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" |
c9eab58f | 30 | #include "assigned-dev.h" |
474a5bb9 | 31 | #include "pmu.h" |
e83d5887 | 32 | #include "hyperv.h" |
313a3dc7 | 33 | |
18068523 | 34 | #include <linux/clocksource.h> |
4d5c5d0f | 35 | #include <linux/interrupt.h> |
313a3dc7 CO |
36 | #include <linux/kvm.h> |
37 | #include <linux/fs.h> | |
38 | #include <linux/vmalloc.h> | |
5fb76f9b | 39 | #include <linux/module.h> |
0de10343 | 40 | #include <linux/mman.h> |
2bacc55c | 41 | #include <linux/highmem.h> |
19de40a8 | 42 | #include <linux/iommu.h> |
62c476c7 | 43 | #include <linux/intel-iommu.h> |
c8076604 | 44 | #include <linux/cpufreq.h> |
18863bdd | 45 | #include <linux/user-return-notifier.h> |
a983fb23 | 46 | #include <linux/srcu.h> |
5a0e3ad6 | 47 | #include <linux/slab.h> |
ff9d07a0 | 48 | #include <linux/perf_event.h> |
7bee342a | 49 | #include <linux/uaccess.h> |
af585b92 | 50 | #include <linux/hash.h> |
a1b60c1c | 51 | #include <linux/pci.h> |
16e8d74d MT |
52 | #include <linux/timekeeper_internal.h> |
53 | #include <linux/pvclock_gtod.h> | |
aec51dc4 | 54 | #include <trace/events/kvm.h> |
2ed152af | 55 | |
229456fc MT |
56 | #define CREATE_TRACE_POINTS |
57 | #include "trace.h" | |
043405e1 | 58 | |
24f1e32c | 59 | #include <asm/debugreg.h> |
d825ed0a | 60 | #include <asm/msr.h> |
a5f61300 | 61 | #include <asm/desc.h> |
890ca9ae | 62 | #include <asm/mce.h> |
f89e32e0 | 63 | #include <linux/kernel_stat.h> |
78f7f1e5 | 64 | #include <asm/fpu/internal.h> /* Ugh! */ |
1d5f066e | 65 | #include <asm/pvclock.h> |
217fc9cf | 66 | #include <asm/div64.h> |
043405e1 | 67 | |
313a3dc7 | 68 | #define MAX_IO_MSRS 256 |
890ca9ae | 69 | #define KVM_MAX_MCE_BANKS 32 |
5854dbca | 70 | #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P) |
890ca9ae | 71 | |
0f65dd70 AK |
72 | #define emul_to_vcpu(ctxt) \ |
73 | container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt) | |
74 | ||
50a37eb4 JR |
75 | /* EFER defaults: |
76 | * - enable syscall per default because its emulated by KVM | |
77 | * - enable LME and LMA per default on 64 bit KVM | |
78 | */ | |
79 | #ifdef CONFIG_X86_64 | |
1260edbe LJ |
80 | static |
81 | u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA)); | |
50a37eb4 | 82 | #else |
1260edbe | 83 | static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE); |
50a37eb4 | 84 | #endif |
313a3dc7 | 85 | |
ba1389b7 AK |
86 | #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM |
87 | #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU | |
417bc304 | 88 | |
cb142eb7 | 89 | static void update_cr8_intercept(struct kvm_vcpu *vcpu); |
7460fb4a | 90 | static void process_nmi(struct kvm_vcpu *vcpu); |
6addfc42 | 91 | static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags); |
674eea0f | 92 | |
97896d04 | 93 | struct kvm_x86_ops *kvm_x86_ops; |
5fdbf976 | 94 | EXPORT_SYMBOL_GPL(kvm_x86_ops); |
97896d04 | 95 | |
476bc001 RR |
96 | static bool ignore_msrs = 0; |
97 | module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR); | |
ed85c068 | 98 | |
9ed96e87 MT |
99 | unsigned int min_timer_period_us = 500; |
100 | module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR); | |
101 | ||
630994b3 MT |
102 | static bool __read_mostly kvmclock_periodic_sync = true; |
103 | module_param(kvmclock_periodic_sync, bool, S_IRUGO); | |
104 | ||
92a1f12d JR |
105 | bool kvm_has_tsc_control; |
106 | EXPORT_SYMBOL_GPL(kvm_has_tsc_control); | |
107 | u32 kvm_max_guest_tsc_khz; | |
108 | EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz); | |
109 | ||
cc578287 ZA |
110 | /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */ |
111 | static u32 tsc_tolerance_ppm = 250; | |
112 | module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR); | |
113 | ||
d0659d94 MT |
114 | /* lapic timer advance (tscdeadline mode only) in nanoseconds */ |
115 | unsigned int lapic_timer_advance_ns = 0; | |
116 | module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR); | |
117 | ||
16a96021 MT |
118 | static bool backwards_tsc_observed = false; |
119 | ||
18863bdd AK |
120 | #define KVM_NR_SHARED_MSRS 16 |
121 | ||
122 | struct kvm_shared_msrs_global { | |
123 | int nr; | |
2bf78fa7 | 124 | u32 msrs[KVM_NR_SHARED_MSRS]; |
18863bdd AK |
125 | }; |
126 | ||
127 | struct kvm_shared_msrs { | |
128 | struct user_return_notifier urn; | |
129 | bool registered; | |
2bf78fa7 SY |
130 | struct kvm_shared_msr_values { |
131 | u64 host; | |
132 | u64 curr; | |
133 | } values[KVM_NR_SHARED_MSRS]; | |
18863bdd AK |
134 | }; |
135 | ||
136 | static struct kvm_shared_msrs_global __read_mostly shared_msrs_global; | |
013f6a5d | 137 | static struct kvm_shared_msrs __percpu *shared_msrs; |
18863bdd | 138 | |
417bc304 | 139 | struct kvm_stats_debugfs_item debugfs_entries[] = { |
ba1389b7 AK |
140 | { "pf_fixed", VCPU_STAT(pf_fixed) }, |
141 | { "pf_guest", VCPU_STAT(pf_guest) }, | |
142 | { "tlb_flush", VCPU_STAT(tlb_flush) }, | |
143 | { "invlpg", VCPU_STAT(invlpg) }, | |
144 | { "exits", VCPU_STAT(exits) }, | |
145 | { "io_exits", VCPU_STAT(io_exits) }, | |
146 | { "mmio_exits", VCPU_STAT(mmio_exits) }, | |
147 | { "signal_exits", VCPU_STAT(signal_exits) }, | |
148 | { "irq_window", VCPU_STAT(irq_window_exits) }, | |
f08864b4 | 149 | { "nmi_window", VCPU_STAT(nmi_window_exits) }, |
ba1389b7 | 150 | { "halt_exits", VCPU_STAT(halt_exits) }, |
f7819512 | 151 | { "halt_successful_poll", VCPU_STAT(halt_successful_poll) }, |
ba1389b7 | 152 | { "halt_wakeup", VCPU_STAT(halt_wakeup) }, |
f11c3a8d | 153 | { "hypercalls", VCPU_STAT(hypercalls) }, |
ba1389b7 AK |
154 | { "request_irq", VCPU_STAT(request_irq_exits) }, |
155 | { "irq_exits", VCPU_STAT(irq_exits) }, | |
156 | { "host_state_reload", VCPU_STAT(host_state_reload) }, | |
157 | { "efer_reload", VCPU_STAT(efer_reload) }, | |
158 | { "fpu_reload", VCPU_STAT(fpu_reload) }, | |
159 | { "insn_emulation", VCPU_STAT(insn_emulation) }, | |
160 | { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) }, | |
fa89a817 | 161 | { "irq_injections", VCPU_STAT(irq_injections) }, |
c4abb7c9 | 162 | { "nmi_injections", VCPU_STAT(nmi_injections) }, |
4cee5764 AK |
163 | { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) }, |
164 | { "mmu_pte_write", VM_STAT(mmu_pte_write) }, | |
165 | { "mmu_pte_updated", VM_STAT(mmu_pte_updated) }, | |
166 | { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) }, | |
167 | { "mmu_flooded", VM_STAT(mmu_flooded) }, | |
168 | { "mmu_recycled", VM_STAT(mmu_recycled) }, | |
dfc5aa00 | 169 | { "mmu_cache_miss", VM_STAT(mmu_cache_miss) }, |
4731d4c7 | 170 | { "mmu_unsync", VM_STAT(mmu_unsync) }, |
0f74a24c | 171 | { "remote_tlb_flush", VM_STAT(remote_tlb_flush) }, |
05da4558 | 172 | { "largepages", VM_STAT(lpages) }, |
417bc304 HB |
173 | { NULL } |
174 | }; | |
175 | ||
2acf923e DC |
176 | u64 __read_mostly host_xcr0; |
177 | ||
b6785def | 178 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt); |
d6aa1000 | 179 | |
af585b92 GN |
180 | static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu) |
181 | { | |
182 | int i; | |
183 | for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++) | |
184 | vcpu->arch.apf.gfns[i] = ~0; | |
185 | } | |
186 | ||
18863bdd AK |
187 | static void kvm_on_user_return(struct user_return_notifier *urn) |
188 | { | |
189 | unsigned slot; | |
18863bdd AK |
190 | struct kvm_shared_msrs *locals |
191 | = container_of(urn, struct kvm_shared_msrs, urn); | |
2bf78fa7 | 192 | struct kvm_shared_msr_values *values; |
18863bdd AK |
193 | |
194 | for (slot = 0; slot < shared_msrs_global.nr; ++slot) { | |
2bf78fa7 SY |
195 | values = &locals->values[slot]; |
196 | if (values->host != values->curr) { | |
197 | wrmsrl(shared_msrs_global.msrs[slot], values->host); | |
198 | values->curr = values->host; | |
18863bdd AK |
199 | } |
200 | } | |
201 | locals->registered = false; | |
202 | user_return_notifier_unregister(urn); | |
203 | } | |
204 | ||
2bf78fa7 | 205 | static void shared_msr_update(unsigned slot, u32 msr) |
18863bdd | 206 | { |
18863bdd | 207 | u64 value; |
013f6a5d MT |
208 | unsigned int cpu = smp_processor_id(); |
209 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
18863bdd | 210 | |
2bf78fa7 SY |
211 | /* only read, and nobody should modify it at this time, |
212 | * so don't need lock */ | |
213 | if (slot >= shared_msrs_global.nr) { | |
214 | printk(KERN_ERR "kvm: invalid MSR slot!"); | |
215 | return; | |
216 | } | |
217 | rdmsrl_safe(msr, &value); | |
218 | smsr->values[slot].host = value; | |
219 | smsr->values[slot].curr = value; | |
220 | } | |
221 | ||
222 | void kvm_define_shared_msr(unsigned slot, u32 msr) | |
223 | { | |
0123be42 | 224 | BUG_ON(slot >= KVM_NR_SHARED_MSRS); |
18863bdd AK |
225 | if (slot >= shared_msrs_global.nr) |
226 | shared_msrs_global.nr = slot + 1; | |
2bf78fa7 SY |
227 | shared_msrs_global.msrs[slot] = msr; |
228 | /* we need ensured the shared_msr_global have been updated */ | |
229 | smp_wmb(); | |
18863bdd AK |
230 | } |
231 | EXPORT_SYMBOL_GPL(kvm_define_shared_msr); | |
232 | ||
233 | static void kvm_shared_msr_cpu_online(void) | |
234 | { | |
235 | unsigned i; | |
18863bdd AK |
236 | |
237 | for (i = 0; i < shared_msrs_global.nr; ++i) | |
2bf78fa7 | 238 | shared_msr_update(i, shared_msrs_global.msrs[i]); |
18863bdd AK |
239 | } |
240 | ||
8b3c3104 | 241 | int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask) |
18863bdd | 242 | { |
013f6a5d MT |
243 | unsigned int cpu = smp_processor_id(); |
244 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
8b3c3104 | 245 | int err; |
18863bdd | 246 | |
2bf78fa7 | 247 | if (((value ^ smsr->values[slot].curr) & mask) == 0) |
8b3c3104 | 248 | return 0; |
2bf78fa7 | 249 | smsr->values[slot].curr = value; |
8b3c3104 AH |
250 | err = wrmsrl_safe(shared_msrs_global.msrs[slot], value); |
251 | if (err) | |
252 | return 1; | |
253 | ||
18863bdd AK |
254 | if (!smsr->registered) { |
255 | smsr->urn.on_user_return = kvm_on_user_return; | |
256 | user_return_notifier_register(&smsr->urn); | |
257 | smsr->registered = true; | |
258 | } | |
8b3c3104 | 259 | return 0; |
18863bdd AK |
260 | } |
261 | EXPORT_SYMBOL_GPL(kvm_set_shared_msr); | |
262 | ||
13a34e06 | 263 | static void drop_user_return_notifiers(void) |
3548bab5 | 264 | { |
013f6a5d MT |
265 | unsigned int cpu = smp_processor_id(); |
266 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
3548bab5 AK |
267 | |
268 | if (smsr->registered) | |
269 | kvm_on_user_return(&smsr->urn); | |
270 | } | |
271 | ||
6866b83e CO |
272 | u64 kvm_get_apic_base(struct kvm_vcpu *vcpu) |
273 | { | |
8a5a87d9 | 274 | return vcpu->arch.apic_base; |
6866b83e CO |
275 | } |
276 | EXPORT_SYMBOL_GPL(kvm_get_apic_base); | |
277 | ||
58cb628d JK |
278 | int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
279 | { | |
280 | u64 old_state = vcpu->arch.apic_base & | |
281 | (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE); | |
282 | u64 new_state = msr_info->data & | |
283 | (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE); | |
284 | u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | | |
285 | 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE); | |
286 | ||
287 | if (!msr_info->host_initiated && | |
288 | ((msr_info->data & reserved_bits) != 0 || | |
289 | new_state == X2APIC_ENABLE || | |
290 | (new_state == MSR_IA32_APICBASE_ENABLE && | |
291 | old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) || | |
292 | (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) && | |
293 | old_state == 0))) | |
294 | return 1; | |
295 | ||
296 | kvm_lapic_set_base(vcpu, msr_info->data); | |
297 | return 0; | |
6866b83e CO |
298 | } |
299 | EXPORT_SYMBOL_GPL(kvm_set_apic_base); | |
300 | ||
2605fc21 | 301 | asmlinkage __visible void kvm_spurious_fault(void) |
e3ba45b8 GL |
302 | { |
303 | /* Fault while not rebooting. We want the trace. */ | |
304 | BUG(); | |
305 | } | |
306 | EXPORT_SYMBOL_GPL(kvm_spurious_fault); | |
307 | ||
3fd28fce ED |
308 | #define EXCPT_BENIGN 0 |
309 | #define EXCPT_CONTRIBUTORY 1 | |
310 | #define EXCPT_PF 2 | |
311 | ||
312 | static int exception_class(int vector) | |
313 | { | |
314 | switch (vector) { | |
315 | case PF_VECTOR: | |
316 | return EXCPT_PF; | |
317 | case DE_VECTOR: | |
318 | case TS_VECTOR: | |
319 | case NP_VECTOR: | |
320 | case SS_VECTOR: | |
321 | case GP_VECTOR: | |
322 | return EXCPT_CONTRIBUTORY; | |
323 | default: | |
324 | break; | |
325 | } | |
326 | return EXCPT_BENIGN; | |
327 | } | |
328 | ||
d6e8c854 NA |
329 | #define EXCPT_FAULT 0 |
330 | #define EXCPT_TRAP 1 | |
331 | #define EXCPT_ABORT 2 | |
332 | #define EXCPT_INTERRUPT 3 | |
333 | ||
334 | static int exception_type(int vector) | |
335 | { | |
336 | unsigned int mask; | |
337 | ||
338 | if (WARN_ON(vector > 31 || vector == NMI_VECTOR)) | |
339 | return EXCPT_INTERRUPT; | |
340 | ||
341 | mask = 1 << vector; | |
342 | ||
343 | /* #DB is trap, as instruction watchpoints are handled elsewhere */ | |
344 | if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR))) | |
345 | return EXCPT_TRAP; | |
346 | ||
347 | if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR))) | |
348 | return EXCPT_ABORT; | |
349 | ||
350 | /* Reserved exceptions will result in fault */ | |
351 | return EXCPT_FAULT; | |
352 | } | |
353 | ||
3fd28fce | 354 | static void kvm_multiple_exception(struct kvm_vcpu *vcpu, |
ce7ddec4 JR |
355 | unsigned nr, bool has_error, u32 error_code, |
356 | bool reinject) | |
3fd28fce ED |
357 | { |
358 | u32 prev_nr; | |
359 | int class1, class2; | |
360 | ||
3842d135 AK |
361 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
362 | ||
3fd28fce ED |
363 | if (!vcpu->arch.exception.pending) { |
364 | queue: | |
3ffb2468 NA |
365 | if (has_error && !is_protmode(vcpu)) |
366 | has_error = false; | |
3fd28fce ED |
367 | vcpu->arch.exception.pending = true; |
368 | vcpu->arch.exception.has_error_code = has_error; | |
369 | vcpu->arch.exception.nr = nr; | |
370 | vcpu->arch.exception.error_code = error_code; | |
3f0fd292 | 371 | vcpu->arch.exception.reinject = reinject; |
3fd28fce ED |
372 | return; |
373 | } | |
374 | ||
375 | /* to check exception */ | |
376 | prev_nr = vcpu->arch.exception.nr; | |
377 | if (prev_nr == DF_VECTOR) { | |
378 | /* triple fault -> shutdown */ | |
a8eeb04a | 379 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
3fd28fce ED |
380 | return; |
381 | } | |
382 | class1 = exception_class(prev_nr); | |
383 | class2 = exception_class(nr); | |
384 | if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY) | |
385 | || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) { | |
386 | /* generate double fault per SDM Table 5-5 */ | |
387 | vcpu->arch.exception.pending = true; | |
388 | vcpu->arch.exception.has_error_code = true; | |
389 | vcpu->arch.exception.nr = DF_VECTOR; | |
390 | vcpu->arch.exception.error_code = 0; | |
391 | } else | |
392 | /* replace previous exception with a new one in a hope | |
393 | that instruction re-execution will regenerate lost | |
394 | exception */ | |
395 | goto queue; | |
396 | } | |
397 | ||
298101da AK |
398 | void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
399 | { | |
ce7ddec4 | 400 | kvm_multiple_exception(vcpu, nr, false, 0, false); |
298101da AK |
401 | } |
402 | EXPORT_SYMBOL_GPL(kvm_queue_exception); | |
403 | ||
ce7ddec4 JR |
404 | void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
405 | { | |
406 | kvm_multiple_exception(vcpu, nr, false, 0, true); | |
407 | } | |
408 | EXPORT_SYMBOL_GPL(kvm_requeue_exception); | |
409 | ||
db8fcefa | 410 | void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err) |
c3c91fee | 411 | { |
db8fcefa AP |
412 | if (err) |
413 | kvm_inject_gp(vcpu, 0); | |
414 | else | |
415 | kvm_x86_ops->skip_emulated_instruction(vcpu); | |
416 | } | |
417 | EXPORT_SYMBOL_GPL(kvm_complete_insn_gp); | |
8df25a32 | 418 | |
6389ee94 | 419 | void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault) |
c3c91fee AK |
420 | { |
421 | ++vcpu->stat.pf_guest; | |
6389ee94 AK |
422 | vcpu->arch.cr2 = fault->address; |
423 | kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code); | |
c3c91fee | 424 | } |
27d6c865 | 425 | EXPORT_SYMBOL_GPL(kvm_inject_page_fault); |
c3c91fee | 426 | |
ef54bcfe | 427 | static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault) |
d4f8cf66 | 428 | { |
6389ee94 AK |
429 | if (mmu_is_nested(vcpu) && !fault->nested_page_fault) |
430 | vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault); | |
d4f8cf66 | 431 | else |
6389ee94 | 432 | vcpu->arch.mmu.inject_page_fault(vcpu, fault); |
ef54bcfe PB |
433 | |
434 | return fault->nested_page_fault; | |
d4f8cf66 JR |
435 | } |
436 | ||
3419ffc8 SY |
437 | void kvm_inject_nmi(struct kvm_vcpu *vcpu) |
438 | { | |
7460fb4a AK |
439 | atomic_inc(&vcpu->arch.nmi_queued); |
440 | kvm_make_request(KVM_REQ_NMI, vcpu); | |
3419ffc8 SY |
441 | } |
442 | EXPORT_SYMBOL_GPL(kvm_inject_nmi); | |
443 | ||
298101da AK |
444 | void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
445 | { | |
ce7ddec4 | 446 | kvm_multiple_exception(vcpu, nr, true, error_code, false); |
298101da AK |
447 | } |
448 | EXPORT_SYMBOL_GPL(kvm_queue_exception_e); | |
449 | ||
ce7ddec4 JR |
450 | void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
451 | { | |
452 | kvm_multiple_exception(vcpu, nr, true, error_code, true); | |
453 | } | |
454 | EXPORT_SYMBOL_GPL(kvm_requeue_exception_e); | |
455 | ||
0a79b009 AK |
456 | /* |
457 | * Checks if cpl <= required_cpl; if true, return true. Otherwise queue | |
458 | * a #GP and return false. | |
459 | */ | |
460 | bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl) | |
298101da | 461 | { |
0a79b009 AK |
462 | if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl) |
463 | return true; | |
464 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
465 | return false; | |
298101da | 466 | } |
0a79b009 | 467 | EXPORT_SYMBOL_GPL(kvm_require_cpl); |
298101da | 468 | |
16f8a6f9 NA |
469 | bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr) |
470 | { | |
471 | if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE)) | |
472 | return true; | |
473 | ||
474 | kvm_queue_exception(vcpu, UD_VECTOR); | |
475 | return false; | |
476 | } | |
477 | EXPORT_SYMBOL_GPL(kvm_require_dr); | |
478 | ||
ec92fe44 JR |
479 | /* |
480 | * This function will be used to read from the physical memory of the currently | |
54bf36aa | 481 | * running guest. The difference to kvm_vcpu_read_guest_page is that this function |
ec92fe44 JR |
482 | * can read from guest physical or from the guest's guest physical memory. |
483 | */ | |
484 | int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, | |
485 | gfn_t ngfn, void *data, int offset, int len, | |
486 | u32 access) | |
487 | { | |
54987b7a | 488 | struct x86_exception exception; |
ec92fe44 JR |
489 | gfn_t real_gfn; |
490 | gpa_t ngpa; | |
491 | ||
492 | ngpa = gfn_to_gpa(ngfn); | |
54987b7a | 493 | real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception); |
ec92fe44 JR |
494 | if (real_gfn == UNMAPPED_GVA) |
495 | return -EFAULT; | |
496 | ||
497 | real_gfn = gpa_to_gfn(real_gfn); | |
498 | ||
54bf36aa | 499 | return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len); |
ec92fe44 JR |
500 | } |
501 | EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu); | |
502 | ||
69b0049a | 503 | static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
3d06b8bf JR |
504 | void *data, int offset, int len, u32 access) |
505 | { | |
506 | return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn, | |
507 | data, offset, len, access); | |
508 | } | |
509 | ||
a03490ed CO |
510 | /* |
511 | * Load the pae pdptrs. Return true is they are all valid. | |
512 | */ | |
ff03a073 | 513 | int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3) |
a03490ed CO |
514 | { |
515 | gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT; | |
516 | unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2; | |
517 | int i; | |
518 | int ret; | |
ff03a073 | 519 | u64 pdpte[ARRAY_SIZE(mmu->pdptrs)]; |
a03490ed | 520 | |
ff03a073 JR |
521 | ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte, |
522 | offset * sizeof(u64), sizeof(pdpte), | |
523 | PFERR_USER_MASK|PFERR_WRITE_MASK); | |
a03490ed CO |
524 | if (ret < 0) { |
525 | ret = 0; | |
526 | goto out; | |
527 | } | |
528 | for (i = 0; i < ARRAY_SIZE(pdpte); ++i) { | |
43a3795a | 529 | if (is_present_gpte(pdpte[i]) && |
20c466b5 | 530 | (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) { |
a03490ed CO |
531 | ret = 0; |
532 | goto out; | |
533 | } | |
534 | } | |
535 | ret = 1; | |
536 | ||
ff03a073 | 537 | memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)); |
6de4f3ad AK |
538 | __set_bit(VCPU_EXREG_PDPTR, |
539 | (unsigned long *)&vcpu->arch.regs_avail); | |
540 | __set_bit(VCPU_EXREG_PDPTR, | |
541 | (unsigned long *)&vcpu->arch.regs_dirty); | |
a03490ed | 542 | out: |
a03490ed CO |
543 | |
544 | return ret; | |
545 | } | |
cc4b6871 | 546 | EXPORT_SYMBOL_GPL(load_pdptrs); |
a03490ed | 547 | |
d835dfec AK |
548 | static bool pdptrs_changed(struct kvm_vcpu *vcpu) |
549 | { | |
ff03a073 | 550 | u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)]; |
d835dfec | 551 | bool changed = true; |
3d06b8bf JR |
552 | int offset; |
553 | gfn_t gfn; | |
d835dfec AK |
554 | int r; |
555 | ||
556 | if (is_long_mode(vcpu) || !is_pae(vcpu)) | |
557 | return false; | |
558 | ||
6de4f3ad AK |
559 | if (!test_bit(VCPU_EXREG_PDPTR, |
560 | (unsigned long *)&vcpu->arch.regs_avail)) | |
561 | return true; | |
562 | ||
9f8fe504 AK |
563 | gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT; |
564 | offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1); | |
3d06b8bf JR |
565 | r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte), |
566 | PFERR_USER_MASK | PFERR_WRITE_MASK); | |
d835dfec AK |
567 | if (r < 0) |
568 | goto out; | |
ff03a073 | 569 | changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0; |
d835dfec | 570 | out: |
d835dfec AK |
571 | |
572 | return changed; | |
573 | } | |
574 | ||
49a9b07e | 575 | int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
a03490ed | 576 | { |
aad82703 | 577 | unsigned long old_cr0 = kvm_read_cr0(vcpu); |
d81135a5 | 578 | unsigned long update_bits = X86_CR0_PG | X86_CR0_WP; |
aad82703 | 579 | |
f9a48e6a AK |
580 | cr0 |= X86_CR0_ET; |
581 | ||
ab344828 | 582 | #ifdef CONFIG_X86_64 |
0f12244f GN |
583 | if (cr0 & 0xffffffff00000000UL) |
584 | return 1; | |
ab344828 GN |
585 | #endif |
586 | ||
587 | cr0 &= ~CR0_RESERVED_BITS; | |
a03490ed | 588 | |
0f12244f GN |
589 | if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) |
590 | return 1; | |
a03490ed | 591 | |
0f12244f GN |
