Commit | Line | Data |
---|---|---|
043405e1 CO |
1 | /* |
2 | * Kernel-based Virtual Machine driver for Linux | |
3 | * | |
4 | * derived from drivers/kvm/kvm_main.c | |
5 | * | |
6 | * Copyright (C) 2006 Qumranet, Inc. | |
4d5c5d0f BAY |
7 | * Copyright (C) 2008 Qumranet, Inc. |
8 | * Copyright IBM Corporation, 2008 | |
9611c187 | 9 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
043405e1 CO |
10 | * |
11 | * Authors: | |
12 | * Avi Kivity <avi@qumranet.com> | |
13 | * Yaniv Kamay <yaniv@qumranet.com> | |
4d5c5d0f BAY |
14 | * Amit Shah <amit.shah@qumranet.com> |
15 | * Ben-Ami Yassour <benami@il.ibm.com> | |
043405e1 CO |
16 | * |
17 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
18 | * the COPYING file in the top-level directory. | |
19 | * | |
20 | */ | |
21 | ||
edf88417 | 22 | #include <linux/kvm_host.h> |
313a3dc7 | 23 | #include "irq.h" |
1d737c8a | 24 | #include "mmu.h" |
7837699f | 25 | #include "i8254.h" |
37817f29 | 26 | #include "tss.h" |
5fdbf976 | 27 | #include "kvm_cache_regs.h" |
26eef70c | 28 | #include "x86.h" |
00b27a3e | 29 | #include "cpuid.h" |
313a3dc7 | 30 | |
18068523 | 31 | #include <linux/clocksource.h> |
4d5c5d0f | 32 | #include <linux/interrupt.h> |
313a3dc7 CO |
33 | #include <linux/kvm.h> |
34 | #include <linux/fs.h> | |
35 | #include <linux/vmalloc.h> | |
5fb76f9b | 36 | #include <linux/module.h> |
0de10343 | 37 | #include <linux/mman.h> |
2bacc55c | 38 | #include <linux/highmem.h> |
19de40a8 | 39 | #include <linux/iommu.h> |
62c476c7 | 40 | #include <linux/intel-iommu.h> |
c8076604 | 41 | #include <linux/cpufreq.h> |
18863bdd | 42 | #include <linux/user-return-notifier.h> |
a983fb23 | 43 | #include <linux/srcu.h> |
5a0e3ad6 | 44 | #include <linux/slab.h> |
ff9d07a0 | 45 | #include <linux/perf_event.h> |
7bee342a | 46 | #include <linux/uaccess.h> |
af585b92 | 47 | #include <linux/hash.h> |
a1b60c1c | 48 | #include <linux/pci.h> |
16e8d74d MT |
49 | #include <linux/timekeeper_internal.h> |
50 | #include <linux/pvclock_gtod.h> | |
aec51dc4 | 51 | #include <trace/events/kvm.h> |
2ed152af | 52 | |
229456fc MT |
53 | #define CREATE_TRACE_POINTS |
54 | #include "trace.h" | |
043405e1 | 55 | |
24f1e32c | 56 | #include <asm/debugreg.h> |
d825ed0a | 57 | #include <asm/msr.h> |
a5f61300 | 58 | #include <asm/desc.h> |
0bed3b56 | 59 | #include <asm/mtrr.h> |
890ca9ae | 60 | #include <asm/mce.h> |
7cf30855 | 61 | #include <asm/i387.h> |
1361b83a | 62 | #include <asm/fpu-internal.h> /* Ugh! */ |
98918833 | 63 | #include <asm/xcr.h> |
1d5f066e | 64 | #include <asm/pvclock.h> |
217fc9cf | 65 | #include <asm/div64.h> |
043405e1 | 66 | |
313a3dc7 | 67 | #define MAX_IO_MSRS 256 |
890ca9ae | 68 | #define KVM_MAX_MCE_BANKS 32 |
5854dbca | 69 | #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P) |
890ca9ae | 70 | |
0f65dd70 AK |
71 | #define emul_to_vcpu(ctxt) \ |
72 | container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt) | |
73 | ||
50a37eb4 JR |
74 | /* EFER defaults: |
75 | * - enable syscall per default because its emulated by KVM | |
76 | * - enable LME and LMA per default on 64 bit KVM | |
77 | */ | |
78 | #ifdef CONFIG_X86_64 | |
1260edbe LJ |
79 | static |
80 | u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA)); | |
50a37eb4 | 81 | #else |
1260edbe | 82 | static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE); |
50a37eb4 | 83 | #endif |
313a3dc7 | 84 | |
ba1389b7 AK |
85 | #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM |
86 | #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU | |
417bc304 | 87 | |
cb142eb7 | 88 | static void update_cr8_intercept(struct kvm_vcpu *vcpu); |
7460fb4a | 89 | static void process_nmi(struct kvm_vcpu *vcpu); |
674eea0f | 90 | |
97896d04 | 91 | struct kvm_x86_ops *kvm_x86_ops; |
5fdbf976 | 92 | EXPORT_SYMBOL_GPL(kvm_x86_ops); |
97896d04 | 93 | |
476bc001 RR |
94 | static bool ignore_msrs = 0; |
95 | module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR); | |
ed85c068 | 96 | |
92a1f12d JR |
97 | bool kvm_has_tsc_control; |
98 | EXPORT_SYMBOL_GPL(kvm_has_tsc_control); | |
99 | u32 kvm_max_guest_tsc_khz; | |
100 | EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz); | |
101 | ||
cc578287 ZA |
102 | /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */ |
103 | static u32 tsc_tolerance_ppm = 250; | |
104 | module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR); | |
105 | ||
18863bdd AK |
106 | #define KVM_NR_SHARED_MSRS 16 |
107 | ||
108 | struct kvm_shared_msrs_global { | |
109 | int nr; | |
2bf78fa7 | 110 | u32 msrs[KVM_NR_SHARED_MSRS]; |
18863bdd AK |
111 | }; |
112 | ||
113 | struct kvm_shared_msrs { | |
114 | struct user_return_notifier urn; | |
115 | bool registered; | |
2bf78fa7 SY |
116 | struct kvm_shared_msr_values { |
117 | u64 host; | |
118 | u64 curr; | |
119 | } values[KVM_NR_SHARED_MSRS]; | |
18863bdd AK |
120 | }; |
121 | ||
122 | static struct kvm_shared_msrs_global __read_mostly shared_msrs_global; | |
013f6a5d | 123 | static struct kvm_shared_msrs __percpu *shared_msrs; |
18863bdd | 124 | |
417bc304 | 125 | struct kvm_stats_debugfs_item debugfs_entries[] = { |
ba1389b7 AK |
126 | { "pf_fixed", VCPU_STAT(pf_fixed) }, |
127 | { "pf_guest", VCPU_STAT(pf_guest) }, | |
128 | { "tlb_flush", VCPU_STAT(tlb_flush) }, | |
129 | { "invlpg", VCPU_STAT(invlpg) }, | |
130 | { "exits", VCPU_STAT(exits) }, | |
131 | { "io_exits", VCPU_STAT(io_exits) }, | |
132 | { "mmio_exits", VCPU_STAT(mmio_exits) }, | |
133 | { "signal_exits", VCPU_STAT(signal_exits) }, | |
134 | { "irq_window", VCPU_STAT(irq_window_exits) }, | |
f08864b4 | 135 | { "nmi_window", VCPU_STAT(nmi_window_exits) }, |
ba1389b7 AK |
136 | { "halt_exits", VCPU_STAT(halt_exits) }, |
137 | { "halt_wakeup", VCPU_STAT(halt_wakeup) }, | |
f11c3a8d | 138 | { "hypercalls", VCPU_STAT(hypercalls) }, |
ba1389b7 AK |
139 | { "request_irq", VCPU_STAT(request_irq_exits) }, |
140 | { "irq_exits", VCPU_STAT(irq_exits) }, | |
141 | { "host_state_reload", VCPU_STAT(host_state_reload) }, | |
142 | { "efer_reload", VCPU_STAT(efer_reload) }, | |
143 | { "fpu_reload", VCPU_STAT(fpu_reload) }, | |
144 | { "insn_emulation", VCPU_STAT(insn_emulation) }, | |
145 | { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) }, | |
fa89a817 | 146 | { "irq_injections", VCPU_STAT(irq_injections) }, |
c4abb7c9 | 147 | { "nmi_injections", VCPU_STAT(nmi_injections) }, |
4cee5764 AK |
148 | { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) }, |
149 | { "mmu_pte_write", VM_STAT(mmu_pte_write) }, | |
150 | { "mmu_pte_updated", VM_STAT(mmu_pte_updated) }, | |
151 | { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) }, | |
152 | { "mmu_flooded", VM_STAT(mmu_flooded) }, | |
153 | { "mmu_recycled", VM_STAT(mmu_recycled) }, | |
dfc5aa00 | 154 | { "mmu_cache_miss", VM_STAT(mmu_cache_miss) }, |
4731d4c7 | 155 | { "mmu_unsync", VM_STAT(mmu_unsync) }, |
0f74a24c | 156 | { "remote_tlb_flush", VM_STAT(remote_tlb_flush) }, |
05da4558 | 157 | { "largepages", VM_STAT(lpages) }, |
417bc304 HB |
158 | { NULL } |
159 | }; | |
160 | ||
2acf923e DC |
161 | u64 __read_mostly host_xcr0; |
162 | ||
b6785def | 163 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt); |
d6aa1000 | 164 | |
af585b92 GN |
165 | static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu) |
166 | { | |
167 | int i; | |
168 | for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++) | |
169 | vcpu->arch.apf.gfns[i] = ~0; | |
170 | } | |
171 | ||
18863bdd AK |
172 | static void kvm_on_user_return(struct user_return_notifier *urn) |
173 | { | |
174 | unsigned slot; | |
18863bdd AK |
175 | struct kvm_shared_msrs *locals |
176 | = container_of(urn, struct kvm_shared_msrs, urn); | |
2bf78fa7 | 177 | struct kvm_shared_msr_values *values; |
18863bdd AK |
178 | |
179 | for (slot = 0; slot < shared_msrs_global.nr; ++slot) { | |
2bf78fa7 SY |
180 | values = &locals->values[slot]; |
181 | if (values->host != values->curr) { | |
182 | wrmsrl(shared_msrs_global.msrs[slot], values->host); | |
183 | values->curr = values->host; | |
18863bdd AK |
184 | } |
185 | } | |
186 | locals->registered = false; | |
187 | user_return_notifier_unregister(urn); | |
188 | } | |
189 | ||
2bf78fa7 | 190 | static void shared_msr_update(unsigned slot, u32 msr) |
18863bdd | 191 | { |
18863bdd | 192 | u64 value; |
013f6a5d MT |
193 | unsigned int cpu = smp_processor_id(); |
194 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
18863bdd | 195 | |
2bf78fa7 SY |
196 | /* only read, and nobody should modify it at this time, |
197 | * so don't need lock */ | |
198 | if (slot >= shared_msrs_global.nr) { | |
199 | printk(KERN_ERR "kvm: invalid MSR slot!"); | |
200 | return; | |
201 | } | |
202 | rdmsrl_safe(msr, &value); | |
203 | smsr->values[slot].host = value; | |
204 | smsr->values[slot].curr = value; | |
205 | } | |
206 | ||
207 | void kvm_define_shared_msr(unsigned slot, u32 msr) | |
208 | { | |
18863bdd AK |
209 | if (slot >= shared_msrs_global.nr) |
210 | shared_msrs_global.nr = slot + 1; | |
2bf78fa7 SY |
211 | shared_msrs_global.msrs[slot] = msr; |
212 | /* we need ensured the shared_msr_global have been updated */ | |
213 | smp_wmb(); | |
18863bdd AK |
214 | } |
215 | EXPORT_SYMBOL_GPL(kvm_define_shared_msr); | |
216 | ||
217 | static void kvm_shared_msr_cpu_online(void) | |
218 | { | |
219 | unsigned i; | |
18863bdd AK |
220 | |
221 | for (i = 0; i < shared_msrs_global.nr; ++i) | |
2bf78fa7 | 222 | shared_msr_update(i, shared_msrs_global.msrs[i]); |
18863bdd AK |
223 | } |
224 | ||
d5696725 | 225 | void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask) |
18863bdd | 226 | { |
013f6a5d MT |
227 | unsigned int cpu = smp_processor_id(); |
228 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
18863bdd | 229 | |
2bf78fa7 | 230 | if (((value ^ smsr->values[slot].curr) & mask) == 0) |
18863bdd | 231 | return; |
2bf78fa7 SY |
232 | smsr->values[slot].curr = value; |
233 | wrmsrl(shared_msrs_global.msrs[slot], value); | |
18863bdd AK |
234 | if (!smsr->registered) { |
235 | smsr->urn.on_user_return = kvm_on_user_return; | |
236 | user_return_notifier_register(&smsr->urn); | |
237 | smsr->registered = true; | |
238 | } | |
239 | } | |
240 | EXPORT_SYMBOL_GPL(kvm_set_shared_msr); | |
241 | ||
3548bab5 AK |
242 | static void drop_user_return_notifiers(void *ignore) |
243 | { | |
013f6a5d MT |
244 | unsigned int cpu = smp_processor_id(); |
245 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
3548bab5 AK |
246 | |
247 | if (smsr->registered) | |
248 | kvm_on_user_return(&smsr->urn); | |
249 | } | |
250 | ||
6866b83e CO |
251 | u64 kvm_get_apic_base(struct kvm_vcpu *vcpu) |
252 | { | |
8a5a87d9 | 253 | return vcpu->arch.apic_base; |
6866b83e CO |
254 | } |
255 | EXPORT_SYMBOL_GPL(kvm_get_apic_base); | |
256 | ||
257 | void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data) | |
258 | { | |
259 | /* TODO: reserve bits check */ | |
8a5a87d9 | 260 | kvm_lapic_set_base(vcpu, data); |
6866b83e CO |
261 | } |
262 | EXPORT_SYMBOL_GPL(kvm_set_apic_base); | |
263 | ||
e3ba45b8 GL |
264 | asmlinkage void kvm_spurious_fault(void) |
265 | { | |
266 | /* Fault while not rebooting. We want the trace. */ | |
267 | BUG(); | |
268 | } | |
269 | EXPORT_SYMBOL_GPL(kvm_spurious_fault); | |
270 | ||
3fd28fce ED |
271 | #define EXCPT_BENIGN 0 |
272 | #define EXCPT_CONTRIBUTORY 1 | |
273 | #define EXCPT_PF 2 | |
274 | ||
275 | static int exception_class(int vector) | |
276 | { | |
277 | switch (vector) { | |
278 | case PF_VECTOR: | |
279 | return EXCPT_PF; | |
280 | case DE_VECTOR: | |
281 | case TS_VECTOR: | |
282 | case NP_VECTOR: | |
283 | case SS_VECTOR: | |
284 | case GP_VECTOR: | |
285 | return EXCPT_CONTRIBUTORY; | |
286 | default: | |
287 | break; | |
288 | } | |
289 | return EXCPT_BENIGN; | |
290 | } | |
291 | ||
292 | static void kvm_multiple_exception(struct kvm_vcpu *vcpu, | |
ce7ddec4 JR |
293 | unsigned nr, bool has_error, u32 error_code, |
294 | bool reinject) | |
3fd28fce ED |
295 | { |
296 | u32 prev_nr; | |
297 | int class1, class2; | |
298 | ||
3842d135 AK |
299 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
300 | ||
3fd28fce ED |
301 | if (!vcpu->arch.exception.pending) { |
302 | queue: | |
303 | vcpu->arch.exception.pending = true; | |
304 | vcpu->arch.exception.has_error_code = has_error; | |
305 | vcpu->arch.exception.nr = nr; | |
306 | vcpu->arch.exception.error_code = error_code; | |
3f0fd292 | 307 | vcpu->arch.exception.reinject = reinject; |
3fd28fce ED |
308 | return; |
309 | } | |
310 | ||
311 | /* to check exception */ | |
312 | prev_nr = vcpu->arch.exception.nr; | |
313 | if (prev_nr == DF_VECTOR) { | |
314 | /* triple fault -> shutdown */ | |
a8eeb04a | 315 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
3fd28fce ED |
316 | return; |
317 | } | |
318 | class1 = exception_class(prev_nr); | |
319 | class2 = exception_class(nr); | |
320 | if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY) | |
321 | || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) { | |
322 | /* generate double fault per SDM Table 5-5 */ | |
323 | vcpu->arch.exception.pending = true; | |
324 | vcpu->arch.exception.has_error_code = true; | |
325 | vcpu->arch.exception.nr = DF_VECTOR; | |
326 | vcpu->arch.exception.error_code = 0; | |
327 | } else | |
328 | /* replace previous exception with a new one in a hope | |
329 | that instruction re-execution will regenerate lost | |
330 | exception */ | |
331 | goto queue; | |
332 | } | |
333 | ||
298101da AK |
334 | void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
335 | { | |
ce7ddec4 | 336 | kvm_multiple_exception(vcpu, nr, false, 0, false); |
298101da AK |
337 | } |
338 | EXPORT_SYMBOL_GPL(kvm_queue_exception); | |
339 | ||
ce7ddec4 JR |
340 | void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
341 | { | |
342 | kvm_multiple_exception(vcpu, nr, false, 0, true); | |
343 | } | |
344 | EXPORT_SYMBOL_GPL(kvm_requeue_exception); | |
345 | ||
db8fcefa | 346 | void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err) |
c3c91fee | 347 | { |
db8fcefa AP |
348 | if (err) |
349 | kvm_inject_gp(vcpu, 0); | |
350 | else | |
351 | kvm_x86_ops->skip_emulated_instruction(vcpu); | |
352 | } | |
353 | EXPORT_SYMBOL_GPL(kvm_complete_insn_gp); | |
8df25a32 | 354 | |
6389ee94 | 355 | void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault) |
c3c91fee AK |
356 | { |
357 | ++vcpu->stat.pf_guest; | |
6389ee94 AK |
358 | vcpu->arch.cr2 = fault->address; |
359 | kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code); | |
c3c91fee | 360 | } |
27d6c865 | 361 | EXPORT_SYMBOL_GPL(kvm_inject_page_fault); |
c3c91fee | 362 | |
6389ee94 | 363 | void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault) |
d4f8cf66 | 364 | { |
6389ee94 AK |
365 | if (mmu_is_nested(vcpu) && !fault->nested_page_fault) |
366 | vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault); | |
d4f8cf66 | 367 | else |
6389ee94 | 368 | vcpu->arch.mmu.inject_page_fault(vcpu, fault); |
d4f8cf66 JR |
369 | } |
370 | ||
3419ffc8 SY |
371 | void kvm_inject_nmi(struct kvm_vcpu *vcpu) |
372 | { | |
7460fb4a AK |
373 | atomic_inc(&vcpu->arch.nmi_queued); |
374 | kvm_make_request(KVM_REQ_NMI, vcpu); | |
3419ffc8 SY |
375 | } |
376 | EXPORT_SYMBOL_GPL(kvm_inject_nmi); | |
377 | ||
298101da AK |
378 | void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
379 | { | |
ce7ddec4 | 380 | kvm_multiple_exception(vcpu, nr, true, error_code, false); |
298101da AK |
381 | } |
382 | EXPORT_SYMBOL_GPL(kvm_queue_exception_e); | |
383 | ||
ce7ddec4 JR |
384 | void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
385 | { | |
386 | kvm_multiple_exception(vcpu, nr, true, error_code, true); | |
387 | } | |
388 | EXPORT_SYMBOL_GPL(kvm_requeue_exception_e); | |
389 | ||
0a79b009 AK |
390 | /* |
391 | * Checks if cpl <= required_cpl; if true, return true. Otherwise queue | |
392 | * a #GP and return false. | |
393 | */ | |
394 | bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl) | |
298101da | 395 | { |
0a79b009 AK |
396 | if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl) |
397 | return true; | |
398 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
399 | return false; | |
298101da | 400 | } |
0a79b009 | 401 | EXPORT_SYMBOL_GPL(kvm_require_cpl); |
298101da | 402 | |
ec92fe44 JR |
403 | /* |
404 | * This function will be used to read from the physical memory of the currently | |
405 | * running guest. The difference to kvm_read_guest_page is that this function | |
406 | * can read from guest physical or from the guest's guest physical memory. | |
407 | */ | |
408 | int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, | |
409 | gfn_t ngfn, void *data, int offset, int len, | |
410 | u32 access) | |
411 | { | |
412 | gfn_t real_gfn; | |
413 | gpa_t ngpa; | |
414 | ||
415 | ngpa = gfn_to_gpa(ngfn); | |
416 | real_gfn = mmu->translate_gpa(vcpu, ngpa, access); | |
417 | if (real_gfn == UNMAPPED_GVA) | |
418 | return -EFAULT; | |
419 | ||
420 | real_gfn = gpa_to_gfn(real_gfn); | |
421 | ||
422 | return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len); | |
423 | } | |
424 | EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu); | |
425 | ||
3d06b8bf JR |
426 | int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
427 | void *data, int offset, int len, u32 access) | |
428 | { | |
429 | return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn, | |
430 | data, offset, len, access); | |
431 | } | |
432 | ||
a03490ed CO |
433 | /* |
434 | * Load the pae pdptrs. Return true is they are all valid. | |
435 | */ | |
ff03a073 | 436 | int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3) |
a03490ed CO |
437 | { |
438 | gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT; | |
439 | unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2; | |
440 | int i; | |
441 | int ret; | |
ff03a073 | 442 | u64 pdpte[ARRAY_SIZE(mmu->pdptrs)]; |
a03490ed | 443 | |
ff03a073 JR |
444 | ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte, |
445 | offset * sizeof(u64), sizeof(pdpte), | |
446 | PFERR_USER_MASK|PFERR_WRITE_MASK); | |
a03490ed CO |
447 | if (ret < 0) { |
448 | ret = 0; | |
449 | goto out; | |
450 | } | |
451 | for (i = 0; i < ARRAY_SIZE(pdpte); ++i) { | |
43a3795a | 452 | if (is_present_gpte(pdpte[i]) && |
20c466b5 | 453 | (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) { |
a03490ed CO |
454 | ret = 0; |
455 | goto out; | |
456 | } | |
457 | } | |
458 | ret = 1; | |
459 | ||
ff03a073 | 460 | memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)); |
6de4f3ad AK |
461 | __set_bit(VCPU_EXREG_PDPTR, |
462 | (unsigned long *)&vcpu->arch.regs_avail); | |
463 | __set_bit(VCPU_EXREG_PDPTR, | |
464 | (unsigned long *)&vcpu->arch.regs_dirty); | |
a03490ed | 465 | out: |
a03490ed CO |
466 | |
467 | return ret; | |
468 | } | |
cc4b6871 | 469 | EXPORT_SYMBOL_GPL(load_pdptrs); |
a03490ed | 470 | |
d835dfec AK |
471 | static bool pdptrs_changed(struct kvm_vcpu *vcpu) |
472 | { | |
ff03a073 | 473 | u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)]; |
d835dfec | 474 | bool changed = true; |
3d06b8bf JR |
475 | int offset; |
476 | gfn_t gfn; | |
d835dfec AK |
477 | int r; |
478 | ||
479 | if (is_long_mode(vcpu) || !is_pae(vcpu)) | |
480 | return false; | |
481 | ||
6de4f3ad AK |
482 | if (!test_bit(VCPU_EXREG_PDPTR, |
483 | (unsigned long *)&vcpu->arch.regs_avail)) | |
484 | return true; | |
485 | ||
9f8fe504 AK |
486 | gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT; |
487 | offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1); | |
3d06b8bf JR |
488 | r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte), |
489 | PFERR_USER_MASK | PFERR_WRITE_MASK); | |
d835dfec AK |
490 | if (r < 0) |
491 | goto out; | |
ff03a073 | 492 | changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0; |
d835dfec | 493 | out: |
d835dfec AK |
494 | |
495 | return changed; | |
496 | } | |
497 | ||
49a9b07e | 498 | int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
a03490ed | 499 | { |
aad82703 SY |
500 | unsigned long old_cr0 = kvm_read_cr0(vcpu); |
501 | unsigned long update_bits = X86_CR0_PG | X86_CR0_WP | | |
502 | X86_CR0_CD | X86_CR0_NW; | |
503 | ||
f9a48e6a AK |
504 | cr0 |= X86_CR0_ET; |
505 | ||
ab344828 | 506 | #ifdef CONFIG_X86_64 |
0f12244f GN |
507 | if (cr0 & 0xffffffff00000000UL) |
508 | return 1; | |
ab344828 GN |
509 | #endif |
510 | ||
511 | cr0 &= ~CR0_RESERVED_BITS; | |
a03490ed | 512 | |
0f12244f GN |
513 | if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) |
514 | return 1; | |
a03490ed | 515 | |
0f12244f GN |
516 | if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) |
517 | return 1; | |
a03490ed CO |
518 | |
519 | if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) { | |
520 | #ifdef CONFIG_X86_64 | |
f6801dff | 521 | if ((vcpu->arch.efer & EFER_LME)) { |
a03490ed CO |
522 | int cs_db, cs_l; |
523 | ||
0f12244f GN |
524 | if (!is_pae(vcpu)) |
525 | return 1; | |
a03490ed | 526 | kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l); |
0f12244f GN |
527 | if (cs_l) |
528 | return 1; | |
a03490ed CO |
529 | } else |
530 | #endif | |
ff03a073 | 531 | if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, |
9f8fe504 | 532 | kvm_read_cr3(vcpu))) |
0f12244f | 533 | return 1; |
a03490ed CO |
534 | } |
535 | ||
ad756a16 MJ |
536 | if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) |
537 | return 1; | |
538 | ||
a03490ed | 539 | kvm_x86_ops->set_cr0(vcpu, cr0); |
a03490ed | 540 | |
d170c419 | 541 | if ((cr0 ^ old_cr0) & X86_CR0_PG) { |
e5f3f027 | 542 | kvm_clear_async_pf_completion_queue(vcpu); |
d170c419 LJ |
543 | kvm_async_pf_hash_reset(vcpu); |
544 | } | |
e5f3f027 | 545 | |
aad82703 SY |
546 | if ((cr0 ^ old_cr0) & update_bits) |
547 | kvm_mmu_reset_context(vcpu); | |
0f12244f GN |
548 | return 0; |
549 | } | |
2d3ad1f4 | 550 | EXPORT_SYMBOL_GPL(kvm_set_cr0); |
a03490ed | 551 | |
2d3ad1f4 | 552 | void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw) |
a03490ed | 553 | { |
49a9b07e | 554 | (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f)); |
a03490ed | 555 | } |
2d3ad1f4 | 556 | EXPORT_SYMBOL_GPL(kvm_lmsw); |
a03490ed | 557 | |
42bdf991 MT |
558 | static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu) |
559 | { | |
560 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) && | |
561 | !vcpu->guest_xcr0_loaded) { | |
562 | /* kvm_set_xcr() also depends on this */ | |
563 | xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0); | |
564 | vcpu->guest_xcr0_loaded = 1; | |
565 | } | |
566 | } | |
567 | ||
568 | static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu) | |
569 | { | |
570 | if (vcpu->guest_xcr0_loaded) { | |
571 | if (vcpu->arch.xcr0 != host_xcr0) | |
572 | xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0); | |
573 | vcpu->guest_xcr0_loaded = 0; | |
574 | } | |
575 | } | |
576 | ||
2acf923e DC |
577 | int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) |
578 | { | |
579 | u64 xcr0; | |
580 | ||
581 | /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */ | |
582 | if (index != XCR_XFEATURE_ENABLED_MASK) | |
583 | return 1; | |
584 | xcr0 = xcr; | |
2acf923e DC |
585 | if (!(xcr0 & XSTATE_FP)) |
586 | return 1; | |
587 | if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE)) | |
588 | return 1; | |
d7876f1b | 589 | if (xcr0 & ~vcpu->arch.guest_supported_xcr0) |
2acf923e | 590 | return 1; |
42bdf991 | 591 | kvm_put_guest_xcr0(vcpu); |
2acf923e | 592 | vcpu->arch.xcr0 = xcr0; |
2acf923e DC |
593 | return 0; |
594 | } | |
595 | ||
596 | int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) | |
597 | { | |
764bcbc5 Z |
598 | if (kvm_x86_ops->get_cpl(vcpu) != 0 || |
599 | __kvm_set_xcr(vcpu, index, xcr)) { | |
2acf923e DC |
600 | kvm_inject_gp(vcpu, 0); |
601 | return 1; | |
602 | } | |
603 | return 0; | |
604 | } | |
605 | EXPORT_SYMBOL_GPL(kvm_set_xcr); | |
606 | ||
a83b29c6 | 607 | int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) |
a03490ed | 608 | { |
fc78f519 | 609 | unsigned long old_cr4 = kvm_read_cr4(vcpu); |
c68b734f YW |
610 | unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | |
611 | X86_CR4_PAE | X86_CR4_SMEP; | |
0f12244f GN |
612 | if (cr4 & CR4_RESERVED_BITS) |
613 | return 1; | |
a03490ed | 614 | |
2acf923e DC |
615 | if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE)) |
616 | return 1; | |
617 | ||
c68b734f YW |
618 | if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP)) |
619 | return 1; | |
620 | ||
afcbf13f | 621 | if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE)) |
74dc2b4f YW |
622 | return 1; |
623 | ||
a03490ed | 624 | if (is_long_mode(vcpu)) { |
0f12244f GN |
625 | if (!(cr4 & X86_CR4_PAE)) |
626 | return 1; | |
a2edf57f AK |
627 | } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE) |
628 | && ((cr4 ^ old_cr4) & pdptr_bits) | |
9f8fe504 AK |
629 | && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, |
630 | kvm_read_cr3(vcpu))) | |
0f12244f GN |
631 | return 1; |
632 | ||
ad756a16 MJ |
633 | if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) { |
634 | if (!guest_cpuid_has_pcid(vcpu)) | |
635 | return 1; | |
636 | ||
637 | /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */ | |
638 | if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu)) | |
639 | return 1; | |
640 | } | |
641 | ||
5e1746d6 | 642 | if (kvm_x86_ops->set_cr4(vcpu, cr4)) |
0f12244f | 643 | return 1; |
a03490ed | 644 | |
ad756a16 MJ |
645 | if (((cr4 ^ old_cr4) & pdptr_bits) || |
646 | (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE))) | |
aad82703 | 647 | kvm_mmu_reset_context(vcpu); |
0f12244f | 648 | |
2acf923e | 649 | if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE) |
00b27a3e | 650 | kvm_update_cpuid(vcpu); |
2acf923e | 651 | |
0f12244f GN |
652 | return 0; |
653 | } | |
2d3ad1f4 | 654 | EXPORT_SYMBOL_GPL(kvm_set_cr4); |
a03490ed | 655 | |
2390218b | 656 | int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3) |
a03490ed | 657 | { |
9f8fe504 | 658 | if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) { |
0ba73cda | 659 | kvm_mmu_sync_roots(vcpu); |
d835dfec | 660 | kvm_mmu_flush_tlb(vcpu); |
0f12244f | 661 | return 0; |
d835dfec AK |
662 | } |
663 | ||
a03490ed | 664 | if (is_long_mode(vcpu)) { |
471842ec | 665 | if (kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) { |
ad756a16 MJ |
666 | if (cr3 & CR3_PCID_ENABLED_RESERVED_BITS) |
667 | return 1; | |
668 | } else | |
669 | if (cr3 & CR3_L_MODE_RESERVED_BITS) | |
670 | return 1; | |
a03490ed CO |
671 | } else { |
672 | if (is_pae(vcpu)) { | |
0f12244f GN |
673 | if (cr3 & CR3_PAE_RESERVED_BITS) |
674 | return 1; | |
ff03a073 JR |
675 | if (is_paging(vcpu) && |
676 | !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) | |
0f12244f | 677 | return 1; |
a03490ed CO |
678 | } |
679 | /* | |
680 | * We don't check reserved bits in nonpae mode, because | |
681 | * this isn't enforced, and VMware depends on this. | |
682 | */ | |
683 | } | |
684 | ||
0f12244f | 685 | vcpu->arch.cr3 = cr3; |
aff48baa | 686 | __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail); |
d8d173da | 687 | kvm_mmu_new_cr3(vcpu); |
0f12244f GN |
688 | return 0; |
689 | } | |
2d3ad1f4 | 690 | EXPORT_SYMBOL_GPL(kvm_set_cr3); |
a03490ed | 691 | |
eea1cff9 | 692 | int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8) |
a03490ed | 693 | { |
0f12244f GN |
694 | if (cr8 & CR8_RESERVED_BITS) |
695 | return 1; | |
a03490ed CO |
696 | if (irqchip_in_kernel(vcpu->kvm)) |
697 | kvm_lapic_set_tpr(vcpu, cr8); | |
698 | else | |
ad312c7c | 699 | vcpu->arch.cr8 = cr8; |
0f12244f GN |
700 | return 0; |
701 | } | |
2d3ad1f4 | 702 | EXPORT_SYMBOL_GPL(kvm_set_cr8); |
a03490ed | 703 | |
2d3ad1f4 | 704 | unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu) |
a03490ed CO |
705 | { |
706 | if (irqchip_in_kernel(vcpu->kvm)) | |
707 | return kvm_lapic_get_cr8(vcpu); | |
708 | else | |
ad312c7c | 709 | return vcpu->arch.cr8; |
a03490ed | 710 | } |
2d3ad1f4 | 711 | EXPORT_SYMBOL_GPL(kvm_get_cr8); |
a03490ed | 712 | |
c8639010 JK |
713 | static void kvm_update_dr7(struct kvm_vcpu *vcpu) |
714 | { | |
715 | unsigned long dr7; | |
716 | ||
717 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) | |
718 | dr7 = vcpu->arch.guest_debug_dr7; | |
719 | else | |
720 | dr7 = vcpu->arch.dr7; | |
721 | kvm_x86_ops->set_dr7(vcpu, dr7); | |
722 | vcpu->arch.switch_db_regs = (dr7 & DR7_BP_EN_MASK); | |
723 | } | |
724 | ||
338dbc97 | 725 | static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) |
020df079 GN |
726 | { |
727 | switch (dr) { | |
728 | case 0 ... 3: | |
729 | vcpu->arch.db[dr] = val; | |
730 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) | |
731 | vcpu->arch.eff_db[dr] = val; | |
732 | break; | |
733 | case 4: | |
338dbc97 GN |
734 | if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) |
735 | return 1; /* #UD */ | |
020df079 GN |
736 | /* fall through */ |
737 | case 6: | |
338dbc97 GN |
738 | if (val & 0xffffffff00000000ULL) |
739 | return -1; /* #GP */ | |
020df079 GN |
740 | vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1; |
741 | break; | |
742 | case 5: | |
338dbc97 GN |
743 | if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) |
744 | return 1; /* #UD */ | |
020df079 GN |
745 | /* fall through */ |
746 | default: /* 7 */ | |
338dbc97 GN |
747 | if (val & 0xffffffff00000000ULL) |
748 | return -1; /* #GP */ | |
020df079 | 749 | vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1; |
c8639010 | 750 | kvm_update_dr7(vcpu); |
020df079 GN |
751 | break; |
752 | } | |
753 | ||
754 | return 0; | |
755 | } | |
338dbc97 GN |
756 | |
757 | int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) | |
758 | { | |
759 | int res; | |
760 | ||
761 | res = __kvm_set_dr(vcpu, dr, val); | |
762 | if (res > 0) | |
763 | kvm_queue_exception(vcpu, UD_VECTOR); | |
764 | else if (res < 0) | |
765 | kvm_inject_gp(vcpu, 0); | |
766 | ||
767 | return res; | |
768 | } | |
020df079 GN |
769 | EXPORT_SYMBOL_GPL(kvm_set_dr); |
770 | ||
338dbc97 | 771 | static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val) |
020df079 GN |
772 | { |
773 | switch (dr) { | |
774 | case 0 ... 3: | |
775 | *val = vcpu->arch.db[dr]; | |
776 | break; | |
777 | case 4: | |
338dbc97 | 778 | if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) |
020df079 | 779 | return 1; |
020df079 GN |
780 | /* fall through */ |
781 | case 6: | |
782 | *val = vcpu->arch.dr6; | |
783 | break; | |
784 | case 5: | |
338dbc97 | 785 | if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) |
020df079 | 786 | return 1; |
020df079 GN |
787 | /* fall through */ |
788 | default: /* 7 */ | |
789 | *val = vcpu->arch.dr7; | |
790 | break; | |
791 | } | |
792 | ||
793 | return 0; | |
794 | } | |
338dbc97 GN |
795 | |
796 | int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val) | |
797 | { | |
798 | if (_kvm_get_dr(vcpu, dr, val)) { | |
799 | kvm_queue_exception(vcpu, UD_VECTOR); | |
800 | return 1; | |
801 | } | |
802 | return 0; | |
803 | } | |
020df079 GN |
804 | EXPORT_SYMBOL_GPL(kvm_get_dr); |
805 | ||
022cd0e8 AK |
806 | bool kvm_rdpmc(struct kvm_vcpu *vcpu) |
807 | { | |
808 | u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
809 | u64 data; | |
810 | int err; | |
811 | ||
812 | err = kvm_pmu_read_pmc(vcpu, ecx, &data); | |
813 | if (err) | |
814 | return err; | |
815 | kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data); | |
816 | kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32); | |
817 | return err; | |
818 | } | |
819 | EXPORT_SYMBOL_GPL(kvm_rdpmc); | |
820 | ||
043405e1 CO |
821 | /* |
822 | * List of msr numbers which we expose to userspace through KVM_GET_MSRS | |
823 | * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST. | |
824 | * | |
825 | * This list is modified at module load time to reflect the | |
e3267cbb GC |
826 | * capabilities of the host cpu. This capabilities test skips MSRs that are |
827 | * kvm-specific. Those are put in the beginning of the list. | |
043405e1 | 828 | */ |
e3267cbb | 829 | |
439793d4 | 830 | #define KVM_SAVE_MSRS_BEGIN 10 |
043405e1 | 831 | static u32 msrs_to_save[] = { |
e3267cbb | 832 | MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK, |
11c6bffa | 833 | MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW, |
55cd8e5a | 834 | HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL, |
c9aaa895 | 835 | HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME, |
ae7a2a3f | 836 | MSR_KVM_PV_EOI_EN, |
043405e1 | 837 | MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, |
8c06585d | 838 | MSR_STAR, |
043405e1 CO |
839 | #ifdef CONFIG_X86_64 |
840 | MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR, | |
841 | #endif | |
b3897a49 NHE |
842 | MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA, |
843 | MSR_IA32_FEATURE_CONTROL | |
043405e1 CO |
844 | }; |
845 | ||
846 | static unsigned num_msrs_to_save; | |
847 | ||
f1d24831 | 848 | static const u32 emulated_msrs[] = { |
ba904635 | 849 | MSR_IA32_TSC_ADJUST, |
a3e06bbe | 850 | MSR_IA32_TSCDEADLINE, |
043405e1 | 851 | MSR_IA32_MISC_ENABLE, |
908e75f3 AK |
852 | MSR_IA32_MCG_STATUS, |
853 | MSR_IA32_MCG_CTL, | |
043405e1 CO |
854 | }; |
855 | ||
384bb783 | 856 | bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer) |
15c4a640 | 857 | { |
b69e8cae | 858 | if (efer & efer_reserved_bits) |
384bb783 | 859 | return false; |
15c4a640 | 860 | |
1b2fd70c AG |
861 | if (efer & EFER_FFXSR) { |
862 | struct kvm_cpuid_entry2 *feat; | |
863 | ||
864 | feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0); | |
b69e8cae | 865 | if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) |
384bb783 | 866 | return false; |
1b2fd70c AG |
867 | } |
868 | ||
d8017474 AG |
869 | if (efer & EFER_SVME) { |
870 | struct kvm_cpuid_entry2 *feat; | |
871 | ||
872 | feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0); | |
b69e8cae | 873 | if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) |
384bb783 | 874 | return false; |
d8017474 AG |
875 | } |
876 | ||
384bb783 JK |
877 | return true; |
878 | } | |
879 | EXPORT_SYMBOL_GPL(kvm_valid_efer); | |
880 | ||
881 | static int set_efer(struct kvm_vcpu *vcpu, u64 efer) | |
882 | { | |
883 | u64 old_efer = vcpu->arch.efer; | |
884 | ||
885 | if (!kvm_valid_efer(vcpu, efer)) | |
886 | return 1; | |
887 | ||
888 | if (is_paging(vcpu) | |
889 | && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) | |
890 | return 1; | |
891 | ||
15c4a640 | 892 | efer &= ~EFER_LMA; |
f6801dff | 893 | efer |= vcpu->arch.efer & EFER_LMA; |
15c4a640 | 894 | |
a3d204e2 SY |
895 | kvm_x86_ops->set_efer(vcpu, efer); |
896 | ||
aad82703 SY |
897 | /* Update reserved bits */ |
898 | if ((efer ^ old_efer) & EFER_NX) | |
899 | kvm_mmu_reset_context(vcpu); | |
900 | ||
b69e8cae | 901 | return 0; |
15c4a640 CO |
902 | } |
903 | ||
f2b4b7dd JR |
904 | void kvm_enable_efer_bits(u64 mask) |
905 | { | |
906 | efer_reserved_bits &= ~mask; | |
907 | } | |
908 | EXPORT_SYMBOL_GPL(kvm_enable_efer_bits); | |
909 | ||
910 | ||
15c4a640 CO |
911 | /* |
912 | * Writes msr value into into the appropriate "register". | |
913 | * Returns 0 on success, non-0 otherwise. | |
914 | * Assumes vcpu_load() was already called. | |
915 | */ | |
8fe8ab46 | 916 | int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr) |
15c4a640 | 917 | { |
8fe8ab46 | 918 | return kvm_x86_ops->set_msr(vcpu, msr); |
15c4a640 CO |
919 | } |
920 | ||
313a3dc7 CO |
921 | /* |
922 | * Adapt set_msr() to msr_io()'s calling convention | |
923 | */ | |
924 | static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) | |
925 | { | |
8fe8ab46 WA |
926 | struct msr_data msr; |
927 | ||
928 | msr.data = *data; | |
929 | msr.index = index; | |
930 | msr.host_initiated = true; | |
931 | return kvm_set_msr(vcpu, &msr); | |
313a3dc7 CO |
932 | } |
933 | ||
16e8d74d MT |
934 | #ifdef CONFIG_X86_64 |
935 | struct pvclock_gtod_data { | |
936 | seqcount_t seq; | |
937 | ||
938 | struct { /* extract of a clocksource struct */ | |
939 | int vclock_mode; | |
940 | cycle_t cycle_last; | |
941 | cycle_t mask; | |
942 | u32 mult; | |
943 | u32 shift; | |
944 | } clock; | |
945 | ||
946 | /* open coded 'struct timespec' */ | |
947 | u64 monotonic_time_snsec; | |
948 | time_t monotonic_time_sec; | |
949 | }; | |
950 | ||
