 AK |
319 | static void skip_emulated_instruction(struct kvm_vcpu *vcpu) |
320 | { | |
a2fa3e9f GH |
321 | struct vcpu_svm *svm = to_svm(vcpu); |
322 | ||
6bc31bdc AP |
323 | if (svm->vmcb->control.next_rip != 0) |
324 | svm->next_rip = svm->vmcb->control.next_rip; | |
325 | ||
a2fa3e9f | 326 | if (!svm->next_rip) { |
851ba692 | 327 | if (emulate_instruction(vcpu, 0, 0, EMULTYPE_SKIP) != |
f629cf84 GN |
328 | EMULATE_DONE) |
329 | printk(KERN_DEBUG "%s: NOP\n", __func__); | |
6aa8b732 AK |
330 | return; |
331 | } | |
5fdbf976 MT |
332 | if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE) |
333 | printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n", | |
334 | __func__, kvm_rip_read(vcpu), svm->next_rip); | |
6aa8b732 | 335 | |
5fdbf976 | 336 | kvm_rip_write(vcpu, svm->next_rip); |
2809f5d2 | 337 | svm_set_interrupt_shadow(vcpu, 0); |
6aa8b732 AK |
338 | } |
339 | ||
116a4752 | 340 | static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr, |
ce7ddec4 JR |
341 | bool has_error_code, u32 error_code, |
342 | bool reinject) | |
116a4752 JK |
343 | { |
344 | struct vcpu_svm *svm = to_svm(vcpu); | |
345 | ||
e0231715 JR |
346 | /* |
347 | * If we are within a nested VM we'd better #VMEXIT and let the guest | |
348 | * handle the exception | |
349 | */ | |
ce7ddec4 JR |
350 | if (!reinject && |
351 | nested_svm_check_exception(svm, nr, has_error_code, error_code)) | |
116a4752 JK |
352 | return; |
353 | ||
66b7138f JK |
354 | if (nr == BP_VECTOR && !svm_has(SVM_FEATURE_NRIP)) { |
355 | unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu); | |
356 | ||
357 | /* | |
358 | * For guest debugging where we have to reinject #BP if some | |
359 | * INT3 is guest-owned: | |
360 | * Emulate nRIP by moving RIP forward. Will fail if injection | |
361 | * raises a fault that is not intercepted. Still better than | |
362 | * failing in all cases. | |
363 | */ | |
364 | skip_emulated_instruction(&svm->vcpu); | |
365 | rip = kvm_rip_read(&svm->vcpu); | |
366 | svm->int3_rip = rip + svm->vmcb->save.cs.base; | |
367 | svm->int3_injected = rip - old_rip; | |
368 | } | |
369 | ||
116a4752 JK |
370 | svm->vmcb->control.event_inj = nr |
371 | | SVM_EVTINJ_VALID | |
372 | | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0) | |
373 | | SVM_EVTINJ_TYPE_EXEPT; | |
374 | svm->vmcb->control.event_inj_err = error_code; | |
375 | } | |
376 | ||
6aa8b732 AK |
377 | static int has_svm(void) |
378 | { | |
63d1142f | 379 | const char *msg; |
6aa8b732 | 380 | |
63d1142f | 381 | if (!cpu_has_svm(&msg)) { |
ff81ff10 | 382 | printk(KERN_INFO "has_svm: %s\n", msg); |
6aa8b732 AK |
383 | return 0; |
384 | } | |
385 | ||
6aa8b732 AK |
386 | return 1; |
387 | } | |
388 | ||
389 | static void svm_hardware_disable(void *garbage) | |
390 | { | |
2c8dceeb | 391 | cpu_svm_disable(); |
6aa8b732 AK |
392 | } |
393 | ||
10474ae8 | 394 | static int svm_hardware_enable(void *garbage) |
6aa8b732 AK |
395 | { |
396 | ||
0fe1e009 | 397 | struct svm_cpu_data *sd; |
6aa8b732 | 398 | uint64_t efer; |
89a27f4d | 399 | struct desc_ptr gdt_descr; |
6aa8b732 AK |
400 | struct desc_struct *gdt; |
401 | int me = raw_smp_processor_id(); | |
402 | ||
10474ae8 AG |
403 | rdmsrl(MSR_EFER, efer); |
404 | if (efer & EFER_SVME) | |
405 | return -EBUSY; | |
406 | ||
6aa8b732 | 407 | if (!has_svm()) { |
e6732a5a ZA |
408 | printk(KERN_ERR "svm_hardware_enable: err EOPNOTSUPP on %d\n", |
409 | me); | |
10474ae8 | 410 | return -EINVAL; |
6aa8b732 | 411 | } |
0fe1e009 | 412 | sd = per_cpu(svm_data, me); |
6aa8b732 | 413 | |
0fe1e009 | 414 | if (!sd) { |
e6732a5a | 415 | printk(KERN_ERR "svm_hardware_enable: svm_data is NULL on %d\n", |
6aa8b732 | 416 | me); |
10474ae8 | 417 | return -EINVAL; |
6aa8b732 AK |
418 | } |
419 | ||
0fe1e009 TH |
420 | sd->asid_generation = 1; |
421 | sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1; | |
422 | sd->next_asid = sd->max_asid + 1; | |
6aa8b732 | 423 | |
d6ab1ed4 | 424 | native_store_gdt(&gdt_descr); |
89a27f4d | 425 | gdt = (struct desc_struct *)gdt_descr.address; |
0fe1e009 | 426 | sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS); |
6aa8b732 | 427 | |
9962d032 | 428 | wrmsrl(MSR_EFER, efer | EFER_SVME); |
6aa8b732 | 429 | |
d0316554 | 430 | wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT); |
10474ae8 AG |
431 | |
432 | return 0; | |
6aa8b732 AK |
433 | } |
434 | ||
0da1db75 JR |
435 | static void svm_cpu_uninit(int cpu) |
436 | { | |
0fe1e009 | 437 | struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id()); |
0da1db75 | 438 | |
0fe1e009 | 439 | if (!sd) |
0da1db75 JR |
440 | return; |
441 | ||
442 | per_cpu(svm_data, raw_smp_processor_id()) = NULL; | |
0fe1e009 TH |
443 | __free_page(sd->save_area); |
444 | kfree(sd); | |
0da1db75 JR |
445 | } |
446 | ||
6aa8b732 AK |
447 | static int svm_cpu_init(int cpu) |
448 | { | |
0fe1e009 | 449 | struct svm_cpu_data *sd; |
6aa8b732 AK |
450 | int r; |
451 | ||
0fe1e009 TH |
452 | sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL); |
453 | if (!sd) | |
6aa8b732 | 454 | return -ENOMEM; |
0fe1e009 TH |
455 | sd->cpu = cpu; |
456 | sd->save_area = alloc_page(GFP_KERNEL); | |
6aa8b732 | 457 | r = -ENOMEM; |
0fe1e009 | 458 | if (!sd->save_area) |
6aa8b732 AK |
459 | goto err_1; |
460 | ||
0fe1e009 | 461 | per_cpu(svm_data, cpu) = sd; |
6aa8b732 AK |
462 | |
463 | return 0; | |
464 | ||
465 | err_1: | |
0fe1e009 | 466 | kfree(sd); |
6aa8b732 AK |
467 | return r; |
468 | ||
469 | } | |
470 | ||
ac72a9b7 JR |
471 | static bool valid_msr_intercept(u32 index) |
472 | { | |
473 | int i; | |
474 | ||
475 | for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) | |
476 | if (direct_access_msrs[i].index == index) | |
477 | return true; | |
478 | ||
479 | return false; | |
480 | } | |
481 | ||
bfc733a7 RR |
482 | static void set_msr_interception(u32 *msrpm, unsigned msr, |
483 | int read, int write) | |
6aa8b732 | 484 | { |
455716fa JR |
485 | u8 bit_read, bit_write; |
486 | unsigned long tmp; | |
487 | u32 offset; | |
6aa8b732 | 488 | |
ac72a9b7 JR |
489 | /* |
490 | * If this warning triggers extend the direct_access_msrs list at the | |
491 | * beginning of the file | |
492 | */ | |
493 | WARN_ON(!valid_msr_intercept(msr)); | |
494 | ||
455716fa JR |
495 | offset = svm_msrpm_offset(msr); |
496 | bit_read = 2 * (msr & 0x0f); | |
497 | bit_write = 2 * (msr & 0x0f) + 1; | |
498 | tmp = msrpm[offset]; | |
499 | ||
500 | BUG_ON(offset == MSR_INVALID); | |
501 | ||
502 | read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp); | |
503 | write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp); | |
504 | ||
505 | msrpm[offset] = tmp; | |
6aa8b732 AK |
506 | } |
507 | ||
f65c229c JR |
508 | static void svm_vcpu_init_msrpm(u32 *msrpm) |
509 | { | |
ac72a9b7 JR |
510 | int i; |
511 | ||
f65c229c JR |
512 | memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER)); |
513 | ||
ac72a9b7 JR |
514 | for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) { |
515 | if (!direct_access_msrs[i].always) | |
516 | continue; | |
517 | ||
518 | set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1); | |
519 | } | |
f65c229c JR |
520 | } |
521 | ||
323c3d80 JR |
522 | static void add_msr_offset(u32 offset) |
523 | { | |
524 | int i; | |
525 | ||
526 | for (i = 0; i < MSRPM_OFFSETS; ++i) { | |
527 | ||
528 | /* Offset already in list? */ | |
529 | if (msrpm_offsets[i] == offset) | |
530 | return; | |
531 | ||
532 | /* Slot used by another offset? */ | |
533 | if (msrpm_offsets[i] != MSR_INVALID) | |
534 | continue; | |
535 | ||
536 | /* Add offset to list */ | |
537 | msrpm_offsets[i] = offset; | |
538 | ||
539 | return; | |
540 | } | |
541 | ||
542 | /* | |
543 | * If this BUG triggers the msrpm_offsets table has an overflow. Just | |
544 | * increase MSRPM_OFFSETS in this case. | |
545 | */ | |
546 | BUG(); | |
547 | } | |
548 | ||
549 | static void init_msrpm_offsets(void) | |
550 | { | |
551 | int i; | |
552 | ||
553 | memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets)); | |
554 | ||
555 | for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) { | |
556 | u32 offset; | |
557 | ||
558 | offset = svm_msrpm_offset(direct_access_msrs[i].index); | |
559 | BUG_ON(offset == MSR_INVALID); | |
560 | ||
561 | add_msr_offset(offset); | |
562 | } | |
563 | } | |
564 | ||
24e09cbf JR |
565 | static void svm_enable_lbrv(struct vcpu_svm *svm) |
566 | { | |
567 | u32 *msrpm = svm->msrpm; | |
568 | ||
569 | svm->vmcb->control.lbr_ctl = 1; | |
570 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1); | |
571 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1); | |
572 | set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1); | |
573 | set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1); | |
574 | } | |
575 | ||
576 | static void svm_disable_lbrv(struct vcpu_svm *svm) | |
577 | { | |
578 | u32 *msrpm = svm->msrpm; | |
579 | ||
580 | svm->vmcb->control.lbr_ctl = 0; | |
581 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0); | |
582 | set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0); | |
583 | set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0); | |
584 | set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0); | |
585 | } | |
586 | ||
6aa8b732 AK |
587 | static __init int svm_hardware_setup(void) |
588 | { | |
589 | int cpu; | |
590 | struct page *iopm_pages; | |
f65c229c | 591 | void *iopm_va; |
6aa8b732 AK |
592 | int r; |
593 | ||
6aa8b732 AK |
594 | iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER); |
595 | ||
596 | if (!iopm_pages) | |
597 | return -ENOMEM; | |
c8681339 AL |
598 | |
599 | iopm_va = page_address(iopm_pages); | |
600 | memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER)); | |
6aa8b732 AK |
601 | iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT; |
602 | ||
323c3d80 JR |
603 | init_msrpm_offsets(); |
604 | ||
50a37eb4 JR |
605 | if (boot_cpu_has(X86_FEATURE_NX)) |
606 | kvm_enable_efer_bits(EFER_NX); | |
607 | ||
1b2fd70c AG |
608 | if (boot_cpu_has(X86_FEATURE_FXSR_OPT)) |
609 | kvm_enable_efer_bits(EFER_FFXSR); | |
610 | ||
236de055 AG |
611 | if (nested) { |
612 | printk(KERN_INFO "kvm: Nested Virtualization enabled\n"); | |
613 | kvm_enable_efer_bits(EFER_SVME); | |
614 | } | |
615 | ||
3230bb47 | 616 | for_each_possible_cpu(cpu) { |
6aa8b732 AK |
617 | r = svm_cpu_init(cpu); |
618 | if (r) | |
f65c229c | 619 | goto err; |
6aa8b732 | 620 | } |
33bd6a0b JR |
621 | |
622 | svm_features = cpuid_edx(SVM_CPUID_FUNC); | |
623 | ||
e3da3acd JR |
624 | if (!svm_has(SVM_FEATURE_NPT)) |
625 | npt_enabled = false; | |
626 | ||
6c7dac72 JR |
627 | if (npt_enabled && !npt) { |
628 | printk(KERN_INFO "kvm: Nested Paging disabled\n"); | |
629 | npt_enabled = false; | |
630 | } | |
631 | ||
18552672 | 632 | if (npt_enabled) { |
e3da3acd | 633 | printk(KERN_INFO "kvm: Nested Paging enabled\n"); |
18552672 | 634 | kvm_enable_tdp(); |
5f4cb662 JR |
635 | } else |
636 | kvm_disable_tdp(); | |
e3da3acd | 637 | |
6aa8b732 AK |
638 | return 0; |
639 | ||
f65c229c | 640 | err: |
6aa8b732 AK |
641 | __free_pages(iopm_pages, IOPM_ALLOC_ORDER); |
642 | iopm_base = 0; | |
643 | return r; | |
644 | } | |
645 | ||
646 | static __exit void svm_hardware_unsetup(void) | |
647 | { | |
0da1db75 JR |
648 | int cpu; |
649 | ||
3230bb47 | 650 | for_each_possible_cpu(cpu) |
0da1db75 JR |
651 | svm_cpu_uninit(cpu); |
652 | ||
6aa8b732 | 653 | __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER); |
f65c229c | 654 | iopm_base = 0; |
6aa8b732 AK |
655 | } |
656 | ||
657 | static void init_seg(struct vmcb_seg *seg) | |
658 | { | |
659 | seg->selector = 0; | |
660 | seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK | | |
e0231715 | 661 | SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */ |
6aa8b732 AK |
662 | seg->limit = 0xffff; |
663 | seg->base = 0; | |
664 | } | |
665 | ||
666 | static void init_sys_seg(struct vmcb_seg *seg, uint32_t type) | |
667 | { | |
668 | seg->selector = 0; | |
669 | seg->attrib = SVM_SELECTOR_P_MASK | type; | |
670 | seg->limit = 0xffff; | |
671 | seg->base = 0; | |
672 | } | |
673 | ||
e6101a96 | 674 | static void init_vmcb(struct vcpu_svm *svm) |
6aa8b732 | 675 | { |
e6101a96 JR |
676 | struct vmcb_control_area *control = &svm->vmcb->control; |
677 | struct vmcb_save_area *save = &svm->vmcb->save; | |
6aa8b732 | 678 | |
bff78274 AK |
679 | svm->vcpu.fpu_active = 1; |
680 | ||
e0231715 | 681 | control->intercept_cr_read = INTERCEPT_CR0_MASK | |
6aa8b732 | 682 | INTERCEPT_CR3_MASK | |
649d6864 | 683 | INTERCEPT_CR4_MASK; |
6aa8b732 | 684 | |
e0231715 | 685 | control->intercept_cr_write = INTERCEPT_CR0_MASK | |
6aa8b732 | 686 | INTERCEPT_CR3_MASK | |
80a8119c AK |
687 | INTERCEPT_CR4_MASK | |
688 | INTERCEPT_CR8_MASK; | |
6aa8b732 | 689 | |
e0231715 | 690 | control->intercept_dr_read = INTERCEPT_DR0_MASK | |
6aa8b732 AK |
691 | INTERCEPT_DR1_MASK | |
692 | INTERCEPT_DR2_MASK | | |
727f5a23 JK |
693 | INTERCEPT_DR3_MASK | |
694 | INTERCEPT_DR4_MASK | | |
695 | INTERCEPT_DR5_MASK | | |
696 | INTERCEPT_DR6_MASK | | |
697 | INTERCEPT_DR7_MASK; | |
6aa8b732 | 698 | |
e0231715 | 699 | control->intercept_dr_write = INTERCEPT_DR0_MASK | |
6aa8b732 AK |
700 | INTERCEPT_DR1_MASK | |
701 | INTERCEPT_DR2_MASK | | |
702 | INTERCEPT_DR3_MASK | | |
727f5a23 | 703 | INTERCEPT_DR4_MASK | |
6aa8b732 | 704 | INTERCEPT_DR5_MASK | |
727f5a23 | 705 | INTERCEPT_DR6_MASK | |
6aa8b732 AK |
706 | INTERCEPT_DR7_MASK; |
707 | ||
7aa81cc0 | 708 | control->intercept_exceptions = (1 << PF_VECTOR) | |
53371b50 JR |
709 | (1 << UD_VECTOR) | |
710 | (1 << MC_VECTOR); | |
6aa8b732 AK |
711 | |
712 | ||
e0231715 | 713 | control->intercept = (1ULL << INTERCEPT_INTR) | |
6aa8b732 | 714 | (1ULL << INTERCEPT_NMI) | |
0152527b | 715 | (1ULL << INTERCEPT_SMI) | |
d225157b | 716 | (1ULL << INTERCEPT_SELECTIVE_CR0) | |
6aa8b732 | 717 | (1ULL << INTERCEPT_CPUID) | |
cf5a94d1 | 718 | (1ULL << INTERCEPT_INVD) | |
6aa8b732 | 719 | (1ULL << INTERCEPT_HLT) | |
a7052897 | 720 | (1ULL << INTERCEPT_INVLPG) | |
6aa8b732 AK |
721 | (1ULL << INTERCEPT_INVLPGA) | |
722 | (1ULL << INTERCEPT_IOIO_PROT) | | |
723 | (1ULL << INTERCEPT_MSR_PROT) | | |
724 | (1ULL << INTERCEPT_TASK_SWITCH) | | |
46fe4ddd | 725 | (1ULL << INTERCEPT_SHUTDOWN) | |
6aa8b732 AK |
726 | (1ULL << INTERCEPT_VMRUN) | |
727 | (1ULL << INTERCEPT_VMMCALL) | | |
728 | (1ULL << INTERCEPT_VMLOAD) | | |
729 | (1ULL << INTERCEPT_VMSAVE) | | |
730 | (1ULL << INTERCEPT_STGI) | | |
731 | (1ULL << INTERCEPT_CLGI) | | |
916ce236 | 732 | (1ULL << INTERCEPT_SKINIT) | |
cf5a94d1 | 733 | (1ULL << INTERCEPT_WBINVD) | |
916ce236 JR |
734 | (1ULL << INTERCEPT_MONITOR) | |
735 | (1ULL << INTERCEPT_MWAIT); | |
6aa8b732 AK |
736 | |
737 | control->iopm_base_pa = iopm_base; | |
f65c229c | 738 | control->msrpm_base_pa = __pa(svm->msrpm); |
0cc5064d | 739 | control->tsc_offset = 0; |
6aa8b732 AK |
740 | control->int_ctl = V_INTR_MASKING_MASK; |
741 | ||
742 | init_seg(&save->es); | |
743 | init_seg(&save->ss); | |