592 | if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) |
593 | return 1; | |
a03490ed CO |
594 | |
595 | if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) { | |
596 | #ifdef CONFIG_X86_64 | |
f6801dff | 597 | if ((vcpu->arch.efer & EFER_LME)) { |
a03490ed CO |
598 | int cs_db, cs_l; |
599 | ||
0f12244f GN |
600 | if (!is_pae(vcpu)) |
601 | return 1; | |
a03490ed | 602 | kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l); |
0f12244f GN |
603 | if (cs_l) |
604 | return 1; | |
a03490ed CO |
605 | } else |
606 | #endif | |
ff03a073 | 607 | if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, |
9f8fe504 | 608 | kvm_read_cr3(vcpu))) |
0f12244f | 609 | return 1; |
a03490ed CO |
610 | } |
611 | ||
ad756a16 MJ |
612 | if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) |
613 | return 1; | |
614 | ||
a03490ed | 615 | kvm_x86_ops->set_cr0(vcpu, cr0); |
a03490ed | 616 | |
d170c419 | 617 | if ((cr0 ^ old_cr0) & X86_CR0_PG) { |
e5f3f027 | 618 | kvm_clear_async_pf_completion_queue(vcpu); |
d170c419 LJ |
619 | kvm_async_pf_hash_reset(vcpu); |
620 | } | |
e5f3f027 | 621 | |
aad82703 SY |
622 | if ((cr0 ^ old_cr0) & update_bits) |
623 | kvm_mmu_reset_context(vcpu); | |
b18d5431 XG |
624 | |
625 | if ((cr0 ^ old_cr0) & X86_CR0_CD) | |
626 | kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL); | |
627 | ||
0f12244f GN |
628 | return 0; |
629 | } | |
2d3ad1f4 | 630 | EXPORT_SYMBOL_GPL(kvm_set_cr0); |
a03490ed | 631 | |
2d3ad1f4 | 632 | void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw) |
a03490ed | 633 | { |
49a9b07e | 634 | (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f)); |
a03490ed | 635 | } |
2d3ad1f4 | 636 | EXPORT_SYMBOL_GPL(kvm_lmsw); |
a03490ed | 637 | |
42bdf991 MT |
638 | static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu) |
639 | { | |
640 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) && | |
641 | !vcpu->guest_xcr0_loaded) { | |
642 | /* kvm_set_xcr() also depends on this */ | |
643 | xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0); | |
644 | vcpu->guest_xcr0_loaded = 1; | |
645 | } | |
646 | } | |
647 | ||
648 | static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu) | |
649 | { | |
650 | if (vcpu->guest_xcr0_loaded) { | |
651 | if (vcpu->arch.xcr0 != host_xcr0) | |
652 | xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0); | |
653 | vcpu->guest_xcr0_loaded = 0; | |
654 | } | |
655 | } | |
656 | ||
69b0049a | 657 | static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) |
2acf923e | 658 | { |
56c103ec LJ |
659 | u64 xcr0 = xcr; |
660 | u64 old_xcr0 = vcpu->arch.xcr0; | |
46c34cb0 | 661 | u64 valid_bits; |
2acf923e DC |
662 | |
663 | /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */ | |
664 | if (index != XCR_XFEATURE_ENABLED_MASK) | |
665 | return 1; | |
2acf923e DC |
666 | if (!(xcr0 & XSTATE_FP)) |
667 | return 1; | |
668 | if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE)) | |
669 | return 1; | |
46c34cb0 PB |
670 | |
671 | /* | |
672 | * Do not allow the guest to set bits that we do not support | |
673 | * saving. However, xcr0 bit 0 is always set, even if the | |
674 | * emulated CPU does not support XSAVE (see fx_init). | |
675 | */ | |
676 | valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP; | |
677 | if (xcr0 & ~valid_bits) | |
2acf923e | 678 | return 1; |
46c34cb0 | 679 | |
390bd528 LJ |
680 | if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR))) |
681 | return 1; | |
682 | ||
612263b3 CP |
683 | if (xcr0 & XSTATE_AVX512) { |
684 | if (!(xcr0 & XSTATE_YMM)) | |
685 | return 1; | |
686 | if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512) | |
687 | return 1; | |
688 | } | |
42bdf991 | 689 | kvm_put_guest_xcr0(vcpu); |
2acf923e | 690 | vcpu->arch.xcr0 = xcr0; |
56c103ec LJ |
691 | |
692 | if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK) | |
693 | kvm_update_cpuid(vcpu); | |
2acf923e DC |
694 | return 0; |
695 | } | |
696 | ||
697 | int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) | |
698 | { | |
764bcbc5 Z |
699 | if (kvm_x86_ops->get_cpl(vcpu) != 0 || |
700 | __kvm_set_xcr(vcpu, index, xcr)) { | |
2acf923e DC |
701 | kvm_inject_gp(vcpu, 0); |
702 | return 1; | |
703 | } | |
704 | return 0; | |
705 | } | |
706 | EXPORT_SYMBOL_GPL(kvm_set_xcr); | |
707 | ||
a83b29c6 | 708 | int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) |
a03490ed | 709 | { |
fc78f519 | 710 | unsigned long old_cr4 = kvm_read_cr4(vcpu); |
0be0226f XG |
711 | unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE | |
712 | X86_CR4_SMEP | X86_CR4_SMAP; | |
713 | ||
0f12244f GN |
714 | if (cr4 & CR4_RESERVED_BITS) |
715 | return 1; | |
a03490ed | 716 | |
2acf923e DC |
717 | if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE)) |
718 | return 1; | |
719 | ||
c68b734f YW |
720 | if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP)) |
721 | return 1; | |
722 | ||
97ec8c06 FW |
723 | if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP)) |
724 | return 1; | |
725 | ||
afcbf13f | 726 | if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE)) |
74dc2b4f YW |
727 | return 1; |
728 | ||
a03490ed | 729 | if (is_long_mode(vcpu)) { |
0f12244f GN |
730 | if (!(cr4 & X86_CR4_PAE)) |
731 | return 1; | |
a2edf57f AK |
732 | } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE) |
733 | && ((cr4 ^ old_cr4) & pdptr_bits) | |
9f8fe504 AK |
734 | && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, |
735 | kvm_read_cr3(vcpu))) | |
0f12244f GN |
736 | return 1; |
737 | ||
ad756a16 MJ |
738 | if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) { |
739 | if (!guest_cpuid_has_pcid(vcpu)) | |
740 | return 1; | |
741 | ||
742 | /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */ | |
743 | if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu)) | |
744 | return 1; | |
745 | } | |
746 | ||
5e1746d6 | 747 | if (kvm_x86_ops->set_cr4(vcpu, cr4)) |
0f12244f | 748 | return 1; |
a03490ed | 749 | |
ad756a16 MJ |
750 | if (((cr4 ^ old_cr4) & pdptr_bits) || |
751 | (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE))) | |
aad82703 | 752 | kvm_mmu_reset_context(vcpu); |
0f12244f | 753 | |
2acf923e | 754 | if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE) |
00b27a3e | 755 | kvm_update_cpuid(vcpu); |
2acf923e | 756 | |
0f12244f GN |
757 | return 0; |
758 | } | |
2d3ad1f4 | 759 | EXPORT_SYMBOL_GPL(kvm_set_cr4); |
a03490ed | 760 | |
2390218b | 761 | int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3) |
a03490ed | 762 | { |
ac146235 | 763 | #ifdef CONFIG_X86_64 |
9d88fca7 | 764 | cr3 &= ~CR3_PCID_INVD; |
ac146235 | 765 | #endif |
9d88fca7 | 766 | |
9f8fe504 | 767 | if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) { |
0ba73cda | 768 | kvm_mmu_sync_roots(vcpu); |
77c3913b | 769 | kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
0f12244f | 770 | return 0; |
d835dfec AK |
771 | } |
772 | ||
a03490ed | 773 | if (is_long_mode(vcpu)) { |
d9f89b88 JK |
774 | if (cr3 & CR3_L_MODE_RESERVED_BITS) |
775 | return 1; | |
776 | } else if (is_pae(vcpu) && is_paging(vcpu) && | |
777 | !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) | |
346874c9 | 778 | return 1; |
a03490ed | 779 | |
0f12244f | 780 | vcpu->arch.cr3 = cr3; |
aff48baa | 781 | __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail); |
d8d173da | 782 | kvm_mmu_new_cr3(vcpu); |
0f12244f GN |
783 | return 0; |
784 | } | |
2d3ad1f4 | 785 | EXPORT_SYMBOL_GPL(kvm_set_cr3); |
a03490ed | 786 | |
eea1cff9 | 787 | int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8) |
a03490ed | 788 | { |
0f12244f GN |
789 | if (cr8 & CR8_RESERVED_BITS) |
790 | return 1; | |
a03490ed CO |
791 | if (irqchip_in_kernel(vcpu->kvm)) |
792 | kvm_lapic_set_tpr(vcpu, cr8); | |
793 | else | |
ad312c7c | 794 | vcpu->arch.cr8 = cr8; |
0f12244f GN |
795 | return 0; |
796 | } | |
2d3ad1f4 | 797 | EXPORT_SYMBOL_GPL(kvm_set_cr8); |
a03490ed | 798 | |
2d3ad1f4 | 799 | unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu) |
a03490ed CO |
800 | { |
801 | if (irqchip_in_kernel(vcpu->kvm)) | |
802 | return kvm_lapic_get_cr8(vcpu); | |
803 | else | |
ad312c7c | 804 | return vcpu->arch.cr8; |
a03490ed | 805 | } |
2d3ad1f4 | 806 | EXPORT_SYMBOL_GPL(kvm_get_cr8); |
a03490ed | 807 | |
ae561ede NA |
808 | static void kvm_update_dr0123(struct kvm_vcpu *vcpu) |
809 | { | |
810 | int i; | |
811 | ||
812 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) { | |
813 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
814 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
815 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD; | |
816 | } | |
817 | } | |
818 | ||
73aaf249 JK |
819 | static void kvm_update_dr6(struct kvm_vcpu *vcpu) |
820 | { | |
821 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) | |
822 | kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6); | |
823 | } | |
824 | ||
c8639010 JK |
825 | static void kvm_update_dr7(struct kvm_vcpu *vcpu) |
826 | { | |
827 | unsigned long dr7; | |
828 | ||
829 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) | |
830 | dr7 = vcpu->arch.guest_debug_dr7; | |
831 | else | |
832 | dr7 = vcpu->arch.dr7; | |
833 | kvm_x86_ops->set_dr7(vcpu, dr7); | |
360b948d PB |
834 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED; |
835 | if (dr7 & DR7_BP_EN_MASK) | |
836 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED; | |
c8639010 JK |
837 | } |
838 | ||
6f43ed01 NA |
839 | static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu) |
840 | { | |
841 | u64 fixed = DR6_FIXED_1; | |
842 | ||
843 | if (!guest_cpuid_has_rtm(vcpu)) | |
844 | fixed |= DR6_RTM; | |
845 | return fixed; | |
846 | } | |
847 | ||
338dbc97 | 848 | static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) |
020df079 GN |
849 | { |
850 | switch (dr) { | |
851 | case 0 ... 3: | |
852 | vcpu->arch.db[dr] = val; | |
853 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) | |
854 | vcpu->arch.eff_db[dr] = val; | |
855 | break; | |
856 | case 4: | |
020df079 GN |
857 | /* fall through */ |
858 | case 6: | |
338dbc97 GN |
859 | if (val & 0xffffffff00000000ULL) |
860 | return -1; /* #GP */ | |
6f43ed01 | 861 | vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu); |
73aaf249 | 862 | kvm_update_dr6(vcpu); |
020df079 GN |
863 | break; |
864 | case 5: | |
020df079 GN |
865 | /* fall through */ |
866 | default: /* 7 */ | |
338dbc97 GN |
867 | if (val & 0xffffffff00000000ULL) |
868 | return -1; /* #GP */ | |
020df079 | 869 | vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1; |
c8639010 | 870 | kvm_update_dr7(vcpu); |
020df079 GN |
871 | break; |
872 | } | |
873 | ||
874 | return 0; | |
875 | } | |
338dbc97 GN |
876 | |
877 | int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) | |
878 | { | |
16f8a6f9 | 879 | if (__kvm_set_dr(vcpu, dr, val)) { |
338dbc97 | 880 | kvm_inject_gp(vcpu, 0); |
16f8a6f9 NA |
881 | return 1; |
882 | } | |
883 | return 0; | |
338dbc97 | 884 | } |
020df079 GN |
885 | EXPORT_SYMBOL_GPL(kvm_set_dr); |
886 | ||
16f8a6f9 | 887 | int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val) |
020df079 GN |
888 | { |
889 | switch (dr) { | |
890 | case 0 ... 3: | |
891 | *val = vcpu->arch.db[dr]; | |
892 | break; | |
893 | case 4: | |
020df079 GN |
894 | /* fall through */ |
895 | case 6: | |
73aaf249 JK |
896 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) |
897 | *val = vcpu->arch.dr6; | |
898 | else | |
899 | *val = kvm_x86_ops->get_dr6(vcpu); | |
020df079 GN |
900 | break; |
901 | case 5: | |
020df079 GN |
902 | /* fall through */ |
903 | default: /* 7 */ | |
904 | *val = vcpu->arch.dr7; | |
905 | break; | |
906 | } | |
338dbc97 GN |
907 | return 0; |
908 | } | |
020df079 GN |
909 | EXPORT_SYMBOL_GPL(kvm_get_dr); |
910 | ||
022cd0e8 AK |
911 | bool kvm_rdpmc(struct kvm_vcpu *vcpu) |
912 | { | |
913 | u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
914 | u64 data; | |
915 | int err; | |
916 | ||
c6702c9d | 917 | err = kvm_pmu_rdpmc(vcpu, ecx, &data); |
022cd0e8 AK |
918 | if (err) |
919 | return err; | |
920 | kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data); | |
921 | kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32); | |
922 | return err; | |
923 | } | |
924 | EXPORT_SYMBOL_GPL(kvm_rdpmc); | |
925 | ||
043405e1 CO |
926 | /* |
927 | * List of msr numbers which we expose to userspace through KVM_GET_MSRS | |
928 | * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST. | |
929 | * | |
930 | * This list is modified at module load time to reflect the | |
e3267cbb | 931 | * capabilities of the host cpu. This capabilities test skips MSRs that are |
62ef68bb PB |
932 | * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs |
933 | * may depend on host virtualization features rather than host cpu features. | |
043405e1 | 934 | */ |
e3267cbb | 935 | |
043405e1 CO |
936 | static u32 msrs_to_save[] = { |
937 | MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, | |
8c06585d | 938 | MSR_STAR, |
043405e1 CO |
939 | #ifdef CONFIG_X86_64 |
940 | MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR, | |
941 | #endif | |
b3897a49 | 942 | MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA, |
0dd376e7 | 943 | MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS |
043405e1 CO |
944 | }; |
945 | ||
946 | static unsigned num_msrs_to_save; | |
947 | ||
62ef68bb PB |
948 | static u32 emulated_msrs[] = { |
949 | MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK, | |
950 | MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW, | |
951 | HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL, | |
952 | HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC, | |
953 | HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME, | |
954 | MSR_KVM_PV_EOI_EN, | |
955 | ||
ba904635 | 956 | MSR_IA32_TSC_ADJUST, |
a3e06bbe | 957 | MSR_IA32_TSCDEADLINE, |
043405e1 | 958 | MSR_IA32_MISC_ENABLE, |
908e75f3 AK |
959 | MSR_IA32_MCG_STATUS, |
960 | MSR_IA32_MCG_CTL, | |
64d60670 | 961 | MSR_IA32_SMBASE, |
043405e1 CO |
962 | }; |
963 | ||
62ef68bb PB |
964 | static unsigned num_emulated_msrs; |
965 | ||
384bb783 | 966 | bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer) |
15c4a640 | 967 | { |
b69e8cae | 968 | if (efer & efer_reserved_bits) |
384bb783 | 969 | return false; |
15c4a640 | 970 | |
1b2fd70c AG |
971 | if (efer & EFER_FFXSR) { |
972 | struct kvm_cpuid_entry2 *feat; | |
973 | ||
974 | feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0); | |
b69e8cae | 975 | if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) |
384bb783 | 976 | return false; |
1b2fd70c AG |
977 | } |
978 | ||
d8017474 AG |
979 | if (efer & EFER_SVME) { |
980 | struct kvm_cpuid_entry2 *feat; | |
981 | ||
982 | feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0); | |
b69e8cae | 983 | if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) |
384bb783 | 984 | return false; |
d8017474 AG |
985 | } |
986 | ||
384bb783 JK |
987 | return true; |
988 | } | |
989 | EXPORT_SYMBOL_GPL(kvm_valid_efer); | |
990 | ||
991 | static int set_efer(struct kvm_vcpu *vcpu, u64 efer) | |
992 | { | |
993 | u64 old_efer = vcpu->arch.efer; | |
994 | ||
995 | if (!kvm_valid_efer(vcpu, efer)) | |
996 | return 1; | |
997 | ||
998 | if (is_paging(vcpu) | |
999 | && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) | |
1000 | return 1; | |
1001 | ||
15c4a640 | 1002 | efer &= ~EFER_LMA; |
f6801dff | 1003 | efer |= vcpu->arch.efer & EFER_LMA; |
15c4a640 | 1004 | |
a3d204e2 SY |
1005 | kvm_x86_ops->set_efer(vcpu, efer); |
1006 | ||
aad82703 SY |
1007 | /* Update reserved bits */ |
1008 | if ((efer ^ old_efer) & EFER_NX) | |
1009 | kvm_mmu_reset_context(vcpu); | |
1010 | ||
b69e8cae | 1011 | return 0; |
15c4a640 CO |
1012 | } |
1013 | ||
f2b4b7dd JR |
1014 | void kvm_enable_efer_bits(u64 mask) |
1015 | { | |
1016 | efer_reserved_bits &= ~mask; | |
1017 | } | |
1018 | EXPORT_SYMBOL_GPL(kvm_enable_efer_bits); | |
1019 | ||
15c4a640 CO |
1020 | /* |
1021 | * Writes msr value into into the appropriate "register". | |
1022 | * Returns 0 on success, non-0 otherwise. | |
1023 | * Assumes vcpu_load() was already called. | |
1024 | */ | |
8fe8ab46 | 1025 | int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr) |
15c4a640 | 1026 | { |
854e8bb1 NA |
1027 | switch (msr->index) { |
1028 | case MSR_FS_BASE: | |
1029 | case MSR_GS_BASE: | |
1030 | case MSR_KERNEL_GS_BASE: | |
1031 | case MSR_CSTAR: | |
1032 | case MSR_LSTAR: | |
1033 | if (is_noncanonical_address(msr->data)) | |
1034 | return 1; | |
1035 | break; | |
1036 | case MSR_IA32_SYSENTER_EIP: | |
1037 | case MSR_IA32_SYSENTER_ESP: | |
1038 | /* | |
1039 | * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if | |
1040 | * non-canonical address is written on Intel but not on | |
1041 | * AMD (which ignores the top 32-bits, because it does | |
1042 | * not implement 64-bit SYSENTER). | |
1043 | * | |
1044 | * 64-bit code should hence be able to write a non-canonical | |
1045 | * value on AMD. Making the address canonical ensures that | |
1046 | * vmentry does not fail on Intel after writing a non-canonical | |
1047 | * value, and that something deterministic happens if the guest | |
1048 | * invokes 64-bit SYSENTER. | |
1049 | */ | |
1050 | msr->data = get_canonical(msr->data); | |
1051 | } | |
8fe8ab46 | 1052 | return kvm_x86_ops->set_msr(vcpu, msr); |
15c4a640 | 1053 | } |
854e8bb1 | 1054 | EXPORT_SYMBOL_GPL(kvm_set_msr); |
15c4a640 | 1055 | |
313a3dc7 CO |
1056 | /* |
1057 | * Adapt set_msr() to msr_io()'s calling convention | |
1058 | */ | |
609e36d3 PB |
1059 | static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) |
1060 | { | |
1061 | struct msr_data msr; | |
1062 | int r; | |
1063 | ||
1064 | msr.index = index; | |
1065 | msr.host_initiated = true; | |
1066 | r = kvm_get_msr(vcpu, &msr); | |
1067 | if (r) | |
1068 | return r; | |
1069 | ||
1070 | *data = msr.data; | |
1071 | return 0; | |
1072 | } | |
1073 | ||
313a3dc7 CO |
1074 | static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) |
1075 | { | |
8fe8ab46 WA |
1076 | struct msr_data msr; |
1077 | ||
1078 | msr.data = *data; | |
1079 | msr.index = index; | |
1080 | msr.host_initiated = true; | |
1081 | return kvm_set_msr(vcpu, &msr); | |
313a3dc7 CO |
1082 | } |
1083 | ||
16e8d74d MT |
1084 | #ifdef CONFIG_X86_64 |
1085 | struct pvclock_gtod_data { | |
1086 | seqcount_t seq; | |
1087 | ||
1088 | struct { /* extract of a clocksource struct */ | |
1089 | int vclock_mode; | |
1090 | cycle_t cycle_last; | |
1091 | cycle_t mask; | |
1092 | u32 mult; | |
1093 | u32 shift; | |
1094 | } clock; | |
1095 | ||
cbcf2dd3 TG |
1096 | u64 boot_ns; |
1097 | u64 nsec_base; | |
16e8d74d MT |
1098 | }; |
1099 | ||
1100 | static struct pvclock_gtod_data pvclock_gtod_data; | |
1101 | ||
1102 | static void update_pvclock_gtod(struct timekeeper *tk) | |
1103 | { | |
1104 | struct pvclock_gtod_data *vdata = &pvclock_gtod_data; | |
cbcf2dd3 TG |
1105 | u64 boot_ns; |
1106 | ||
876e7881 | 1107 | boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot)); |
16e8d74d MT |
1108 | |
1109 | write_seqcount_begin(&vdata->seq); | |
1110 | ||
1111 | /* copy pvclock gtod data */ | |
876e7881 PZ |
1112 | vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode; |
1113 | vdata->clock.cycle_last = tk->tkr_mono.cycle_last; | |
1114 | vdata->clock.mask = tk->tkr_mono.mask; | |
1115 | vdata->clock.mult = tk->tkr_mono.mult; | |
1116 | vdata->clock.shift = tk->tkr_mono.shift; | |
16e8d74d | 1117 | |
cbcf2dd3 | 1118 | vdata->boot_ns = boot_ns; |
876e7881 | 1119 | vdata->nsec_base = tk->tkr_mono.xtime_nsec; |
16e8d74d MT |
1120 | |
1121 | write_seqcount_end(&vdata->seq); | |
1122 | } | |
1123 | #endif | |
1124 | ||
bab5bb39 NK |
1125 | void kvm_set_pending_timer(struct kvm_vcpu *vcpu) |
1126 | { | |
1127 | /* | |
1128 | * Note: KVM_REQ_PENDING_TIMER is implicitly checked in | |
1129 | * vcpu_enter_guest. This function is only called from | |
1130 | * the physical CPU that is running vcpu. | |
1131 | */ | |
1132 | kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu); | |
1133 | } | |
16e8d74d | 1134 | |
18068523 GOC |
1135 | static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock) |
1136 | { | |
9ed3c444 AK |
1137 | int version; |
1138 | int r; | |
50d0a0f9 | 1139 | struct pvclock_wall_clock wc; |
923de3cf | 1140 | struct timespec boot; |
18068523 GOC |
1141 | |
1142 | if (!wall_clock) | |
1143 | return; | |
1144 | ||
9ed3c444 AK |
1145 | r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version)); |
1146 | if (r) | |
1147 | return; | |
1148 | ||
1149 | if (version & 1) | |
1150 | ++version; /* first time write, random junk */ | |
1151 | ||
1152 | ++version; | |
18068523 | 1153 | |
18068523 GOC |
1154 | kvm_write_guest(kvm, wall_clock, &version, sizeof(version)); |
1155 | ||
50d0a0f9 GH |
1156 | /* |
1157 | * The guest calculates current wall clock time by adding | |
34c238a1 | 1158 | * system time (updated by kvm_guest_time_update below) to the |
50d0a0f9 GH |
1159 | * wall clock specified here. guest system time equals host |
1160 | * system time for us, thus we must fill in host boot time here. | |
1161 | */ | |
923de3cf | 1162 | getboottime(&boot); |
50d0a0f9 | 1163 | |
4b648665 BR |
1164 | if (kvm->arch.kvmclock_offset) { |
1165 | struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset); | |
1166 | boot = timespec_sub(boot, ts); | |
1167 | } | |
50d0a0f9 GH |
1168 | wc.sec = boot.tv_sec; |
1169 | wc.nsec = boot.tv_nsec; | |
1170 | wc.version = version; | |
18068523 GOC |
1171 | |
1172 | kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc)); | |
1173 | ||
1174 | version++; | |
1175 | kvm_write_guest(kvm, wall_clock, &version, sizeof(version)); | |
18068523 GOC |
1176 | } |
1177 | ||
50d0a0f9 GH |
1178 | static uint32_t div_frac(uint32_t dividend, uint32_t divisor) |
1179 | { | |
1180 | uint32_t quotient, remainder; | |
1181 | ||
1182 | /* Don't try to replace with do_div(), this one calculates | |
1183 | * "(dividend << 32) / divisor" */ | |
1184 | __asm__ ( "divl %4" | |
1185 | : "=a" (quotient), "=d" (remainder) | |
1186 | : "0" (0), "1" (dividend), "r" (divisor) ); | |
1187 | return quotient; | |
1188 | } | |
1189 | ||
5f4e3f88 ZA |
1190 | static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz, |
1191 | s8 *pshift, u32 *pmultiplier) | |
50d0a0f9 | 1192 | { |
5f4e3f88 | 1193 | uint64_t scaled64; |
50d0a0f9 GH |
1194 | int32_t shift = 0; |
1195 | uint64_t tps64; | |
1196 | uint32_t tps32; | |
1197 | ||
5f4e3f88 ZA |
1198 | tps64 = base_khz * 1000LL; |
1199 | scaled64 = scaled_khz * 1000LL; | |
50933623 | 1200 | while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) { |
50d0a0f9 GH |
1201 | tps64 >>= 1; |
1202 | shift--; | |
1203 | } | |
1204 | ||
1205 | tps32 = (uint32_t)tps64; | |