951 | static struct pvclock_gtod_data pvclock_gtod_data; | |
952 | ||
953 | static void update_pvclock_gtod(struct timekeeper *tk) | |
954 | { | |
955 | struct pvclock_gtod_data *vdata = &pvclock_gtod_data; | |
956 | ||
957 | write_seqcount_begin(&vdata->seq); | |
958 | ||
959 | /* copy pvclock gtod data */ | |
960 | vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode; | |
961 | vdata->clock.cycle_last = tk->clock->cycle_last; | |
962 | vdata->clock.mask = tk->clock->mask; | |
963 | vdata->clock.mult = tk->mult; | |
964 | vdata->clock.shift = tk->shift; | |
965 | ||
966 | vdata->monotonic_time_sec = tk->xtime_sec | |
967 | + tk->wall_to_monotonic.tv_sec; | |
968 | vdata->monotonic_time_snsec = tk->xtime_nsec | |
969 | + (tk->wall_to_monotonic.tv_nsec | |
970 | << tk->shift); | |
971 | while (vdata->monotonic_time_snsec >= | |
972 | (((u64)NSEC_PER_SEC) << tk->shift)) { | |
973 | vdata->monotonic_time_snsec -= | |
974 | ((u64)NSEC_PER_SEC) << tk->shift; | |
975 | vdata->monotonic_time_sec++; | |
976 | } | |
977 | ||
978 | write_seqcount_end(&vdata->seq); | |
979 | } | |
980 | #endif | |
981 | ||
982 | ||
18068523 GOC |
983 | static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock) |
984 | { | |
9ed3c444 AK |
985 | int version; |
986 | int r; | |
50d0a0f9 | 987 | struct pvclock_wall_clock wc; |
923de3cf | 988 | struct timespec boot; |
18068523 GOC |
989 | |
990 | if (!wall_clock) | |
991 | return; | |
992 | ||
9ed3c444 AK |
993 | r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version)); |
994 | if (r) | |
995 | return; | |
996 | ||
997 | if (version & 1) | |
998 | ++version; /* first time write, random junk */ | |
999 | ||
1000 | ++version; | |
18068523 | 1001 | |
18068523 GOC |
1002 | kvm_write_guest(kvm, wall_clock, &version, sizeof(version)); |
1003 | ||
50d0a0f9 GH |
1004 | /* |
1005 | * The guest calculates current wall clock time by adding | |
34c238a1 | 1006 | * system time (updated by kvm_guest_time_update below) to the |
50d0a0f9 GH |
1007 | * wall clock specified here. guest system time equals host |
1008 | * system time for us, thus we must fill in host boot time here. | |
1009 | */ | |
923de3cf | 1010 | getboottime(&boot); |
50d0a0f9 | 1011 | |
4b648665 BR |
1012 | if (kvm->arch.kvmclock_offset) { |
1013 | struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset); | |
1014 | boot = timespec_sub(boot, ts); | |
1015 | } | |
50d0a0f9 GH |
1016 | wc.sec = boot.tv_sec; |
1017 | wc.nsec = boot.tv_nsec; | |
1018 | wc.version = version; | |
18068523 GOC |
1019 | |
1020 | kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc)); | |
1021 | ||
1022 | version++; | |
1023 | kvm_write_guest(kvm, wall_clock, &version, sizeof(version)); | |
18068523 GOC |
1024 | } |
1025 | ||
50d0a0f9 GH |
1026 | static uint32_t div_frac(uint32_t dividend, uint32_t divisor) |
1027 | { | |
1028 | uint32_t quotient, remainder; | |
1029 | ||
1030 | /* Don't try to replace with do_div(), this one calculates | |
1031 | * "(dividend << 32) / divisor" */ | |
1032 | __asm__ ( "divl %4" | |
1033 | : "=a" (quotient), "=d" (remainder) | |
1034 | : "0" (0), "1" (dividend), "r" (divisor) ); | |
1035 | return quotient; | |
1036 | } | |
1037 | ||
5f4e3f88 ZA |
1038 | static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz, |
1039 | s8 *pshift, u32 *pmultiplier) | |
50d0a0f9 | 1040 | { |
5f4e3f88 | 1041 | uint64_t scaled64; |
50d0a0f9 GH |
1042 | int32_t shift = 0; |
1043 | uint64_t tps64; | |
1044 | uint32_t tps32; | |
1045 | ||
5f4e3f88 ZA |
1046 | tps64 = base_khz * 1000LL; |
1047 | scaled64 = scaled_khz * 1000LL; | |
50933623 | 1048 | while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) { |
50d0a0f9 GH |
1049 | tps64 >>= 1; |
1050 | shift--; | |
1051 | } | |
1052 | ||
1053 | tps32 = (uint32_t)tps64; | |
50933623 JK |
1054 | while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) { |
1055 | if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000) | |
5f4e3f88 ZA |
1056 | scaled64 >>= 1; |
1057 | else | |
1058 | tps32 <<= 1; | |
50d0a0f9 GH |
1059 | shift++; |
1060 | } | |
1061 | ||
5f4e3f88 ZA |
1062 | *pshift = shift; |
1063 | *pmultiplier = div_frac(scaled64, tps32); | |
50d0a0f9 | 1064 | |
5f4e3f88 ZA |
1065 | pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n", |
1066 | __func__, base_khz, scaled_khz, shift, *pmultiplier); | |
50d0a0f9 GH |
1067 | } |
1068 | ||
759379dd ZA |
1069 | static inline u64 get_kernel_ns(void) |
1070 | { | |
1071 | struct timespec ts; | |
1072 | ||
1073 | WARN_ON(preemptible()); | |
1074 | ktime_get_ts(&ts); | |
1075 | monotonic_to_bootbased(&ts); | |
1076 | return timespec_to_ns(&ts); | |
50d0a0f9 GH |
1077 | } |
1078 | ||
d828199e | 1079 | #ifdef CONFIG_X86_64 |
16e8d74d | 1080 | static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0); |
d828199e | 1081 | #endif |
16e8d74d | 1082 | |
c8076604 | 1083 | static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz); |
c285545f | 1084 | unsigned long max_tsc_khz; |
c8076604 | 1085 | |
cc578287 | 1086 | static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec) |
8cfdc000 | 1087 | { |
cc578287 ZA |
1088 | return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult, |
1089 | vcpu->arch.virtual_tsc_shift); | |
8cfdc000 ZA |
1090 | } |
1091 | ||
cc578287 | 1092 | static u32 adjust_tsc_khz(u32 khz, s32 ppm) |
1e993611 | 1093 | { |
cc578287 ZA |
1094 | u64 v = (u64)khz * (1000000 + ppm); |
1095 | do_div(v, 1000000); | |
1096 | return v; | |
1e993611 JR |
1097 | } |
1098 | ||
cc578287 | 1099 | static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz) |
759379dd | 1100 | { |
cc578287 ZA |
1101 | u32 thresh_lo, thresh_hi; |
1102 | int use_scaling = 0; | |
217fc9cf | 1103 | |
03ba32ca MT |
1104 | /* tsc_khz can be zero if TSC calibration fails */ |
1105 | if (this_tsc_khz == 0) | |
1106 | return; | |
1107 | ||
c285545f ZA |
1108 | /* Compute a scale to convert nanoseconds in TSC cycles */ |
1109 | kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000, | |
cc578287 ZA |
1110 | &vcpu->arch.virtual_tsc_shift, |
1111 | &vcpu->arch.virtual_tsc_mult); | |
1112 | vcpu->arch.virtual_tsc_khz = this_tsc_khz; | |
1113 | ||
1114 | /* | |
1115 | * Compute the variation in TSC rate which is acceptable | |
1116 | * within the range of tolerance and decide if the | |
1117 | * rate being applied is within that bounds of the hardware | |
1118 | * rate. If so, no scaling or compensation need be done. | |
1119 | */ | |
1120 | thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm); | |
1121 | thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm); | |
1122 | if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) { | |
1123 | pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi); | |
1124 | use_scaling = 1; | |
1125 | } | |
1126 | kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling); | |
c285545f ZA |
1127 | } |
1128 | ||
1129 | static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns) | |
1130 | { | |
e26101b1 | 1131 | u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec, |
cc578287 ZA |
1132 | vcpu->arch.virtual_tsc_mult, |
1133 | vcpu->arch.virtual_tsc_shift); | |
e26101b1 | 1134 | tsc += vcpu->arch.this_tsc_write; |
c285545f ZA |
1135 | return tsc; |
1136 | } | |
1137 | ||
b48aa97e MT |
1138 | void kvm_track_tsc_matching(struct kvm_vcpu *vcpu) |
1139 | { | |
1140 | #ifdef CONFIG_X86_64 | |
1141 | bool vcpus_matched; | |
1142 | bool do_request = false; | |
1143 | struct kvm_arch *ka = &vcpu->kvm->arch; | |
1144 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
1145 | ||
1146 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
1147 | atomic_read(&vcpu->kvm->online_vcpus)); | |
1148 | ||
1149 | if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC) | |
1150 | if (!ka->use_master_clock) | |
1151 | do_request = 1; | |
1152 | ||
1153 | if (!vcpus_matched && ka->use_master_clock) | |
1154 | do_request = 1; | |
1155 | ||
1156 | if (do_request) | |
1157 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); | |
1158 | ||
1159 | trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc, | |
1160 | atomic_read(&vcpu->kvm->online_vcpus), | |
1161 | ka->use_master_clock, gtod->clock.vclock_mode); | |
1162 | #endif | |
1163 | } | |
1164 | ||
ba904635 WA |
1165 | static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset) |
1166 | { | |
1167 | u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu); | |
1168 | vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset; | |
1169 | } | |
1170 | ||
8fe8ab46 | 1171 | void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr) |
99e3e30a ZA |
1172 | { |
1173 | struct kvm *kvm = vcpu->kvm; | |
f38e098f | 1174 | u64 offset, ns, elapsed; |
99e3e30a | 1175 | unsigned long flags; |
02626b6a | 1176 | s64 usdiff; |
b48aa97e | 1177 | bool matched; |
8fe8ab46 | 1178 | u64 data = msr->data; |
99e3e30a | 1179 | |
038f8c11 | 1180 | raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags); |
857e4099 | 1181 | offset = kvm_x86_ops->compute_tsc_offset(vcpu, data); |
759379dd | 1182 | ns = get_kernel_ns(); |
f38e098f | 1183 | elapsed = ns - kvm->arch.last_tsc_nsec; |
5d3cb0f6 | 1184 | |
03ba32ca | 1185 | if (vcpu->arch.virtual_tsc_khz) { |
8915aa27 MT |
1186 | int faulted = 0; |
1187 | ||
03ba32ca MT |
1188 | /* n.b - signed multiplication and division required */ |
1189 | usdiff = data - kvm->arch.last_tsc_write; | |
5d3cb0f6 | 1190 | #ifdef CONFIG_X86_64 |
03ba32ca | 1191 | usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz; |
5d3cb0f6 | 1192 | #else |
03ba32ca | 1193 | /* do_div() only does unsigned */ |
8915aa27 MT |
1194 | asm("1: idivl %[divisor]\n" |
1195 | "2: xor %%edx, %%edx\n" | |
1196 | " movl $0, %[faulted]\n" | |
1197 | "3:\n" | |
1198 | ".section .fixup,\"ax\"\n" | |
1199 | "4: movl $1, %[faulted]\n" | |
1200 | " jmp 3b\n" | |
1201 | ".previous\n" | |
1202 | ||
1203 | _ASM_EXTABLE(1b, 4b) | |
1204 | ||
1205 | : "=A"(usdiff), [faulted] "=r" (faulted) | |
1206 | : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz)); | |
1207 | ||
5d3cb0f6 | 1208 | #endif |
03ba32ca MT |
1209 | do_div(elapsed, 1000); |
1210 | usdiff -= elapsed; | |
1211 | if (usdiff < 0) | |
1212 | usdiff = -usdiff; | |
8915aa27 MT |
1213 | |
1214 | /* idivl overflow => difference is larger than USEC_PER_SEC */ | |
1215 | if (faulted) | |
1216 | usdiff = USEC_PER_SEC; | |
03ba32ca MT |
1217 | } else |
1218 | usdiff = USEC_PER_SEC; /* disable TSC match window below */ | |
f38e098f ZA |
1219 | |
1220 | /* | |
5d3cb0f6 ZA |
1221 | * Special case: TSC write with a small delta (1 second) of virtual |
1222 | * cycle time against real time is interpreted as an attempt to | |
1223 | * synchronize the CPU. | |
1224 | * | |
1225 | * For a reliable TSC, we can match TSC offsets, and for an unstable | |
1226 | * TSC, we add elapsed time in this computation. We could let the | |
1227 | * compensation code attempt to catch up if we fall behind, but | |
1228 | * it's better to try to match offsets from the beginning. | |
1229 | */ | |
02626b6a | 1230 | if (usdiff < USEC_PER_SEC && |
5d3cb0f6 | 1231 | vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) { |
f38e098f | 1232 | if (!check_tsc_unstable()) { |
e26101b1 | 1233 | offset = kvm->arch.cur_tsc_offset; |
f38e098f ZA |
1234 | pr_debug("kvm: matched tsc offset for %llu\n", data); |
1235 | } else { | |
857e4099 | 1236 | u64 delta = nsec_to_cycles(vcpu, elapsed); |
5d3cb0f6 ZA |
1237 | data += delta; |
1238 | offset = kvm_x86_ops->compute_tsc_offset(vcpu, data); | |
759379dd | 1239 | pr_debug("kvm: adjusted tsc offset by %llu\n", delta); |
f38e098f | 1240 | } |
b48aa97e | 1241 | matched = true; |
e26101b1 ZA |
1242 | } else { |
1243 | /* | |
1244 | * We split periods of matched TSC writes into generations. | |
1245 | * For each generation, we track the original measured | |
1246 | * nanosecond time, offset, and write, so if TSCs are in | |
1247 | * sync, we can match exact offset, and if not, we can match | |
4a969980 | 1248 | * exact software computation in compute_guest_tsc() |
e26101b1 ZA |
1249 | * |
1250 | * These values are tracked in kvm->arch.cur_xxx variables. | |
1251 | */ | |
1252 | kvm->arch.cur_tsc_generation++; | |
1253 | kvm->arch.cur_tsc_nsec = ns; | |
1254 | kvm->arch.cur_tsc_write = data; | |
1255 | kvm->arch.cur_tsc_offset = offset; | |
b48aa97e | 1256 | matched = false; |
e26101b1 ZA |
1257 | pr_debug("kvm: new tsc generation %u, clock %llu\n", |
1258 | kvm->arch.cur_tsc_generation, data); | |
f38e098f | 1259 | } |
e26101b1 ZA |
1260 | |
1261 | /* | |
1262 | * We also track th most recent recorded KHZ, write and time to | |
1263 | * allow the matching interval to be extended at each write. | |
1264 | */ | |
f38e098f ZA |
1265 | kvm->arch.last_tsc_nsec = ns; |
1266 | kvm->arch.last_tsc_write = data; | |
5d3cb0f6 | 1267 | kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz; |
99e3e30a ZA |
1268 | |
1269 | /* Reset of TSC must disable overshoot protection below */ | |
1270 | vcpu->arch.hv_clock.tsc_timestamp = 0; | |
b183aa58 | 1271 | vcpu->arch.last_guest_tsc = data; |
e26101b1 ZA |
1272 | |
1273 | /* Keep track of which generation this VCPU has synchronized to */ | |
1274 | vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation; | |
1275 | vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec; | |
1276 | vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write; | |
1277 | ||
ba904635 WA |
1278 | if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated) |
1279 | update_ia32_tsc_adjust_msr(vcpu, offset); | |
e26101b1 ZA |
1280 | kvm_x86_ops->write_tsc_offset(vcpu, offset); |
1281 | raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags); | |
b48aa97e MT |
1282 | |
1283 | spin_lock(&kvm->arch.pvclock_gtod_sync_lock); | |
1284 | if (matched) | |
1285 | kvm->arch.nr_vcpus_matched_tsc++; | |
1286 | else | |
1287 | kvm->arch.nr_vcpus_matched_tsc = 0; | |
1288 | ||
1289 | kvm_track_tsc_matching(vcpu); | |
1290 | spin_unlock(&kvm->arch.pvclock_gtod_sync_lock); | |
99e3e30a | 1291 | } |
e26101b1 | 1292 | |
99e3e30a ZA |
1293 | EXPORT_SYMBOL_GPL(kvm_write_tsc); |
1294 | ||
d828199e MT |
1295 | #ifdef CONFIG_X86_64 |
1296 | ||
1297 | static cycle_t read_tsc(void) | |
1298 | { | |
1299 | cycle_t ret; | |
1300 | u64 last; | |
1301 | ||
1302 | /* | |
1303 | * Empirically, a fence (of type that depends on the CPU) | |
1304 | * before rdtsc is enough to ensure that rdtsc is ordered | |
1305 | * with respect to loads. The various CPU manuals are unclear | |
1306 | * as to whether rdtsc can be reordered with later loads, | |
1307 | * but no one has ever seen it happen. | |
1308 | */ | |
1309 | rdtsc_barrier(); | |
1310 | ret = (cycle_t)vget_cycles(); | |
1311 | ||
1312 | last = pvclock_gtod_data.clock.cycle_last; | |
1313 | ||
1314 | if (likely(ret >= last)) | |
1315 | return ret; | |
1316 | ||
1317 | /* | |
1318 | * GCC likes to generate cmov here, but this branch is extremely | |
1319 | * predictable (it's just a funciton of time and the likely is | |
1320 | * very likely) and there's a data dependence, so force GCC | |
1321 | * to generate a branch instead. I don't barrier() because | |
1322 | * we don't actually need a barrier, and if this function | |
1323 | * ever gets inlined it will generate worse code. | |
1324 | */ | |
1325 | asm volatile (""); | |
1326 | return last; | |
1327 | } | |
1328 | ||
1329 | static inline u64 vgettsc(cycle_t *cycle_now) | |
1330 | { | |
1331 | long v; | |
1332 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
1333 | ||
1334 | *cycle_now = read_tsc(); | |
1335 | ||
1336 | v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask; | |
1337 | return v * gtod->clock.mult; | |
1338 | } | |
1339 | ||
1340 | static int do_monotonic(struct timespec *ts, cycle_t *cycle_now) | |
1341 | { | |
1342 | unsigned long seq; | |
1343 | u64 ns; | |
1344 | int mode; | |
1345 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
1346 | ||
1347 | ts->tv_nsec = 0; | |
1348 | do { | |
1349 | seq = read_seqcount_begin(>od->seq); | |
1350 | mode = gtod->clock.vclock_mode; | |
1351 | ts->tv_sec = gtod->monotonic_time_sec; | |
1352 | ns = gtod->monotonic_time_snsec; | |
1353 | ns += vgettsc(cycle_now); | |
1354 | ns >>= gtod->clock.shift; | |
1355 | } while (unlikely(read_seqcount_retry(>od->seq, seq))); | |
1356 | timespec_add_ns(ts, ns); | |
1357 | ||
1358 | return mode; | |
1359 | } | |
1360 | ||
1361 | /* returns true if host is using tsc clocksource */ | |
1362 | static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now) | |
1363 | { | |
1364 | struct timespec ts; | |
1365 | ||
1366 | /* checked again under seqlock below */ | |
1367 | if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC) | |
1368 | return false; | |
1369 | ||
1370 | if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC) | |
1371 | return false; | |
1372 | ||
1373 | monotonic_to_bootbased(&ts); | |
1374 | *kernel_ns = timespec_to_ns(&ts); | |
1375 | ||
1376 | return true; | |
1377 | } | |
1378 | #endif | |
1379 | ||
1380 | /* | |
1381 | * | |
b48aa97e MT |
1382 | * Assuming a stable TSC across physical CPUS, and a stable TSC |
1383 | * across virtual CPUs, the following condition is possible. | |
1384 | * Each numbered line represents an event visible to both | |
d828199e MT |
1385 | * CPUs at the next numbered event. |
1386 | * | |
1387 | * "timespecX" represents host monotonic time. "tscX" represents | |
1388 | * RDTSC value. | |
1389 | * | |
1390 | * VCPU0 on CPU0 | VCPU1 on CPU1 | |
1391 | * | |
1392 | * 1. read timespec0,tsc0 | |
1393 | * 2. | timespec1 = timespec0 + N | |
1394 | * | tsc1 = tsc0 + M | |
1395 | * 3. transition to guest | transition to guest | |
1396 | * 4. ret0 = timespec0 + (rdtsc - tsc0) | | |
1397 | * 5. | ret1 = timespec1 + (rdtsc - tsc1) | |
1398 | * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M)) | |
1399 | * | |
1400 | * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity: | |
1401 | * | |
1402 | * - ret0 < ret1 | |
1403 | * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M)) | |
1404 | * ... | |
1405 | * - 0 < N - M => M < N | |
1406 | * | |
1407 | * That is, when timespec0 != timespec1, M < N. Unfortunately that is not | |
1408 | * always the case (the difference between two distinct xtime instances | |
1409 | * might be smaller then the difference between corresponding TSC reads, | |
1410 | * when updating guest vcpus pvclock areas). | |
1411 | * | |
1412 | * To avoid that problem, do not allow visibility of distinct | |
1413 | * system_timestamp/tsc_timestamp values simultaneously: use a master | |
1414 | * copy of host monotonic time values. Update that master copy | |
1415 | * in lockstep. | |
1416 | * | |
b48aa97e | 1417 | * Rely on synchronization of host TSCs and guest TSCs for monotonicity. |
d828199e MT |
1418 | * |
1419 | */ | |
1420 | ||
1421 | static void pvclock_update_vm_gtod_copy(struct kvm *kvm) | |
1422 | { | |
1423 | #ifdef CONFIG_X86_64 | |
1424 | struct kvm_arch *ka = &kvm->arch; | |
1425 | int vclock_mode; | |
b48aa97e MT |
1426 | bool host_tsc_clocksource, vcpus_matched; |
1427 | ||
1428 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
1429 | atomic_read(&kvm->online_vcpus)); | |
d828199e MT |
1430 | |
1431 | /* | |
1432 | * If the host uses TSC clock, then passthrough TSC as stable | |
1433 | * to the guest. | |
1434 | */ | |
b48aa97e | 1435 | host_tsc_clocksource = kvm_get_time_and_clockread( |
d828199e MT |
1436 | &ka->master_kernel_ns, |
1437 | &ka->master_cycle_now); | |
1438 | ||
b48aa97e MT |
1439 | ka->use_master_clock = host_tsc_clocksource & vcpus_matched; |
1440 | ||
d828199e MT |
1441 | if (ka->use_master_clock) |
1442 | atomic_set(&kvm_guest_has_master_clock, 1); | |
1443 | ||
1444 | vclock_mode = pvclock_gtod_data.clock.vclock_mode; | |
b48aa97e MT |
1445 | trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode, |
1446 | vcpus_matched); | |
d828199e MT |
1447 | #endif |
1448 | } | |
1449 | ||
2e762ff7 MT |
1450 | static void kvm_gen_update_masterclock(struct kvm *kvm) |
1451 | { | |
1452 | #ifdef CONFIG_X86_64 | |
1453 | int i; | |
1454 | struct kvm_vcpu *vcpu; | |
1455 | struct kvm_arch *ka = &kvm->arch; | |
1456 | ||
1457 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
1458 | kvm_make_mclock_inprogress_request(kvm); | |
1459 | /* no guest entries from this point */ | |
1460 | pvclock_update_vm_gtod_copy(kvm); | |
1461 | ||
1462 | kvm_for_each_vcpu(i, vcpu, kvm) | |
1463 | set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests); | |
1464 | ||
1465 | /* guest entries allowed */ | |
1466 | kvm_for_each_vcpu(i, vcpu, kvm) | |
1467 | clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests); | |
1468 | ||
1469 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
1470 | #endif | |
1471 | } | |
1472 | ||
34c238a1 | 1473 | static int kvm_guest_time_update(struct kvm_vcpu *v) |
18068523 | 1474 | { |
d828199e | 1475 | unsigned long flags, this_tsc_khz; |
18068523 | 1476 | struct kvm_vcpu_arch *vcpu = &v->arch; |
d828199e | 1477 | struct kvm_arch *ka = &v->kvm->arch; |
1d5f066e | 1478 | s64 kernel_ns, max_kernel_ns; |
d828199e | 1479 | u64 tsc_timestamp, host_tsc; |
0b79459b | 1480 | struct pvclock_vcpu_time_info guest_hv_clock; |
51d59c6b | 1481 | u8 pvclock_flags; |
d828199e MT |
1482 | bool use_master_clock; |
1483 | ||
1484 | kernel_ns = 0; | |
1485 | host_tsc = 0; | |
18068523 | 1486 | |
d828199e MT |
1487 | /* |
1488 | * If the host uses TSC clock, then passthrough TSC as stable | |
1489 | * to the guest. | |
1490 | */ | |
1491 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
1492 | use_master_clock = ka->use_master_clock; | |
1493 | if (use_master_clock) { | |
1494 | host_tsc = ka->master_cycle_now; | |
1495 | kernel_ns = ka->master_kernel_ns; | |
1496 | } | |
1497 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
c09664bb MT |
1498 | |
1499 | /* Keep irq disabled to prevent changes to the clock */ | |
1500 | local_irq_save(flags); | |
1501 | this_tsc_khz = __get_cpu_var(cpu_tsc_khz); | |
1502 | if (unlikely(this_tsc_khz == 0)) { | |
1503 | local_irq_restore(flags); | |
1504 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); | |
1505 | return 1; | |
1506 | } | |
d828199e MT |
1507 | if (!use_master_clock) { |
1508 | host_tsc = native_read_tsc(); | |
1509 | kernel_ns = get_kernel_ns(); | |
1510 | } | |
1511 | ||
1512 | tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc); | |
1513 | ||
c285545f ZA |
1514 | /* |
1515 | * We may have to catch up the TSC to match elapsed wall clock | |
1516 | * time for two reasons, even if kvmclock is used. | |
1517 | * 1) CPU could have been running below the maximum TSC rate | |
1518 | * 2) Broken TSC compensation resets the base at each VCPU | |
1519 | * entry to avoid unknown leaps of TSC even when running | |
1520 | * again on the same CPU. This may cause apparent elapsed | |
1521 | * time to disappear, and the guest to stand still or run | |
1522 | * very slowly. | |
1523 | */ | |
1524 | if (vcpu->tsc_catchup) { | |
1525 | u64 tsc = compute_guest_tsc(v, kernel_ns); | |
1526 | if (tsc > tsc_timestamp) { | |
f1e2b260 | 1527 | adjust_tsc_offset_guest(v, tsc - tsc_timestamp); |
c285545f ZA |
1528 | tsc_timestamp = tsc; |
1529 | } | |
50d0a0f9 GH |
1530 | } |
1531 | ||
18068523 GOC |
1532 | local_irq_restore(flags); |
1533 | ||
0b79459b | 1534 | if (!vcpu->pv_time_enabled) |
c285545f | 1535 | return 0; |
18068523 | 1536 | |
1d5f066e ZA |
1537 | /* |
1538 | * Time as measured by the TSC may go backwards when resetting the base | |
1539 | * tsc_timestamp. The reason for this is that the TSC resolution is | |
1540 | * higher than the resolution of the other clock scales. Thus, many | |
1541 | * possible measurments of the TSC correspond to one measurement of any | |
1542 | * other clock, and so a spread of values is possible. This is not a | |
1543 | * problem for the computation of the nanosecond clock; with TSC rates | |
1544 | * around 1GHZ, there can only be a few cycles which correspond to one | |
1545 | * nanosecond value, and any path through this code will inevitably | |
1546 | * take longer than that. However, with the kernel_ns value itself, | |
1547 | * the precision may be much lower, down to HZ granularity. If the | |
1548 | * first sampling of TSC against kernel_ns ends in the low part of the | |
1549 | * range, and the second in the high end of the range, we can get: | |
1550 | * | |
1551 | * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new | |
1552 | * | |
1553 | * As the sampling errors potentially range in the thousands of cycles, | |
1554 | * it is possible such a time value has already been observed by the | |
1555 | * guest. To protect against this, we must compute the system time as | |
1556 | * observed by the guest and ensure the new system time is greater. | |
1557 | */ | |
1558 | max_kernel_ns = 0; | |
b183aa58 | 1559 | if (vcpu->hv_clock.tsc_timestamp) { |
1d5f066e ZA |
1560 | max_kernel_ns = vcpu->last_guest_tsc - |
1561 | vcpu->hv_clock.tsc_timestamp; | |
1562 | max_kernel_ns = pvclock_scale_delta(max_kernel_ns, | |
1563 | vcpu->hv_clock.tsc_to_system_mul, | |
1564 | vcpu->hv_clock.tsc_shift); | |
1565 | max_kernel_ns += vcpu->last_kernel_ns; | |
1566 | } | |
afbcf7ab | 1567 | |
e48672fa | 1568 | if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) { |
5f4e3f88 ZA |
1569 | kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz, |
1570 | &vcpu->hv_clock.tsc_shift, | |
1571 | &vcpu->hv_clock.tsc_to_system_mul); | |
e48672fa | 1572 | vcpu->hw_tsc_khz = this_tsc_khz; |
8cfdc000 ZA |
1573 | } |
1574 | ||
d828199e MT |
1575 | /* with a master <monotonic time, tsc value> tuple, |
1576 | * pvclock clock reads always increase at the (scaled) rate | |
1577 | * of guest TSC - no need to deal with sampling errors. | |
1578 | */ | |
1579 | if (!use_master_clock) { | |
1580 | if (max_kernel_ns > kernel_ns) | |
1581 | kernel_ns = max_kernel_ns; | |
1582 | } | |
8cfdc000 | 1583 | /* With all the info we got, fill in the values */ |
1d5f066e | 1584 | vcpu->hv_clock.tsc_timestamp = tsc_timestamp; |
759379dd | 1585 | vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset; |
1d5f066e | 1586 | vcpu->last_kernel_ns = kernel_ns; |
28e4639a | 1587 | vcpu->last_guest_tsc = tsc_timestamp; |
51d59c6b | 1588 | |
18068523 GOC |
1589 | /* |
1590 | * The interface expects us to write an even number signaling that the | |
1591 | * update is finished. Since the guest won't see the intermediate | |
50d0a0f9 | 1592 | * state, we just increase by 2 at the end. |
18068523 | 1593 | */ |
50d0a0f9 | 1594 | vcpu->hv_clock.version += 2; |
18068523 | 1595 | |
0b79459b AH |
1596 | if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time, |
1597 | &guest_hv_clock, sizeof(guest_hv_clock)))) | |
1598 | return 0; | |
78c0337a MT |
1599 | |
1600 | /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */ | |
0b79459b | 1601 | pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED); |
78c0337a MT |
1602 | |
1603 | if (vcpu->pvclock_set_guest_stopped_request) { | |
1604 | pvclock_flags |= PVCLOCK_GUEST_STOPPED; | |
1605 | vcpu->pvclock_set_guest_stopped_request = false; | |
1606 | } | |
1607 | ||
d828199e MT |
1608 | /* If the host uses TSC clocksource, then it is stable */ |
1609 | if (use_master_clock) | |
1610 | pvclock_flags |= PVCLOCK_TSC_STABLE_BIT; | |
1611 | ||
78c0337a MT |
1612 | vcpu->hv_clock.flags = pvclock_flags; |
1613 | ||
0b79459b AH |
1614 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, |
1615 | &vcpu->hv_clock, | |
1616 | sizeof(vcpu->hv_clock)); | |
8cfdc000 | 1617 | return 0; |
c8076604 GH |
1618 | } |
1619 | ||
0061d53d MT |
1620 | /* |
1621 | * kvmclock updates which are isolated to a given vcpu, such as | |
1622 | * vcpu->cpu migration, should not allow system_timestamp from | |
1623 | * the rest of the vcpus to remain static. Otherwise ntp frequency | |
1624 | * correction applies to one vcpu's system_timestamp but not | |
1625 | * the others. | |
1626 | * | |
1627 | * So in those cases, request a kvmclock update for all vcpus. | |
1628 | * The worst case for a remote vcpu to update its kvmclock | |
1629 | * is then bounded by maximum nohz sleep latency. | |
1630 | */ | |
1631 | ||
1632 | static void kvm_gen_kvmclock_update(struct kvm_vcpu *v) | |
1633 | { | |
1634 | int i; | |
1635 | struct kvm *kvm = v->kvm; | |
1636 | struct kvm_vcpu *vcpu; | |
1637 | ||
1638 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
1639 | set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests); | |
1640 | kvm_vcpu_kick(vcpu); | |
1641 | } | |
1642 | } | |
1643 | ||
9ba075a6 AK |
1644 | static bool msr_mtrr_valid(unsigned msr) |
1645 | { | |
1646 | switch (msr) { | |
1647 | case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1: | |
1648 | case MSR_MTRRfix64K_00000: | |
1649 | case MSR_MTRRfix16K_80000: | |
1650 | case MSR_MTRRfix16K_A0000: | |
1651 | case MSR_MTRRfix4K_C0000: | |
1652 | case MSR_MTRRfix4K_C8000: | |
1653 | case MSR_MTRRfix4K_D0000: | |
1654 | case MSR_MTRRfix4K_D8000: | |
1655 | case MSR_MTRRfix4K_E0000: | |
1656 | case MSR_MTRRfix4K_E8000: | |
1657 | case MSR_MTRRfix4K_F0000: | |
1658 | case MSR_MTRRfix4K_F8000: | |
1659 | case MSR_MTRRdefType: | |
1660 | case MSR_IA32_CR_PAT: | |
1661 | return true; | |
1662 | case 0x2f8: | |
1663 | return true; | |
1664 | } | |
1665 | return false; | |
1666 | } | |
1667 | ||
d6289b93 MT |
1668 | static bool valid_pat_type(unsigned t) |
1669 | { | |
1670 | return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */ | |
1671 | } | |
1672 | ||
1673 | static bool valid_mtrr_type(unsigned t) | |
1674 | { | |
1675 | return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */ | |
1676 | } | |
1677 | ||
1678 | static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data) | |
1679 | { | |
1680 | int i; | |
1681 | ||
1682 | if (!msr_mtrr_valid(msr)) | |
1683 | return false; | |
1684 | ||
1685 | if (msr == MSR_IA32_CR_PAT) { | |
1686 | for (i = 0; i < 8; i++) | |
1687 | if (!valid_pat_type((data >> (i * 8)) & 0xff)) | |
1688 | return false; | |
1689 | return true; | |
1690 | } else if (msr == MSR_MTRRdefType) { | |
1691 | if (data & ~0xcff) | |
1692 | return false; | |
1693 | return valid_mtrr_type(data & 0xff); | |
1694 | } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) { | |
1695 | for (i = 0; i < 8 ; i++) | |
1696 | if (!valid_mtrr_type((data >> (i * 8)) & 0xff)) | |
1697 | return false; | |
1698 | return true; | |
1699 | } | |
1700 | ||
1701 | /* variable MTRRs */ | |
1702 | return valid_mtrr_type(data & 0xff); | |
1703 | } | |
1704 | ||
9ba075a6 AK |
1705 | static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data) |
1706 | { | |
0bed3b56 SY |
1707 | u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges; |
1708 | ||
d6289b93 | 1709 | if (!mtrr_valid(vcpu, msr, data)) |
9ba075a6 AK |
1710 | return 1; |
1711 | ||
0bed3b56 SY |
1712 | if (msr == MSR_MTRRdefType) { |
1713 | vcpu->arch.mtrr_state.def_type = data; | |
1714 | vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10; | |
1715 | } else if (msr == MSR_MTRRfix64K_00000) | |
1716 | p[0] = data; | |
1717 | else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000) | |
1718 | p[1 + msr - MSR_MTRRfix16K_80000] = data; | |
1719 | else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000) | |
1720 | p[3 + msr - MSR_MTRRfix4K_C0000] = data; | |
1721 | else if (msr == MSR_IA32_CR_PAT) | |
1722 | vcpu->arch.pat = data; | |
1723 | else { /* Variable MTRRs */ | |
1724 | int idx, is_mtrr_mask; | |
1725 | u64 *pt; | |
1726 | ||
1727 | idx = (msr - 0x200) / 2; | |
1728 | is_mtrr_mask = msr - 0x200 - 2 * idx; | |
1729 | if (!is_mtrr_mask) | |
1730 | pt = | |
1731 | (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo; | |
1732 | else | |
1733 | pt = | |
1734 | (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo; | |
1735 | *pt = data; | |
1736 | } | |
1737 | ||
1738 | kvm_mmu_reset_context(vcpu); | |
9ba075a6 AK |
1739 | return 0; |
1740 | } | |
15c4a640 | 1741 | |
890ca9ae | 1742 | static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data) |
15c4a640 | 1743 | { |
890ca9ae HY |
1744 | u64 mcg_cap = vcpu->arch.mcg_cap; |
1745 | unsigned bank_num = mcg_cap & 0xff; | |
1746 | ||
15c4a640 | 1747 | switch (msr) { |
15c4a640 | 1748 | case MSR_IA32_MCG_STATUS: |
890ca9ae | 1749 | vcpu->arch.mcg_status = data; |
15c4a640 | 1750 | break; |
c7ac679c | 1751 | case MSR_IA32_MCG_CTL: |
890ca9ae HY |
1752 | if (!(mcg_cap & MCG_CTL_P)) |
1753 | return 1; | |
1754 | if (data != 0 && data != ~(u64)0) | |
1755 | return -1; | |
1756 | vcpu->arch.mcg_ctl = data; | |
1757 | break; | |
1758 | default: | |
1759 | if (msr >= MSR_IA32_MC0_CTL && | |
1760 | msr < MSR_IA32_MC0_CTL + 4 * bank_num) { | |
1761 | u32 offset = msr - MSR_IA32_MC0_CTL; | |
114be429 AP |
1762 | /* only 0 or all 1s can be written to IA32_MCi_CTL |
1763 | * some Linux kernels though clear bit 10 in bank 4 to | |
1764 | * workaround a BIOS/GART TBL issue on AMD K8s, ignore | |
1765 | * this to avoid an uncatched #GP in the guest | |
1766 | */ | |
890ca9ae | 1767 | if ((offset & 0x3) == 0 && |
114be429 | 1768 | data != 0 && (data | (1 << 10)) != ~(u64)0) |
890ca9ae HY |
1769 | return -1; |
1770 | vcpu->arch.mce_banks[offset] = data; | |
1771 | break; | |
1772 | } | |
1773 | return 1; | |
1774 | } | |
1775 | return 0; | |
1776 | } | |
1777 | ||
ffde22ac ES |
1778 | static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data) |
1779 | { | |
1780 | struct kvm *kvm = vcpu->kvm; | |
1781 | int lm = is_long_mode(vcpu); | |
1782 | u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64 | |
1783 | : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32; | |
1784 | u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64 | |
1785 | : kvm->arch.xen_hvm_config.blob_size_32; | |
1786 | u32 page_num = data & ~PAGE_MASK; | |
1787 | u64 page_addr = data & PAGE_MASK; | |
1788 | u8 *page; | |
1789 | int r; | |
1790 | ||
1791 | r = -E2BIG; | |
1792 | if (page_num >= blob_size) | |
1793 | goto out; | |
1794 | r = -ENOMEM; | |
ff5c2c03 SL |
1795 | page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE); |
1796 | if (IS_ERR(page)) { | |
1797 | r = PTR_ERR(page); | |
ffde22ac | 1798 | goto out; |
ff5c2c03 | 1799 | } |
ffde22ac ES |
1800 | if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE)) |
1801 | goto out_free; | |
1802 | r = 0; | |
1803 | out_free: | |
1804 | kfree(page); | |
1805 | out: | |
1806 | return r; | |
1807 | } | |
1808 | ||
55cd8e5a GN |
1809 | static bool kvm_hv_hypercall_enabled(struct kvm *kvm) |
1810 | { | |
1811 | return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE; | |
1812 | } | |
1813 | ||
1814 | static bool kvm_hv_msr_partition_wide(u32 msr) | |
1815 | { | |
1816 | bool r = false; | |
1817 | switch (msr) { | |
1818 | case HV_X64_MSR_GUEST_OS_ID: | |
1819 | case HV_X64_MSR_HYPERCALL: | |
1820 | r = true; | |
1821 | break; | |
1822 | } | |
1823 | ||
1824 | return r; | |
1825 | } | |
1826 | ||
1827 | static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data) | |
1828 | { | |
1829 | struct kvm *kvm = vcpu->kvm; | |
1830 | ||
1831 | switch (msr) { | |
1832 | case HV_X64_MSR_GUEST_OS_ID: | |
1833 | kvm->arch.hv_guest_os_id = data; | |
1834 | /* setting guest os id to zero disables hypercall page */ | |
1835 | if (!kvm->arch.hv_guest_os_id) | |
1836 | kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE; | |
1837 | break; | |
1838 | case HV_X64_MSR_HYPERCALL: { | |
1839 | u64 gfn; | |
1840 | unsigned long addr; | |
1841 | u8 instructions[4]; | |
1842 | ||
1843 | /* if guest os id is not set hypercall should remain disabled */ | |
1844 | if (!kvm->arch.hv_guest_os_id) | |
1845 | break; | |
1846 | if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) { | |
1847 | kvm->arch.hv_hypercall = data; | |
1848 | break; | |
1849 | } | |
1850 | gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT; | |
1851 | addr = gfn_to_hva(kvm, gfn); | |
1852 | if (kvm_is_error_hva(addr)) | |
1853 | return 1; | |
1854 | kvm_x86_ops->patch_hypercall(vcpu, instructions); | |
1855 | ((unsigned char *)instructions)[3] = 0xc3; /* ret */ | |
8b0cedff | 1856 | if (__copy_to_user((void __user *)addr, instructions, 4)) |
55cd8e5a GN |
1857 | return 1; |