744 | init_seg(&save->ds); | |
745 | init_seg(&save->fs); | |
746 | init_seg(&save->gs); | |
747 | ||
748 | save->cs.selector = 0xf000; | |
749 | /* Executable/Readable Code Segment */ | |
750 | save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK | | |
751 | SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK; | |
752 | save->cs.limit = 0xffff; | |
d92899a0 AK |
753 | /* |
754 | * cs.base should really be 0xffff0000, but vmx can't handle that, so | |
755 | * be consistent with it. | |
756 | * | |
757 | * Replace when we have real mode working for vmx. | |
758 | */ | |
759 | save->cs.base = 0xf0000; | |
6aa8b732 AK |
760 | |
761 | save->gdtr.limit = 0xffff; | |
762 | save->idtr.limit = 0xffff; | |
763 | ||
764 | init_sys_seg(&save->ldtr, SEG_TYPE_LDT); | |
765 | init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16); | |
766 | ||
9962d032 | 767 | save->efer = EFER_SVME; |
d77c26fc | 768 | save->dr6 = 0xffff0ff0; |
6aa8b732 AK |
769 | save->dr7 = 0x400; |
770 | save->rflags = 2; | |
771 | save->rip = 0x0000fff0; | |
5fdbf976 | 772 | svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip; |
6aa8b732 | 773 | |
e0231715 JR |
774 | /* |
775 | * This is the guest-visible cr0 value. | |
18fa000a | 776 | * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0. |
6aa8b732 | 777 | */ |
18fa000a EH |
778 | svm->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET; |
779 | kvm_set_cr0(&svm->vcpu, svm->vcpu.arch.cr0); | |
780 | ||
66aee91a | 781 | save->cr4 = X86_CR4_PAE; |
6aa8b732 | 782 | /* rdx = ?? */ |
709ddebf JR |
783 | |
784 | if (npt_enabled) { | |
785 | /* Setup VMCB for Nested Paging */ | |
786 | control->nested_ctl = 1; | |
a7052897 MT |
787 | control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) | |
788 | (1ULL << INTERCEPT_INVLPG)); | |
709ddebf | 789 | control->intercept_exceptions &= ~(1 << PF_VECTOR); |
888f9f3e AK |
790 | control->intercept_cr_read &= ~INTERCEPT_CR3_MASK; |
791 | control->intercept_cr_write &= ~INTERCEPT_CR3_MASK; | |
709ddebf | 792 | save->g_pat = 0x0007040600070406ULL; |
709ddebf JR |
793 | save->cr3 = 0; |
794 | save->cr4 = 0; | |
795 | } | |
a79d2f18 | 796 | force_new_asid(&svm->vcpu); |
1371d904 | 797 | |
e6aa9abd | 798 | svm->nested.vmcb = 0; |
2af9194d JR |
799 | svm->vcpu.arch.hflags = 0; |
800 | ||
565d0998 ML |
801 | if (svm_has(SVM_FEATURE_PAUSE_FILTER)) { |
802 | control->pause_filter_count = 3000; | |
803 | control->intercept |= (1ULL << INTERCEPT_PAUSE); | |
804 | } | |
805 | ||
2af9194d | 806 | enable_gif(svm); |
6aa8b732 AK |
807 | } |
808 | ||
e00c8cf2 | 809 | static int svm_vcpu_reset(struct kvm_vcpu *vcpu) |
04d2cc77 AK |
810 | { |
811 | struct vcpu_svm *svm = to_svm(vcpu); | |
812 | ||
e6101a96 | 813 | init_vmcb(svm); |
70433389 | 814 | |
c5af89b6 | 815 | if (!kvm_vcpu_is_bsp(vcpu)) { |
5fdbf976 | 816 | kvm_rip_write(vcpu, 0); |
ad312c7c ZX |
817 | svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12; |
818 | svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8; | |
70433389 | 819 | } |
5fdbf976 MT |
820 | vcpu->arch.regs_avail = ~0; |
821 | vcpu->arch.regs_dirty = ~0; | |
e00c8cf2 AK |
822 | |
823 | return 0; | |
04d2cc77 AK |
824 | } |
825 | ||
fb3f0f51 | 826 | static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id) |
6aa8b732 | 827 | { |
a2fa3e9f | 828 | struct vcpu_svm *svm; |
6aa8b732 | 829 | struct page *page; |
f65c229c | 830 | struct page *msrpm_pages; |
b286d5d8 | 831 | struct page *hsave_page; |
3d6368ef | 832 | struct page *nested_msrpm_pages; |
fb3f0f51 | 833 | int err; |
6aa8b732 | 834 | |
c16f862d | 835 | svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); |
fb3f0f51 RR |
836 | if (!svm) { |
837 | err = -ENOMEM; | |
838 | goto out; | |
839 | } | |
840 | ||
841 | err = kvm_vcpu_init(&svm->vcpu, kvm, id); | |
842 | if (err) | |
843 | goto free_svm; | |
844 | ||
b7af4043 | 845 | err = -ENOMEM; |
6aa8b732 | 846 | page = alloc_page(GFP_KERNEL); |
b7af4043 | 847 | if (!page) |
fb3f0f51 | 848 | goto uninit; |
6aa8b732 | 849 | |
f65c229c JR |
850 | msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER); |
851 | if (!msrpm_pages) | |
b7af4043 | 852 | goto free_page1; |
3d6368ef AG |
853 | |
854 | nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER); | |
855 | if (!nested_msrpm_pages) | |
b7af4043 | 856 | goto free_page2; |
f65c229c | 857 | |
b286d5d8 AG |
858 | hsave_page = alloc_page(GFP_KERNEL); |
859 | if (!hsave_page) | |
b7af4043 TY |
860 | goto free_page3; |
861 | ||
e6aa9abd | 862 | svm->nested.hsave = page_address(hsave_page); |
b286d5d8 | 863 | |
b7af4043 TY |
864 | svm->msrpm = page_address(msrpm_pages); |
865 | svm_vcpu_init_msrpm(svm->msrpm); | |
866 | ||
e6aa9abd | 867 | svm->nested.msrpm = page_address(nested_msrpm_pages); |
323c3d80 | 868 | svm_vcpu_init_msrpm(svm->nested.msrpm); |
3d6368ef | 869 | |
a2fa3e9f GH |
870 | svm->vmcb = page_address(page); |
871 | clear_page(svm->vmcb); | |
872 | svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT; | |
873 | svm->asid_generation = 0; | |
e6101a96 | 874 | init_vmcb(svm); |
a2fa3e9f | 875 | |
fb3f0f51 | 876 | fx_init(&svm->vcpu); |
ad312c7c | 877 | svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE; |
c5af89b6 | 878 | if (kvm_vcpu_is_bsp(&svm->vcpu)) |
ad312c7c | 879 | svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP; |
6aa8b732 | 880 | |
fb3f0f51 | 881 | return &svm->vcpu; |
36241b8c | 882 | |
b7af4043 TY |
883 | free_page3: |
884 | __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER); | |
885 | free_page2: | |
886 | __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER); | |
887 | free_page1: | |
888 | __free_page(page); | |
fb3f0f51 RR |
889 | uninit: |
890 | kvm_vcpu_uninit(&svm->vcpu); | |
891 | free_svm: | |
a4770347 | 892 | kmem_cache_free(kvm_vcpu_cache, svm); |
fb3f0f51 RR |
893 | out: |
894 | return ERR_PTR(err); | |
6aa8b732 AK |
895 | } |
896 | ||
897 | static void svm_free_vcpu(struct kvm_vcpu *vcpu) | |
898 | { | |
a2fa3e9f GH |
899 | struct vcpu_svm *svm = to_svm(vcpu); |
900 | ||
fb3f0f51 | 901 | __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT)); |
f65c229c | 902 | __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER); |
e6aa9abd JR |
903 | __free_page(virt_to_page(svm->nested.hsave)); |
904 | __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER); | |
fb3f0f51 | 905 | kvm_vcpu_uninit(vcpu); |
a4770347 | 906 | kmem_cache_free(kvm_vcpu_cache, svm); |
6aa8b732 AK |
907 | } |
908 | ||
15ad7146 | 909 | static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
6aa8b732 | 910 | { |
a2fa3e9f | 911 | struct vcpu_svm *svm = to_svm(vcpu); |
15ad7146 | 912 | int i; |
0cc5064d | 913 | |
0cc5064d | 914 | if (unlikely(cpu != vcpu->cpu)) { |
e935d48e | 915 | u64 delta; |
0cc5064d | 916 | |
953899b6 JR |
917 | if (check_tsc_unstable()) { |
918 | /* | |
919 | * Make sure that the guest sees a monotonically | |
920 | * increasing TSC. | |
921 | */ | |
922 | delta = vcpu->arch.host_tsc - native_read_tsc(); | |
923 | svm->vmcb->control.tsc_offset += delta; | |
924 | if (is_nested(svm)) | |
925 | svm->nested.hsave->control.tsc_offset += delta; | |
926 | } | |
0cc5064d | 927 | vcpu->cpu = cpu; |
2f599714 | 928 | kvm_migrate_timers(vcpu); |
4b656b12 | 929 | svm->asid_generation = 0; |
0cc5064d | 930 | } |
94dfbdb3 AL |
931 | |
932 | for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++) | |
a2fa3e9f | 933 | rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]); |
6aa8b732 AK |
934 | } |
935 | ||
936 | static void svm_vcpu_put(struct kvm_vcpu *vcpu) | |
937 | { | |
a2fa3e9f | 938 | struct vcpu_svm *svm = to_svm(vcpu); |
94dfbdb3 AL |
939 | int i; |
940 | ||
e1beb1d3 | 941 | ++vcpu->stat.host_state_reload; |
94dfbdb3 | 942 | for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++) |
a2fa3e9f | 943 | wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]); |
94dfbdb3 | 944 | |
e935d48e | 945 | vcpu->arch.host_tsc = native_read_tsc(); |
6aa8b732 AK |
946 | } |
947 | ||
6aa8b732 AK |
948 | static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu) |
949 | { | |
a2fa3e9f | 950 | return to_svm(vcpu)->vmcb->save.rflags; |
6aa8b732 AK |
951 | } |
952 | ||
953 | static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
954 | { | |
a2fa3e9f | 955 | to_svm(vcpu)->vmcb->save.rflags = rflags; |
6aa8b732 AK |
956 | } |
957 | ||
6de4f3ad AK |
958 | static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg) |
959 | { | |
960 | switch (reg) { | |
961 | case VCPU_EXREG_PDPTR: | |
962 | BUG_ON(!npt_enabled); | |
963 | load_pdptrs(vcpu, vcpu->arch.cr3); | |
964 | break; | |
965 | default: | |
966 | BUG(); | |
967 | } | |
968 | } | |
969 | ||
f0b85051 AG |
970 | static void svm_set_vintr(struct vcpu_svm *svm) |
971 | { | |
972 | svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR; | |
973 | } | |
974 | ||
975 | static void svm_clear_vintr(struct vcpu_svm *svm) | |
976 | { | |
977 | svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR); | |
978 | } | |
979 | ||
6aa8b732 AK |
980 | static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg) |
981 | { | |
a2fa3e9f | 982 | struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save; |
6aa8b732 AK |
983 | |
984 | switch (seg) { | |
985 | case VCPU_SREG_CS: return &save->cs; | |
986 | case VCPU_SREG_DS: return &save->ds; | |
987 | case VCPU_SREG_ES: return &save->es; | |
988 | case VCPU_SREG_FS: return &save->fs; | |
989 | case VCPU_SREG_GS: return &save->gs; | |
990 | case VCPU_SREG_SS: return &save->ss; | |
991 | case VCPU_SREG_TR: return &save->tr; | |
992 | case VCPU_SREG_LDTR: return &save->ldtr; | |
993 | } | |
994 | BUG(); | |
8b6d44c7 | 995 | return NULL; |
6aa8b732 AK |
996 | } |
997 | ||
998 | static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg) | |
999 | { | |
1000 | struct vmcb_seg *s = svm_seg(vcpu, seg); | |
1001 | ||
1002 | return s->base; | |
1003 | } | |
1004 | ||
1005 | static void svm_get_segment(struct kvm_vcpu *vcpu, | |
1006 | struct kvm_segment *var, int seg) | |
1007 | { | |
1008 | struct vmcb_seg *s = svm_seg(vcpu, seg); | |
1009 | ||
1010 | var->base = s->base; | |
1011 | var->limit = s->limit; | |
1012 | var->selector = s->selector; | |
1013 | var->type = s->attrib & SVM_SELECTOR_TYPE_MASK; | |
1014 | var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1; | |
1015 | var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3; | |
1016 | var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1; | |
1017 | var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1; | |
1018 | var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1; | |
1019 | var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1; | |
1020 | var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1; | |
25022acc | 1021 | |
e0231715 JR |
1022 | /* |
1023 | * AMD's VMCB does not have an explicit unusable field, so emulate it | |
19bca6ab AP |
1024 | * for cross vendor migration purposes by "not present" |
1025 | */ | |
1026 | var->unusable = !var->present || (var->type == 0); | |
1027 | ||
1fbdc7a5 AP |
1028 | switch (seg) { |
1029 | case VCPU_SREG_CS: | |
1030 | /* | |
1031 | * SVM always stores 0 for the 'G' bit in the CS selector in | |
1032 | * the VMCB on a VMEXIT. This hurts cross-vendor migration: | |
1033 | * Intel's VMENTRY has a check on the 'G' bit. | |
1034 | */ | |
25022acc | 1035 | var->g = s->limit > 0xfffff; |
1fbdc7a5 AP |
1036 | break; |
1037 | case VCPU_SREG_TR: | |
1038 | /* | |
1039 | * Work around a bug where the busy flag in the tr selector | |
1040 | * isn't exposed | |
1041 | */ | |
c0d09828 | 1042 | var->type |= 0x2; |
1fbdc7a5 AP |
1043 | break; |
1044 | case VCPU_SREG_DS: | |
1045 | case VCPU_SREG_ES: | |
1046 | case VCPU_SREG_FS: | |
1047 | case VCPU_SREG_GS: | |
1048 | /* | |
1049 | * The accessed bit must always be set in the segment | |
1050 | * descriptor cache, although it can be cleared in the | |
1051 | * descriptor, the cached bit always remains at 1. Since | |
1052 | * Intel has a check on this, set it here to support | |
1053 | * cross-vendor migration. | |
1054 | */ | |
1055 | if (!var->unusable) | |
1056 | var->type |= 0x1; | |
1057 | break; | |
b586eb02 | 1058 | case VCPU_SREG_SS: |
e0231715 JR |
1059 | /* |
1060 | * On AMD CPUs sometimes the DB bit in the segment | |
b586eb02 AP |
1061 | * descriptor is left as 1, although the whole segment has |
1062 | * been made unusable. Clear it here to pass an Intel VMX | |
1063 | * entry check when cross vendor migrating. | |
1064 | */ | |
1065 | if (var->unusable) | |
1066 | var->db = 0; | |
1067 | break; | |
1fbdc7a5 | 1068 | } |
6aa8b732 AK |
1069 | } |
1070 | ||
2e4d2653 IE |
1071 | static int svm_get_cpl(struct kvm_vcpu *vcpu) |
1072 | { | |
1073 | struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save; | |
1074 | ||
1075 | return save->cpl; | |
1076 | } | |
1077 | ||
89a27f4d | 1078 | static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 1079 | { |
a2fa3e9f GH |
1080 | struct vcpu_svm *svm = to_svm(vcpu); |
1081 | ||
89a27f4d GN |
1082 | dt->size = svm->vmcb->save.idtr.limit; |
1083 | dt->address = svm->vmcb->save.idtr.base; | |
6aa8b732 AK |
1084 | } |
1085 | ||
89a27f4d | 1086 | static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 1087 | { |
a2fa3e9f GH |
1088 | struct vcpu_svm *svm = to_svm(vcpu); |
1089 | ||
89a27f4d GN |
1090 | svm->vmcb->save.idtr.limit = dt->size; |
1091 | svm->vmcb->save.idtr.base = dt->address ; | |
6aa8b732 AK |
1092 | } |
1093 | ||
89a27f4d | 1094 | static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 1095 | { |
a2fa3e9f GH |
1096 | struct vcpu_svm *svm = to_svm(vcpu); |
1097 | ||
89a27f4d GN |
1098 | dt->size = svm->vmcb->save.gdtr.limit; |
1099 | dt->address = svm->vmcb->save.gdtr.base; | |
6aa8b732 AK |
1100 | } |
1101 | ||
89a27f4d | 1102 | static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt) |
6aa8b732 | 1103 | { |
a2fa3e9f GH |
1104 | struct vcpu_svm *svm = to_svm(vcpu); |
1105 | ||
89a27f4d GN |
1106 | svm->vmcb->save.gdtr.limit = dt->size; |
1107 | svm->vmcb->save.gdtr.base = dt->address ; | |
6aa8b732 AK |
1108 | } |
1109 | ||
e8467fda AK |
1110 | static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu) |
1111 | { | |
1112 | } | |
1113 | ||
25c4c276 | 1114 | static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu) |
399badf3 AK |
1115 | { |
1116 | } | |
1117 | ||
d225157b AK |
1118 | static void update_cr0_intercept(struct vcpu_svm *svm) |
1119 | { | |
66a562f7 | 1120 | struct vmcb *vmcb = svm->vmcb; |
d225157b AK |
1121 | ulong gcr0 = svm->vcpu.arch.cr0; |
1122 | u64 *hcr0 = &svm->vmcb->save.cr0; | |
1123 | ||
1124 | if (!svm->vcpu.fpu_active) | |
1125 | *hcr0 |= SVM_CR0_SELECTIVE_MASK; | |
1126 | else | |
1127 | *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK) | |
1128 | | (gcr0 & SVM_CR0_SELECTIVE_MASK); | |
1129 | ||
1130 | ||
1131 | if (gcr0 == *hcr0 && svm->vcpu.fpu_active) { | |
66a562f7 JR |
1132 | vmcb->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK; |