50933623 JK |
1206 | while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) { |
1207 | if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000) | |
5f4e3f88 ZA |
1208 | scaled64 >>= 1; |
1209 | else | |
1210 | tps32 <<= 1; | |
50d0a0f9 GH |
1211 | shift++; |
1212 | } | |
1213 | ||
5f4e3f88 ZA |
1214 | *pshift = shift; |
1215 | *pmultiplier = div_frac(scaled64, tps32); | |
50d0a0f9 | 1216 | |
5f4e3f88 ZA |
1217 | pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n", |
1218 | __func__, base_khz, scaled_khz, shift, *pmultiplier); | |
50d0a0f9 GH |
1219 | } |
1220 | ||
d828199e | 1221 | #ifdef CONFIG_X86_64 |
16e8d74d | 1222 | static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0); |
d828199e | 1223 | #endif |
16e8d74d | 1224 | |
c8076604 | 1225 | static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz); |
69b0049a | 1226 | static unsigned long max_tsc_khz; |
c8076604 | 1227 | |
cc578287 | 1228 | static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec) |
8cfdc000 | 1229 | { |
cc578287 ZA |
1230 | return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult, |
1231 | vcpu->arch.virtual_tsc_shift); | |
8cfdc000 ZA |
1232 | } |
1233 | ||
cc578287 | 1234 | static u32 adjust_tsc_khz(u32 khz, s32 ppm) |
1e993611 | 1235 | { |
cc578287 ZA |
1236 | u64 v = (u64)khz * (1000000 + ppm); |
1237 | do_div(v, 1000000); | |
1238 | return v; | |
1e993611 JR |
1239 | } |
1240 | ||
cc578287 | 1241 | static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz) |
759379dd | 1242 | { |
cc578287 ZA |
1243 | u32 thresh_lo, thresh_hi; |
1244 | int use_scaling = 0; | |
217fc9cf | 1245 | |
03ba32ca MT |
1246 | /* tsc_khz can be zero if TSC calibration fails */ |
1247 | if (this_tsc_khz == 0) | |
1248 | return; | |
1249 | ||
c285545f ZA |
1250 | /* Compute a scale to convert nanoseconds in TSC cycles */ |
1251 | kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000, | |
cc578287 ZA |
1252 | &vcpu->arch.virtual_tsc_shift, |
1253 | &vcpu->arch.virtual_tsc_mult); | |
1254 | vcpu->arch.virtual_tsc_khz = this_tsc_khz; | |
1255 | ||
1256 | /* | |
1257 | * Compute the variation in TSC rate which is acceptable | |
1258 | * within the range of tolerance and decide if the | |
1259 | * rate being applied is within that bounds of the hardware | |
1260 | * rate. If so, no scaling or compensation need be done. | |
1261 | */ | |
1262 | thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm); | |
1263 | thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm); | |
1264 | if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) { | |
1265 | pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi); | |
1266 | use_scaling = 1; | |
1267 | } | |
1268 | kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling); | |
c285545f ZA |
1269 | } |
1270 | ||
1271 | static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns) | |
1272 | { | |
e26101b1 | 1273 | u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec, |
cc578287 ZA |
1274 | vcpu->arch.virtual_tsc_mult, |
1275 | vcpu->arch.virtual_tsc_shift); | |
e26101b1 | 1276 | tsc += vcpu->arch.this_tsc_write; |
c285545f ZA |
1277 | return tsc; |
1278 | } | |
1279 | ||
69b0049a | 1280 | static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu) |
b48aa97e MT |
1281 | { |
1282 | #ifdef CONFIG_X86_64 | |
1283 | bool vcpus_matched; | |
b48aa97e MT |
1284 | struct kvm_arch *ka = &vcpu->kvm->arch; |
1285 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
1286 | ||
1287 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
1288 | atomic_read(&vcpu->kvm->online_vcpus)); | |
1289 | ||
7f187922 MT |
1290 | /* |
1291 | * Once the masterclock is enabled, always perform request in | |
1292 | * order to update it. | |
1293 | * | |
1294 | * In order to enable masterclock, the host clocksource must be TSC | |
1295 | * and the vcpus need to have matched TSCs. When that happens, | |
1296 | * perform request to enable masterclock. | |
1297 | */ | |
1298 | if (ka->use_master_clock || | |
1299 | (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched)) | |
b48aa97e MT |
1300 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
1301 | ||
1302 | trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc, | |
1303 | atomic_read(&vcpu->kvm->online_vcpus), | |
1304 | ka->use_master_clock, gtod->clock.vclock_mode); | |
1305 | #endif | |
1306 | } | |
1307 | ||
ba904635 WA |
1308 | static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset) |
1309 | { | |
1310 | u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu); | |
1311 | vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset; | |
1312 | } | |
1313 | ||
8fe8ab46 | 1314 | void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr) |
99e3e30a ZA |
1315 | { |
1316 | struct kvm *kvm = vcpu->kvm; | |
f38e098f | 1317 | u64 offset, ns, elapsed; |
99e3e30a | 1318 | unsigned long flags; |
02626b6a | 1319 | s64 usdiff; |
b48aa97e | 1320 | bool matched; |
0d3da0d2 | 1321 | bool already_matched; |
8fe8ab46 | 1322 | u64 data = msr->data; |
99e3e30a | 1323 | |
038f8c11 | 1324 | raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags); |
857e4099 | 1325 | offset = kvm_x86_ops->compute_tsc_offset(vcpu, data); |
759379dd | 1326 | ns = get_kernel_ns(); |
f38e098f | 1327 | elapsed = ns - kvm->arch.last_tsc_nsec; |
5d3cb0f6 | 1328 | |
03ba32ca | 1329 | if (vcpu->arch.virtual_tsc_khz) { |
8915aa27 MT |
1330 | int faulted = 0; |
1331 | ||
03ba32ca MT |
1332 | /* n.b - signed multiplication and division required */ |
1333 | usdiff = data - kvm->arch.last_tsc_write; | |
5d3cb0f6 | 1334 | #ifdef CONFIG_X86_64 |
03ba32ca | 1335 | usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz; |
5d3cb0f6 | 1336 | #else |
03ba32ca | 1337 | /* do_div() only does unsigned */ |
8915aa27 MT |
1338 | asm("1: idivl %[divisor]\n" |
1339 | "2: xor %%edx, %%edx\n" | |
1340 | " movl $0, %[faulted]\n" | |
1341 | "3:\n" | |
1342 | ".section .fixup,\"ax\"\n" | |
1343 | "4: movl $1, %[faulted]\n" | |
1344 | " jmp 3b\n" | |
1345 | ".previous\n" | |
1346 | ||
1347 | _ASM_EXTABLE(1b, 4b) | |
1348 | ||
1349 | : "=A"(usdiff), [faulted] "=r" (faulted) | |
1350 | : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz)); | |
1351 | ||
5d3cb0f6 | 1352 | #endif |
03ba32ca MT |
1353 | do_div(elapsed, 1000); |
1354 | usdiff -= elapsed; | |
1355 | if (usdiff < 0) | |
1356 | usdiff = -usdiff; | |
8915aa27 MT |
1357 | |
1358 | /* idivl overflow => difference is larger than USEC_PER_SEC */ | |
1359 | if (faulted) | |
1360 | usdiff = USEC_PER_SEC; | |
03ba32ca MT |
1361 | } else |
1362 | usdiff = USEC_PER_SEC; /* disable TSC match window below */ | |
f38e098f ZA |
1363 | |
1364 | /* | |
5d3cb0f6 ZA |
1365 | * Special case: TSC write with a small delta (1 second) of virtual |
1366 | * cycle time against real time is interpreted as an attempt to | |
1367 | * synchronize the CPU. | |
1368 | * | |
1369 | * For a reliable TSC, we can match TSC offsets, and for an unstable | |
1370 | * TSC, we add elapsed time in this computation. We could let the | |
1371 | * compensation code attempt to catch up if we fall behind, but | |
1372 | * it's better to try to match offsets from the beginning. | |
1373 | */ | |
02626b6a | 1374 | if (usdiff < USEC_PER_SEC && |
5d3cb0f6 | 1375 | vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) { |
f38e098f | 1376 | if (!check_tsc_unstable()) { |
e26101b1 | 1377 | offset = kvm->arch.cur_tsc_offset; |
f38e098f ZA |
1378 | pr_debug("kvm: matched tsc offset for %llu\n", data); |
1379 | } else { | |
857e4099 | 1380 | u64 delta = nsec_to_cycles(vcpu, elapsed); |
5d3cb0f6 ZA |
1381 | data += delta; |
1382 | offset = kvm_x86_ops->compute_tsc_offset(vcpu, data); | |
759379dd | 1383 | pr_debug("kvm: adjusted tsc offset by %llu\n", delta); |
f38e098f | 1384 | } |
b48aa97e | 1385 | matched = true; |
0d3da0d2 | 1386 | already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation); |
e26101b1 ZA |
1387 | } else { |
1388 | /* | |
1389 | * We split periods of matched TSC writes into generations. | |
1390 | * For each generation, we track the original measured | |
1391 | * nanosecond time, offset, and write, so if TSCs are in | |
1392 | * sync, we can match exact offset, and if not, we can match | |
4a969980 | 1393 | * exact software computation in compute_guest_tsc() |
e26101b1 ZA |
1394 | * |
1395 | * These values are tracked in kvm->arch.cur_xxx variables. | |
1396 | */ | |
1397 | kvm->arch.cur_tsc_generation++; | |
1398 | kvm->arch.cur_tsc_nsec = ns; | |
1399 | kvm->arch.cur_tsc_write = data; | |
1400 | kvm->arch.cur_tsc_offset = offset; | |
b48aa97e | 1401 | matched = false; |
0d3da0d2 | 1402 | pr_debug("kvm: new tsc generation %llu, clock %llu\n", |
e26101b1 | 1403 | kvm->arch.cur_tsc_generation, data); |
f38e098f | 1404 | } |
e26101b1 ZA |
1405 | |
1406 | /* | |
1407 | * We also track th most recent recorded KHZ, write and time to | |
1408 | * allow the matching interval to be extended at each write. | |
1409 | */ | |
f38e098f ZA |
1410 | kvm->arch.last_tsc_nsec = ns; |
1411 | kvm->arch.last_tsc_write = data; | |
5d3cb0f6 | 1412 | kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz; |
99e3e30a | 1413 | |
b183aa58 | 1414 | vcpu->arch.last_guest_tsc = data; |
e26101b1 ZA |
1415 | |
1416 | /* Keep track of which generation this VCPU has synchronized to */ | |
1417 | vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation; | |
1418 | vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec; | |
1419 | vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write; | |
1420 | ||
ba904635 WA |
1421 | if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated) |
1422 | update_ia32_tsc_adjust_msr(vcpu, offset); | |
e26101b1 ZA |
1423 | kvm_x86_ops->write_tsc_offset(vcpu, offset); |
1424 | raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags); | |
b48aa97e MT |
1425 | |
1426 | spin_lock(&kvm->arch.pvclock_gtod_sync_lock); | |
0d3da0d2 | 1427 | if (!matched) { |
b48aa97e | 1428 | kvm->arch.nr_vcpus_matched_tsc = 0; |
0d3da0d2 TG |
1429 | } else if (!already_matched) { |
1430 | kvm->arch.nr_vcpus_matched_tsc++; | |
1431 | } | |
b48aa97e MT |
1432 | |
1433 | kvm_track_tsc_matching(vcpu); | |
1434 | spin_unlock(&kvm->arch.pvclock_gtod_sync_lock); | |
99e3e30a | 1435 | } |
e26101b1 | 1436 | |
99e3e30a ZA |
1437 | EXPORT_SYMBOL_GPL(kvm_write_tsc); |
1438 | ||
d828199e MT |
1439 | #ifdef CONFIG_X86_64 |
1440 | ||
1441 | static cycle_t read_tsc(void) | |
1442 | { | |
1443 | cycle_t ret; | |
1444 | u64 last; | |
1445 | ||
1446 | /* | |
1447 | * Empirically, a fence (of type that depends on the CPU) | |
1448 | * before rdtsc is enough to ensure that rdtsc is ordered | |
1449 | * with respect to loads. The various CPU manuals are unclear | |
1450 | * as to whether rdtsc can be reordered with later loads, | |
1451 | * but no one has ever seen it happen. | |
1452 | */ | |
1453 | rdtsc_barrier(); | |
1454 | ret = (cycle_t)vget_cycles(); | |
1455 | ||
1456 | last = pvclock_gtod_data.clock.cycle_last; | |
1457 | ||
1458 | if (likely(ret >= last)) | |
1459 | return ret; | |
1460 | ||
1461 | /* | |
1462 | * GCC likes to generate cmov here, but this branch is extremely | |
1463 | * predictable (it's just a funciton of time and the likely is | |
1464 | * very likely) and there's a data dependence, so force GCC | |
1465 | * to generate a branch instead. I don't barrier() because | |
1466 | * we don't actually need a barrier, and if this function | |
1467 | * ever gets inlined it will generate worse code. | |
1468 | */ | |
1469 | asm volatile (""); | |
1470 | return last; | |
1471 | } | |
1472 | ||
1473 | static inline u64 vgettsc(cycle_t *cycle_now) | |
1474 | { | |
1475 | long v; | |
1476 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
1477 | ||
1478 | *cycle_now = read_tsc(); | |
1479 | ||
1480 | v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask; | |
1481 | return v * gtod->clock.mult; | |
1482 | } | |
1483 | ||
cbcf2dd3 | 1484 | static int do_monotonic_boot(s64 *t, cycle_t *cycle_now) |
d828199e | 1485 | { |
cbcf2dd3 | 1486 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; |
d828199e | 1487 | unsigned long seq; |
d828199e | 1488 | int mode; |
cbcf2dd3 | 1489 | u64 ns; |
d828199e | 1490 | |
d828199e MT |
1491 | do { |
1492 | seq = read_seqcount_begin(>od->seq); | |
1493 | mode = gtod->clock.vclock_mode; | |
cbcf2dd3 | 1494 | ns = gtod->nsec_base; |
d828199e MT |
1495 | ns += vgettsc(cycle_now); |
1496 | ns >>= gtod->clock.shift; | |
cbcf2dd3 | 1497 | ns += gtod->boot_ns; |
d828199e | 1498 | } while (unlikely(read_seqcount_retry(>od->seq, seq))); |
cbcf2dd3 | 1499 | *t = ns; |
d828199e MT |
1500 | |
1501 | return mode; | |
1502 | } | |
1503 | ||
1504 | /* returns true if host is using tsc clocksource */ | |
1505 | static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now) | |
1506 | { | |
d828199e MT |
1507 | /* checked again under seqlock below */ |
1508 | if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC) | |
1509 | return false; | |
1510 | ||
cbcf2dd3 | 1511 | return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC; |
d828199e MT |
1512 | } |
1513 | #endif | |
1514 | ||
1515 | /* | |
1516 | * | |
b48aa97e MT |
1517 | * Assuming a stable TSC across physical CPUS, and a stable TSC |
1518 | * across virtual CPUs, the following condition is possible. | |
1519 | * Each numbered line represents an event visible to both | |
d828199e MT |
1520 | * CPUs at the next numbered event. |
1521 | * | |
1522 | * "timespecX" represents host monotonic time. "tscX" represents | |
1523 | * RDTSC value. | |
1524 | * | |
1525 | * VCPU0 on CPU0 | VCPU1 on CPU1 | |
1526 | * | |
1527 | * 1. read timespec0,tsc0 | |
1528 | * 2. | timespec1 = timespec0 + N | |
1529 | * | tsc1 = tsc0 + M | |
1530 | * 3. transition to guest | transition to guest | |
1531 | * 4. ret0 = timespec0 + (rdtsc - tsc0) | | |
1532 | * 5. | ret1 = timespec1 + (rdtsc - tsc1) | |
1533 | * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M)) | |
1534 | * | |
1535 | * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity: | |
1536 | * | |
1537 | * - ret0 < ret1 | |
1538 | * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M)) | |
1539 | * ... | |
1540 | * - 0 < N - M => M < N | |
1541 | * | |
1542 | * That is, when timespec0 != timespec1, M < N. Unfortunately that is not | |
1543 | * always the case (the difference between two distinct xtime instances | |
1544 | * might be smaller then the difference between corresponding TSC reads, | |
1545 | * when updating guest vcpus pvclock areas). | |
1546 | * | |
1547 | * To avoid that problem, do not allow visibility of distinct | |
1548 | * system_timestamp/tsc_timestamp values simultaneously: use a master | |
1549 | * copy of host monotonic time values. Update that master copy | |
1550 | * in lockstep. | |
1551 | * | |
b48aa97e | 1552 | * Rely on synchronization of host TSCs and guest TSCs for monotonicity. |
d828199e MT |
1553 | * |
1554 | */ | |
1555 | ||
1556 | static void pvclock_update_vm_gtod_copy(struct kvm *kvm) | |
1557 | { | |
1558 | #ifdef CONFIG_X86_64 | |
1559 | struct kvm_arch *ka = &kvm->arch; | |
1560 | int vclock_mode; | |
b48aa97e MT |
1561 | bool host_tsc_clocksource, vcpus_matched; |
1562 | ||
1563 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
1564 | atomic_read(&kvm->online_vcpus)); | |
d828199e MT |
1565 | |
1566 | /* | |
1567 | * If the host uses TSC clock, then passthrough TSC as stable | |
1568 | * to the guest. | |
1569 | */ | |
b48aa97e | 1570 | host_tsc_clocksource = kvm_get_time_and_clockread( |
d828199e MT |
1571 | &ka->master_kernel_ns, |
1572 | &ka->master_cycle_now); | |
1573 | ||
16a96021 | 1574 | ka->use_master_clock = host_tsc_clocksource && vcpus_matched |
54750f2c MT |
1575 | && !backwards_tsc_observed |
1576 | && !ka->boot_vcpu_runs_old_kvmclock; | |
b48aa97e | 1577 | |
d828199e MT |
1578 | if (ka->use_master_clock) |
1579 | atomic_set(&kvm_guest_has_master_clock, 1); | |
1580 | ||
1581 | vclock_mode = pvclock_gtod_data.clock.vclock_mode; | |
b48aa97e MT |
1582 | trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode, |
1583 | vcpus_matched); | |
d828199e MT |
1584 | #endif |
1585 | } | |
1586 | ||
2e762ff7 MT |
1587 | static void kvm_gen_update_masterclock(struct kvm *kvm) |
1588 | { | |
1589 | #ifdef CONFIG_X86_64 | |
1590 | int i; | |
1591 | struct kvm_vcpu *vcpu; | |
1592 | struct kvm_arch *ka = &kvm->arch; | |
1593 | ||
1594 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
1595 | kvm_make_mclock_inprogress_request(kvm); | |
1596 | /* no guest entries from this point */ | |
1597 | pvclock_update_vm_gtod_copy(kvm); | |
1598 | ||
1599 | kvm_for_each_vcpu(i, vcpu, kvm) | |
105b21bb | 1600 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
2e762ff7 MT |
1601 | |
1602 | /* guest entries allowed */ | |
1603 | kvm_for_each_vcpu(i, vcpu, kvm) | |
1604 | clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests); | |
1605 | ||
1606 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
1607 | #endif | |
1608 | } | |
1609 | ||
34c238a1 | 1610 | static int kvm_guest_time_update(struct kvm_vcpu *v) |
18068523 | 1611 | { |
d828199e | 1612 | unsigned long flags, this_tsc_khz; |
18068523 | 1613 | struct kvm_vcpu_arch *vcpu = &v->arch; |
d828199e | 1614 | struct kvm_arch *ka = &v->kvm->arch; |
f25e656d | 1615 | s64 kernel_ns; |
d828199e | 1616 | u64 tsc_timestamp, host_tsc; |
0b79459b | 1617 | struct pvclock_vcpu_time_info guest_hv_clock; |
51d59c6b | 1618 | u8 pvclock_flags; |
d828199e MT |
1619 | bool use_master_clock; |
1620 | ||
1621 | kernel_ns = 0; | |
1622 | host_tsc = 0; | |
18068523 | 1623 | |
d828199e MT |
1624 | /* |
1625 | * If the host uses TSC clock, then passthrough TSC as stable | |
1626 | * to the guest. | |
1627 | */ | |
1628 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
1629 | use_master_clock = ka->use_master_clock; | |
1630 | if (use_master_clock) { | |
1631 | host_tsc = ka->master_cycle_now; | |
1632 | kernel_ns = ka->master_kernel_ns; | |
1633 | } | |
1634 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
c09664bb MT |
1635 | |
1636 | /* Keep irq disabled to prevent changes to the clock */ | |
1637 | local_irq_save(flags); | |
89cbc767 | 1638 | this_tsc_khz = __this_cpu_read(cpu_tsc_khz); |
c09664bb MT |
1639 | if (unlikely(this_tsc_khz == 0)) { |
1640 | local_irq_restore(flags); | |
1641 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); | |
1642 | return 1; | |
1643 | } | |
d828199e MT |
1644 | if (!use_master_clock) { |
1645 | host_tsc = native_read_tsc(); | |
1646 | kernel_ns = get_kernel_ns(); | |
1647 | } | |
1648 | ||
1649 | tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc); | |
1650 | ||
c285545f ZA |
1651 | /* |
1652 | * We may have to catch up the TSC to match elapsed wall clock | |
1653 | * time for two reasons, even if kvmclock is used. | |
1654 | * 1) CPU could have been running below the maximum TSC rate | |
1655 | * 2) Broken TSC compensation resets the base at each VCPU | |
1656 | * entry to avoid unknown leaps of TSC even when running | |
1657 | * again on the same CPU. This may cause apparent elapsed | |
1658 | * time to disappear, and the guest to stand still or run | |
1659 | * very slowly. | |
1660 | */ | |
1661 | if (vcpu->tsc_catchup) { | |
1662 | u64 tsc = compute_guest_tsc(v, kernel_ns); | |
1663 | if (tsc > tsc_timestamp) { | |
f1e2b260 | 1664 | adjust_tsc_offset_guest(v, tsc - tsc_timestamp); |
c285545f ZA |
1665 | tsc_timestamp = tsc; |
1666 | } | |
50d0a0f9 GH |
1667 | } |
1668 | ||
18068523 GOC |
1669 | local_irq_restore(flags); |
1670 | ||
0b79459b | 1671 | if (!vcpu->pv_time_enabled) |
c285545f | 1672 | return 0; |
18068523 | 1673 | |
e48672fa | 1674 | if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) { |
5f4e3f88 ZA |
1675 | kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz, |
1676 | &vcpu->hv_clock.tsc_shift, | |
1677 | &vcpu->hv_clock.tsc_to_system_mul); | |
e48672fa | 1678 | vcpu->hw_tsc_khz = this_tsc_khz; |
8cfdc000 ZA |
1679 | } |
1680 | ||
1681 | /* With all the info we got, fill in the values */ | |
1d5f066e | 1682 | vcpu->hv_clock.tsc_timestamp = tsc_timestamp; |
759379dd | 1683 | vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset; |
28e4639a | 1684 | vcpu->last_guest_tsc = tsc_timestamp; |
51d59c6b | 1685 | |
09a0c3f1 OH |
1686 | if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time, |
1687 | &guest_hv_clock, sizeof(guest_hv_clock)))) | |
1688 | return 0; | |
1689 | ||
5dca0d91 RK |
1690 | /* This VCPU is paused, but it's legal for a guest to read another |
1691 | * VCPU's kvmclock, so we really have to follow the specification where | |
1692 | * it says that version is odd if data is being modified, and even after | |
1693 | * it is consistent. | |
1694 | * | |
1695 | * Version field updates must be kept separate. This is because | |
1696 | * kvm_write_guest_cached might use a "rep movs" instruction, and | |
1697 | * writes within a string instruction are weakly ordered. So there | |
1698 | * are three writes overall. | |
1699 | * | |
1700 | * As a small optimization, only write the version field in the first | |
1701 | * and third write. The vcpu->pv_time cache is still valid, because the | |
1702 | * version field is the first in the struct. | |
18068523 | 1703 | */ |
5dca0d91 RK |
1704 | BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0); |
1705 | ||
1706 | vcpu->hv_clock.version = guest_hv_clock.version + 1; | |
1707 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, | |
1708 | &vcpu->hv_clock, | |
1709 | sizeof(vcpu->hv_clock.version)); | |
1710 | ||
1711 | smp_wmb(); | |
78c0337a MT |
1712 | |
1713 | /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */ | |
0b79459b | 1714 | pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED); |
78c0337a MT |
1715 | |
1716 | if (vcpu->pvclock_set_guest_stopped_request) { | |
1717 | pvclock_flags |= PVCLOCK_GUEST_STOPPED; | |
1718 | vcpu->pvclock_set_guest_stopped_request = false; | |
1719 | } | |
1720 | ||
b7e60c5a MT |
1721 | pvclock_flags |= PVCLOCK_COUNTS_FROM_ZERO; |
1722 | ||
d828199e MT |
1723 | /* If the host uses TSC clocksource, then it is stable */ |
1724 | if (use_master_clock) | |
1725 | pvclock_flags |= PVCLOCK_TSC_STABLE_BIT; | |
1726 | ||
78c0337a MT |
1727 | vcpu->hv_clock.flags = pvclock_flags; |
1728 | ||
ce1a5e60 DM |
1729 | trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock); |
1730 | ||
0b79459b AH |
1731 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, |
1732 | &vcpu->hv_clock, | |
1733 | sizeof(vcpu->hv_clock)); | |
5dca0d91 RK |
1734 | |
1735 | smp_wmb(); | |
1736 | ||
1737 | vcpu->hv_clock.version++; | |
1738 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, | |