1858 | kvm->arch.hv_hypercall = data; | |
1859 | break; | |
1860 | } | |
1861 | default: | |
a737f256 CD |
1862 | vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x " |
1863 | "data 0x%llx\n", msr, data); | |
55cd8e5a GN |
1864 | return 1; |
1865 | } | |
1866 | return 0; | |
1867 | } | |
1868 | ||
1869 | static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data) | |
1870 | { | |
10388a07 GN |
1871 | switch (msr) { |
1872 | case HV_X64_MSR_APIC_ASSIST_PAGE: { | |
1873 | unsigned long addr; | |
55cd8e5a | 1874 | |
10388a07 GN |
1875 | if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) { |
1876 | vcpu->arch.hv_vapic = data; | |
1877 | break; | |
1878 | } | |
1879 | addr = gfn_to_hva(vcpu->kvm, data >> | |
1880 | HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT); | |
1881 | if (kvm_is_error_hva(addr)) | |
1882 | return 1; | |
8b0cedff | 1883 | if (__clear_user((void __user *)addr, PAGE_SIZE)) |
10388a07 GN |
1884 | return 1; |
1885 | vcpu->arch.hv_vapic = data; | |
1886 | break; | |
1887 | } | |
1888 | case HV_X64_MSR_EOI: | |
1889 | return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data); | |
1890 | case HV_X64_MSR_ICR: | |
1891 | return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data); | |
1892 | case HV_X64_MSR_TPR: | |
1893 | return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data); | |
1894 | default: | |
a737f256 CD |
1895 | vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x " |
1896 | "data 0x%llx\n", msr, data); | |
10388a07 GN |
1897 | return 1; |
1898 | } | |
1899 | ||
1900 | return 0; | |
55cd8e5a GN |
1901 | } |
1902 | ||
344d9588 GN |
1903 | static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data) |
1904 | { | |
1905 | gpa_t gpa = data & ~0x3f; | |
1906 | ||
4a969980 | 1907 | /* Bits 2:5 are reserved, Should be zero */ |
6adba527 | 1908 | if (data & 0x3c) |
344d9588 GN |
1909 | return 1; |
1910 | ||
1911 | vcpu->arch.apf.msr_val = data; | |
1912 | ||
1913 | if (!(data & KVM_ASYNC_PF_ENABLED)) { | |
1914 | kvm_clear_async_pf_completion_queue(vcpu); | |
1915 | kvm_async_pf_hash_reset(vcpu); | |
1916 | return 0; | |
1917 | } | |
1918 | ||
8f964525 AH |
1919 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa, |
1920 | sizeof(u32))) | |
344d9588 GN |
1921 | return 1; |
1922 | ||
6adba527 | 1923 | vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS); |
344d9588 GN |
1924 | kvm_async_pf_wakeup_all(vcpu); |
1925 | return 0; | |
1926 | } | |
1927 | ||
12f9a48f GC |
1928 | static void kvmclock_reset(struct kvm_vcpu *vcpu) |
1929 | { | |
0b79459b | 1930 | vcpu->arch.pv_time_enabled = false; |
12f9a48f GC |
1931 | } |
1932 | ||
c9aaa895 GC |
1933 | static void accumulate_steal_time(struct kvm_vcpu *vcpu) |
1934 | { | |
1935 | u64 delta; | |
1936 | ||
1937 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) | |
1938 | return; | |
1939 | ||
1940 | delta = current->sched_info.run_delay - vcpu->arch.st.last_steal; | |
1941 | vcpu->arch.st.last_steal = current->sched_info.run_delay; | |
1942 | vcpu->arch.st.accum_steal = delta; | |
1943 | } | |
1944 | ||
1945 | static void record_steal_time(struct kvm_vcpu *vcpu) | |
1946 | { | |
1947 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) | |
1948 | return; | |
1949 | ||
1950 | if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime, | |
1951 | &vcpu->arch.st.steal, sizeof(struct kvm_steal_time)))) | |
1952 | return; | |
1953 | ||
1954 | vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal; | |
1955 | vcpu->arch.st.steal.version += 2; | |
1956 | vcpu->arch.st.accum_steal = 0; | |
1957 | ||
1958 | kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime, | |
1959 | &vcpu->arch.st.steal, sizeof(struct kvm_steal_time)); | |
1960 | } | |
1961 | ||
8fe8ab46 | 1962 | int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
15c4a640 | 1963 | { |
5753785f | 1964 | bool pr = false; |
8fe8ab46 WA |
1965 | u32 msr = msr_info->index; |
1966 | u64 data = msr_info->data; | |
5753785f | 1967 | |
15c4a640 | 1968 | switch (msr) { |
2e32b719 BP |
1969 | case MSR_AMD64_NB_CFG: |
1970 | case MSR_IA32_UCODE_REV: | |
1971 | case MSR_IA32_UCODE_WRITE: | |
1972 | case MSR_VM_HSAVE_PA: | |
1973 | case MSR_AMD64_PATCH_LOADER: | |
1974 | case MSR_AMD64_BU_CFG2: | |
1975 | break; | |
1976 | ||
15c4a640 | 1977 | case MSR_EFER: |
b69e8cae | 1978 | return set_efer(vcpu, data); |
8f1589d9 AP |
1979 | case MSR_K7_HWCR: |
1980 | data &= ~(u64)0x40; /* ignore flush filter disable */ | |
82494028 | 1981 | data &= ~(u64)0x100; /* ignore ignne emulation enable */ |
a223c313 | 1982 | data &= ~(u64)0x8; /* ignore TLB cache disable */ |
8f1589d9 | 1983 | if (data != 0) { |
a737f256 CD |
1984 | vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n", |
1985 | data); | |
8f1589d9 AP |
1986 | return 1; |
1987 | } | |
15c4a640 | 1988 | break; |
f7c6d140 AP |
1989 | case MSR_FAM10H_MMIO_CONF_BASE: |
1990 | if (data != 0) { | |
a737f256 CD |
1991 | vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: " |
1992 | "0x%llx\n", data); | |
f7c6d140 AP |
1993 | return 1; |
1994 | } | |
15c4a640 | 1995 | break; |
b5e2fec0 AG |
1996 | case MSR_IA32_DEBUGCTLMSR: |
1997 | if (!data) { | |
1998 | /* We support the non-activated case already */ | |
1999 | break; | |
2000 | } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) { | |
2001 | /* Values other than LBR and BTF are vendor-specific, | |
2002 | thus reserved and should throw a #GP */ | |
2003 | return 1; | |
2004 | } | |
a737f256 CD |
2005 | vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n", |
2006 | __func__, data); | |
b5e2fec0 | 2007 | break; |
9ba075a6 AK |
2008 | case 0x200 ... 0x2ff: |
2009 | return set_msr_mtrr(vcpu, msr, data); | |
15c4a640 CO |
2010 | case MSR_IA32_APICBASE: |
2011 | kvm_set_apic_base(vcpu, data); | |
2012 | break; | |
0105d1a5 GN |
2013 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff: |
2014 | return kvm_x2apic_msr_write(vcpu, msr, data); | |
a3e06bbe LJ |
2015 | case MSR_IA32_TSCDEADLINE: |
2016 | kvm_set_lapic_tscdeadline_msr(vcpu, data); | |
2017 | break; | |
ba904635 WA |
2018 | case MSR_IA32_TSC_ADJUST: |
2019 | if (guest_cpuid_has_tsc_adjust(vcpu)) { | |
2020 | if (!msr_info->host_initiated) { | |
2021 | u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr; | |
2022 | kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true); | |
2023 | } | |
2024 | vcpu->arch.ia32_tsc_adjust_msr = data; | |
2025 | } | |
2026 | break; | |
15c4a640 | 2027 | case MSR_IA32_MISC_ENABLE: |
ad312c7c | 2028 | vcpu->arch.ia32_misc_enable_msr = data; |
15c4a640 | 2029 | break; |
11c6bffa | 2030 | case MSR_KVM_WALL_CLOCK_NEW: |
18068523 GOC |
2031 | case MSR_KVM_WALL_CLOCK: |
2032 | vcpu->kvm->arch.wall_clock = data; | |
2033 | kvm_write_wall_clock(vcpu->kvm, data); | |
2034 | break; | |
11c6bffa | 2035 | case MSR_KVM_SYSTEM_TIME_NEW: |
18068523 | 2036 | case MSR_KVM_SYSTEM_TIME: { |
0b79459b | 2037 | u64 gpa_offset; |
12f9a48f | 2038 | kvmclock_reset(vcpu); |
18068523 GOC |
2039 | |
2040 | vcpu->arch.time = data; | |
0061d53d | 2041 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); |
18068523 GOC |
2042 | |
2043 | /* we verify if the enable bit is set... */ | |
2044 | if (!(data & 1)) | |
2045 | break; | |
2046 | ||
0b79459b | 2047 | gpa_offset = data & ~(PAGE_MASK | 1); |
18068523 | 2048 | |
0b79459b | 2049 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, |
8f964525 AH |
2050 | &vcpu->arch.pv_time, data & ~1ULL, |
2051 | sizeof(struct pvclock_vcpu_time_info))) | |
0b79459b AH |
2052 | vcpu->arch.pv_time_enabled = false; |
2053 | else | |
2054 | vcpu->arch.pv_time_enabled = true; | |
32cad84f | 2055 | |
18068523 GOC |
2056 | break; |
2057 | } | |
344d9588 GN |
2058 | case MSR_KVM_ASYNC_PF_EN: |
2059 | if (kvm_pv_enable_async_pf(vcpu, data)) | |
2060 | return 1; | |
2061 | break; | |
c9aaa895 GC |
2062 | case MSR_KVM_STEAL_TIME: |
2063 | ||
2064 | if (unlikely(!sched_info_on())) | |
2065 | return 1; | |
2066 | ||
2067 | if (data & KVM_STEAL_RESERVED_MASK) | |
2068 | return 1; | |
2069 | ||
2070 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime, | |
8f964525 AH |
2071 | data & KVM_STEAL_VALID_BITS, |
2072 | sizeof(struct kvm_steal_time))) | |
c9aaa895 GC |
2073 | return 1; |
2074 | ||
2075 | vcpu->arch.st.msr_val = data; | |
2076 | ||
2077 | if (!(data & KVM_MSR_ENABLED)) | |
2078 | break; | |
2079 | ||
2080 | vcpu->arch.st.last_steal = current->sched_info.run_delay; | |
2081 | ||
2082 | preempt_disable(); | |
2083 | accumulate_steal_time(vcpu); | |
2084 | preempt_enable(); | |
2085 | ||
2086 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); | |
2087 | ||
2088 | break; | |
ae7a2a3f MT |
2089 | case MSR_KVM_PV_EOI_EN: |
2090 | if (kvm_lapic_enable_pv_eoi(vcpu, data)) | |
2091 | return 1; | |
2092 | break; | |
c9aaa895 | 2093 | |
890ca9ae HY |
2094 | case MSR_IA32_MCG_CTL: |
2095 | case MSR_IA32_MCG_STATUS: | |
2096 | case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1: | |
2097 | return set_msr_mce(vcpu, msr, data); | |
71db6023 AP |
2098 | |
2099 | /* Performance counters are not protected by a CPUID bit, | |
2100 | * so we should check all of them in the generic path for the sake of | |
2101 | * cross vendor migration. | |
2102 | * Writing a zero into the event select MSRs disables them, | |
2103 | * which we perfectly emulate ;-). Any other value should be at least | |
2104 | * reported, some guests depend on them. | |
2105 | */ | |
71db6023 AP |
2106 | case MSR_K7_EVNTSEL0: |
2107 | case MSR_K7_EVNTSEL1: | |
2108 | case MSR_K7_EVNTSEL2: | |
2109 | case MSR_K7_EVNTSEL3: | |
2110 | if (data != 0) | |
a737f256 CD |
2111 | vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: " |
2112 | "0x%x data 0x%llx\n", msr, data); | |
71db6023 AP |
2113 | break; |
2114 | /* at least RHEL 4 unconditionally writes to the perfctr registers, | |
2115 | * so we ignore writes to make it happy. | |
2116 | */ | |
71db6023 AP |
2117 | case MSR_K7_PERFCTR0: |
2118 | case MSR_K7_PERFCTR1: | |
2119 | case MSR_K7_PERFCTR2: | |
2120 | case MSR_K7_PERFCTR3: | |
a737f256 CD |
2121 | vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: " |
2122 | "0x%x data 0x%llx\n", msr, data); | |
71db6023 | 2123 | break; |
5753785f GN |
2124 | case MSR_P6_PERFCTR0: |
2125 | case MSR_P6_PERFCTR1: | |
2126 | pr = true; | |
2127 | case MSR_P6_EVNTSEL0: | |
2128 | case MSR_P6_EVNTSEL1: | |
2129 | if (kvm_pmu_msr(vcpu, msr)) | |
afd80d85 | 2130 | return kvm_pmu_set_msr(vcpu, msr_info); |
5753785f GN |
2131 | |
2132 | if (pr || data != 0) | |
a737f256 CD |
2133 | vcpu_unimpl(vcpu, "disabled perfctr wrmsr: " |
2134 | "0x%x data 0x%llx\n", msr, data); | |
5753785f | 2135 | break; |
84e0cefa JS |
2136 | case MSR_K7_CLK_CTL: |
2137 | /* | |
2138 | * Ignore all writes to this no longer documented MSR. | |
2139 | * Writes are only relevant for old K7 processors, | |
2140 | * all pre-dating SVM, but a recommended workaround from | |
4a969980 | 2141 | * AMD for these chips. It is possible to specify the |
84e0cefa JS |
2142 | * affected processor models on the command line, hence |
2143 | * the need to ignore the workaround. | |
2144 | */ | |
2145 | break; | |
55cd8e5a GN |
2146 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
2147 | if (kvm_hv_msr_partition_wide(msr)) { | |
2148 | int r; | |
2149 | mutex_lock(&vcpu->kvm->lock); | |
2150 | r = set_msr_hyperv_pw(vcpu, msr, data); | |
2151 | mutex_unlock(&vcpu->kvm->lock); | |
2152 | return r; | |
2153 | } else | |
2154 | return set_msr_hyperv(vcpu, msr, data); | |
2155 | break; | |
91c9c3ed | 2156 | case MSR_IA32_BBL_CR_CTL3: |
2157 | /* Drop writes to this legacy MSR -- see rdmsr | |
2158 | * counterpart for further detail. | |
2159 | */ | |
a737f256 | 2160 | vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data); |
91c9c3ed | 2161 | break; |
2b036c6b BO |
2162 | case MSR_AMD64_OSVW_ID_LENGTH: |
2163 | if (!guest_cpuid_has_osvw(vcpu)) | |
2164 | return 1; | |
2165 | vcpu->arch.osvw.length = data; | |
2166 | break; | |
2167 | case MSR_AMD64_OSVW_STATUS: | |
2168 | if (!guest_cpuid_has_osvw(vcpu)) | |
2169 | return 1; | |
2170 | vcpu->arch.osvw.status = data; | |
2171 | break; | |
15c4a640 | 2172 | default: |
ffde22ac ES |
2173 | if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr)) |
2174 | return xen_hvm_config(vcpu, data); | |
f5132b01 | 2175 | if (kvm_pmu_msr(vcpu, msr)) |
afd80d85 | 2176 | return kvm_pmu_set_msr(vcpu, msr_info); |
ed85c068 | 2177 | if (!ignore_msrs) { |
a737f256 CD |
2178 | vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", |
2179 | msr, data); | |
ed85c068 AP |
2180 | return 1; |
2181 | } else { | |
a737f256 CD |
2182 | vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", |
2183 | msr, data); | |
ed85c068 AP |
2184 | break; |
2185 | } | |
15c4a640 CO |
2186 | } |
2187 | return 0; | |
2188 | } | |
2189 | EXPORT_SYMBOL_GPL(kvm_set_msr_common); | |
2190 | ||
2191 | ||
2192 | /* | |
2193 | * Reads an msr value (of 'msr_index') into 'pdata'. | |
2194 | * Returns 0 on success, non-0 otherwise. | |
2195 | * Assumes vcpu_load() was already called. | |
2196 | */ | |
2197 | int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata) | |
2198 | { | |
2199 | return kvm_x86_ops->get_msr(vcpu, msr_index, pdata); | |
2200 | } | |
2201 | ||
9ba075a6 AK |
2202 | static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata) |
2203 | { | |
0bed3b56 SY |
2204 | u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges; |
2205 | ||
9ba075a6 AK |
2206 | if (!msr_mtrr_valid(msr)) |
2207 | return 1; | |
2208 | ||
0bed3b56 SY |
2209 | if (msr == MSR_MTRRdefType) |
2210 | *pdata = vcpu->arch.mtrr_state.def_type + | |
2211 | (vcpu->arch.mtrr_state.enabled << 10); | |
2212 | else if (msr == MSR_MTRRfix64K_00000) | |
2213 | *pdata = p[0]; | |
2214 | else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000) | |
2215 | *pdata = p[1 + msr - MSR_MTRRfix16K_80000]; | |
2216 | else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000) | |
2217 | *pdata = p[3 + msr - MSR_MTRRfix4K_C0000]; | |
2218 | else if (msr == MSR_IA32_CR_PAT) | |
2219 | *pdata = vcpu->arch.pat; | |
2220 | else { /* Variable MTRRs */ | |
2221 | int idx, is_mtrr_mask; | |
2222 | u64 *pt; | |
2223 | ||
2224 | idx = (msr - 0x200) / 2; | |
2225 | is_mtrr_mask = msr - 0x200 - 2 * idx; | |
2226 | if (!is_mtrr_mask) | |
2227 | pt = | |
2228 | (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo; | |
2229 | else | |
2230 | pt = | |
2231 | (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo; | |
2232 | *pdata = *pt; | |
2233 | } | |
2234 | ||
9ba075a6 AK |
2235 | return 0; |
2236 | } | |
2237 | ||
890ca9ae | 2238 | static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata) |
15c4a640 CO |
2239 | { |
2240 | u64 data; | |
890ca9ae HY |
2241 | u64 mcg_cap = vcpu->arch.mcg_cap; |
2242 | unsigned bank_num = mcg_cap & 0xff; | |
15c4a640 CO |
2243 | |
2244 | switch (msr) { | |
15c4a640 CO |
2245 | case MSR_IA32_P5_MC_ADDR: |
2246 | case MSR_IA32_P5_MC_TYPE: | |
890ca9ae HY |
2247 | data = 0; |
2248 | break; | |
15c4a640 | 2249 | case MSR_IA32_MCG_CAP: |
890ca9ae HY |
2250 | data = vcpu->arch.mcg_cap; |
2251 | break; | |
c7ac679c | 2252 | case MSR_IA32_MCG_CTL: |
890ca9ae HY |
2253 | if (!(mcg_cap & MCG_CTL_P)) |
2254 | return 1; | |
2255 | data = vcpu->arch.mcg_ctl; | |
2256 | break; | |
2257 | case MSR_IA32_MCG_STATUS: | |
2258 | data = vcpu->arch.mcg_status; | |
2259 | break; | |
2260 | default: | |
2261 | if (msr >= MSR_IA32_MC0_CTL && | |
2262 | msr < MSR_IA32_MC0_CTL + 4 * bank_num) { | |
2263 | u32 offset = msr - MSR_IA32_MC0_CTL; | |
2264 | data = vcpu->arch.mce_banks[offset]; | |
2265 | break; | |
2266 | } | |
2267 | return 1; | |
2268 | } | |
2269 | *pdata = data; | |
2270 | return 0; | |
2271 | } | |
2272 | ||
55cd8e5a GN |
2273 | static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata) |
2274 | { | |
2275 | u64 data = 0; | |
2276 | struct kvm *kvm = vcpu->kvm; | |
2277 | ||
2278 | switch (msr) { | |
2279 | case HV_X64_MSR_GUEST_OS_ID: | |
2280 | data = kvm->arch.hv_guest_os_id; | |
2281 | break; | |
2282 | case HV_X64_MSR_HYPERCALL: | |
2283 | data = kvm->arch.hv_hypercall; | |
2284 | break; | |
2285 | default: | |
a737f256 | 2286 | vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr); |
55cd8e5a GN |
2287 | return 1; |
2288 | } | |
2289 | ||
2290 | *pdata = data; | |
2291 | return 0; | |
2292 | } | |
2293 | ||
2294 | static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata) | |
2295 | { | |
2296 | u64 data = 0; | |
2297 | ||
2298 | switch (msr) { | |
2299 | case HV_X64_MSR_VP_INDEX: { | |
2300 | int r; | |
2301 | struct kvm_vcpu *v; | |
2302 | kvm_for_each_vcpu(r, v, vcpu->kvm) | |
2303 | if (v == vcpu) | |
2304 | data = r; | |
2305 | break; | |
2306 | } | |
10388a07 GN |
2307 | case HV_X64_MSR_EOI: |
2308 | return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata); | |
2309 | case HV_X64_MSR_ICR: | |
2310 | return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata); | |
2311 | case HV_X64_MSR_TPR: | |
2312 | return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata); | |
14fa67ee | 2313 | case HV_X64_MSR_APIC_ASSIST_PAGE: |
d1613ad5 MW |
2314 | data = vcpu->arch.hv_vapic; |
2315 | break; | |
55cd8e5a | 2316 | default: |
a737f256 | 2317 | vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr); |
55cd8e5a GN |
2318 | return 1; |
2319 | } | |
2320 | *pdata = data; | |
2321 | return 0; | |
2322 | } | |
2323 | ||
890ca9ae HY |
2324 | int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata) |
2325 | { | |
2326 | u64 data; | |
2327 | ||
2328 | switch (msr) { | |
890ca9ae | 2329 | case MSR_IA32_PLATFORM_ID: |
15c4a640 | 2330 | case MSR_IA32_EBL_CR_POWERON: |
b5e2fec0 AG |
2331 | case MSR_IA32_DEBUGCTLMSR: |
2332 | case MSR_IA32_LASTBRANCHFROMIP: | |
2333 | case MSR_IA32_LASTBRANCHTOIP: | |
2334 | case MSR_IA32_LASTINTFROMIP: | |
2335 | case MSR_IA32_LASTINTTOIP: | |
60af2ecd JSR |
2336 | case MSR_K8_SYSCFG: |
2337 | case MSR_K7_HWCR: | |
61a6bd67 | 2338 | case MSR_VM_HSAVE_PA: |
9e699624 | 2339 | case MSR_K7_EVNTSEL0: |
1f3ee616 | 2340 | case MSR_K7_PERFCTR0: |
1fdbd48c | 2341 | case MSR_K8_INT_PENDING_MSG: |
c323c0e5 | 2342 | case MSR_AMD64_NB_CFG: |
f7c6d140 | 2343 | case MSR_FAM10H_MMIO_CONF_BASE: |
2e32b719 | 2344 | case MSR_AMD64_BU_CFG2: |
15c4a640 CO |
2345 | data = 0; |
2346 | break; | |
5753785f GN |
2347 | case MSR_P6_PERFCTR0: |
2348 | case MSR_P6_PERFCTR1: | |
2349 | case MSR_P6_EVNTSEL0: | |
2350 | case MSR_P6_EVNTSEL1: | |
2351 | if (kvm_pmu_msr(vcpu, msr)) | |
2352 | return kvm_pmu_get_msr(vcpu, msr, pdata); | |
2353 | data = 0; | |
2354 | break; | |
742bc670 MT |
2355 | case MSR_IA32_UCODE_REV: |
2356 | data = 0x100000000ULL; | |
2357 | break; | |
9ba075a6 AK |
2358 | case MSR_MTRRcap: |
2359 | data = 0x500 | KVM_NR_VAR_MTRR; | |
2360 | break; | |
2361 | case 0x200 ... 0x2ff: | |
2362 | return get_msr_mtrr(vcpu, msr, pdata); | |
15c4a640 CO |
2363 | case 0xcd: /* fsb frequency */ |
2364 | data = 3; | |
2365 | break; | |
7b914098 JS |
2366 | /* |
2367 | * MSR_EBC_FREQUENCY_ID | |
2368 | * Conservative value valid for even the basic CPU models. | |
2369 | * Models 0,1: 000 in bits 23:21 indicating a bus speed of | |
2370 | * 100MHz, model 2 000 in bits 18:16 indicating 100MHz, | |
2371 | * and 266MHz for model 3, or 4. Set Core Clock | |
2372 | * Frequency to System Bus Frequency Ratio to 1 (bits | |
2373 | * 31:24) even though these are only valid for CPU | |
2374 | * models > 2, however guests may end up dividing or | |
2375 | * multiplying by zero otherwise. | |
2376 | */ | |
2377 | case MSR_EBC_FREQUENCY_ID: | |
2378 | data = 1 << 24; | |
2379 | break; | |
15c4a640 CO |
2380 | case MSR_IA32_APICBASE: |
2381 | data = kvm_get_apic_base(vcpu); | |
2382 | break; | |
0105d1a5 GN |
2383 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff: |
2384 | return kvm_x2apic_msr_read(vcpu, msr, pdata); | |
2385 | break; | |
a3e06bbe LJ |
2386 | case MSR_IA32_TSCDEADLINE: |
2387 | data = kvm_get_lapic_tscdeadline_msr(vcpu); | |
2388 | break; | |
ba904635 WA |
2389 | case MSR_IA32_TSC_ADJUST: |
2390 | data = (u64)vcpu->arch.ia32_tsc_adjust_msr; | |
2391 | break; | |
15c4a640 | 2392 | case MSR_IA32_MISC_ENABLE: |
ad312c7c | 2393 | data = vcpu->arch.ia32_misc_enable_msr; |
15c4a640 | 2394 | break; |
847f0ad8 AG |
2395 | case MSR_IA32_PERF_STATUS: |
2396 | /* TSC increment by tick */ | |
2397 | data = 1000ULL; | |
2398 | /* CPU multiplier */ | |
2399 | data |= (((uint64_t)4ULL) << 40); | |
2400 | break; | |
15c4a640 | 2401 | case MSR_EFER: |
f6801dff | 2402 | data = vcpu->arch.efer; |
15c4a640 | 2403 | break; |
18068523 | 2404 | case MSR_KVM_WALL_CLOCK: |
11c6bffa | 2405 | case MSR_KVM_WALL_CLOCK_NEW: |
18068523 GOC |
2406 | data = vcpu->kvm->arch.wall_clock; |
2407 | break; | |
2408 | case MSR_KVM_SYSTEM_TIME: | |
11c6bffa | 2409 | case MSR_KVM_SYSTEM_TIME_NEW: |
18068523 GOC |
2410 | data = vcpu->arch.time; |
2411 | break; | |
344d9588 GN |
2412 | case MSR_KVM_ASYNC_PF_EN: |
2413 | data = vcpu->arch.apf.msr_val; | |
2414 | break; | |
c9aaa895 GC |
2415 | case MSR_KVM_STEAL_TIME: |
2416 | data = vcpu->arch.st.msr_val; | |
2417 | break; | |
1d92128f MT |
2418 | case MSR_KVM_PV_EOI_EN: |
2419 | data = vcpu->arch.pv_eoi.msr_val; | |
2420 | break; | |
890ca9ae HY |
2421 | case MSR_IA32_P5_MC_ADDR: |
2422 | case MSR_IA32_P5_MC_TYPE: | |
2423 | case MSR_IA32_MCG_CAP: | |
2424 | case MSR_IA32_MCG_CTL: | |
2425 | case MSR_IA32_MCG_STATUS: | |
2426 | case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1: | |
2427 | return get_msr_mce(vcpu, msr, pdata); | |
84e0cefa JS |
2428 | case MSR_K7_CLK_CTL: |
2429 | /* | |
2430 | * Provide expected ramp-up count for K7. All other | |
2431 | * are set to zero, indicating minimum divisors for | |
2432 | * every field. | |
2433 | * | |
2434 | * This prevents guest kernels on AMD host with CPU | |
2435 | * type 6, model 8 and higher from exploding due to | |
2436 | * the rdmsr failing. | |
2437 | */ | |
2438 | data = 0x20000000; | |
2439 | break; | |
55cd8e5a GN |
2440 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
2441 | if (kvm_hv_msr_partition_wide(msr)) { | |
2442 | int r; | |
2443 | mutex_lock(&vcpu->kvm->lock); | |
2444 | r = get_msr_hyperv_pw(vcpu, msr, pdata); | |
2445 | mutex_unlock(&vcpu->kvm->lock); | |
2446 | return r; | |
2447 | } else | |
2448 | return get_msr_hyperv(vcpu, msr, pdata); | |
2449 | break; | |
91c9c3ed | 2450 | case MSR_IA32_BBL_CR_CTL3: |
2451 | /* This legacy MSR exists but isn't fully documented in current | |
2452 | * silicon. It is however accessed by winxp in very narrow | |
2453 | * scenarios where it sets bit #19, itself documented as | |
2454 | * a "reserved" bit. Best effort attempt to source coherent | |
2455 | * read data here should the balance of the register be | |
2456 | * interpreted by the guest: | |
2457 | * | |
2458 | * L2 cache control register 3: 64GB range, 256KB size, | |
2459 | * enabled, latency 0x1, configured | |
2460 | */ | |
2461 | data = 0xbe702111; | |
2462 | break; | |
2b036c6b BO |
2463 | case MSR_AMD64_OSVW_ID_LENGTH: |
2464 | if (!guest_cpuid_has_osvw(vcpu)) | |
2465 | return 1; | |
2466 | data = vcpu->arch.osvw.length; | |
2467 | break; | |
2468 | case MSR_AMD64_OSVW_STATUS: | |
2469 | if (!guest_cpuid_has_osvw(vcpu)) | |
2470 | return 1; | |
2471 | data = vcpu->arch.osvw.status; | |
2472 | break; | |
15c4a640 | 2473 | default: |
f5132b01 GN |
2474 | if (kvm_pmu_msr(vcpu, msr)) |
2475 | return kvm_pmu_get_msr(vcpu, msr, pdata); | |
ed85c068 | 2476 | if (!ignore_msrs) { |
a737f256 | 2477 | vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr); |
ed85c068 AP |
2478 | return 1; |
2479 | } else { | |
a737f256 | 2480 | vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr); |
ed85c068 AP |
2481 | data = 0; |
2482 | } | |
2483 | break; | |
15c4a640 CO |
2484 | } |
2485 | *pdata = data; | |
2486 | return 0; | |
2487 | } | |
2488 | EXPORT_SYMBOL_GPL(kvm_get_msr_common); | |
2489 | ||
313a3dc7 CO |
2490 | /* |
2491 | * Read or write a bunch of msrs. All parameters are kernel addresses. | |
2492 | * | |
2493 | * @return number of msrs set successfully. | |
2494 | */ | |
2495 | static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs, | |
2496 | struct kvm_msr_entry *entries, | |
2497 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
2498 | unsigned index, u64 *data)) | |
2499 | { | |
f656ce01 | 2500 | int i, idx; |
313a3dc7 | 2501 | |
f656ce01 | 2502 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
313a3dc7 CO |
2503 | for (i = 0; i < msrs->nmsrs; ++i) |
2504 | if (do_msr(vcpu, entries[i].index, &entries[i].data)) | |
2505 | break; | |
f656ce01 | 2506 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
313a3dc7 | 2507 | |
313a3dc7 CO |
2508 | return i; |
2509 | } | |
2510 | ||
2511 | /* | |
2512 | * Read or write a bunch of msrs. Parameters are user addresses. | |
2513 | * | |
2514 | * @return number of msrs set successfully. | |
2515 | */ | |
2516 | static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs, | |
2517 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
2518 | unsigned index, u64 *data), | |
2519 | int writeback) | |
2520 | { | |
2521 | struct kvm_msrs msrs; | |
2522 | struct kvm_msr_entry *entries; | |
2523 | int r, n; | |
2524 | unsigned size; | |
2525 | ||
2526 | r = -EFAULT; | |
2527 | if (copy_from_user(&msrs, user_msrs, sizeof msrs)) | |
2528 | goto out; | |
2529 | ||
2530 | r = -E2BIG; | |
2531 | if (msrs.nmsrs >= MAX_IO_MSRS) | |
2532 | goto out; | |
2533 | ||
313a3dc7 | 2534 | size = sizeof(struct kvm_msr_entry) * msrs.nmsrs; |
ff5c2c03 SL |
2535 | entries = memdup_user(user_msrs->entries, size); |
2536 | if (IS_ERR(entries)) { | |
2537 | r = PTR_ERR(entries); | |
313a3dc7 | 2538 | goto out; |
ff5c2c03 | 2539 | } |
313a3dc7 CO |
2540 | |
2541 | r = n = __msr_io(vcpu, &msrs, entries, do_msr); | |
2542 | if (r < 0) | |
2543 | goto out_free; | |
2544 | ||
2545 | r = -EFAULT; | |
2546 | if (writeback && copy_to_user(user_msrs->entries, entries, size)) | |
2547 | goto out_free; | |
2548 | ||
2549 | r = n; | |
2550 | ||
2551 | out_free: | |
7a73c028 | 2552 | kfree(entries); |
313a3dc7 CO |
2553 | out: |
2554 | return r; | |
2555 | } | |
2556 | ||
018d00d2 ZX |
2557 | int kvm_dev_ioctl_check_extension(long ext) |
2558 | { | |
2559 | int r; | |
2560 | ||
2561 | switch (ext) { | |
2562 | case KVM_CAP_IRQCHIP: | |
2563 | case KVM_CAP_HLT: | |
2564 | case KVM_CAP_MMU_SHADOW_CACHE_CONTROL: | |
018d00d2 | 2565 | case KVM_CAP_SET_TSS_ADDR: |
07716717 | 2566 | case KVM_CAP_EXT_CPUID: |
9c15bb1d | 2567 | case KVM_CAP_EXT_EMUL_CPUID: |
c8076604 | 2568 | case KVM_CAP_CLOCKSOURCE: |
7837699f | 2569 | case KVM_CAP_PIT: |
a28e4f5a | 2570 | case KVM_CAP_NOP_IO_DELAY: |
62d9f0db | 2571 | case KVM_CAP_MP_STATE: |
ed848624 | 2572 | case KVM_CAP_SYNC_MMU: |
a355c85c | 2573 | case KVM_CAP_USER_NMI: |
52d939a0 | 2574 | case KVM_CAP_REINJECT_CONTROL: |
4925663a | 2575 | case KVM_CAP_IRQ_INJECT_STATUS: |
721eecbf | 2576 | case KVM_CAP_IRQFD: |
d34e6b17 | 2577 | case KVM_CAP_IOEVENTFD: |
c5ff41ce | 2578 | case KVM_CAP_PIT2: |
e9f42757 | 2579 | case KVM_CAP_PIT_STATE2: |
b927a3ce | 2580 | case KVM_CAP_SET_IDENTITY_MAP_ADDR: |
ffde22ac | 2581 | case KVM_CAP_XEN_HVM: |
afbcf7ab | 2582 | case KVM_CAP_ADJUST_CLOCK: |
3cfc3092 | 2583 | case KVM_CAP_VCPU_EVENTS: |
55cd8e5a | 2584 | case KVM_CAP_HYPERV: |
10388a07 | 2585 | case KVM_CAP_HYPERV_VAPIC: |
c25bc163 | 2586 | case KVM_CAP_HYPERV_SPIN: |
ab9f4ecb | 2587 | case KVM_CAP_PCI_SEGMENT: |
a1efbe77 | 2588 | case KVM_CAP_DEBUGREGS: |
d2be1651 | 2589 | case KVM_CAP_X86_ROBUST_SINGLESTEP: |
2d5b5a66 | 2590 | case KVM_CAP_XSAVE: |
344d9588 | 2591 | case KVM_CAP_ASYNC_PF: |
92a1f12d | 2592 | case KVM_CAP_GET_TSC_KHZ: |
1c0b28c2 | 2593 | case KVM_CAP_KVMCLOCK_CTRL: |
4d8b81ab | 2594 | case KVM_CAP_READONLY_MEM: |
2a5bab10 AW |
2595 | #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT |
2596 | case KVM_CAP_ASSIGN_DEV_IRQ: | |
2597 | case KVM_CAP_PCI_2_3: | |
2598 | #endif | |
018d00d2 ZX |
2599 | r = 1; |
2600 | break; | |
542472b5 LV |
2601 | case KVM_CAP_COALESCED_MMIO: |
2602 | r = KVM_COALESCED_MMIO_PAGE_OFFSET; | |
2603 | break; | |
774ead3a AK |
2604 | case KVM_CAP_VAPIC: |
2605 | r = !kvm_x86_ops->cpu_has_accelerated_tpr(); | |
2606 | break; | |
f725230a | 2607 | case KVM_CAP_NR_VCPUS: |
8c3ba334 SL |
2608 | r = KVM_SOFT_MAX_VCPUS; |
2609 | break; | |
2610 | case KVM_CAP_MAX_VCPUS: | |
f725230a AK |
2611 | r = KVM_MAX_VCPUS; |
2612 | break; | |
a988b910 | 2613 | case KVM_CAP_NR_MEMSLOTS: |
bbacc0c1 | 2614 | r = KVM_USER_MEM_SLOTS; |
a988b910 | 2615 | break; |
a68a6a72 MT |
2616 | case KVM_CAP_PV_MMU: /* obsolete */ |
2617 | r = 0; | |
2f333bcb | 2618 | break; |
4cee4b72 | 2619 | #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT |
62c476c7 | 2620 | case KVM_CAP_IOMMU: |
a1b60c1c | 2621 | r = iommu_present(&pci_bus_type); |
62c476c7 | 2622 | break; |
4cee4b72 | 2623 | #endif |
890ca9ae HY |
2624 | case KVM_CAP_MCE: |
2625 | r = KVM_MAX_MCE_BANKS; | |
2626 | break; | |
2d5b5a66 SY |
2627 | case KVM_CAP_XCRS: |
2628 | r = cpu_has_xsave; | |
2629 | break; | |
92a1f12d JR |
2630 | case KVM_CAP_TSC_CONTROL: |
2631 | r = kvm_has_tsc_control; | |
2632 | break; | |
4d25a066 JK |
2633 | case KVM_CAP_TSC_DEADLINE_TIMER: |
2634 | r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER); | |
2635 | break; | |
018d00d2 ZX |
2636 | default: |
2637 | r = 0; | |
2638 | break; | |
2639 | } | |
2640 | return r; | |
2641 | ||
2642 | } | |
2643 | ||
043405e1 CO |
2644 | long kvm_arch_dev_ioctl(struct file *filp, |
2645 | unsigned int ioctl, unsigned long arg) | |
2646 | { | |
2647 | void __user *argp = (void __user *)arg; | |
2648 | long r; | |
2649 | ||
2650 | switch (ioctl) { | |
2651 | case KVM_GET_MSR_INDEX_LIST: { | |
2652 | struct kvm_msr_list __user *user_msr_list = argp; | |
2653 | struct kvm_msr_list msr_list; | |
2654 | unsigned n; | |
2655 | ||
2656 | r = -EFAULT; | |
2657 | if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list)) | |
2658 | goto out; | |
2659 | n = msr_list.nmsrs; | |
2660 | msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs); | |
2661 | if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list)) | |
2662 | goto out; | |
2663 | r = -E2BIG; | |
e125e7b6 | 2664 | if (n < msr_list.nmsrs) |
043405e1 CO |
2665 | goto out; |
2666 | r = -EFAULT; | |
2667 | if (copy_to_user(user_msr_list->indices, &msrs_to_save, | |
2668 | num_msrs_to_save * sizeof(u32))) | |
2669 | goto out; | |
e125e7b6 | 2670 | if (copy_to_user(user_msr_list->indices + num_msrs_to_save, |
043405e1 CO |
2671 | &emulated_msrs, |
2672 | ARRAY_SIZE(emulated_msrs) * sizeof(u32))) | |
2673 | goto out; | |
2674 | r = 0; | |
2675 | break; | |
2676 | } | |
9c15bb1d BP |
2677 | case KVM_GET_SUPPORTED_CPUID: |
2678 | case KVM_GET_EMULATED_CPUID: { | |
674eea0f AK |
2679 | struct kvm_cpuid2 __user *cpuid_arg = argp; |
2680 | struct kvm_cpuid2 cpuid; | |
2681 | ||
2682 | r = -EFAULT; | |
2683 | if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid)) | |
2684 | goto out; | |
9c15bb1d BP |
2685 | |
2686 | r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries, | |
2687 | ioctl); | |
674eea0f AK |
2688 | if (r) |
2689 | goto out; | |
2690 | ||
2691 | r = -EFAULT; | |
2692 | if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid)) | |
2693 | goto out; | |
2694 | r = 0; | |
2695 | break; | |
2696 | } | |
890ca9ae HY |
2697 | case KVM_X86_GET_MCE_CAP_SUPPORTED: { |
2698 | u64 mce_cap; | |
2699 | ||
2700 | mce_cap = KVM_MCE_CAP_SUPPORTED; | |
2701 | r = -EFAULT; | |
2702 | if (copy_to_user(argp, &mce_cap, sizeof mce_cap)) | |
2703 | goto out; | |
2704 | r = 0; | |
2705 | break; | |
2706 | } | |
043405e1 CO |
2707 | default: |
2708 | r = -EINVAL; | |
2709 | } | |
2710 | out: | |
2711 | return r; | |
2712 | } | |
2713 | ||
f5f48ee1 SY |
2714 | static void wbinvd_ipi(void *garbage) |
2715 | { | |
2716 | wbinvd(); | |
2717 | } | |
2718 | ||
2719 | static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
2720 | { | |
e0f0bbc5 | 2721 | return kvm_arch_has_noncoherent_dma(vcpu->kvm); |
f5f48ee1 SY |
2722 | } |
2723 | ||
313a3dc7 CO |
2724 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
2725 | { | |
f5f48ee1 SY |
2726 | /* Address WBINVD may be executed by guest */ |
2727 | if (need_emulate_wbinvd(vcpu)) { | |
2728 | if (kvm_x86_ops->has_wbinvd_exit()) | |
2729 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); | |
2730 | else if (vcpu->cpu != -1 && vcpu->cpu != cpu) | |
2731 | smp_call_function_single(vcpu->cpu, | |
2732 | wbinvd_ipi, NULL, 1); | |
2733 | } | |
2734 | ||
313a3dc7 | 2735 | kvm_x86_ops->vcpu_load(vcpu, cpu); |
8f6055cb | 2736 | |
0dd6a6ed ZA |
2737 | /* Apply any externally detected TSC adjustments (due to suspend) */ |
2738 | if (unlikely(vcpu->arch.tsc_offset_adjustment)) { | |
2739 | adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment); | |
2740 | vcpu->arch.tsc_offset_adjustment = 0; | |
2741 | set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests); | |
2742 | } | |
8f6055cb | 2743 | |
48434c20 | 2744 | if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) { |
6f526ec5 ZA |
2745 | s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 : |
2746 | native_read_tsc() - vcpu->arch.last_host_tsc; | |
e48672fa ZA |
2747 | if (tsc_delta < 0) |
2748 | mark_tsc_unstable("KVM discovered backwards TSC"); | |
c285545f | 2749 | if (check_tsc_unstable()) { |
b183aa58 ZA |
2750 | u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu, |
2751 | vcpu->arch.last_guest_tsc); | |
2752 | kvm_x86_ops->write_tsc_offset(vcpu, offset); | |
c285545f | 2753 | vcpu->arch.tsc_catchup = 1; |
c285545f | 2754 | } |
d98d07ca MT |
2755 | /* |
2756 | * On a host with synchronized TSC, there is no need to update | |
2757 | * kvmclock on vcpu->cpu migration | |
2758 | */ | |
2759 | if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1) | |
0061d53d | 2760 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); |
c285545f ZA |
2761 | if (vcpu->cpu != cpu) |
2762 | kvm_migrate_timers(vcpu); | |
e48672fa | 2763 | vcpu->cpu = cpu; |
6b7d7e76 | 2764 | } |
c9aaa895 GC |
2765 | |
2766 | accumulate_steal_time(vcpu); | |
2767 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); | |
313a3dc7 CO |
2768 | } |
2769 | ||
2770 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
2771 | { | |
02daab21 | 2772 | kvm_x86_ops->vcpu_put(vcpu); |
1c11e713 | 2773 | kvm_put_guest_fpu(vcpu); |
6f526ec5 | 2774 | vcpu->arch.last_host_tsc = native_read_tsc(); |
313a3dc7 CO |
2775 | } |
2776 | ||
313a3dc7 CO |
2777 | static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu, |
2778 | struct kvm_lapic_state *s) | |
2779 | { | |
5a71785d | 2780 | kvm_x86_ops->sync_pir_to_irr(vcpu); |
ad312c7c | 2781 | memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s); |
313a3dc7 CO |
2782 | |
2783 | return 0; | |
2784 | } | |
2785 | ||
2786 | static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu, | |
2787 | struct kvm_lapic_state *s) | |
2788 | { | |
64eb0620 | 2789 | kvm_apic_post_state_restore(vcpu, s); |
cb142eb7 | 2790 | update_cr8_intercept(vcpu); |
313a3dc7 CO |
2791 | |
2792 | return 0; | |
2793 | } | |
2794 | ||
f77bc6a4 ZX |
2795 | static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, |
2796 | struct kvm_interrupt *irq) | |
2797 | { | |
02cdb50f | 2798 | if (irq->irq >= KVM_NR_INTERRUPTS) |
f77bc6a4 ZX |
2799 | return -EINVAL; |
2800 | if (irqchip_in_kernel(vcpu->kvm)) | |
2801 | return -ENXIO; | |
f77bc6a4 | 2802 | |
66fd3f7f | 2803 | kvm_queue_interrupt(vcpu, irq->irq, false); |
3842d135 | 2804 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
f77bc6a4 | 2805 | |
f77bc6a4 ZX |
2806 | return 0; |
2807 | } | |
2808 | ||
c4abb7c9 JK |
2809 | static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu) |
2810 | { | |
c4abb7c9 | 2811 | kvm_inject_nmi(vcpu); |
c4abb7c9 JK |
2812 | |
2813 | return 0; | |
2814 | } | |
2815 | ||
b209749f AK |
2816 | static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu, |
2817 | struct kvm_tpr_access_ctl *tac) | |
2818 | { | |
2819 | if (tac->flags) | |
2820 | return -EINVAL; | |
2821 | vcpu->arch.tpr_access_reporting = !!tac->enabled; | |
2822 | return 0; | |
2823 | } | |
2824 | ||