1133 | vmcb->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK; | |
1134 | if (is_nested(svm)) { | |
1135 | struct vmcb *hsave = svm->nested.hsave; | |
1136 | ||
1137 | hsave->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK; | |
1138 | hsave->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK; | |
1139 | vmcb->control.intercept_cr_read |= svm->nested.intercept_cr_read; | |
1140 | vmcb->control.intercept_cr_write |= svm->nested.intercept_cr_write; | |
1141 | } | |
d225157b AK |
1142 | } else { |
1143 | svm->vmcb->control.intercept_cr_read |= INTERCEPT_CR0_MASK; | |
1144 | svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR0_MASK; | |
66a562f7 JR |
1145 | if (is_nested(svm)) { |
1146 | struct vmcb *hsave = svm->nested.hsave; | |
1147 | ||
1148 | hsave->control.intercept_cr_read |= INTERCEPT_CR0_MASK; | |
1149 | hsave->control.intercept_cr_write |= INTERCEPT_CR0_MASK; | |
1150 | } | |
d225157b AK |
1151 | } |
1152 | } | |
1153 | ||
6aa8b732 AK |
1154 | static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
1155 | { | |
a2fa3e9f GH |
1156 | struct vcpu_svm *svm = to_svm(vcpu); |
1157 | ||
7f5d8b56 JR |
1158 | if (is_nested(svm)) { |
1159 | /* | |
1160 | * We are here because we run in nested mode, the host kvm | |
1161 | * intercepts cr0 writes but the l1 hypervisor does not. | |
1162 | * But the L1 hypervisor may intercept selective cr0 writes. | |
1163 | * This needs to be checked here. | |
1164 | */ | |
1165 | unsigned long old, new; | |
1166 | ||
1167 | /* Remove bits that would trigger a real cr0 write intercept */ | |
1168 | old = vcpu->arch.cr0 & SVM_CR0_SELECTIVE_MASK; | |
1169 | new = cr0 & SVM_CR0_SELECTIVE_MASK; | |
1170 | ||
1171 | if (old == new) { | |
1172 | /* cr0 write with ts and mp unchanged */ | |
1173 | svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE; | |
1174 | if (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE) | |
1175 | return; | |
1176 | } | |
1177 | } | |
1178 | ||
05b3e0c2 | 1179 | #ifdef CONFIG_X86_64 |
f6801dff | 1180 | if (vcpu->arch.efer & EFER_LME) { |
707d92fa | 1181 | if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) { |
f6801dff | 1182 | vcpu->arch.efer |= EFER_LMA; |
2b5203ee | 1183 | svm->vmcb->save.efer |= EFER_LMA | EFER_LME; |
6aa8b732 AK |
1184 | } |
1185 | ||
d77c26fc | 1186 | if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) { |
f6801dff | 1187 | vcpu->arch.efer &= ~EFER_LMA; |
2b5203ee | 1188 | svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME); |
6aa8b732 AK |
1189 | } |
1190 | } | |
1191 | #endif | |
ad312c7c | 1192 | vcpu->arch.cr0 = cr0; |
888f9f3e AK |
1193 | |
1194 | if (!npt_enabled) | |
1195 | cr0 |= X86_CR0_PG | X86_CR0_WP; | |
02daab21 AK |
1196 | |
1197 | if (!vcpu->fpu_active) | |
334df50a | 1198 | cr0 |= X86_CR0_TS; |
709ddebf JR |
1199 | /* |
1200 | * re-enable caching here because the QEMU bios | |
1201 | * does not do it - this results in some delay at | |
1202 | * reboot | |
1203 | */ | |
1204 | cr0 &= ~(X86_CR0_CD | X86_CR0_NW); | |
a2fa3e9f | 1205 | svm->vmcb->save.cr0 = cr0; |
d225157b | 1206 | update_cr0_intercept(svm); |
6aa8b732 AK |
1207 | } |
1208 | ||
1209 | static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) | |
1210 | { | |
6394b649 | 1211 | unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE; |
e5eab0ce JR |
1212 | unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4; |
1213 | ||
1214 | if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE)) | |
1215 | force_new_asid(vcpu); | |
6394b649 | 1216 | |
ec077263 JR |
1217 | vcpu->arch.cr4 = cr4; |
1218 | if (!npt_enabled) | |
1219 | cr4 |= X86_CR4_PAE; | |
6394b649 | 1220 | cr4 |= host_cr4_mce; |
ec077263 | 1221 | to_svm(vcpu)->vmcb->save.cr4 = cr4; |
6aa8b732 AK |
1222 | } |
1223 | ||
1224 | static void svm_set_segment(struct kvm_vcpu *vcpu, | |
1225 | struct kvm_segment *var, int seg) | |
1226 | { | |
a2fa3e9f | 1227 | struct vcpu_svm *svm = to_svm(vcpu); |
6aa8b732 AK |
1228 | struct vmcb_seg *s = svm_seg(vcpu, seg); |
1229 | ||
1230 | s->base = var->base; | |
1231 | s->limit = var->limit; | |
1232 | s->selector = var->selector; | |
1233 | if (var->unusable) | |
1234 | s->attrib = 0; | |
1235 | else { | |
1236 | s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK); | |
1237 | s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT; | |
1238 | s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT; | |
1239 | s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT; | |
1240 | s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT; | |
1241 | s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT; | |
1242 | s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT; | |
1243 | s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT; | |
1244 | } | |
1245 | if (seg == VCPU_SREG_CS) | |
a2fa3e9f GH |
1246 | svm->vmcb->save.cpl |
1247 | = (svm->vmcb->save.cs.attrib | |
6aa8b732 AK |
1248 | >> SVM_SELECTOR_DPL_SHIFT) & 3; |
1249 | ||
1250 | } | |
1251 | ||
44c11430 | 1252 | static void update_db_intercept(struct kvm_vcpu *vcpu) |
6aa8b732 | 1253 | { |
d0bfb940 JK |
1254 | struct vcpu_svm *svm = to_svm(vcpu); |
1255 | ||
d0bfb940 JK |
1256 | svm->vmcb->control.intercept_exceptions &= |
1257 | ~((1 << DB_VECTOR) | (1 << BP_VECTOR)); | |
44c11430 | 1258 | |
6be7d306 | 1259 | if (svm->nmi_singlestep) |
44c11430 GN |
1260 | svm->vmcb->control.intercept_exceptions |= (1 << DB_VECTOR); |
1261 | ||
d0bfb940 JK |
1262 | if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) { |
1263 | if (vcpu->guest_debug & | |
1264 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) | |
1265 | svm->vmcb->control.intercept_exceptions |= | |
1266 | 1 << DB_VECTOR; | |
1267 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) | |
1268 | svm->vmcb->control.intercept_exceptions |= | |
1269 | 1 << BP_VECTOR; | |
1270 | } else | |
1271 | vcpu->guest_debug = 0; | |
44c11430 GN |
1272 | } |
1273 | ||
355be0b9 | 1274 | static void svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg) |
44c11430 | 1275 | { |
44c11430 GN |
1276 | struct vcpu_svm *svm = to_svm(vcpu); |
1277 | ||
ae675ef0 JK |
1278 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) |
1279 | svm->vmcb->save.dr7 = dbg->arch.debugreg[7]; | |
1280 | else | |
1281 | svm->vmcb->save.dr7 = vcpu->arch.dr7; | |
1282 | ||
355be0b9 | 1283 | update_db_intercept(vcpu); |
6aa8b732 AK |
1284 | } |
1285 | ||
1286 | static void load_host_msrs(struct kvm_vcpu *vcpu) | |
1287 | { | |
94dfbdb3 | 1288 | #ifdef CONFIG_X86_64 |
a2fa3e9f | 1289 | wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base); |
94dfbdb3 | 1290 | #endif |
6aa8b732 AK |
1291 | } |
1292 | ||
1293 | static void save_host_msrs(struct kvm_vcpu *vcpu) | |
1294 | { | |
94dfbdb3 | 1295 | #ifdef CONFIG_X86_64 |
a2fa3e9f | 1296 | rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base); |
94dfbdb3 | 1297 | #endif |
6aa8b732 AK |
1298 | } |
1299 | ||
0fe1e009 | 1300 | static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd) |
6aa8b732 | 1301 | { |
0fe1e009 TH |
1302 | if (sd->next_asid > sd->max_asid) { |
1303 | ++sd->asid_generation; | |
1304 | sd->next_asid = 1; | |
a2fa3e9f | 1305 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID; |
6aa8b732 AK |
1306 | } |
1307 | ||
0fe1e009 TH |
1308 | svm->asid_generation = sd->asid_generation; |
1309 | svm->vmcb->control.asid = sd->next_asid++; | |
6aa8b732 AK |
1310 | } |
1311 | ||
020df079 | 1312 | static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value) |
6aa8b732 | 1313 | { |
42dbaa5a | 1314 | struct vcpu_svm *svm = to_svm(vcpu); |
42dbaa5a | 1315 | |
020df079 | 1316 | svm->vmcb->save.dr7 = value; |
6aa8b732 AK |
1317 | } |
1318 | ||
851ba692 | 1319 | static int pf_interception(struct vcpu_svm *svm) |
6aa8b732 | 1320 | { |
6aa8b732 AK |
1321 | u64 fault_address; |
1322 | u32 error_code; | |
6aa8b732 | 1323 | |
a2fa3e9f GH |
1324 | fault_address = svm->vmcb->control.exit_info_2; |
1325 | error_code = svm->vmcb->control.exit_info_1; | |
af9ca2d7 | 1326 | |
229456fc | 1327 | trace_kvm_page_fault(fault_address, error_code); |
52c7847d AK |
1328 | if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu)) |
1329 | kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address); | |
3067714c | 1330 | return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code); |
6aa8b732 AK |
1331 | } |
1332 | ||
851ba692 | 1333 | static int db_interception(struct vcpu_svm *svm) |
d0bfb940 | 1334 | { |
851ba692 AK |
1335 | struct kvm_run *kvm_run = svm->vcpu.run; |
1336 | ||
d0bfb940 | 1337 | if (!(svm->vcpu.guest_debug & |
44c11430 | 1338 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) && |
6be7d306 | 1339 | !svm->nmi_singlestep) { |
d0bfb940 JK |
1340 | kvm_queue_exception(&svm->vcpu, DB_VECTOR); |
1341 | return 1; | |
1342 | } | |
44c11430 | 1343 | |
6be7d306 JK |
1344 | if (svm->nmi_singlestep) { |
1345 | svm->nmi_singlestep = false; | |
44c11430 GN |
1346 | if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) |
1347 | svm->vmcb->save.rflags &= | |
1348 | ~(X86_EFLAGS_TF | X86_EFLAGS_RF); | |
1349 | update_db_intercept(&svm->vcpu); | |
1350 | } | |
1351 | ||
1352 | if (svm->vcpu.guest_debug & | |
e0231715 | 1353 | (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) { |
44c11430 GN |
1354 | kvm_run->exit_reason = KVM_EXIT_DEBUG; |
1355 | kvm_run->debug.arch.pc = | |
1356 | svm->vmcb->save.cs.base + svm->vmcb->save.rip; | |
1357 | kvm_run->debug.arch.exception = DB_VECTOR; | |
1358 | return 0; | |
1359 | } | |
1360 | ||
1361 | return 1; | |
d0bfb940 JK |
1362 | } |
1363 | ||
851ba692 | 1364 | static int bp_interception(struct vcpu_svm *svm) |
d0bfb940 | 1365 | { |
851ba692 AK |
1366 | struct kvm_run *kvm_run = svm->vcpu.run; |
1367 | ||
d0bfb940 JK |
1368 | kvm_run->exit_reason = KVM_EXIT_DEBUG; |
1369 | kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip; | |
1370 | kvm_run->debug.arch.exception = BP_VECTOR; | |
1371 | return 0; | |
1372 | } | |
1373 | ||
851ba692 | 1374 | static int ud_interception(struct vcpu_svm *svm) |
7aa81cc0 AL |
1375 | { |
1376 | int er; | |
1377 | ||
851ba692 | 1378 | er = emulate_instruction(&svm->vcpu, 0, 0, EMULTYPE_TRAP_UD); |
7aa81cc0 | 1379 | if (er != EMULATE_DONE) |
7ee5d940 | 1380 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); |
7aa81cc0 AL |
1381 | return 1; |
1382 | } | |
1383 | ||
6b52d186 | 1384 | static void svm_fpu_activate(struct kvm_vcpu *vcpu) |
7807fa6c | 1385 | { |
6b52d186 | 1386 | struct vcpu_svm *svm = to_svm(vcpu); |
66a562f7 JR |
1387 | u32 excp; |
1388 | ||
1389 | if (is_nested(svm)) { | |
1390 | u32 h_excp, n_excp; | |
1391 | ||
1392 | h_excp = svm->nested.hsave->control.intercept_exceptions; | |
1393 | n_excp = svm->nested.intercept_exceptions; | |
1394 | h_excp &= ~(1 << NM_VECTOR); | |
1395 | excp = h_excp | n_excp; | |
1396 | } else { | |
1397 | excp = svm->vmcb->control.intercept_exceptions; | |
e0231715 | 1398 | excp &= ~(1 << NM_VECTOR); |
66a562f7 JR |
1399 | } |
1400 | ||
1401 | svm->vmcb->control.intercept_exceptions = excp; | |
1402 | ||
e756fc62 | 1403 | svm->vcpu.fpu_active = 1; |
d225157b | 1404 | update_cr0_intercept(svm); |
6b52d186 | 1405 | } |
a2fa3e9f | 1406 | |
6b52d186 AK |
1407 | static int nm_interception(struct vcpu_svm *svm) |
1408 | { | |
1409 | svm_fpu_activate(&svm->vcpu); | |
a2fa3e9f | 1410 | return 1; |
7807fa6c AL |
1411 | } |
1412 | ||
851ba692 | 1413 | static int mc_interception(struct vcpu_svm *svm) |
53371b50 JR |
1414 | { |
1415 | /* | |
1416 | * On an #MC intercept the MCE handler is not called automatically in | |
1417 | * the host. So do it by hand here. | |
1418 | */ | |
1419 | asm volatile ( | |
1420 | "int $0x12\n"); | |
1421 | /* not sure if we ever come back to this point */ | |
1422 | ||
1423 | return 1; | |
1424 | } | |
1425 | ||
851ba692 | 1426 | static int shutdown_interception(struct vcpu_svm *svm) |
46fe4ddd | 1427 | { |
851ba692 AK |
1428 | struct kvm_run *kvm_run = svm->vcpu.run; |
1429 | ||
46fe4ddd JR |
1430 | /* |
1431 | * VMCB is undefined after a SHUTDOWN intercept | |
1432 | * so reinitialize it. | |
1433 | */ | |
a2fa3e9f | 1434 | clear_page(svm->vmcb); |
e6101a96 | 1435 | init_vmcb(svm); |
46fe4ddd JR |
1436 | |
1437 | kvm_run->exit_reason = KVM_EXIT_SHUTDOWN; | |
1438 | return 0; | |
1439 | } | |
1440 | ||
851ba692 | 1441 | static int io_interception(struct vcpu_svm *svm) |
6aa8b732 | 1442 | { |
cf8f70bf | 1443 | struct kvm_vcpu *vcpu = &svm->vcpu; |
d77c26fc | 1444 | u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */ |
34c33d16 | 1445 | int size, in, string; |
039576c0 | 1446 | unsigned port; |
6aa8b732 | 1447 | |
e756fc62 | 1448 | ++svm->vcpu.stat.io_exits; |
e70669ab | 1449 | string = (io_info & SVM_IOIO_STR_MASK) != 0; |
039576c0 | 1450 | in = (io_info & SVM_IOIO_TYPE_MASK) != 0; |
cf8f70bf GN |
1451 | if (string || in) |
1452 | return !(emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DO_MMIO); | |
1453 | ||
039576c0 AK |
1454 | port = io_info >> 16; |
1455 | size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT; | |
cf8f70bf | 1456 | svm->next_rip = svm->vmcb->control.exit_info_2; |
e93f36bc | 1457 | skip_emulated_instruction(&svm->vcpu); |
cf8f70bf GN |
1458 | |
1459 | return kvm_fast_pio_out(vcpu, size, port); | |
6aa8b732 AK |
1460 | } |
1461 | ||
851ba692 | 1462 | static int nmi_interception(struct vcpu_svm *svm) |
c47f098d JR |
1463 | { |
1464 | return 1; | |
1465 | } | |
1466 | ||
851ba692 | 1467 | static int intr_interception(struct vcpu_svm *svm) |
a0698055 JR |
1468 | { |
1469 | ++svm->vcpu.stat.irq_exits; | |
1470 | return 1; | |
1471 | } | |
1472 | ||
851ba692 | 1473 | static int nop_on_interception(struct vcpu_svm *svm) |
6aa8b732 AK |
1474 | { |
1475 | return 1; | |
1476 | } | |
1477 | ||
851ba692 | 1478 | static int halt_interception(struct vcpu_svm *svm) |
6aa8b732 | 1479 | { |
5fdbf976 | 1480 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 1; |
e756fc62 RR |
1481 | skip_emulated_instruction(&svm->vcpu); |
1482 | return kvm_emulate_halt(&svm->vcpu); | |
6aa8b732 AK |
1483 | } |
1484 | ||
851ba692 | 1485 | static int vmmcall_interception(struct vcpu_svm *svm) |
02e235bc | 1486 | { |
5fdbf976 | 1487 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; |
e756fc62 | 1488 | skip_emulated_instruction(&svm->vcpu); |
7aa81cc0 AL |
1489 | kvm_emulate_hypercall(&svm->vcpu); |
1490 | return 1; | |
02e235bc AK |
1491 | } |
1492 | ||
c0725420 AG |
1493 | static int nested_svm_check_permissions(struct vcpu_svm *svm) |
1494 | { | |
f6801dff | 1495 | if (!(svm->vcpu.arch.efer & EFER_SVME) |
c0725420 AG |
1496 | || !is_paging(&svm->vcpu)) { |
1497 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
1498 | return 1; | |
1499 | } | |
1500 | ||
1501 | if (svm->vmcb->save.cpl) { | |
1502 | kvm_inject_gp(&svm->vcpu, 0); | |
1503 | return 1; | |
1504 | } | |
1505 | ||
1506 | return 0; | |
1507 | } | |
1508 | ||
cf74a78b AG |
1509 | static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr, |
1510 | bool has_error_code, u32 error_code) | |