1739 | &vcpu->hv_clock, | |
1740 | sizeof(vcpu->hv_clock.version)); | |
8cfdc000 | 1741 | return 0; |
c8076604 GH |
1742 | } |
1743 | ||
0061d53d MT |
1744 | /* |
1745 | * kvmclock updates which are isolated to a given vcpu, such as | |
1746 | * vcpu->cpu migration, should not allow system_timestamp from | |
1747 | * the rest of the vcpus to remain static. Otherwise ntp frequency | |
1748 | * correction applies to one vcpu's system_timestamp but not | |
1749 | * the others. | |
1750 | * | |
1751 | * So in those cases, request a kvmclock update for all vcpus. | |
7e44e449 AJ |
1752 | * We need to rate-limit these requests though, as they can |
1753 | * considerably slow guests that have a large number of vcpus. | |
1754 | * The time for a remote vcpu to update its kvmclock is bound | |
1755 | * by the delay we use to rate-limit the updates. | |
0061d53d MT |
1756 | */ |
1757 | ||
7e44e449 AJ |
1758 | #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100) |
1759 | ||
1760 | static void kvmclock_update_fn(struct work_struct *work) | |
0061d53d MT |
1761 | { |
1762 | int i; | |
7e44e449 AJ |
1763 | struct delayed_work *dwork = to_delayed_work(work); |
1764 | struct kvm_arch *ka = container_of(dwork, struct kvm_arch, | |
1765 | kvmclock_update_work); | |
1766 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
0061d53d MT |
1767 | struct kvm_vcpu *vcpu; |
1768 | ||
1769 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
105b21bb | 1770 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0061d53d MT |
1771 | kvm_vcpu_kick(vcpu); |
1772 | } | |
1773 | } | |
1774 | ||
7e44e449 AJ |
1775 | static void kvm_gen_kvmclock_update(struct kvm_vcpu *v) |
1776 | { | |
1777 | struct kvm *kvm = v->kvm; | |
1778 | ||
105b21bb | 1779 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); |
7e44e449 AJ |
1780 | schedule_delayed_work(&kvm->arch.kvmclock_update_work, |
1781 | KVMCLOCK_UPDATE_DELAY); | |
1782 | } | |
1783 | ||
332967a3 AJ |
1784 | #define KVMCLOCK_SYNC_PERIOD (300 * HZ) |
1785 | ||
1786 | static void kvmclock_sync_fn(struct work_struct *work) | |
1787 | { | |
1788 | struct delayed_work *dwork = to_delayed_work(work); | |
1789 | struct kvm_arch *ka = container_of(dwork, struct kvm_arch, | |
1790 | kvmclock_sync_work); | |
1791 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
1792 | ||
630994b3 MT |
1793 | if (!kvmclock_periodic_sync) |
1794 | return; | |
1795 | ||
332967a3 AJ |
1796 | schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0); |
1797 | schedule_delayed_work(&kvm->arch.kvmclock_sync_work, | |
1798 | KVMCLOCK_SYNC_PERIOD); | |
1799 | } | |
1800 | ||
890ca9ae | 1801 | static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data) |
15c4a640 | 1802 | { |
890ca9ae HY |
1803 | u64 mcg_cap = vcpu->arch.mcg_cap; |
1804 | unsigned bank_num = mcg_cap & 0xff; | |
1805 | ||
15c4a640 | 1806 | switch (msr) { |
15c4a640 | 1807 | case MSR_IA32_MCG_STATUS: |
890ca9ae | 1808 | vcpu->arch.mcg_status = data; |
15c4a640 | 1809 | break; |
c7ac679c | 1810 | case MSR_IA32_MCG_CTL: |
890ca9ae HY |
1811 | if (!(mcg_cap & MCG_CTL_P)) |
1812 | return 1; | |
1813 | if (data != 0 && data != ~(u64)0) | |
1814 | return -1; | |
1815 | vcpu->arch.mcg_ctl = data; | |
1816 | break; | |
1817 | default: | |
1818 | if (msr >= MSR_IA32_MC0_CTL && | |
81760dcc | 1819 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
890ca9ae | 1820 | u32 offset = msr - MSR_IA32_MC0_CTL; |
114be429 AP |
1821 | /* only 0 or all 1s can be written to IA32_MCi_CTL |
1822 | * some Linux kernels though clear bit 10 in bank 4 to | |
1823 | * workaround a BIOS/GART TBL issue on AMD K8s, ignore | |
1824 | * this to avoid an uncatched #GP in the guest | |
1825 | */ | |
890ca9ae | 1826 | if ((offset & 0x3) == 0 && |
114be429 | 1827 | data != 0 && (data | (1 << 10)) != ~(u64)0) |
890ca9ae HY |
1828 | return -1; |
1829 | vcpu->arch.mce_banks[offset] = data; | |
1830 | break; | |
1831 | } | |
1832 | return 1; | |
1833 | } | |
1834 | return 0; | |
1835 | } | |
1836 | ||
ffde22ac ES |
1837 | static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data) |
1838 | { | |
1839 | struct kvm *kvm = vcpu->kvm; | |
1840 | int lm = is_long_mode(vcpu); | |
1841 | u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64 | |
1842 | : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32; | |
1843 | u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64 | |
1844 | : kvm->arch.xen_hvm_config.blob_size_32; | |
1845 | u32 page_num = data & ~PAGE_MASK; | |
1846 | u64 page_addr = data & PAGE_MASK; | |
1847 | u8 *page; | |
1848 | int r; | |
1849 | ||
1850 | r = -E2BIG; | |
1851 | if (page_num >= blob_size) | |
1852 | goto out; | |
1853 | r = -ENOMEM; | |
ff5c2c03 SL |
1854 | page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE); |
1855 | if (IS_ERR(page)) { | |
1856 | r = PTR_ERR(page); | |
ffde22ac | 1857 | goto out; |
ff5c2c03 | 1858 | } |
54bf36aa | 1859 | if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) |
ffde22ac ES |
1860 | goto out_free; |
1861 | r = 0; | |
1862 | out_free: | |
1863 | kfree(page); | |
1864 | out: | |
1865 | return r; | |
1866 | } | |
1867 | ||
344d9588 GN |
1868 | static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data) |
1869 | { | |
1870 | gpa_t gpa = data & ~0x3f; | |
1871 | ||
4a969980 | 1872 | /* Bits 2:5 are reserved, Should be zero */ |
6adba527 | 1873 | if (data & 0x3c) |
344d9588 GN |
1874 | return 1; |
1875 | ||
1876 | vcpu->arch.apf.msr_val = data; | |
1877 | ||
1878 | if (!(data & KVM_ASYNC_PF_ENABLED)) { | |
1879 | kvm_clear_async_pf_completion_queue(vcpu); | |
1880 | kvm_async_pf_hash_reset(vcpu); | |
1881 | return 0; | |
1882 | } | |
1883 | ||
8f964525 AH |
1884 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa, |
1885 | sizeof(u32))) | |
344d9588 GN |
1886 | return 1; |
1887 | ||
6adba527 | 1888 | vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS); |
344d9588 GN |
1889 | kvm_async_pf_wakeup_all(vcpu); |
1890 | return 0; | |
1891 | } | |
1892 | ||
12f9a48f GC |
1893 | static void kvmclock_reset(struct kvm_vcpu *vcpu) |
1894 | { | |
0b79459b | 1895 | vcpu->arch.pv_time_enabled = false; |
12f9a48f GC |
1896 | } |
1897 | ||
c9aaa895 GC |
1898 | static void accumulate_steal_time(struct kvm_vcpu *vcpu) |
1899 | { | |
1900 | u64 delta; | |
1901 | ||
1902 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) | |
1903 | return; | |
1904 | ||
1905 | delta = current->sched_info.run_delay - vcpu->arch.st.last_steal; | |
1906 | vcpu->arch.st.last_steal = current->sched_info.run_delay; | |
1907 | vcpu->arch.st.accum_steal = delta; | |
1908 | } | |
1909 | ||
1910 | static void record_steal_time(struct kvm_vcpu *vcpu) | |
1911 | { | |
1912 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) | |
1913 | return; | |
1914 | ||
1915 | if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime, | |
1916 | &vcpu->arch.st.steal, sizeof(struct kvm_steal_time)))) | |
1917 | return; | |
1918 | ||
1919 | vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal; | |
1920 | vcpu->arch.st.steal.version += 2; | |
1921 | vcpu->arch.st.accum_steal = 0; | |
1922 | ||
1923 | kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime, | |
1924 | &vcpu->arch.st.steal, sizeof(struct kvm_steal_time)); | |
1925 | } | |
1926 | ||
8fe8ab46 | 1927 | int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
15c4a640 | 1928 | { |
5753785f | 1929 | bool pr = false; |
8fe8ab46 WA |
1930 | u32 msr = msr_info->index; |
1931 | u64 data = msr_info->data; | |
5753785f | 1932 | |
15c4a640 | 1933 | switch (msr) { |
2e32b719 BP |
1934 | case MSR_AMD64_NB_CFG: |
1935 | case MSR_IA32_UCODE_REV: | |
1936 | case MSR_IA32_UCODE_WRITE: | |
1937 | case MSR_VM_HSAVE_PA: | |
1938 | case MSR_AMD64_PATCH_LOADER: | |
1939 | case MSR_AMD64_BU_CFG2: | |
1940 | break; | |
1941 | ||
15c4a640 | 1942 | case MSR_EFER: |
b69e8cae | 1943 | return set_efer(vcpu, data); |
8f1589d9 AP |
1944 | case MSR_K7_HWCR: |
1945 | data &= ~(u64)0x40; /* ignore flush filter disable */ | |
82494028 | 1946 | data &= ~(u64)0x100; /* ignore ignne emulation enable */ |
a223c313 | 1947 | data &= ~(u64)0x8; /* ignore TLB cache disable */ |
22d48b2d | 1948 | data &= ~(u64)0x40000; /* ignore Mc status write enable */ |
8f1589d9 | 1949 | if (data != 0) { |
a737f256 CD |
1950 | vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n", |
1951 | data); | |
8f1589d9 AP |
1952 | return 1; |
1953 | } | |
15c4a640 | 1954 | break; |
f7c6d140 AP |
1955 | case MSR_FAM10H_MMIO_CONF_BASE: |
1956 | if (data != 0) { | |
a737f256 CD |
1957 | vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: " |
1958 | "0x%llx\n", data); | |
f7c6d140 AP |
1959 | return 1; |
1960 | } | |
15c4a640 | 1961 | break; |
b5e2fec0 AG |
1962 | case MSR_IA32_DEBUGCTLMSR: |
1963 | if (!data) { | |
1964 | /* We support the non-activated case already */ | |
1965 | break; | |
1966 | } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) { | |
1967 | /* Values other than LBR and BTF are vendor-specific, | |
1968 | thus reserved and should throw a #GP */ | |
1969 | return 1; | |
1970 | } | |
a737f256 CD |
1971 | vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n", |
1972 | __func__, data); | |
b5e2fec0 | 1973 | break; |
9ba075a6 | 1974 | case 0x200 ... 0x2ff: |
ff53604b | 1975 | return kvm_mtrr_set_msr(vcpu, msr, data); |
15c4a640 | 1976 | case MSR_IA32_APICBASE: |
58cb628d | 1977 | return kvm_set_apic_base(vcpu, msr_info); |
0105d1a5 GN |
1978 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff: |
1979 | return kvm_x2apic_msr_write(vcpu, msr, data); | |
a3e06bbe LJ |
1980 | case MSR_IA32_TSCDEADLINE: |
1981 | kvm_set_lapic_tscdeadline_msr(vcpu, data); | |
1982 | break; | |
ba904635 WA |
1983 | case MSR_IA32_TSC_ADJUST: |
1984 | if (guest_cpuid_has_tsc_adjust(vcpu)) { | |
1985 | if (!msr_info->host_initiated) { | |
d913b904 | 1986 | s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr; |
ba904635 WA |
1987 | kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true); |
1988 | } | |
1989 | vcpu->arch.ia32_tsc_adjust_msr = data; | |
1990 | } | |
1991 | break; | |
15c4a640 | 1992 | case MSR_IA32_MISC_ENABLE: |
ad312c7c | 1993 | vcpu->arch.ia32_misc_enable_msr = data; |
15c4a640 | 1994 | break; |
64d60670 PB |
1995 | case MSR_IA32_SMBASE: |
1996 | if (!msr_info->host_initiated) | |
1997 | return 1; | |
1998 | vcpu->arch.smbase = data; | |
1999 | break; | |
11c6bffa | 2000 | case MSR_KVM_WALL_CLOCK_NEW: |
18068523 GOC |
2001 | case MSR_KVM_WALL_CLOCK: |
2002 | vcpu->kvm->arch.wall_clock = data; | |
2003 | kvm_write_wall_clock(vcpu->kvm, data); | |
2004 | break; | |
11c6bffa | 2005 | case MSR_KVM_SYSTEM_TIME_NEW: |
18068523 | 2006 | case MSR_KVM_SYSTEM_TIME: { |
0b79459b | 2007 | u64 gpa_offset; |
54750f2c MT |
2008 | struct kvm_arch *ka = &vcpu->kvm->arch; |
2009 | ||
12f9a48f | 2010 | kvmclock_reset(vcpu); |
18068523 | 2011 | |
54750f2c MT |
2012 | if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) { |
2013 | bool tmp = (msr == MSR_KVM_SYSTEM_TIME); | |
2014 | ||
2015 | if (ka->boot_vcpu_runs_old_kvmclock != tmp) | |
2016 | set_bit(KVM_REQ_MASTERCLOCK_UPDATE, | |
2017 | &vcpu->requests); | |
2018 | ||
2019 | ka->boot_vcpu_runs_old_kvmclock = tmp; | |
b7e60c5a MT |
2020 | |
2021 | ka->kvmclock_offset = -get_kernel_ns(); | |
54750f2c MT |
2022 | } |
2023 | ||
18068523 | 2024 | vcpu->arch.time = data; |
0061d53d | 2025 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); |
18068523 GOC |
2026 | |
2027 | /* we verify if the enable bit is set... */ | |
2028 | if (!(data & 1)) | |
2029 | break; | |
2030 | ||
0b79459b | 2031 | gpa_offset = data & ~(PAGE_MASK | 1); |
18068523 | 2032 | |
0b79459b | 2033 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, |
8f964525 AH |
2034 | &vcpu->arch.pv_time, data & ~1ULL, |
2035 | sizeof(struct pvclock_vcpu_time_info))) | |
0b79459b AH |
2036 | vcpu->arch.pv_time_enabled = false; |
2037 | else | |
2038 | vcpu->arch.pv_time_enabled = true; | |
32cad84f | 2039 | |
18068523 GOC |
2040 | break; |
2041 | } | |
344d9588 GN |
2042 | case MSR_KVM_ASYNC_PF_EN: |
2043 | if (kvm_pv_enable_async_pf(vcpu, data)) | |
2044 | return 1; | |
2045 | break; | |
c9aaa895 GC |
2046 | case MSR_KVM_STEAL_TIME: |
2047 | ||
2048 | if (unlikely(!sched_info_on())) | |
2049 | return 1; | |
2050 | ||
2051 | if (data & KVM_STEAL_RESERVED_MASK) | |
2052 | return 1; | |
2053 | ||
2054 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime, | |
8f964525 AH |
2055 | data & KVM_STEAL_VALID_BITS, |
2056 | sizeof(struct kvm_steal_time))) | |
c9aaa895 GC |
2057 | return 1; |
2058 | ||
2059 | vcpu->arch.st.msr_val = data; | |
2060 | ||
2061 | if (!(data & KVM_MSR_ENABLED)) | |
2062 | break; | |
2063 | ||
2064 | vcpu->arch.st.last_steal = current->sched_info.run_delay; | |
2065 | ||
2066 | preempt_disable(); | |
2067 | accumulate_steal_time(vcpu); | |
2068 | preempt_enable(); | |
2069 | ||
2070 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); | |
2071 | ||
2072 | break; | |
ae7a2a3f MT |
2073 | case MSR_KVM_PV_EOI_EN: |
2074 | if (kvm_lapic_enable_pv_eoi(vcpu, data)) | |
2075 | return 1; | |
2076 | break; | |
c9aaa895 | 2077 | |
890ca9ae HY |
2078 | case MSR_IA32_MCG_CTL: |
2079 | case MSR_IA32_MCG_STATUS: | |
81760dcc | 2080 | case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1: |
890ca9ae | 2081 | return set_msr_mce(vcpu, msr, data); |
71db6023 | 2082 | |
6912ac32 WH |
2083 | case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3: |
2084 | case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1: | |
2085 | pr = true; /* fall through */ | |
2086 | case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3: | |
2087 | case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1: | |
c6702c9d | 2088 | if (kvm_pmu_is_valid_msr(vcpu, msr)) |
afd80d85 | 2089 | return kvm_pmu_set_msr(vcpu, msr_info); |
5753785f GN |
2090 | |
2091 | if (pr || data != 0) | |
a737f256 CD |
2092 | vcpu_unimpl(vcpu, "disabled perfctr wrmsr: " |
2093 | "0x%x data 0x%llx\n", msr, data); | |
5753785f | 2094 | break; |
84e0cefa JS |
2095 | case MSR_K7_CLK_CTL: |
2096 | /* | |
2097 | * Ignore all writes to this no longer documented MSR. | |
2098 | * Writes are only relevant for old K7 processors, | |
2099 | * all pre-dating SVM, but a recommended workaround from | |
4a969980 | 2100 | * AMD for these chips. It is possible to specify the |
84e0cefa JS |
2101 | * affected processor models on the command line, hence |
2102 | * the need to ignore the workaround. | |
2103 | */ | |
2104 | break; | |
55cd8e5a | 2105 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
e83d5887 | 2106 | return kvm_hv_set_msr_common(vcpu, msr, data); |
91c9c3ed | 2107 | case MSR_IA32_BBL_CR_CTL3: |
2108 | /* Drop writes to this legacy MSR -- see rdmsr | |
2109 | * counterpart for further detail. | |
2110 | */ | |
a737f256 | 2111 | vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data); |
91c9c3ed | 2112 | break; |
2b036c6b BO |
2113 | case MSR_AMD64_OSVW_ID_LENGTH: |
2114 | if (!guest_cpuid_has_osvw(vcpu)) | |
2115 | return 1; | |
2116 | vcpu->arch.osvw.length = data; | |
2117 | break; | |
2118 | case MSR_AMD64_OSVW_STATUS: | |
2119 | if (!guest_cpuid_has_osvw(vcpu)) | |
2120 | return 1; | |
2121 | vcpu->arch.osvw.status = data; | |
2122 | break; | |
15c4a640 | 2123 | default: |
ffde22ac ES |
2124 | if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr)) |
2125 | return xen_hvm_config(vcpu, data); | |
c6702c9d | 2126 | if (kvm_pmu_is_valid_msr(vcpu, msr)) |
afd80d85 | 2127 | return kvm_pmu_set_msr(vcpu, msr_info); |
ed85c068 | 2128 | if (!ignore_msrs) { |
a737f256 CD |
2129 | vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", |
2130 | msr, data); | |
ed85c068 AP |
2131 | return 1; |
2132 | } else { | |
a737f256 CD |
2133 | vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", |
2134 | msr, data); | |
ed85c068 AP |
2135 | break; |
2136 | } | |
15c4a640 CO |
2137 | } |
2138 | return 0; | |
2139 | } | |
2140 | EXPORT_SYMBOL_GPL(kvm_set_msr_common); | |
2141 | ||
2142 | ||
2143 | /* | |
2144 | * Reads an msr value (of 'msr_index') into 'pdata'. | |
2145 | * Returns 0 on success, non-0 otherwise. | |
2146 | * Assumes vcpu_load() was already called. | |
2147 | */ | |
609e36d3 | 2148 | int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr) |
15c4a640 | 2149 | { |
609e36d3 | 2150 | return kvm_x86_ops->get_msr(vcpu, msr); |
15c4a640 | 2151 | } |
ff651cb6 | 2152 | EXPORT_SYMBOL_GPL(kvm_get_msr); |
15c4a640 | 2153 | |
890ca9ae | 2154 | static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata) |
15c4a640 CO |
2155 | { |
2156 | u64 data; | |
890ca9ae HY |
2157 | u64 mcg_cap = vcpu->arch.mcg_cap; |
2158 | unsigned bank_num = mcg_cap & 0xff; | |
15c4a640 CO |
2159 | |
2160 | switch (msr) { | |
15c4a640 CO |
2161 | case MSR_IA32_P5_MC_ADDR: |
2162 | case MSR_IA32_P5_MC_TYPE: | |
890ca9ae HY |
2163 | data = 0; |
2164 | break; | |
15c4a640 | 2165 | case MSR_IA32_MCG_CAP: |
890ca9ae HY |
2166 | data = vcpu->arch.mcg_cap; |
2167 | break; | |
c7ac679c | 2168 | case MSR_IA32_MCG_CTL: |
890ca9ae HY |
2169 | if (!(mcg_cap & MCG_CTL_P)) |
2170 | return 1; | |
2171 | data = vcpu->arch.mcg_ctl; | |
2172 | break; | |
2173 | case MSR_IA32_MCG_STATUS: | |
2174 | data = vcpu->arch.mcg_status; | |
2175 | break; | |
2176 | default: | |
2177 | if (msr >= MSR_IA32_MC0_CTL && | |
81760dcc | 2178 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
890ca9ae HY |
2179 | u32 offset = msr - MSR_IA32_MC0_CTL; |
2180 | data = vcpu->arch.mce_banks[offset]; | |
2181 | break; | |
2182 | } | |
2183 | return 1; | |
2184 | } | |
2185 | *pdata = data; | |
2186 | return 0; | |
2187 | } | |
2188 | ||
609e36d3 | 2189 | int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
890ca9ae | 2190 | { |
609e36d3 | 2191 | switch (msr_info->index) { |
890ca9ae | 2192 | case MSR_IA32_PLATFORM_ID: |
15c4a640 | 2193 | case MSR_IA32_EBL_CR_POWERON: |
b5e2fec0 AG |
2194 | case MSR_IA32_DEBUGCTLMSR: |
2195 | case MSR_IA32_LASTBRANCHFROMIP: | |
2196 | case MSR_IA32_LASTBRANCHTOIP: | |
2197 | case MSR_IA32_LASTINTFROMIP: | |
2198 | case MSR_IA32_LASTINTTOIP: | |
60af2ecd JSR |
2199 | case MSR_K8_SYSCFG: |
2200 | case MSR_K7_HWCR: | |
61a6bd67 | 2201 | case MSR_VM_HSAVE_PA: |
1fdbd48c | 2202 | case MSR_K8_INT_PENDING_MSG: |
c323c0e5 | 2203 | case MSR_AMD64_NB_CFG: |
f7c6d140 | 2204 | case MSR_FAM10H_MMIO_CONF_BASE: |
2e32b719 | 2205 | case MSR_AMD64_BU_CFG2: |
609e36d3 | 2206 | msr_info->data = 0; |
15c4a640 | 2207 | break; |
6912ac32 WH |
2208 | case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3: |
2209 | case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3: | |
2210 | case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1: | |
2211 | case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1: | |
c6702c9d | 2212 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
609e36d3 PB |
2213 | return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data); |
2214 | msr_info->data = 0; | |
5753785f | 2215 | break; |
742bc670 | 2216 | case MSR_IA32_UCODE_REV: |
609e36d3 | 2217 | msr_info->data = 0x100000000ULL; |
742bc670 | 2218 | break; |
9ba075a6 | 2219 | case MSR_MTRRcap: |
9ba075a6 | 2220 | case 0x200 ... 0x2ff: |
ff53604b | 2221 | return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data); |
15c4a640 | 2222 | case 0xcd: /* fsb frequency */ |
609e36d3 | 2223 | msr_info->data = 3; |
15c4a640 | 2224 | break; |
7b914098 JS |
2225 | /* |
2226 | * MSR_EBC_FREQUENCY_ID | |
2227 | * Conservative value valid for even the basic CPU models. | |
2228 | * Models 0,1: 000 in bits 23:21 indicating a bus speed of | |
2229 | * 100MHz, model 2 000 in bits 18:16 indicating 100MHz, | |
2230 | * and 266MHz for model 3, or 4. Set Core Clock | |
2231 | * Frequency to System Bus Frequency Ratio to 1 (bits | |
2232 | * 31:24) even though these are only valid for CPU | |
2233 | * models > 2, however guests may end up dividing or | |
2234 | * multiplying by zero otherwise. | |
2235 | */ | |
2236 | case MSR_EBC_FREQUENCY_ID: | |
609e36d3 | 2237 | msr_info->data = 1 << 24; |
7b914098 | 2238 | break; |
15c4a640 | 2239 | case MSR_IA32_APICBASE: |
609e36d3 | 2240 | msr_info->data = kvm_get_apic_base(vcpu); |
15c4a640 | 2241 | break; |
0105d1a5 | 2242 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff: |
609e36d3 | 2243 | return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data); |
0105d1a5 | 2244 | break; |
a3e06bbe | 2245 | case MSR_IA32_TSCDEADLINE: |
609e36d3 | 2246 | msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu); |
a3e06bbe | 2247 | break; |
ba904635 | 2248 | case MSR_IA32_TSC_ADJUST: |
609e36d3 | 2249 | msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr; |
ba904635 | 2250 | break; |
15c4a640 | 2251 | case MSR_IA32_MISC_ENABLE: |
609e36d3 | 2252 | msr_info->data = vcpu->arch.ia32_misc_enable_msr; |
15c4a640 | 2253 | break; |
64d60670 PB |
2254 | case MSR_IA32_SMBASE: |
2255 | if (!msr_info->host_initiated) | |
2256 | return 1; | |
2257 | msr_info->data = vcpu->arch.smbase; | |
15c4a640 | 2258 | break; |
847f0ad8 AG |
2259 | case MSR_IA32_PERF_STATUS: |
2260 | /* TSC increment by tick */ | |
609e36d3 | 2261 | msr_info->data = 1000ULL; |
847f0ad8 | 2262 | /* CPU multiplier */ |
b0996ae4 | 2263 | msr_info->data |= (((uint64_t)4ULL) << 40); |
847f0ad8 | 2264 | break; |
15c4a640 | 2265 | case MSR_EFER: |
609e36d3 | 2266 | msr_info->data = vcpu->arch.efer; |
15c4a640 | 2267 | break; |
18068523 | 2268 | case MSR_KVM_WALL_CLOCK: |
11c6bffa | 2269 | case MSR_KVM_WALL_CLOCK_NEW: |
609e36d3 | 2270 | msr_info->data = vcpu->kvm->arch.wall_clock; |
18068523 GOC |
2271 | break; |
2272 | case MSR_KVM_SYSTEM_TIME: | |
11c6bffa | 2273 | case MSR_KVM_SYSTEM_TIME_NEW: |
609e36d3 | 2274 | msr_info->data = vcpu->arch.time; |
18068523 | 2275 | break; |
344d9588 | 2276 | case MSR_KVM_ASYNC_PF_EN: |
609e36d3 | 2277 | msr_info->data = vcpu->arch.apf.msr_val; |
344d9588 | 2278 | break; |
c9aaa895 | 2279 | case MSR_KVM_STEAL_TIME: |
609e36d3 | 2280 | msr_info->data = vcpu->arch.st.msr_val; |
c9aaa895 | 2281 | break; |
1d92128f | 2282 | case MSR_KVM_PV_EOI_EN: |
609e36d3 | 2283 | msr_info->data = vcpu->arch.pv_eoi.msr_val; |
1d92128f | 2284 | break; |
890ca9ae HY |
2285 | case MSR_IA32_P5_MC_ADDR: |
2286 | case MSR_IA32_P5_MC_TYPE: | |
2287 | case MSR_IA32_MCG_CAP: | |
2288 | case MSR_IA32_MCG_CTL: | |
2289 | case MSR_IA32_MCG_STATUS: | |
81760dcc | 2290 | case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1: |
609e36d3 | 2291 | return get_msr_mce(vcpu, msr_info->index, &msr_info->data); |
84e0cefa JS |
2292 | case MSR_K7_CLK_CTL: |
2293 | /* | |
2294 | * Provide expected ramp-up count for K7. All other | |
2295 | * are set to zero, indicating minimum divisors for | |