890ca9ae HY |
2825 | static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu, |
2826 | u64 mcg_cap) | |
2827 | { | |
2828 | int r; | |
2829 | unsigned bank_num = mcg_cap & 0xff, bank; | |
2830 | ||
2831 | r = -EINVAL; | |
a9e38c3e | 2832 | if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS) |
890ca9ae HY |
2833 | goto out; |
2834 | if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000)) | |
2835 | goto out; | |
2836 | r = 0; | |
2837 | vcpu->arch.mcg_cap = mcg_cap; | |
2838 | /* Init IA32_MCG_CTL to all 1s */ | |
2839 | if (mcg_cap & MCG_CTL_P) | |
2840 | vcpu->arch.mcg_ctl = ~(u64)0; | |
2841 | /* Init IA32_MCi_CTL to all 1s */ | |
2842 | for (bank = 0; bank < bank_num; bank++) | |
2843 | vcpu->arch.mce_banks[bank*4] = ~(u64)0; | |
2844 | out: | |
2845 | return r; | |
2846 | } | |
2847 | ||
2848 | static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu, | |
2849 | struct kvm_x86_mce *mce) | |
2850 | { | |
2851 | u64 mcg_cap = vcpu->arch.mcg_cap; | |
2852 | unsigned bank_num = mcg_cap & 0xff; | |
2853 | u64 *banks = vcpu->arch.mce_banks; | |
2854 | ||
2855 | if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL)) | |
2856 | return -EINVAL; | |
2857 | /* | |
2858 | * if IA32_MCG_CTL is not all 1s, the uncorrected error | |
2859 | * reporting is disabled | |
2860 | */ | |
2861 | if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) && | |
2862 | vcpu->arch.mcg_ctl != ~(u64)0) | |
2863 | return 0; | |
2864 | banks += 4 * mce->bank; | |
2865 | /* | |
2866 | * if IA32_MCi_CTL is not all 1s, the uncorrected error | |
2867 | * reporting is disabled for the bank | |
2868 | */ | |
2869 | if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0) | |
2870 | return 0; | |
2871 | if (mce->status & MCI_STATUS_UC) { | |
2872 | if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) || | |
fc78f519 | 2873 | !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) { |
a8eeb04a | 2874 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
890ca9ae HY |
2875 | return 0; |
2876 | } | |
2877 | if (banks[1] & MCI_STATUS_VAL) | |
2878 | mce->status |= MCI_STATUS_OVER; | |
2879 | banks[2] = mce->addr; | |
2880 | banks[3] = mce->misc; | |
2881 | vcpu->arch.mcg_status = mce->mcg_status; | |
2882 | banks[1] = mce->status; | |
2883 | kvm_queue_exception(vcpu, MC_VECTOR); | |
2884 | } else if (!(banks[1] & MCI_STATUS_VAL) | |
2885 | || !(banks[1] & MCI_STATUS_UC)) { | |
2886 | if (banks[1] & MCI_STATUS_VAL) | |
2887 | mce->status |= MCI_STATUS_OVER; | |
2888 | banks[2] = mce->addr; | |
2889 | banks[3] = mce->misc; | |
2890 | banks[1] = mce->status; | |
2891 | } else | |
2892 | banks[1] |= MCI_STATUS_OVER; | |
2893 | return 0; | |
2894 | } | |
2895 | ||
3cfc3092 JK |
2896 | static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu, |
2897 | struct kvm_vcpu_events *events) | |
2898 | { | |
7460fb4a | 2899 | process_nmi(vcpu); |
03b82a30 JK |
2900 | events->exception.injected = |
2901 | vcpu->arch.exception.pending && | |
2902 | !kvm_exception_is_soft(vcpu->arch.exception.nr); | |
3cfc3092 JK |
2903 | events->exception.nr = vcpu->arch.exception.nr; |
2904 | events->exception.has_error_code = vcpu->arch.exception.has_error_code; | |
97e69aa6 | 2905 | events->exception.pad = 0; |
3cfc3092 JK |
2906 | events->exception.error_code = vcpu->arch.exception.error_code; |
2907 | ||
03b82a30 JK |
2908 | events->interrupt.injected = |
2909 | vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft; | |
3cfc3092 | 2910 | events->interrupt.nr = vcpu->arch.interrupt.nr; |
03b82a30 | 2911 | events->interrupt.soft = 0; |
48005f64 JK |
2912 | events->interrupt.shadow = |
2913 | kvm_x86_ops->get_interrupt_shadow(vcpu, | |
2914 | KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI); | |
3cfc3092 JK |
2915 | |
2916 | events->nmi.injected = vcpu->arch.nmi_injected; | |
7460fb4a | 2917 | events->nmi.pending = vcpu->arch.nmi_pending != 0; |
3cfc3092 | 2918 | events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu); |
97e69aa6 | 2919 | events->nmi.pad = 0; |
3cfc3092 | 2920 | |
66450a21 | 2921 | events->sipi_vector = 0; /* never valid when reporting to user space */ |
3cfc3092 | 2922 | |
dab4b911 | 2923 | events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING |
48005f64 | 2924 | | KVM_VCPUEVENT_VALID_SHADOW); |
97e69aa6 | 2925 | memset(&events->reserved, 0, sizeof(events->reserved)); |
3cfc3092 JK |
2926 | } |
2927 | ||
2928 | static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu, | |
2929 | struct kvm_vcpu_events *events) | |
2930 | { | |
dab4b911 | 2931 | if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING |
48005f64 JK |
2932 | | KVM_VCPUEVENT_VALID_SIPI_VECTOR |
2933 | | KVM_VCPUEVENT_VALID_SHADOW)) | |
3cfc3092 JK |
2934 | return -EINVAL; |
2935 | ||
7460fb4a | 2936 | process_nmi(vcpu); |
3cfc3092 JK |
2937 | vcpu->arch.exception.pending = events->exception.injected; |
2938 | vcpu->arch.exception.nr = events->exception.nr; | |
2939 | vcpu->arch.exception.has_error_code = events->exception.has_error_code; | |
2940 | vcpu->arch.exception.error_code = events->exception.error_code; | |
2941 | ||
2942 | vcpu->arch.interrupt.pending = events->interrupt.injected; | |
2943 | vcpu->arch.interrupt.nr = events->interrupt.nr; | |
2944 | vcpu->arch.interrupt.soft = events->interrupt.soft; | |
48005f64 JK |
2945 | if (events->flags & KVM_VCPUEVENT_VALID_SHADOW) |
2946 | kvm_x86_ops->set_interrupt_shadow(vcpu, | |
2947 | events->interrupt.shadow); | |
3cfc3092 JK |
2948 | |
2949 | vcpu->arch.nmi_injected = events->nmi.injected; | |
dab4b911 JK |
2950 | if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING) |
2951 | vcpu->arch.nmi_pending = events->nmi.pending; | |
3cfc3092 JK |
2952 | kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked); |
2953 | ||
66450a21 JK |
2954 | if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR && |
2955 | kvm_vcpu_has_lapic(vcpu)) | |
2956 | vcpu->arch.apic->sipi_vector = events->sipi_vector; | |
3cfc3092 | 2957 | |
3842d135 AK |
2958 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
2959 | ||
3cfc3092 JK |
2960 | return 0; |
2961 | } | |
2962 | ||
a1efbe77 JK |
2963 | static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu, |
2964 | struct kvm_debugregs *dbgregs) | |
2965 | { | |
a1efbe77 JK |
2966 | memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db)); |
2967 | dbgregs->dr6 = vcpu->arch.dr6; | |
2968 | dbgregs->dr7 = vcpu->arch.dr7; | |
2969 | dbgregs->flags = 0; | |
97e69aa6 | 2970 | memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved)); |
a1efbe77 JK |
2971 | } |
2972 | ||
2973 | static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu, | |
2974 | struct kvm_debugregs *dbgregs) | |
2975 | { | |
2976 | if (dbgregs->flags) | |
2977 | return -EINVAL; | |
2978 | ||
a1efbe77 JK |
2979 | memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db)); |
2980 | vcpu->arch.dr6 = dbgregs->dr6; | |
2981 | vcpu->arch.dr7 = dbgregs->dr7; | |
2982 | ||
a1efbe77 JK |
2983 | return 0; |
2984 | } | |
2985 | ||
2d5b5a66 SY |
2986 | static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu, |
2987 | struct kvm_xsave *guest_xsave) | |
2988 | { | |
4344ee98 | 2989 | if (cpu_has_xsave) { |
2d5b5a66 SY |
2990 | memcpy(guest_xsave->region, |
2991 | &vcpu->arch.guest_fpu.state->xsave, | |
4344ee98 PB |
2992 | vcpu->arch.guest_xstate_size); |
2993 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] &= | |
2994 | vcpu->arch.guest_supported_xcr0 | XSTATE_FPSSE; | |
2995 | } else { | |
2d5b5a66 SY |
2996 | memcpy(guest_xsave->region, |
2997 | &vcpu->arch.guest_fpu.state->fxsave, | |
2998 | sizeof(struct i387_fxsave_struct)); | |
2999 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] = | |
3000 | XSTATE_FPSSE; | |
3001 | } | |
3002 | } | |
3003 | ||
3004 | static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu, | |
3005 | struct kvm_xsave *guest_xsave) | |
3006 | { | |
3007 | u64 xstate_bv = | |
3008 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)]; | |
3009 | ||
d7876f1b PB |
3010 | if (cpu_has_xsave) { |
3011 | /* | |
3012 | * Here we allow setting states that are not present in | |
3013 | * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility | |
3014 | * with old userspace. | |
3015 | */ | |
3016 | if (xstate_bv & ~KVM_SUPPORTED_XCR0) | |
3017 | return -EINVAL; | |
3018 | if (xstate_bv & ~host_xcr0) | |
3019 | return -EINVAL; | |
2d5b5a66 | 3020 | memcpy(&vcpu->arch.guest_fpu.state->xsave, |
4344ee98 | 3021 | guest_xsave->region, vcpu->arch.guest_xstate_size); |
d7876f1b | 3022 | } else { |
2d5b5a66 SY |
3023 | if (xstate_bv & ~XSTATE_FPSSE) |
3024 | return -EINVAL; | |
3025 | memcpy(&vcpu->arch.guest_fpu.state->fxsave, | |
3026 | guest_xsave->region, sizeof(struct i387_fxsave_struct)); | |
3027 | } | |
3028 | return 0; | |
3029 | } | |
3030 | ||
3031 | static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu, | |
3032 | struct kvm_xcrs *guest_xcrs) | |
3033 | { | |
3034 | if (!cpu_has_xsave) { | |
3035 | guest_xcrs->nr_xcrs = 0; | |
3036 | return; | |
3037 | } | |
3038 | ||
3039 | guest_xcrs->nr_xcrs = 1; | |
3040 | guest_xcrs->flags = 0; | |
3041 | guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK; | |
3042 | guest_xcrs->xcrs[0].value = vcpu->arch.xcr0; | |
3043 | } | |
3044 | ||
3045 | static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu, | |
3046 | struct kvm_xcrs *guest_xcrs) | |
3047 | { | |
3048 | int i, r = 0; | |
3049 | ||
3050 | if (!cpu_has_xsave) | |
3051 | return -EINVAL; | |
3052 | ||
3053 | if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags) | |
3054 | return -EINVAL; | |
3055 | ||
3056 | for (i = 0; i < guest_xcrs->nr_xcrs; i++) | |
3057 | /* Only support XCR0 currently */ | |
3058 | if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) { | |
3059 | r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK, | |
3060 | guest_xcrs->xcrs[0].value); | |
3061 | break; | |
3062 | } | |
3063 | if (r) | |
3064 | r = -EINVAL; | |
3065 | return r; | |
3066 | } | |
3067 | ||
1c0b28c2 EM |
3068 | /* |
3069 | * kvm_set_guest_paused() indicates to the guest kernel that it has been | |
3070 | * stopped by the hypervisor. This function will be called from the host only. | |
3071 | * EINVAL is returned when the host attempts to set the flag for a guest that | |
3072 | * does not support pv clocks. | |
3073 | */ | |
3074 | static int kvm_set_guest_paused(struct kvm_vcpu *vcpu) | |
3075 | { | |
0b79459b | 3076 | if (!vcpu->arch.pv_time_enabled) |
1c0b28c2 | 3077 | return -EINVAL; |
51d59c6b | 3078 | vcpu->arch.pvclock_set_guest_stopped_request = true; |
1c0b28c2 EM |
3079 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
3080 | return 0; | |
3081 | } | |
3082 | ||
313a3dc7 CO |
3083 | long kvm_arch_vcpu_ioctl(struct file *filp, |
3084 | unsigned int ioctl, unsigned long arg) | |
3085 | { | |
3086 | struct kvm_vcpu *vcpu = filp->private_data; | |
3087 | void __user *argp = (void __user *)arg; | |
3088 | int r; | |
d1ac91d8 AK |
3089 | union { |
3090 | struct kvm_lapic_state *lapic; | |
3091 | struct kvm_xsave *xsave; | |
3092 | struct kvm_xcrs *xcrs; | |
3093 | void *buffer; | |
3094 | } u; | |
3095 | ||
3096 | u.buffer = NULL; | |
313a3dc7 CO |
3097 | switch (ioctl) { |
3098 | case KVM_GET_LAPIC: { | |
2204ae3c MT |
3099 | r = -EINVAL; |
3100 | if (!vcpu->arch.apic) | |
3101 | goto out; | |
d1ac91d8 | 3102 | u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL); |
313a3dc7 | 3103 | |
b772ff36 | 3104 | r = -ENOMEM; |
d1ac91d8 | 3105 | if (!u.lapic) |
b772ff36 | 3106 | goto out; |
d1ac91d8 | 3107 | r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic); |
313a3dc7 CO |
3108 | if (r) |
3109 | goto out; | |
3110 | r = -EFAULT; | |
d1ac91d8 | 3111 | if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state))) |
313a3dc7 CO |
3112 | goto out; |
3113 | r = 0; | |
3114 | break; | |
3115 | } | |
3116 | case KVM_SET_LAPIC: { | |
2204ae3c MT |
3117 | r = -EINVAL; |
3118 | if (!vcpu->arch.apic) | |
3119 | goto out; | |
ff5c2c03 | 3120 | u.lapic = memdup_user(argp, sizeof(*u.lapic)); |
18595411 GC |
3121 | if (IS_ERR(u.lapic)) |
3122 | return PTR_ERR(u.lapic); | |
ff5c2c03 | 3123 | |
d1ac91d8 | 3124 | r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic); |
313a3dc7 CO |
3125 | break; |
3126 | } | |
f77bc6a4 ZX |
3127 | case KVM_INTERRUPT: { |
3128 | struct kvm_interrupt irq; | |
3129 | ||
3130 | r = -EFAULT; | |
3131 | if (copy_from_user(&irq, argp, sizeof irq)) | |
3132 | goto out; | |
3133 | r = kvm_vcpu_ioctl_interrupt(vcpu, &irq); | |
f77bc6a4 ZX |
3134 | break; |
3135 | } | |
c4abb7c9 JK |
3136 | case KVM_NMI: { |
3137 | r = kvm_vcpu_ioctl_nmi(vcpu); | |
c4abb7c9 JK |
3138 | break; |
3139 | } | |
313a3dc7 CO |
3140 | case KVM_SET_CPUID: { |
3141 | struct kvm_cpuid __user *cpuid_arg = argp; | |
3142 | struct kvm_cpuid cpuid; | |
3143 | ||
3144 | r = -EFAULT; | |
3145 | if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid)) | |
3146 | goto out; | |
3147 | r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries); | |
313a3dc7 CO |
3148 | break; |
3149 | } | |
07716717 DK |
3150 | case KVM_SET_CPUID2: { |
3151 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
3152 | struct kvm_cpuid2 cpuid; | |
3153 | ||
3154 | r = -EFAULT; | |
3155 | if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid)) | |
3156 | goto out; | |
3157 | r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid, | |
19355475 | 3158 | cpuid_arg->entries); |
07716717 DK |
3159 | break; |
3160 | } | |
3161 | case KVM_GET_CPUID2: { | |
3162 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
3163 | struct kvm_cpuid2 cpuid; | |
3164 | ||
3165 | r = -EFAULT; | |
3166 | if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid)) | |
3167 | goto out; | |
3168 | r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid, | |
19355475 | 3169 | cpuid_arg->entries); |
07716717 DK |
3170 | if (r) |
3171 | goto out; | |
3172 | r = -EFAULT; | |
3173 | if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid)) | |
3174 | goto out; | |
3175 | r = 0; | |
3176 | break; | |
3177 | } | |
313a3dc7 CO |
3178 | case KVM_GET_MSRS: |
3179 | r = msr_io(vcpu, argp, kvm_get_msr, 1); | |
3180 | break; | |
3181 | case KVM_SET_MSRS: | |
3182 | r = msr_io(vcpu, argp, do_set_msr, 0); | |
3183 | break; | |
b209749f AK |
3184 | case KVM_TPR_ACCESS_REPORTING: { |
3185 | struct kvm_tpr_access_ctl tac; | |
3186 | ||
3187 | r = -EFAULT; | |
3188 | if (copy_from_user(&tac, argp, sizeof tac)) | |
3189 | goto out; | |
3190 | r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac); | |
3191 | if (r) | |
3192 | goto out; | |
3193 | r = -EFAULT; | |
3194 | if (copy_to_user(argp, &tac, sizeof tac)) | |
3195 | goto out; | |
3196 | r = 0; | |
3197 | break; | |
3198 | }; | |
b93463aa AK |
3199 | case KVM_SET_VAPIC_ADDR: { |
3200 | struct kvm_vapic_addr va; | |
3201 | ||
3202 | r = -EINVAL; | |
3203 | if (!irqchip_in_kernel(vcpu->kvm)) | |
3204 | goto out; | |
3205 | r = -EFAULT; | |
3206 | if (copy_from_user(&va, argp, sizeof va)) | |
3207 | goto out; | |
3208 | r = 0; | |
3209 | kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr); | |
3210 | break; | |
3211 | } | |
890ca9ae HY |
3212 | case KVM_X86_SETUP_MCE: { |
3213 | u64 mcg_cap; | |
3214 | ||
3215 | r = -EFAULT; | |
3216 | if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap)) | |
3217 | goto out; | |
3218 | r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap); | |
3219 | break; | |
3220 | } | |
3221 | case KVM_X86_SET_MCE: { | |
3222 | struct kvm_x86_mce mce; | |
3223 | ||
3224 | r = -EFAULT; | |
3225 | if (copy_from_user(&mce, argp, sizeof mce)) | |
3226 | goto out; | |
3227 | r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce); | |
3228 | break; | |
3229 | } | |
3cfc3092 JK |
3230 | case KVM_GET_VCPU_EVENTS: { |
3231 | struct kvm_vcpu_events events; | |
3232 | ||
3233 | kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events); | |
3234 | ||
3235 | r = -EFAULT; | |
3236 | if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events))) | |
3237 | break; | |
3238 | r = 0; | |
3239 | break; | |
3240 | } | |
3241 | case KVM_SET_VCPU_EVENTS: { | |
3242 | struct kvm_vcpu_events events; | |
3243 | ||
3244 | r = -EFAULT; | |
3245 | if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events))) | |
3246 | break; | |
3247 | ||
3248 | r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events); | |
3249 | break; | |
3250 | } | |
a1efbe77 JK |
3251 | case KVM_GET_DEBUGREGS: { |
3252 | struct kvm_debugregs dbgregs; | |
3253 | ||
3254 | kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs); | |
3255 | ||
3256 | r = -EFAULT; | |
3257 | if (copy_to_user(argp, &dbgregs, | |
3258 | sizeof(struct kvm_debugregs))) | |
3259 | break; | |
3260 | r = 0; | |
3261 | break; | |
3262 | } | |
3263 | case KVM_SET_DEBUGREGS: { | |
3264 | struct kvm_debugregs dbgregs; | |
3265 | ||
3266 | r = -EFAULT; | |
3267 | if (copy_from_user(&dbgregs, argp, | |
3268 | sizeof(struct kvm_debugregs))) | |
3269 | break; | |
3270 | ||
3271 | r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs); | |
3272 | break; | |
3273 | } | |
2d5b5a66 | 3274 | case KVM_GET_XSAVE: { |
d1ac91d8 | 3275 | u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL); |
2d5b5a66 | 3276 | r = -ENOMEM; |
d1ac91d8 | 3277 | if (!u.xsave) |
2d5b5a66 SY |
3278 | break; |
3279 | ||
d1ac91d8 | 3280 | kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
3281 | |
3282 | r = -EFAULT; | |
d1ac91d8 | 3283 | if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave))) |
2d5b5a66 SY |
3284 | break; |
3285 | r = 0; | |
3286 | break; | |
3287 | } | |
3288 | case KVM_SET_XSAVE: { | |
ff5c2c03 | 3289 | u.xsave = memdup_user(argp, sizeof(*u.xsave)); |
18595411 GC |
3290 | if (IS_ERR(u.xsave)) |
3291 | return PTR_ERR(u.xsave); | |
2d5b5a66 | 3292 | |
d1ac91d8 | 3293 | r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
3294 | break; |
3295 | } | |
3296 | case KVM_GET_XCRS: { | |
d1ac91d8 | 3297 | u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL); |
2d5b5a66 | 3298 | r = -ENOMEM; |
d1ac91d8 | 3299 | if (!u.xcrs) |
2d5b5a66 SY |
3300 | break; |
3301 | ||
d1ac91d8 | 3302 | kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
3303 | |
3304 | r = -EFAULT; | |
d1ac91d8 | 3305 | if (copy_to_user(argp, u.xcrs, |
2d5b5a66 SY |
3306 | sizeof(struct kvm_xcrs))) |
3307 | break; | |
3308 | r = 0; | |
3309 | break; | |
3310 | } | |
3311 | case KVM_SET_XCRS: { | |
ff5c2c03 | 3312 | u.xcrs = memdup_user(argp, sizeof(*u.xcrs)); |
18595411 GC |
3313 | if (IS_ERR(u.xcrs)) |
3314 | return PTR_ERR(u.xcrs); | |
2d5b5a66 | 3315 | |
d1ac91d8 | 3316 | r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
3317 | break; |
3318 | } | |
92a1f12d JR |
3319 | case KVM_SET_TSC_KHZ: { |
3320 | u32 user_tsc_khz; | |
3321 | ||
3322 | r = -EINVAL; | |
92a1f12d JR |
3323 | user_tsc_khz = (u32)arg; |
3324 | ||
3325 | if (user_tsc_khz >= kvm_max_guest_tsc_khz) | |
3326 | goto out; | |
3327 | ||
cc578287 ZA |
3328 | if (user_tsc_khz == 0) |
3329 | user_tsc_khz = tsc_khz; | |
3330 | ||
3331 | kvm_set_tsc_khz(vcpu, user_tsc_khz); | |
92a1f12d JR |
3332 | |
3333 | r = 0; | |
3334 | goto out; | |
3335 | } | |
3336 | case KVM_GET_TSC_KHZ: { | |
cc578287 | 3337 | r = vcpu->arch.virtual_tsc_khz; |
92a1f12d JR |
3338 | goto out; |
3339 | } | |
1c0b28c2 EM |
3340 | case KVM_KVMCLOCK_CTRL: { |
3341 | r = kvm_set_guest_paused(vcpu); | |
3342 | goto out; | |
3343 | } | |
313a3dc7 CO |
3344 | default: |
3345 | r = -EINVAL; | |
3346 | } | |
3347 | out: | |
d1ac91d8 | 3348 | kfree(u.buffer); |
313a3dc7 CO |
3349 | return r; |
3350 | } | |
3351 | ||
5b1c1493 CO |
3352 | int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
3353 | { | |
3354 | return VM_FAULT_SIGBUS; | |
3355 | } | |
3356 | ||
1fe779f8 CO |
3357 | static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr) |
3358 | { | |
3359 | int ret; | |
3360 | ||
3361 | if (addr > (unsigned int)(-3 * PAGE_SIZE)) | |
951179ce | 3362 | return -EINVAL; |
1fe779f8 CO |
3363 | ret = kvm_x86_ops->set_tss_addr(kvm, addr); |
3364 | return ret; | |
3365 | } | |
3366 | ||
b927a3ce SY |
3367 | static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm, |
3368 | u64 ident_addr) | |
3369 | { | |
3370 | kvm->arch.ept_identity_map_addr = ident_addr; | |
3371 | return 0; | |
3372 | } | |
3373 | ||
1fe779f8 CO |
3374 | static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm, |
3375 | u32 kvm_nr_mmu_pages) | |
3376 | { | |
3377 | if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES) | |
3378 | return -EINVAL; | |
3379 | ||
79fac95e | 3380 | mutex_lock(&kvm->slots_lock); |
1fe779f8 CO |
3381 | |
3382 | kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages); | |
f05e70ac | 3383 | kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages; |
1fe779f8 | 3384 | |
79fac95e | 3385 | mutex_unlock(&kvm->slots_lock); |
1fe779f8 CO |
3386 | return 0; |
3387 | } | |
3388 | ||
3389 | static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm) | |
3390 | { | |
39de71ec | 3391 | return kvm->arch.n_max_mmu_pages; |
1fe779f8 CO |
3392 | } |
3393 | ||
1fe779f8 CO |
3394 | static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) |
3395 | { | |
3396 | int r; | |
3397 | ||
3398 | r = 0; | |
3399 | switch (chip->chip_id) { | |
3400 | case KVM_IRQCHIP_PIC_MASTER: | |
3401 | memcpy(&chip->chip.pic, | |
3402 | &pic_irqchip(kvm)->pics[0], | |
3403 | sizeof(struct kvm_pic_state)); | |
3404 | break; | |
3405 | case KVM_IRQCHIP_PIC_SLAVE: | |
3406 | memcpy(&chip->chip.pic, | |
3407 | &pic_irqchip(kvm)->pics[1], | |
3408 | sizeof(struct kvm_pic_state)); | |
3409 | break; | |
3410 | case KVM_IRQCHIP_IOAPIC: | |
eba0226b | 3411 | r = kvm_get_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
3412 | break; |
3413 | default: | |
3414 | r = -EINVAL; | |
3415 | break; | |
3416 | } | |
3417 | return r; | |
3418 | } | |
3419 | ||
3420 | static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) | |
3421 | { | |
3422 | int r; | |
3423 | ||
3424 | r = 0; | |
3425 | switch (chip->chip_id) { | |
3426 | case KVM_IRQCHIP_PIC_MASTER: | |
f4f51050 | 3427 | spin_lock(&pic_irqchip(kvm)->lock); |
1fe779f8 CO |
3428 | memcpy(&pic_irqchip(kvm)->pics[0], |
3429 | &chip->chip.pic, | |
3430 | sizeof(struct kvm_pic_state)); | |
f4f51050 | 3431 | spin_unlock(&pic_irqchip(kvm)->lock); |
1fe779f8 CO |
3432 | break; |
3433 | case KVM_IRQCHIP_PIC_SLAVE: | |
f4f51050 | 3434 | spin_lock(&pic_irqchip(kvm)->lock); |
1fe779f8 CO |
3435 | memcpy(&pic_irqchip(kvm)->pics[1], |
3436 | &chip->chip.pic, | |
3437 | sizeof(struct kvm_pic_state)); | |
f4f51050 | 3438 | spin_unlock(&pic_irqchip(kvm)->lock); |
1fe779f8 CO |
3439 | break; |
3440 | case KVM_IRQCHIP_IOAPIC: | |
eba0226b | 3441 | r = kvm_set_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
3442 | break; |
3443 | default: | |
3444 | r = -EINVAL; | |
3445 | break; | |
3446 | } | |
3447 | kvm_pic_update_irq(pic_irqchip(kvm)); | |
3448 | return r; | |
3449 | } | |
3450 | ||
e0f63cb9 SY |
3451 | static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps) |
3452 | { | |
3453 | int r = 0; | |
3454 | ||
894a9c55 | 3455 | mutex_lock(&kvm->arch.vpit->pit_state.lock); |
e0f63cb9 | 3456 | memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state)); |
894a9c55 | 3457 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); |
e0f63cb9 SY |
3458 | return r; |
3459 | } | |
3460 | ||
3461 | static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps) | |
3462 | { | |
3463 | int r = 0; | |
3464 | ||
894a9c55 | 3465 | mutex_lock(&kvm->arch.vpit->pit_state.lock); |
e0f63cb9 | 3466 | memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state)); |
e9f42757 BK |
3467 | kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0); |
3468 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); | |
3469 | return r; | |
3470 | } | |
3471 | ||
3472 | static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
3473 | { | |
3474 | int r = 0; | |
3475 | ||
3476 | mutex_lock(&kvm->arch.vpit->pit_state.lock); | |
3477 | memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels, | |
3478 | sizeof(ps->channels)); | |
3479 | ps->flags = kvm->arch.vpit->pit_state.flags; | |
3480 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); | |
97e69aa6 | 3481 | memset(&ps->reserved, 0, sizeof(ps->reserved)); |
e9f42757 BK |
3482 | return r; |
3483 | } | |
3484 | ||
3485 | static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
3486 | { | |
3487 | int r = 0, start = 0; | |
3488 | u32 prev_legacy, cur_legacy; | |
3489 | mutex_lock(&kvm->arch.vpit->pit_state.lock); | |
3490 | prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY; | |
3491 | cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY; | |
3492 | if (!prev_legacy && cur_legacy) | |
3493 | start = 1; | |
3494 | memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels, | |
3495 | sizeof(kvm->arch.vpit->pit_state.channels)); | |
3496 | kvm->arch.vpit->pit_state.flags = ps->flags; | |
3497 | kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start); | |
894a9c55 | 3498 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); |
e0f63cb9 SY |
3499 | return r; |
3500 | } | |
3501 | ||
52d939a0 MT |
3502 | static int kvm_vm_ioctl_reinject(struct kvm *kvm, |
3503 | struct kvm_reinject_control *control) | |
3504 | { | |
3505 | if (!kvm->arch.vpit) | |
3506 | return -ENXIO; | |
894a9c55 | 3507 | mutex_lock(&kvm->arch.vpit->pit_state.lock); |
26ef1924 | 3508 | kvm->arch.vpit->pit_state.reinject = control->pit_reinject; |
894a9c55 | 3509 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); |
52d939a0 MT |
3510 | return 0; |
3511 | } | |
3512 | ||
95d4c16c | 3513 | /** |
60c34612 TY |
3514 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot |
3515 | * @kvm: kvm instance | |
3516 | * @log: slot id and address to which we copy the log | |
95d4c16c | 3517 | * |
60c34612 TY |
3518 | * We need to keep it in mind that VCPU threads can write to the bitmap |
3519 | * concurrently. So, to avoid losing data, we keep the following order for | |
3520 | * each bit: | |
95d4c16c | 3521 | * |
60c34612 TY |
3522 | * 1. Take a snapshot of the bit and clear it if needed. |
3523 | * 2. Write protect the corresponding page. | |
3524 | * 3. Flush TLB's if needed. | |
3525 | * 4. Copy the snapshot to the userspace. | |
95d4c16c | 3526 | * |
60c34612 TY |
3527 | * Between 2 and 3, the guest may write to the page using the remaining TLB |
3528 | * entry. This is not a problem because the page will be reported dirty at | |
3529 | * step 4 using the snapshot taken before and step 3 ensures that successive | |
3530 | * writes will be logged for the next call. | |
5bb064dc | 3531 | */ |
60c34612 | 3532 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
5bb064dc | 3533 | { |
7850ac54 | 3534 | int r; |
5bb064dc | 3535 | struct kvm_memory_slot *memslot; |
60c34612 TY |
3536 | unsigned long n, i; |
3537 | unsigned long *dirty_bitmap; | |
3538 | unsigned long *dirty_bitmap_buffer; | |
3539 | bool is_dirty = false; | |
5bb064dc | 3540 | |
79fac95e | 3541 | mutex_lock(&kvm->slots_lock); |
5bb064dc | 3542 | |
b050b015 | 3543 | r = -EINVAL; |
bbacc0c1 | 3544 | if (log->slot >= KVM_USER_MEM_SLOTS) |
b050b015 MT |
3545 | goto out; |
3546 | ||
28a37544 | 3547 | memslot = id_to_memslot(kvm->memslots, log->slot); |
60c34612 TY |
3548 | |
3549 | dirty_bitmap = memslot->dirty_bitmap; | |
b050b015 | 3550 | r = -ENOENT; |
60c34612 | 3551 | if (!dirty_bitmap) |
b050b015 MT |
3552 | goto out; |
3553 | ||
87bf6e7d | 3554 | n = kvm_dirty_bitmap_bytes(memslot); |
b050b015 | 3555 | |
60c34612 TY |
3556 | dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long); |
3557 | memset(dirty_bitmap_buffer, 0, n); | |
b050b015 | 3558 | |
60c34612 | 3559 | spin_lock(&kvm->mmu_lock); |
b050b015 | 3560 | |
60c34612 TY |
3561 | for (i = 0; i < n / sizeof(long); i++) { |
3562 | unsigned long mask; | |
3563 | gfn_t offset; | |
cdfca7b3 | 3564 | |
60c34612 TY |
3565 | if (!dirty_bitmap[i]) |
3566 | continue; | |
b050b015 | 3567 | |
60c34612 | 3568 | is_dirty = true; |
914ebccd | 3569 | |
60c34612 TY |
3570 | mask = xchg(&dirty_bitmap[i], 0); |
3571 | dirty_bitmap_buffer[i] = mask; | |
edde99ce | 3572 | |
60c34612 TY |
3573 | offset = i * BITS_PER_LONG; |
3574 | kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask); | |
5bb064dc | 3575 | } |
60c34612 TY |
3576 | if (is_dirty) |
3577 | kvm_flush_remote_tlbs(kvm); | |
3578 | ||
3579 | spin_unlock(&kvm->mmu_lock); | |
3580 | ||
3581 | r = -EFAULT; | |
3582 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) | |
3583 | goto out; | |
b050b015 | 3584 | |
5bb064dc ZX |
3585 | r = 0; |
3586 | out: | |
79fac95e | 3587 | mutex_unlock(&kvm->slots_lock); |
5bb064dc ZX |
3588 | return r; |
3589 | } | |
3590 | ||
aa2fbe6d YZ |
3591 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event, |
3592 | bool line_status) | |
23d43cf9 CD |
3593 | { |
3594 | if (!irqchip_in_kernel(kvm)) | |
3595 | return -ENXIO; | |
3596 | ||
3597 | irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, | |
aa2fbe6d YZ |
3598 | irq_event->irq, irq_event->level, |
3599 | line_status); | |
23d43cf9 CD |
3600 | return 0; |
3601 | } | |
3602 | ||
1fe779f8 CO |
3603 | long kvm_arch_vm_ioctl(struct file *filp, |
3604 | unsigned int ioctl, unsigned long arg) | |
3605 | { | |
3606 | struct kvm *kvm = filp->private_data; | |
3607 | void __user *argp = (void __user *)arg; | |
367e1319 | 3608 | int r = -ENOTTY; |
f0d66275 DH |
3609 | /* |
3610 | * This union makes it completely explicit to gcc-3.x | |
3611 | * that these two variables' stack usage should be | |
3612 | * combined, not added together. | |
3613 | */ | |
3614 | union { | |
3615 | struct kvm_pit_state ps; | |
e9f42757 | 3616 | struct kvm_pit_state2 ps2; |
c5ff41ce | 3617 | struct kvm_pit_config pit_config; |
f0d66275 | 3618 | } u; |
1fe779f8 CO |
3619 | |
3620 | switch (ioctl) { | |
3621 | case KVM_SET_TSS_ADDR: | |
3622 | r = kvm_vm_ioctl_set_tss_addr(kvm, arg); | |
1fe779f8 | 3623 | break; |
b927a3ce SY |
3624 | case KVM_SET_IDENTITY_MAP_ADDR: { |
3625 | u64 ident_addr; | |
3626 | ||
3627 | r = -EFAULT; | |
3628 | if (copy_from_user(&ident_addr, argp, sizeof ident_addr)) | |
3629 | goto out; | |
3630 | r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr); | |
b927a3ce SY |
3631 | break; |
3632 | } | |
1fe779f8 CO |
3633 | case KVM_SET_NR_MMU_PAGES: |
3634 | r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg); | |
1fe779f8 CO |
3635 | break; |
3636 | case KVM_GET_NR_MMU_PAGES: | |
3637 | r = kvm_vm_ioctl_get_nr_mmu_pages(kvm); | |
3638 | break; | |
3ddea128 MT |
3639 | case KVM_CREATE_IRQCHIP: { |
3640 | struct kvm_pic *vpic; | |
3641 | ||
3642 | mutex_lock(&kvm->lock); | |
3643 | r = -EEXIST; | |
3644 | if (kvm->arch.vpic) | |
3645 | goto create_irqchip_unlock; | |
3e515705 AK |
3646 | r = -EINVAL; |
3647 | if (atomic_read(&kvm->online_vcpus)) | |
3648 | goto create_irqchip_unlock; | |
1fe779f8 | 3649 | r = -ENOMEM; |
3ddea128 MT |
3650 | vpic = kvm_create_pic(kvm); |
3651 | if (vpic) { | |
1fe779f8 CO |
3652 | r = kvm_ioapic_init(kvm); |
3653 | if (r) { | |
175504cd | 3654 | mutex_lock(&kvm->slots_lock); |
72bb2fcd | 3655 | kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, |
743eeb0b SL |
3656 | &vpic->dev_master); |
3657 | kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, | |
3658 | &vpic->dev_slave); | |
3659 | kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, | |
3660 | &vpic->dev_eclr); | |
175504cd | 3661 | mutex_unlock(&kvm->slots_lock); |
3ddea128 MT |
3662 | kfree(vpic); |
3663 | goto create_irqchip_unlock; | |
1fe779f8 CO |
3664 | } |
3665 | } else | |
3ddea128 MT |
3666 | goto create_irqchip_unlock; |
3667 | smp_wmb(); | |
3668 | kvm->arch.vpic = vpic; | |
3669 | smp_wmb(); | |
399ec807 AK |
3670 | r = kvm_setup_default_irq_routing(kvm); |
3671 | if (r) { | |
175504cd | 3672 | mutex_lock(&kvm->slots_lock); |
3ddea128 | 3673 | mutex_lock(&kvm->irq_lock); |
72bb2fcd WY |
3674 | kvm_ioapic_destroy(kvm); |
3675 | kvm_destroy_pic(kvm); | |
3ddea128 | 3676 | mutex_unlock(&kvm->irq_lock); |
175504cd | 3677 | mutex_unlock(&kvm->slots_lock); |
399ec807 | 3678 | } |
3ddea128 MT |
3679 | create_irqchip_unlock: |
3680 | mutex_unlock(&kvm->lock); | |
1fe779f8 | 3681 | break; |
3ddea128 | 3682 | } |
7837699f | 3683 | case KVM_CREATE_PIT: |
c5ff41ce JK |
3684 | u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY; |
3685 | goto create_pit; | |
3686 | case KVM_CREATE_PIT2: | |
3687 | r = -EFAULT; | |
3688 | if (copy_from_user(&u.pit_config, argp, | |
3689 | sizeof(struct kvm_pit_config))) | |
3690 | goto out; | |
3691 | create_pit: | |
79fac95e | 3692 | mutex_lock(&kvm->slots_lock); |
269e05e4 AK |
3693 | r = -EEXIST; |
3694 | if (kvm->arch.vpit) | |
3695 | goto create_pit_unlock; | |
7837699f | 3696 | r = -ENOMEM; |
c5ff41ce | 3697 | kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags); |
7837699f SY |
3698 | if (kvm->arch.vpit) |
3699 | r = 0; | |
269e05e4 | 3700 | create_pit_unlock: |
79fac95e | 3701 | mutex_unlock(&kvm->slots_lock); |
7837699f | 3702 | break; |
1fe779f8 CO |
3703 | case KVM_GET_IRQCHIP: { |
3704 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 3705 | struct kvm_irqchip *chip; |
1fe779f8 | 3706 | |
ff5c2c03 SL |
3707 | chip = memdup_user(argp, sizeof(*chip)); |
3708 | if (IS_ERR(chip)) { | |
3709 | r = PTR_ERR(chip); | |
1fe779f8 | 3710 | goto out; |
ff5c2c03 SL |
3711 | } |
3712 | ||
1fe779f8 CO |
3713 | r = -ENXIO; |
3714 | if (!irqchip_in_kernel(kvm)) | |
f0d66275 DH |
3715 | goto get_irqchip_out; |
3716 | r = kvm_vm_ioctl_get_irqchip(kvm, chip); | |
1fe779f8 | 3717 | if (r) |
f0d66275 | 3718 | goto get_irqchip_out; |
1fe779f8 | 3719 | r = -EFAULT; |
f0d66275 DH |
3720 | if (copy_to_user(argp, chip, sizeof *chip)) |
3721 | goto get_irqchip_out; | |
1fe779f8 | 3722 | r = 0; |
f0d66275 DH |
3723 | get_irqchip_out: |
3724 | kfree(chip); | |
1fe779f8 CO |
3725 | break; |
3726 | } | |
3727 | case KVM_SET_IRQCHIP: { | |
3728 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 3729 | struct kvm_irqchip *chip; |
1fe779f8 | 3730 | |
ff5c2c03 SL |
3731 | chip = memdup_user(argp, sizeof(*chip)); |
3732 | if (IS_ERR(chip)) { | |
3733 | r = PTR_ERR(chip); | |
1fe779f8 | 3734 | goto out; |
ff5c2c03 SL |
3735 | } |
3736 | ||
1fe779f8 CO |
3737 | r = -ENXIO; |
3738 | if (!irqchip_in_kernel(kvm)) | |
f0d66275 DH |
3739 | goto set_irqchip_out; |
3740 | r = kvm_vm_ioctl_set_irqchip(kvm, chip); | |
1fe779f8 | 3741 | if (r) |
f0d66275 | 3742 | goto set_irqchip_out; |
1fe779f8 | 3743 | r = 0; |
f0d66275 DH |
3744 | set_irqchip_out: |
3745 | kfree(chip); | |
1fe779f8 CO |
3746 | break; |
3747 | } | |
e0f63cb9 | 3748 | case KVM_GET_PIT: { |
e0f63cb9 | 3749 | r = -EFAULT; |
f0d66275 | 3750 | if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
3751 | goto out; |
3752 | r = -ENXIO; | |
3753 | if (!kvm->arch.vpit) | |
3754 | goto out; | |
f0d66275 | 3755 | r = kvm_vm_ioctl_get_pit(kvm, &u.ps); |
e0f63cb9 SY |
3756 | if (r) |
3757 | goto out; | |
3758 | r = -EFAULT; | |
f0d66275 | 3759 | if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
3760 | goto out; |
3761 | r = 0; | |
3762 | break; | |
3763 | } | |
3764 | case KVM_SET_PIT: { | |
e0f63cb9 | 3765 | r = -EFAULT; |
f0d66275 | 3766 | if (copy_from_user(&u.ps, argp, sizeof u.ps)) |
e0f63cb9 SY |
3767 | goto out; |
3768 | r = -ENXIO; | |
3769 | if (!kvm->arch.vpit) | |
3770 | goto out; | |
f0d66275 | 3771 | r = kvm_vm_ioctl_set_pit(kvm, &u.ps); |
e0f63cb9 SY |
3772 | break; |
3773 | } | |
e9f42757 BK |
3774 | case KVM_GET_PIT2: { |
3775 | r = -ENXIO; | |
3776 | if (!kvm->arch.vpit) | |
3777 | goto out; | |
3778 | r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2); | |
3779 | if (r) | |
3780 | goto out; | |
3781 | r = -EFAULT; | |
3782 | if (copy_to_user(argp, &u.ps2, sizeof(u.ps2))) | |
3783 | goto out; | |
3784 | r = 0; | |
3785 | break; | |
3786 | } | |
3787 | case KVM_SET_PIT2: { | |
3788 | r = -EFAULT; | |
3789 | if (copy_from_user(&u.ps2, argp, sizeof(u.ps2))) | |
3790 | goto out; | |
3791 | r = -ENXIO; | |
3792 | if (!kvm->arch.vpit) | |
3793 | goto out; | |
3794 | r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2); | |
e9f42757 BK |
3795 | break; |
3796 | } | |
52d939a0 MT |