1511 | { | |
b8e88bc8 JR |
1512 | int vmexit; |
1513 | ||
0295ad7d JR |
1514 | if (!is_nested(svm)) |
1515 | return 0; | |
cf74a78b | 1516 | |
0295ad7d JR |
1517 | svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr; |
1518 | svm->vmcb->control.exit_code_hi = 0; | |
1519 | svm->vmcb->control.exit_info_1 = error_code; | |
1520 | svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2; | |
1521 | ||
b8e88bc8 JR |
1522 | vmexit = nested_svm_intercept(svm); |
1523 | if (vmexit == NESTED_EXIT_DONE) | |
1524 | svm->nested.exit_required = true; | |
1525 | ||
1526 | return vmexit; | |
cf74a78b AG |
1527 | } |
1528 | ||
8fe54654 JR |
1529 | /* This function returns true if it is save to enable the irq window */ |
1530 | static inline bool nested_svm_intr(struct vcpu_svm *svm) | |
cf74a78b | 1531 | { |
26666957 | 1532 | if (!is_nested(svm)) |
8fe54654 | 1533 | return true; |
cf74a78b | 1534 | |
26666957 | 1535 | if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK)) |
8fe54654 | 1536 | return true; |
cf74a78b | 1537 | |
26666957 | 1538 | if (!(svm->vcpu.arch.hflags & HF_HIF_MASK)) |
8fe54654 | 1539 | return false; |
cf74a78b | 1540 | |
197717d5 JR |
1541 | svm->vmcb->control.exit_code = SVM_EXIT_INTR; |
1542 | svm->vmcb->control.exit_info_1 = 0; | |
1543 | svm->vmcb->control.exit_info_2 = 0; | |
26666957 | 1544 | |
cd3ff653 JR |
1545 | if (svm->nested.intercept & 1ULL) { |
1546 | /* | |
1547 | * The #vmexit can't be emulated here directly because this | |
1548 | * code path runs with irqs and preemtion disabled. A | |
1549 | * #vmexit emulation might sleep. Only signal request for | |
1550 | * the #vmexit here. | |
1551 | */ | |
1552 | svm->nested.exit_required = true; | |
236649de | 1553 | trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip); |
8fe54654 | 1554 | return false; |
cf74a78b AG |
1555 | } |
1556 | ||
8fe54654 | 1557 | return true; |
cf74a78b AG |
1558 | } |
1559 | ||
887f500c JR |
1560 | /* This function returns true if it is save to enable the nmi window */ |
1561 | static inline bool nested_svm_nmi(struct vcpu_svm *svm) | |
1562 | { | |
1563 | if (!is_nested(svm)) | |
1564 | return true; | |
1565 | ||
1566 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI))) | |
1567 | return true; | |
1568 | ||
1569 | svm->vmcb->control.exit_code = SVM_EXIT_NMI; | |
1570 | svm->nested.exit_required = true; | |
1571 | ||
1572 | return false; | |
1573 | } | |
1574 | ||
7597f129 | 1575 | static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page) |
34f80cfa JR |
1576 | { |
1577 | struct page *page; | |
1578 | ||
6c3bd3d7 JR |
1579 | might_sleep(); |
1580 | ||
34f80cfa | 1581 | page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT); |
34f80cfa JR |
1582 | if (is_error_page(page)) |
1583 | goto error; | |
1584 | ||
7597f129 JR |
1585 | *_page = page; |
1586 | ||
1587 | return kmap(page); | |
34f80cfa JR |
1588 | |
1589 | error: | |
1590 | kvm_release_page_clean(page); | |
1591 | kvm_inject_gp(&svm->vcpu, 0); | |
1592 | ||
1593 | return NULL; | |
1594 | } | |
1595 | ||
7597f129 | 1596 | static void nested_svm_unmap(struct page *page) |
34f80cfa | 1597 | { |
7597f129 | 1598 | kunmap(page); |
34f80cfa JR |
1599 | kvm_release_page_dirty(page); |
1600 | } | |
1601 | ||
ce2ac085 JR |
1602 | static int nested_svm_intercept_ioio(struct vcpu_svm *svm) |
1603 | { | |
1604 | unsigned port; | |
1605 | u8 val, bit; | |
1606 | u64 gpa; | |
1607 | ||
1608 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT))) | |
1609 | return NESTED_EXIT_HOST; | |
1610 | ||
1611 | port = svm->vmcb->control.exit_info_1 >> 16; | |
1612 | gpa = svm->nested.vmcb_iopm + (port / 8); | |
1613 | bit = port % 8; | |
1614 | val = 0; | |
1615 | ||
1616 | if (kvm_read_guest(svm->vcpu.kvm, gpa, &val, 1)) | |
1617 | val &= (1 << bit); | |
1618 | ||
1619 | return val ? NESTED_EXIT_DONE : NESTED_EXIT_HOST; | |
1620 | } | |
1621 | ||
d2477826 | 1622 | static int nested_svm_exit_handled_msr(struct vcpu_svm *svm) |
4c2161ae | 1623 | { |
0d6b3537 JR |
1624 | u32 offset, msr, value; |
1625 | int write, mask; | |
4c2161ae | 1626 | |
3d62d9aa | 1627 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT))) |
d2477826 | 1628 | return NESTED_EXIT_HOST; |
3d62d9aa | 1629 | |
0d6b3537 JR |
1630 | msr = svm->vcpu.arch.regs[VCPU_REGS_RCX]; |
1631 | offset = svm_msrpm_offset(msr); | |
1632 | write = svm->vmcb->control.exit_info_1 & 1; | |
1633 | mask = 1 << ((2 * (msr & 0xf)) + write); | |
4c2161ae | 1634 | |
0d6b3537 JR |
1635 | if (offset == MSR_INVALID) |
1636 | return NESTED_EXIT_DONE; | |
3d62d9aa | 1637 | |
0d6b3537 JR |
1638 | /* Offset is in 32 bit units but need in 8 bit units */ |
1639 | offset *= 4; | |
1640 | ||
1641 | if (kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + offset, &value, 4)) | |
1642 | return NESTED_EXIT_DONE; | |
1643 | ||
1644 | return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST; | |
4c2161ae JR |
1645 | } |
1646 | ||
410e4d57 | 1647 | static int nested_svm_exit_special(struct vcpu_svm *svm) |
cf74a78b | 1648 | { |
cf74a78b | 1649 | u32 exit_code = svm->vmcb->control.exit_code; |
4c2161ae | 1650 | |
410e4d57 JR |
1651 | switch (exit_code) { |
1652 | case SVM_EXIT_INTR: | |
1653 | case SVM_EXIT_NMI: | |
ff47a49b | 1654 | case SVM_EXIT_EXCP_BASE + MC_VECTOR: |
410e4d57 | 1655 | return NESTED_EXIT_HOST; |
410e4d57 | 1656 | case SVM_EXIT_NPF: |
e0231715 | 1657 | /* For now we are always handling NPFs when using them */ |
410e4d57 JR |
1658 | if (npt_enabled) |
1659 | return NESTED_EXIT_HOST; | |
1660 | break; | |
410e4d57 | 1661 | case SVM_EXIT_EXCP_BASE + PF_VECTOR: |
e0231715 | 1662 | /* When we're shadowing, trap PFs */ |
410e4d57 JR |
1663 | if (!npt_enabled) |
1664 | return NESTED_EXIT_HOST; | |
1665 | break; | |
66a562f7 JR |
1666 | case SVM_EXIT_EXCP_BASE + NM_VECTOR: |
1667 | nm_interception(svm); | |
1668 | break; | |
410e4d57 JR |
1669 | default: |
1670 | break; | |
cf74a78b AG |
1671 | } |
1672 | ||
410e4d57 JR |
1673 | return NESTED_EXIT_CONTINUE; |
1674 | } | |
1675 | ||
1676 | /* | |
1677 | * If this function returns true, this #vmexit was already handled | |
1678 | */ | |
b8e88bc8 | 1679 | static int nested_svm_intercept(struct vcpu_svm *svm) |
410e4d57 JR |
1680 | { |
1681 | u32 exit_code = svm->vmcb->control.exit_code; | |
1682 | int vmexit = NESTED_EXIT_HOST; | |
1683 | ||
cf74a78b | 1684 | switch (exit_code) { |
9c4e40b9 | 1685 | case SVM_EXIT_MSR: |
3d62d9aa | 1686 | vmexit = nested_svm_exit_handled_msr(svm); |
9c4e40b9 | 1687 | break; |
ce2ac085 JR |
1688 | case SVM_EXIT_IOIO: |
1689 | vmexit = nested_svm_intercept_ioio(svm); | |
1690 | break; | |
cf74a78b AG |
1691 | case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: { |
1692 | u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0); | |
aad42c64 | 1693 | if (svm->nested.intercept_cr_read & cr_bits) |
410e4d57 | 1694 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
1695 | break; |
1696 | } | |
1697 | case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: { | |
1698 | u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0); | |
aad42c64 | 1699 | if (svm->nested.intercept_cr_write & cr_bits) |
410e4d57 | 1700 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
1701 | break; |
1702 | } | |
1703 | case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: { | |
1704 | u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0); | |
aad42c64 | 1705 | if (svm->nested.intercept_dr_read & dr_bits) |
410e4d57 | 1706 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
1707 | break; |
1708 | } | |
1709 | case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: { | |
1710 | u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0); | |
aad42c64 | 1711 | if (svm->nested.intercept_dr_write & dr_bits) |
410e4d57 | 1712 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
1713 | break; |
1714 | } | |
1715 | case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: { | |
1716 | u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE); | |
aad42c64 | 1717 | if (svm->nested.intercept_exceptions & excp_bits) |
410e4d57 | 1718 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
1719 | break; |
1720 | } | |
228070b1 JR |
1721 | case SVM_EXIT_ERR: { |
1722 | vmexit = NESTED_EXIT_DONE; | |
1723 | break; | |
1724 | } | |
cf74a78b AG |
1725 | default: { |
1726 | u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR); | |
aad42c64 | 1727 | if (svm->nested.intercept & exit_bits) |
410e4d57 | 1728 | vmexit = NESTED_EXIT_DONE; |
cf74a78b AG |
1729 | } |
1730 | } | |
1731 | ||
b8e88bc8 JR |
1732 | return vmexit; |
1733 | } | |
1734 | ||
1735 | static int nested_svm_exit_handled(struct vcpu_svm *svm) | |
1736 | { | |
1737 | int vmexit; | |
1738 | ||
1739 | vmexit = nested_svm_intercept(svm); | |
1740 | ||
1741 | if (vmexit == NESTED_EXIT_DONE) | |
9c4e40b9 | 1742 | nested_svm_vmexit(svm); |
9c4e40b9 JR |
1743 | |
1744 | return vmexit; | |
cf74a78b AG |
1745 | } |
1746 | ||
0460a979 JR |
1747 | static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb) |
1748 | { | |
1749 | struct vmcb_control_area *dst = &dst_vmcb->control; | |
1750 | struct vmcb_control_area *from = &from_vmcb->control; | |
1751 | ||
1752 | dst->intercept_cr_read = from->intercept_cr_read; | |
1753 | dst->intercept_cr_write = from->intercept_cr_write; | |
1754 | dst->intercept_dr_read = from->intercept_dr_read; | |
1755 | dst->intercept_dr_write = from->intercept_dr_write; | |
1756 | dst->intercept_exceptions = from->intercept_exceptions; | |
1757 | dst->intercept = from->intercept; | |
1758 | dst->iopm_base_pa = from->iopm_base_pa; | |
1759 | dst->msrpm_base_pa = from->msrpm_base_pa; | |
1760 | dst->tsc_offset = from->tsc_offset; | |
1761 | dst->asid = from->asid; | |
1762 | dst->tlb_ctl = from->tlb_ctl; | |
1763 | dst->int_ctl = from->int_ctl; | |
1764 | dst->int_vector = from->int_vector; | |
1765 | dst->int_state = from->int_state; | |
1766 | dst->exit_code = from->exit_code; | |
1767 | dst->exit_code_hi = from->exit_code_hi; | |
1768 | dst->exit_info_1 = from->exit_info_1; | |
1769 | dst->exit_info_2 = from->exit_info_2; | |
1770 | dst->exit_int_info = from->exit_int_info; | |
1771 | dst->exit_int_info_err = from->exit_int_info_err; | |
1772 | dst->nested_ctl = from->nested_ctl; | |
1773 | dst->event_inj = from->event_inj; | |
1774 | dst->event_inj_err = from->event_inj_err; | |
1775 | dst->nested_cr3 = from->nested_cr3; | |
1776 | dst->lbr_ctl = from->lbr_ctl; | |
1777 | } | |
1778 | ||
34f80cfa | 1779 | static int nested_svm_vmexit(struct vcpu_svm *svm) |
cf74a78b | 1780 | { |
34f80cfa | 1781 | struct vmcb *nested_vmcb; |
e6aa9abd | 1782 | struct vmcb *hsave = svm->nested.hsave; |
33740e40 | 1783 | struct vmcb *vmcb = svm->vmcb; |
7597f129 | 1784 | struct page *page; |
cf74a78b | 1785 | |
17897f36 JR |
1786 | trace_kvm_nested_vmexit_inject(vmcb->control.exit_code, |
1787 | vmcb->control.exit_info_1, | |
1788 | vmcb->control.exit_info_2, | |
1789 | vmcb->control.exit_int_info, | |
1790 | vmcb->control.exit_int_info_err); | |
1791 | ||
7597f129 | 1792 | nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page); |
34f80cfa JR |
1793 | if (!nested_vmcb) |
1794 | return 1; | |
1795 | ||
06fc7772 JR |
1796 | /* Exit nested SVM mode */ |
1797 | svm->nested.vmcb = 0; | |
1798 | ||
cf74a78b | 1799 | /* Give the current vmcb to the guest */ |
33740e40 JR |
1800 | disable_gif(svm); |
1801 | ||
1802 | nested_vmcb->save.es = vmcb->save.es; | |
1803 | nested_vmcb->save.cs = vmcb->save.cs; | |
1804 | nested_vmcb->save.ss = vmcb->save.ss; | |
1805 | nested_vmcb->save.ds = vmcb->save.ds; | |
1806 | nested_vmcb->save.gdtr = vmcb->save.gdtr; | |
1807 | nested_vmcb->save.idtr = vmcb->save.idtr; | |
cdbbdc12 | 1808 | nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu); |
2be4fc7a | 1809 | nested_vmcb->save.cr3 = svm->vcpu.arch.cr3; |
33740e40 | 1810 | nested_vmcb->save.cr2 = vmcb->save.cr2; |
cdbbdc12 | 1811 | nested_vmcb->save.cr4 = svm->vcpu.arch.cr4; |
33740e40 JR |
1812 | nested_vmcb->save.rflags = vmcb->save.rflags; |
1813 | nested_vmcb->save.rip = vmcb->save.rip; | |
1814 | nested_vmcb->save.rsp = vmcb->save.rsp; | |
1815 | nested_vmcb->save.rax = vmcb->save.rax; | |
1816 | nested_vmcb->save.dr7 = vmcb->save.dr7; | |
1817 | nested_vmcb->save.dr6 = vmcb->save.dr6; | |
1818 | nested_vmcb->save.cpl = vmcb->save.cpl; | |
1819 | ||
1820 | nested_vmcb->control.int_ctl = vmcb->control.int_ctl; | |
1821 | nested_vmcb->control.int_vector = vmcb->control.int_vector; | |
1822 | nested_vmcb->control.int_state = vmcb->control.int_state; | |
1823 | nested_vmcb->control.exit_code = vmcb->control.exit_code; | |
1824 | nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi; | |
1825 | nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1; | |
1826 | nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2; | |
1827 | nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info; | |
1828 | nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err; | |
8d23c466 AG |
1829 | |
1830 | /* | |
1831 | * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have | |
1832 | * to make sure that we do not lose injected events. So check event_inj | |
1833 | * here and copy it to exit_int_info if it is valid. | |
1834 | * Exit_int_info and event_inj can't be both valid because the case | |
1835 | * below only happens on a VMRUN instruction intercept which has | |
1836 | * no valid exit_int_info set. | |
1837 | */ | |
1838 | if (vmcb->control.event_inj & SVM_EVTINJ_VALID) { | |
1839 | struct vmcb_control_area *nc = &nested_vmcb->control; | |
1840 | ||
1841 | nc->exit_int_info = vmcb->control.event_inj; | |
1842 | nc->exit_int_info_err = vmcb->control.event_inj_err; | |
1843 | } | |
1844 | ||
33740e40 JR |
1845 | nested_vmcb->control.tlb_ctl = 0; |
1846 | nested_vmcb->control.event_inj = 0; | |
1847 | nested_vmcb->control.event_inj_err = 0; | |
cf74a78b AG |
1848 | |
1849 | /* We always set V_INTR_MASKING and remember the old value in hflags */ | |
1850 | if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK)) | |
1851 | nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK; | |
1852 | ||
cf74a78b | 1853 | /* Restore the original control entries */ |
0460a979 | 1854 | copy_vmcb_control_area(vmcb, hsave); |
cf74a78b | 1855 | |
219b65dc AG |
1856 | kvm_clear_exception_queue(&svm->vcpu); |
1857 | kvm_clear_interrupt_queue(&svm->vcpu); | |
cf74a78b AG |
1858 | |
1859 | /* Restore selected save entries */ | |
1860 | svm->vmcb->save.es = hsave->save.es; | |
1861 | svm->vmcb->save.cs = hsave->save.cs; | |
1862 | svm->vmcb->save.ss = hsave->save.ss; | |
1863 | svm->vmcb->save.ds = hsave->save.ds; | |
1864 | svm->vmcb->save.gdtr = hsave->save.gdtr; | |
1865 | svm->vmcb->save.idtr = hsave->save.idtr; | |
1866 | svm->vmcb->save.rflags = hsave->save.rflags; | |
1867 | svm_set_efer(&svm->vcpu, hsave->save.efer); | |
1868 | svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE); | |
1869 | svm_set_cr4(&svm->vcpu, hsave->save.cr4); | |