2296 | * every field. | |
2297 | * | |
2298 | * This prevents guest kernels on AMD host with CPU | |
2299 | * type 6, model 8 and higher from exploding due to | |
2300 | * the rdmsr failing. | |
2301 | */ | |
609e36d3 | 2302 | msr_info->data = 0x20000000; |
84e0cefa | 2303 | break; |
55cd8e5a | 2304 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
e83d5887 AS |
2305 | return kvm_hv_get_msr_common(vcpu, |
2306 | msr_info->index, &msr_info->data); | |
55cd8e5a | 2307 | break; |
91c9c3ed | 2308 | case MSR_IA32_BBL_CR_CTL3: |
2309 | /* This legacy MSR exists but isn't fully documented in current | |
2310 | * silicon. It is however accessed by winxp in very narrow | |
2311 | * scenarios where it sets bit #19, itself documented as | |
2312 | * a "reserved" bit. Best effort attempt to source coherent | |
2313 | * read data here should the balance of the register be | |
2314 | * interpreted by the guest: | |
2315 | * | |
2316 | * L2 cache control register 3: 64GB range, 256KB size, | |
2317 | * enabled, latency 0x1, configured | |
2318 | */ | |
609e36d3 | 2319 | msr_info->data = 0xbe702111; |
91c9c3ed | 2320 | break; |
2b036c6b BO |
2321 | case MSR_AMD64_OSVW_ID_LENGTH: |
2322 | if (!guest_cpuid_has_osvw(vcpu)) | |
2323 | return 1; | |
609e36d3 | 2324 | msr_info->data = vcpu->arch.osvw.length; |
2b036c6b BO |
2325 | break; |
2326 | case MSR_AMD64_OSVW_STATUS: | |
2327 | if (!guest_cpuid_has_osvw(vcpu)) | |
2328 | return 1; | |
609e36d3 | 2329 | msr_info->data = vcpu->arch.osvw.status; |
2b036c6b | 2330 | break; |
15c4a640 | 2331 | default: |
c6702c9d | 2332 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
609e36d3 | 2333 | return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data); |
ed85c068 | 2334 | if (!ignore_msrs) { |
609e36d3 | 2335 | vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index); |
ed85c068 AP |
2336 | return 1; |
2337 | } else { | |
609e36d3 PB |
2338 | vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index); |
2339 | msr_info->data = 0; | |
ed85c068 AP |
2340 | } |
2341 | break; | |
15c4a640 | 2342 | } |
15c4a640 CO |
2343 | return 0; |
2344 | } | |
2345 | EXPORT_SYMBOL_GPL(kvm_get_msr_common); | |
2346 | ||
313a3dc7 CO |
2347 | /* |
2348 | * Read or write a bunch of msrs. All parameters are kernel addresses. | |
2349 | * | |
2350 | * @return number of msrs set successfully. | |
2351 | */ | |
2352 | static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs, | |
2353 | struct kvm_msr_entry *entries, | |
2354 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
2355 | unsigned index, u64 *data)) | |
2356 | { | |
f656ce01 | 2357 | int i, idx; |
313a3dc7 | 2358 | |
f656ce01 | 2359 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
313a3dc7 CO |
2360 | for (i = 0; i < msrs->nmsrs; ++i) |
2361 | if (do_msr(vcpu, entries[i].index, &entries[i].data)) | |
2362 | break; | |
f656ce01 | 2363 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
313a3dc7 | 2364 | |
313a3dc7 CO |
2365 | return i; |
2366 | } | |
2367 | ||
2368 | /* | |
2369 | * Read or write a bunch of msrs. Parameters are user addresses. | |
2370 | * | |
2371 | * @return number of msrs set successfully. | |
2372 | */ | |
2373 | static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs, | |
2374 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
2375 | unsigned index, u64 *data), | |
2376 | int writeback) | |
2377 | { | |
2378 | struct kvm_msrs msrs; | |
2379 | struct kvm_msr_entry *entries; | |
2380 | int r, n; | |
2381 | unsigned size; | |
2382 | ||
2383 | r = -EFAULT; | |
2384 | if (copy_from_user(&msrs, user_msrs, sizeof msrs)) | |
2385 | goto out; | |
2386 | ||
2387 | r = -E2BIG; | |
2388 | if (msrs.nmsrs >= MAX_IO_MSRS) | |
2389 | goto out; | |
2390 | ||
313a3dc7 | 2391 | size = sizeof(struct kvm_msr_entry) * msrs.nmsrs; |
ff5c2c03 SL |
2392 | entries = memdup_user(user_msrs->entries, size); |
2393 | if (IS_ERR(entries)) { | |
2394 | r = PTR_ERR(entries); | |
313a3dc7 | 2395 | goto out; |
ff5c2c03 | 2396 | } |
313a3dc7 CO |
2397 | |
2398 | r = n = __msr_io(vcpu, &msrs, entries, do_msr); | |
2399 | if (r < 0) | |
2400 | goto out_free; | |
2401 | ||
2402 | r = -EFAULT; | |
2403 | if (writeback && copy_to_user(user_msrs->entries, entries, size)) | |
2404 | goto out_free; | |
2405 | ||
2406 | r = n; | |
2407 | ||
2408 | out_free: | |
7a73c028 | 2409 | kfree(entries); |
313a3dc7 CO |
2410 | out: |
2411 | return r; | |
2412 | } | |
2413 | ||
784aa3d7 | 2414 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
018d00d2 ZX |
2415 | { |
2416 | int r; | |
2417 | ||
2418 | switch (ext) { | |
2419 | case KVM_CAP_IRQCHIP: | |
2420 | case KVM_CAP_HLT: | |
2421 | case KVM_CAP_MMU_SHADOW_CACHE_CONTROL: | |
018d00d2 | 2422 | case KVM_CAP_SET_TSS_ADDR: |
07716717 | 2423 | case KVM_CAP_EXT_CPUID: |
9c15bb1d | 2424 | case KVM_CAP_EXT_EMUL_CPUID: |
c8076604 | 2425 | case KVM_CAP_CLOCKSOURCE: |
7837699f | 2426 | case KVM_CAP_PIT: |
a28e4f5a | 2427 | case KVM_CAP_NOP_IO_DELAY: |
62d9f0db | 2428 | case KVM_CAP_MP_STATE: |
ed848624 | 2429 | case KVM_CAP_SYNC_MMU: |
a355c85c | 2430 | case KVM_CAP_USER_NMI: |
52d939a0 | 2431 | case KVM_CAP_REINJECT_CONTROL: |
4925663a | 2432 | case KVM_CAP_IRQ_INJECT_STATUS: |
d34e6b17 | 2433 | case KVM_CAP_IOEVENTFD: |
f848a5a8 | 2434 | case KVM_CAP_IOEVENTFD_NO_LENGTH: |
c5ff41ce | 2435 | case KVM_CAP_PIT2: |
e9f42757 | 2436 | case KVM_CAP_PIT_STATE2: |
b927a3ce | 2437 | case KVM_CAP_SET_IDENTITY_MAP_ADDR: |
ffde22ac | 2438 | case KVM_CAP_XEN_HVM: |
afbcf7ab | 2439 | case KVM_CAP_ADJUST_CLOCK: |
3cfc3092 | 2440 | case KVM_CAP_VCPU_EVENTS: |
55cd8e5a | 2441 | case KVM_CAP_HYPERV: |
10388a07 | 2442 | case KVM_CAP_HYPERV_VAPIC: |
c25bc163 | 2443 | case KVM_CAP_HYPERV_SPIN: |
ab9f4ecb | 2444 | case KVM_CAP_PCI_SEGMENT: |
a1efbe77 | 2445 | case KVM_CAP_DEBUGREGS: |
d2be1651 | 2446 | case KVM_CAP_X86_ROBUST_SINGLESTEP: |
2d5b5a66 | 2447 | case KVM_CAP_XSAVE: |
344d9588 | 2448 | case KVM_CAP_ASYNC_PF: |
92a1f12d | 2449 | case KVM_CAP_GET_TSC_KHZ: |
1c0b28c2 | 2450 | case KVM_CAP_KVMCLOCK_CTRL: |
4d8b81ab | 2451 | case KVM_CAP_READONLY_MEM: |
5f66b620 | 2452 | case KVM_CAP_HYPERV_TIME: |
100943c5 | 2453 | case KVM_CAP_IOAPIC_POLARITY_IGNORED: |
defcf51f | 2454 | case KVM_CAP_TSC_DEADLINE_TIMER: |
90de4a18 NA |
2455 | case KVM_CAP_ENABLE_CAP_VM: |
2456 | case KVM_CAP_DISABLE_QUIRKS: | |
2a5bab10 AW |
2457 | #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT |
2458 | case KVM_CAP_ASSIGN_DEV_IRQ: | |
2459 | case KVM_CAP_PCI_2_3: | |
2460 | #endif | |
018d00d2 ZX |
2461 | r = 1; |
2462 | break; | |
6d396b55 PB |
2463 | case KVM_CAP_X86_SMM: |
2464 | /* SMBASE is usually relocated above 1M on modern chipsets, | |
2465 | * and SMM handlers might indeed rely on 4G segment limits, | |
2466 | * so do not report SMM to be available if real mode is | |
2467 | * emulated via vm86 mode. Still, do not go to great lengths | |
2468 | * to avoid userspace's usage of the feature, because it is a | |
2469 | * fringe case that is not enabled except via specific settings | |
2470 | * of the module parameters. | |
2471 | */ | |
2472 | r = kvm_x86_ops->cpu_has_high_real_mode_segbase(); | |
2473 | break; | |
542472b5 LV |
2474 | case KVM_CAP_COALESCED_MMIO: |
2475 | r = KVM_COALESCED_MMIO_PAGE_OFFSET; | |
2476 | break; | |
774ead3a AK |
2477 | case KVM_CAP_VAPIC: |
2478 | r = !kvm_x86_ops->cpu_has_accelerated_tpr(); | |
2479 | break; | |
f725230a | 2480 | case KVM_CAP_NR_VCPUS: |
8c3ba334 SL |
2481 | r = KVM_SOFT_MAX_VCPUS; |
2482 | break; | |
2483 | case KVM_CAP_MAX_VCPUS: | |
f725230a AK |
2484 | r = KVM_MAX_VCPUS; |
2485 | break; | |
a988b910 | 2486 | case KVM_CAP_NR_MEMSLOTS: |
bbacc0c1 | 2487 | r = KVM_USER_MEM_SLOTS; |
a988b910 | 2488 | break; |
a68a6a72 MT |
2489 | case KVM_CAP_PV_MMU: /* obsolete */ |
2490 | r = 0; | |
2f333bcb | 2491 | break; |
4cee4b72 | 2492 | #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT |
62c476c7 | 2493 | case KVM_CAP_IOMMU: |
a1b60c1c | 2494 | r = iommu_present(&pci_bus_type); |
62c476c7 | 2495 | break; |
4cee4b72 | 2496 | #endif |
890ca9ae HY |
2497 | case KVM_CAP_MCE: |
2498 | r = KVM_MAX_MCE_BANKS; | |
2499 | break; | |
2d5b5a66 SY |
2500 | case KVM_CAP_XCRS: |
2501 | r = cpu_has_xsave; | |
2502 | break; | |
92a1f12d JR |
2503 | case KVM_CAP_TSC_CONTROL: |
2504 | r = kvm_has_tsc_control; | |
2505 | break; | |
018d00d2 ZX |
2506 | default: |
2507 | r = 0; | |
2508 | break; | |
2509 | } | |
2510 | return r; | |
2511 | ||
2512 | } | |
2513 | ||
043405e1 CO |
2514 | long kvm_arch_dev_ioctl(struct file *filp, |
2515 | unsigned int ioctl, unsigned long arg) | |
2516 | { | |
2517 | void __user *argp = (void __user *)arg; | |
2518 | long r; | |
2519 | ||
2520 | switch (ioctl) { | |
2521 | case KVM_GET_MSR_INDEX_LIST: { | |
2522 | struct kvm_msr_list __user *user_msr_list = argp; | |
2523 | struct kvm_msr_list msr_list; | |
2524 | unsigned n; | |
2525 | ||
2526 | r = -EFAULT; | |
2527 | if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list)) | |
2528 | goto out; | |
2529 | n = msr_list.nmsrs; | |
62ef68bb | 2530 | msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs; |
043405e1 CO |
2531 | if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list)) |
2532 | goto out; | |
2533 | r = -E2BIG; | |
e125e7b6 | 2534 | if (n < msr_list.nmsrs) |
043405e1 CO |
2535 | goto out; |
2536 | r = -EFAULT; | |
2537 | if (copy_to_user(user_msr_list->indices, &msrs_to_save, | |
2538 | num_msrs_to_save * sizeof(u32))) | |
2539 | goto out; | |
e125e7b6 | 2540 | if (copy_to_user(user_msr_list->indices + num_msrs_to_save, |
043405e1 | 2541 | &emulated_msrs, |
62ef68bb | 2542 | num_emulated_msrs * sizeof(u32))) |
043405e1 CO |
2543 | goto out; |
2544 | r = 0; | |
2545 | break; | |
2546 | } | |
9c15bb1d BP |
2547 | case KVM_GET_SUPPORTED_CPUID: |
2548 | case KVM_GET_EMULATED_CPUID: { | |
674eea0f AK |
2549 | struct kvm_cpuid2 __user *cpuid_arg = argp; |
2550 | struct kvm_cpuid2 cpuid; | |
2551 | ||
2552 | r = -EFAULT; | |
2553 | if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid)) | |
2554 | goto out; | |
9c15bb1d BP |
2555 | |
2556 | r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries, | |
2557 | ioctl); | |
674eea0f AK |
2558 | if (r) |
2559 | goto out; | |
2560 | ||
2561 | r = -EFAULT; | |
2562 | if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid)) | |
2563 | goto out; | |
2564 | r = 0; | |
2565 | break; | |
2566 | } | |
890ca9ae HY |
2567 | case KVM_X86_GET_MCE_CAP_SUPPORTED: { |
2568 | u64 mce_cap; | |
2569 | ||
2570 | mce_cap = KVM_MCE_CAP_SUPPORTED; | |
2571 | r = -EFAULT; | |
2572 | if (copy_to_user(argp, &mce_cap, sizeof mce_cap)) | |
2573 | goto out; | |
2574 | r = 0; | |
2575 | break; | |
2576 | } | |
043405e1 CO |
2577 | default: |
2578 | r = -EINVAL; | |
2579 | } | |
2580 | out: | |
2581 | return r; | |
2582 | } | |
2583 | ||
f5f48ee1 SY |
2584 | static void wbinvd_ipi(void *garbage) |
2585 | { | |
2586 | wbinvd(); | |
2587 | } | |
2588 | ||
2589 | static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
2590 | { | |
e0f0bbc5 | 2591 | return kvm_arch_has_noncoherent_dma(vcpu->kvm); |
f5f48ee1 SY |
2592 | } |
2593 | ||
313a3dc7 CO |
2594 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
2595 | { | |
f5f48ee1 SY |
2596 | /* Address WBINVD may be executed by guest */ |
2597 | if (need_emulate_wbinvd(vcpu)) { | |
2598 | if (kvm_x86_ops->has_wbinvd_exit()) | |
2599 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); | |
2600 | else if (vcpu->cpu != -1 && vcpu->cpu != cpu) | |
2601 | smp_call_function_single(vcpu->cpu, | |
2602 | wbinvd_ipi, NULL, 1); | |
2603 | } | |
2604 | ||
313a3dc7 | 2605 | kvm_x86_ops->vcpu_load(vcpu, cpu); |
8f6055cb | 2606 | |
0dd6a6ed ZA |
2607 | /* Apply any externally detected TSC adjustments (due to suspend) */ |
2608 | if (unlikely(vcpu->arch.tsc_offset_adjustment)) { | |
2609 | adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment); | |
2610 | vcpu->arch.tsc_offset_adjustment = 0; | |
105b21bb | 2611 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0dd6a6ed | 2612 | } |
8f6055cb | 2613 | |
48434c20 | 2614 | if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) { |
6f526ec5 ZA |
2615 | s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 : |
2616 | native_read_tsc() - vcpu->arch.last_host_tsc; | |
e48672fa ZA |
2617 | if (tsc_delta < 0) |
2618 | mark_tsc_unstable("KVM discovered backwards TSC"); | |
c285545f | 2619 | if (check_tsc_unstable()) { |
b183aa58 ZA |
2620 | u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu, |
2621 | vcpu->arch.last_guest_tsc); | |
2622 | kvm_x86_ops->write_tsc_offset(vcpu, offset); | |
c285545f | 2623 | vcpu->arch.tsc_catchup = 1; |
c285545f | 2624 | } |
d98d07ca MT |
2625 | /* |
2626 | * On a host with synchronized TSC, there is no need to update | |
2627 | * kvmclock on vcpu->cpu migration | |
2628 | */ | |
2629 | if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1) | |
0061d53d | 2630 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); |
c285545f ZA |
2631 | if (vcpu->cpu != cpu) |
2632 | kvm_migrate_timers(vcpu); | |
e48672fa | 2633 | vcpu->cpu = cpu; |
6b7d7e76 | 2634 | } |
c9aaa895 GC |
2635 | |
2636 | accumulate_steal_time(vcpu); | |
2637 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); | |
313a3dc7 CO |
2638 | } |
2639 | ||
2640 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) | |
2641 | { | |
02daab21 | 2642 | kvm_x86_ops->vcpu_put(vcpu); |
1c11e713 | 2643 | kvm_put_guest_fpu(vcpu); |
6f526ec5 | 2644 | vcpu->arch.last_host_tsc = native_read_tsc(); |
313a3dc7 CO |
2645 | } |
2646 | ||
313a3dc7 CO |
2647 | static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu, |
2648 | struct kvm_lapic_state *s) | |
2649 | { | |
5a71785d | 2650 | kvm_x86_ops->sync_pir_to_irr(vcpu); |
ad312c7c | 2651 | memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s); |
313a3dc7 CO |
2652 | |
2653 | return 0; | |
2654 | } | |
2655 | ||
2656 | static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu, | |
2657 | struct kvm_lapic_state *s) | |
2658 | { | |
64eb0620 | 2659 | kvm_apic_post_state_restore(vcpu, s); |
cb142eb7 | 2660 | update_cr8_intercept(vcpu); |
313a3dc7 CO |
2661 | |
2662 | return 0; | |
2663 | } | |
2664 | ||
f77bc6a4 ZX |
2665 | static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, |
2666 | struct kvm_interrupt *irq) | |
2667 | { | |
02cdb50f | 2668 | if (irq->irq >= KVM_NR_INTERRUPTS) |
f77bc6a4 ZX |
2669 | return -EINVAL; |
2670 | if (irqchip_in_kernel(vcpu->kvm)) | |
2671 | return -ENXIO; | |
f77bc6a4 | 2672 | |
66fd3f7f | 2673 | kvm_queue_interrupt(vcpu, irq->irq, false); |
3842d135 | 2674 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
f77bc6a4 | 2675 | |
f77bc6a4 ZX |
2676 | return 0; |
2677 | } | |
2678 | ||
c4abb7c9 JK |
2679 | static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu) |
2680 | { | |
c4abb7c9 | 2681 | kvm_inject_nmi(vcpu); |
c4abb7c9 JK |
2682 | |
2683 | return 0; | |
2684 | } | |
2685 | ||
f077825a PB |
2686 | static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu) |
2687 | { | |
64d60670 PB |
2688 | kvm_make_request(KVM_REQ_SMI, vcpu); |
2689 | ||
f077825a PB |
2690 | return 0; |
2691 | } | |
2692 | ||
b209749f AK |
2693 | static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu, |
2694 | struct kvm_tpr_access_ctl *tac) | |
2695 | { | |
2696 | if (tac->flags) | |
2697 | return -EINVAL; | |
2698 | vcpu->arch.tpr_access_reporting = !!tac->enabled; | |
2699 | return 0; | |
2700 | } | |
2701 | ||
890ca9ae HY |
2702 | static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu, |
2703 | u64 mcg_cap) | |
2704 | { | |
2705 | int r; | |
2706 | unsigned bank_num = mcg_cap & 0xff, bank; | |
2707 | ||
2708 | r = -EINVAL; | |
a9e38c3e | 2709 | if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS) |
890ca9ae HY |
2710 | goto out; |
2711 | if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000)) | |
2712 | goto out; | |
2713 | r = 0; | |
2714 | vcpu->arch.mcg_cap = mcg_cap; | |
2715 | /* Init IA32_MCG_CTL to all 1s */ | |
2716 | if (mcg_cap & MCG_CTL_P) | |
2717 | vcpu->arch.mcg_ctl = ~(u64)0; | |
2718 | /* Init IA32_MCi_CTL to all 1s */ | |
2719 | for (bank = 0; bank < bank_num; bank++) | |
2720 | vcpu->arch.mce_banks[bank*4] = ~(u64)0; | |
2721 | out: | |
2722 | return r; | |
2723 | } | |
2724 | ||
2725 | static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu, | |
2726 | struct kvm_x86_mce *mce) | |
2727 | { | |
2728 | u64 mcg_cap = vcpu->arch.mcg_cap; | |
2729 | unsigned bank_num = mcg_cap & 0xff; | |
2730 | u64 *banks = vcpu->arch.mce_banks; | |
2731 | ||
2732 | if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL)) | |
2733 | return -EINVAL; | |
2734 | /* | |
2735 | * if IA32_MCG_CTL is not all 1s, the uncorrected error | |
2736 | * reporting is disabled | |
2737 | */ | |
2738 | if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) && | |
2739 | vcpu->arch.mcg_ctl != ~(u64)0) | |
2740 | return 0; | |
2741 | banks += 4 * mce->bank; | |
2742 | /* | |
2743 | * if IA32_MCi_CTL is not all 1s, the uncorrected error | |
2744 | * reporting is disabled for the bank | |
2745 | */ | |
2746 | if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0) | |
2747 | return 0; | |
2748 | if (mce->status & MCI_STATUS_UC) { | |
2749 | if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) || | |
fc78f519 | 2750 | !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) { |
a8eeb04a | 2751 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
890ca9ae HY |
2752 | return 0; |
2753 | } | |
2754 | if (banks[1] & MCI_STATUS_VAL) | |
2755 | mce->status |= MCI_STATUS_OVER; | |
2756 | banks[2] = mce->addr; | |
2757 | banks[3] = mce->misc; | |
2758 | vcpu->arch.mcg_status = mce->mcg_status; | |
2759 | banks[1] = mce->status; | |
2760 | kvm_queue_exception(vcpu, MC_VECTOR); | |
2761 | } else if (!(banks[1] & MCI_STATUS_VAL) | |
2762 | || !(banks[1] & MCI_STATUS_UC)) { | |
2763 | if (banks[1] & MCI_STATUS_VAL) | |
2764 | mce->status |= MCI_STATUS_OVER; | |
2765 | banks[2] = mce->addr; | |
2766 | banks[3] = mce->misc; | |
2767 | banks[1] = mce->status; | |
2768 | } else | |
2769 | banks[1] |= MCI_STATUS_OVER; | |
2770 | return 0; | |
2771 | } | |
2772 | ||
3cfc3092 JK |
2773 | static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu, |
2774 | struct kvm_vcpu_events *events) | |
2775 | { | |
7460fb4a | 2776 | process_nmi(vcpu); |
03b82a30 JK |
2777 | events->exception.injected = |
2778 | vcpu->arch.exception.pending && | |
2779 | !kvm_exception_is_soft(vcpu->arch.exception.nr); | |
3cfc3092 JK |
2780 | events->exception.nr = vcpu->arch.exception.nr; |
2781 | events->exception.has_error_code = vcpu->arch.exception.has_error_code; | |
97e69aa6 | 2782 | events->exception.pad = 0; |
3cfc3092 JK |
2783 | events->exception.error_code = vcpu->arch.exception.error_code; |
2784 | ||
03b82a30 JK |
2785 | events->interrupt.injected = |
2786 | vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft; | |
3cfc3092 | 2787 | events->interrupt.nr = vcpu->arch.interrupt.nr; |
03b82a30 | 2788 | events->interrupt.soft = 0; |
37ccdcbe | 2789 | events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu); |
3cfc3092 JK |
2790 | |
2791 | events->nmi.injected = vcpu->arch.nmi_injected; | |
7460fb4a | 2792 | events->nmi.pending = vcpu->arch.nmi_pending != 0; |
3cfc3092 | 2793 | events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu); |
97e69aa6 | 2794 | events->nmi.pad = 0; |
3cfc3092 | 2795 | |
66450a21 | 2796 | events->sipi_vector = 0; /* never valid when reporting to user space */ |
3cfc3092 | 2797 | |
f077825a PB |
2798 | events->smi.smm = is_smm(vcpu); |
2799 | events->smi.pending = vcpu->arch.smi_pending; | |
2800 | events->smi.smm_inside_nmi = | |
2801 | !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK); | |
2802 | events->smi.latched_init = kvm_lapic_latched_init(vcpu); | |
2803 | ||
dab4b911 | 2804 | events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING |
f077825a PB |
2805 | | KVM_VCPUEVENT_VALID_SHADOW |
2806 | | KVM_VCPUEVENT_VALID_SMM); | |
97e69aa6 | 2807 | memset(&events->reserved, 0, sizeof(events->reserved)); |
3cfc3092 JK |
2808 | } |
2809 | ||
2810 | static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu, | |
2811 | struct kvm_vcpu_events *events) | |
2812 | { | |
dab4b911 | 2813 | if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING |
48005f64 | 2814 | | KVM_VCPUEVENT_VALID_SIPI_VECTOR |
f077825a PB |
2815 | | KVM_VCPUEVENT_VALID_SHADOW |
2816 | | KVM_VCPUEVENT_VALID_SMM)) | |
3cfc3092 JK |
2817 | return -EINVAL; |
2818 | ||
7460fb4a | 2819 | process_nmi(vcpu); |
3cfc3092 JK |
2820 | vcpu->arch.exception.pending = events->exception.injected; |
2821 | vcpu->arch.exception.nr = events->exception.nr; | |
2822 | vcpu->arch.exception.has_error_code = events->exception.has_error_code; | |
2823 | vcpu->arch.exception.error_code = events->exception.error_code; | |
2824 | ||
2825 | vcpu->arch.interrupt.pending = events->interrupt.injected; | |
2826 | vcpu->arch.interrupt.nr = events->interrupt.nr; | |
2827 | vcpu->arch.interrupt.soft = events->interrupt.soft; | |
48005f64 JK |
2828 | if (events->flags & KVM_VCPUEVENT_VALID_SHADOW) |
2829 | kvm_x86_ops->set_interrupt_shadow(vcpu, | |
2830 | events->interrupt.shadow); | |
3cfc3092 JK |
2831 | |
2832 | vcpu->arch.nmi_injected = events->nmi.injected; | |
dab4b911 JK |
2833 | if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING) |
2834 | vcpu->arch.nmi_pending = events->nmi.pending; | |
3cfc3092 JK |
2835 | kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked); |
2836 | ||
66450a21 JK |
2837 | if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR && |
2838 | kvm_vcpu_has_lapic(vcpu)) | |
2839 | vcpu->arch.apic->sipi_vector = events->sipi_vector; | |
3cfc3092 | 2840 | |
f077825a PB |
2841 | if (events->flags & KVM_VCPUEVENT_VALID_SMM) { |
2842 | if (events->smi.smm) | |
2843 | vcpu->arch.hflags |= HF_SMM_MASK; | |
2844 | else | |
2845 | vcpu->arch.hflags &= ~HF_SMM_MASK; | |
2846 | vcpu->arch.smi_pending = events->smi.pending; | |
2847 | if (events->smi.smm_inside_nmi) | |
2848 | vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK; | |
2849 | else | |
2850 | vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK; | |
2851 | if (kvm_vcpu_has_lapic(vcpu)) { | |
2852 | if (events->smi.latched_init) | |
2853 | set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events); | |
2854 | else | |
2855 | clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events); | |
2856 | } | |
2857 | } | |
2858 | ||
3842d135 AK |
2859 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
2860 | ||
3cfc3092 JK |
2861 | return 0; |
2862 | } | |
2863 | ||
a1efbe77 JK |
2864 | static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu, |