3797 | case KVM_REINJECT_CONTROL: { |
3798 | struct kvm_reinject_control control; | |
3799 | r = -EFAULT; | |
3800 | if (copy_from_user(&control, argp, sizeof(control))) | |
3801 | goto out; | |
3802 | r = kvm_vm_ioctl_reinject(kvm, &control); | |
52d939a0 MT |
3803 | break; |
3804 | } | |
ffde22ac ES |
3805 | case KVM_XEN_HVM_CONFIG: { |
3806 | r = -EFAULT; | |
3807 | if (copy_from_user(&kvm->arch.xen_hvm_config, argp, | |
3808 | sizeof(struct kvm_xen_hvm_config))) | |
3809 | goto out; | |
3810 | r = -EINVAL; | |
3811 | if (kvm->arch.xen_hvm_config.flags) | |
3812 | goto out; | |
3813 | r = 0; | |
3814 | break; | |
3815 | } | |
afbcf7ab | 3816 | case KVM_SET_CLOCK: { |
afbcf7ab GC |
3817 | struct kvm_clock_data user_ns; |
3818 | u64 now_ns; | |
3819 | s64 delta; | |
3820 | ||
3821 | r = -EFAULT; | |
3822 | if (copy_from_user(&user_ns, argp, sizeof(user_ns))) | |
3823 | goto out; | |
3824 | ||
3825 | r = -EINVAL; | |
3826 | if (user_ns.flags) | |
3827 | goto out; | |
3828 | ||
3829 | r = 0; | |
395c6b0a | 3830 | local_irq_disable(); |
759379dd | 3831 | now_ns = get_kernel_ns(); |
afbcf7ab | 3832 | delta = user_ns.clock - now_ns; |
395c6b0a | 3833 | local_irq_enable(); |
afbcf7ab | 3834 | kvm->arch.kvmclock_offset = delta; |
2e762ff7 | 3835 | kvm_gen_update_masterclock(kvm); |
afbcf7ab GC |
3836 | break; |
3837 | } | |
3838 | case KVM_GET_CLOCK: { | |
afbcf7ab GC |
3839 | struct kvm_clock_data user_ns; |
3840 | u64 now_ns; | |
3841 | ||
395c6b0a | 3842 | local_irq_disable(); |
759379dd | 3843 | now_ns = get_kernel_ns(); |
afbcf7ab | 3844 | user_ns.clock = kvm->arch.kvmclock_offset + now_ns; |
395c6b0a | 3845 | local_irq_enable(); |
afbcf7ab | 3846 | user_ns.flags = 0; |
97e69aa6 | 3847 | memset(&user_ns.pad, 0, sizeof(user_ns.pad)); |
afbcf7ab GC |
3848 | |
3849 | r = -EFAULT; | |
3850 | if (copy_to_user(argp, &user_ns, sizeof(user_ns))) | |
3851 | goto out; | |
3852 | r = 0; | |
3853 | break; | |
3854 | } | |
3855 | ||
1fe779f8 CO |
3856 | default: |
3857 | ; | |
3858 | } | |
3859 | out: | |
3860 | return r; | |
3861 | } | |
3862 | ||
a16b043c | 3863 | static void kvm_init_msr_list(void) |
043405e1 CO |
3864 | { |
3865 | u32 dummy[2]; | |
3866 | unsigned i, j; | |
3867 | ||
e3267cbb GC |
3868 | /* skip the first msrs in the list. KVM-specific */ |
3869 | for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) { | |
043405e1 CO |
3870 | if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0) |
3871 | continue; | |
3872 | if (j < i) | |
3873 | msrs_to_save[j] = msrs_to_save[i]; | |
3874 | j++; | |
3875 | } | |
3876 | num_msrs_to_save = j; | |
3877 | } | |
3878 | ||
bda9020e MT |
3879 | static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len, |
3880 | const void *v) | |
bbd9b64e | 3881 | { |
70252a10 AK |
3882 | int handled = 0; |
3883 | int n; | |
3884 | ||
3885 | do { | |
3886 | n = min(len, 8); | |
3887 | if (!(vcpu->arch.apic && | |
3888 | !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v)) | |
3889 | && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v)) | |
3890 | break; | |
3891 | handled += n; | |
3892 | addr += n; | |
3893 | len -= n; | |
3894 | v += n; | |
3895 | } while (len); | |
bbd9b64e | 3896 | |
70252a10 | 3897 | return handled; |
bbd9b64e CO |
3898 | } |
3899 | ||
bda9020e | 3900 | static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v) |
bbd9b64e | 3901 | { |
70252a10 AK |
3902 | int handled = 0; |
3903 | int n; | |
3904 | ||
3905 | do { | |
3906 | n = min(len, 8); | |
3907 | if (!(vcpu->arch.apic && | |
3908 | !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v)) | |
3909 | && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v)) | |
3910 | break; | |
3911 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v); | |
3912 | handled += n; | |
3913 | addr += n; | |
3914 | len -= n; | |
3915 | v += n; | |
3916 | } while (len); | |
bbd9b64e | 3917 | |
70252a10 | 3918 | return handled; |
bbd9b64e CO |
3919 | } |
3920 | ||
2dafc6c2 GN |
3921 | static void kvm_set_segment(struct kvm_vcpu *vcpu, |
3922 | struct kvm_segment *var, int seg) | |
3923 | { | |
3924 | kvm_x86_ops->set_segment(vcpu, var, seg); | |
3925 | } | |
3926 | ||
3927 | void kvm_get_segment(struct kvm_vcpu *vcpu, | |
3928 | struct kvm_segment *var, int seg) | |
3929 | { | |
3930 | kvm_x86_ops->get_segment(vcpu, var, seg); | |
3931 | } | |
3932 | ||
e459e322 | 3933 | gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access) |
02f59dc9 JR |
3934 | { |
3935 | gpa_t t_gpa; | |
ab9ae313 | 3936 | struct x86_exception exception; |
02f59dc9 JR |
3937 | |
3938 | BUG_ON(!mmu_is_nested(vcpu)); | |
3939 | ||
3940 | /* NPT walks are always user-walks */ | |
3941 | access |= PFERR_USER_MASK; | |
ab9ae313 | 3942 | t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception); |
02f59dc9 JR |
3943 | |
3944 | return t_gpa; | |
3945 | } | |
3946 | ||
ab9ae313 AK |
3947 | gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, |
3948 | struct x86_exception *exception) | |
1871c602 GN |
3949 | { |
3950 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
ab9ae313 | 3951 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
3952 | } |
3953 | ||
ab9ae313 AK |
3954 | gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, |
3955 | struct x86_exception *exception) | |
1871c602 GN |
3956 | { |
3957 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
3958 | access |= PFERR_FETCH_MASK; | |
ab9ae313 | 3959 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
3960 | } |
3961 | ||
ab9ae313 AK |
3962 | gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, |
3963 | struct x86_exception *exception) | |
1871c602 GN |
3964 | { |
3965 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
3966 | access |= PFERR_WRITE_MASK; | |
ab9ae313 | 3967 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
3968 | } |
3969 | ||
3970 | /* uses this to access any guest's mapped memory without checking CPL */ | |
ab9ae313 AK |
3971 | gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, |
3972 | struct x86_exception *exception) | |
1871c602 | 3973 | { |
ab9ae313 | 3974 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception); |
1871c602 GN |
3975 | } |
3976 | ||
3977 | static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes, | |
3978 | struct kvm_vcpu *vcpu, u32 access, | |
bcc55cba | 3979 | struct x86_exception *exception) |
bbd9b64e CO |
3980 | { |
3981 | void *data = val; | |
10589a46 | 3982 | int r = X86EMUL_CONTINUE; |
bbd9b64e CO |
3983 | |
3984 | while (bytes) { | |
14dfe855 | 3985 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access, |
ab9ae313 | 3986 | exception); |
bbd9b64e | 3987 | unsigned offset = addr & (PAGE_SIZE-1); |
77c2002e | 3988 | unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset); |
bbd9b64e CO |
3989 | int ret; |
3990 | ||
bcc55cba | 3991 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 3992 | return X86EMUL_PROPAGATE_FAULT; |
77c2002e | 3993 | ret = kvm_read_guest(vcpu->kvm, gpa, data, toread); |
10589a46 | 3994 | if (ret < 0) { |
c3cd7ffa | 3995 | r = X86EMUL_IO_NEEDED; |
10589a46 MT |
3996 | goto out; |
3997 | } | |
bbd9b64e | 3998 | |
77c2002e IE |
3999 | bytes -= toread; |
4000 | data += toread; | |
4001 | addr += toread; | |
bbd9b64e | 4002 | } |
10589a46 | 4003 | out: |
10589a46 | 4004 | return r; |
bbd9b64e | 4005 | } |
77c2002e | 4006 | |
1871c602 | 4007 | /* used for instruction fetching */ |
0f65dd70 AK |
4008 | static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt, |
4009 | gva_t addr, void *val, unsigned int bytes, | |
bcc55cba | 4010 | struct x86_exception *exception) |
1871c602 | 4011 | { |
0f65dd70 | 4012 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
1871c602 | 4013 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
0f65dd70 | 4014 | |
1871c602 | 4015 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, |
bcc55cba AK |
4016 | access | PFERR_FETCH_MASK, |
4017 | exception); | |
1871c602 GN |
4018 | } |
4019 | ||
064aea77 | 4020 | int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt, |
0f65dd70 | 4021 | gva_t addr, void *val, unsigned int bytes, |
bcc55cba | 4022 | struct x86_exception *exception) |
1871c602 | 4023 | { |
0f65dd70 | 4024 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
1871c602 | 4025 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
0f65dd70 | 4026 | |
1871c602 | 4027 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, |
bcc55cba | 4028 | exception); |
1871c602 | 4029 | } |
064aea77 | 4030 | EXPORT_SYMBOL_GPL(kvm_read_guest_virt); |
1871c602 | 4031 | |
0f65dd70 AK |
4032 | static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt, |
4033 | gva_t addr, void *val, unsigned int bytes, | |
bcc55cba | 4034 | struct x86_exception *exception) |
1871c602 | 4035 | { |
0f65dd70 | 4036 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
bcc55cba | 4037 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception); |
1871c602 GN |
4038 | } |
4039 | ||
6a4d7550 | 4040 | int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt, |
0f65dd70 | 4041 | gva_t addr, void *val, |
2dafc6c2 | 4042 | unsigned int bytes, |
bcc55cba | 4043 | struct x86_exception *exception) |
77c2002e | 4044 | { |
0f65dd70 | 4045 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
77c2002e IE |
4046 | void *data = val; |
4047 | int r = X86EMUL_CONTINUE; | |
4048 | ||
4049 | while (bytes) { | |
14dfe855 JR |
4050 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, |
4051 | PFERR_WRITE_MASK, | |
ab9ae313 | 4052 | exception); |
77c2002e IE |
4053 | unsigned offset = addr & (PAGE_SIZE-1); |
4054 | unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset); | |
4055 | int ret; | |
4056 | ||
bcc55cba | 4057 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 4058 | return X86EMUL_PROPAGATE_FAULT; |
77c2002e IE |
4059 | ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite); |
4060 | if (ret < 0) { | |
c3cd7ffa | 4061 | r = X86EMUL_IO_NEEDED; |
77c2002e IE |
4062 | goto out; |
4063 | } | |
4064 | ||
4065 | bytes -= towrite; | |
4066 | data += towrite; | |
4067 | addr += towrite; | |
4068 | } | |
4069 | out: | |
4070 | return r; | |
4071 | } | |
6a4d7550 | 4072 | EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system); |
77c2002e | 4073 | |
af7cc7d1 XG |
4074 | static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva, |
4075 | gpa_t *gpa, struct x86_exception *exception, | |
4076 | bool write) | |
4077 | { | |
97d64b78 AK |
4078 | u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0) |
4079 | | (write ? PFERR_WRITE_MASK : 0); | |
af7cc7d1 | 4080 | |
97d64b78 AK |
4081 | if (vcpu_match_mmio_gva(vcpu, gva) |
4082 | && !permission_fault(vcpu->arch.walk_mmu, vcpu->arch.access, access)) { | |
bebb106a XG |
4083 | *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT | |
4084 | (gva & (PAGE_SIZE - 1)); | |
4f022648 | 4085 | trace_vcpu_match_mmio(gva, *gpa, write, false); |
bebb106a XG |
4086 | return 1; |
4087 | } | |
4088 | ||
af7cc7d1 XG |
4089 | *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
4090 | ||
4091 | if (*gpa == UNMAPPED_GVA) | |
4092 | return -1; | |
4093 | ||
4094 | /* For APIC access vmexit */ | |
4095 | if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
4096 | return 1; | |
4097 | ||
4f022648 XG |
4098 | if (vcpu_match_mmio_gpa(vcpu, *gpa)) { |
4099 | trace_vcpu_match_mmio(gva, *gpa, write, true); | |
bebb106a | 4100 | return 1; |
4f022648 | 4101 | } |
bebb106a | 4102 | |
af7cc7d1 XG |
4103 | return 0; |
4104 | } | |
4105 | ||
3200f405 | 4106 | int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa, |
bcc55cba | 4107 | const void *val, int bytes) |
bbd9b64e CO |
4108 | { |
4109 | int ret; | |
4110 | ||
4111 | ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes); | |
9f811285 | 4112 | if (ret < 0) |
bbd9b64e | 4113 | return 0; |
f57f2ef5 | 4114 | kvm_mmu_pte_write(vcpu, gpa, val, bytes); |
bbd9b64e CO |
4115 | return 1; |
4116 | } | |
4117 | ||
77d197b2 XG |
4118 | struct read_write_emulator_ops { |
4119 | int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val, | |
4120 | int bytes); | |
4121 | int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4122 | void *val, int bytes); | |
4123 | int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4124 | int bytes, void *val); | |
4125 | int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4126 | void *val, int bytes); | |
4127 | bool write; | |
4128 | }; | |
4129 | ||
4130 | static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes) | |
4131 | { | |
4132 | if (vcpu->mmio_read_completed) { | |
77d197b2 | 4133 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, |
f78146b0 | 4134 | vcpu->mmio_fragments[0].gpa, *(u64 *)val); |
77d197b2 XG |
4135 | vcpu->mmio_read_completed = 0; |
4136 | return 1; | |
4137 | } | |
4138 | ||
4139 | return 0; | |
4140 | } | |
4141 | ||
4142 | static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4143 | void *val, int bytes) | |
4144 | { | |
4145 | return !kvm_read_guest(vcpu->kvm, gpa, val, bytes); | |
4146 | } | |
4147 | ||
4148 | static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4149 | void *val, int bytes) | |
4150 | { | |
4151 | return emulator_write_phys(vcpu, gpa, val, bytes); | |
4152 | } | |
4153 | ||
4154 | static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val) | |
4155 | { | |
4156 | trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val); | |
4157 | return vcpu_mmio_write(vcpu, gpa, bytes, val); | |
4158 | } | |
4159 | ||
4160 | static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4161 | void *val, int bytes) | |
4162 | { | |
4163 | trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0); | |
4164 | return X86EMUL_IO_NEEDED; | |
4165 | } | |
4166 | ||
4167 | static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4168 | void *val, int bytes) | |
4169 | { | |
f78146b0 AK |
4170 | struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0]; |
4171 | ||
87da7e66 | 4172 | memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len)); |
77d197b2 XG |
4173 | return X86EMUL_CONTINUE; |
4174 | } | |
4175 | ||
0fbe9b0b | 4176 | static const struct read_write_emulator_ops read_emultor = { |
77d197b2 XG |
4177 | .read_write_prepare = read_prepare, |
4178 | .read_write_emulate = read_emulate, | |
4179 | .read_write_mmio = vcpu_mmio_read, | |
4180 | .read_write_exit_mmio = read_exit_mmio, | |
4181 | }; | |
4182 | ||
0fbe9b0b | 4183 | static const struct read_write_emulator_ops write_emultor = { |
77d197b2 XG |
4184 | .read_write_emulate = write_emulate, |
4185 | .read_write_mmio = write_mmio, | |
4186 | .read_write_exit_mmio = write_exit_mmio, | |
4187 | .write = true, | |
4188 | }; | |
4189 | ||
22388a3c XG |
4190 | static int emulator_read_write_onepage(unsigned long addr, void *val, |
4191 | unsigned int bytes, | |
4192 | struct x86_exception *exception, | |
4193 | struct kvm_vcpu *vcpu, | |
0fbe9b0b | 4194 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 4195 | { |
af7cc7d1 XG |
4196 | gpa_t gpa; |
4197 | int handled, ret; | |
22388a3c | 4198 | bool write = ops->write; |
f78146b0 | 4199 | struct kvm_mmio_fragment *frag; |
10589a46 | 4200 | |
22388a3c | 4201 | ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write); |
bbd9b64e | 4202 | |
af7cc7d1 | 4203 | if (ret < 0) |
bbd9b64e | 4204 | return X86EMUL_PROPAGATE_FAULT; |
bbd9b64e CO |
4205 | |
4206 | /* For APIC access vmexit */ | |
af7cc7d1 | 4207 | if (ret) |
bbd9b64e CO |
4208 | goto mmio; |
4209 | ||
22388a3c | 4210 | if (ops->read_write_emulate(vcpu, gpa, val, bytes)) |
bbd9b64e CO |
4211 | return X86EMUL_CONTINUE; |
4212 | ||
4213 | mmio: | |
4214 | /* | |
4215 | * Is this MMIO handled locally? | |
4216 | */ | |
22388a3c | 4217 | handled = ops->read_write_mmio(vcpu, gpa, bytes, val); |
70252a10 | 4218 | if (handled == bytes) |
bbd9b64e | 4219 | return X86EMUL_CONTINUE; |
bbd9b64e | 4220 | |
70252a10 AK |
4221 | gpa += handled; |
4222 | bytes -= handled; | |
4223 | val += handled; | |
4224 | ||
87da7e66 XG |
4225 | WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS); |
4226 | frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++]; | |
4227 | frag->gpa = gpa; | |
4228 | frag->data = val; | |
4229 | frag->len = bytes; | |
f78146b0 | 4230 | return X86EMUL_CONTINUE; |
bbd9b64e CO |
4231 | } |
4232 | ||
22388a3c XG |
4233 | int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr, |
4234 | void *val, unsigned int bytes, | |
4235 | struct x86_exception *exception, | |
0fbe9b0b | 4236 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 4237 | { |
0f65dd70 | 4238 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
f78146b0 AK |
4239 | gpa_t gpa; |
4240 | int rc; | |
4241 | ||
4242 | if (ops->read_write_prepare && | |
4243 | ops->read_write_prepare(vcpu, val, bytes)) | |
4244 | return X86EMUL_CONTINUE; | |
4245 | ||
4246 | vcpu->mmio_nr_fragments = 0; | |
0f65dd70 | 4247 | |
bbd9b64e CO |
4248 | /* Crossing a page boundary? */ |
4249 | if (((addr + bytes - 1) ^ addr) & PAGE_MASK) { | |
f78146b0 | 4250 | int now; |
bbd9b64e CO |
4251 | |
4252 | now = -addr & ~PAGE_MASK; | |
22388a3c XG |
4253 | rc = emulator_read_write_onepage(addr, val, now, exception, |
4254 | vcpu, ops); | |
4255 | ||
bbd9b64e CO |
4256 | if (rc != X86EMUL_CONTINUE) |
4257 | return rc; | |
4258 | addr += now; | |
4259 | val += now; | |
4260 | bytes -= now; | |
4261 | } | |
22388a3c | 4262 | |
f78146b0 AK |
4263 | rc = emulator_read_write_onepage(addr, val, bytes, exception, |
4264 | vcpu, ops); | |
4265 | if (rc != X86EMUL_CONTINUE) | |
4266 | return rc; | |
4267 | ||
4268 | if (!vcpu->mmio_nr_fragments) | |
4269 | return rc; | |
4270 | ||
4271 | gpa = vcpu->mmio_fragments[0].gpa; | |
4272 | ||
4273 | vcpu->mmio_needed = 1; | |
4274 | vcpu->mmio_cur_fragment = 0; | |
4275 | ||
87da7e66 | 4276 | vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len); |
f78146b0 AK |
4277 | vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write; |
4278 | vcpu->run->exit_reason = KVM_EXIT_MMIO; | |
4279 | vcpu->run->mmio.phys_addr = gpa; | |
4280 | ||
4281 | return ops->read_write_exit_mmio(vcpu, gpa, val, bytes); | |
22388a3c XG |
4282 | } |
4283 | ||
4284 | static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt, | |
4285 | unsigned long addr, | |
4286 | void *val, | |
4287 | unsigned int bytes, | |
4288 | struct x86_exception *exception) | |
4289 | { | |
4290 | return emulator_read_write(ctxt, addr, val, bytes, | |
4291 | exception, &read_emultor); | |
4292 | } | |
4293 | ||
4294 | int emulator_write_emulated(struct x86_emulate_ctxt *ctxt, | |
4295 | unsigned long addr, | |
4296 | const void *val, | |
4297 | unsigned int bytes, | |
4298 | struct x86_exception *exception) | |
4299 | { | |
4300 | return emulator_read_write(ctxt, addr, (void *)val, bytes, | |
4301 | exception, &write_emultor); | |
bbd9b64e | 4302 | } |
bbd9b64e | 4303 | |
daea3e73 AK |
4304 | #define CMPXCHG_TYPE(t, ptr, old, new) \ |
4305 | (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old)) | |
4306 | ||
4307 | #ifdef CONFIG_X86_64 | |
4308 | # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new) | |
4309 | #else | |
4310 | # define CMPXCHG64(ptr, old, new) \ | |
9749a6c0 | 4311 | (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old)) |
daea3e73 AK |
4312 | #endif |
4313 | ||
0f65dd70 AK |
4314 | static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt, |
4315 | unsigned long addr, | |
bbd9b64e CO |
4316 | const void *old, |
4317 | const void *new, | |
4318 | unsigned int bytes, | |
0f65dd70 | 4319 | struct x86_exception *exception) |
bbd9b64e | 4320 | { |
0f65dd70 | 4321 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
daea3e73 AK |
4322 | gpa_t gpa; |
4323 | struct page *page; | |
4324 | char *kaddr; | |
4325 | bool exchanged; | |
2bacc55c | 4326 | |
daea3e73 AK |
4327 | /* guests cmpxchg8b have to be emulated atomically */ |
4328 | if (bytes > 8 || (bytes & (bytes - 1))) | |
4329 | goto emul_write; | |
10589a46 | 4330 | |
daea3e73 | 4331 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL); |
2bacc55c | 4332 | |
daea3e73 AK |
4333 | if (gpa == UNMAPPED_GVA || |
4334 | (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
4335 | goto emul_write; | |
2bacc55c | 4336 | |
daea3e73 AK |
4337 | if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK)) |
4338 | goto emul_write; | |
72dc67a6 | 4339 | |
daea3e73 | 4340 | page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT); |
32cad84f | 4341 | if (is_error_page(page)) |
c19b8bd6 | 4342 | goto emul_write; |
72dc67a6 | 4343 | |
8fd75e12 | 4344 | kaddr = kmap_atomic(page); |
daea3e73 AK |
4345 | kaddr += offset_in_page(gpa); |
4346 | switch (bytes) { | |
4347 | case 1: | |
4348 | exchanged = CMPXCHG_TYPE(u8, kaddr, old, new); | |
4349 | break; | |
4350 | case 2: | |
4351 | exchanged = CMPXCHG_TYPE(u16, kaddr, old, new); | |
4352 | break; | |
4353 | case 4: | |
4354 | exchanged = CMPXCHG_TYPE(u32, kaddr, old, new); | |
4355 | break; | |
4356 | case 8: | |
4357 | exchanged = CMPXCHG64(kaddr, old, new); | |
4358 | break; | |
4359 | default: | |
4360 | BUG(); | |
2bacc55c | 4361 | } |
8fd75e12 | 4362 | kunmap_atomic(kaddr); |
daea3e73 AK |
4363 | kvm_release_page_dirty(page); |
4364 | ||
4365 | if (!exchanged) | |
4366 | return X86EMUL_CMPXCHG_FAILED; | |
4367 | ||
f57f2ef5 | 4368 | kvm_mmu_pte_write(vcpu, gpa, new, bytes); |
8f6abd06 GN |
4369 | |
4370 | return X86EMUL_CONTINUE; | |
4a5f48f6 | 4371 | |
3200f405 | 4372 | emul_write: |
daea3e73 | 4373 | printk_once(KERN_WARNING "kvm: emulating exchange as write\n"); |
2bacc55c | 4374 | |
0f65dd70 | 4375 | return emulator_write_emulated(ctxt, addr, new, bytes, exception); |
bbd9b64e CO |
4376 | } |
4377 | ||
cf8f70bf GN |
4378 | static int kernel_pio(struct kvm_vcpu *vcpu, void *pd) |
4379 | { | |
4380 | /* TODO: String I/O for in kernel device */ | |
4381 | int r; | |
4382 | ||
4383 | if (vcpu->arch.pio.in) | |
4384 | r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port, | |
4385 | vcpu->arch.pio.size, pd); | |
4386 | else | |
4387 | r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS, | |
4388 | vcpu->arch.pio.port, vcpu->arch.pio.size, | |
4389 | pd); | |
4390 | return r; | |
4391 | } | |
4392 | ||
6f6fbe98 XG |
4393 | static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size, |
4394 | unsigned short port, void *val, | |
4395 | unsigned int count, bool in) | |
cf8f70bf | 4396 | { |
6f6fbe98 | 4397 | trace_kvm_pio(!in, port, size, count); |
cf8f70bf GN |
4398 | |
4399 | vcpu->arch.pio.port = port; | |
6f6fbe98 | 4400 | vcpu->arch.pio.in = in; |
7972995b | 4401 | vcpu->arch.pio.count = count; |
cf8f70bf GN |
4402 | vcpu->arch.pio.size = size; |
4403 | ||
4404 | if (!kernel_pio(vcpu, vcpu->arch.pio_data)) { | |
7972995b | 4405 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
4406 | return 1; |
4407 | } | |
4408 | ||
4409 | vcpu->run->exit_reason = KVM_EXIT_IO; | |
6f6fbe98 | 4410 | vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT; |
cf8f70bf GN |
4411 | vcpu->run->io.size = size; |
4412 | vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE; | |
4413 | vcpu->run->io.count = count; | |
4414 | vcpu->run->io.port = port; | |
4415 | ||
4416 | return 0; | |
4417 | } | |
4418 | ||
6f6fbe98 XG |
4419 | static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt, |
4420 | int size, unsigned short port, void *val, | |
4421 | unsigned int count) | |
cf8f70bf | 4422 | { |
ca1d4a9e | 4423 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
6f6fbe98 | 4424 | int ret; |
ca1d4a9e | 4425 | |
6f6fbe98 XG |
4426 | if (vcpu->arch.pio.count) |
4427 | goto data_avail; | |
cf8f70bf | 4428 | |
6f6fbe98 XG |
4429 | ret = emulator_pio_in_out(vcpu, size, port, val, count, true); |
4430 | if (ret) { | |
4431 | data_avail: | |
4432 | memcpy(val, vcpu->arch.pio_data, size * count); | |
7972995b | 4433 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
4434 | return 1; |
4435 | } | |
4436 | ||
cf8f70bf GN |
4437 | return 0; |
4438 | } | |
4439 | ||
6f6fbe98 XG |
4440 | static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt, |
4441 | int size, unsigned short port, | |
4442 | const void *val, unsigned int count) | |
4443 | { | |
4444 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
4445 | ||
4446 | memcpy(vcpu->arch.pio_data, val, size * count); | |
4447 | return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false); | |
4448 | } | |
4449 | ||
bbd9b64e CO |
4450 | static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg) |
4451 | { | |
4452 | return kvm_x86_ops->get_segment_base(vcpu, seg); | |
4453 | } | |
4454 | ||
3cb16fe7 | 4455 | static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address) |
bbd9b64e | 4456 | { |
3cb16fe7 | 4457 | kvm_mmu_invlpg(emul_to_vcpu(ctxt), address); |
bbd9b64e CO |
4458 | } |
4459 | ||
f5f48ee1 SY |
4460 | int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu) |
4461 | { | |
4462 | if (!need_emulate_wbinvd(vcpu)) | |
4463 | return X86EMUL_CONTINUE; | |
4464 | ||
4465 | if (kvm_x86_ops->has_wbinvd_exit()) { | |
2eec7343 JK |
4466 | int cpu = get_cpu(); |
4467 | ||
4468 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); | |
f5f48ee1 SY |
4469 | smp_call_function_many(vcpu->arch.wbinvd_dirty_mask, |
4470 | wbinvd_ipi, NULL, 1); | |
2eec7343 | 4471 | put_cpu(); |
f5f48ee1 | 4472 | cpumask_clear(vcpu->arch.wbinvd_dirty_mask); |
2eec7343 JK |
4473 | } else |
4474 | wbinvd(); | |
f5f48ee1 SY |
4475 | return X86EMUL_CONTINUE; |
4476 | } | |
4477 | EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd); | |
4478 | ||
bcaf5cc5 AK |
4479 | static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt) |
4480 | { | |
4481 | kvm_emulate_wbinvd(emul_to_vcpu(ctxt)); | |
4482 | } | |
4483 | ||
717746e3 | 4484 | int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest) |
bbd9b64e | 4485 | { |
717746e3 | 4486 | return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest); |
bbd9b64e CO |
4487 | } |
4488 | ||
717746e3 | 4489 | int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value) |
bbd9b64e | 4490 | { |
338dbc97 | 4491 | |
717746e3 | 4492 | return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value); |
bbd9b64e CO |
4493 | } |
4494 | ||
52a46617 | 4495 | static u64 mk_cr_64(u64 curr_cr, u32 new_val) |
5fdbf976 | 4496 | { |
52a46617 | 4497 | return (curr_cr & ~((1ULL << 32) - 1)) | new_val; |
5fdbf976 MT |
4498 | } |
4499 | ||
717746e3 | 4500 | static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr) |
bbd9b64e | 4501 | { |
717746e3 | 4502 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
52a46617 GN |
4503 | unsigned long value; |
4504 | ||
4505 | switch (cr) { | |
4506 | case 0: | |
4507 | value = kvm_read_cr0(vcpu); | |
4508 | break; | |
4509 | case 2: | |
4510 | value = vcpu->arch.cr2; | |
4511 | break; | |
4512 | case 3: | |
9f8fe504 | 4513 | value = kvm_read_cr3(vcpu); |
52a46617 GN |
4514 | break; |
4515 | case 4: | |
4516 | value = kvm_read_cr4(vcpu); | |
4517 | break; | |
4518 | case 8: | |
4519 | value = kvm_get_cr8(vcpu); | |
4520 | break; | |
4521 | default: | |
a737f256 | 4522 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
52a46617 GN |
4523 | return 0; |
4524 | } | |
4525 | ||
4526 | return value; | |
4527 | } | |
4528 | ||
717746e3 | 4529 | static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val) |
52a46617 | 4530 | { |
717746e3 | 4531 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
0f12244f GN |
4532 | int res = 0; |
4533 | ||
52a46617 GN |
4534 | switch (cr) { |
4535 | case 0: | |
49a9b07e | 4536 | res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val)); |
52a46617 GN |
4537 | break; |
4538 | case 2: | |
4539 | vcpu->arch.cr2 = val; | |
4540 | break; | |
4541 | case 3: | |
2390218b | 4542 | res = kvm_set_cr3(vcpu, val); |
52a46617 GN |
4543 | break; |
4544 | case 4: | |
a83b29c6 | 4545 | res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val)); |
52a46617 GN |
4546 | break; |
4547 | case 8: | |
eea1cff9 | 4548 | res = kvm_set_cr8(vcpu, val); |
52a46617 GN |
4549 | break; |
4550 | default: | |
a737f256 | 4551 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
0f12244f | 4552 | res = -1; |
52a46617 | 4553 | } |
0f12244f GN |
4554 | |
4555 | return res; | |
52a46617 GN |
4556 | } |
4557 | ||
4cee4798 KW |
4558 | static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val) |
4559 | { | |
4560 | kvm_set_rflags(emul_to_vcpu(ctxt), val); | |
4561 | } | |
4562 | ||
717746e3 | 4563 | static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt) |
9c537244 | 4564 | { |
717746e3 | 4565 | return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt)); |
9c537244 GN |
4566 | } |
4567 | ||
4bff1e86 | 4568 | static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
2dafc6c2 | 4569 | { |
4bff1e86 | 4570 | kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt); |
2dafc6c2 GN |
4571 | } |
4572 | ||
4bff1e86 | 4573 | static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
160ce1f1 | 4574 | { |
4bff1e86 | 4575 | kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt); |
160ce1f1 MG |
4576 | } |
4577 | ||
1ac9d0cf AK |
4578 | static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
4579 | { | |
4580 | kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt); | |
4581 | } | |
4582 | ||
4583 | static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) | |
4584 | { | |
4585 | kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt); | |
4586 | } | |
4587 | ||
4bff1e86 AK |
4588 | static unsigned long emulator_get_cached_segment_base( |
4589 | struct x86_emulate_ctxt *ctxt, int seg) | |
5951c442 | 4590 | { |
4bff1e86 | 4591 | return get_segment_base(emul_to_vcpu(ctxt), seg); |
5951c442 GN |
4592 | } |
4593 | ||
1aa36616 AK |
4594 | static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector, |
4595 | struct desc_struct *desc, u32 *base3, | |
4596 | int seg) | |
2dafc6c2 GN |
4597 | { |
4598 | struct kvm_segment var; | |
4599 | ||
4bff1e86 | 4600 | kvm_get_segment(emul_to_vcpu(ctxt), &var, seg); |
1aa36616 | 4601 | *selector = var.selector; |
2dafc6c2 | 4602 | |
378a8b09 GN |
4603 | if (var.unusable) { |
4604 | memset(desc, 0, sizeof(*desc)); | |
2dafc6c2 | 4605 | return false; |
378a8b09 | 4606 | } |
2dafc6c2 GN |
4607 | |
4608 | if (var.g) | |
4609 | var.limit >>= 12; | |
4610 | set_desc_limit(desc, var.limit); | |
4611 | set_desc_base(desc, (unsigned long)var.base); | |
5601d05b GN |
4612 | #ifdef CONFIG_X86_64 |
4613 | if (base3) | |
4614 | *base3 = var.base >> 32; | |
4615 | #endif | |
2dafc6c2 GN |
4616 | desc->type = var.type; |
4617 | desc->s = var.s; | |
4618 | desc->dpl = var.dpl; | |
4619 | desc->p = var.present; | |
4620 | desc->avl = var.avl; | |
4621 | desc->l = var.l; | |
4622 | desc->d = var.db; | |
4623 | desc->g = var.g; | |
4624 | ||
4625 | return true; | |
4626 | } | |
4627 | ||
1aa36616 AK |
4628 | static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector, |
4629 | struct desc_struct *desc, u32 base3, | |
4630 | int seg) | |
2dafc6c2 | 4631 | { |
4bff1e86 | 4632 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
2dafc6c2 GN |
4633 | struct kvm_segment var; |
4634 | ||
1aa36616 | 4635 | var.selector = selector; |
2dafc6c2 | 4636 | var.base = get_desc_base(desc); |
5601d05b GN |
4637 | #ifdef CONFIG_X86_64 |
4638 | var.base |= ((u64)base3) << 32; | |
4639 | #endif | |
2dafc6c2 GN |
4640 | var.limit = get_desc_limit(desc); |
4641 | if (desc->g) | |
4642 | var.limit = (var.limit << 12) | 0xfff; | |
4643 | var.type = desc->type; | |
4644 | var.present = desc->p; | |
4645 | var.dpl = desc->dpl; | |
4646 | var.db = desc->d; | |
4647 | var.s = desc->s; | |
4648 | var.l = desc->l; | |
4649 | var.g = desc->g; | |
4650 | var.avl = desc->avl; | |
4651 | var.present = desc->p; | |
4652 | var.unusable = !var.present; | |
4653 | var.padding = 0; | |
4654 | ||
4655 | kvm_set_segment(vcpu, &var, seg); | |
4656 | return; | |
4657 | } | |
4658 | ||
717746e3 AK |
4659 | static int emulator_get_msr(struct x86_emulate_ctxt *ctxt, |
4660 | u32 msr_index, u64 *pdata) | |
4661 | { | |
4662 | return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata); | |
4663 | } | |
4664 | ||
4665 | static int emulator_set_msr(struct x86_emulate_ctxt *ctxt, | |
4666 | u32 msr_index, u64 data) | |
4667 | { | |
8fe8ab46 WA |
4668 | struct msr_data msr; |
4669 | ||
4670 | msr.data = data; | |
4671 | msr.index = msr_index; | |
4672 | msr.host_initiated = false; | |
4673 | return kvm_set_msr(emul_to_vcpu(ctxt), &msr); | |
717746e3 AK |
4674 | } |
4675 | ||
222d21aa AK |
4676 | static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt, |
4677 | u32 pmc, u64 *pdata) | |
4678 | { | |
4679 | return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata); | |
4680 | } | |
4681 | ||
6c3287f7 AK |
4682 | static void emulator_halt(struct x86_emulate_ctxt *ctxt) |
4683 | { | |
4684 | emul_to_vcpu(ctxt)->arch.halt_request = 1; | |
4685 | } | |
4686 | ||
5037f6f3 AK |
4687 | static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt) |
4688 | { | |
4689 | preempt_disable(); | |
5197b808 | 4690 | kvm_load_guest_fpu(emul_to_vcpu(ctxt)); |
5037f6f3 AK |
4691 | /* |
4692 | * CR0.TS may reference the host fpu state, not the guest fpu state, | |
4693 | * so it may be clear at this point. | |
4694 | */ | |
4695 | clts(); | |
4696 | } | |
4697 | ||
4698 | static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt) | |
4699 | { | |
4700 | preempt_enable(); | |
4701 | } | |
4702 | ||
2953538e | 4703 | static int emulator_intercept(struct x86_emulate_ctxt *ctxt, |
8a76d7f2 | 4704 | struct x86_instruction_info *info, |
c4f035c6 AK |
4705 | enum x86_intercept_stage stage) |
4706 | { | |
2953538e | 4707 | return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage); |
c4f035c6 AK |
4708 | } |
4709 | ||
0017f93a | 4710 | static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt, |
bdb42f5a SB |
4711 | u32 *eax, u32 *ebx, u32 *ecx, u32 *edx) |
4712 | { | |
0017f93a | 4713 | kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx); |
bdb42f5a SB |
4714 | } |
4715 | ||
dd856efa AK |
4716 | static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg) |
4717 | { | |
4718 | return kvm_register_read(emul_to_vcpu(ctxt), reg); | |
4719 | } | |
4720 | ||
4721 | static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val) | |
4722 | { | |
4723 | kvm_register_write(emul_to_vcpu(ctxt), reg, val); | |
4724 | } | |
4725 | ||
0225fb50 | 4726 | static const struct x86_emulate_ops emulate_ops = { |
dd856efa AK |
4727 | .read_gpr = emulator_read_gpr, |
4728 | .write_gpr = emulator_write_gpr, | |
1871c602 | 4729 | .read_std = kvm_read_guest_virt_system, |
2dafc6c2 | 4730 | .write_std = kvm_write_guest_virt_system, |
1871c602 | 4731 | .fetch = kvm_fetch_guest_virt, |
bbd9b64e CO |
4732 | .read_emulated = emulator_read_emulated, |
4733 | .write_emulated = emulator_write_emulated, | |
4734 | .cmpxchg_emulated = emulator_cmpxchg_emulated, | |
3cb16fe7 | 4735 | .invlpg = emulator_invlpg, |
cf8f70bf GN |
4736 | .pio_in_emulated = emulator_pio_in_emulated, |
4737 | .pio_out_emulated = emulator_pio_out_emulated, | |
1aa36616 AK |
4738 | .get_segment = emulator_get_segment, |
4739 | .set_segment = emulator_set_segment, | |
5951c442 | 4740 | .get_cached_segment_base = emulator_get_cached_segment_base, |
2dafc6c2 | 4741 | .get_gdt = emulator_get_gdt, |
160ce1f1 | 4742 | .get_idt = emulator_get_idt, |
1ac9d0cf AK |
4743 | .set_gdt = emulator_set_gdt, |
4744 | .set_idt = emulator_set_idt, | |
52a46617 GN |
4745 | .get_cr = emulator_get_cr, |
4746 | .set_cr = emulator_set_cr, | |
4cee4798 | 4747 | .set_rflags = emulator_set_rflags, |
9c537244 | 4748 | .cpl = emulator_get_cpl, |
35aa5375 GN |
4749 | .get_dr = emulator_get_dr, |
4750 | .set_dr = emulator_set_dr, | |
717746e3 AK |
4751 | .set_msr = emulator_set_msr, |
4752 | .get_msr = emulator_get_msr, | |
222d21aa | 4753 | .read_pmc = emulator_read_pmc, |
6c3287f7 | 4754 | .halt = emulator_halt, |
bcaf5cc5 | 4755 | .wbinvd = emulator_wbinvd, |
d6aa1000 | 4756 | .fix_hypercall = emulator_fix_hypercall, |
5037f6f3 AK |
4757 | .get_fpu = emulator_get_fpu, |
4758 | .put_fpu = emulator_put_fpu, | |
c4f035c6 | 4759 | .intercept = emulator_intercept, |
bdb42f5a | 4760 | .get_cpuid = emulator_get_cpuid, |
bbd9b64e CO |