1870 | if (npt_enabled) { | |
1871 | svm->vmcb->save.cr3 = hsave->save.cr3; | |
1872 | svm->vcpu.arch.cr3 = hsave->save.cr3; | |
1873 | } else { | |
1874 | kvm_set_cr3(&svm->vcpu, hsave->save.cr3); | |
1875 | } | |
1876 | kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax); | |
1877 | kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp); | |
1878 | kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip); | |
1879 | svm->vmcb->save.dr7 = 0; | |
1880 | svm->vmcb->save.cpl = 0; | |
1881 | svm->vmcb->control.exit_int_info = 0; | |
1882 | ||
7597f129 | 1883 | nested_svm_unmap(page); |
cf74a78b AG |
1884 | |
1885 | kvm_mmu_reset_context(&svm->vcpu); | |
1886 | kvm_mmu_load(&svm->vcpu); | |
1887 | ||
1888 | return 0; | |
1889 | } | |
3d6368ef | 1890 | |
9738b2c9 | 1891 | static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm) |
3d6368ef | 1892 | { |
323c3d80 JR |
1893 | /* |
1894 | * This function merges the msr permission bitmaps of kvm and the | |
1895 | * nested vmcb. It is omptimized in that it only merges the parts where | |
1896 | * the kvm msr permission bitmap may contain zero bits | |
1897 | */ | |
3d6368ef | 1898 | int i; |
9738b2c9 | 1899 | |
323c3d80 JR |
1900 | if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT))) |
1901 | return true; | |
9738b2c9 | 1902 | |
323c3d80 JR |
1903 | for (i = 0; i < MSRPM_OFFSETS; i++) { |
1904 | u32 value, p; | |
1905 | u64 offset; | |
9738b2c9 | 1906 | |
323c3d80 JR |
1907 | if (msrpm_offsets[i] == 0xffffffff) |
1908 | break; | |
3d6368ef | 1909 | |
0d6b3537 JR |
1910 | p = msrpm_offsets[i]; |
1911 | offset = svm->nested.vmcb_msrpm + (p * 4); | |
323c3d80 JR |
1912 | |
1913 | if (kvm_read_guest(svm->vcpu.kvm, offset, &value, 4)) | |
1914 | return false; | |
1915 | ||
1916 | svm->nested.msrpm[p] = svm->msrpm[p] | value; | |
1917 | } | |
1918 | ||
1919 | svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm); | |
9738b2c9 JR |
1920 | |
1921 | return true; | |
3d6368ef AG |
1922 | } |
1923 | ||
9738b2c9 | 1924 | static bool nested_svm_vmrun(struct vcpu_svm *svm) |
3d6368ef | 1925 | { |
9738b2c9 | 1926 | struct vmcb *nested_vmcb; |
e6aa9abd | 1927 | struct vmcb *hsave = svm->nested.hsave; |
defbba56 | 1928 | struct vmcb *vmcb = svm->vmcb; |
7597f129 | 1929 | struct page *page; |
06fc7772 JR |
1930 | u64 vmcb_gpa; |
1931 | ||
1932 | vmcb_gpa = svm->vmcb->save.rax; | |
3d6368ef | 1933 | |
7597f129 | 1934 | nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page); |
9738b2c9 JR |
1935 | if (!nested_vmcb) |
1936 | return false; | |
1937 | ||
ecf1405d | 1938 | trace_kvm_nested_vmrun(svm->vmcb->save.rip - 3, vmcb_gpa, |
0ac406de JR |
1939 | nested_vmcb->save.rip, |
1940 | nested_vmcb->control.int_ctl, | |
1941 | nested_vmcb->control.event_inj, | |
1942 | nested_vmcb->control.nested_ctl); | |
1943 | ||
2e554e8d JR |
1944 | trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr_read, |
1945 | nested_vmcb->control.intercept_cr_write, | |
1946 | nested_vmcb->control.intercept_exceptions, | |
1947 | nested_vmcb->control.intercept); | |
1948 | ||
3d6368ef | 1949 | /* Clear internal status */ |
219b65dc AG |
1950 | kvm_clear_exception_queue(&svm->vcpu); |
1951 | kvm_clear_interrupt_queue(&svm->vcpu); | |
3d6368ef | 1952 | |
e0231715 JR |
1953 | /* |
1954 | * Save the old vmcb, so we don't need to pick what we save, but can | |
1955 | * restore everything when a VMEXIT occurs | |
1956 | */ | |
defbba56 JR |
1957 | hsave->save.es = vmcb->save.es; |
1958 | hsave->save.cs = vmcb->save.cs; | |
1959 | hsave->save.ss = vmcb->save.ss; | |
1960 | hsave->save.ds = vmcb->save.ds; | |
1961 | hsave->save.gdtr = vmcb->save.gdtr; | |
1962 | hsave->save.idtr = vmcb->save.idtr; | |
f6801dff | 1963 | hsave->save.efer = svm->vcpu.arch.efer; |
4d4ec087 | 1964 | hsave->save.cr0 = kvm_read_cr0(&svm->vcpu); |
defbba56 JR |
1965 | hsave->save.cr4 = svm->vcpu.arch.cr4; |
1966 | hsave->save.rflags = vmcb->save.rflags; | |
1967 | hsave->save.rip = svm->next_rip; | |
1968 | hsave->save.rsp = vmcb->save.rsp; | |
1969 | hsave->save.rax = vmcb->save.rax; | |
1970 | if (npt_enabled) | |
1971 | hsave->save.cr3 = vmcb->save.cr3; | |
1972 | else | |
1973 | hsave->save.cr3 = svm->vcpu.arch.cr3; | |
1974 | ||
0460a979 | 1975 | copy_vmcb_control_area(hsave, vmcb); |
3d6368ef AG |
1976 | |
1977 | if (svm->vmcb->save.rflags & X86_EFLAGS_IF) | |
1978 | svm->vcpu.arch.hflags |= HF_HIF_MASK; | |
1979 | else | |
1980 | svm->vcpu.arch.hflags &= ~HF_HIF_MASK; | |
1981 | ||
1982 | /* Load the nested guest state */ | |
1983 | svm->vmcb->save.es = nested_vmcb->save.es; | |
1984 | svm->vmcb->save.cs = nested_vmcb->save.cs; | |
1985 | svm->vmcb->save.ss = nested_vmcb->save.ss; | |
1986 | svm->vmcb->save.ds = nested_vmcb->save.ds; | |
1987 | svm->vmcb->save.gdtr = nested_vmcb->save.gdtr; | |
1988 | svm->vmcb->save.idtr = nested_vmcb->save.idtr; | |
1989 | svm->vmcb->save.rflags = nested_vmcb->save.rflags; | |
1990 | svm_set_efer(&svm->vcpu, nested_vmcb->save.efer); | |
1991 | svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0); | |
1992 | svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4); | |
1993 | if (npt_enabled) { | |
1994 | svm->vmcb->save.cr3 = nested_vmcb->save.cr3; | |
1995 | svm->vcpu.arch.cr3 = nested_vmcb->save.cr3; | |
0e5cbe36 | 1996 | } else |
3d6368ef | 1997 | kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3); |
0e5cbe36 JR |
1998 | |
1999 | /* Guest paging mode is active - reset mmu */ | |
2000 | kvm_mmu_reset_context(&svm->vcpu); | |
2001 | ||
defbba56 | 2002 | svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2; |
3d6368ef AG |
2003 | kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax); |
2004 | kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp); | |
2005 | kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip); | |
e0231715 | 2006 | |
3d6368ef AG |
2007 | /* In case we don't even reach vcpu_run, the fields are not updated */ |
2008 | svm->vmcb->save.rax = nested_vmcb->save.rax; | |
2009 | svm->vmcb->save.rsp = nested_vmcb->save.rsp; | |
2010 | svm->vmcb->save.rip = nested_vmcb->save.rip; | |
2011 | svm->vmcb->save.dr7 = nested_vmcb->save.dr7; | |
2012 | svm->vmcb->save.dr6 = nested_vmcb->save.dr6; | |
2013 | svm->vmcb->save.cpl = nested_vmcb->save.cpl; | |
2014 | ||
f7138538 | 2015 | svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL; |
ce2ac085 | 2016 | svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL; |
3d6368ef | 2017 | |
aad42c64 JR |
2018 | /* cache intercepts */ |
2019 | svm->nested.intercept_cr_read = nested_vmcb->control.intercept_cr_read; | |
2020 | svm->nested.intercept_cr_write = nested_vmcb->control.intercept_cr_write; | |
2021 | svm->nested.intercept_dr_read = nested_vmcb->control.intercept_dr_read; | |
2022 | svm->nested.intercept_dr_write = nested_vmcb->control.intercept_dr_write; | |
2023 | svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions; | |
2024 | svm->nested.intercept = nested_vmcb->control.intercept; | |
2025 | ||
3d6368ef | 2026 | force_new_asid(&svm->vcpu); |
3d6368ef | 2027 | svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK; |
3d6368ef AG |
2028 | if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK) |
2029 | svm->vcpu.arch.hflags |= HF_VINTR_MASK; | |
2030 | else | |
2031 | svm->vcpu.arch.hflags &= ~HF_VINTR_MASK; | |
2032 | ||
88ab24ad JR |
2033 | if (svm->vcpu.arch.hflags & HF_VINTR_MASK) { |
2034 | /* We only want the cr8 intercept bits of the guest */ | |
2035 | svm->vmcb->control.intercept_cr_read &= ~INTERCEPT_CR8_MASK; | |
2036 | svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK; | |
2037 | } | |
2038 | ||
e0231715 JR |
2039 | /* |
2040 | * We don't want a nested guest to be more powerful than the guest, so | |
2041 | * all intercepts are ORed | |
2042 | */ | |
88ab24ad JR |
2043 | svm->vmcb->control.intercept_cr_read |= |
2044 | nested_vmcb->control.intercept_cr_read; | |
2045 | svm->vmcb->control.intercept_cr_write |= | |
2046 | nested_vmcb->control.intercept_cr_write; | |
2047 | svm->vmcb->control.intercept_dr_read |= | |
2048 | nested_vmcb->control.intercept_dr_read; | |
2049 | svm->vmcb->control.intercept_dr_write |= | |
2050 | nested_vmcb->control.intercept_dr_write; | |
2051 | svm->vmcb->control.intercept_exceptions |= | |
2052 | nested_vmcb->control.intercept_exceptions; | |
2053 | ||
2054 | svm->vmcb->control.intercept |= nested_vmcb->control.intercept; | |
2055 | ||
2056 | svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl; | |
3d6368ef AG |
2057 | svm->vmcb->control.int_vector = nested_vmcb->control.int_vector; |
2058 | svm->vmcb->control.int_state = nested_vmcb->control.int_state; | |
2059 | svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset; | |
3d6368ef AG |
2060 | svm->vmcb->control.event_inj = nested_vmcb->control.event_inj; |
2061 | svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err; | |
2062 | ||
7597f129 | 2063 | nested_svm_unmap(page); |
9738b2c9 | 2064 | |
06fc7772 JR |
2065 | /* nested_vmcb is our indicator if nested SVM is activated */ |
2066 | svm->nested.vmcb = vmcb_gpa; | |
2067 | ||
2af9194d | 2068 | enable_gif(svm); |
3d6368ef | 2069 | |
9738b2c9 | 2070 | return true; |
3d6368ef AG |
2071 | } |
2072 | ||
9966bf68 | 2073 | static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb) |
5542675b AG |
2074 | { |
2075 | to_vmcb->save.fs = from_vmcb->save.fs; | |
2076 | to_vmcb->save.gs = from_vmcb->save.gs; | |
2077 | to_vmcb->save.tr = from_vmcb->save.tr; | |
2078 | to_vmcb->save.ldtr = from_vmcb->save.ldtr; | |
2079 | to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base; | |
2080 | to_vmcb->save.star = from_vmcb->save.star; | |
2081 | to_vmcb->save.lstar = from_vmcb->save.lstar; | |
2082 | to_vmcb->save.cstar = from_vmcb->save.cstar; | |
2083 | to_vmcb->save.sfmask = from_vmcb->save.sfmask; | |
2084 | to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs; | |
2085 | to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp; | |
2086 | to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip; | |
5542675b AG |
2087 | } |
2088 | ||
851ba692 | 2089 | static int vmload_interception(struct vcpu_svm *svm) |
5542675b | 2090 | { |
9966bf68 | 2091 | struct vmcb *nested_vmcb; |
7597f129 | 2092 | struct page *page; |
9966bf68 | 2093 | |
5542675b AG |
2094 | if (nested_svm_check_permissions(svm)) |
2095 | return 1; | |
2096 | ||
2097 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
2098 | skip_emulated_instruction(&svm->vcpu); | |
2099 | ||
7597f129 | 2100 | nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page); |
9966bf68 JR |
2101 | if (!nested_vmcb) |
2102 | return 1; | |
2103 | ||
2104 | nested_svm_vmloadsave(nested_vmcb, svm->vmcb); | |
7597f129 | 2105 | nested_svm_unmap(page); |
5542675b AG |
2106 | |
2107 | return 1; | |
2108 | } | |
2109 | ||
851ba692 | 2110 | static int vmsave_interception(struct vcpu_svm *svm) |
5542675b | 2111 | { |
9966bf68 | 2112 | struct vmcb *nested_vmcb; |
7597f129 | 2113 | struct page *page; |
9966bf68 | 2114 | |
5542675b AG |
2115 | if (nested_svm_check_permissions(svm)) |
2116 | return 1; | |
2117 | ||
2118 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
2119 | skip_emulated_instruction(&svm->vcpu); | |
2120 | ||
7597f129 | 2121 | nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page); |
9966bf68 JR |
2122 | if (!nested_vmcb) |
2123 | return 1; | |
2124 | ||
2125 | nested_svm_vmloadsave(svm->vmcb, nested_vmcb); | |
7597f129 | 2126 | nested_svm_unmap(page); |
5542675b AG |
2127 | |
2128 | return 1; | |
2129 | } | |
2130 | ||
851ba692 | 2131 | static int vmrun_interception(struct vcpu_svm *svm) |
3d6368ef | 2132 | { |
3d6368ef AG |
2133 | if (nested_svm_check_permissions(svm)) |
2134 | return 1; | |
2135 | ||
2136 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
2137 | skip_emulated_instruction(&svm->vcpu); | |
2138 | ||
9738b2c9 | 2139 | if (!nested_svm_vmrun(svm)) |
3d6368ef AG |
2140 | return 1; |
2141 | ||
9738b2c9 | 2142 | if (!nested_svm_vmrun_msrpm(svm)) |
1f8da478 JR |
2143 | goto failed; |
2144 | ||
2145 | return 1; | |
2146 | ||
2147 | failed: | |
2148 | ||
2149 | svm->vmcb->control.exit_code = SVM_EXIT_ERR; | |
2150 | svm->vmcb->control.exit_code_hi = 0; | |
2151 | svm->vmcb->control.exit_info_1 = 0; | |
2152 | svm->vmcb->control.exit_info_2 = 0; | |
2153 | ||
2154 | nested_svm_vmexit(svm); | |
3d6368ef AG |
2155 | |
2156 | return 1; | |
2157 | } | |
2158 | ||
851ba692 | 2159 | static int stgi_interception(struct vcpu_svm *svm) |
1371d904 AG |
2160 | { |
2161 | if (nested_svm_check_permissions(svm)) | |
2162 | return 1; | |
2163 | ||
2164 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
2165 | skip_emulated_instruction(&svm->vcpu); | |
2166 | ||
2af9194d | 2167 | enable_gif(svm); |
1371d904 AG |
2168 | |
2169 | return 1; | |
2170 | } | |
2171 | ||
851ba692 | 2172 | static int clgi_interception(struct vcpu_svm *svm) |
1371d904 AG |
2173 | { |
2174 | if (nested_svm_check_permissions(svm)) | |
2175 | return 1; | |
2176 | ||
2177 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
2178 | skip_emulated_instruction(&svm->vcpu); | |
2179 | ||
2af9194d | 2180 | disable_gif(svm); |
1371d904 AG |
2181 | |
2182 | /* After a CLGI no interrupts should come */ | |
2183 | svm_clear_vintr(svm); | |
2184 | svm->vmcb->control.int_ctl &= ~V_IRQ_MASK; | |
2185 | ||
2186 | return 1; | |
2187 | } | |
2188 | ||
851ba692 | 2189 | static int invlpga_interception(struct vcpu_svm *svm) |
ff092385 AG |
2190 | { |
2191 | struct kvm_vcpu *vcpu = &svm->vcpu; | |
ff092385 | 2192 | |
ec1ff790 JR |
2193 | trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX], |
2194 | vcpu->arch.regs[VCPU_REGS_RAX]); | |
2195 | ||
ff092385 AG |
2196 | /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */ |
2197 | kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]); | |
2198 | ||
2199 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 3; | |
2200 | skip_emulated_instruction(&svm->vcpu); | |
2201 | return 1; | |
2202 | } | |
2203 | ||
532a46b9 JR |
2204 | static int skinit_interception(struct vcpu_svm *svm) |
2205 | { | |
2206 | trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]); | |
2207 | ||
2208 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); | |
2209 | return 1; | |
2210 | } | |
2211 | ||
851ba692 | 2212 | static int invalid_op_interception(struct vcpu_svm *svm) |
6aa8b732 | 2213 | { |
7ee5d940 | 2214 | kvm_queue_exception(&svm->vcpu, UD_VECTOR); |
6aa8b732 AK |
2215 | return 1; |
2216 | } | |
2217 | ||
851ba692 | 2218 | static int task_switch_interception(struct vcpu_svm *svm) |
6aa8b732 | 2219 | { |
37817f29 | 2220 | u16 tss_selector; |
64a7ec06 GN |
2221 | int reason; |
2222 | int int_type = svm->vmcb->control.exit_int_info & | |
2223 | SVM_EXITINTINFO_TYPE_MASK; | |
8317c298 | 2224 | int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK; |
fe8e7f83 GN |
2225 | uint32_t type = |
2226 | svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK; | |
2227 | uint32_t idt_v = | |