2865 | struct kvm_debugregs *dbgregs) | |
2866 | { | |
73aaf249 JK |
2867 | unsigned long val; |
2868 | ||
a1efbe77 | 2869 | memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db)); |
16f8a6f9 | 2870 | kvm_get_dr(vcpu, 6, &val); |
73aaf249 | 2871 | dbgregs->dr6 = val; |
a1efbe77 JK |
2872 | dbgregs->dr7 = vcpu->arch.dr7; |
2873 | dbgregs->flags = 0; | |
97e69aa6 | 2874 | memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved)); |
a1efbe77 JK |
2875 | } |
2876 | ||
2877 | static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu, | |
2878 | struct kvm_debugregs *dbgregs) | |
2879 | { | |
2880 | if (dbgregs->flags) | |
2881 | return -EINVAL; | |
2882 | ||
a1efbe77 | 2883 | memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db)); |
ae561ede | 2884 | kvm_update_dr0123(vcpu); |
a1efbe77 | 2885 | vcpu->arch.dr6 = dbgregs->dr6; |
73aaf249 | 2886 | kvm_update_dr6(vcpu); |
a1efbe77 | 2887 | vcpu->arch.dr7 = dbgregs->dr7; |
9926c9fd | 2888 | kvm_update_dr7(vcpu); |
a1efbe77 | 2889 | |
a1efbe77 JK |
2890 | return 0; |
2891 | } | |
2892 | ||
df1daba7 PB |
2893 | #define XSTATE_COMPACTION_ENABLED (1ULL << 63) |
2894 | ||
2895 | static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu) | |
2896 | { | |
c47ada30 | 2897 | struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave; |
400e4b20 | 2898 | u64 xstate_bv = xsave->header.xfeatures; |
df1daba7 PB |
2899 | u64 valid; |
2900 | ||
2901 | /* | |
2902 | * Copy legacy XSAVE area, to avoid complications with CPUID | |
2903 | * leaves 0 and 1 in the loop below. | |
2904 | */ | |
2905 | memcpy(dest, xsave, XSAVE_HDR_OFFSET); | |
2906 | ||
2907 | /* Set XSTATE_BV */ | |
2908 | *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv; | |
2909 | ||
2910 | /* | |
2911 | * Copy each region from the possibly compacted offset to the | |
2912 | * non-compacted offset. | |
2913 | */ | |
2914 | valid = xstate_bv & ~XSTATE_FPSSE; | |
2915 | while (valid) { | |
2916 | u64 feature = valid & -valid; | |
2917 | int index = fls64(feature) - 1; | |
2918 | void *src = get_xsave_addr(xsave, feature); | |
2919 | ||
2920 | if (src) { | |
2921 | u32 size, offset, ecx, edx; | |
2922 | cpuid_count(XSTATE_CPUID, index, | |
2923 | &size, &offset, &ecx, &edx); | |
2924 | memcpy(dest + offset, src, size); | |
2925 | } | |
2926 | ||
2927 | valid -= feature; | |
2928 | } | |
2929 | } | |
2930 | ||
2931 | static void load_xsave(struct kvm_vcpu *vcpu, u8 *src) | |
2932 | { | |
c47ada30 | 2933 | struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave; |
df1daba7 PB |
2934 | u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET); |
2935 | u64 valid; | |
2936 | ||
2937 | /* | |
2938 | * Copy legacy XSAVE area, to avoid complications with CPUID | |
2939 | * leaves 0 and 1 in the loop below. | |
2940 | */ | |
2941 | memcpy(xsave, src, XSAVE_HDR_OFFSET); | |
2942 | ||
2943 | /* Set XSTATE_BV and possibly XCOMP_BV. */ | |
400e4b20 | 2944 | xsave->header.xfeatures = xstate_bv; |
df1daba7 | 2945 | if (cpu_has_xsaves) |
3a54450b | 2946 | xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED; |
df1daba7 PB |
2947 | |
2948 | /* | |
2949 | * Copy each region from the non-compacted offset to the | |
2950 | * possibly compacted offset. | |
2951 | */ | |
2952 | valid = xstate_bv & ~XSTATE_FPSSE; | |
2953 | while (valid) { | |
2954 | u64 feature = valid & -valid; | |
2955 | int index = fls64(feature) - 1; | |
2956 | void *dest = get_xsave_addr(xsave, feature); | |
2957 | ||
2958 | if (dest) { | |
2959 | u32 size, offset, ecx, edx; | |
2960 | cpuid_count(XSTATE_CPUID, index, | |
2961 | &size, &offset, &ecx, &edx); | |
2962 | memcpy(dest, src + offset, size); | |
ee4100da | 2963 | } |
df1daba7 PB |
2964 | |
2965 | valid -= feature; | |
2966 | } | |
2967 | } | |
2968 | ||
2d5b5a66 SY |
2969 | static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu, |
2970 | struct kvm_xsave *guest_xsave) | |
2971 | { | |
4344ee98 | 2972 | if (cpu_has_xsave) { |
df1daba7 PB |
2973 | memset(guest_xsave, 0, sizeof(struct kvm_xsave)); |
2974 | fill_xsave((u8 *) guest_xsave->region, vcpu); | |
4344ee98 | 2975 | } else { |
2d5b5a66 | 2976 | memcpy(guest_xsave->region, |
7366ed77 | 2977 | &vcpu->arch.guest_fpu.state.fxsave, |
c47ada30 | 2978 | sizeof(struct fxregs_state)); |
2d5b5a66 SY |
2979 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] = |
2980 | XSTATE_FPSSE; | |
2981 | } | |
2982 | } | |
2983 | ||
2984 | static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu, | |
2985 | struct kvm_xsave *guest_xsave) | |
2986 | { | |
2987 | u64 xstate_bv = | |
2988 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)]; | |
2989 | ||
d7876f1b PB |
2990 | if (cpu_has_xsave) { |
2991 | /* | |
2992 | * Here we allow setting states that are not present in | |
2993 | * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility | |
2994 | * with old userspace. | |
2995 | */ | |
4ff41732 | 2996 | if (xstate_bv & ~kvm_supported_xcr0()) |
d7876f1b | 2997 | return -EINVAL; |
df1daba7 | 2998 | load_xsave(vcpu, (u8 *)guest_xsave->region); |
d7876f1b | 2999 | } else { |
2d5b5a66 SY |
3000 | if (xstate_bv & ~XSTATE_FPSSE) |
3001 | return -EINVAL; | |
7366ed77 | 3002 | memcpy(&vcpu->arch.guest_fpu.state.fxsave, |
c47ada30 | 3003 | guest_xsave->region, sizeof(struct fxregs_state)); |
2d5b5a66 SY |
3004 | } |
3005 | return 0; | |
3006 | } | |
3007 | ||
3008 | static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu, | |
3009 | struct kvm_xcrs *guest_xcrs) | |
3010 | { | |
3011 | if (!cpu_has_xsave) { | |
3012 | guest_xcrs->nr_xcrs = 0; | |
3013 | return; | |
3014 | } | |
3015 | ||
3016 | guest_xcrs->nr_xcrs = 1; | |
3017 | guest_xcrs->flags = 0; | |
3018 | guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK; | |
3019 | guest_xcrs->xcrs[0].value = vcpu->arch.xcr0; | |
3020 | } | |
3021 | ||
3022 | static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu, | |
3023 | struct kvm_xcrs *guest_xcrs) | |
3024 | { | |
3025 | int i, r = 0; | |
3026 | ||
3027 | if (!cpu_has_xsave) | |
3028 | return -EINVAL; | |
3029 | ||
3030 | if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags) | |
3031 | return -EINVAL; | |
3032 | ||
3033 | for (i = 0; i < guest_xcrs->nr_xcrs; i++) | |
3034 | /* Only support XCR0 currently */ | |
c67a04cb | 3035 | if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) { |
2d5b5a66 | 3036 | r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK, |
c67a04cb | 3037 | guest_xcrs->xcrs[i].value); |
2d5b5a66 SY |
3038 | break; |
3039 | } | |
3040 | if (r) | |
3041 | r = -EINVAL; | |
3042 | return r; | |
3043 | } | |
3044 | ||
1c0b28c2 EM |
3045 | /* |
3046 | * kvm_set_guest_paused() indicates to the guest kernel that it has been | |
3047 | * stopped by the hypervisor. This function will be called from the host only. | |
3048 | * EINVAL is returned when the host attempts to set the flag for a guest that | |
3049 | * does not support pv clocks. | |
3050 | */ | |
3051 | static int kvm_set_guest_paused(struct kvm_vcpu *vcpu) | |
3052 | { | |
0b79459b | 3053 | if (!vcpu->arch.pv_time_enabled) |
1c0b28c2 | 3054 | return -EINVAL; |
51d59c6b | 3055 | vcpu->arch.pvclock_set_guest_stopped_request = true; |
1c0b28c2 EM |
3056 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
3057 | return 0; | |
3058 | } | |
3059 | ||
313a3dc7 CO |
3060 | long kvm_arch_vcpu_ioctl(struct file *filp, |
3061 | unsigned int ioctl, unsigned long arg) | |
3062 | { | |
3063 | struct kvm_vcpu *vcpu = filp->private_data; | |
3064 | void __user *argp = (void __user *)arg; | |
3065 | int r; | |
d1ac91d8 AK |
3066 | union { |
3067 | struct kvm_lapic_state *lapic; | |
3068 | struct kvm_xsave *xsave; | |
3069 | struct kvm_xcrs *xcrs; | |
3070 | void *buffer; | |
3071 | } u; | |
3072 | ||
3073 | u.buffer = NULL; | |
313a3dc7 CO |
3074 | switch (ioctl) { |
3075 | case KVM_GET_LAPIC: { | |
2204ae3c MT |
3076 | r = -EINVAL; |
3077 | if (!vcpu->arch.apic) | |
3078 | goto out; | |
d1ac91d8 | 3079 | u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL); |
313a3dc7 | 3080 | |
b772ff36 | 3081 | r = -ENOMEM; |
d1ac91d8 | 3082 | if (!u.lapic) |
b772ff36 | 3083 | goto out; |
d1ac91d8 | 3084 | r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic); |
313a3dc7 CO |
3085 | if (r) |
3086 | goto out; | |
3087 | r = -EFAULT; | |
d1ac91d8 | 3088 | if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state))) |
313a3dc7 CO |
3089 | goto out; |
3090 | r = 0; | |
3091 | break; | |
3092 | } | |
3093 | case KVM_SET_LAPIC: { | |
2204ae3c MT |
3094 | r = -EINVAL; |
3095 | if (!vcpu->arch.apic) | |
3096 | goto out; | |
ff5c2c03 | 3097 | u.lapic = memdup_user(argp, sizeof(*u.lapic)); |
18595411 GC |
3098 | if (IS_ERR(u.lapic)) |
3099 | return PTR_ERR(u.lapic); | |
ff5c2c03 | 3100 | |
d1ac91d8 | 3101 | r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic); |
313a3dc7 CO |
3102 | break; |
3103 | } | |
f77bc6a4 ZX |
3104 | case KVM_INTERRUPT: { |
3105 | struct kvm_interrupt irq; | |
3106 | ||
3107 | r = -EFAULT; | |
3108 | if (copy_from_user(&irq, argp, sizeof irq)) | |
3109 | goto out; | |
3110 | r = kvm_vcpu_ioctl_interrupt(vcpu, &irq); | |
f77bc6a4 ZX |
3111 | break; |
3112 | } | |
c4abb7c9 JK |
3113 | case KVM_NMI: { |
3114 | r = kvm_vcpu_ioctl_nmi(vcpu); | |
c4abb7c9 JK |
3115 | break; |
3116 | } | |
f077825a PB |
3117 | case KVM_SMI: { |
3118 | r = kvm_vcpu_ioctl_smi(vcpu); | |
3119 | break; | |
3120 | } | |
313a3dc7 CO |
3121 | case KVM_SET_CPUID: { |
3122 | struct kvm_cpuid __user *cpuid_arg = argp; | |
3123 | struct kvm_cpuid cpuid; | |
3124 | ||
3125 | r = -EFAULT; | |
3126 | if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid)) | |
3127 | goto out; | |
3128 | r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries); | |
313a3dc7 CO |
3129 | break; |
3130 | } | |
07716717 DK |
3131 | case KVM_SET_CPUID2: { |
3132 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
3133 | struct kvm_cpuid2 cpuid; | |
3134 | ||
3135 | r = -EFAULT; | |
3136 | if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid)) | |
3137 | goto out; | |
3138 | r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid, | |
19355475 | 3139 | cpuid_arg->entries); |
07716717 DK |
3140 | break; |
3141 | } | |
3142 | case KVM_GET_CPUID2: { | |
3143 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
3144 | struct kvm_cpuid2 cpuid; | |
3145 | ||
3146 | r = -EFAULT; | |
3147 | if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid)) | |
3148 | goto out; | |
3149 | r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid, | |
19355475 | 3150 | cpuid_arg->entries); |
07716717 DK |
3151 | if (r) |
3152 | goto out; | |
3153 | r = -EFAULT; | |
3154 | if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid)) | |
3155 | goto out; | |
3156 | r = 0; | |
3157 | break; | |
3158 | } | |
313a3dc7 | 3159 | case KVM_GET_MSRS: |
609e36d3 | 3160 | r = msr_io(vcpu, argp, do_get_msr, 1); |
313a3dc7 CO |
3161 | break; |
3162 | case KVM_SET_MSRS: | |
3163 | r = msr_io(vcpu, argp, do_set_msr, 0); | |
3164 | break; | |
b209749f AK |
3165 | case KVM_TPR_ACCESS_REPORTING: { |
3166 | struct kvm_tpr_access_ctl tac; | |
3167 | ||
3168 | r = -EFAULT; | |
3169 | if (copy_from_user(&tac, argp, sizeof tac)) | |
3170 | goto out; | |
3171 | r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac); | |
3172 | if (r) | |
3173 | goto out; | |
3174 | r = -EFAULT; | |
3175 | if (copy_to_user(argp, &tac, sizeof tac)) | |
3176 | goto out; | |
3177 | r = 0; | |
3178 | break; | |
3179 | }; | |
b93463aa AK |
3180 | case KVM_SET_VAPIC_ADDR: { |
3181 | struct kvm_vapic_addr va; | |
3182 | ||
3183 | r = -EINVAL; | |
3184 | if (!irqchip_in_kernel(vcpu->kvm)) | |
3185 | goto out; | |
3186 | r = -EFAULT; | |
3187 | if (copy_from_user(&va, argp, sizeof va)) | |
3188 | goto out; | |
fda4e2e8 | 3189 | r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr); |
b93463aa AK |
3190 | break; |
3191 | } | |
890ca9ae HY |
3192 | case KVM_X86_SETUP_MCE: { |
3193 | u64 mcg_cap; | |
3194 | ||
3195 | r = -EFAULT; | |
3196 | if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap)) | |
3197 | goto out; | |
3198 | r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap); | |
3199 | break; | |
3200 | } | |
3201 | case KVM_X86_SET_MCE: { | |
3202 | struct kvm_x86_mce mce; | |
3203 | ||
3204 | r = -EFAULT; | |
3205 | if (copy_from_user(&mce, argp, sizeof mce)) | |
3206 | goto out; | |
3207 | r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce); | |
3208 | break; | |
3209 | } | |
3cfc3092 JK |
3210 | case KVM_GET_VCPU_EVENTS: { |
3211 | struct kvm_vcpu_events events; | |
3212 | ||
3213 | kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events); | |
3214 | ||
3215 | r = -EFAULT; | |
3216 | if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events))) | |
3217 | break; | |
3218 | r = 0; | |
3219 | break; | |
3220 | } | |
3221 | case KVM_SET_VCPU_EVENTS: { | |
3222 | struct kvm_vcpu_events events; | |
3223 | ||
3224 | r = -EFAULT; | |
3225 | if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events))) | |
3226 | break; | |
3227 | ||
3228 | r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events); | |
3229 | break; | |
3230 | } | |
a1efbe77 JK |
3231 | case KVM_GET_DEBUGREGS: { |
3232 | struct kvm_debugregs dbgregs; | |
3233 | ||
3234 | kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs); | |
3235 | ||
3236 | r = -EFAULT; | |
3237 | if (copy_to_user(argp, &dbgregs, | |
3238 | sizeof(struct kvm_debugregs))) | |
3239 | break; | |
3240 | r = 0; | |
3241 | break; | |
3242 | } | |
3243 | case KVM_SET_DEBUGREGS: { | |
3244 | struct kvm_debugregs dbgregs; | |
3245 | ||
3246 | r = -EFAULT; | |
3247 | if (copy_from_user(&dbgregs, argp, | |
3248 | sizeof(struct kvm_debugregs))) | |
3249 | break; | |
3250 | ||
3251 | r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs); | |
3252 | break; | |
3253 | } | |
2d5b5a66 | 3254 | case KVM_GET_XSAVE: { |
d1ac91d8 | 3255 | u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL); |
2d5b5a66 | 3256 | r = -ENOMEM; |
d1ac91d8 | 3257 | if (!u.xsave) |
2d5b5a66 SY |
3258 | break; |
3259 | ||
d1ac91d8 | 3260 | kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
3261 | |
3262 | r = -EFAULT; | |
d1ac91d8 | 3263 | if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave))) |
2d5b5a66 SY |
3264 | break; |
3265 | r = 0; | |
3266 | break; | |
3267 | } | |
3268 | case KVM_SET_XSAVE: { | |
ff5c2c03 | 3269 | u.xsave = memdup_user(argp, sizeof(*u.xsave)); |
18595411 GC |
3270 | if (IS_ERR(u.xsave)) |
3271 | return PTR_ERR(u.xsave); | |
2d5b5a66 | 3272 | |
d1ac91d8 | 3273 | r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
3274 | break; |
3275 | } | |
3276 | case KVM_GET_XCRS: { | |
d1ac91d8 | 3277 | u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL); |
2d5b5a66 | 3278 | r = -ENOMEM; |
d1ac91d8 | 3279 | if (!u.xcrs) |
2d5b5a66 SY |
3280 | break; |
3281 | ||
d1ac91d8 | 3282 | kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
3283 | |
3284 | r = -EFAULT; | |
d1ac91d8 | 3285 | if (copy_to_user(argp, u.xcrs, |
2d5b5a66 SY |
3286 | sizeof(struct kvm_xcrs))) |
3287 | break; | |
3288 | r = 0; | |
3289 | break; | |
3290 | } | |
3291 | case KVM_SET_XCRS: { | |
ff5c2c03 | 3292 | u.xcrs = memdup_user(argp, sizeof(*u.xcrs)); |
18595411 GC |
3293 | if (IS_ERR(u.xcrs)) |
3294 | return PTR_ERR(u.xcrs); | |
2d5b5a66 | 3295 | |
d1ac91d8 | 3296 | r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
3297 | break; |
3298 | } | |
92a1f12d JR |
3299 | case KVM_SET_TSC_KHZ: { |
3300 | u32 user_tsc_khz; | |
3301 | ||
3302 | r = -EINVAL; | |
92a1f12d JR |
3303 | user_tsc_khz = (u32)arg; |
3304 | ||
3305 | if (user_tsc_khz >= kvm_max_guest_tsc_khz) | |
3306 | goto out; | |
3307 | ||
cc578287 ZA |
3308 | if (user_tsc_khz == 0) |
3309 | user_tsc_khz = tsc_khz; | |
3310 | ||
3311 | kvm_set_tsc_khz(vcpu, user_tsc_khz); | |
92a1f12d JR |
3312 | |
3313 | r = 0; | |
3314 | goto out; | |
3315 | } | |
3316 | case KVM_GET_TSC_KHZ: { | |
cc578287 | 3317 | r = vcpu->arch.virtual_tsc_khz; |
92a1f12d JR |
3318 | goto out; |
3319 | } | |
1c0b28c2 EM |
3320 | case KVM_KVMCLOCK_CTRL: { |
3321 | r = kvm_set_guest_paused(vcpu); | |
3322 | goto out; | |
3323 | } | |
313a3dc7 CO |
3324 | default: |
3325 | r = -EINVAL; | |
3326 | } | |
3327 | out: | |
d1ac91d8 | 3328 | kfree(u.buffer); |
313a3dc7 CO |
3329 | return r; |
3330 | } | |
3331 | ||
5b1c1493 CO |
3332 | int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
3333 | { | |
3334 | return VM_FAULT_SIGBUS; | |
3335 | } | |
3336 | ||
1fe779f8 CO |
3337 | static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr) |
3338 | { | |
3339 | int ret; | |
3340 | ||
3341 | if (addr > (unsigned int)(-3 * PAGE_SIZE)) | |
951179ce | 3342 | return -EINVAL; |
1fe779f8 CO |
3343 | ret = kvm_x86_ops->set_tss_addr(kvm, addr); |
3344 | return ret; | |
3345 | } | |
3346 | ||
b927a3ce SY |
3347 | static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm, |
3348 | u64 ident_addr) | |
3349 | { | |
3350 | kvm->arch.ept_identity_map_addr = ident_addr; | |
3351 | return 0; | |
3352 | } | |
3353 | ||
1fe779f8 CO |
3354 | static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm, |
3355 | u32 kvm_nr_mmu_pages) | |
3356 | { | |
3357 | if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES) | |
3358 | return -EINVAL; | |
3359 | ||
79fac95e | 3360 | mutex_lock(&kvm->slots_lock); |
1fe779f8 CO |
3361 | |
3362 | kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages); | |
f05e70ac | 3363 | kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages; |
1fe779f8 | 3364 | |
79fac95e | 3365 | mutex_unlock(&kvm->slots_lock); |
1fe779f8 CO |
3366 | return 0; |
3367 | } | |
3368 | ||
3369 | static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm) | |
3370 | { | |
39de71ec | 3371 | return kvm->arch.n_max_mmu_pages; |
1fe779f8 CO |
3372 | } |
3373 | ||
1fe779f8 CO |
3374 | static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) |
3375 | { | |
3376 | int r; | |
3377 | ||
3378 | r = 0; | |
3379 | switch (chip->chip_id) { | |
3380 | case KVM_IRQCHIP_PIC_MASTER: | |
3381 | memcpy(&chip->chip.pic, | |
3382 | &pic_irqchip(kvm)->pics[0], | |
3383 | sizeof(struct kvm_pic_state)); | |
3384 | break; | |
3385 | case KVM_IRQCHIP_PIC_SLAVE: | |
3386 | memcpy(&chip->chip.pic, | |
3387 | &pic_irqchip(kvm)->pics[1], | |
3388 | sizeof(struct kvm_pic_state)); | |
3389 | break; | |
3390 | case KVM_IRQCHIP_IOAPIC: | |
eba0226b | 3391 | r = kvm_get_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
3392 | break; |
3393 | default: | |
3394 | r = -EINVAL; | |
3395 | break; | |
3396 | } | |
3397 | return r; | |
3398 | } | |
3399 | ||
3400 | static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) | |
3401 | { | |
3402 | int r; | |
3403 | ||
3404 | r = 0; | |
3405 | switch (chip->chip_id) { | |
3406 | case KVM_IRQCHIP_PIC_MASTER: | |
f4f51050 | 3407 | spin_lock(&pic_irqchip(kvm)->lock); |
1fe779f8 CO |
3408 | memcpy(&pic_irqchip(kvm)->pics[0], |
3409 | &chip->chip.pic, | |
3410 | sizeof(struct kvm_pic_state)); | |
f4f51050 | 3411 | spin_unlock(&pic_irqchip(kvm)->lock); |
1fe779f8 CO |
3412 | break; |
3413 | case KVM_IRQCHIP_PIC_SLAVE: | |
f4f51050 | 3414 | spin_lock(&pic_irqchip(kvm)->lock); |
1fe779f8 CO |
3415 | memcpy(&pic_irqchip(kvm)->pics[1], |
3416 | &chip->chip.pic, | |
3417 | sizeof(struct kvm_pic_state)); | |
f4f51050 | 3418 | spin_unlock(&pic_irqchip(kvm)->lock); |
1fe779f8 CO |
3419 | break; |
3420 | case KVM_IRQCHIP_IOAPIC: | |
eba0226b | 3421 | r = kvm_set_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
3422 | break; |
3423 | default: | |
3424 | r = -EINVAL; | |
3425 | break; | |
3426 | } | |
3427 | kvm_pic_update_irq(pic_irqchip(kvm)); | |
3428 | return r; | |
3429 | } | |
3430 | ||
e0f63cb9 SY |
3431 | static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps) |
3432 | { | |
3433 | int r = 0; | |
3434 | ||
894a9c55 | 3435 | mutex_lock(&kvm->arch.vpit->pit_state.lock); |
e0f63cb9 | 3436 | memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state)); |
894a9c55 | 3437 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); |
e0f63cb9 SY |
3438 | return r; |
3439 | } | |
3440 | ||
3441 | static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps) | |
3442 | { | |
3443 | int r = 0; | |
3444 | ||
894a9c55 | 3445 | mutex_lock(&kvm->arch.vpit->pit_state.lock); |
e0f63cb9 | 3446 | memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state)); |
e9f42757 BK |
3447 | kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0); |
3448 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); | |
3449 | return r; | |
3450 | } | |
3451 | ||
3452 | static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
3453 | { | |
3454 | int r = 0; | |
3455 | ||
3456 | mutex_lock(&kvm->arch.vpit->pit_state.lock); | |
3457 | memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels, | |
3458 | sizeof(ps->channels)); | |
3459 | ps->flags = kvm->arch.vpit->pit_state.flags; | |
3460 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); | |
97e69aa6 | 3461 | memset(&ps->reserved, 0, sizeof(ps->reserved)); |
e9f42757 BK |
3462 | return r; |
3463 | } | |
3464 | ||
3465 | static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
3466 | { | |
3467 | int r = 0, start = 0; | |
3468 | u32 prev_legacy, cur_legacy; | |
3469 | mutex_lock(&kvm->arch.vpit->pit_state.lock); | |
3470 | prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY; | |
3471 | cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY; | |
3472 | if (!prev_legacy && cur_legacy) | |
3473 | start = 1; | |
3474 | memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels, | |
3475 | sizeof(kvm->arch.vpit->pit_state.channels)); | |
3476 | kvm->arch.vpit->pit_state.flags = ps->flags; | |
3477 | kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start); | |
894a9c55 | 3478 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); |
e0f63cb9 SY |
3479 | return r; |
3480 | } | |
3481 | ||
52d939a0 MT |
3482 | static int kvm_vm_ioctl_reinject(struct kvm *kvm, |
3483 | struct kvm_reinject_control *control) | |
3484 | { | |
3485 | if (!kvm->arch.vpit) | |
3486 | return -ENXIO; | |
894a9c55 | 3487 | mutex_lock(&kvm->arch.vpit->pit_state.lock); |
26ef1924 | 3488 | kvm->arch.vpit->pit_state.reinject = control->pit_reinject; |
894a9c55 | 3489 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); |
52d939a0 MT |
3490 | return 0; |
3491 | } | |
3492 | ||
95d4c16c | 3493 | /** |
60c34612 TY |
3494 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot |
3495 | * @kvm: kvm instance | |
3496 | * @log: slot id and address to which we copy the log | |
95d4c16c | 3497 | * |
e108ff2f PB |
3498 | * Steps 1-4 below provide general overview of dirty page logging. See |