4761 | }; |
4762 | ||
95cb2295 GN |
4763 | static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask) |
4764 | { | |
4765 | u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask); | |
4766 | /* | |
4767 | * an sti; sti; sequence only disable interrupts for the first | |
4768 | * instruction. So, if the last instruction, be it emulated or | |
4769 | * not, left the system with the INT_STI flag enabled, it | |
4770 | * means that the last instruction is an sti. We should not | |
4771 | * leave the flag on in this case. The same goes for mov ss | |
4772 | */ | |
4773 | if (!(int_shadow & mask)) | |
4774 | kvm_x86_ops->set_interrupt_shadow(vcpu, mask); | |
4775 | } | |
4776 | ||
54b8486f GN |
4777 | static void inject_emulated_exception(struct kvm_vcpu *vcpu) |
4778 | { | |
4779 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; | |
da9cb575 | 4780 | if (ctxt->exception.vector == PF_VECTOR) |
6389ee94 | 4781 | kvm_propagate_fault(vcpu, &ctxt->exception); |
da9cb575 AK |
4782 | else if (ctxt->exception.error_code_valid) |
4783 | kvm_queue_exception_e(vcpu, ctxt->exception.vector, | |
4784 | ctxt->exception.error_code); | |
54b8486f | 4785 | else |
da9cb575 | 4786 | kvm_queue_exception(vcpu, ctxt->exception.vector); |
54b8486f GN |
4787 | } |
4788 | ||
dd856efa | 4789 | static void init_decode_cache(struct x86_emulate_ctxt *ctxt) |
b5c9ff73 | 4790 | { |
1ce19dc1 BP |
4791 | memset(&ctxt->opcode_len, 0, |
4792 | (void *)&ctxt->_regs - (void *)&ctxt->opcode_len); | |
b5c9ff73 | 4793 | |
9dac77fa AK |
4794 | ctxt->fetch.start = 0; |
4795 | ctxt->fetch.end = 0; | |
4796 | ctxt->io_read.pos = 0; | |
4797 | ctxt->io_read.end = 0; | |
4798 | ctxt->mem_read.pos = 0; | |
4799 | ctxt->mem_read.end = 0; | |
b5c9ff73 TY |
4800 | } |
4801 | ||
8ec4722d MG |
4802 | static void init_emulate_ctxt(struct kvm_vcpu *vcpu) |
4803 | { | |
adf52235 | 4804 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
8ec4722d MG |
4805 | int cs_db, cs_l; |
4806 | ||
8ec4722d MG |
4807 | kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l); |
4808 | ||
adf52235 TY |
4809 | ctxt->eflags = kvm_get_rflags(vcpu); |
4810 | ctxt->eip = kvm_rip_read(vcpu); | |
4811 | ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL : | |
4812 | (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 : | |
4813 | cs_l ? X86EMUL_MODE_PROT64 : | |
4814 | cs_db ? X86EMUL_MODE_PROT32 : | |
4815 | X86EMUL_MODE_PROT16; | |
4816 | ctxt->guest_mode = is_guest_mode(vcpu); | |
4817 | ||
dd856efa | 4818 | init_decode_cache(ctxt); |
7ae441ea | 4819 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; |
8ec4722d MG |
4820 | } |
4821 | ||
71f9833b | 4822 | int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip) |
63995653 | 4823 | { |
9d74191a | 4824 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
63995653 MG |
4825 | int ret; |
4826 | ||
4827 | init_emulate_ctxt(vcpu); | |
4828 | ||
9dac77fa AK |
4829 | ctxt->op_bytes = 2; |
4830 | ctxt->ad_bytes = 2; | |
4831 | ctxt->_eip = ctxt->eip + inc_eip; | |
9d74191a | 4832 | ret = emulate_int_real(ctxt, irq); |
63995653 MG |
4833 | |
4834 | if (ret != X86EMUL_CONTINUE) | |
4835 | return EMULATE_FAIL; | |
4836 | ||
9dac77fa | 4837 | ctxt->eip = ctxt->_eip; |
9d74191a TY |
4838 | kvm_rip_write(vcpu, ctxt->eip); |
4839 | kvm_set_rflags(vcpu, ctxt->eflags); | |
63995653 MG |
4840 | |
4841 | if (irq == NMI_VECTOR) | |
7460fb4a | 4842 | vcpu->arch.nmi_pending = 0; |
63995653 MG |
4843 | else |
4844 | vcpu->arch.interrupt.pending = false; | |
4845 | ||
4846 | return EMULATE_DONE; | |
4847 | } | |
4848 | EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt); | |
4849 | ||
6d77dbfc GN |
4850 | static int handle_emulation_failure(struct kvm_vcpu *vcpu) |
4851 | { | |
fc3a9157 JR |
4852 | int r = EMULATE_DONE; |
4853 | ||
6d77dbfc GN |
4854 | ++vcpu->stat.insn_emulation_fail; |
4855 | trace_kvm_emulate_insn_failed(vcpu); | |
fc3a9157 JR |
4856 | if (!is_guest_mode(vcpu)) { |
4857 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
4858 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
4859 | vcpu->run->internal.ndata = 0; | |
4860 | r = EMULATE_FAIL; | |
4861 | } | |
6d77dbfc | 4862 | kvm_queue_exception(vcpu, UD_VECTOR); |
fc3a9157 JR |
4863 | |
4864 | return r; | |
6d77dbfc GN |
4865 | } |
4866 | ||
93c05d3e | 4867 | static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2, |
991eebf9 GN |
4868 | bool write_fault_to_shadow_pgtable, |
4869 | int emulation_type) | |
a6f177ef | 4870 | { |
95b3cf69 | 4871 | gpa_t gpa = cr2; |
8e3d9d06 | 4872 | pfn_t pfn; |
a6f177ef | 4873 | |
991eebf9 GN |
4874 | if (emulation_type & EMULTYPE_NO_REEXECUTE) |
4875 | return false; | |
4876 | ||
95b3cf69 XG |
4877 | if (!vcpu->arch.mmu.direct_map) { |
4878 | /* | |
4879 | * Write permission should be allowed since only | |
4880 | * write access need to be emulated. | |
4881 | */ | |
4882 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL); | |
a6f177ef | 4883 | |
95b3cf69 XG |
4884 | /* |
4885 | * If the mapping is invalid in guest, let cpu retry | |
4886 | * it to generate fault. | |
4887 | */ | |
4888 | if (gpa == UNMAPPED_GVA) | |
4889 | return true; | |
4890 | } | |
a6f177ef | 4891 | |
8e3d9d06 XG |
4892 | /* |
4893 | * Do not retry the unhandleable instruction if it faults on the | |
4894 | * readonly host memory, otherwise it will goto a infinite loop: | |
4895 | * retry instruction -> write #PF -> emulation fail -> retry | |
4896 | * instruction -> ... | |
4897 | */ | |
4898 | pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa)); | |
95b3cf69 XG |
4899 | |
4900 | /* | |
4901 | * If the instruction failed on the error pfn, it can not be fixed, | |
4902 | * report the error to userspace. | |
4903 | */ | |
4904 | if (is_error_noslot_pfn(pfn)) | |
4905 | return false; | |
4906 | ||
4907 | kvm_release_pfn_clean(pfn); | |
4908 | ||
4909 | /* The instructions are well-emulated on direct mmu. */ | |
4910 | if (vcpu->arch.mmu.direct_map) { | |
4911 | unsigned int indirect_shadow_pages; | |
4912 | ||
4913 | spin_lock(&vcpu->kvm->mmu_lock); | |
4914 | indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages; | |
4915 | spin_unlock(&vcpu->kvm->mmu_lock); | |
4916 | ||
4917 | if (indirect_shadow_pages) | |
4918 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
4919 | ||
a6f177ef | 4920 | return true; |
8e3d9d06 | 4921 | } |
a6f177ef | 4922 | |
95b3cf69 XG |
4923 | /* |
4924 | * if emulation was due to access to shadowed page table | |
4925 | * and it failed try to unshadow page and re-enter the | |
4926 | * guest to let CPU execute the instruction. | |
4927 | */ | |
4928 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
93c05d3e XG |
4929 | |
4930 | /* | |
4931 | * If the access faults on its page table, it can not | |
4932 | * be fixed by unprotecting shadow page and it should | |
4933 | * be reported to userspace. | |
4934 | */ | |
4935 | return !write_fault_to_shadow_pgtable; | |
a6f177ef GN |
4936 | } |
4937 | ||
1cb3f3ae XG |
4938 | static bool retry_instruction(struct x86_emulate_ctxt *ctxt, |
4939 | unsigned long cr2, int emulation_type) | |
4940 | { | |
4941 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
4942 | unsigned long last_retry_eip, last_retry_addr, gpa = cr2; | |
4943 | ||
4944 | last_retry_eip = vcpu->arch.last_retry_eip; | |
4945 | last_retry_addr = vcpu->arch.last_retry_addr; | |
4946 | ||
4947 | /* | |
4948 | * If the emulation is caused by #PF and it is non-page_table | |
4949 | * writing instruction, it means the VM-EXIT is caused by shadow | |
4950 | * page protected, we can zap the shadow page and retry this | |
4951 | * instruction directly. | |
4952 | * | |
4953 | * Note: if the guest uses a non-page-table modifying instruction | |
4954 | * on the PDE that points to the instruction, then we will unmap | |
4955 | * the instruction and go to an infinite loop. So, we cache the | |
4956 | * last retried eip and the last fault address, if we meet the eip | |
4957 | * and the address again, we can break out of the potential infinite | |
4958 | * loop. | |
4959 | */ | |
4960 | vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0; | |
4961 | ||
4962 | if (!(emulation_type & EMULTYPE_RETRY)) | |
4963 | return false; | |
4964 | ||
4965 | if (x86_page_table_writing_insn(ctxt)) | |
4966 | return false; | |
4967 | ||
4968 | if (ctxt->eip == last_retry_eip && last_retry_addr == cr2) | |
4969 | return false; | |
4970 | ||
4971 | vcpu->arch.last_retry_eip = ctxt->eip; | |
4972 | vcpu->arch.last_retry_addr = cr2; | |
4973 | ||
4974 | if (!vcpu->arch.mmu.direct_map) | |
4975 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL); | |
4976 | ||
22368028 | 4977 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); |
1cb3f3ae XG |
4978 | |
4979 | return true; | |
4980 | } | |
4981 | ||
716d51ab GN |
4982 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu); |
4983 | static int complete_emulated_pio(struct kvm_vcpu *vcpu); | |
4984 | ||
4a1e10d5 PB |
4985 | static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7, |
4986 | unsigned long *db) | |
4987 | { | |
4988 | u32 dr6 = 0; | |
4989 | int i; | |
4990 | u32 enable, rwlen; | |
4991 | ||
4992 | enable = dr7; | |
4993 | rwlen = dr7 >> 16; | |
4994 | for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4) | |
4995 | if ((enable & 3) && (rwlen & 15) == type && db[i] == addr) | |
4996 | dr6 |= (1 << i); | |
4997 | return dr6; | |
4998 | } | |
4999 | ||
663f4c61 PB |
5000 | static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, int *r) |
5001 | { | |
5002 | struct kvm_run *kvm_run = vcpu->run; | |
5003 | ||
5004 | /* | |
5005 | * Use the "raw" value to see if TF was passed to the processor. | |
5006 | * Note that the new value of the flags has not been saved yet. | |
5007 | * | |
5008 | * This is correct even for TF set by the guest, because "the | |
5009 | * processor will not generate this exception after the instruction | |
5010 | * that sets the TF flag". | |
5011 | */ | |
5012 | unsigned long rflags = kvm_x86_ops->get_rflags(vcpu); | |
5013 | ||
5014 | if (unlikely(rflags & X86_EFLAGS_TF)) { | |
5015 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) { | |
5016 | kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1; | |
5017 | kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip; | |
5018 | kvm_run->debug.arch.exception = DB_VECTOR; | |
5019 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
5020 | *r = EMULATE_USER_EXIT; | |
5021 | } else { | |
5022 | vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF; | |
5023 | /* | |
5024 | * "Certain debug exceptions may clear bit 0-3. The | |
5025 | * remaining contents of the DR6 register are never | |
5026 | * cleared by the processor". | |
5027 | */ | |
5028 | vcpu->arch.dr6 &= ~15; | |
5029 | vcpu->arch.dr6 |= DR6_BS; | |
5030 | kvm_queue_exception(vcpu, DB_VECTOR); | |
5031 | } | |
5032 | } | |
5033 | } | |
5034 | ||
4a1e10d5 PB |
5035 | static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r) |
5036 | { | |
5037 | struct kvm_run *kvm_run = vcpu->run; | |
5038 | unsigned long eip = vcpu->arch.emulate_ctxt.eip; | |
5039 | u32 dr6 = 0; | |
5040 | ||
5041 | if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) && | |
5042 | (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) { | |
5043 | dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
5044 | vcpu->arch.guest_debug_dr7, | |
5045 | vcpu->arch.eff_db); | |
5046 | ||
5047 | if (dr6 != 0) { | |
5048 | kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1; | |
5049 | kvm_run->debug.arch.pc = kvm_rip_read(vcpu) + | |
5050 | get_segment_base(vcpu, VCPU_SREG_CS); | |
5051 | ||
5052 | kvm_run->debug.arch.exception = DB_VECTOR; | |
5053 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
5054 | *r = EMULATE_USER_EXIT; | |
5055 | return true; | |
5056 | } | |
5057 | } | |
5058 | ||
5059 | if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK)) { | |
5060 | dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
5061 | vcpu->arch.dr7, | |
5062 | vcpu->arch.db); | |
5063 | ||
5064 | if (dr6 != 0) { | |
5065 | vcpu->arch.dr6 &= ~15; | |
5066 | vcpu->arch.dr6 |= dr6; | |
5067 | kvm_queue_exception(vcpu, DB_VECTOR); | |
5068 | *r = EMULATE_DONE; | |
5069 | return true; | |
5070 | } | |
5071 | } | |
5072 | ||
5073 | return false; | |
5074 | } | |
5075 | ||
51d8b661 AP |
5076 | int x86_emulate_instruction(struct kvm_vcpu *vcpu, |
5077 | unsigned long cr2, | |
dc25e89e AP |
5078 | int emulation_type, |
5079 | void *insn, | |
5080 | int insn_len) | |
bbd9b64e | 5081 | { |
95cb2295 | 5082 | int r; |
9d74191a | 5083 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
7ae441ea | 5084 | bool writeback = true; |
93c05d3e | 5085 | bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable; |
bbd9b64e | 5086 | |
93c05d3e XG |
5087 | /* |
5088 | * Clear write_fault_to_shadow_pgtable here to ensure it is | |
5089 | * never reused. | |
5090 | */ | |
5091 | vcpu->arch.write_fault_to_shadow_pgtable = false; | |
26eef70c | 5092 | kvm_clear_exception_queue(vcpu); |
8d7d8102 | 5093 | |
571008da | 5094 | if (!(emulation_type & EMULTYPE_NO_DECODE)) { |
8ec4722d | 5095 | init_emulate_ctxt(vcpu); |
4a1e10d5 PB |
5096 | |
5097 | /* | |
5098 | * We will reenter on the same instruction since | |
5099 | * we do not set complete_userspace_io. This does not | |
5100 | * handle watchpoints yet, those would be handled in | |
5101 | * the emulate_ops. | |
5102 | */ | |
5103 | if (kvm_vcpu_check_breakpoint(vcpu, &r)) | |
5104 | return r; | |
5105 | ||
9d74191a TY |
5106 | ctxt->interruptibility = 0; |
5107 | ctxt->have_exception = false; | |
5108 | ctxt->perm_ok = false; | |
bbd9b64e | 5109 | |
b51e974f | 5110 | ctxt->ud = emulation_type & EMULTYPE_TRAP_UD; |
4005996e | 5111 | |
9d74191a | 5112 | r = x86_decode_insn(ctxt, insn, insn_len); |
bbd9b64e | 5113 | |
e46479f8 | 5114 | trace_kvm_emulate_insn_start(vcpu); |
f2b5756b | 5115 | ++vcpu->stat.insn_emulation; |
1d2887e2 | 5116 | if (r != EMULATION_OK) { |
4005996e AK |
5117 | if (emulation_type & EMULTYPE_TRAP_UD) |
5118 | return EMULATE_FAIL; | |
991eebf9 GN |
5119 | if (reexecute_instruction(vcpu, cr2, write_fault_to_spt, |
5120 | emulation_type)) | |
bbd9b64e | 5121 | return EMULATE_DONE; |
6d77dbfc GN |
5122 | if (emulation_type & EMULTYPE_SKIP) |
5123 | return EMULATE_FAIL; | |
5124 | return handle_emulation_failure(vcpu); | |
bbd9b64e CO |
5125 | } |
5126 | } | |
5127 | ||
ba8afb6b | 5128 | if (emulation_type & EMULTYPE_SKIP) { |
9dac77fa | 5129 | kvm_rip_write(vcpu, ctxt->_eip); |
ba8afb6b GN |
5130 | return EMULATE_DONE; |
5131 | } | |
5132 | ||
1cb3f3ae XG |
5133 | if (retry_instruction(ctxt, cr2, emulation_type)) |
5134 | return EMULATE_DONE; | |
5135 | ||
7ae441ea | 5136 | /* this is needed for vmware backdoor interface to work since it |
4d2179e1 | 5137 | changes registers values during IO operation */ |
7ae441ea GN |
5138 | if (vcpu->arch.emulate_regs_need_sync_from_vcpu) { |
5139 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; | |
dd856efa | 5140 | emulator_invalidate_register_cache(ctxt); |
7ae441ea | 5141 | } |
4d2179e1 | 5142 | |
5cd21917 | 5143 | restart: |
9d74191a | 5144 | r = x86_emulate_insn(ctxt); |
bbd9b64e | 5145 | |
775fde86 JR |
5146 | if (r == EMULATION_INTERCEPTED) |
5147 | return EMULATE_DONE; | |
5148 | ||
d2ddd1c4 | 5149 | if (r == EMULATION_FAILED) { |
991eebf9 GN |
5150 | if (reexecute_instruction(vcpu, cr2, write_fault_to_spt, |
5151 | emulation_type)) | |
c3cd7ffa GN |
5152 | return EMULATE_DONE; |
5153 | ||
6d77dbfc | 5154 | return handle_emulation_failure(vcpu); |
bbd9b64e CO |
5155 | } |
5156 | ||
9d74191a | 5157 | if (ctxt->have_exception) { |
54b8486f | 5158 | inject_emulated_exception(vcpu); |
d2ddd1c4 GN |
5159 | r = EMULATE_DONE; |
5160 | } else if (vcpu->arch.pio.count) { | |
0912c977 PB |
5161 | if (!vcpu->arch.pio.in) { |
5162 | /* FIXME: return into emulator if single-stepping. */ | |
3457e419 | 5163 | vcpu->arch.pio.count = 0; |
0912c977 | 5164 | } else { |
7ae441ea | 5165 | writeback = false; |
716d51ab GN |
5166 | vcpu->arch.complete_userspace_io = complete_emulated_pio; |
5167 | } | |
ac0a48c3 | 5168 | r = EMULATE_USER_EXIT; |
7ae441ea GN |
5169 | } else if (vcpu->mmio_needed) { |
5170 | if (!vcpu->mmio_is_write) | |
5171 | writeback = false; | |
ac0a48c3 | 5172 | r = EMULATE_USER_EXIT; |
716d51ab | 5173 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; |
7ae441ea | 5174 | } else if (r == EMULATION_RESTART) |
5cd21917 | 5175 | goto restart; |
d2ddd1c4 GN |
5176 | else |
5177 | r = EMULATE_DONE; | |
f850e2e6 | 5178 | |
7ae441ea | 5179 | if (writeback) { |
9d74191a | 5180 | toggle_interruptibility(vcpu, ctxt->interruptibility); |
7ae441ea | 5181 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
7ae441ea | 5182 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
9d74191a | 5183 | kvm_rip_write(vcpu, ctxt->eip); |
663f4c61 PB |
5184 | if (r == EMULATE_DONE) |
5185 | kvm_vcpu_check_singlestep(vcpu, &r); | |
5186 | kvm_set_rflags(vcpu, ctxt->eflags); | |
7ae441ea GN |
5187 | } else |
5188 | vcpu->arch.emulate_regs_need_sync_to_vcpu = true; | |
e85d28f8 GN |
5189 | |
5190 | return r; | |
de7d789a | 5191 | } |
51d8b661 | 5192 | EXPORT_SYMBOL_GPL(x86_emulate_instruction); |
de7d789a | 5193 | |
cf8f70bf | 5194 | int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port) |
de7d789a | 5195 | { |
cf8f70bf | 5196 | unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX); |
ca1d4a9e AK |
5197 | int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt, |
5198 | size, port, &val, 1); | |
cf8f70bf | 5199 | /* do not return to emulator after return from userspace */ |
7972995b | 5200 | vcpu->arch.pio.count = 0; |
de7d789a CO |
5201 | return ret; |
5202 | } | |
cf8f70bf | 5203 | EXPORT_SYMBOL_GPL(kvm_fast_pio_out); |
de7d789a | 5204 | |
8cfdc000 ZA |
5205 | static void tsc_bad(void *info) |
5206 | { | |
0a3aee0d | 5207 | __this_cpu_write(cpu_tsc_khz, 0); |
8cfdc000 ZA |
5208 | } |
5209 | ||
5210 | static void tsc_khz_changed(void *data) | |
c8076604 | 5211 | { |
8cfdc000 ZA |
5212 | struct cpufreq_freqs *freq = data; |
5213 | unsigned long khz = 0; | |
5214 | ||
5215 | if (data) | |
5216 | khz = freq->new; | |
5217 | else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) | |
5218 | khz = cpufreq_quick_get(raw_smp_processor_id()); | |
5219 | if (!khz) | |
5220 | khz = tsc_khz; | |
0a3aee0d | 5221 | __this_cpu_write(cpu_tsc_khz, khz); |
c8076604 GH |
5222 | } |
5223 | ||
c8076604 GH |
5224 | static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val, |
5225 | void *data) | |
5226 | { | |
5227 | struct cpufreq_freqs *freq = data; | |
5228 | struct kvm *kvm; | |
5229 | struct kvm_vcpu *vcpu; | |
5230 | int i, send_ipi = 0; | |
5231 | ||
8cfdc000 ZA |
5232 | /* |
5233 | * We allow guests to temporarily run on slowing clocks, | |
5234 | * provided we notify them after, or to run on accelerating | |
5235 | * clocks, provided we notify them before. Thus time never | |
5236 | * goes backwards. | |
5237 | * | |
5238 | * However, we have a problem. We can't atomically update | |
5239 | * the frequency of a given CPU from this function; it is | |
5240 | * merely a notifier, which can be called from any CPU. | |
5241 | * Changing the TSC frequency at arbitrary points in time | |
5242 | * requires a recomputation of local variables related to | |
5243 | * the TSC for each VCPU. We must flag these local variables | |
5244 | * to be updated and be sure the update takes place with the | |
5245 | * new frequency before any guests proceed. | |
5246 | * | |
5247 | * Unfortunately, the combination of hotplug CPU and frequency | |
5248 | * change creates an intractable locking scenario; the order | |
5249 | * of when these callouts happen is undefined with respect to | |
5250 | * CPU hotplug, and they can race with each other. As such, | |
5251 | * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is | |
5252 | * undefined; you can actually have a CPU frequency change take | |
5253 | * place in between the computation of X and the setting of the | |
5254 | * variable. To protect against this problem, all updates of | |
5255 | * the per_cpu tsc_khz variable are done in an interrupt | |
5256 | * protected IPI, and all callers wishing to update the value | |
5257 | * must wait for a synchronous IPI to complete (which is trivial | |
5258 | * if the caller is on the CPU already). This establishes the | |
5259 | * necessary total order on variable updates. | |
5260 | * | |
5261 | * Note that because a guest time update may take place | |
5262 | * anytime after the setting of the VCPU's request bit, the | |
5263 | * correct TSC value must be set before the request. However, | |
5264 | * to ensure the update actually makes it to any guest which | |
5265 | * starts running in hardware virtualization between the set | |
5266 | * and the acquisition of the spinlock, we must also ping the | |
5267 | * CPU after setting the request bit. | |
5268 | * | |
5269 | */ | |
5270 | ||
c8076604 GH |
5271 | if (val == CPUFREQ_PRECHANGE && freq->old > freq->new) |
5272 | return 0; | |
5273 | if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new) | |
5274 | return 0; | |
8cfdc000 ZA |
5275 | |
5276 | smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1); | |
c8076604 | 5277 | |
2f303b74 | 5278 | spin_lock(&kvm_lock); |
c8076604 | 5279 | list_for_each_entry(kvm, &vm_list, vm_list) { |
988a2cae | 5280 | kvm_for_each_vcpu(i, vcpu, kvm) { |
c8076604 GH |
5281 | if (vcpu->cpu != freq->cpu) |
5282 | continue; | |
c285545f | 5283 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
c8076604 | 5284 | if (vcpu->cpu != smp_processor_id()) |
8cfdc000 | 5285 | send_ipi = 1; |
c8076604 GH |
5286 | } |
5287 | } | |
2f303b74 | 5288 | spin_unlock(&kvm_lock); |
c8076604 GH |
5289 | |
5290 | if (freq->old < freq->new && send_ipi) { | |
5291 | /* | |
5292 | * We upscale the frequency. Must make the guest | |
5293 | * doesn't see old kvmclock values while running with | |
5294 | * the new frequency, otherwise we risk the guest sees | |
5295 | * time go backwards. | |
5296 | * | |
5297 | * In case we update the frequency for another cpu | |
5298 | * (which might be in guest context) send an interrupt | |
5299 | * to kick the cpu out of guest context. Next time | |
5300 | * guest context is entered kvmclock will be updated, | |
5301 | * so the guest will not see stale values. | |
5302 | */ | |
8cfdc000 | 5303 | smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1); |
c8076604 GH |
5304 | } |
5305 | return 0; | |
5306 | } | |
5307 | ||
5308 | static struct notifier_block kvmclock_cpufreq_notifier_block = { | |
8cfdc000 ZA |
5309 | .notifier_call = kvmclock_cpufreq_notifier |
5310 | }; | |
5311 | ||
5312 | static int kvmclock_cpu_notifier(struct notifier_block *nfb, | |
5313 | unsigned long action, void *hcpu) | |
5314 | { | |
5315 | unsigned int cpu = (unsigned long)hcpu; | |
5316 | ||
5317 | switch (action) { | |
5318 | case CPU_ONLINE: | |
5319 | case CPU_DOWN_FAILED: | |
5320 | smp_call_function_single(cpu, tsc_khz_changed, NULL, 1); | |
5321 | break; | |
5322 | case CPU_DOWN_PREPARE: | |
5323 | smp_call_function_single(cpu, tsc_bad, NULL, 1); | |
5324 | break; | |
5325 | } | |
5326 | return NOTIFY_OK; | |
5327 | } | |
5328 | ||
5329 | static struct notifier_block kvmclock_cpu_notifier_block = { | |
5330 | .notifier_call = kvmclock_cpu_notifier, | |
5331 | .priority = -INT_MAX | |
c8076604 GH |
5332 | }; |
5333 | ||
b820cc0c ZA |
5334 | static void kvm_timer_init(void) |
5335 | { | |
5336 | int cpu; | |
5337 | ||
c285545f | 5338 | max_tsc_khz = tsc_khz; |
8cfdc000 | 5339 | register_hotcpu_notifier(&kvmclock_cpu_notifier_block); |
b820cc0c | 5340 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) { |
c285545f ZA |
5341 | #ifdef CONFIG_CPU_FREQ |
5342 | struct cpufreq_policy policy; | |
5343 | memset(&policy, 0, sizeof(policy)); | |
3e26f230 AK |
5344 | cpu = get_cpu(); |
5345 | cpufreq_get_policy(&policy, cpu); | |
c285545f ZA |
5346 | if (policy.cpuinfo.max_freq) |
5347 | max_tsc_khz = policy.cpuinfo.max_freq; | |
3e26f230 | 5348 | put_cpu(); |
c285545f | 5349 | #endif |
b820cc0c ZA |
5350 | cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block, |
5351 | CPUFREQ_TRANSITION_NOTIFIER); | |
5352 | } | |
c285545f | 5353 | pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz); |
8cfdc000 ZA |
5354 | for_each_online_cpu(cpu) |
5355 | smp_call_function_single(cpu, tsc_khz_changed, NULL, 1); | |
b820cc0c ZA |
5356 | } |
5357 | ||
ff9d07a0 ZY |
5358 | static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu); |
5359 | ||
f5132b01 | 5360 | int kvm_is_in_guest(void) |
ff9d07a0 | 5361 | { |
086c9855 | 5362 | return __this_cpu_read(current_vcpu) != NULL; |
ff9d07a0 ZY |
5363 | } |
5364 | ||
5365 | static int kvm_is_user_mode(void) | |
5366 | { | |
5367 | int user_mode = 3; | |
dcf46b94 | 5368 | |
086c9855 AS |
5369 | if (__this_cpu_read(current_vcpu)) |
5370 | user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu)); | |
dcf46b94 | 5371 | |
ff9d07a0 ZY |
5372 | return user_mode != 0; |
5373 | } | |
5374 | ||
5375 | static unsigned long kvm_get_guest_ip(void) | |
5376 | { | |
5377 | unsigned long ip = 0; | |
dcf46b94 | 5378 | |
086c9855 AS |
5379 | if (__this_cpu_read(current_vcpu)) |
5380 | ip = kvm_rip_read(__this_cpu_read(current_vcpu)); | |
dcf46b94 | 5381 | |
ff9d07a0 ZY |
5382 | return ip; |
5383 | } | |
5384 | ||
5385 | static struct perf_guest_info_callbacks kvm_guest_cbs = { | |
5386 | .is_in_guest = kvm_is_in_guest, | |
5387 | .is_user_mode = kvm_is_user_mode, | |
5388 | .get_guest_ip = kvm_get_guest_ip, | |
5389 | }; | |
5390 | ||
5391 | void kvm_before_handle_nmi(struct kvm_vcpu *vcpu) | |
5392 | { | |
086c9855 | 5393 | __this_cpu_write(current_vcpu, vcpu); |
ff9d07a0 ZY |
5394 | } |
5395 | EXPORT_SYMBOL_GPL(kvm_before_handle_nmi); | |
5396 | ||
5397 | void kvm_after_handle_nmi(struct kvm_vcpu *vcpu) | |
5398 | { | |
086c9855 | 5399 | __this_cpu_write(current_vcpu, NULL); |
ff9d07a0 ZY |
5400 | } |
5401 | EXPORT_SYMBOL_GPL(kvm_after_handle_nmi); | |
5402 | ||
ce88decf XG |
5403 | static void kvm_set_mmio_spte_mask(void) |
5404 | { | |
5405 | u64 mask; | |
5406 | int maxphyaddr = boot_cpu_data.x86_phys_bits; | |
5407 | ||
5408 | /* | |
5409 | * Set the reserved bits and the present bit of an paging-structure | |
5410 | * entry to generate page fault with PFER.RSV = 1. | |
5411 | */ | |
885032b9 XG |
5412 | /* Mask the reserved physical address bits. */ |
5413 | mask = ((1ull << (51 - maxphyaddr + 1)) - 1) << maxphyaddr; | |
5414 | ||
5415 | /* Bit 62 is always reserved for 32bit host. */ | |
5416 | mask |= 0x3ull << 62; | |
5417 | ||
5418 | /* Set the present bit. */ | |
ce88decf XG |
5419 | mask |= 1ull; |
5420 | ||
5421 | #ifdef CONFIG_X86_64 | |
5422 | /* | |
5423 | * If reserved bit is not supported, clear the present bit to disable | |
5424 | * mmio page fault. | |
5425 | */ | |
5426 | if (maxphyaddr == 52) | |
5427 | mask &= ~1ull; | |
5428 | #endif | |
5429 | ||
5430 | kvm_mmu_set_mmio_spte_mask(mask); | |
5431 | } | |
5432 | ||
16e8d74d MT |
5433 | #ifdef CONFIG_X86_64 |
5434 | static void pvclock_gtod_update_fn(struct work_struct *work) | |
5435 | { | |
d828199e MT |
5436 | struct kvm *kvm; |
5437 | ||
5438 | struct kvm_vcpu *vcpu; | |
5439 | int i; | |
5440 | ||
2f303b74 | 5441 | spin_lock(&kvm_lock); |
d828199e MT |
5442 | list_for_each_entry(kvm, &vm_list, vm_list) |
5443 | kvm_for_each_vcpu(i, vcpu, kvm) | |
5444 | set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests); | |
5445 | atomic_set(&kvm_guest_has_master_clock, 0); | |
2f303b74 | 5446 | spin_unlock(&kvm_lock); |
16e8d74d MT |
5447 | } |
5448 | ||
5449 | static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn); | |
5450 | ||
5451 | /* | |
5452 | * Notification about pvclock gtod data update. | |
5453 | */ | |
5454 | static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused, | |
5455 | void *priv) | |
5456 | { | |
5457 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
5458 | struct timekeeper *tk = priv; | |
5459 | ||
5460 | update_pvclock_gtod(tk); | |
5461 | ||
5462 | /* disable master clock if host does not trust, or does not | |
5463 | * use, TSC clocksource | |
5464 | */ | |
5465 | if (gtod->clock.vclock_mode != VCLOCK_TSC && | |
5466 | atomic_read(&kvm_guest_has_master_clock) != 0) | |
5467 | queue_work(system_long_wq, &pvclock_gtod_work); | |
5468 | ||
5469 | return 0; | |
5470 | } | |
5471 | ||
5472 | static struct notifier_block pvclock_gtod_notifier = { | |
5473 | .notifier_call = pvclock_gtod_notify, | |
5474 | }; | |
5475 | #endif | |
5476 | ||
f8c16bba | 5477 | int kvm_arch_init(void *opaque) |
043405e1 | 5478 | { |
b820cc0c | 5479 | int r; |
6b61edf7 | 5480 | struct kvm_x86_ops *ops = opaque; |
f8c16bba | 5481 | |
f8c16bba ZX |
5482 | if (kvm_x86_ops) { |
5483 | printk(KERN_ERR "kvm: already loaded the other module\n"); | |
56c6d28a ZX |
5484 | r = -EEXIST; |
5485 | goto out; | |
f8c16bba ZX |
5486 | } |
5487 | ||
5488 | if (!ops->cpu_has_kvm_support()) { | |
5489 | printk(KERN_ERR "kvm: no hardware support\n"); | |
56c6d28a ZX |
5490 | r = -EOPNOTSUPP; |
5491 | goto out; | |
f8c16bba ZX |
5492 | } |
5493 | if (ops->disabled_by_bios()) { | |
5494 | printk(KERN_ERR "kvm: disabled by bios\n"); | |
56c6d28a ZX |
5495 | r = -EOPNOTSUPP; |
5496 | goto out; | |
f8c16bba ZX |
5497 | } |
5498 | ||
013f6a5d MT |
5499 | r = -ENOMEM; |
5500 | shared_msrs = alloc_percpu(struct kvm_shared_msrs); | |
5501 | if (!shared_msrs) { | |
5502 | printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n"); | |
5503 | goto out; | |
5504 | } | |
5505 | ||
97db56ce AK |
5506 | r = kvm_mmu_module_init(); |
5507 | if (r) | |
013f6a5d | 5508 | goto out_free_percpu; |
97db56ce | 5509 | |
ce88decf | 5510 | kvm_set_mmio_spte_mask(); |
97db56ce AK |
5511 | kvm_init_msr_list(); |
5512 | ||
f8c16bba | 5513 | kvm_x86_ops = ops; |
7b52345e | 5514 | kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK, |
4b12f0de | 5515 | PT_DIRTY_MASK, PT64_NX_MASK, 0); |
c8076604 | 5516 | |
b820cc0c | 5517 | kvm_timer_init(); |
c8076604 | 5518 | |
ff9d07a0 ZY |
5519 | perf_register_guest_info_callbacks(&kvm_guest_cbs); |
5520 | ||
2acf923e DC |
5521 | if (cpu_has_xsave) |
5522 | host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK); | |
5523 | ||
c5cc421b | 5524 | kvm_lapic_init(); |
16e8d74d MT |
5525 | #ifdef CONFIG_X86_64 |
5526 | pvclock_gtod_register_notifier(&pvclock_gtod_notifier); | |
5527 | #endif | |
5528 | ||
f8c16bba | 5529 | return 0; |
56c6d28a | 5530 | |
013f6a5d MT |
5531 | out_free_percpu: |
5532 | free_percpu(shared_msrs); | |
56c6d28a | 5533 | out: |
56c6d28a | 5534 | return r; |
043405e1 | 5535 | } |
8776e519 | 5536 | |
f8c16bba ZX |
5537 | void kvm_arch_exit(void) |
5538 | { | |
ff9d07a0 ZY |
5539 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); |
5540 | ||
888d256e JK |
5541 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) |
5542 | cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block, | |
5543 | CPUFREQ_TRANSITION_NOTIFIER); | |
8cfdc000 | 5544 | unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block); |
16e8d74d MT |
5545 | #ifdef CONFIG_X86_64 |
5546 | pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier); | |
5547 | #endif | |
f8c16bba | 5548 | kvm_x86_ops = NULL; |
56c6d28a | 5549 | kvm_mmu_module_exit(); |
013f6a5d | 5550 | free_percpu(shared_msrs); |
56c6d28a | 5551 | } |
f8c16bba | 5552 | |
8776e519 HB |
5553 | int kvm_emulate_halt(struct kvm_vcpu *vcpu) |
5554 | { | |
5555 | ++vcpu->stat.halt_exits; | |
5556 | if (irqchip_in_kernel(vcpu->kvm)) { | |
a4535290 | 5557 | vcpu->arch.mp_state = KVM_MP_STATE_HALTED; |
8776e519 HB |
5558 | return 1; |
5559 | } else { | |
5560 | vcpu->run->exit_reason = KVM_EXIT_HLT; | |
5561 | return 0; | |
5562 | } | |
5563 | } | |
5564 | EXPORT_SYMBOL_GPL(kvm_emulate_halt); | |
5565 | ||
55cd8e5a GN |
5566 | int kvm_hv_hypercall(struct kvm_vcpu *vcpu) |
5567 | { | |
5568 | u64 param, ingpa, outgpa, ret; | |
5569 | uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0; | |
5570 | bool fast, longmode; | |
5571 | int cs_db, cs_l; | |
5572 | ||
5573 | /* | |
5574 | * hypercall generates UD from non zero cpl and real mode | |
5575 | * per HYPER-V spec | |
5576 | */ | |
3eeb3288 | 5577 | if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) { |
55cd8e5a GN |
5578 | kvm_queue_exception(vcpu, UD_VECTOR); |
5579 | return 0; | |
5580 | } | |
5581 | ||
5582 | kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l); | |
5583 | longmode = is_long_mode(vcpu) && cs_l == 1; | |
5584 | ||
5585 | if (!longmode) { | |
ccd46936 GN |
5586 | param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) | |
5587 | (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff); | |
5588 | ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) | | |
5589 | (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff); | |
5590 | outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) | | |
5591 | (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff); | |
55cd8e5a GN |
5592 | } |
5593 | #ifdef CONFIG_X86_64 | |
5594 | else { | |
5595 | param = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
5596 | ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX); | |
5597 | outgpa = kvm_register_read(vcpu, VCPU_REGS_R8); | |
5598 | } | |
5599 | #endif | |
5600 | ||
5601 | code = param & 0xffff; | |
5602 | fast = (param >> 16) & 0x1; | |
5603 | rep_cnt = (param >> 32) & 0xfff; | |
5604 | rep_idx = (param >> 48) & 0xfff; | |
5605 | ||
5606 | trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa); | |
5607 | ||
c25bc163 GN |
5608 | switch (code) { |
5609 | case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT: | |
5610 | kvm_vcpu_on_spin(vcpu); | |
5611 | break; | |
5612 | default: | |
5613 | res = HV_STATUS_INVALID_HYPERCALL_CODE; | |
5614 | break; | |
5615 | } | |
55cd8e5a GN |
5616 | |
5617 | ret = res | (((u64)rep_done & 0xfff) << 32); | |
5618 | if (longmode) { | |
5619 | kvm_register_write(vcpu, VCPU_REGS_RAX, ret); | |
5620 | } else { | |
5621 | kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32); | |
5622 | kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff); | |
5623 | } | |
5624 | ||
5625 | return 1; | |
5626 | } | |
5627 | ||
6aef266c SV |
5628 | /* |
5629 | * kvm_pv_kick_cpu_op: Kick a vcpu. | |
5630 | * | |
5631 | * @apicid - apicid of vcpu to be kicked. | |
5632 | */ | |
5633 | static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid) | |
5634 | { | |
24d2166b | 5635 | struct kvm_lapic_irq lapic_irq; |
6aef266c | 5636 | |
24d2166b R |
5637 | lapic_irq.shorthand = 0; |
5638 | lapic_irq.dest_mode = 0; | |
5639 | lapic_irq.dest_id = apicid; | |
6aef266c | 5640 | |
24d2166b R |
5641 | lapic_irq.delivery_mode = APIC_DM_REMRD; |
5642 | kvm_irq_delivery_to_apic(kvm, 0, &lapic_irq, NULL); | |
6aef266c SV |
5643 | } |
5644 | ||
8776e519 HB |
5645 | int kvm_emulate_hypercall(struct kvm_vcpu *vcpu) |
5646 | { | |
5647 | unsigned long nr, a0, a1, a2, a3, ret; | |
2f333bcb | 5648 | int r = 1; |
8776e519 | 5649 | |
55cd8e5a GN |
5650 | if (kvm_hv_hypercall_enabled(vcpu->kvm)) |
5651 | return kvm_hv_hypercall(vcpu); | |
5652 | ||
5fdbf976 MT |
5653 | nr = kvm_register_read(vcpu, VCPU_REGS_RAX); |
5654 | a0 = kvm_register_read(vcpu, VCPU_REGS_RBX); | |
5655 | a1 = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
5656 | a2 = kvm_register_read(vcpu, VCPU_REGS_RDX); | |
5657 | a3 = kvm_register_read(vcpu, VCPU_REGS_RSI); | |
8776e519 | 5658 | |
229456fc | 5659 | trace_kvm_hypercall(nr, a0, a1, a2, a3); |
2714d1d3 | 5660 | |
8776e519 HB |
5661 | if (!is_long_mode(vcpu)) { |
5662 | nr &= 0xFFFFFFFF; | |
5663 | a0 &= 0xFFFFFFFF; | |
5664 | a1 &= 0xFFFFFFFF; | |
5665 | a2 &= 0xFFFFFFFF; | |
5666 | a3 &= 0xFFFFFFFF; | |
5667 | } | |
5668 | ||
07708c4a JK |
5669 | if (kvm_x86_ops->get_cpl(vcpu) != 0) { |
5670 | ret = -KVM_EPERM; | |
5671 | goto out; | |
5672 | } | |
5673 | ||
8776e519 | 5674 | switch (nr) { |
b93463aa AK |
5675 | case KVM_HC_VAPIC_POLL_IRQ: |
5676 | ret = 0; | |
5677 | break; | |
6aef266c SV |
5678 | case KVM_HC_KICK_CPU: |
5679 | kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1); | |