2228 | svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID; | |
e269fb21 JK |
2229 | bool has_error_code = false; |
2230 | u32 error_code = 0; | |
37817f29 IE |
2231 | |
2232 | tss_selector = (u16)svm->vmcb->control.exit_info_1; | |
64a7ec06 | 2233 | |
37817f29 IE |
2234 | if (svm->vmcb->control.exit_info_2 & |
2235 | (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET)) | |
64a7ec06 GN |
2236 | reason = TASK_SWITCH_IRET; |
2237 | else if (svm->vmcb->control.exit_info_2 & | |
2238 | (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP)) | |
2239 | reason = TASK_SWITCH_JMP; | |
fe8e7f83 | 2240 | else if (idt_v) |
64a7ec06 GN |
2241 | reason = TASK_SWITCH_GATE; |
2242 | else | |
2243 | reason = TASK_SWITCH_CALL; | |
2244 | ||
fe8e7f83 GN |
2245 | if (reason == TASK_SWITCH_GATE) { |
2246 | switch (type) { | |
2247 | case SVM_EXITINTINFO_TYPE_NMI: | |
2248 | svm->vcpu.arch.nmi_injected = false; | |
2249 | break; | |
2250 | case SVM_EXITINTINFO_TYPE_EXEPT: | |
e269fb21 JK |
2251 | if (svm->vmcb->control.exit_info_2 & |
2252 | (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) { | |
2253 | has_error_code = true; | |
2254 | error_code = | |
2255 | (u32)svm->vmcb->control.exit_info_2; | |
2256 | } | |
fe8e7f83 GN |
2257 | kvm_clear_exception_queue(&svm->vcpu); |
2258 | break; | |
2259 | case SVM_EXITINTINFO_TYPE_INTR: | |
2260 | kvm_clear_interrupt_queue(&svm->vcpu); | |
2261 | break; | |
2262 | default: | |
2263 | break; | |
2264 | } | |
2265 | } | |
64a7ec06 | 2266 | |
8317c298 GN |
2267 | if (reason != TASK_SWITCH_GATE || |
2268 | int_type == SVM_EXITINTINFO_TYPE_SOFT || | |
2269 | (int_type == SVM_EXITINTINFO_TYPE_EXEPT && | |
f629cf84 GN |
2270 | (int_vec == OF_VECTOR || int_vec == BP_VECTOR))) |
2271 | skip_emulated_instruction(&svm->vcpu); | |
64a7ec06 | 2272 | |
acb54517 GN |
2273 | if (kvm_task_switch(&svm->vcpu, tss_selector, reason, |
2274 | has_error_code, error_code) == EMULATE_FAIL) { | |
2275 | svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
2276 | svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
2277 | svm->vcpu.run->internal.ndata = 0; | |
2278 | return 0; | |
2279 | } | |
2280 | return 1; | |
6aa8b732 AK |
2281 | } |
2282 | ||
851ba692 | 2283 | static int cpuid_interception(struct vcpu_svm *svm) |
6aa8b732 | 2284 | { |
5fdbf976 | 2285 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 2; |
e756fc62 | 2286 | kvm_emulate_cpuid(&svm->vcpu); |
06465c5a | 2287 | return 1; |
6aa8b732 AK |
2288 | } |
2289 | ||
851ba692 | 2290 | static int iret_interception(struct vcpu_svm *svm) |
95ba8273 GN |
2291 | { |
2292 | ++svm->vcpu.stat.nmi_window_exits; | |
2293 | svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET); | |
44c11430 | 2294 | svm->vcpu.arch.hflags |= HF_IRET_MASK; |
95ba8273 GN |
2295 | return 1; |
2296 | } | |
2297 | ||
851ba692 | 2298 | static int invlpg_interception(struct vcpu_svm *svm) |
a7052897 | 2299 | { |
851ba692 | 2300 | if (emulate_instruction(&svm->vcpu, 0, 0, 0) != EMULATE_DONE) |
a7052897 MT |
2301 | pr_unimpl(&svm->vcpu, "%s: failed\n", __func__); |
2302 | return 1; | |
2303 | } | |
2304 | ||
851ba692 | 2305 | static int emulate_on_interception(struct vcpu_svm *svm) |
6aa8b732 | 2306 | { |
851ba692 | 2307 | if (emulate_instruction(&svm->vcpu, 0, 0, 0) != EMULATE_DONE) |
b8688d51 | 2308 | pr_unimpl(&svm->vcpu, "%s: failed\n", __func__); |
6aa8b732 AK |
2309 | return 1; |
2310 | } | |
2311 | ||
851ba692 | 2312 | static int cr8_write_interception(struct vcpu_svm *svm) |
1d075434 | 2313 | { |
851ba692 AK |
2314 | struct kvm_run *kvm_run = svm->vcpu.run; |
2315 | ||
0a5fff19 GN |
2316 | u8 cr8_prev = kvm_get_cr8(&svm->vcpu); |
2317 | /* instruction emulation calls kvm_set_cr8() */ | |
851ba692 | 2318 | emulate_instruction(&svm->vcpu, 0, 0, 0); |
95ba8273 GN |
2319 | if (irqchip_in_kernel(svm->vcpu.kvm)) { |
2320 | svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK; | |
1d075434 | 2321 | return 1; |
95ba8273 | 2322 | } |
0a5fff19 GN |
2323 | if (cr8_prev <= kvm_get_cr8(&svm->vcpu)) |
2324 | return 1; | |
1d075434 JR |
2325 | kvm_run->exit_reason = KVM_EXIT_SET_TPR; |
2326 | return 0; | |
2327 | } | |
2328 | ||
6aa8b732 AK |
2329 | static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data) |
2330 | { | |
a2fa3e9f GH |
2331 | struct vcpu_svm *svm = to_svm(vcpu); |
2332 | ||
6aa8b732 | 2333 | switch (ecx) { |
af24a4e4 | 2334 | case MSR_IA32_TSC: { |
20824f30 | 2335 | u64 tsc_offset; |
6aa8b732 | 2336 | |
20824f30 JR |
2337 | if (is_nested(svm)) |
2338 | tsc_offset = svm->nested.hsave->control.tsc_offset; | |
2339 | else | |
2340 | tsc_offset = svm->vmcb->control.tsc_offset; | |
2341 | ||
2342 | *data = tsc_offset + native_read_tsc(); | |
6aa8b732 AK |
2343 | break; |
2344 | } | |
0e859cac | 2345 | case MSR_K6_STAR: |
a2fa3e9f | 2346 | *data = svm->vmcb->save.star; |
6aa8b732 | 2347 | break; |
0e859cac | 2348 | #ifdef CONFIG_X86_64 |
6aa8b732 | 2349 | case MSR_LSTAR: |
a2fa3e9f | 2350 | *data = svm->vmcb->save.lstar; |
6aa8b732 AK |
2351 | break; |
2352 | case MSR_CSTAR: | |
a2fa3e9f | 2353 | *data = svm->vmcb->save.cstar; |
6aa8b732 AK |
2354 | break; |
2355 | case MSR_KERNEL_GS_BASE: | |
a2fa3e9f | 2356 | *data = svm->vmcb->save.kernel_gs_base; |
6aa8b732 AK |
2357 | break; |
2358 | case MSR_SYSCALL_MASK: | |
a2fa3e9f | 2359 | *data = svm->vmcb->save.sfmask; |
6aa8b732 AK |
2360 | break; |
2361 | #endif | |
2362 | case MSR_IA32_SYSENTER_CS: | |
a2fa3e9f | 2363 | *data = svm->vmcb->save.sysenter_cs; |
6aa8b732 AK |
2364 | break; |
2365 | case MSR_IA32_SYSENTER_EIP: | |
017cb99e | 2366 | *data = svm->sysenter_eip; |
6aa8b732 AK |
2367 | break; |
2368 | case MSR_IA32_SYSENTER_ESP: | |
017cb99e | 2369 | *data = svm->sysenter_esp; |
6aa8b732 | 2370 | break; |
e0231715 JR |
2371 | /* |
2372 | * Nobody will change the following 5 values in the VMCB so we can | |
2373 | * safely return them on rdmsr. They will always be 0 until LBRV is | |
2374 | * implemented. | |
2375 | */ | |
a2938c80 JR |
2376 | case MSR_IA32_DEBUGCTLMSR: |
2377 | *data = svm->vmcb->save.dbgctl; | |
2378 | break; | |
2379 | case MSR_IA32_LASTBRANCHFROMIP: | |
2380 | *data = svm->vmcb->save.br_from; | |
2381 | break; | |
2382 | case MSR_IA32_LASTBRANCHTOIP: | |
2383 | *data = svm->vmcb->save.br_to; | |
2384 | break; | |
2385 | case MSR_IA32_LASTINTFROMIP: | |
2386 | *data = svm->vmcb->save.last_excp_from; | |
2387 | break; | |
2388 | case MSR_IA32_LASTINTTOIP: | |
2389 | *data = svm->vmcb->save.last_excp_to; | |
2390 | break; | |
b286d5d8 | 2391 | case MSR_VM_HSAVE_PA: |
e6aa9abd | 2392 | *data = svm->nested.hsave_msr; |
b286d5d8 | 2393 | break; |
eb6f302e | 2394 | case MSR_VM_CR: |
4a810181 | 2395 | *data = svm->nested.vm_cr_msr; |
eb6f302e | 2396 | break; |
c8a73f18 AG |
2397 | case MSR_IA32_UCODE_REV: |
2398 | *data = 0x01000065; | |
2399 | break; | |
6aa8b732 | 2400 | default: |
3bab1f5d | 2401 | return kvm_get_msr_common(vcpu, ecx, data); |
6aa8b732 AK |
2402 | } |
2403 | return 0; | |
2404 | } | |
2405 | ||
851ba692 | 2406 | static int rdmsr_interception(struct vcpu_svm *svm) |
6aa8b732 | 2407 | { |
ad312c7c | 2408 | u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX]; |
6aa8b732 AK |
2409 | u64 data; |
2410 | ||
59200273 AK |
2411 | if (svm_get_msr(&svm->vcpu, ecx, &data)) { |
2412 | trace_kvm_msr_read_ex(ecx); | |
c1a5d4f9 | 2413 | kvm_inject_gp(&svm->vcpu, 0); |
59200273 | 2414 | } else { |
229456fc | 2415 | trace_kvm_msr_read(ecx, data); |
af9ca2d7 | 2416 | |
5fdbf976 | 2417 | svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff; |
ad312c7c | 2418 | svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32; |
5fdbf976 | 2419 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 2; |
e756fc62 | 2420 | skip_emulated_instruction(&svm->vcpu); |
6aa8b732 AK |
2421 | } |
2422 | return 1; | |
2423 | } | |
2424 | ||
4a810181 JR |
2425 | static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data) |
2426 | { | |
2427 | struct vcpu_svm *svm = to_svm(vcpu); | |
2428 | int svm_dis, chg_mask; | |
2429 | ||
2430 | if (data & ~SVM_VM_CR_VALID_MASK) | |
2431 | return 1; | |
2432 | ||
2433 | chg_mask = SVM_VM_CR_VALID_MASK; | |
2434 | ||
2435 | if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK) | |
2436 | chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK); | |
2437 | ||
2438 | svm->nested.vm_cr_msr &= ~chg_mask; | |
2439 | svm->nested.vm_cr_msr |= (data & chg_mask); | |
2440 | ||
2441 | svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK; | |
2442 | ||
2443 | /* check for svm_disable while efer.svme is set */ | |
2444 | if (svm_dis && (vcpu->arch.efer & EFER_SVME)) | |
2445 | return 1; | |
2446 | ||
2447 | return 0; | |
2448 | } | |
2449 | ||
6aa8b732 AK |
2450 | static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data) |
2451 | { | |
a2fa3e9f GH |
2452 | struct vcpu_svm *svm = to_svm(vcpu); |
2453 | ||
6aa8b732 | 2454 | switch (ecx) { |
af24a4e4 | 2455 | case MSR_IA32_TSC: { |
20824f30 JR |
2456 | u64 tsc_offset = data - native_read_tsc(); |
2457 | u64 g_tsc_offset = 0; | |
2458 | ||
2459 | if (is_nested(svm)) { | |
2460 | g_tsc_offset = svm->vmcb->control.tsc_offset - | |
2461 | svm->nested.hsave->control.tsc_offset; | |
2462 | svm->nested.hsave->control.tsc_offset = tsc_offset; | |
2463 | } | |
2464 | ||
2465 | svm->vmcb->control.tsc_offset = tsc_offset + g_tsc_offset; | |
6aa8b732 | 2466 | |
6aa8b732 AK |
2467 | break; |
2468 | } | |
0e859cac | 2469 | case MSR_K6_STAR: |
a2fa3e9f | 2470 | svm->vmcb->save.star = data; |
6aa8b732 | 2471 | break; |
49b14f24 | 2472 | #ifdef CONFIG_X86_64 |
6aa8b732 | 2473 | case MSR_LSTAR: |
a2fa3e9f | 2474 | svm->vmcb->save.lstar = data; |
6aa8b732 AK |
2475 | break; |
2476 | case MSR_CSTAR: | |
a2fa3e9f | 2477 | svm->vmcb->save.cstar = data; |
6aa8b732 AK |
2478 | break; |
2479 | case MSR_KERNEL_GS_BASE: | |
a2fa3e9f | 2480 | svm->vmcb->save.kernel_gs_base = data; |
6aa8b732 AK |
2481 | break; |
2482 | case MSR_SYSCALL_MASK: | |
a2fa3e9f | 2483 | svm->vmcb->save.sfmask = data; |
6aa8b732 AK |
2484 | break; |
2485 | #endif | |
2486 | case MSR_IA32_SYSENTER_CS: | |
a2fa3e9f | 2487 | svm->vmcb->save.sysenter_cs = data; |
6aa8b732 AK |
2488 | break; |
2489 | case MSR_IA32_SYSENTER_EIP: | |
017cb99e | 2490 | svm->sysenter_eip = data; |
a2fa3e9f | 2491 | svm->vmcb->save.sysenter_eip = data; |
6aa8b732 AK |
2492 | break; |
2493 | case MSR_IA32_SYSENTER_ESP: | |
017cb99e | 2494 | svm->sysenter_esp = data; |
a2fa3e9f | 2495 | svm->vmcb->save.sysenter_esp = data; |
6aa8b732 | 2496 | break; |
a2938c80 | 2497 | case MSR_IA32_DEBUGCTLMSR: |
24e09cbf JR |
2498 | if (!svm_has(SVM_FEATURE_LBRV)) { |
2499 | pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n", | |
b8688d51 | 2500 | __func__, data); |
24e09cbf JR |
2501 | break; |
2502 | } | |
2503 | if (data & DEBUGCTL_RESERVED_BITS) | |
2504 | return 1; | |
2505 | ||
2506 | svm->vmcb->save.dbgctl = data; | |
2507 | if (data & (1ULL<<0)) | |
2508 | svm_enable_lbrv(svm); | |
2509 | else | |
2510 | svm_disable_lbrv(svm); | |
a2938c80 | 2511 | break; |
b286d5d8 | 2512 | case MSR_VM_HSAVE_PA: |
e6aa9abd | 2513 | svm->nested.hsave_msr = data; |
62b9abaa | 2514 | break; |
3c5d0a44 | 2515 | case MSR_VM_CR: |
4a810181 | 2516 | return svm_set_vm_cr(vcpu, data); |
3c5d0a44 | 2517 | case MSR_VM_IGNNE: |
3c5d0a44 AG |
2518 | pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data); |
2519 | break; | |
6aa8b732 | 2520 | default: |
3bab1f5d | 2521 | return kvm_set_msr_common(vcpu, ecx, data); |
6aa8b732 AK |
2522 | } |
2523 | return 0; | |
2524 | } | |
2525 | ||
851ba692 | 2526 | static int wrmsr_interception(struct vcpu_svm *svm) |
6aa8b732 | 2527 | { |
ad312c7c | 2528 | u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX]; |
5fdbf976 | 2529 | u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u) |
ad312c7c | 2530 | | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32); |
af9ca2d7 | 2531 | |
af9ca2d7 | 2532 | |
5fdbf976 | 2533 | svm->next_rip = kvm_rip_read(&svm->vcpu) + 2; |
59200273 AK |
2534 | if (svm_set_msr(&svm->vcpu, ecx, data)) { |
2535 | trace_kvm_msr_write_ex(ecx, data); | |
c1a5d4f9 | 2536 | kvm_inject_gp(&svm->vcpu, 0); |
59200273 AK |
2537 | } else { |
2538 | trace_kvm_msr_write(ecx, data); | |
e756fc62 | 2539 | skip_emulated_instruction(&svm->vcpu); |
59200273 | 2540 | } |
6aa8b732 AK |
2541 | return 1; |
2542 | } | |
2543 | ||
851ba692 | 2544 | static int msr_interception(struct vcpu_svm *svm) |
6aa8b732 | 2545 | { |
e756fc62 | 2546 | if (svm->vmcb->control.exit_info_1) |
851ba692 | 2547 | return wrmsr_interception(svm); |
6aa8b732 | 2548 | else |
851ba692 | 2549 | return rdmsr_interception(svm); |
6aa8b732 AK |
2550 | } |
2551 | ||
851ba692 | 2552 | static int interrupt_window_interception(struct vcpu_svm *svm) |
c1150d8c | 2553 | { |
851ba692 AK |
2554 | struct kvm_run *kvm_run = svm->vcpu.run; |
2555 | ||
f0b85051 | 2556 | svm_clear_vintr(svm); |
85f455f7 | 2557 | svm->vmcb->control.int_ctl &= ~V_IRQ_MASK; |
c1150d8c DL |
2558 | /* |
2559 | * If the user space waits to inject interrupts, exit as soon as | |
2560 | * possible | |
2561 | */ | |
8061823a GN |
2562 | if (!irqchip_in_kernel(svm->vcpu.kvm) && |
2563 | kvm_run->request_interrupt_window && | |
2564 | !kvm_cpu_has_interrupt(&svm->vcpu)) { | |
e756fc62 | 2565 | ++svm->vcpu.stat.irq_window_exits; |
c1150d8c DL |
2566 | kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN; |
2567 | return 0; | |
2568 | } | |
2569 | ||
2570 | return 1; | |
2571 | } | |
2572 | ||
565d0998 ML |
2573 | static int pause_interception(struct vcpu_svm *svm) |
2574 | { | |
2575 | kvm_vcpu_on_spin(&(svm->vcpu)); | |
2576 | return 1; | |
2577 | } | |
2578 | ||
851ba692 | 2579 | static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = { |
e0231715 JR |
2580 | [SVM_EXIT_READ_CR0] = emulate_on_interception, |
2581 | [SVM_EXIT_READ_CR3] = emulate_on_interception, | |
2582 | [SVM_EXIT_READ_CR4] = emulate_on_interception, | |
2583 | [SVM_EXIT_READ_CR8] = emulate_on_interception, | |
d225157b | 2584 | [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, |
e0231715 JR |
2585 | [SVM_EXIT_WRITE_CR0] = emulate_on_interception, |
2586 | [SVM_EXIT_WRITE_CR3] = emulate_on_interception, | |
2587 | [SVM_EXIT_WRITE_CR4] = emulate_on_interception, | |
2588 | [SVM_EXIT_WRITE_CR8] = cr8_write_interception, | |
2589 | [SVM_EXIT_READ_DR0] = emulate_on_interception, | |
6aa8b732 AK |
2590 | [SVM_EXIT_READ_DR1] = emulate_on_interception, |
2591 | [SVM_EXIT_READ_DR2] = emulate_on_interception, | |
2592 | [SVM_EXIT_READ_DR3] = emulate_on_interception, | |
727f5a23 JK |
2593 | [SVM_EXIT_READ_DR4] = emulate_on_interception, |
2594 | [SVM_EXIT_READ_DR5] = emulate_on_interception, | |
2595 | [SVM_EXIT_READ_DR6] = emulate_on_interception, | |
2596 | [SVM_EXIT_READ_DR7] = emulate_on_interception, | |
6aa8b732 AK |
2597 | [SVM_EXIT_WRITE_DR0] = emulate_on_interception, |
2598 | [SVM_EXIT_WRITE_DR1] = emulate_on_interception, | |
2599 | [SVM_EXIT_WRITE_DR2] = emulate_on_interception, | |