3499 | * kvm_get_dirty_log_protect() function description for additional details. | |
3500 | * | |
3501 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
3502 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
3503 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
3504 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
3505 | * writes will be marked dirty for next log read. | |
95d4c16c | 3506 | * |
60c34612 TY |
3507 | * 1. Take a snapshot of the bit and clear it if needed. |
3508 | * 2. Write protect the corresponding page. | |
e108ff2f PB |
3509 | * 3. Copy the snapshot to the userspace. |
3510 | * 4. Flush TLB's if needed. | |
5bb064dc | 3511 | */ |
60c34612 | 3512 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
5bb064dc | 3513 | { |
60c34612 | 3514 | bool is_dirty = false; |
e108ff2f | 3515 | int r; |
5bb064dc | 3516 | |
79fac95e | 3517 | mutex_lock(&kvm->slots_lock); |
5bb064dc | 3518 | |
88178fd4 KH |
3519 | /* |
3520 | * Flush potentially hardware-cached dirty pages to dirty_bitmap. | |
3521 | */ | |
3522 | if (kvm_x86_ops->flush_log_dirty) | |
3523 | kvm_x86_ops->flush_log_dirty(kvm); | |
3524 | ||
e108ff2f | 3525 | r = kvm_get_dirty_log_protect(kvm, log, &is_dirty); |
198c74f4 XG |
3526 | |
3527 | /* | |
3528 | * All the TLBs can be flushed out of mmu lock, see the comments in | |
3529 | * kvm_mmu_slot_remove_write_access(). | |
3530 | */ | |
e108ff2f | 3531 | lockdep_assert_held(&kvm->slots_lock); |
198c74f4 XG |
3532 | if (is_dirty) |
3533 | kvm_flush_remote_tlbs(kvm); | |
3534 | ||
79fac95e | 3535 | mutex_unlock(&kvm->slots_lock); |
5bb064dc ZX |
3536 | return r; |
3537 | } | |
3538 | ||
aa2fbe6d YZ |
3539 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event, |
3540 | bool line_status) | |
23d43cf9 CD |
3541 | { |
3542 | if (!irqchip_in_kernel(kvm)) | |
3543 | return -ENXIO; | |
3544 | ||
3545 | irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, | |
aa2fbe6d YZ |
3546 | irq_event->irq, irq_event->level, |
3547 | line_status); | |
23d43cf9 CD |
3548 | return 0; |
3549 | } | |
3550 | ||
90de4a18 NA |
3551 | static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
3552 | struct kvm_enable_cap *cap) | |
3553 | { | |
3554 | int r; | |
3555 | ||
3556 | if (cap->flags) | |
3557 | return -EINVAL; | |
3558 | ||
3559 | switch (cap->cap) { | |
3560 | case KVM_CAP_DISABLE_QUIRKS: | |
3561 | kvm->arch.disabled_quirks = cap->args[0]; | |
3562 | r = 0; | |
3563 | break; | |
3564 | default: | |
3565 | r = -EINVAL; | |
3566 | break; | |
3567 | } | |
3568 | return r; | |
3569 | } | |
3570 | ||
1fe779f8 CO |
3571 | long kvm_arch_vm_ioctl(struct file *filp, |
3572 | unsigned int ioctl, unsigned long arg) | |
3573 | { | |
3574 | struct kvm *kvm = filp->private_data; | |
3575 | void __user *argp = (void __user *)arg; | |
367e1319 | 3576 | int r = -ENOTTY; |
f0d66275 DH |
3577 | /* |
3578 | * This union makes it completely explicit to gcc-3.x | |
3579 | * that these two variables' stack usage should be | |
3580 | * combined, not added together. | |
3581 | */ | |
3582 | union { | |
3583 | struct kvm_pit_state ps; | |
e9f42757 | 3584 | struct kvm_pit_state2 ps2; |
c5ff41ce | 3585 | struct kvm_pit_config pit_config; |
f0d66275 | 3586 | } u; |
1fe779f8 CO |
3587 | |
3588 | switch (ioctl) { | |
3589 | case KVM_SET_TSS_ADDR: | |
3590 | r = kvm_vm_ioctl_set_tss_addr(kvm, arg); | |
1fe779f8 | 3591 | break; |
b927a3ce SY |
3592 | case KVM_SET_IDENTITY_MAP_ADDR: { |
3593 | u64 ident_addr; | |
3594 | ||
3595 | r = -EFAULT; | |
3596 | if (copy_from_user(&ident_addr, argp, sizeof ident_addr)) | |
3597 | goto out; | |
3598 | r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr); | |
b927a3ce SY |
3599 | break; |
3600 | } | |
1fe779f8 CO |
3601 | case KVM_SET_NR_MMU_PAGES: |
3602 | r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg); | |
1fe779f8 CO |
3603 | break; |
3604 | case KVM_GET_NR_MMU_PAGES: | |
3605 | r = kvm_vm_ioctl_get_nr_mmu_pages(kvm); | |
3606 | break; | |
3ddea128 MT |
3607 | case KVM_CREATE_IRQCHIP: { |
3608 | struct kvm_pic *vpic; | |
3609 | ||
3610 | mutex_lock(&kvm->lock); | |
3611 | r = -EEXIST; | |
3612 | if (kvm->arch.vpic) | |
3613 | goto create_irqchip_unlock; | |
3e515705 AK |
3614 | r = -EINVAL; |
3615 | if (atomic_read(&kvm->online_vcpus)) | |
3616 | goto create_irqchip_unlock; | |
1fe779f8 | 3617 | r = -ENOMEM; |
3ddea128 MT |
3618 | vpic = kvm_create_pic(kvm); |
3619 | if (vpic) { | |
1fe779f8 CO |
3620 | r = kvm_ioapic_init(kvm); |
3621 | if (r) { | |
175504cd | 3622 | mutex_lock(&kvm->slots_lock); |
72bb2fcd | 3623 | kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, |
743eeb0b SL |
3624 | &vpic->dev_master); |
3625 | kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, | |
3626 | &vpic->dev_slave); | |
3627 | kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, | |
3628 | &vpic->dev_eclr); | |
175504cd | 3629 | mutex_unlock(&kvm->slots_lock); |
3ddea128 MT |
3630 | kfree(vpic); |
3631 | goto create_irqchip_unlock; | |
1fe779f8 CO |
3632 | } |
3633 | } else | |
3ddea128 MT |
3634 | goto create_irqchip_unlock; |
3635 | smp_wmb(); | |
3636 | kvm->arch.vpic = vpic; | |
3637 | smp_wmb(); | |
399ec807 AK |
3638 | r = kvm_setup_default_irq_routing(kvm); |
3639 | if (r) { | |
175504cd | 3640 | mutex_lock(&kvm->slots_lock); |
3ddea128 | 3641 | mutex_lock(&kvm->irq_lock); |
72bb2fcd WY |
3642 | kvm_ioapic_destroy(kvm); |
3643 | kvm_destroy_pic(kvm); | |
3ddea128 | 3644 | mutex_unlock(&kvm->irq_lock); |
175504cd | 3645 | mutex_unlock(&kvm->slots_lock); |
399ec807 | 3646 | } |
3ddea128 MT |
3647 | create_irqchip_unlock: |
3648 | mutex_unlock(&kvm->lock); | |
1fe779f8 | 3649 | break; |
3ddea128 | 3650 | } |
7837699f | 3651 | case KVM_CREATE_PIT: |
c5ff41ce JK |
3652 | u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY; |
3653 | goto create_pit; | |
3654 | case KVM_CREATE_PIT2: | |
3655 | r = -EFAULT; | |
3656 | if (copy_from_user(&u.pit_config, argp, | |
3657 | sizeof(struct kvm_pit_config))) | |
3658 | goto out; | |
3659 | create_pit: | |
79fac95e | 3660 | mutex_lock(&kvm->slots_lock); |
269e05e4 AK |
3661 | r = -EEXIST; |
3662 | if (kvm->arch.vpit) | |
3663 | goto create_pit_unlock; | |
7837699f | 3664 | r = -ENOMEM; |
c5ff41ce | 3665 | kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags); |
7837699f SY |
3666 | if (kvm->arch.vpit) |
3667 | r = 0; | |
269e05e4 | 3668 | create_pit_unlock: |
79fac95e | 3669 | mutex_unlock(&kvm->slots_lock); |
7837699f | 3670 | break; |
1fe779f8 CO |
3671 | case KVM_GET_IRQCHIP: { |
3672 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 3673 | struct kvm_irqchip *chip; |
1fe779f8 | 3674 | |
ff5c2c03 SL |
3675 | chip = memdup_user(argp, sizeof(*chip)); |
3676 | if (IS_ERR(chip)) { | |
3677 | r = PTR_ERR(chip); | |
1fe779f8 | 3678 | goto out; |
ff5c2c03 SL |
3679 | } |
3680 | ||
1fe779f8 CO |
3681 | r = -ENXIO; |
3682 | if (!irqchip_in_kernel(kvm)) | |
f0d66275 DH |
3683 | goto get_irqchip_out; |
3684 | r = kvm_vm_ioctl_get_irqchip(kvm, chip); | |
1fe779f8 | 3685 | if (r) |
f0d66275 | 3686 | goto get_irqchip_out; |
1fe779f8 | 3687 | r = -EFAULT; |
f0d66275 DH |
3688 | if (copy_to_user(argp, chip, sizeof *chip)) |
3689 | goto get_irqchip_out; | |
1fe779f8 | 3690 | r = 0; |
f0d66275 DH |
3691 | get_irqchip_out: |
3692 | kfree(chip); | |
1fe779f8 CO |
3693 | break; |
3694 | } | |
3695 | case KVM_SET_IRQCHIP: { | |
3696 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 3697 | struct kvm_irqchip *chip; |
1fe779f8 | 3698 | |
ff5c2c03 SL |
3699 | chip = memdup_user(argp, sizeof(*chip)); |
3700 | if (IS_ERR(chip)) { | |
3701 | r = PTR_ERR(chip); | |
1fe779f8 | 3702 | goto out; |
ff5c2c03 SL |
3703 | } |
3704 | ||
1fe779f8 CO |
3705 | r = -ENXIO; |
3706 | if (!irqchip_in_kernel(kvm)) | |
f0d66275 DH |
3707 | goto set_irqchip_out; |
3708 | r = kvm_vm_ioctl_set_irqchip(kvm, chip); | |
1fe779f8 | 3709 | if (r) |
f0d66275 | 3710 | goto set_irqchip_out; |
1fe779f8 | 3711 | r = 0; |
f0d66275 DH |
3712 | set_irqchip_out: |
3713 | kfree(chip); | |
1fe779f8 CO |
3714 | break; |
3715 | } | |
e0f63cb9 | 3716 | case KVM_GET_PIT: { |
e0f63cb9 | 3717 | r = -EFAULT; |
f0d66275 | 3718 | if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
3719 | goto out; |
3720 | r = -ENXIO; | |
3721 | if (!kvm->arch.vpit) | |
3722 | goto out; | |
f0d66275 | 3723 | r = kvm_vm_ioctl_get_pit(kvm, &u.ps); |
e0f63cb9 SY |
3724 | if (r) |
3725 | goto out; | |
3726 | r = -EFAULT; | |
f0d66275 | 3727 | if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
3728 | goto out; |
3729 | r = 0; | |
3730 | break; | |
3731 | } | |
3732 | case KVM_SET_PIT: { | |
e0f63cb9 | 3733 | r = -EFAULT; |
f0d66275 | 3734 | if (copy_from_user(&u.ps, argp, sizeof u.ps)) |
e0f63cb9 SY |
3735 | goto out; |
3736 | r = -ENXIO; | |
3737 | if (!kvm->arch.vpit) | |
3738 | goto out; | |
f0d66275 | 3739 | r = kvm_vm_ioctl_set_pit(kvm, &u.ps); |
e0f63cb9 SY |
3740 | break; |
3741 | } | |
e9f42757 BK |
3742 | case KVM_GET_PIT2: { |
3743 | r = -ENXIO; | |
3744 | if (!kvm->arch.vpit) | |
3745 | goto out; | |
3746 | r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2); | |
3747 | if (r) | |
3748 | goto out; | |
3749 | r = -EFAULT; | |
3750 | if (copy_to_user(argp, &u.ps2, sizeof(u.ps2))) | |
3751 | goto out; | |
3752 | r = 0; | |
3753 | break; | |
3754 | } | |
3755 | case KVM_SET_PIT2: { | |
3756 | r = -EFAULT; | |
3757 | if (copy_from_user(&u.ps2, argp, sizeof(u.ps2))) | |
3758 | goto out; | |
3759 | r = -ENXIO; | |
3760 | if (!kvm->arch.vpit) | |
3761 | goto out; | |
3762 | r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2); | |
e9f42757 BK |
3763 | break; |
3764 | } | |
52d939a0 MT |
3765 | case KVM_REINJECT_CONTROL: { |
3766 | struct kvm_reinject_control control; | |
3767 | r = -EFAULT; | |
3768 | if (copy_from_user(&control, argp, sizeof(control))) | |
3769 | goto out; | |
3770 | r = kvm_vm_ioctl_reinject(kvm, &control); | |
52d939a0 MT |
3771 | break; |
3772 | } | |
ffde22ac ES |
3773 | case KVM_XEN_HVM_CONFIG: { |
3774 | r = -EFAULT; | |
3775 | if (copy_from_user(&kvm->arch.xen_hvm_config, argp, | |
3776 | sizeof(struct kvm_xen_hvm_config))) | |
3777 | goto out; | |
3778 | r = -EINVAL; | |
3779 | if (kvm->arch.xen_hvm_config.flags) | |
3780 | goto out; | |
3781 | r = 0; | |
3782 | break; | |
3783 | } | |
afbcf7ab | 3784 | case KVM_SET_CLOCK: { |
afbcf7ab GC |
3785 | struct kvm_clock_data user_ns; |
3786 | u64 now_ns; | |
3787 | s64 delta; | |
3788 | ||
3789 | r = -EFAULT; | |
3790 | if (copy_from_user(&user_ns, argp, sizeof(user_ns))) | |
3791 | goto out; | |
3792 | ||
3793 | r = -EINVAL; | |
3794 | if (user_ns.flags) | |
3795 | goto out; | |
3796 | ||
3797 | r = 0; | |
395c6b0a | 3798 | local_irq_disable(); |
759379dd | 3799 | now_ns = get_kernel_ns(); |
afbcf7ab | 3800 | delta = user_ns.clock - now_ns; |
395c6b0a | 3801 | local_irq_enable(); |
afbcf7ab | 3802 | kvm->arch.kvmclock_offset = delta; |
2e762ff7 | 3803 | kvm_gen_update_masterclock(kvm); |
afbcf7ab GC |
3804 | break; |
3805 | } | |
3806 | case KVM_GET_CLOCK: { | |
afbcf7ab GC |
3807 | struct kvm_clock_data user_ns; |
3808 | u64 now_ns; | |
3809 | ||
395c6b0a | 3810 | local_irq_disable(); |
759379dd | 3811 | now_ns = get_kernel_ns(); |
afbcf7ab | 3812 | user_ns.clock = kvm->arch.kvmclock_offset + now_ns; |
395c6b0a | 3813 | local_irq_enable(); |
afbcf7ab | 3814 | user_ns.flags = 0; |
97e69aa6 | 3815 | memset(&user_ns.pad, 0, sizeof(user_ns.pad)); |
afbcf7ab GC |
3816 | |
3817 | r = -EFAULT; | |
3818 | if (copy_to_user(argp, &user_ns, sizeof(user_ns))) | |
3819 | goto out; | |
3820 | r = 0; | |
3821 | break; | |
3822 | } | |
90de4a18 NA |
3823 | case KVM_ENABLE_CAP: { |
3824 | struct kvm_enable_cap cap; | |
afbcf7ab | 3825 | |
90de4a18 NA |
3826 | r = -EFAULT; |
3827 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
3828 | goto out; | |
3829 | r = kvm_vm_ioctl_enable_cap(kvm, &cap); | |
3830 | break; | |
3831 | } | |
1fe779f8 | 3832 | default: |
c274e03a | 3833 | r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg); |
1fe779f8 CO |
3834 | } |
3835 | out: | |
3836 | return r; | |
3837 | } | |
3838 | ||
a16b043c | 3839 | static void kvm_init_msr_list(void) |
043405e1 CO |
3840 | { |
3841 | u32 dummy[2]; | |
3842 | unsigned i, j; | |
3843 | ||
62ef68bb | 3844 | for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) { |
043405e1 CO |
3845 | if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0) |
3846 | continue; | |
93c4adc7 PB |
3847 | |
3848 | /* | |
3849 | * Even MSRs that are valid in the host may not be exposed | |
3850 | * to the guests in some cases. We could work around this | |
3851 | * in VMX with the generic MSR save/load machinery, but it | |
3852 | * is not really worthwhile since it will really only | |
3853 | * happen with nested virtualization. | |
3854 | */ | |
3855 | switch (msrs_to_save[i]) { | |
3856 | case MSR_IA32_BNDCFGS: | |
3857 | if (!kvm_x86_ops->mpx_supported()) | |
3858 | continue; | |
3859 | break; | |
3860 | default: | |
3861 | break; | |
3862 | } | |
3863 | ||
043405e1 CO |
3864 | if (j < i) |
3865 | msrs_to_save[j] = msrs_to_save[i]; | |
3866 | j++; | |
3867 | } | |
3868 | num_msrs_to_save = j; | |
62ef68bb PB |
3869 | |
3870 | for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) { | |
3871 | switch (emulated_msrs[i]) { | |
6d396b55 PB |
3872 | case MSR_IA32_SMBASE: |
3873 | if (!kvm_x86_ops->cpu_has_high_real_mode_segbase()) | |
3874 | continue; | |
3875 | break; | |
62ef68bb PB |
3876 | default: |
3877 | break; | |
3878 | } | |
3879 | ||
3880 | if (j < i) | |
3881 | emulated_msrs[j] = emulated_msrs[i]; | |
3882 | j++; | |
3883 | } | |
3884 | num_emulated_msrs = j; | |
043405e1 CO |
3885 | } |
3886 | ||
bda9020e MT |
3887 | static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len, |
3888 | const void *v) | |
bbd9b64e | 3889 | { |
70252a10 AK |
3890 | int handled = 0; |
3891 | int n; | |
3892 | ||
3893 | do { | |
3894 | n = min(len, 8); | |
3895 | if (!(vcpu->arch.apic && | |
e32edf4f NN |
3896 | !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v)) |
3897 | && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v)) | |
70252a10 AK |
3898 | break; |
3899 | handled += n; | |
3900 | addr += n; | |
3901 | len -= n; | |
3902 | v += n; | |
3903 | } while (len); | |
bbd9b64e | 3904 | |
70252a10 | 3905 | return handled; |
bbd9b64e CO |
3906 | } |
3907 | ||
bda9020e | 3908 | static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v) |
bbd9b64e | 3909 | { |
70252a10 AK |
3910 | int handled = 0; |
3911 | int n; | |
3912 | ||
3913 | do { | |
3914 | n = min(len, 8); | |
3915 | if (!(vcpu->arch.apic && | |
e32edf4f NN |
3916 | !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev, |
3917 | addr, n, v)) | |
3918 | && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v)) | |
70252a10 AK |
3919 | break; |
3920 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v); | |
3921 | handled += n; | |
3922 | addr += n; | |
3923 | len -= n; | |
3924 | v += n; | |
3925 | } while (len); | |
bbd9b64e | 3926 | |
70252a10 | 3927 | return handled; |
bbd9b64e CO |
3928 | } |
3929 | ||
2dafc6c2 GN |
3930 | static void kvm_set_segment(struct kvm_vcpu *vcpu, |
3931 | struct kvm_segment *var, int seg) | |
3932 | { | |
3933 | kvm_x86_ops->set_segment(vcpu, var, seg); | |
3934 | } | |
3935 | ||
3936 | void kvm_get_segment(struct kvm_vcpu *vcpu, | |
3937 | struct kvm_segment *var, int seg) | |
3938 | { | |
3939 | kvm_x86_ops->get_segment(vcpu, var, seg); | |
3940 | } | |
3941 | ||
54987b7a PB |
3942 | gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access, |
3943 | struct x86_exception *exception) | |
02f59dc9 JR |
3944 | { |
3945 | gpa_t t_gpa; | |
02f59dc9 JR |
3946 | |
3947 | BUG_ON(!mmu_is_nested(vcpu)); | |
3948 | ||
3949 | /* NPT walks are always user-walks */ | |
3950 | access |= PFERR_USER_MASK; | |
54987b7a | 3951 | t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception); |
02f59dc9 JR |
3952 | |
3953 | return t_gpa; | |
3954 | } | |
3955 | ||
ab9ae313 AK |
3956 | gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, |
3957 | struct x86_exception *exception) | |
1871c602 GN |
3958 | { |
3959 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
ab9ae313 | 3960 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
3961 | } |
3962 | ||
ab9ae313 AK |
3963 | gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, |
3964 | struct x86_exception *exception) | |
1871c602 GN |
3965 | { |
3966 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
3967 | access |= PFERR_FETCH_MASK; | |
ab9ae313 | 3968 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
3969 | } |
3970 | ||
ab9ae313 AK |
3971 | gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, |
3972 | struct x86_exception *exception) | |
1871c602 GN |
3973 | { |
3974 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
3975 | access |= PFERR_WRITE_MASK; | |
ab9ae313 | 3976 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
3977 | } |
3978 | ||
3979 | /* uses this to access any guest's mapped memory without checking CPL */ | |
ab9ae313 AK |
3980 | gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, |
3981 | struct x86_exception *exception) | |
1871c602 | 3982 | { |
ab9ae313 | 3983 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception); |
1871c602 GN |
3984 | } |
3985 | ||
3986 | static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes, | |
3987 | struct kvm_vcpu *vcpu, u32 access, | |
bcc55cba | 3988 | struct x86_exception *exception) |
bbd9b64e CO |
3989 | { |
3990 | void *data = val; | |
10589a46 | 3991 | int r = X86EMUL_CONTINUE; |
bbd9b64e CO |
3992 | |
3993 | while (bytes) { | |
14dfe855 | 3994 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access, |
ab9ae313 | 3995 | exception); |
bbd9b64e | 3996 | unsigned offset = addr & (PAGE_SIZE-1); |
77c2002e | 3997 | unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset); |
bbd9b64e CO |
3998 | int ret; |
3999 | ||
bcc55cba | 4000 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 4001 | return X86EMUL_PROPAGATE_FAULT; |
54bf36aa PB |
4002 | ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data, |
4003 | offset, toread); | |
10589a46 | 4004 | if (ret < 0) { |
c3cd7ffa | 4005 | r = X86EMUL_IO_NEEDED; |
10589a46 MT |
4006 | goto out; |
4007 | } | |
bbd9b64e | 4008 | |
77c2002e IE |
4009 | bytes -= toread; |
4010 | data += toread; | |
4011 | addr += toread; | |
bbd9b64e | 4012 | } |
10589a46 | 4013 | out: |
10589a46 | 4014 | return r; |
bbd9b64e | 4015 | } |
77c2002e | 4016 | |
1871c602 | 4017 | /* used for instruction fetching */ |
0f65dd70 AK |
4018 | static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt, |
4019 | gva_t addr, void *val, unsigned int bytes, | |
bcc55cba | 4020 | struct x86_exception *exception) |
1871c602 | 4021 | { |
0f65dd70 | 4022 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
1871c602 | 4023 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
44583cba PB |
4024 | unsigned offset; |
4025 | int ret; | |
0f65dd70 | 4026 | |
44583cba PB |
4027 | /* Inline kvm_read_guest_virt_helper for speed. */ |
4028 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK, | |
4029 | exception); | |
4030 | if (unlikely(gpa == UNMAPPED_GVA)) | |
4031 | return X86EMUL_PROPAGATE_FAULT; | |
4032 | ||
4033 | offset = addr & (PAGE_SIZE-1); | |
4034 | if (WARN_ON(offset + bytes > PAGE_SIZE)) | |
4035 | bytes = (unsigned)PAGE_SIZE - offset; | |
54bf36aa PB |
4036 | ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val, |
4037 | offset, bytes); | |
44583cba PB |
4038 | if (unlikely(ret < 0)) |
4039 | return X86EMUL_IO_NEEDED; | |
4040 | ||
4041 | return X86EMUL_CONTINUE; | |
1871c602 GN |
4042 | } |
4043 | ||
064aea77 | 4044 | int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt, |
0f65dd70 | 4045 | gva_t addr, void *val, unsigned int bytes, |
bcc55cba | 4046 | struct x86_exception *exception) |
1871c602 | 4047 | { |
0f65dd70 | 4048 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
1871c602 | 4049 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
0f65dd70 | 4050 | |
1871c602 | 4051 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, |
bcc55cba | 4052 | exception); |
1871c602 | 4053 | } |
064aea77 | 4054 | EXPORT_SYMBOL_GPL(kvm_read_guest_virt); |
1871c602 | 4055 | |
0f65dd70 AK |
4056 | static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt, |
4057 | gva_t addr, void *val, unsigned int bytes, | |
bcc55cba | 4058 | struct x86_exception *exception) |
1871c602 | 4059 | { |
0f65dd70 | 4060 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
bcc55cba | 4061 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception); |
1871c602 GN |
4062 | } |
4063 | ||
6a4d7550 | 4064 | int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt, |
0f65dd70 | 4065 | gva_t addr, void *val, |
2dafc6c2 | 4066 | unsigned int bytes, |
bcc55cba | 4067 | struct x86_exception *exception) |
77c2002e | 4068 | { |
0f65dd70 | 4069 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
77c2002e IE |
4070 | void *data = val; |
4071 | int r = X86EMUL_CONTINUE; | |
4072 | ||
4073 | while (bytes) { | |
14dfe855 JR |
4074 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, |
4075 | PFERR_WRITE_MASK, | |
ab9ae313 | 4076 | exception); |
77c2002e IE |
4077 | unsigned offset = addr & (PAGE_SIZE-1); |
4078 | unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset); | |
4079 | int ret; | |
4080 | ||
bcc55cba | 4081 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 4082 | return X86EMUL_PROPAGATE_FAULT; |
54bf36aa | 4083 | ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite); |
77c2002e | 4084 | if (ret < 0) { |
c3cd7ffa | 4085 | r = X86EMUL_IO_NEEDED; |
77c2002e IE |
4086 | goto out; |
4087 | } | |
4088 | ||
4089 | bytes -= towrite; | |
4090 | data += towrite; | |
4091 | addr += towrite; | |
4092 | } | |
4093 | out: | |
4094 | return r; | |
4095 | } | |
6a4d7550 | 4096 | EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system); |
77c2002e | 4097 | |
af7cc7d1 XG |
4098 | static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva, |
4099 | gpa_t *gpa, struct x86_exception *exception, | |
4100 | bool write) | |
4101 | { | |
97d64b78 AK |
4102 | u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0) |
4103 | | (write ? PFERR_WRITE_MASK : 0); | |
af7cc7d1 | 4104 | |
97d64b78 | 4105 | if (vcpu_match_mmio_gva(vcpu, gva) |
97ec8c06 FW |
4106 | && !permission_fault(vcpu, vcpu->arch.walk_mmu, |
4107 | vcpu->arch.access, access)) { | |
bebb106a XG |
4108 | *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT | |
4109 | (gva & (PAGE_SIZE - 1)); | |
4f022648 | 4110 | trace_vcpu_match_mmio(gva, *gpa, write, false); |
bebb106a XG |
4111 | return 1; |
4112 | } | |
4113 | ||
af7cc7d1 XG |
4114 | *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