5680 | ret = 0; | |
5681 | break; | |
8776e519 HB |
5682 | default: |
5683 | ret = -KVM_ENOSYS; | |
5684 | break; | |
5685 | } | |
07708c4a | 5686 | out: |
5fdbf976 | 5687 | kvm_register_write(vcpu, VCPU_REGS_RAX, ret); |
f11c3a8d | 5688 | ++vcpu->stat.hypercalls; |
2f333bcb | 5689 | return r; |
8776e519 HB |
5690 | } |
5691 | EXPORT_SYMBOL_GPL(kvm_emulate_hypercall); | |
5692 | ||
b6785def | 5693 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt) |
8776e519 | 5694 | { |
d6aa1000 | 5695 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
8776e519 | 5696 | char instruction[3]; |
5fdbf976 | 5697 | unsigned long rip = kvm_rip_read(vcpu); |
8776e519 | 5698 | |
8776e519 | 5699 | kvm_x86_ops->patch_hypercall(vcpu, instruction); |
8776e519 | 5700 | |
9d74191a | 5701 | return emulator_write_emulated(ctxt, rip, instruction, 3, NULL); |
8776e519 HB |
5702 | } |
5703 | ||
b6c7a5dc HB |
5704 | /* |
5705 | * Check if userspace requested an interrupt window, and that the | |
5706 | * interrupt window is open. | |
5707 | * | |
5708 | * No need to exit to userspace if we already have an interrupt queued. | |
5709 | */ | |
851ba692 | 5710 | static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu) |
b6c7a5dc | 5711 | { |
8061823a | 5712 | return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) && |
851ba692 | 5713 | vcpu->run->request_interrupt_window && |
5df56646 | 5714 | kvm_arch_interrupt_allowed(vcpu)); |
b6c7a5dc HB |
5715 | } |
5716 | ||
851ba692 | 5717 | static void post_kvm_run_save(struct kvm_vcpu *vcpu) |
b6c7a5dc | 5718 | { |
851ba692 AK |
5719 | struct kvm_run *kvm_run = vcpu->run; |
5720 | ||
91586a3b | 5721 | kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0; |
2d3ad1f4 | 5722 | kvm_run->cr8 = kvm_get_cr8(vcpu); |
b6c7a5dc | 5723 | kvm_run->apic_base = kvm_get_apic_base(vcpu); |
4531220b | 5724 | if (irqchip_in_kernel(vcpu->kvm)) |
b6c7a5dc | 5725 | kvm_run->ready_for_interrupt_injection = 1; |
4531220b | 5726 | else |
b6c7a5dc | 5727 | kvm_run->ready_for_interrupt_injection = |
fa9726b0 GN |
5728 | kvm_arch_interrupt_allowed(vcpu) && |
5729 | !kvm_cpu_has_interrupt(vcpu) && | |
5730 | !kvm_event_needs_reinjection(vcpu); | |
b6c7a5dc HB |
5731 | } |
5732 | ||
4484141a | 5733 | static int vapic_enter(struct kvm_vcpu *vcpu) |
b93463aa AK |
5734 | { |
5735 | struct kvm_lapic *apic = vcpu->arch.apic; | |
5736 | struct page *page; | |
5737 | ||
5738 | if (!apic || !apic->vapic_addr) | |
4484141a | 5739 | return 0; |
b93463aa AK |
5740 | |
5741 | page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT); | |
4484141a XG |
5742 | if (is_error_page(page)) |
5743 | return -EFAULT; | |
72dc67a6 IE |
5744 | |
5745 | vcpu->arch.apic->vapic_page = page; | |
4484141a | 5746 | return 0; |
b93463aa AK |
5747 | } |
5748 | ||
5749 | static void vapic_exit(struct kvm_vcpu *vcpu) | |
5750 | { | |
5751 | struct kvm_lapic *apic = vcpu->arch.apic; | |
f656ce01 | 5752 | int idx; |
b93463aa AK |
5753 | |
5754 | if (!apic || !apic->vapic_addr) | |
5755 | return; | |
5756 | ||
f656ce01 | 5757 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
b93463aa AK |
5758 | kvm_release_page_dirty(apic->vapic_page); |
5759 | mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT); | |
f656ce01 | 5760 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
b93463aa AK |
5761 | } |
5762 | ||
95ba8273 GN |
5763 | static void update_cr8_intercept(struct kvm_vcpu *vcpu) |
5764 | { | |
5765 | int max_irr, tpr; | |
5766 | ||
5767 | if (!kvm_x86_ops->update_cr8_intercept) | |
5768 | return; | |
5769 | ||
88c808fd AK |
5770 | if (!vcpu->arch.apic) |
5771 | return; | |
5772 | ||
8db3baa2 GN |
5773 | if (!vcpu->arch.apic->vapic_addr) |
5774 | max_irr = kvm_lapic_find_highest_irr(vcpu); | |
5775 | else | |
5776 | max_irr = -1; | |
95ba8273 GN |
5777 | |
5778 | if (max_irr != -1) | |
5779 | max_irr >>= 4; | |
5780 | ||
5781 | tpr = kvm_lapic_get_cr8(vcpu); | |
5782 | ||
5783 | kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr); | |
5784 | } | |
5785 | ||
851ba692 | 5786 | static void inject_pending_event(struct kvm_vcpu *vcpu) |
95ba8273 GN |
5787 | { |
5788 | /* try to reinject previous events if any */ | |
b59bb7bd | 5789 | if (vcpu->arch.exception.pending) { |
5c1c85d0 AK |
5790 | trace_kvm_inj_exception(vcpu->arch.exception.nr, |
5791 | vcpu->arch.exception.has_error_code, | |
5792 | vcpu->arch.exception.error_code); | |
b59bb7bd GN |
5793 | kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr, |
5794 | vcpu->arch.exception.has_error_code, | |
ce7ddec4 JR |
5795 | vcpu->arch.exception.error_code, |
5796 | vcpu->arch.exception.reinject); | |
b59bb7bd GN |
5797 | return; |
5798 | } | |
5799 | ||
95ba8273 GN |
5800 | if (vcpu->arch.nmi_injected) { |
5801 | kvm_x86_ops->set_nmi(vcpu); | |
5802 | return; | |
5803 | } | |
5804 | ||
5805 | if (vcpu->arch.interrupt.pending) { | |
66fd3f7f | 5806 | kvm_x86_ops->set_irq(vcpu); |
95ba8273 GN |
5807 | return; |
5808 | } | |
5809 | ||
5810 | /* try to inject new event if pending */ | |
5811 | if (vcpu->arch.nmi_pending) { | |
5812 | if (kvm_x86_ops->nmi_allowed(vcpu)) { | |
7460fb4a | 5813 | --vcpu->arch.nmi_pending; |
95ba8273 GN |
5814 | vcpu->arch.nmi_injected = true; |
5815 | kvm_x86_ops->set_nmi(vcpu); | |
5816 | } | |
c7c9c56c | 5817 | } else if (kvm_cpu_has_injectable_intr(vcpu)) { |
95ba8273 | 5818 | if (kvm_x86_ops->interrupt_allowed(vcpu)) { |
66fd3f7f GN |
5819 | kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), |
5820 | false); | |
5821 | kvm_x86_ops->set_irq(vcpu); | |
95ba8273 GN |
5822 | } |
5823 | } | |
5824 | } | |
5825 | ||
7460fb4a AK |
5826 | static void process_nmi(struct kvm_vcpu *vcpu) |
5827 | { | |
5828 | unsigned limit = 2; | |
5829 | ||
5830 | /* | |
5831 | * x86 is limited to one NMI running, and one NMI pending after it. | |
5832 | * If an NMI is already in progress, limit further NMIs to just one. | |
5833 | * Otherwise, allow two (and we'll inject the first one immediately). | |
5834 | */ | |
5835 | if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected) | |
5836 | limit = 1; | |
5837 | ||
5838 | vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0); | |
5839 | vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit); | |
5840 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
5841 | } | |
5842 | ||
3d81bc7e | 5843 | static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu) |
c7c9c56c YZ |
5844 | { |
5845 | u64 eoi_exit_bitmap[4]; | |
cf9e65b7 | 5846 | u32 tmr[8]; |
c7c9c56c | 5847 | |
3d81bc7e YZ |
5848 | if (!kvm_apic_hw_enabled(vcpu->arch.apic)) |
5849 | return; | |
c7c9c56c YZ |
5850 | |
5851 | memset(eoi_exit_bitmap, 0, 32); | |
cf9e65b7 | 5852 | memset(tmr, 0, 32); |
c7c9c56c | 5853 | |
cf9e65b7 | 5854 | kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr); |
c7c9c56c | 5855 | kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap); |
cf9e65b7 | 5856 | kvm_apic_update_tmr(vcpu, tmr); |
c7c9c56c YZ |
5857 | } |
5858 | ||
851ba692 | 5859 | static int vcpu_enter_guest(struct kvm_vcpu *vcpu) |
b6c7a5dc HB |
5860 | { |
5861 | int r; | |
6a8b1d13 | 5862 | bool req_int_win = !irqchip_in_kernel(vcpu->kvm) && |
851ba692 | 5863 | vcpu->run->request_interrupt_window; |
730dca42 | 5864 | bool req_immediate_exit = false; |
b6c7a5dc | 5865 | |
3e007509 | 5866 | if (vcpu->requests) { |
a8eeb04a | 5867 | if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) |
2e53d63a | 5868 | kvm_mmu_unload(vcpu); |
a8eeb04a | 5869 | if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu)) |
2f599714 | 5870 | __kvm_migrate_timers(vcpu); |
d828199e MT |
5871 | if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu)) |
5872 | kvm_gen_update_masterclock(vcpu->kvm); | |
0061d53d MT |
5873 | if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu)) |
5874 | kvm_gen_kvmclock_update(vcpu); | |
34c238a1 ZA |
5875 | if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) { |
5876 | r = kvm_guest_time_update(vcpu); | |
8cfdc000 ZA |
5877 | if (unlikely(r)) |
5878 | goto out; | |
5879 | } | |
a8eeb04a | 5880 | if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu)) |
4731d4c7 | 5881 | kvm_mmu_sync_roots(vcpu); |
a8eeb04a | 5882 | if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) |
d4acf7e7 | 5883 | kvm_x86_ops->tlb_flush(vcpu); |
a8eeb04a | 5884 | if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) { |
851ba692 | 5885 | vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS; |
b93463aa AK |
5886 | r = 0; |
5887 | goto out; | |
5888 | } | |
a8eeb04a | 5889 | if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) { |
851ba692 | 5890 | vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN; |
71c4dfaf JR |
5891 | r = 0; |
5892 | goto out; | |
5893 | } | |
a8eeb04a | 5894 | if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) { |
02daab21 AK |
5895 | vcpu->fpu_active = 0; |
5896 | kvm_x86_ops->fpu_deactivate(vcpu); | |
5897 | } | |
af585b92 GN |
5898 | if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) { |
5899 | /* Page is swapped out. Do synthetic halt */ | |
5900 | vcpu->arch.apf.halted = true; | |
5901 | r = 1; | |
5902 | goto out; | |
5903 | } | |
c9aaa895 GC |
5904 | if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu)) |
5905 | record_steal_time(vcpu); | |
7460fb4a AK |
5906 | if (kvm_check_request(KVM_REQ_NMI, vcpu)) |
5907 | process_nmi(vcpu); | |
f5132b01 GN |
5908 | if (kvm_check_request(KVM_REQ_PMU, vcpu)) |
5909 | kvm_handle_pmu_event(vcpu); | |
5910 | if (kvm_check_request(KVM_REQ_PMI, vcpu)) | |
5911 | kvm_deliver_pmi(vcpu); | |
3d81bc7e YZ |
5912 | if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu)) |
5913 | vcpu_scan_ioapic(vcpu); | |
2f52d58c | 5914 | } |
b93463aa | 5915 | |
b463a6f7 | 5916 | if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) { |
66450a21 JK |
5917 | kvm_apic_accept_events(vcpu); |
5918 | if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) { | |
5919 | r = 1; | |
5920 | goto out; | |
5921 | } | |
5922 | ||
b463a6f7 AK |
5923 | inject_pending_event(vcpu); |
5924 | ||
5925 | /* enable NMI/IRQ window open exits if needed */ | |
7460fb4a | 5926 | if (vcpu->arch.nmi_pending) |
03b28f81 JK |
5927 | req_immediate_exit = |
5928 | kvm_x86_ops->enable_nmi_window(vcpu) != 0; | |
c7c9c56c | 5929 | else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win) |
730dca42 JK |
5930 | req_immediate_exit = |
5931 | kvm_x86_ops->enable_irq_window(vcpu) != 0; | |
b463a6f7 AK |
5932 | |
5933 | if (kvm_lapic_enabled(vcpu)) { | |
c7c9c56c YZ |
5934 | /* |
5935 | * Update architecture specific hints for APIC | |
5936 | * virtual interrupt delivery. | |
5937 | */ | |
5938 | if (kvm_x86_ops->hwapic_irr_update) | |
5939 | kvm_x86_ops->hwapic_irr_update(vcpu, | |
5940 | kvm_lapic_find_highest_irr(vcpu)); | |
b463a6f7 AK |
5941 | update_cr8_intercept(vcpu); |
5942 | kvm_lapic_sync_to_vapic(vcpu); | |
5943 | } | |
5944 | } | |
5945 | ||
d8368af8 AK |
5946 | r = kvm_mmu_reload(vcpu); |
5947 | if (unlikely(r)) { | |
d905c069 | 5948 | goto cancel_injection; |
d8368af8 AK |
5949 | } |
5950 | ||
b6c7a5dc HB |
5951 | preempt_disable(); |
5952 | ||
5953 | kvm_x86_ops->prepare_guest_switch(vcpu); | |
2608d7a1 AK |
5954 | if (vcpu->fpu_active) |
5955 | kvm_load_guest_fpu(vcpu); | |
2acf923e | 5956 | kvm_load_guest_xcr0(vcpu); |
b6c7a5dc | 5957 | |
6b7e2d09 XG |
5958 | vcpu->mode = IN_GUEST_MODE; |
5959 | ||
5960 | /* We should set ->mode before check ->requests, | |
5961 | * see the comment in make_all_cpus_request. | |
5962 | */ | |
5963 | smp_mb(); | |
b6c7a5dc | 5964 | |
d94e1dc9 | 5965 | local_irq_disable(); |
32f88400 | 5966 | |
6b7e2d09 | 5967 | if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests |
d94e1dc9 | 5968 | || need_resched() || signal_pending(current)) { |
6b7e2d09 | 5969 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 5970 | smp_wmb(); |
6c142801 AK |
5971 | local_irq_enable(); |
5972 | preempt_enable(); | |
5973 | r = 1; | |
d905c069 | 5974 | goto cancel_injection; |
6c142801 AK |
5975 | } |
5976 | ||
f656ce01 | 5977 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); |
3200f405 | 5978 | |
d6185f20 NHE |
5979 | if (req_immediate_exit) |
5980 | smp_send_reschedule(vcpu->cpu); | |
5981 | ||
b6c7a5dc HB |
5982 | kvm_guest_enter(); |
5983 | ||
42dbaa5a | 5984 | if (unlikely(vcpu->arch.switch_db_regs)) { |
42dbaa5a JK |
5985 | set_debugreg(0, 7); |
5986 | set_debugreg(vcpu->arch.eff_db[0], 0); | |
5987 | set_debugreg(vcpu->arch.eff_db[1], 1); | |
5988 | set_debugreg(vcpu->arch.eff_db[2], 2); | |
5989 | set_debugreg(vcpu->arch.eff_db[3], 3); | |
5990 | } | |
b6c7a5dc | 5991 | |
229456fc | 5992 | trace_kvm_entry(vcpu->vcpu_id); |
851ba692 | 5993 | kvm_x86_ops->run(vcpu); |
b6c7a5dc | 5994 | |
24f1e32c FW |
5995 | /* |
5996 | * If the guest has used debug registers, at least dr7 | |
5997 | * will be disabled while returning to the host. | |
5998 | * If we don't have active breakpoints in the host, we don't | |
5999 | * care about the messed up debug address registers. But if | |
6000 | * we have some of them active, restore the old state. | |
6001 | */ | |
59d8eb53 | 6002 | if (hw_breakpoint_active()) |
24f1e32c | 6003 | hw_breakpoint_restore(); |
42dbaa5a | 6004 | |
886b470c MT |
6005 | vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, |
6006 | native_read_tsc()); | |
1d5f066e | 6007 | |
6b7e2d09 | 6008 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 6009 | smp_wmb(); |
a547c6db YZ |
6010 | |
6011 | /* Interrupt is enabled by handle_external_intr() */ | |
6012 | kvm_x86_ops->handle_external_intr(vcpu); | |
b6c7a5dc HB |
6013 | |
6014 | ++vcpu->stat.exits; | |
6015 | ||
6016 | /* | |
6017 | * We must have an instruction between local_irq_enable() and | |
6018 | * kvm_guest_exit(), so the timer interrupt isn't delayed by | |
6019 | * the interrupt shadow. The stat.exits increment will do nicely. | |
6020 | * But we need to prevent reordering, hence this barrier(): | |
6021 | */ | |
6022 | barrier(); | |
6023 | ||
6024 | kvm_guest_exit(); | |
6025 | ||
6026 | preempt_enable(); | |
6027 | ||
f656ce01 | 6028 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
3200f405 | 6029 | |
b6c7a5dc HB |
6030 | /* |
6031 | * Profile KVM exit RIPs: | |
6032 | */ | |
6033 | if (unlikely(prof_on == KVM_PROFILING)) { | |
5fdbf976 MT |
6034 | unsigned long rip = kvm_rip_read(vcpu); |
6035 | profile_hit(KVM_PROFILING, (void *)rip); | |
b6c7a5dc HB |
6036 | } |
6037 | ||
cc578287 ZA |
6038 | if (unlikely(vcpu->arch.tsc_always_catchup)) |
6039 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); | |
298101da | 6040 | |
5cfb1d5a MT |
6041 | if (vcpu->arch.apic_attention) |
6042 | kvm_lapic_sync_from_vapic(vcpu); | |
b93463aa | 6043 | |
851ba692 | 6044 | r = kvm_x86_ops->handle_exit(vcpu); |
d905c069 MT |
6045 | return r; |
6046 | ||
6047 | cancel_injection: | |
6048 | kvm_x86_ops->cancel_injection(vcpu); | |
ae7a2a3f MT |
6049 | if (unlikely(vcpu->arch.apic_attention)) |
6050 | kvm_lapic_sync_from_vapic(vcpu); | |
d7690175 MT |
6051 | out: |
6052 | return r; | |
6053 | } | |
b6c7a5dc | 6054 | |
09cec754 | 6055 | |
851ba692 | 6056 | static int __vcpu_run(struct kvm_vcpu *vcpu) |
d7690175 MT |
6057 | { |
6058 | int r; | |
f656ce01 | 6059 | struct kvm *kvm = vcpu->kvm; |
d7690175 | 6060 | |
f656ce01 | 6061 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
4484141a XG |
6062 | r = vapic_enter(vcpu); |
6063 | if (r) { | |
6064 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); | |
6065 | return r; | |
6066 | } | |
d7690175 MT |
6067 | |
6068 | r = 1; | |
6069 | while (r > 0) { | |
af585b92 GN |
6070 | if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE && |
6071 | !vcpu->arch.apf.halted) | |
851ba692 | 6072 | r = vcpu_enter_guest(vcpu); |
d7690175 | 6073 | else { |
f656ce01 | 6074 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
d7690175 | 6075 | kvm_vcpu_block(vcpu); |
f656ce01 | 6076 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
66450a21 JK |
6077 | if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) { |
6078 | kvm_apic_accept_events(vcpu); | |
09cec754 GN |
6079 | switch(vcpu->arch.mp_state) { |
6080 | case KVM_MP_STATE_HALTED: | |
6aef266c | 6081 | vcpu->arch.pv.pv_unhalted = false; |
d7690175 | 6082 | vcpu->arch.mp_state = |
09cec754 GN |
6083 | KVM_MP_STATE_RUNNABLE; |
6084 | case KVM_MP_STATE_RUNNABLE: | |
af585b92 | 6085 | vcpu->arch.apf.halted = false; |
09cec754 | 6086 | break; |
66450a21 JK |
6087 | case KVM_MP_STATE_INIT_RECEIVED: |
6088 | break; | |
09cec754 GN |
6089 | default: |
6090 | r = -EINTR; | |
6091 | break; | |
6092 | } | |
6093 | } | |
d7690175 MT |
6094 | } |
6095 | ||
09cec754 GN |
6096 | if (r <= 0) |
6097 | break; | |
6098 | ||
6099 | clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests); | |
6100 | if (kvm_cpu_has_pending_timer(vcpu)) | |
6101 | kvm_inject_pending_timer_irqs(vcpu); | |
6102 | ||
851ba692 | 6103 | if (dm_request_for_irq_injection(vcpu)) { |
09cec754 | 6104 | r = -EINTR; |
851ba692 | 6105 | vcpu->run->exit_reason = KVM_EXIT_INTR; |
09cec754 GN |
6106 | ++vcpu->stat.request_irq_exits; |
6107 | } | |
af585b92 GN |
6108 | |
6109 | kvm_check_async_pf_completion(vcpu); | |
6110 | ||
09cec754 GN |
6111 | if (signal_pending(current)) { |
6112 | r = -EINTR; | |
851ba692 | 6113 | vcpu->run->exit_reason = KVM_EXIT_INTR; |
09cec754 GN |
6114 | ++vcpu->stat.signal_exits; |
6115 | } | |
6116 | if (need_resched()) { | |
f656ce01 | 6117 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
09cec754 | 6118 | kvm_resched(vcpu); |
f656ce01 | 6119 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
d7690175 | 6120 | } |
b6c7a5dc HB |
6121 | } |
6122 | ||
f656ce01 | 6123 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
b6c7a5dc | 6124 | |
b93463aa AK |
6125 | vapic_exit(vcpu); |
6126 | ||
b6c7a5dc HB |
6127 | return r; |
6128 | } | |
6129 | ||
716d51ab GN |
6130 | static inline int complete_emulated_io(struct kvm_vcpu *vcpu) |
6131 | { | |
6132 | int r; | |
6133 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); | |
6134 | r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE); | |
6135 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); | |
6136 | if (r != EMULATE_DONE) | |
6137 | return 0; | |
6138 | return 1; | |
6139 | } | |
6140 | ||
6141 | static int complete_emulated_pio(struct kvm_vcpu *vcpu) | |
6142 | { | |
6143 | BUG_ON(!vcpu->arch.pio.count); | |
6144 | ||
6145 | return complete_emulated_io(vcpu); | |
6146 | } | |
6147 | ||
f78146b0 AK |
6148 | /* |
6149 | * Implements the following, as a state machine: | |
6150 | * | |
6151 | * read: | |
6152 | * for each fragment | |
87da7e66 XG |
6153 | * for each mmio piece in the fragment |
6154 | * write gpa, len | |
6155 | * exit | |
6156 | * copy data | |
f78146b0 AK |
6157 | * execute insn |
6158 | * | |
6159 | * write: | |
6160 | * for each fragment | |
87da7e66 XG |
6161 | * for each mmio piece in the fragment |
6162 | * write gpa, len | |
6163 | * copy data | |
6164 | * exit | |
f78146b0 | 6165 | */ |
716d51ab | 6166 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu) |
5287f194 AK |
6167 | { |
6168 | struct kvm_run *run = vcpu->run; | |
f78146b0 | 6169 | struct kvm_mmio_fragment *frag; |
87da7e66 | 6170 | unsigned len; |
5287f194 | 6171 | |
716d51ab | 6172 | BUG_ON(!vcpu->mmio_needed); |
5287f194 | 6173 | |
716d51ab | 6174 | /* Complete previous fragment */ |
87da7e66 XG |
6175 | frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment]; |
6176 | len = min(8u, frag->len); | |
716d51ab | 6177 | if (!vcpu->mmio_is_write) |
87da7e66 XG |
6178 | memcpy(frag->data, run->mmio.data, len); |
6179 | ||
6180 | if (frag->len <= 8) { | |
6181 | /* Switch to the next fragment. */ | |
6182 | frag++; | |
6183 | vcpu->mmio_cur_fragment++; | |
6184 | } else { | |
6185 | /* Go forward to the next mmio piece. */ | |
6186 | frag->data += len; | |
6187 | frag->gpa += len; | |
6188 | frag->len -= len; | |
6189 | } | |
6190 | ||
716d51ab GN |
6191 | if (vcpu->mmio_cur_fragment == vcpu->mmio_nr_fragments) { |
6192 | vcpu->mmio_needed = 0; | |
0912c977 PB |
6193 | |
6194 | /* FIXME: return into emulator if single-stepping. */ | |
cef4dea0 | 6195 | if (vcpu->mmio_is_write) |
716d51ab GN |
6196 | return 1; |
6197 | vcpu->mmio_read_completed = 1; | |
6198 | return complete_emulated_io(vcpu); | |
6199 | } | |
87da7e66 | 6200 | |
716d51ab GN |
6201 | run->exit_reason = KVM_EXIT_MMIO; |
6202 | run->mmio.phys_addr = frag->gpa; | |
6203 | if (vcpu->mmio_is_write) | |
87da7e66 XG |
6204 | memcpy(run->mmio.data, frag->data, min(8u, frag->len)); |
6205 | run->mmio.len = min(8u, frag->len); | |
716d51ab GN |
6206 | run->mmio.is_write = vcpu->mmio_is_write; |
6207 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; | |
6208 | return 0; | |
5287f194 AK |
6209 | } |
6210 | ||
716d51ab | 6211 | |
b6c7a5dc HB |
6212 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) |
6213 | { | |
6214 | int r; | |
6215 | sigset_t sigsaved; | |
6216 | ||
e5c30142 AK |
6217 | if (!tsk_used_math(current) && init_fpu(current)) |
6218 | return -ENOMEM; | |
6219 | ||
ac9f6dc0 AK |
6220 | if (vcpu->sigset_active) |
6221 | sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); | |
6222 | ||
a4535290 | 6223 | if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) { |
b6c7a5dc | 6224 | kvm_vcpu_block(vcpu); |
66450a21 | 6225 | kvm_apic_accept_events(vcpu); |
d7690175 | 6226 | clear_bit(KVM_REQ_UNHALT, &vcpu->requests); |
ac9f6dc0 AK |
6227 | r = -EAGAIN; |
6228 | goto out; | |
b6c7a5dc HB |
6229 | } |
6230 | ||
b6c7a5dc | 6231 | /* re-sync apic's tpr */ |
eea1cff9 AP |
6232 | if (!irqchip_in_kernel(vcpu->kvm)) { |
6233 | if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) { | |
6234 | r = -EINVAL; | |
6235 | goto out; | |
6236 | } | |
6237 | } | |
b6c7a5dc | 6238 | |
716d51ab GN |
6239 | if (unlikely(vcpu->arch.complete_userspace_io)) { |
6240 | int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io; | |
6241 | vcpu->arch.complete_userspace_io = NULL; | |
6242 | r = cui(vcpu); | |
6243 | if (r <= 0) | |
6244 | goto out; | |
6245 | } else | |
6246 | WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed); | |
5287f194 | 6247 | |
851ba692 | 6248 | r = __vcpu_run(vcpu); |
b6c7a5dc HB |
6249 | |
6250 | out: | |
f1d86e46 | 6251 | post_kvm_run_save(vcpu); |
b6c7a5dc HB |
6252 | if (vcpu->sigset_active) |
6253 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
6254 | ||
b6c7a5dc HB |
6255 | return r; |
6256 | } | |
6257 | ||
6258 | int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) | |
6259 | { | |
7ae441ea GN |
6260 | if (vcpu->arch.emulate_regs_need_sync_to_vcpu) { |
6261 | /* | |
6262 | * We are here if userspace calls get_regs() in the middle of | |
6263 | * instruction emulation. Registers state needs to be copied | |
4a969980 | 6264 | * back from emulation context to vcpu. Userspace shouldn't do |
7ae441ea GN |
6265 | * that usually, but some bad designed PV devices (vmware |
6266 | * backdoor interface) need this to work | |
6267 | */ | |
dd856efa | 6268 | emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt); |
7ae441ea GN |
6269 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
6270 | } | |
5fdbf976 MT |
6271 | regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX); |
6272 | regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX); | |
6273 | regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
6274 | regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX); | |
6275 | regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI); | |
6276 | regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI); | |
6277 | regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP); | |
6278 | regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP); | |
b6c7a5dc | 6279 | #ifdef CONFIG_X86_64 |
5fdbf976 MT |
6280 | regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8); |
6281 | regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9); | |
6282 | regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10); | |
6283 | regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11); | |
6284 | regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12); | |
6285 | regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13); | |
6286 | regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14); | |
6287 | regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15); | |
b6c7a5dc HB |
6288 | #endif |
6289 | ||
5fdbf976 | 6290 | regs->rip = kvm_rip_read(vcpu); |
91586a3b | 6291 | regs->rflags = kvm_get_rflags(vcpu); |
b6c7a5dc | 6292 | |
b6c7a5dc HB |
6293 | return 0; |
6294 | } | |
6295 | ||
6296 | int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) | |
6297 | { | |
7ae441ea GN |
6298 | vcpu->arch.emulate_regs_need_sync_from_vcpu = true; |
6299 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; | |
6300 | ||
5fdbf976 MT |
6301 | kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax); |
6302 | kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx); | |
6303 | kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx); | |
6304 | kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx); | |
6305 | kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi); | |
6306 | kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi); | |
6307 | kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp); | |
6308 | kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp); | |
b6c7a5dc | 6309 | #ifdef CONFIG_X86_64 |
5fdbf976 MT |
6310 | kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8); |
6311 | kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9); | |
6312 | kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10); | |
6313 | kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11); | |
6314 | kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12); | |
6315 | kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13); | |
6316 | kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14); | |
6317 | kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15); | |
b6c7a5dc HB |
6318 | #endif |
6319 | ||
5fdbf976 | 6320 | kvm_rip_write(vcpu, regs->rip); |
91586a3b | 6321 | kvm_set_rflags(vcpu, regs->rflags); |
b6c7a5dc | 6322 | |
b4f14abd JK |
6323 | vcpu->arch.exception.pending = false; |
6324 | ||
3842d135 AK |
6325 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
6326 | ||
b6c7a5dc HB |
6327 | return 0; |
6328 | } | |
6329 | ||
b6c7a5dc HB |
6330 | void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l) |
6331 | { | |
6332 | struct kvm_segment cs; | |
6333 | ||
3e6e0aab | 6334 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); |
b6c7a5dc HB |
6335 | *db = cs.db; |
6336 | *l = cs.l; | |
6337 | } | |
6338 | EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits); | |
6339 | ||
6340 | int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, | |
6341 | struct kvm_sregs *sregs) | |
6342 | { | |
89a27f4d | 6343 | struct desc_ptr dt; |
b6c7a5dc | 6344 | |
3e6e0aab GT |
6345 | kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
6346 | kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
6347 | kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
6348 | kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
6349 | kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
6350 | kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 6351 | |
3e6e0aab GT |
6352 | kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
6353 | kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc HB |
6354 | |
6355 | kvm_x86_ops->get_idt(vcpu, &dt); | |
89a27f4d GN |
6356 | sregs->idt.limit = dt.size; |
6357 | sregs->idt.base = dt.address; | |
b6c7a5dc | 6358 | kvm_x86_ops->get_gdt(vcpu, &dt); |
89a27f4d GN |
6359 | sregs->gdt.limit = dt.size; |
6360 | sregs->gdt.base = dt.address; | |
b6c7a5dc | 6361 | |
4d4ec087 | 6362 | sregs->cr0 = kvm_read_cr0(vcpu); |
ad312c7c | 6363 | sregs->cr2 = vcpu->arch.cr2; |
9f8fe504 | 6364 | sregs->cr3 = kvm_read_cr3(vcpu); |
fc78f519 | 6365 | sregs->cr4 = kvm_read_cr4(vcpu); |
2d3ad1f4 | 6366 | sregs->cr8 = kvm_get_cr8(vcpu); |
f6801dff | 6367 | sregs->efer = vcpu->arch.efer; |
b6c7a5dc HB |
6368 | sregs->apic_base = kvm_get_apic_base(vcpu); |
6369 | ||
923c61bb | 6370 | memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap); |
b6c7a5dc | 6371 | |
36752c9b | 6372 | if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft) |
14d0bc1f GN |
6373 | set_bit(vcpu->arch.interrupt.nr, |
6374 | (unsigned long *)sregs->interrupt_bitmap); | |
16d7a191 | 6375 | |
b6c7a5dc HB |
6376 | return 0; |
6377 | } | |
6378 | ||
62d9f0db MT |
6379 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
6380 | struct kvm_mp_state *mp_state) | |
6381 | { | |
66450a21 | 6382 | kvm_apic_accept_events(vcpu); |
6aef266c SV |
6383 | if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED && |
6384 | vcpu->arch.pv.pv_unhalted) | |
6385 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
6386 | else | |
6387 | mp_state->mp_state = vcpu->arch.mp_state; | |
6388 | ||
62d9f0db MT |
6389 | return 0; |
6390 | } | |
6391 | ||
6392 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
6393 | struct kvm_mp_state *mp_state) | |
6394 | { | |
66450a21 JK |
6395 | if (!kvm_vcpu_has_lapic(vcpu) && |
6396 | mp_state->mp_state != KVM_MP_STATE_RUNNABLE) | |
6397 | return -EINVAL; | |
6398 | ||
6399 | if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) { | |
6400 | vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED; | |
6401 | set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events); | |
6402 | } else | |
6403 | vcpu->arch.mp_state = mp_state->mp_state; | |
3842d135 | 6404 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
62d9f0db MT |
6405 | return 0; |
6406 | } | |
6407 | ||
7f3d35fd KW |
6408 | int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index, |
6409 | int reason, bool has_error_code, u32 error_code) | |
b6c7a5dc | 6410 | { |
9d74191a | 6411 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
8ec4722d | 6412 | int ret; |
e01c2426 | 6413 | |
8ec4722d | 6414 | init_emulate_ctxt(vcpu); |
c697518a | 6415 | |
7f3d35fd | 6416 | ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason, |
9d74191a | 6417 | has_error_code, error_code); |
c697518a | 6418 | |
c697518a | 6419 | if (ret) |
19d04437 | 6420 | return EMULATE_FAIL; |
37817f29 | 6421 | |
9d74191a TY |
6422 | kvm_rip_write(vcpu, ctxt->eip); |
6423 | kvm_set_rflags(vcpu, ctxt->eflags); | |
3842d135 | 6424 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
19d04437 | 6425 | return EMULATE_DONE; |
37817f29 IE |
6426 | } |
6427 | EXPORT_SYMBOL_GPL(kvm_task_switch); | |
6428 | ||
b6c7a5dc HB |
6429 | int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, |
6430 | struct kvm_sregs *sregs) | |
6431 | { | |
6432 | int mmu_reset_needed = 0; | |
63f42e02 | 6433 | int pending_vec, max_bits, idx; |
89a27f4d | 6434 | struct desc_ptr dt; |
b6c7a5dc | 6435 | |
6d1068b3 PM |
6436 | if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE)) |
6437 | return -EINVAL; | |
6438 | ||
89a27f4d GN |
6439 | dt.size = sregs->idt.limit; |
6440 | dt.address = sregs->idt.base; | |
b6c7a5dc | 6441 | kvm_x86_ops->set_idt(vcpu, &dt); |
89a27f4d GN |
6442 | dt.size = sregs->gdt.limit; |
6443 | dt.address = sregs->gdt.base; | |
b6c7a5dc HB |
6444 | kvm_x86_ops->set_gdt(vcpu, &dt); |
6445 | ||
ad312c7c | 6446 | vcpu->arch.cr2 = sregs->cr2; |
9f8fe504 | 6447 | mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3; |
dc7e795e | 6448 | vcpu->arch.cr3 = sregs->cr3; |
aff48baa | 6449 | __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail); |
b6c7a5dc | 6450 | |
2d3ad1f4 | 6451 | kvm_set_cr8(vcpu, sregs->cr8); |
b6c7a5dc | 6452 | |
f6801dff | 6453 | mmu_reset_needed |= vcpu->arch.efer != sregs->efer; |
b6c7a5dc | 6454 | kvm_x86_ops->set_efer(vcpu, sregs->efer); |
b6c7a5dc HB |
6455 | kvm_set_apic_base(vcpu, sregs->apic_base); |
6456 | ||
4d4ec087 | 6457 | mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0; |
b6c7a5dc | 6458 | kvm_x86_ops->set_cr0(vcpu, sregs->cr0); |
d7306163 | 6459 | vcpu->arch.cr0 = sregs->cr0; |
b6c7a5dc | 6460 | |
fc78f519 | 6461 | mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4; |
b6c7a5dc | 6462 | kvm_x86_ops->set_cr4(vcpu, sregs->cr4); |
3ea3aa8c | 6463 | if (sregs->cr4 & X86_CR4_OSXSAVE) |
00b27a3e | 6464 | kvm_update_cpuid(vcpu); |
63f42e02 XG |
6465 | |
6466 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
7c93be44 | 6467 | if (!is_long_mode(vcpu) && is_pae(vcpu)) { |
9f8fe504 | 6468 | load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)); |
7c93be44 MT |
6469 | mmu_reset_needed = 1; |
6470 | } | |
63f42e02 | 6471 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
b6c7a5dc HB |
6472 | |
6473 | if (mmu_reset_needed) | |
6474 | kvm_mmu_reset_context(vcpu); | |
6475 | ||
a50abc3b | 6476 | max_bits = KVM_NR_INTERRUPTS; |
923c61bb GN |
6477 | pending_vec = find_first_bit( |
6478 | (const unsigned long *)sregs->interrupt_bitmap, max_bits); | |
6479 | if (pending_vec < max_bits) { | |
66fd3f7f | 6480 | kvm_queue_interrupt(vcpu, pending_vec, false); |
923c61bb | 6481 | pr_debug("Set back pending irq %d\n", pending_vec); |
b6c7a5dc HB |
6482 | } |
6483 | ||
3e6e0aab GT |
6484 | kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
6485 | kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
6486 | kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
6487 | kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
6488 | kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
6489 | kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 6490 | |
3e6e0aab GT |
6491 | kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
6492 | kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc | 6493 | |
5f0269f5 ME |
6494 | update_cr8_intercept(vcpu); |
6495 | ||
9c3e4aab | 6496 | /* Older userspace won't unhalt the vcpu on reset. */ |
c5af89b6 | 6497 | if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 && |
9c3e4aab | 6498 | sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 && |
3eeb3288 | 6499 | !is_protmode(vcpu)) |
9c3e4aab MT |
6500 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
6501 | ||
3842d135 AK |
6502 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
6503 | ||
b6c7a5dc HB |
6504 | return 0; |
6505 | } | |
6506 | ||
d0bfb940 JK |
6507 | int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, |
6508 | struct kvm_guest_debug *dbg) | |
b6c7a5dc | 6509 | { |
355be0b9 | 6510 | unsigned long rflags; |
ae675ef0 | 6511 | int i, r; |
b6c7a5dc | 6512 | |
4f926bf2 JK |
6513 | if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) { |
6514 | r = -EBUSY; | |
6515 | if (vcpu->arch.exception.pending) | |
2122ff5e | 6516 | goto out; |
4f926bf2 JK |
6517 | if (dbg->control & KVM_GUESTDBG_INJECT_DB) |
6518 | kvm_queue_exception(vcpu, DB_VECTOR); | |
6519 | else | |
6520 | kvm_queue_exception(vcpu, BP_VECTOR); | |
6521 | } | |
6522 | ||
91586a3b JK |
6523 | /* |
6524 | * Read rflags as long as potentially injected trace flags are still | |
6525 | * filtered out. | |
6526 | */ | |
6527 | rflags = kvm_get_rflags(vcpu); | |
355be0b9 JK |
6528 | |
6529 | vcpu->guest_debug = dbg->control; | |
6530 | if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE)) | |
6531 | vcpu->guest_debug = 0; | |
6532 | ||
6533 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) { | |
ae675ef0 JK |
6534 | for (i = 0; i < KVM_NR_DB_REGS; ++i) |
6535 | vcpu->arch.eff_db[i] = dbg->arch.debugreg[i]; | |
c8639010 | 6536 | vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7]; |
ae675ef0 JK |
6537 | } else { |
6538 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
6539 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
ae675ef0 | 6540 | } |
c8639010 | 6541 | kvm_update_dr7(vcpu); |
ae675ef0 | 6542 | |
f92653ee JK |
6543 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) |
6544 | vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) + | |
6545 | get_segment_base(vcpu, VCPU_SREG_CS); | |
94fe45da | 6546 | |
91586a3b JK |
6547 | /* |
6548 | * Trigger an rflags update that will inject or remove the trace | |
6549 | * flags. | |
6550 | */ | |
6551 | kvm_set_rflags(vcpu, rflags); | |
b6c7a5dc | 6552 | |
c8639010 | 6553 | kvm_x86_ops->update_db_bp_intercept(vcpu); |