2600 | [SVM_EXIT_WRITE_DR3] = emulate_on_interception, | |
727f5a23 | 2601 | [SVM_EXIT_WRITE_DR4] = emulate_on_interception, |
6aa8b732 | 2602 | [SVM_EXIT_WRITE_DR5] = emulate_on_interception, |
727f5a23 | 2603 | [SVM_EXIT_WRITE_DR6] = emulate_on_interception, |
6aa8b732 | 2604 | [SVM_EXIT_WRITE_DR7] = emulate_on_interception, |
d0bfb940 JK |
2605 | [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception, |
2606 | [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception, | |
7aa81cc0 | 2607 | [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception, |
e0231715 JR |
2608 | [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception, |
2609 | [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception, | |
2610 | [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception, | |
2611 | [SVM_EXIT_INTR] = intr_interception, | |
c47f098d | 2612 | [SVM_EXIT_NMI] = nmi_interception, |
6aa8b732 AK |
2613 | [SVM_EXIT_SMI] = nop_on_interception, |
2614 | [SVM_EXIT_INIT] = nop_on_interception, | |
c1150d8c | 2615 | [SVM_EXIT_VINTR] = interrupt_window_interception, |
6aa8b732 | 2616 | [SVM_EXIT_CPUID] = cpuid_interception, |
95ba8273 | 2617 | [SVM_EXIT_IRET] = iret_interception, |
cf5a94d1 | 2618 | [SVM_EXIT_INVD] = emulate_on_interception, |
565d0998 | 2619 | [SVM_EXIT_PAUSE] = pause_interception, |
6aa8b732 | 2620 | [SVM_EXIT_HLT] = halt_interception, |
a7052897 | 2621 | [SVM_EXIT_INVLPG] = invlpg_interception, |
ff092385 | 2622 | [SVM_EXIT_INVLPGA] = invlpga_interception, |
e0231715 | 2623 | [SVM_EXIT_IOIO] = io_interception, |
6aa8b732 AK |
2624 | [SVM_EXIT_MSR] = msr_interception, |
2625 | [SVM_EXIT_TASK_SWITCH] = task_switch_interception, | |
46fe4ddd | 2626 | [SVM_EXIT_SHUTDOWN] = shutdown_interception, |
3d6368ef | 2627 | [SVM_EXIT_VMRUN] = vmrun_interception, |
02e235bc | 2628 | [SVM_EXIT_VMMCALL] = vmmcall_interception, |
5542675b AG |
2629 | [SVM_EXIT_VMLOAD] = vmload_interception, |
2630 | [SVM_EXIT_VMSAVE] = vmsave_interception, | |
1371d904 AG |
2631 | [SVM_EXIT_STGI] = stgi_interception, |
2632 | [SVM_EXIT_CLGI] = clgi_interception, | |
532a46b9 | 2633 | [SVM_EXIT_SKINIT] = skinit_interception, |
cf5a94d1 | 2634 | [SVM_EXIT_WBINVD] = emulate_on_interception, |
916ce236 JR |
2635 | [SVM_EXIT_MONITOR] = invalid_op_interception, |
2636 | [SVM_EXIT_MWAIT] = invalid_op_interception, | |
709ddebf | 2637 | [SVM_EXIT_NPF] = pf_interception, |
6aa8b732 AK |
2638 | }; |
2639 | ||
851ba692 | 2640 | static int handle_exit(struct kvm_vcpu *vcpu) |
6aa8b732 | 2641 | { |
04d2cc77 | 2642 | struct vcpu_svm *svm = to_svm(vcpu); |
851ba692 | 2643 | struct kvm_run *kvm_run = vcpu->run; |
a2fa3e9f | 2644 | u32 exit_code = svm->vmcb->control.exit_code; |
6aa8b732 | 2645 | |
5bfd8b54 | 2646 | trace_kvm_exit(exit_code, vcpu); |
af9ca2d7 | 2647 | |
2be4fc7a JR |
2648 | if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR0_MASK)) |
2649 | vcpu->arch.cr0 = svm->vmcb->save.cr0; | |
2650 | if (npt_enabled) | |
2651 | vcpu->arch.cr3 = svm->vmcb->save.cr3; | |
2652 | ||
cd3ff653 JR |
2653 | if (unlikely(svm->nested.exit_required)) { |
2654 | nested_svm_vmexit(svm); | |
2655 | svm->nested.exit_required = false; | |
2656 | ||
2657 | return 1; | |
2658 | } | |
2659 | ||
cf74a78b | 2660 | if (is_nested(svm)) { |
410e4d57 JR |
2661 | int vmexit; |
2662 | ||
d8cabddf JR |
2663 | trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code, |
2664 | svm->vmcb->control.exit_info_1, | |
2665 | svm->vmcb->control.exit_info_2, | |
2666 | svm->vmcb->control.exit_int_info, | |
2667 | svm->vmcb->control.exit_int_info_err); | |
2668 | ||
410e4d57 JR |
2669 | vmexit = nested_svm_exit_special(svm); |
2670 | ||
2671 | if (vmexit == NESTED_EXIT_CONTINUE) | |
2672 | vmexit = nested_svm_exit_handled(svm); | |
2673 | ||
2674 | if (vmexit == NESTED_EXIT_DONE) | |
cf74a78b | 2675 | return 1; |
cf74a78b AG |
2676 | } |
2677 | ||
a5c3832d JR |
2678 | svm_complete_interrupts(svm); |
2679 | ||
04d2cc77 AK |
2680 | if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) { |
2681 | kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY; | |
2682 | kvm_run->fail_entry.hardware_entry_failure_reason | |
2683 | = svm->vmcb->control.exit_code; | |
2684 | return 0; | |
2685 | } | |
2686 | ||
a2fa3e9f | 2687 | if (is_external_interrupt(svm->vmcb->control.exit_int_info) && |
709ddebf | 2688 | exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR && |
fe8e7f83 | 2689 | exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH) |
6aa8b732 AK |
2690 | printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x " |
2691 | "exit_code 0x%x\n", | |
b8688d51 | 2692 | __func__, svm->vmcb->control.exit_int_info, |
6aa8b732 AK |
2693 | exit_code); |
2694 | ||
9d8f549d | 2695 | if (exit_code >= ARRAY_SIZE(svm_exit_handlers) |
56919c5c | 2696 | || !svm_exit_handlers[exit_code]) { |
6aa8b732 | 2697 | kvm_run->exit_reason = KVM_EXIT_UNKNOWN; |
364b625b | 2698 | kvm_run->hw.hardware_exit_reason = exit_code; |
6aa8b732 AK |
2699 | return 0; |
2700 | } | |
2701 | ||
851ba692 | 2702 | return svm_exit_handlers[exit_code](svm); |
6aa8b732 AK |
2703 | } |
2704 | ||
2705 | static void reload_tss(struct kvm_vcpu *vcpu) | |
2706 | { | |
2707 | int cpu = raw_smp_processor_id(); | |
2708 | ||
0fe1e009 TH |
2709 | struct svm_cpu_data *sd = per_cpu(svm_data, cpu); |
2710 | sd->tss_desc->type = 9; /* available 32/64-bit TSS */ | |
6aa8b732 AK |
2711 | load_TR_desc(); |
2712 | } | |
2713 | ||
e756fc62 | 2714 | static void pre_svm_run(struct vcpu_svm *svm) |
6aa8b732 AK |
2715 | { |
2716 | int cpu = raw_smp_processor_id(); | |
2717 | ||
0fe1e009 | 2718 | struct svm_cpu_data *sd = per_cpu(svm_data, cpu); |
6aa8b732 | 2719 | |
a2fa3e9f | 2720 | svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING; |
4b656b12 | 2721 | /* FIXME: handle wraparound of asid_generation */ |
0fe1e009 TH |
2722 | if (svm->asid_generation != sd->asid_generation) |
2723 | new_asid(svm, sd); | |
6aa8b732 AK |
2724 | } |
2725 | ||
95ba8273 GN |
2726 | static void svm_inject_nmi(struct kvm_vcpu *vcpu) |
2727 | { | |
2728 | struct vcpu_svm *svm = to_svm(vcpu); | |
2729 | ||
2730 | svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI; | |
2731 | vcpu->arch.hflags |= HF_NMI_MASK; | |
2732 | svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET); | |
2733 | ++vcpu->stat.nmi_injections; | |
2734 | } | |
6aa8b732 | 2735 | |
85f455f7 | 2736 | static inline void svm_inject_irq(struct vcpu_svm *svm, int irq) |
6aa8b732 AK |
2737 | { |
2738 | struct vmcb_control_area *control; | |
2739 | ||
229456fc | 2740 | trace_kvm_inj_virq(irq); |
af9ca2d7 | 2741 | |
fa89a817 | 2742 | ++svm->vcpu.stat.irq_injections; |
e756fc62 | 2743 | control = &svm->vmcb->control; |
85f455f7 | 2744 | control->int_vector = irq; |
6aa8b732 AK |
2745 | control->int_ctl &= ~V_INTR_PRIO_MASK; |
2746 | control->int_ctl |= V_IRQ_MASK | | |
2747 | ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT); | |
2748 | } | |
2749 | ||
66fd3f7f | 2750 | static void svm_set_irq(struct kvm_vcpu *vcpu) |
2a8067f1 ED |
2751 | { |
2752 | struct vcpu_svm *svm = to_svm(vcpu); | |
2753 | ||
2af9194d | 2754 | BUG_ON(!(gif_set(svm))); |
cf74a78b | 2755 | |
219b65dc AG |
2756 | svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr | |
2757 | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR; | |
2a8067f1 ED |
2758 | } |
2759 | ||
95ba8273 | 2760 | static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr) |
aaacfc9a JR |
2761 | { |
2762 | struct vcpu_svm *svm = to_svm(vcpu); | |
aaacfc9a | 2763 | |
88ab24ad JR |
2764 | if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK)) |
2765 | return; | |
2766 | ||
95ba8273 | 2767 | if (irr == -1) |
aaacfc9a JR |
2768 | return; |
2769 | ||
95ba8273 GN |
2770 | if (tpr >= irr) |
2771 | svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK; | |
2772 | } | |
aaacfc9a | 2773 | |
95ba8273 GN |
2774 | static int svm_nmi_allowed(struct kvm_vcpu *vcpu) |
2775 | { | |
2776 | struct vcpu_svm *svm = to_svm(vcpu); | |
2777 | struct vmcb *vmcb = svm->vmcb; | |
924584cc JR |
2778 | int ret; |
2779 | ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) && | |
2780 | !(svm->vcpu.arch.hflags & HF_NMI_MASK); | |
2781 | ret = ret && gif_set(svm) && nested_svm_nmi(svm); | |
2782 | ||
2783 | return ret; | |
aaacfc9a JR |
2784 | } |
2785 | ||
3cfc3092 JK |
2786 | static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu) |
2787 | { | |
2788 | struct vcpu_svm *svm = to_svm(vcpu); | |
2789 | ||
2790 | return !!(svm->vcpu.arch.hflags & HF_NMI_MASK); | |
2791 | } | |
2792 | ||
2793 | static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked) | |
2794 | { | |
2795 | struct vcpu_svm *svm = to_svm(vcpu); | |
2796 | ||
2797 | if (masked) { | |
2798 | svm->vcpu.arch.hflags |= HF_NMI_MASK; | |
2799 | svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET); | |
2800 | } else { | |
2801 | svm->vcpu.arch.hflags &= ~HF_NMI_MASK; | |
2802 | svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET); | |
2803 | } | |
2804 | } | |
2805 | ||
78646121 GN |
2806 | static int svm_interrupt_allowed(struct kvm_vcpu *vcpu) |
2807 | { | |
2808 | struct vcpu_svm *svm = to_svm(vcpu); | |
2809 | struct vmcb *vmcb = svm->vmcb; | |
7fcdb510 JR |
2810 | int ret; |
2811 | ||
2812 | if (!gif_set(svm) || | |
2813 | (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)) | |
2814 | return 0; | |
2815 | ||
2816 | ret = !!(vmcb->save.rflags & X86_EFLAGS_IF); | |
2817 | ||
2818 | if (is_nested(svm)) | |
2819 | return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK); | |
2820 | ||
2821 | return ret; | |
78646121 GN |
2822 | } |
2823 | ||
9222be18 | 2824 | static void enable_irq_window(struct kvm_vcpu *vcpu) |
6aa8b732 | 2825 | { |
219b65dc | 2826 | struct vcpu_svm *svm = to_svm(vcpu); |
219b65dc | 2827 | |
e0231715 JR |
2828 | /* |
2829 | * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes | |
2830 | * 1, because that's a separate STGI/VMRUN intercept. The next time we | |
2831 | * get that intercept, this function will be called again though and | |
2832 | * we'll get the vintr intercept. | |
2833 | */ | |
8fe54654 | 2834 | if (gif_set(svm) && nested_svm_intr(svm)) { |
219b65dc AG |
2835 | svm_set_vintr(svm); |
2836 | svm_inject_irq(svm, 0x0); | |
2837 | } | |
85f455f7 ED |
2838 | } |
2839 | ||
95ba8273 | 2840 | static void enable_nmi_window(struct kvm_vcpu *vcpu) |
c1150d8c | 2841 | { |
04d2cc77 | 2842 | struct vcpu_svm *svm = to_svm(vcpu); |
c1150d8c | 2843 | |
44c11430 GN |
2844 | if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK)) |
2845 | == HF_NMI_MASK) | |
2846 | return; /* IRET will cause a vm exit */ | |
2847 | ||
e0231715 JR |
2848 | /* |
2849 | * Something prevents NMI from been injected. Single step over possible | |
2850 | * problem (IRET or exception injection or interrupt shadow) | |
2851 | */ | |
924584cc JR |
2852 | svm->nmi_singlestep = true; |
2853 | svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF); | |
2854 | update_db_intercept(vcpu); | |
c1150d8c DL |
2855 | } |
2856 | ||
cbc94022 IE |
2857 | static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr) |
2858 | { | |
2859 | return 0; | |
2860 | } | |
2861 | ||
d9e368d6 AK |
2862 | static void svm_flush_tlb(struct kvm_vcpu *vcpu) |
2863 | { | |
2864 | force_new_asid(vcpu); | |
2865 | } | |
2866 | ||
04d2cc77 AK |
2867 | static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu) |
2868 | { | |
2869 | } | |
2870 | ||
d7bf8221 JR |
2871 | static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu) |
2872 | { | |
2873 | struct vcpu_svm *svm = to_svm(vcpu); | |
2874 | ||
88ab24ad JR |
2875 | if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK)) |
2876 | return; | |
2877 | ||
d7bf8221 JR |
2878 | if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) { |
2879 | int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK; | |
615d5193 | 2880 | kvm_set_cr8(vcpu, cr8); |
d7bf8221 JR |
2881 | } |
2882 | } | |
2883 | ||
649d6864 JR |
2884 | static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu) |
2885 | { | |
2886 | struct vcpu_svm *svm = to_svm(vcpu); | |
2887 | u64 cr8; | |
2888 | ||
88ab24ad JR |
2889 | if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK)) |
2890 | return; | |
2891 | ||
649d6864 JR |
2892 | cr8 = kvm_get_cr8(vcpu); |
2893 | svm->vmcb->control.int_ctl &= ~V_TPR_MASK; | |
2894 | svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK; | |
2895 | } | |
2896 | ||
9222be18 GN |
2897 | static void svm_complete_interrupts(struct vcpu_svm *svm) |
2898 | { | |
2899 | u8 vector; | |
2900 | int type; | |
2901 | u32 exitintinfo = svm->vmcb->control.exit_int_info; | |
66b7138f JK |
2902 | unsigned int3_injected = svm->int3_injected; |
2903 | ||
2904 | svm->int3_injected = 0; | |
9222be18 | 2905 | |
44c11430 GN |
2906 | if (svm->vcpu.arch.hflags & HF_IRET_MASK) |
2907 | svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK); | |
2908 | ||
9222be18 GN |
2909 | svm->vcpu.arch.nmi_injected = false; |
2910 | kvm_clear_exception_queue(&svm->vcpu); | |
2911 | kvm_clear_interrupt_queue(&svm->vcpu); | |
2912 | ||
2913 | if (!(exitintinfo & SVM_EXITINTINFO_VALID)) | |
2914 | return; | |
2915 | ||
2916 | vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK; | |
2917 | type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK; | |
2918 | ||
2919 | switch (type) { | |
2920 | case SVM_EXITINTINFO_TYPE_NMI: | |
2921 | svm->vcpu.arch.nmi_injected = true; | |
2922 | break; | |
2923 | case SVM_EXITINTINFO_TYPE_EXEPT: | |
66b7138f JK |
2924 | /* |
2925 | * In case of software exceptions, do not reinject the vector, | |
2926 | * but re-execute the instruction instead. Rewind RIP first | |
2927 | * if we emulated INT3 before. | |
2928 | */ | |
2929 | if (kvm_exception_is_soft(vector)) { | |
2930 | if (vector == BP_VECTOR && int3_injected && | |
2931 | kvm_is_linear_rip(&svm->vcpu, svm->int3_rip)) | |
2932 | kvm_rip_write(&svm->vcpu, | |
2933 | kvm_rip_read(&svm->vcpu) - | |
2934 | int3_injected); | |
9222be18 | 2935 | break; |
66b7138f | 2936 | } |
9222be18 GN |
2937 | if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) { |
2938 | u32 err = svm->vmcb->control.exit_int_info_err; | |
ce7ddec4 | 2939 | kvm_requeue_exception_e(&svm->vcpu, vector, err); |
9222be18 GN |
2940 | |
2941 | } else | |
ce7ddec4 | 2942 | kvm_requeue_exception(&svm->vcpu, vector); |
9222be18 GN |
2943 | break; |
2944 | case SVM_EXITINTINFO_TYPE_INTR: | |
66fd3f7f | 2945 | kvm_queue_interrupt(&svm->vcpu, vector, false); |
9222be18 GN |
2946 | break; |
2947 | default: | |
2948 | break; | |
2949 | } | |
2950 | } | |
2951 | ||
80e31d4f AK |
2952 | #ifdef CONFIG_X86_64 |
2953 | #define R "r" | |
2954 | #else | |
2955 | #define R "e" | |
2956 | #endif | |
2957 | ||
851ba692 | 2958 | static void svm_vcpu_run(struct kvm_vcpu *vcpu) |
6aa8b732 | 2959 | { |
a2fa3e9f | 2960 | struct vcpu_svm *svm = to_svm(vcpu); |
6aa8b732 AK |
2961 | u16 fs_selector; |
2962 | u16 gs_selector; | |