b6c7a5dc | 6554 | |
4f926bf2 | 6555 | r = 0; |
d0bfb940 | 6556 | |
2122ff5e | 6557 | out: |
b6c7a5dc HB |
6558 | |
6559 | return r; | |
6560 | } | |
6561 | ||
8b006791 ZX |
6562 | /* |
6563 | * Translate a guest virtual address to a guest physical address. | |
6564 | */ | |
6565 | int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, | |
6566 | struct kvm_translation *tr) | |
6567 | { | |
6568 | unsigned long vaddr = tr->linear_address; | |
6569 | gpa_t gpa; | |
f656ce01 | 6570 | int idx; |
8b006791 | 6571 | |
f656ce01 | 6572 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
1871c602 | 6573 | gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL); |
f656ce01 | 6574 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
8b006791 ZX |
6575 | tr->physical_address = gpa; |
6576 | tr->valid = gpa != UNMAPPED_GVA; | |
6577 | tr->writeable = 1; | |
6578 | tr->usermode = 0; | |
8b006791 ZX |
6579 | |
6580 | return 0; | |
6581 | } | |
6582 | ||
d0752060 HB |
6583 | int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) |
6584 | { | |
98918833 SY |
6585 | struct i387_fxsave_struct *fxsave = |
6586 | &vcpu->arch.guest_fpu.state->fxsave; | |
d0752060 | 6587 | |
d0752060 HB |
6588 | memcpy(fpu->fpr, fxsave->st_space, 128); |
6589 | fpu->fcw = fxsave->cwd; | |
6590 | fpu->fsw = fxsave->swd; | |
6591 | fpu->ftwx = fxsave->twd; | |
6592 | fpu->last_opcode = fxsave->fop; | |
6593 | fpu->last_ip = fxsave->rip; | |
6594 | fpu->last_dp = fxsave->rdp; | |
6595 | memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space); | |
6596 | ||
d0752060 HB |
6597 | return 0; |
6598 | } | |
6599 | ||
6600 | int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) | |
6601 | { | |
98918833 SY |
6602 | struct i387_fxsave_struct *fxsave = |
6603 | &vcpu->arch.guest_fpu.state->fxsave; | |
d0752060 | 6604 | |
d0752060 HB |
6605 | memcpy(fxsave->st_space, fpu->fpr, 128); |
6606 | fxsave->cwd = fpu->fcw; | |
6607 | fxsave->swd = fpu->fsw; | |
6608 | fxsave->twd = fpu->ftwx; | |
6609 | fxsave->fop = fpu->last_opcode; | |
6610 | fxsave->rip = fpu->last_ip; | |
6611 | fxsave->rdp = fpu->last_dp; | |
6612 | memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space); | |
6613 | ||
d0752060 HB |
6614 | return 0; |
6615 | } | |
6616 | ||
10ab25cd | 6617 | int fx_init(struct kvm_vcpu *vcpu) |
d0752060 | 6618 | { |
10ab25cd JK |
6619 | int err; |
6620 | ||
6621 | err = fpu_alloc(&vcpu->arch.guest_fpu); | |
6622 | if (err) | |
6623 | return err; | |
6624 | ||
98918833 | 6625 | fpu_finit(&vcpu->arch.guest_fpu); |
d0752060 | 6626 | |
2acf923e DC |
6627 | /* |
6628 | * Ensure guest xcr0 is valid for loading | |
6629 | */ | |
6630 | vcpu->arch.xcr0 = XSTATE_FP; | |
6631 | ||
ad312c7c | 6632 | vcpu->arch.cr0 |= X86_CR0_ET; |
10ab25cd JK |
6633 | |
6634 | return 0; | |
d0752060 HB |
6635 | } |
6636 | EXPORT_SYMBOL_GPL(fx_init); | |
6637 | ||
98918833 SY |
6638 | static void fx_free(struct kvm_vcpu *vcpu) |
6639 | { | |
6640 | fpu_free(&vcpu->arch.guest_fpu); | |
6641 | } | |
6642 | ||
d0752060 HB |
6643 | void kvm_load_guest_fpu(struct kvm_vcpu *vcpu) |
6644 | { | |
2608d7a1 | 6645 | if (vcpu->guest_fpu_loaded) |
d0752060 HB |
6646 | return; |
6647 | ||
2acf923e DC |
6648 | /* |
6649 | * Restore all possible states in the guest, | |
6650 | * and assume host would use all available bits. | |
6651 | * Guest xcr0 would be loaded later. | |
6652 | */ | |
6653 | kvm_put_guest_xcr0(vcpu); | |
d0752060 | 6654 | vcpu->guest_fpu_loaded = 1; |
b1a74bf8 | 6655 | __kernel_fpu_begin(); |
98918833 | 6656 | fpu_restore_checking(&vcpu->arch.guest_fpu); |
0c04851c | 6657 | trace_kvm_fpu(1); |
d0752060 | 6658 | } |
d0752060 HB |
6659 | |
6660 | void kvm_put_guest_fpu(struct kvm_vcpu *vcpu) | |
6661 | { | |
2acf923e DC |
6662 | kvm_put_guest_xcr0(vcpu); |
6663 | ||
d0752060 HB |
6664 | if (!vcpu->guest_fpu_loaded) |
6665 | return; | |
6666 | ||
6667 | vcpu->guest_fpu_loaded = 0; | |
98918833 | 6668 | fpu_save_init(&vcpu->arch.guest_fpu); |
b1a74bf8 | 6669 | __kernel_fpu_end(); |
f096ed85 | 6670 | ++vcpu->stat.fpu_reload; |
a8eeb04a | 6671 | kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu); |
0c04851c | 6672 | trace_kvm_fpu(0); |
d0752060 | 6673 | } |
e9b11c17 ZX |
6674 | |
6675 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
6676 | { | |
12f9a48f | 6677 | kvmclock_reset(vcpu); |
7f1ea208 | 6678 | |
f5f48ee1 | 6679 | free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); |
98918833 | 6680 | fx_free(vcpu); |
e9b11c17 ZX |
6681 | kvm_x86_ops->vcpu_free(vcpu); |
6682 | } | |
6683 | ||
6684 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, | |
6685 | unsigned int id) | |
6686 | { | |
6755bae8 ZA |
6687 | if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0) |
6688 | printk_once(KERN_WARNING | |
6689 | "kvm: SMP vm created on host with unstable TSC; " | |
6690 | "guest TSC will not be reliable\n"); | |
26e5215f AK |
6691 | return kvm_x86_ops->vcpu_create(kvm, id); |
6692 | } | |
e9b11c17 | 6693 | |
26e5215f AK |
6694 | int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu) |
6695 | { | |
6696 | int r; | |
e9b11c17 | 6697 | |
0bed3b56 | 6698 | vcpu->arch.mtrr_state.have_fixed = 1; |
9fc77441 MT |
6699 | r = vcpu_load(vcpu); |
6700 | if (r) | |
6701 | return r; | |
57f252f2 | 6702 | kvm_vcpu_reset(vcpu); |
8a3c1a33 | 6703 | kvm_mmu_setup(vcpu); |
e9b11c17 | 6704 | vcpu_put(vcpu); |
e9b11c17 | 6705 | |
26e5215f | 6706 | return r; |
e9b11c17 ZX |
6707 | } |
6708 | ||
42897d86 MT |
6709 | int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
6710 | { | |
6711 | int r; | |
8fe8ab46 | 6712 | struct msr_data msr; |
42897d86 MT |
6713 | |
6714 | r = vcpu_load(vcpu); | |
6715 | if (r) | |
6716 | return r; | |
8fe8ab46 WA |
6717 | msr.data = 0x0; |
6718 | msr.index = MSR_IA32_TSC; | |
6719 | msr.host_initiated = true; | |
6720 | kvm_write_tsc(vcpu, &msr); | |
42897d86 MT |
6721 | vcpu_put(vcpu); |
6722 | ||
6723 | return r; | |
6724 | } | |
6725 | ||
d40ccc62 | 6726 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) |
e9b11c17 | 6727 | { |
9fc77441 | 6728 | int r; |
344d9588 GN |
6729 | vcpu->arch.apf.msr_val = 0; |
6730 | ||
9fc77441 MT |
6731 | r = vcpu_load(vcpu); |
6732 | BUG_ON(r); | |
e9b11c17 ZX |
6733 | kvm_mmu_unload(vcpu); |
6734 | vcpu_put(vcpu); | |
6735 | ||
98918833 | 6736 | fx_free(vcpu); |
e9b11c17 ZX |
6737 | kvm_x86_ops->vcpu_free(vcpu); |
6738 | } | |
6739 | ||
66450a21 | 6740 | void kvm_vcpu_reset(struct kvm_vcpu *vcpu) |
e9b11c17 | 6741 | { |
7460fb4a AK |
6742 | atomic_set(&vcpu->arch.nmi_queued, 0); |
6743 | vcpu->arch.nmi_pending = 0; | |
448fa4a9 JK |
6744 | vcpu->arch.nmi_injected = false; |
6745 | ||
42dbaa5a JK |
6746 | memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db)); |
6747 | vcpu->arch.dr6 = DR6_FIXED_1; | |
6748 | vcpu->arch.dr7 = DR7_FIXED_1; | |
c8639010 | 6749 | kvm_update_dr7(vcpu); |
42dbaa5a | 6750 | |
3842d135 | 6751 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
344d9588 | 6752 | vcpu->arch.apf.msr_val = 0; |
c9aaa895 | 6753 | vcpu->arch.st.msr_val = 0; |
3842d135 | 6754 | |
12f9a48f GC |
6755 | kvmclock_reset(vcpu); |
6756 | ||
af585b92 GN |
6757 | kvm_clear_async_pf_completion_queue(vcpu); |
6758 | kvm_async_pf_hash_reset(vcpu); | |
6759 | vcpu->arch.apf.halted = false; | |
3842d135 | 6760 | |
f5132b01 GN |
6761 | kvm_pmu_reset(vcpu); |
6762 | ||
66f7b72e JS |
6763 | memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs)); |
6764 | vcpu->arch.regs_avail = ~0; | |
6765 | vcpu->arch.regs_dirty = ~0; | |
6766 | ||
57f252f2 | 6767 | kvm_x86_ops->vcpu_reset(vcpu); |
e9b11c17 ZX |
6768 | } |
6769 | ||
66450a21 JK |
6770 | void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, unsigned int vector) |
6771 | { | |
6772 | struct kvm_segment cs; | |
6773 | ||
6774 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
6775 | cs.selector = vector << 8; | |
6776 | cs.base = vector << 12; | |
6777 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
6778 | kvm_rip_write(vcpu, 0); | |
e9b11c17 ZX |
6779 | } |
6780 | ||
10474ae8 | 6781 | int kvm_arch_hardware_enable(void *garbage) |
e9b11c17 | 6782 | { |
ca84d1a2 ZA |
6783 | struct kvm *kvm; |
6784 | struct kvm_vcpu *vcpu; | |
6785 | int i; | |
0dd6a6ed ZA |
6786 | int ret; |
6787 | u64 local_tsc; | |
6788 | u64 max_tsc = 0; | |
6789 | bool stable, backwards_tsc = false; | |
18863bdd AK |
6790 | |
6791 | kvm_shared_msr_cpu_online(); | |
0dd6a6ed ZA |
6792 | ret = kvm_x86_ops->hardware_enable(garbage); |
6793 | if (ret != 0) | |
6794 | return ret; | |
6795 | ||
6796 | local_tsc = native_read_tsc(); | |
6797 | stable = !check_tsc_unstable(); | |
6798 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
6799 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
6800 | if (!stable && vcpu->cpu == smp_processor_id()) | |
6801 | set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests); | |
6802 | if (stable && vcpu->arch.last_host_tsc > local_tsc) { | |
6803 | backwards_tsc = true; | |
6804 | if (vcpu->arch.last_host_tsc > max_tsc) | |
6805 | max_tsc = vcpu->arch.last_host_tsc; | |
6806 | } | |
6807 | } | |
6808 | } | |
6809 | ||
6810 | /* | |
6811 | * Sometimes, even reliable TSCs go backwards. This happens on | |
6812 | * platforms that reset TSC during suspend or hibernate actions, but | |
6813 | * maintain synchronization. We must compensate. Fortunately, we can | |
6814 | * detect that condition here, which happens early in CPU bringup, | |
6815 | * before any KVM threads can be running. Unfortunately, we can't | |
6816 | * bring the TSCs fully up to date with real time, as we aren't yet far | |
6817 | * enough into CPU bringup that we know how much real time has actually | |
6818 | * elapsed; our helper function, get_kernel_ns() will be using boot | |
6819 | * variables that haven't been updated yet. | |
6820 | * | |
6821 | * So we simply find the maximum observed TSC above, then record the | |
6822 | * adjustment to TSC in each VCPU. When the VCPU later gets loaded, | |
6823 | * the adjustment will be applied. Note that we accumulate | |
6824 | * adjustments, in case multiple suspend cycles happen before some VCPU | |
6825 | * gets a chance to run again. In the event that no KVM threads get a | |
6826 | * chance to run, we will miss the entire elapsed period, as we'll have | |
6827 | * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may | |
6828 | * loose cycle time. This isn't too big a deal, since the loss will be | |
6829 | * uniform across all VCPUs (not to mention the scenario is extremely | |
6830 | * unlikely). It is possible that a second hibernate recovery happens | |
6831 | * much faster than a first, causing the observed TSC here to be | |
6832 | * smaller; this would require additional padding adjustment, which is | |
6833 | * why we set last_host_tsc to the local tsc observed here. | |
6834 | * | |
6835 | * N.B. - this code below runs only on platforms with reliable TSC, | |
6836 | * as that is the only way backwards_tsc is set above. Also note | |
6837 | * that this runs for ALL vcpus, which is not a bug; all VCPUs should | |
6838 | * have the same delta_cyc adjustment applied if backwards_tsc | |
6839 | * is detected. Note further, this adjustment is only done once, | |
6840 | * as we reset last_host_tsc on all VCPUs to stop this from being | |
6841 | * called multiple times (one for each physical CPU bringup). | |
6842 | * | |
4a969980 | 6843 | * Platforms with unreliable TSCs don't have to deal with this, they |
0dd6a6ed ZA |
6844 | * will be compensated by the logic in vcpu_load, which sets the TSC to |
6845 | * catchup mode. This will catchup all VCPUs to real time, but cannot | |
6846 | * guarantee that they stay in perfect synchronization. | |
6847 | */ | |
6848 | if (backwards_tsc) { | |
6849 | u64 delta_cyc = max_tsc - local_tsc; | |
6850 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
6851 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
6852 | vcpu->arch.tsc_offset_adjustment += delta_cyc; | |
6853 | vcpu->arch.last_host_tsc = local_tsc; | |
d828199e MT |
6854 | set_bit(KVM_REQ_MASTERCLOCK_UPDATE, |
6855 | &vcpu->requests); | |
0dd6a6ed ZA |
6856 | } |
6857 | ||
6858 | /* | |
6859 | * We have to disable TSC offset matching.. if you were | |
6860 | * booting a VM while issuing an S4 host suspend.... | |
6861 | * you may have some problem. Solving this issue is | |
6862 | * left as an exercise to the reader. | |
6863 | */ | |
6864 | kvm->arch.last_tsc_nsec = 0; | |
6865 | kvm->arch.last_tsc_write = 0; | |
6866 | } | |
6867 | ||
6868 | } | |
6869 | return 0; | |
e9b11c17 ZX |
6870 | } |
6871 | ||
6872 | void kvm_arch_hardware_disable(void *garbage) | |
6873 | { | |
6874 | kvm_x86_ops->hardware_disable(garbage); | |
3548bab5 | 6875 | drop_user_return_notifiers(garbage); |
e9b11c17 ZX |
6876 | } |
6877 | ||
6878 | int kvm_arch_hardware_setup(void) | |
6879 | { | |
6880 | return kvm_x86_ops->hardware_setup(); | |
6881 | } | |
6882 | ||
6883 | void kvm_arch_hardware_unsetup(void) | |
6884 | { | |
6885 | kvm_x86_ops->hardware_unsetup(); | |
6886 | } | |
6887 | ||
6888 | void kvm_arch_check_processor_compat(void *rtn) | |
6889 | { | |
6890 | kvm_x86_ops->check_processor_compatibility(rtn); | |
6891 | } | |
6892 | ||
3e515705 AK |
6893 | bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu) |
6894 | { | |
6895 | return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL); | |
6896 | } | |
6897 | ||
54e9818f GN |
6898 | struct static_key kvm_no_apic_vcpu __read_mostly; |
6899 | ||
e9b11c17 ZX |
6900 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
6901 | { | |
6902 | struct page *page; | |
6903 | struct kvm *kvm; | |
6904 | int r; | |
6905 | ||
6906 | BUG_ON(vcpu->kvm == NULL); | |
6907 | kvm = vcpu->kvm; | |
6908 | ||
6aef266c | 6909 | vcpu->arch.pv.pv_unhalted = false; |
9aabc88f | 6910 | vcpu->arch.emulate_ctxt.ops = &emulate_ops; |
c5af89b6 | 6911 | if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu)) |
a4535290 | 6912 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
e9b11c17 | 6913 | else |
a4535290 | 6914 | vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED; |
e9b11c17 ZX |
6915 | |
6916 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
6917 | if (!page) { | |
6918 | r = -ENOMEM; | |
6919 | goto fail; | |
6920 | } | |
ad312c7c | 6921 | vcpu->arch.pio_data = page_address(page); |
e9b11c17 | 6922 | |
cc578287 | 6923 | kvm_set_tsc_khz(vcpu, max_tsc_khz); |
c285545f | 6924 | |
e9b11c17 ZX |
6925 | r = kvm_mmu_create(vcpu); |
6926 | if (r < 0) | |
6927 | goto fail_free_pio_data; | |
6928 | ||
6929 | if (irqchip_in_kernel(kvm)) { | |
6930 | r = kvm_create_lapic(vcpu); | |
6931 | if (r < 0) | |
6932 | goto fail_mmu_destroy; | |
54e9818f GN |
6933 | } else |
6934 | static_key_slow_inc(&kvm_no_apic_vcpu); | |
e9b11c17 | 6935 | |
890ca9ae HY |
6936 | vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4, |
6937 | GFP_KERNEL); | |
6938 | if (!vcpu->arch.mce_banks) { | |
6939 | r = -ENOMEM; | |
443c39bc | 6940 | goto fail_free_lapic; |
890ca9ae HY |
6941 | } |
6942 | vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS; | |
6943 | ||
f1797359 WY |
6944 | if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) { |
6945 | r = -ENOMEM; | |
f5f48ee1 | 6946 | goto fail_free_mce_banks; |
f1797359 | 6947 | } |
f5f48ee1 | 6948 | |
66f7b72e JS |
6949 | r = fx_init(vcpu); |
6950 | if (r) | |
6951 | goto fail_free_wbinvd_dirty_mask; | |
6952 | ||
ba904635 | 6953 | vcpu->arch.ia32_tsc_adjust_msr = 0x0; |
0b79459b | 6954 | vcpu->arch.pv_time_enabled = false; |
d7876f1b PB |
6955 | |
6956 | vcpu->arch.guest_supported_xcr0 = 0; | |
4344ee98 | 6957 | vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; |
d7876f1b | 6958 | |
af585b92 | 6959 | kvm_async_pf_hash_reset(vcpu); |
f5132b01 | 6960 | kvm_pmu_init(vcpu); |
af585b92 | 6961 | |
e9b11c17 | 6962 | return 0; |
66f7b72e JS |
6963 | fail_free_wbinvd_dirty_mask: |
6964 | free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); | |
f5f48ee1 SY |
6965 | fail_free_mce_banks: |
6966 | kfree(vcpu->arch.mce_banks); | |
443c39bc WY |
6967 | fail_free_lapic: |
6968 | kvm_free_lapic(vcpu); | |
e9b11c17 ZX |
6969 | fail_mmu_destroy: |
6970 | kvm_mmu_destroy(vcpu); | |
6971 | fail_free_pio_data: | |
ad312c7c | 6972 | free_page((unsigned long)vcpu->arch.pio_data); |
e9b11c17 ZX |
6973 | fail: |
6974 | return r; | |
6975 | } | |
6976 | ||
6977 | void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) | |
6978 | { | |
f656ce01 MT |
6979 | int idx; |
6980 | ||
f5132b01 | 6981 | kvm_pmu_destroy(vcpu); |
36cb93fd | 6982 | kfree(vcpu->arch.mce_banks); |
e9b11c17 | 6983 | kvm_free_lapic(vcpu); |
f656ce01 | 6984 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
e9b11c17 | 6985 | kvm_mmu_destroy(vcpu); |
f656ce01 | 6986 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
ad312c7c | 6987 | free_page((unsigned long)vcpu->arch.pio_data); |
54e9818f GN |
6988 | if (!irqchip_in_kernel(vcpu->kvm)) |
6989 | static_key_slow_dec(&kvm_no_apic_vcpu); | |
e9b11c17 | 6990 | } |
d19a9cd2 | 6991 | |
e08b9637 | 6992 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
d19a9cd2 | 6993 | { |
e08b9637 CO |
6994 | if (type) |
6995 | return -EINVAL; | |
6996 | ||
f05e70ac | 6997 | INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); |
365c8868 | 6998 | INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages); |
4d5c5d0f | 6999 | INIT_LIST_HEAD(&kvm->arch.assigned_dev_head); |
e0f0bbc5 | 7000 | atomic_set(&kvm->arch.noncoherent_dma_count, 0); |
d19a9cd2 | 7001 | |
5550af4d SY |
7002 | /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */ |
7003 | set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap); | |
7a84428a AW |
7004 | /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */ |
7005 | set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, | |
7006 | &kvm->arch.irq_sources_bitmap); | |
5550af4d | 7007 | |
038f8c11 | 7008 | raw_spin_lock_init(&kvm->arch.tsc_write_lock); |
1e08ec4a | 7009 | mutex_init(&kvm->arch.apic_map_lock); |
d828199e MT |
7010 | spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock); |
7011 | ||
7012 | pvclock_update_vm_gtod_copy(kvm); | |
53f658b3 | 7013 | |
d89f5eff | 7014 | return 0; |
d19a9cd2 ZX |
7015 | } |
7016 | ||
7017 | static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu) | |
7018 | { | |
9fc77441 MT |
7019 | int r; |
7020 | r = vcpu_load(vcpu); | |
7021 | BUG_ON(r); | |
d19a9cd2 ZX |
7022 | kvm_mmu_unload(vcpu); |
7023 | vcpu_put(vcpu); | |
7024 | } | |
7025 | ||
7026 | static void kvm_free_vcpus(struct kvm *kvm) | |
7027 | { | |
7028 | unsigned int i; | |
988a2cae | 7029 | struct kvm_vcpu *vcpu; |
d19a9cd2 ZX |
7030 | |
7031 | /* | |
7032 | * Unpin any mmu pages first. | |
7033 | */ | |
af585b92 GN |
7034 | kvm_for_each_vcpu(i, vcpu, kvm) { |
7035 | kvm_clear_async_pf_completion_queue(vcpu); | |
988a2cae | 7036 | kvm_unload_vcpu_mmu(vcpu); |
af585b92 | 7037 | } |
988a2cae GN |
7038 | kvm_for_each_vcpu(i, vcpu, kvm) |
7039 | kvm_arch_vcpu_free(vcpu); | |
7040 | ||
7041 | mutex_lock(&kvm->lock); | |
7042 | for (i = 0; i < atomic_read(&kvm->online_vcpus); i++) | |
7043 | kvm->vcpus[i] = NULL; | |
d19a9cd2 | 7044 | |
988a2cae GN |
7045 | atomic_set(&kvm->online_vcpus, 0); |
7046 | mutex_unlock(&kvm->lock); | |
d19a9cd2 ZX |
7047 | } |
7048 | ||
ad8ba2cd SY |
7049 | void kvm_arch_sync_events(struct kvm *kvm) |
7050 | { | |
ba4cef31 | 7051 | kvm_free_all_assigned_devices(kvm); |
aea924f6 | 7052 | kvm_free_pit(kvm); |
ad8ba2cd SY |
7053 | } |
7054 | ||
d19a9cd2 ZX |
7055 | void kvm_arch_destroy_vm(struct kvm *kvm) |
7056 | { | |
27469d29 AH |
7057 | if (current->mm == kvm->mm) { |
7058 | /* | |
7059 | * Free memory regions allocated on behalf of userspace, | |
7060 | * unless the the memory map has changed due to process exit | |
7061 | * or fd copying. | |
7062 | */ | |
7063 | struct kvm_userspace_memory_region mem; | |
7064 | memset(&mem, 0, sizeof(mem)); | |
7065 | mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT; | |
7066 | kvm_set_memory_region(kvm, &mem); | |
7067 | ||
7068 | mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT; | |
7069 | kvm_set_memory_region(kvm, &mem); | |
7070 | ||
7071 | mem.slot = TSS_PRIVATE_MEMSLOT; | |
7072 | kvm_set_memory_region(kvm, &mem); | |
7073 | } | |
6eb55818 | 7074 | kvm_iommu_unmap_guest(kvm); |
d7deeeb0 ZX |
7075 | kfree(kvm->arch.vpic); |
7076 | kfree(kvm->arch.vioapic); | |
d19a9cd2 | 7077 | kvm_free_vcpus(kvm); |
3d45830c AK |
7078 | if (kvm->arch.apic_access_page) |
7079 | put_page(kvm->arch.apic_access_page); | |
b7ebfb05 SY |
7080 | if (kvm->arch.ept_identity_pagetable) |
7081 | put_page(kvm->arch.ept_identity_pagetable); | |
1e08ec4a | 7082 | kfree(rcu_dereference_check(kvm->arch.apic_map, 1)); |
d19a9cd2 | 7083 | } |
0de10343 | 7084 | |
db3fe4eb TY |
7085 | void kvm_arch_free_memslot(struct kvm_memory_slot *free, |
7086 | struct kvm_memory_slot *dont) | |
7087 | { | |
7088 | int i; | |
7089 | ||
d89cc617 TY |
7090 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
7091 | if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) { | |
7092 | kvm_kvfree(free->arch.rmap[i]); | |
7093 | free->arch.rmap[i] = NULL; | |
77d11309 | 7094 | } |
d89cc617 TY |
7095 | if (i == 0) |
7096 | continue; | |
7097 | ||
7098 | if (!dont || free->arch.lpage_info[i - 1] != | |
7099 | dont->arch.lpage_info[i - 1]) { | |
7100 | kvm_kvfree(free->arch.lpage_info[i - 1]); | |
7101 | free->arch.lpage_info[i - 1] = NULL; | |
db3fe4eb TY |
7102 | } |
7103 | } | |
7104 | } | |
7105 | ||
7106 | int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages) | |
7107 | { | |
7108 | int i; | |
7109 | ||
d89cc617 | 7110 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
db3fe4eb TY |
7111 | unsigned long ugfn; |
7112 | int lpages; | |
d89cc617 | 7113 | int level = i + 1; |
db3fe4eb TY |
7114 | |
7115 | lpages = gfn_to_index(slot->base_gfn + npages - 1, | |
7116 | slot->base_gfn, level) + 1; | |
7117 | ||
d89cc617 TY |
7118 | slot->arch.rmap[i] = |
7119 | kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i])); | |
7120 | if (!slot->arch.rmap[i]) | |
77d11309 | 7121 | goto out_free; |
d89cc617 TY |
7122 | if (i == 0) |
7123 | continue; | |
77d11309 | 7124 | |
d89cc617 TY |
7125 | slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages * |
7126 | sizeof(*slot->arch.lpage_info[i - 1])); | |
7127 | if (!slot->arch.lpage_info[i - 1]) | |
db3fe4eb TY |
7128 | goto out_free; |
7129 | ||
7130 | if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1)) | |
d89cc617 | 7131 | slot->arch.lpage_info[i - 1][0].write_count = 1; |
db3fe4eb | 7132 | if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1)) |
d89cc617 | 7133 | slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1; |
db3fe4eb TY |
7134 | ugfn = slot->userspace_addr >> PAGE_SHIFT; |
7135 | /* | |
7136 | * If the gfn and userspace address are not aligned wrt each | |
7137 | * other, or if explicitly asked to, disable large page | |
7138 | * support for this slot | |
7139 | */ | |
7140 | if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) || | |
7141 | !kvm_largepages_enabled()) { | |
7142 | unsigned long j; | |
7143 | ||
7144 | for (j = 0; j < lpages; ++j) | |
d89cc617 | 7145 | slot->arch.lpage_info[i - 1][j].write_count = 1; |
db3fe4eb TY |
7146 | } |
7147 | } | |
7148 | ||
7149 | return 0; | |
7150 | ||
7151 | out_free: | |
d89cc617 TY |
7152 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
7153 | kvm_kvfree(slot->arch.rmap[i]); | |
7154 | slot->arch.rmap[i] = NULL; | |
7155 | if (i == 0) | |
7156 | continue; | |
7157 | ||
7158 | kvm_kvfree(slot->arch.lpage_info[i - 1]); | |
7159 | slot->arch.lpage_info[i - 1] = NULL; | |
db3fe4eb TY |
7160 | } |
7161 | return -ENOMEM; | |
7162 | } | |
7163 | ||
e59dbe09 TY |
7164 | void kvm_arch_memslots_updated(struct kvm *kvm) |
7165 | { | |
e6dff7d1 TY |
7166 | /* |
7167 | * memslots->generation has been incremented. | |
7168 | * mmio generation may have reached its maximum value. | |
7169 | */ | |
7170 | kvm_mmu_invalidate_mmio_sptes(kvm); | |
e59dbe09 TY |
7171 | } |
7172 | ||
f7784b8e MT |
7173 | int kvm_arch_prepare_memory_region(struct kvm *kvm, |
7174 | struct kvm_memory_slot *memslot, | |
f7784b8e | 7175 | struct kvm_userspace_memory_region *mem, |
7b6195a9 | 7176 | enum kvm_mr_change change) |
0de10343 | 7177 | { |
7a905b14 TY |
7178 | /* |
7179 | * Only private memory slots need to be mapped here since | |
7180 | * KVM_SET_MEMORY_REGION ioctl is no longer supported. | |
0de10343 | 7181 | */ |
7b6195a9 | 7182 | if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) { |
7a905b14 | 7183 | unsigned long userspace_addr; |
604b38ac | 7184 | |
7a905b14 TY |
7185 | /* |
7186 | * MAP_SHARED to prevent internal slot pages from being moved | |
7187 | * by fork()/COW. | |
7188 | */ | |
7b6195a9 | 7189 | userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE, |
7a905b14 TY |
7190 | PROT_READ | PROT_WRITE, |
7191 | MAP_SHARED | MAP_ANONYMOUS, 0); | |
0de10343 | 7192 | |
7a905b14 TY |
7193 | if (IS_ERR((void *)userspace_addr)) |
7194 | return PTR_ERR((void *)userspace_addr); | |
604b38ac | 7195 | |
7a905b14 | 7196 | memslot->userspace_addr = userspace_addr; |
0de10343 ZX |
7197 | } |
7198 | ||
f7784b8e MT |
7199 | return 0; |
7200 | } | |
7201 | ||
7202 | void kvm_arch_commit_memory_region(struct kvm *kvm, | |
7203 | struct kvm_userspace_memory_region *mem, | |
8482644a TY |
7204 | const struct kvm_memory_slot *old, |
7205 | enum kvm_mr_change change) | |
f7784b8e MT |
7206 | { |
7207 | ||
8482644a | 7208 | int nr_mmu_pages = 0; |
f7784b8e | 7209 | |
8482644a | 7210 | if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) { |
f7784b8e MT |
7211 | int ret; |
7212 | ||
8482644a TY |
7213 | ret = vm_munmap(old->userspace_addr, |
7214 | old->npages * PAGE_SIZE); | |
f7784b8e MT |
7215 | if (ret < 0) |
7216 | printk(KERN_WARNING | |
7217 | "kvm_vm_ioctl_set_memory_region: " | |
7218 | "failed to munmap memory\n"); | |
7219 | } | |
7220 | ||
48c0e4e9 XG |
7221 | if (!kvm->arch.n_requested_mmu_pages) |
7222 | nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm); | |
7223 | ||
48c0e4e9 | 7224 | if (nr_mmu_pages) |
0de10343 | 7225 | kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages); |
c972f3b1 TY |
7226 | /* |
7227 | * Write protect all pages for dirty logging. | |
7228 | * Existing largepage mappings are destroyed here and new ones will | |
7229 | * not be created until the end of the logging. | |
7230 | */ | |
8482644a | 7231 | if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES)) |
c972f3b1 | 7232 | kvm_mmu_slot_remove_write_access(kvm, mem->slot); |
0de10343 | 7233 | } |
1d737c8a | 7234 | |
2df72e9b | 7235 | void kvm_arch_flush_shadow_all(struct kvm *kvm) |
34d4cb8f | 7236 | { |
6ca18b69 | 7237 | kvm_mmu_invalidate_zap_all_pages(kvm); |
34d4cb8f MT |
7238 | } |
7239 | ||
2df72e9b MT |
7240 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, |
7241 | struct kvm_memory_slot *slot) | |
7242 | { | |
6ca18b69 | 7243 | kvm_mmu_invalidate_zap_all_pages(kvm); |
2df72e9b MT |
7244 | } |
7245 | ||
1d737c8a ZX |
7246 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) |
7247 | { | |
af585b92 GN |
7248 | return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE && |
7249 | !vcpu->arch.apf.halted) | |
7250 | || !list_empty_careful(&vcpu->async_pf.done) | |
66450a21 | 7251 | || kvm_apic_has_events(vcpu) |
6aef266c | 7252 | || vcpu->arch.pv.pv_unhalted |
7460fb4a | 7253 | || atomic_read(&vcpu->arch.nmi_queued) || |
a1b37100 GN |
7254 | (kvm_arch_interrupt_allowed(vcpu) && |
7255 | kvm_cpu_has_interrupt(vcpu)); | |
1d737c8a | 7256 | } |
5736199a | 7257 | |
b6d33834 | 7258 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
5736199a | 7259 | { |
b6d33834 | 7260 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; |
5736199a | 7261 | } |
78646121 GN |
7262 | |
7263 | int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu) | |
7264 | { | |
7265 | return kvm_x86_ops->interrupt_allowed(vcpu); | |
7266 | } | |
229456fc | 7267 | |
f92653ee JK |
7268 | bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip) |
7269 | { | |
7270 | unsigned long current_rip = kvm_rip_read(vcpu) + | |
7271 | get_segment_base(vcpu, VCPU_SREG_CS); | |
7272 | ||
7273 | return current_rip == linear_rip; | |
7274 | } | |
7275 | EXPORT_SYMBOL_GPL(kvm_is_linear_rip); | |
7276 | ||
94fe45da JK |
7277 | unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu) |
7278 | { | |
7279 | unsigned long rflags; | |
7280 | ||
7281 | rflags = kvm_x86_ops->get_rflags(vcpu); | |
7282 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) | |
c310bac5 | 7283 | rflags &= ~X86_EFLAGS_TF; |
94fe45da JK |
7284 | return rflags; |
7285 | } | |
7286 | EXPORT_SYMBOL_GPL(kvm_get_rflags); | |
7287 | ||
7288 | void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
7289 | { | |
7290 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP && | |
f92653ee | 7291 | kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip)) |
c310bac5 | 7292 | rflags |= X86_EFLAGS_TF; |
94fe45da | 7293 | kvm_x86_ops->set_rflags(vcpu, rflags); |
3842d135 | 7294 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
94fe45da JK |
7295 | } |
7296 | EXPORT_SYMBOL_GPL(kvm_set_rflags); | |
7297 | ||
56028d08 GN |
7298 | void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work) |
7299 | { | |
7300 | int r; | |
7301 | ||
fb67e14f | 7302 | if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) || |
f2e10669 | 7303 | work->wakeup_all) |
56028d08 GN |
7304 | return; |
7305 | ||
7306 | r = kvm_mmu_reload(vcpu); | |
7307 | if (unlikely(r)) | |
7308 | return; | |
7309 | ||
fb67e14f XG |
7310 | if (!vcpu->arch.mmu.direct_map && |
7311 | work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu)) | |
7312 | return; | |
7313 | ||
56028d08 GN |
7314 | vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true); |
7315 | } | |
7316 | ||
af585b92 GN |
7317 | static inline u32 kvm_async_pf_hash_fn(gfn_t gfn) |
7318 | { | |
7319 | return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU)); | |
7320 | } | |
7321 | ||
7322 | static inline u32 kvm_async_pf_next_probe(u32 key) | |
7323 | { | |
7324 | return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1); | |
7325 | } | |
7326 | ||
7327 | static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
7328 | { | |
7329 | u32 key = kvm_async_pf_hash_fn(gfn); | |
7330 | ||
7331 | while (vcpu->arch.apf.gfns[key] != ~0) | |
7332 | key = kvm_async_pf_next_probe(key); | |
7333 | ||
7334 | vcpu->arch.apf.gfns[key] = gfn; | |
7335 | } | |
7336 | ||
7337 | static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn) | |
7338 | { | |
7339 | int i; | |
7340 | u32 key = kvm_async_pf_hash_fn(gfn); | |
7341 | ||
7342 | for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) && | |
c7d28c24 XG |
7343 | (vcpu->arch.apf.gfns[key] != gfn && |
7344 | vcpu->arch.apf.gfns[key] != ~0); i++) | |
af585b92 GN |
7345 | key = kvm_async_pf_next_probe(key); |
7346 | ||
7347 | return key; | |
7348 | } | |
7349 | ||
7350 | bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
7351 | { | |
7352 | return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn; | |
7353 | } | |
7354 | ||
7355 | static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
7356 | { | |
7357 | u32 i, j, k; | |
7358 | ||
7359 | i = j = kvm_async_pf_gfn_slot(vcpu, gfn); | |
7360 | while (true) { | |
7361 | vcpu->arch.apf.gfns[i] = ~0; | |
7362 | do { | |
7363 | j = kvm_async_pf_next_probe(j); | |
7364 | if (vcpu->arch.apf.gfns[j] == ~0) | |
7365 | return; | |
7366 | k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]); | |
7367 | /* | |
7368 | * k lies cyclically in ]i,j] | |
7369 | * | i.k.j | | |
7370 | * |....j i.k.| or |.k..j i...| | |
7371 | */ | |
7372 | } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j)); | |
7373 | vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j]; | |
7374 | i = j; | |
7375 | } | |
7376 | } | |
7377 | ||
7c90705b GN |
7378 | static int apf_put_user(struct kvm_vcpu *vcpu, u32 val) |
7379 | { | |
7380 | ||
7381 | return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val, | |
7382 | sizeof(val)); | |
7383 | } | |
7384 | ||
af585b92 GN |
7385 | void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu, |
7386 | struct kvm_async_pf *work) | |
7387 | { | |
6389ee94 AK |
7388 | struct x86_exception fault; |
7389 | ||
7c90705b | 7390 | trace_kvm_async_pf_not_present(work->arch.token, work->gva); |
af585b92 | 7391 | kvm_add_async_pf_gfn(vcpu, work->arch.gfn); |
7c90705b GN |
7392 | |
7393 | if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) || | |
fc5f06fa GN |
7394 | (vcpu->arch.apf.send_user_only && |
7395 | kvm_x86_ops->get_cpl(vcpu) == 0)) | |
7c90705b GN |
7396 | kvm_make_request(KVM_REQ_APF_HALT, vcpu); |
7397 | else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) { | |
6389ee94 AK |
7398 | fault.vector = PF_VECTOR; |
7399 | fault.error_code_valid = true; | |
7400 | fault.error_code = 0; | |
7401 | fault.nested_page_fault = false; | |
7402 | fault.address = work->arch.token; | |
7403 | kvm_inject_page_fault(vcpu, &fault); | |
7c90705b | 7404 | } |
af585b92 GN |
7405 | } |
7406 | ||
7407 | void kvm_arch_async_page_present(struct kvm_vcpu *vcpu, | |
7408 | struct kvm_async_pf *work) | |
7409 | { | |
6389ee94 AK |
7410 | struct x86_exception fault; |
7411 | ||
7c90705b | 7412 | trace_kvm_async_pf_ready(work->arch.token, work->gva); |
f2e10669 | 7413 | if (work->wakeup_all) |
7c90705b GN |
7414 | work->arch.token = ~0; /* broadcast wakeup */ |
7415 | else | |
7416 | kvm_del_async_pf_gfn(vcpu, work->arch.gfn); | |
7417 | ||
7418 | if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) && | |
7419 | !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) { | |
6389ee94 AK |
7420 | fault.vector = PF_VECTOR; |
7421 | fault.error_code_valid = true; | |
7422 | fault.error_code = 0; | |
7423 | fault.nested_page_fault = false; | |
7424 | fault.address = work->arch.token; | |
7425 | kvm_inject_page_fault(vcpu, &fault); | |
7c90705b | 7426 | } |
e6d53e3b | 7427 | vcpu->arch.apf.halted = false; |
a4fa1635 | 7428 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
7c90705b GN |
7429 | } |
7430 | ||
7431 | bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu) | |
7432 | { | |
7433 | if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED)) | |
7434 | return true; | |
7435 | else | |
7436 | return !kvm_event_needs_reinjection(vcpu) && | |
7437 | kvm_x86_ops->interrupt_allowed(vcpu); | |
af585b92 GN |
7438 | } |
7439 | ||
e0f0bbc5 AW |
7440 | void kvm_arch_register_noncoherent_dma(struct kvm *kvm) |
7441 | { | |
7442 | atomic_inc(&kvm->arch.noncoherent_dma_count); | |
7443 | } | |
7444 | EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma); | |
7445 | ||
7446 | void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm) | |
7447 | { | |
7448 | atomic_dec(&kvm->arch.noncoherent_dma_count); | |
7449 | } | |
7450 | EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma); | |
7451 | ||
7452 | bool kvm_arch_has_noncoherent_dma(struct kvm *kvm) | |
7453 | { | |
7454 | return atomic_read(&kvm->arch.noncoherent_dma_count); | |
7455 | } | |
7456 | EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma); | |
7457 | ||
229456fc MT |
7458 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit); |
7459 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq); | |
7460 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault); | |
7461 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr); | |
7462 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr); | |
0ac406de | 7463 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun); |
d8cabddf | 7464 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit); |
17897f36 | 7465 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject); |
236649de | 7466 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit); |
ec1ff790 | 7467 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga); |
532a46b9 | 7468 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit); |
2e554e8d | 7469 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts); |
489223ed | 7470 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset); |