2963 | u16 ldt_selector; | |
d9e368d6 | 2964 | |
2041a06a JR |
2965 | svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX]; |
2966 | svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP]; | |
2967 | svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP]; | |
2968 | ||
cd3ff653 JR |
2969 | /* |
2970 | * A vmexit emulation is required before the vcpu can be executed | |
2971 | * again. | |
2972 | */ | |
2973 | if (unlikely(svm->nested.exit_required)) | |
2974 | return; | |
2975 | ||
e756fc62 | 2976 | pre_svm_run(svm); |
6aa8b732 | 2977 | |
649d6864 JR |
2978 | sync_lapic_to_cr8(vcpu); |
2979 | ||
6aa8b732 | 2980 | save_host_msrs(vcpu); |
d6e88aec AK |
2981 | fs_selector = kvm_read_fs(); |
2982 | gs_selector = kvm_read_gs(); | |
2983 | ldt_selector = kvm_read_ldt(); | |
cda0ffdd | 2984 | svm->vmcb->save.cr2 = vcpu->arch.cr2; |
709ddebf JR |
2985 | /* required for live migration with NPT */ |
2986 | if (npt_enabled) | |
2987 | svm->vmcb->save.cr3 = vcpu->arch.cr3; | |
6aa8b732 | 2988 | |
04d2cc77 AK |
2989 | clgi(); |
2990 | ||
2991 | local_irq_enable(); | |
36241b8c | 2992 | |
6aa8b732 | 2993 | asm volatile ( |
80e31d4f AK |
2994 | "push %%"R"bp; \n\t" |
2995 | "mov %c[rbx](%[svm]), %%"R"bx \n\t" | |
2996 | "mov %c[rcx](%[svm]), %%"R"cx \n\t" | |
2997 | "mov %c[rdx](%[svm]), %%"R"dx \n\t" | |
2998 | "mov %c[rsi](%[svm]), %%"R"si \n\t" | |
2999 | "mov %c[rdi](%[svm]), %%"R"di \n\t" | |
3000 | "mov %c[rbp](%[svm]), %%"R"bp \n\t" | |
05b3e0c2 | 3001 | #ifdef CONFIG_X86_64 |
fb3f0f51 RR |
3002 | "mov %c[r8](%[svm]), %%r8 \n\t" |
3003 | "mov %c[r9](%[svm]), %%r9 \n\t" | |
3004 | "mov %c[r10](%[svm]), %%r10 \n\t" | |
3005 | "mov %c[r11](%[svm]), %%r11 \n\t" | |
3006 | "mov %c[r12](%[svm]), %%r12 \n\t" | |
3007 | "mov %c[r13](%[svm]), %%r13 \n\t" | |
3008 | "mov %c[r14](%[svm]), %%r14 \n\t" | |
3009 | "mov %c[r15](%[svm]), %%r15 \n\t" | |
6aa8b732 AK |
3010 | #endif |
3011 | ||
6aa8b732 | 3012 | /* Enter guest mode */ |
80e31d4f AK |
3013 | "push %%"R"ax \n\t" |
3014 | "mov %c[vmcb](%[svm]), %%"R"ax \n\t" | |
4ecac3fd AK |
3015 | __ex(SVM_VMLOAD) "\n\t" |
3016 | __ex(SVM_VMRUN) "\n\t" | |
3017 | __ex(SVM_VMSAVE) "\n\t" | |
80e31d4f | 3018 | "pop %%"R"ax \n\t" |
6aa8b732 AK |
3019 | |
3020 | /* Save guest registers, load host registers */ | |
80e31d4f AK |
3021 | "mov %%"R"bx, %c[rbx](%[svm]) \n\t" |
3022 | "mov %%"R"cx, %c[rcx](%[svm]) \n\t" | |
3023 | "mov %%"R"dx, %c[rdx](%[svm]) \n\t" | |
3024 | "mov %%"R"si, %c[rsi](%[svm]) \n\t" | |
3025 | "mov %%"R"di, %c[rdi](%[svm]) \n\t" | |
3026 | "mov %%"R"bp, %c[rbp](%[svm]) \n\t" | |
05b3e0c2 | 3027 | #ifdef CONFIG_X86_64 |
fb3f0f51 RR |
3028 | "mov %%r8, %c[r8](%[svm]) \n\t" |
3029 | "mov %%r9, %c[r9](%[svm]) \n\t" | |
3030 | "mov %%r10, %c[r10](%[svm]) \n\t" | |
3031 | "mov %%r11, %c[r11](%[svm]) \n\t" | |
3032 | "mov %%r12, %c[r12](%[svm]) \n\t" | |
3033 | "mov %%r13, %c[r13](%[svm]) \n\t" | |
3034 | "mov %%r14, %c[r14](%[svm]) \n\t" | |
3035 | "mov %%r15, %c[r15](%[svm]) \n\t" | |
6aa8b732 | 3036 | #endif |
80e31d4f | 3037 | "pop %%"R"bp" |
6aa8b732 | 3038 | : |
fb3f0f51 | 3039 | : [svm]"a"(svm), |
6aa8b732 | 3040 | [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)), |
ad312c7c ZX |
3041 | [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])), |
3042 | [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])), | |
3043 | [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])), | |
3044 | [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])), | |
3045 | [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])), | |
3046 | [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP])) | |
05b3e0c2 | 3047 | #ifdef CONFIG_X86_64 |
ad312c7c ZX |
3048 | , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])), |
3049 | [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])), | |
3050 | [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])), | |
3051 | [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])), | |
3052 | [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])), | |
3053 | [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])), | |
3054 | [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])), | |
3055 | [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15])) | |
6aa8b732 | 3056 | #endif |
54a08c04 | 3057 | : "cc", "memory" |
80e31d4f | 3058 | , R"bx", R"cx", R"dx", R"si", R"di" |
54a08c04 | 3059 | #ifdef CONFIG_X86_64 |
54a08c04 LV |
3060 | , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15" |
3061 | #endif | |
3062 | ); | |
6aa8b732 | 3063 | |
ad312c7c | 3064 | vcpu->arch.cr2 = svm->vmcb->save.cr2; |
5fdbf976 MT |
3065 | vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax; |
3066 | vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp; | |
3067 | vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip; | |
6aa8b732 | 3068 | |
d6e88aec AK |
3069 | kvm_load_fs(fs_selector); |
3070 | kvm_load_gs(gs_selector); | |
3071 | kvm_load_ldt(ldt_selector); | |
6aa8b732 AK |
3072 | load_host_msrs(vcpu); |
3073 | ||
3074 | reload_tss(vcpu); | |
3075 | ||
56ba47dd AK |
3076 | local_irq_disable(); |
3077 | ||
3078 | stgi(); | |
3079 | ||
d7bf8221 JR |
3080 | sync_cr8_to_lapic(vcpu); |
3081 | ||
a2fa3e9f | 3082 | svm->next_rip = 0; |
9222be18 | 3083 | |
6de4f3ad AK |
3084 | if (npt_enabled) { |
3085 | vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR); | |
3086 | vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR); | |
3087 | } | |
6aa8b732 AK |
3088 | } |
3089 | ||
80e31d4f AK |
3090 | #undef R |
3091 | ||
6aa8b732 AK |
3092 | static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root) |
3093 | { | |
a2fa3e9f GH |
3094 | struct vcpu_svm *svm = to_svm(vcpu); |
3095 | ||
709ddebf JR |
3096 | if (npt_enabled) { |
3097 | svm->vmcb->control.nested_cr3 = root; | |
3098 | force_new_asid(vcpu); | |
3099 | return; | |
3100 | } | |
3101 | ||
a2fa3e9f | 3102 | svm->vmcb->save.cr3 = root; |
6aa8b732 AK |
3103 | force_new_asid(vcpu); |
3104 | } | |
3105 | ||
6aa8b732 AK |
3106 | static int is_disabled(void) |
3107 | { | |
6031a61c JR |
3108 | u64 vm_cr; |
3109 | ||
3110 | rdmsrl(MSR_VM_CR, vm_cr); | |
3111 | if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE)) | |
3112 | return 1; | |
3113 | ||
6aa8b732 AK |
3114 | return 0; |
3115 | } | |
3116 | ||
102d8325 IM |
3117 | static void |
3118 | svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall) | |
3119 | { | |
3120 | /* | |
3121 | * Patch in the VMMCALL instruction: | |
3122 | */ | |
3123 | hypercall[0] = 0x0f; | |
3124 | hypercall[1] = 0x01; | |
3125 | hypercall[2] = 0xd9; | |
102d8325 IM |
3126 | } |
3127 | ||
002c7f7c YS |
3128 | static void svm_check_processor_compat(void *rtn) |
3129 | { | |
3130 | *(int *)rtn = 0; | |
3131 | } | |
3132 | ||
774ead3a AK |
3133 | static bool svm_cpu_has_accelerated_tpr(void) |
3134 | { | |
3135 | return false; | |
3136 | } | |
3137 | ||
67253af5 SY |
3138 | static int get_npt_level(void) |
3139 | { | |
3140 | #ifdef CONFIG_X86_64 | |
3141 | return PT64_ROOT_LEVEL; | |
3142 | #else | |
3143 | return PT32E_ROOT_LEVEL; | |
3144 | #endif | |
3145 | } | |
3146 | ||
4b12f0de | 3147 | static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio) |
64d4d521 SY |
3148 | { |
3149 | return 0; | |
3150 | } | |
3151 | ||
0e851880 SY |
3152 | static void svm_cpuid_update(struct kvm_vcpu *vcpu) |
3153 | { | |
3154 | } | |
3155 | ||
d4330ef2 JR |
3156 | static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry) |
3157 | { | |
c2c63a49 JR |
3158 | switch (func) { |
3159 | case 0x8000000A: | |
3160 | entry->eax = 1; /* SVM revision 1 */ | |
3161 | entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper | |
3162 | ASID emulation to nested SVM */ | |
3163 | entry->ecx = 0; /* Reserved */ | |
3164 | entry->edx = 0; /* Do not support any additional features */ | |
3165 | ||
3166 | break; | |
3167 | } | |
d4330ef2 JR |
3168 | } |
3169 | ||
229456fc | 3170 | static const struct trace_print_flags svm_exit_reasons_str[] = { |
e0231715 JR |
3171 | { SVM_EXIT_READ_CR0, "read_cr0" }, |
3172 | { SVM_EXIT_READ_CR3, "read_cr3" }, | |
3173 | { SVM_EXIT_READ_CR4, "read_cr4" }, | |
3174 | { SVM_EXIT_READ_CR8, "read_cr8" }, | |
3175 | { SVM_EXIT_WRITE_CR0, "write_cr0" }, | |
3176 | { SVM_EXIT_WRITE_CR3, "write_cr3" }, | |
3177 | { SVM_EXIT_WRITE_CR4, "write_cr4" }, | |
3178 | { SVM_EXIT_WRITE_CR8, "write_cr8" }, | |
3179 | { SVM_EXIT_READ_DR0, "read_dr0" }, | |
3180 | { SVM_EXIT_READ_DR1, "read_dr1" }, | |
3181 | { SVM_EXIT_READ_DR2, "read_dr2" }, | |
3182 | { SVM_EXIT_READ_DR3, "read_dr3" }, | |
3183 | { SVM_EXIT_WRITE_DR0, "write_dr0" }, | |
3184 | { SVM_EXIT_WRITE_DR1, "write_dr1" }, | |
3185 | { SVM_EXIT_WRITE_DR2, "write_dr2" }, | |
3186 | { SVM_EXIT_WRITE_DR3, "write_dr3" }, | |
3187 | { SVM_EXIT_WRITE_DR5, "write_dr5" }, | |
3188 | { SVM_EXIT_WRITE_DR7, "write_dr7" }, | |
229456fc MT |
3189 | { SVM_EXIT_EXCP_BASE + DB_VECTOR, "DB excp" }, |
3190 | { SVM_EXIT_EXCP_BASE + BP_VECTOR, "BP excp" }, | |
3191 | { SVM_EXIT_EXCP_BASE + UD_VECTOR, "UD excp" }, | |
3192 | { SVM_EXIT_EXCP_BASE + PF_VECTOR, "PF excp" }, | |
3193 | { SVM_EXIT_EXCP_BASE + NM_VECTOR, "NM excp" }, | |
3194 | { SVM_EXIT_EXCP_BASE + MC_VECTOR, "MC excp" }, | |
3195 | { SVM_EXIT_INTR, "interrupt" }, | |
3196 | { SVM_EXIT_NMI, "nmi" }, | |
3197 | { SVM_EXIT_SMI, "smi" }, | |
3198 | { SVM_EXIT_INIT, "init" }, | |
3199 | { SVM_EXIT_VINTR, "vintr" }, | |
3200 | { SVM_EXIT_CPUID, "cpuid" }, | |
3201 | { SVM_EXIT_INVD, "invd" }, | |
3202 | { SVM_EXIT_HLT, "hlt" }, | |
3203 | { SVM_EXIT_INVLPG, "invlpg" }, | |
3204 | { SVM_EXIT_INVLPGA, "invlpga" }, | |
3205 | { SVM_EXIT_IOIO, "io" }, | |
3206 | { SVM_EXIT_MSR, "msr" }, | |
3207 | { SVM_EXIT_TASK_SWITCH, "task_switch" }, | |
3208 | { SVM_EXIT_SHUTDOWN, "shutdown" }, | |
3209 | { SVM_EXIT_VMRUN, "vmrun" }, | |
3210 | { SVM_EXIT_VMMCALL, "hypercall" }, | |
3211 | { SVM_EXIT_VMLOAD, "vmload" }, | |
3212 | { SVM_EXIT_VMSAVE, "vmsave" }, | |
3213 | { SVM_EXIT_STGI, "stgi" }, | |
3214 | { SVM_EXIT_CLGI, "clgi" }, | |
3215 | { SVM_EXIT_SKINIT, "skinit" }, | |
3216 | { SVM_EXIT_WBINVD, "wbinvd" }, | |
3217 | { SVM_EXIT_MONITOR, "monitor" }, | |
3218 | { SVM_EXIT_MWAIT, "mwait" }, | |
3219 | { SVM_EXIT_NPF, "npf" }, | |
3220 | { -1, NULL } | |
3221 | }; | |
3222 | ||
17cc3935 | 3223 | static int svm_get_lpage_level(void) |
344f414f | 3224 | { |
17cc3935 | 3225 | return PT_PDPE_LEVEL; |
344f414f JR |
3226 | } |
3227 | ||
4e47c7a6 SY |
3228 | static bool svm_rdtscp_supported(void) |
3229 | { | |
3230 | return false; | |
3231 | } | |
3232 | ||
02daab21 AK |
3233 | static void svm_fpu_deactivate(struct kvm_vcpu *vcpu) |
3234 | { | |
3235 | struct vcpu_svm *svm = to_svm(vcpu); | |
3236 | ||
02daab21 | 3237 | svm->vmcb->control.intercept_exceptions |= 1 << NM_VECTOR; |
66a562f7 JR |
3238 | if (is_nested(svm)) |
3239 | svm->nested.hsave->control.intercept_exceptions |= 1 << NM_VECTOR; | |
3240 | update_cr0_intercept(svm); | |
02daab21 AK |
3241 | } |
3242 | ||
cbdd1bea | 3243 | static struct kvm_x86_ops svm_x86_ops = { |
6aa8b732 AK |
3244 | .cpu_has_kvm_support = has_svm, |
3245 | .disabled_by_bios = is_disabled, | |
3246 | .hardware_setup = svm_hardware_setup, | |
3247 | .hardware_unsetup = svm_hardware_unsetup, | |
002c7f7c | 3248 | .check_processor_compatibility = svm_check_processor_compat, |
6aa8b732 AK |
3249 | .hardware_enable = svm_hardware_enable, |
3250 | .hardware_disable = svm_hardware_disable, | |
774ead3a | 3251 | .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr, |
6aa8b732 AK |
3252 | |
3253 | .vcpu_create = svm_create_vcpu, | |
3254 | .vcpu_free = svm_free_vcpu, | |
04d2cc77 | 3255 | .vcpu_reset = svm_vcpu_reset, |
6aa8b732 | 3256 | |
04d2cc77 | 3257 | .prepare_guest_switch = svm_prepare_guest_switch, |
6aa8b732 AK |
3258 | .vcpu_load = svm_vcpu_load, |
3259 | .vcpu_put = svm_vcpu_put, | |
3260 | ||
3261 | .set_guest_debug = svm_guest_debug, | |
3262 | .get_msr = svm_get_msr, | |
3263 | .set_msr = svm_set_msr, | |
3264 | .get_segment_base = svm_get_segment_base, | |
3265 | .get_segment = svm_get_segment, | |
3266 | .set_segment = svm_set_segment, | |
2e4d2653 | 3267 | .get_cpl = svm_get_cpl, |
1747fb71 | 3268 | .get_cs_db_l_bits = kvm_get_cs_db_l_bits, |
e8467fda | 3269 | .decache_cr0_guest_bits = svm_decache_cr0_guest_bits, |
25c4c276 | 3270 | .decache_cr4_guest_bits = svm_decache_cr4_guest_bits, |
6aa8b732 | 3271 | .set_cr0 = svm_set_cr0, |
6aa8b732 AK |
3272 | .set_cr3 = svm_set_cr3, |
3273 | .set_cr4 = svm_set_cr4, | |
3274 | .set_efer = svm_set_efer, | |
3275 | .get_idt = svm_get_idt, | |
3276 | .set_idt = svm_set_idt, | |
3277 | .get_gdt = svm_get_gdt, | |
3278 | .set_gdt = svm_set_gdt, | |
020df079 | 3279 | .set_dr7 = svm_set_dr7, |
6de4f3ad | 3280 | .cache_reg = svm_cache_reg, |
6aa8b732 AK |
3281 | .get_rflags = svm_get_rflags, |
3282 | .set_rflags = svm_set_rflags, | |
6b52d186 | 3283 | .fpu_activate = svm_fpu_activate, |
02daab21 | 3284 | .fpu_deactivate = svm_fpu_deactivate, |
6aa8b732 | 3285 | |
6aa8b732 | 3286 | .tlb_flush = svm_flush_tlb, |
6aa8b732 | 3287 | |
6aa8b732 | 3288 | .run = svm_vcpu_run, |
04d2cc77 | 3289 | .handle_exit = handle_exit, |
6aa8b732 | 3290 | .skip_emulated_instruction = skip_emulated_instruction, |
2809f5d2 GC |
3291 | .set_interrupt_shadow = svm_set_interrupt_shadow, |
3292 | .get_interrupt_shadow = svm_get_interrupt_shadow, | |
102d8325 | 3293 | .patch_hypercall = svm_patch_hypercall, |
2a8067f1 | 3294 | .set_irq = svm_set_irq, |
95ba8273 | 3295 | .set_nmi = svm_inject_nmi, |
298101da | 3296 | .queue_exception = svm_queue_exception, |
78646121 | 3297 | .interrupt_allowed = svm_interrupt_allowed, |
95ba8273 | 3298 | .nmi_allowed = svm_nmi_allowed, |
3cfc3092 JK |
3299 | .get_nmi_mask = svm_get_nmi_mask, |
3300 | .set_nmi_mask = svm_set_nmi_mask, | |
95ba8273 GN |
3301 | .enable_nmi_window = enable_nmi_window, |
3302 | .enable_irq_window = enable_irq_window, | |
3303 | .update_cr8_intercept = update_cr8_intercept, | |
cbc94022 IE |
3304 | |
3305 | .set_tss_addr = svm_set_tss_addr, | |
67253af5 | 3306 | .get_tdp_level = get_npt_level, |
4b12f0de | 3307 | .get_mt_mask = svm_get_mt_mask, |
229456fc MT |
3308 | |
3309 | .exit_reasons_str = svm_exit_reasons_str, | |
17cc3935 | 3310 | .get_lpage_level = svm_get_lpage_level, |
0e851880 SY |
3311 | |
3312 | .cpuid_update = svm_cpuid_update, | |
4e47c7a6 SY |
3313 | |
3314 | .rdtscp_supported = svm_rdtscp_supported, | |
d4330ef2 JR |
3315 | |
3316 | .set_supported_cpuid = svm_set_supported_cpuid, | |
6aa8b732 AK |
3317 | }; |
3318 | ||
3319 | static int __init svm_init(void) | |
3320 | { | |
cb498ea2 | 3321 | return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm), |
0ee75bea | 3322 | __alignof__(struct vcpu_svm), THIS_MODULE); |
6aa8b732 AK |
3323 | } |
3324 | ||
3325 | static void __exit svm_exit(void) | |
3326 | { | |
cb498ea2 | 3327 | kvm_exit(); |
6aa8b732 AK |
3328 | } |
3329 | ||
3330 | module_init(svm_init) | |
3331 | module_exit(svm_exit) |