4115 | ||
4116 | if (*gpa == UNMAPPED_GVA) | |
4117 | return -1; | |
4118 | ||
4119 | /* For APIC access vmexit */ | |
4120 | if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
4121 | return 1; | |
4122 | ||
4f022648 XG |
4123 | if (vcpu_match_mmio_gpa(vcpu, *gpa)) { |
4124 | trace_vcpu_match_mmio(gva, *gpa, write, true); | |
bebb106a | 4125 | return 1; |
4f022648 | 4126 | } |
bebb106a | 4127 | |
af7cc7d1 XG |
4128 | return 0; |
4129 | } | |
4130 | ||
3200f405 | 4131 | int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa, |
bcc55cba | 4132 | const void *val, int bytes) |
bbd9b64e CO |
4133 | { |
4134 | int ret; | |
4135 | ||
54bf36aa | 4136 | ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes); |
9f811285 | 4137 | if (ret < 0) |
bbd9b64e | 4138 | return 0; |
f57f2ef5 | 4139 | kvm_mmu_pte_write(vcpu, gpa, val, bytes); |
bbd9b64e CO |
4140 | return 1; |
4141 | } | |
4142 | ||
77d197b2 XG |
4143 | struct read_write_emulator_ops { |
4144 | int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val, | |
4145 | int bytes); | |
4146 | int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4147 | void *val, int bytes); | |
4148 | int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4149 | int bytes, void *val); | |
4150 | int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4151 | void *val, int bytes); | |
4152 | bool write; | |
4153 | }; | |
4154 | ||
4155 | static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes) | |
4156 | { | |
4157 | if (vcpu->mmio_read_completed) { | |
77d197b2 | 4158 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, |
f78146b0 | 4159 | vcpu->mmio_fragments[0].gpa, *(u64 *)val); |
77d197b2 XG |
4160 | vcpu->mmio_read_completed = 0; |
4161 | return 1; | |
4162 | } | |
4163 | ||
4164 | return 0; | |
4165 | } | |
4166 | ||
4167 | static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4168 | void *val, int bytes) | |
4169 | { | |
54bf36aa | 4170 | return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes); |
77d197b2 XG |
4171 | } |
4172 | ||
4173 | static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4174 | void *val, int bytes) | |
4175 | { | |
4176 | return emulator_write_phys(vcpu, gpa, val, bytes); | |
4177 | } | |
4178 | ||
4179 | static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val) | |
4180 | { | |
4181 | trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val); | |
4182 | return vcpu_mmio_write(vcpu, gpa, bytes, val); | |
4183 | } | |
4184 | ||
4185 | static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4186 | void *val, int bytes) | |
4187 | { | |
4188 | trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0); | |
4189 | return X86EMUL_IO_NEEDED; | |
4190 | } | |
4191 | ||
4192 | static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
4193 | void *val, int bytes) | |
4194 | { | |
f78146b0 AK |
4195 | struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0]; |
4196 | ||
87da7e66 | 4197 | memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len)); |
77d197b2 XG |
4198 | return X86EMUL_CONTINUE; |
4199 | } | |
4200 | ||
0fbe9b0b | 4201 | static const struct read_write_emulator_ops read_emultor = { |
77d197b2 XG |
4202 | .read_write_prepare = read_prepare, |
4203 | .read_write_emulate = read_emulate, | |
4204 | .read_write_mmio = vcpu_mmio_read, | |
4205 | .read_write_exit_mmio = read_exit_mmio, | |
4206 | }; | |
4207 | ||
0fbe9b0b | 4208 | static const struct read_write_emulator_ops write_emultor = { |
77d197b2 XG |
4209 | .read_write_emulate = write_emulate, |
4210 | .read_write_mmio = write_mmio, | |
4211 | .read_write_exit_mmio = write_exit_mmio, | |
4212 | .write = true, | |
4213 | }; | |
4214 | ||
22388a3c XG |
4215 | static int emulator_read_write_onepage(unsigned long addr, void *val, |
4216 | unsigned int bytes, | |
4217 | struct x86_exception *exception, | |
4218 | struct kvm_vcpu *vcpu, | |
0fbe9b0b | 4219 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 4220 | { |
af7cc7d1 XG |
4221 | gpa_t gpa; |
4222 | int handled, ret; | |
22388a3c | 4223 | bool write = ops->write; |
f78146b0 | 4224 | struct kvm_mmio_fragment *frag; |
10589a46 | 4225 | |
22388a3c | 4226 | ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write); |
bbd9b64e | 4227 | |
af7cc7d1 | 4228 | if (ret < 0) |
bbd9b64e | 4229 | return X86EMUL_PROPAGATE_FAULT; |
bbd9b64e CO |
4230 | |
4231 | /* For APIC access vmexit */ | |
af7cc7d1 | 4232 | if (ret) |
bbd9b64e CO |
4233 | goto mmio; |
4234 | ||
22388a3c | 4235 | if (ops->read_write_emulate(vcpu, gpa, val, bytes)) |
bbd9b64e CO |
4236 | return X86EMUL_CONTINUE; |
4237 | ||
4238 | mmio: | |
4239 | /* | |
4240 | * Is this MMIO handled locally? | |
4241 | */ | |
22388a3c | 4242 | handled = ops->read_write_mmio(vcpu, gpa, bytes, val); |
70252a10 | 4243 | if (handled == bytes) |
bbd9b64e | 4244 | return X86EMUL_CONTINUE; |
bbd9b64e | 4245 | |
70252a10 AK |
4246 | gpa += handled; |
4247 | bytes -= handled; | |
4248 | val += handled; | |
4249 | ||
87da7e66 XG |
4250 | WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS); |
4251 | frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++]; | |
4252 | frag->gpa = gpa; | |
4253 | frag->data = val; | |
4254 | frag->len = bytes; | |
f78146b0 | 4255 | return X86EMUL_CONTINUE; |
bbd9b64e CO |
4256 | } |
4257 | ||
52eb5a6d XL |
4258 | static int emulator_read_write(struct x86_emulate_ctxt *ctxt, |
4259 | unsigned long addr, | |
22388a3c XG |
4260 | void *val, unsigned int bytes, |
4261 | struct x86_exception *exception, | |
0fbe9b0b | 4262 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 4263 | { |
0f65dd70 | 4264 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
f78146b0 AK |
4265 | gpa_t gpa; |
4266 | int rc; | |
4267 | ||
4268 | if (ops->read_write_prepare && | |
4269 | ops->read_write_prepare(vcpu, val, bytes)) | |
4270 | return X86EMUL_CONTINUE; | |
4271 | ||
4272 | vcpu->mmio_nr_fragments = 0; | |
0f65dd70 | 4273 | |
bbd9b64e CO |
4274 | /* Crossing a page boundary? */ |
4275 | if (((addr + bytes - 1) ^ addr) & PAGE_MASK) { | |
f78146b0 | 4276 | int now; |
bbd9b64e CO |
4277 | |
4278 | now = -addr & ~PAGE_MASK; | |
22388a3c XG |
4279 | rc = emulator_read_write_onepage(addr, val, now, exception, |
4280 | vcpu, ops); | |
4281 | ||
bbd9b64e CO |
4282 | if (rc != X86EMUL_CONTINUE) |
4283 | return rc; | |
4284 | addr += now; | |
bac15531 NA |
4285 | if (ctxt->mode != X86EMUL_MODE_PROT64) |
4286 | addr = (u32)addr; | |
bbd9b64e CO |
4287 | val += now; |
4288 | bytes -= now; | |
4289 | } | |
22388a3c | 4290 | |
f78146b0 AK |
4291 | rc = emulator_read_write_onepage(addr, val, bytes, exception, |
4292 | vcpu, ops); | |
4293 | if (rc != X86EMUL_CONTINUE) | |
4294 | return rc; | |
4295 | ||
4296 | if (!vcpu->mmio_nr_fragments) | |
4297 | return rc; | |
4298 | ||
4299 | gpa = vcpu->mmio_fragments[0].gpa; | |
4300 | ||
4301 | vcpu->mmio_needed = 1; | |
4302 | vcpu->mmio_cur_fragment = 0; | |
4303 | ||
87da7e66 | 4304 | vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len); |
f78146b0 AK |
4305 | vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write; |
4306 | vcpu->run->exit_reason = KVM_EXIT_MMIO; | |
4307 | vcpu->run->mmio.phys_addr = gpa; | |
4308 | ||
4309 | return ops->read_write_exit_mmio(vcpu, gpa, val, bytes); | |
22388a3c XG |
4310 | } |
4311 | ||
4312 | static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt, | |
4313 | unsigned long addr, | |
4314 | void *val, | |
4315 | unsigned int bytes, | |
4316 | struct x86_exception *exception) | |
4317 | { | |
4318 | return emulator_read_write(ctxt, addr, val, bytes, | |
4319 | exception, &read_emultor); | |
4320 | } | |
4321 | ||
52eb5a6d | 4322 | static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt, |
22388a3c XG |
4323 | unsigned long addr, |
4324 | const void *val, | |
4325 | unsigned int bytes, | |
4326 | struct x86_exception *exception) | |
4327 | { | |
4328 | return emulator_read_write(ctxt, addr, (void *)val, bytes, | |
4329 | exception, &write_emultor); | |
bbd9b64e | 4330 | } |
bbd9b64e | 4331 | |
daea3e73 AK |
4332 | #define CMPXCHG_TYPE(t, ptr, old, new) \ |
4333 | (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old)) | |
4334 | ||
4335 | #ifdef CONFIG_X86_64 | |
4336 | # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new) | |
4337 | #else | |
4338 | # define CMPXCHG64(ptr, old, new) \ | |
9749a6c0 | 4339 | (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old)) |
daea3e73 AK |
4340 | #endif |
4341 | ||
0f65dd70 AK |
4342 | static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt, |
4343 | unsigned long addr, | |
bbd9b64e CO |
4344 | const void *old, |
4345 | const void *new, | |
4346 | unsigned int bytes, | |
0f65dd70 | 4347 | struct x86_exception *exception) |
bbd9b64e | 4348 | { |
0f65dd70 | 4349 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
daea3e73 AK |
4350 | gpa_t gpa; |
4351 | struct page *page; | |
4352 | char *kaddr; | |
4353 | bool exchanged; | |
2bacc55c | 4354 | |
daea3e73 AK |
4355 | /* guests cmpxchg8b have to be emulated atomically */ |
4356 | if (bytes > 8 || (bytes & (bytes - 1))) | |
4357 | goto emul_write; | |
10589a46 | 4358 | |
daea3e73 | 4359 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL); |
2bacc55c | 4360 | |
daea3e73 AK |
4361 | if (gpa == UNMAPPED_GVA || |
4362 | (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
4363 | goto emul_write; | |
2bacc55c | 4364 | |
daea3e73 AK |
4365 | if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK)) |
4366 | goto emul_write; | |
72dc67a6 | 4367 | |
54bf36aa | 4368 | page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT); |
32cad84f | 4369 | if (is_error_page(page)) |
c19b8bd6 | 4370 | goto emul_write; |
72dc67a6 | 4371 | |
8fd75e12 | 4372 | kaddr = kmap_atomic(page); |
daea3e73 AK |
4373 | kaddr += offset_in_page(gpa); |
4374 | switch (bytes) { | |
4375 | case 1: | |
4376 | exchanged = CMPXCHG_TYPE(u8, kaddr, old, new); | |
4377 | break; | |
4378 | case 2: | |
4379 | exchanged = CMPXCHG_TYPE(u16, kaddr, old, new); | |
4380 | break; | |
4381 | case 4: | |
4382 | exchanged = CMPXCHG_TYPE(u32, kaddr, old, new); | |
4383 | break; | |
4384 | case 8: | |
4385 | exchanged = CMPXCHG64(kaddr, old, new); | |
4386 | break; | |
4387 | default: | |
4388 | BUG(); | |
2bacc55c | 4389 | } |
8fd75e12 | 4390 | kunmap_atomic(kaddr); |
daea3e73 AK |
4391 | kvm_release_page_dirty(page); |
4392 | ||
4393 | if (!exchanged) | |
4394 | return X86EMUL_CMPXCHG_FAILED; | |
4395 | ||
54bf36aa | 4396 | kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT); |
f57f2ef5 | 4397 | kvm_mmu_pte_write(vcpu, gpa, new, bytes); |
8f6abd06 GN |
4398 | |
4399 | return X86EMUL_CONTINUE; | |
4a5f48f6 | 4400 | |
3200f405 | 4401 | emul_write: |
daea3e73 | 4402 | printk_once(KERN_WARNING "kvm: emulating exchange as write\n"); |
2bacc55c | 4403 | |
0f65dd70 | 4404 | return emulator_write_emulated(ctxt, addr, new, bytes, exception); |
bbd9b64e CO |
4405 | } |
4406 | ||
cf8f70bf GN |
4407 | static int kernel_pio(struct kvm_vcpu *vcpu, void *pd) |
4408 | { | |
4409 | /* TODO: String I/O for in kernel device */ | |
4410 | int r; | |
4411 | ||
4412 | if (vcpu->arch.pio.in) | |
e32edf4f | 4413 | r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port, |
cf8f70bf GN |
4414 | vcpu->arch.pio.size, pd); |
4415 | else | |
e32edf4f | 4416 | r = kvm_io_bus_write(vcpu, KVM_PIO_BUS, |
cf8f70bf GN |
4417 | vcpu->arch.pio.port, vcpu->arch.pio.size, |
4418 | pd); | |
4419 | return r; | |
4420 | } | |
4421 | ||
6f6fbe98 XG |
4422 | static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size, |
4423 | unsigned short port, void *val, | |
4424 | unsigned int count, bool in) | |
cf8f70bf | 4425 | { |
cf8f70bf | 4426 | vcpu->arch.pio.port = port; |
6f6fbe98 | 4427 | vcpu->arch.pio.in = in; |
7972995b | 4428 | vcpu->arch.pio.count = count; |
cf8f70bf GN |
4429 | vcpu->arch.pio.size = size; |
4430 | ||
4431 | if (!kernel_pio(vcpu, vcpu->arch.pio_data)) { | |
7972995b | 4432 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
4433 | return 1; |
4434 | } | |
4435 | ||
4436 | vcpu->run->exit_reason = KVM_EXIT_IO; | |
6f6fbe98 | 4437 | vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT; |
cf8f70bf GN |
4438 | vcpu->run->io.size = size; |
4439 | vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE; | |
4440 | vcpu->run->io.count = count; | |
4441 | vcpu->run->io.port = port; | |
4442 | ||
4443 | return 0; | |
4444 | } | |
4445 | ||
6f6fbe98 XG |
4446 | static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt, |
4447 | int size, unsigned short port, void *val, | |
4448 | unsigned int count) | |
cf8f70bf | 4449 | { |
ca1d4a9e | 4450 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
6f6fbe98 | 4451 | int ret; |
ca1d4a9e | 4452 | |
6f6fbe98 XG |
4453 | if (vcpu->arch.pio.count) |
4454 | goto data_avail; | |
cf8f70bf | 4455 | |
6f6fbe98 XG |
4456 | ret = emulator_pio_in_out(vcpu, size, port, val, count, true); |
4457 | if (ret) { | |
4458 | data_avail: | |
4459 | memcpy(val, vcpu->arch.pio_data, size * count); | |
1171903d | 4460 | trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data); |
7972995b | 4461 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
4462 | return 1; |
4463 | } | |
4464 | ||
cf8f70bf GN |
4465 | return 0; |
4466 | } | |
4467 | ||
6f6fbe98 XG |
4468 | static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt, |
4469 | int size, unsigned short port, | |
4470 | const void *val, unsigned int count) | |
4471 | { | |
4472 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
4473 | ||
4474 | memcpy(vcpu->arch.pio_data, val, size * count); | |
1171903d | 4475 | trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data); |
6f6fbe98 XG |
4476 | return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false); |
4477 | } | |
4478 | ||
bbd9b64e CO |
4479 | static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg) |
4480 | { | |
4481 | return kvm_x86_ops->get_segment_base(vcpu, seg); | |
4482 | } | |
4483 | ||
3cb16fe7 | 4484 | static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address) |
bbd9b64e | 4485 | { |
3cb16fe7 | 4486 | kvm_mmu_invlpg(emul_to_vcpu(ctxt), address); |
bbd9b64e CO |
4487 | } |
4488 | ||
5cb56059 | 4489 | int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu) |
f5f48ee1 SY |
4490 | { |
4491 | if (!need_emulate_wbinvd(vcpu)) | |
4492 | return X86EMUL_CONTINUE; | |
4493 | ||
4494 | if (kvm_x86_ops->has_wbinvd_exit()) { | |
2eec7343 JK |
4495 | int cpu = get_cpu(); |
4496 | ||
4497 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); | |
f5f48ee1 SY |
4498 | smp_call_function_many(vcpu->arch.wbinvd_dirty_mask, |
4499 | wbinvd_ipi, NULL, 1); | |
2eec7343 | 4500 | put_cpu(); |
f5f48ee1 | 4501 | cpumask_clear(vcpu->arch.wbinvd_dirty_mask); |
2eec7343 JK |
4502 | } else |
4503 | wbinvd(); | |
f5f48ee1 SY |
4504 | return X86EMUL_CONTINUE; |
4505 | } | |
5cb56059 JS |
4506 | |
4507 | int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
4508 | { | |
4509 | kvm_x86_ops->skip_emulated_instruction(vcpu); | |
4510 | return kvm_emulate_wbinvd_noskip(vcpu); | |
4511 | } | |
f5f48ee1 SY |
4512 | EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd); |
4513 | ||
5cb56059 JS |
4514 | |
4515 | ||
bcaf5cc5 AK |
4516 | static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt) |
4517 | { | |
5cb56059 | 4518 | kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt)); |
bcaf5cc5 AK |
4519 | } |
4520 | ||
52eb5a6d XL |
4521 | static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, |
4522 | unsigned long *dest) | |
bbd9b64e | 4523 | { |
16f8a6f9 | 4524 | return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest); |
bbd9b64e CO |
4525 | } |
4526 | ||
52eb5a6d XL |
4527 | static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, |
4528 | unsigned long value) | |
bbd9b64e | 4529 | { |
338dbc97 | 4530 | |
717746e3 | 4531 | return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value); |
bbd9b64e CO |
4532 | } |
4533 | ||
52a46617 | 4534 | static u64 mk_cr_64(u64 curr_cr, u32 new_val) |
5fdbf976 | 4535 | { |
52a46617 | 4536 | return (curr_cr & ~((1ULL << 32) - 1)) | new_val; |
5fdbf976 MT |
4537 | } |
4538 | ||
717746e3 | 4539 | static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr) |
bbd9b64e | 4540 | { |
717746e3 | 4541 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
52a46617 GN |
4542 | unsigned long value; |
4543 | ||
4544 | switch (cr) { | |
4545 | case 0: | |
4546 | value = kvm_read_cr0(vcpu); | |
4547 | break; | |
4548 | case 2: | |
4549 | value = vcpu->arch.cr2; | |
4550 | break; | |
4551 | case 3: | |
9f8fe504 | 4552 | value = kvm_read_cr3(vcpu); |
52a46617 GN |
4553 | break; |
4554 | case 4: | |
4555 | value = kvm_read_cr4(vcpu); | |
4556 | break; | |
4557 | case 8: | |
4558 | value = kvm_get_cr8(vcpu); | |
4559 | break; | |
4560 | default: | |
a737f256 | 4561 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
52a46617 GN |
4562 | return 0; |
4563 | } | |
4564 | ||
4565 | return value; | |
4566 | } | |
4567 | ||
717746e3 | 4568 | static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val) |
52a46617 | 4569 | { |
717746e3 | 4570 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
0f12244f GN |
4571 | int res = 0; |
4572 | ||
52a46617 GN |
4573 | switch (cr) { |
4574 | case 0: | |
49a9b07e | 4575 | res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val)); |
52a46617 GN |
4576 | break; |
4577 | case 2: | |
4578 | vcpu->arch.cr2 = val; | |
4579 | break; | |
4580 | case 3: | |
2390218b | 4581 | res = kvm_set_cr3(vcpu, val); |
52a46617 GN |
4582 | break; |
4583 | case 4: | |
a83b29c6 | 4584 | res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val)); |
52a46617 GN |
4585 | break; |
4586 | case 8: | |
eea1cff9 | 4587 | res = kvm_set_cr8(vcpu, val); |
52a46617 GN |
4588 | break; |
4589 | default: | |
a737f256 | 4590 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
0f12244f | 4591 | res = -1; |
52a46617 | 4592 | } |
0f12244f GN |
4593 | |
4594 | return res; | |
52a46617 GN |
4595 | } |
4596 | ||
717746e3 | 4597 | static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt) |
9c537244 | 4598 | { |
717746e3 | 4599 | return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt)); |
9c537244 GN |
4600 | } |
4601 | ||
4bff1e86 | 4602 | static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
2dafc6c2 | 4603 | { |
4bff1e86 | 4604 | kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt); |
2dafc6c2 GN |
4605 | } |
4606 | ||
4bff1e86 | 4607 | static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
160ce1f1 | 4608 | { |
4bff1e86 | 4609 | kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt); |
160ce1f1 MG |
4610 | } |
4611 | ||
1ac9d0cf AK |
4612 | static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
4613 | { | |
4614 | kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt); | |
4615 | } | |
4616 | ||
4617 | static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) | |
4618 | { | |
4619 | kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt); | |
4620 | } | |
4621 | ||
4bff1e86 AK |
4622 | static unsigned long emulator_get_cached_segment_base( |
4623 | struct x86_emulate_ctxt *ctxt, int seg) | |
5951c442 | 4624 | { |
4bff1e86 | 4625 | return get_segment_base(emul_to_vcpu(ctxt), seg); |
5951c442 GN |
4626 | } |
4627 | ||
1aa36616 AK |
4628 | static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector, |
4629 | struct desc_struct *desc, u32 *base3, | |
4630 | int seg) | |
2dafc6c2 GN |
4631 | { |
4632 | struct kvm_segment var; | |
4633 | ||
4bff1e86 | 4634 | kvm_get_segment(emul_to_vcpu(ctxt), &var, seg); |
1aa36616 | 4635 | *selector = var.selector; |
2dafc6c2 | 4636 | |
378a8b09 GN |
4637 | if (var.unusable) { |
4638 | memset(desc, 0, sizeof(*desc)); | |
2dafc6c2 | 4639 | return false; |
378a8b09 | 4640 | } |
2dafc6c2 GN |
4641 | |
4642 | if (var.g) | |
4643 | var.limit >>= 12; | |
4644 | set_desc_limit(desc, var.limit); | |
4645 | set_desc_base(desc, (unsigned long)var.base); | |
5601d05b GN |
4646 | #ifdef CONFIG_X86_64 |
4647 | if (base3) | |
4648 | *base3 = var.base >> 32; | |
4649 | #endif | |
2dafc6c2 GN |
4650 | desc->type = var.type; |
4651 | desc->s = var.s; | |
4652 | desc->dpl = var.dpl; | |
4653 | desc->p = var.present; | |
4654 | desc->avl = var.avl; | |
4655 | desc->l = var.l; | |
4656 | desc->d = var.db; | |
4657 | desc->g = var.g; | |
4658 | ||
4659 | return true; | |
4660 | } | |
4661 | ||
1aa36616 AK |
4662 | static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector, |
4663 | struct desc_struct *desc, u32 base3, | |
4664 | int seg) | |
2dafc6c2 | 4665 | { |
4bff1e86 | 4666 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
2dafc6c2 GN |
4667 | struct kvm_segment var; |
4668 | ||
1aa36616 | 4669 | var.selector = selector; |
2dafc6c2 | 4670 | var.base = get_desc_base(desc); |
5601d05b GN |
4671 | #ifdef CONFIG_X86_64 |
4672 | var.base |= ((u64)base3) << 32; | |
4673 | #endif | |
2dafc6c2 GN |
4674 | var.limit = get_desc_limit(desc); |
4675 | if (desc->g) | |
4676 | var.limit = (var.limit << 12) | 0xfff; | |
4677 | var.type = desc->type; | |
2dafc6c2 GN |
4678 | var.dpl = desc->dpl; |
4679 | var.db = desc->d; | |
4680 | var.s = desc->s; | |
4681 | var.l = desc->l; | |
4682 | var.g = desc->g; | |
4683 | var.avl = desc->avl; | |
4684 | var.present = desc->p; | |
4685 | var.unusable = !var.present; | |
4686 | var.padding = 0; | |
4687 | ||
4688 | kvm_set_segment(vcpu, &var, seg); | |
4689 | return; | |
4690 | } | |
4691 | ||
717746e3 AK |
4692 | static int emulator_get_msr(struct x86_emulate_ctxt *ctxt, |
4693 | u32 msr_index, u64 *pdata) | |
4694 | { | |
609e36d3 PB |
4695 | struct msr_data msr; |
4696 | int r; | |
4697 | ||
4698 | msr.index = msr_index; | |
4699 | msr.host_initiated = false; | |
4700 | r = kvm_get_msr(emul_to_vcpu(ctxt), &msr); | |
4701 | if (r) | |
4702 | return r; | |
4703 | ||
4704 | *pdata = msr.data; | |
4705 | return 0; | |
717746e3 AK |
4706 | } |
4707 | ||
4708 | static int emulator_set_msr(struct x86_emulate_ctxt *ctxt, | |
4709 | u32 msr_index, u64 data) | |
4710 | { | |
8fe8ab46 WA |
4711 | struct msr_data msr; |
4712 | ||
4713 | msr.data = data; | |
4714 | msr.index = msr_index; | |
4715 | msr.host_initiated = false; | |
4716 | return kvm_set_msr(emul_to_vcpu(ctxt), &msr); | |
717746e3 AK |
4717 | } |
4718 | ||
64d60670 PB |
4719 | static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt) |
4720 | { | |
4721 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
4722 | ||
4723 | return vcpu->arch.smbase; | |
4724 | } | |
4725 | ||
4726 | static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase) | |
4727 | { | |
4728 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
4729 | ||
4730 | vcpu->arch.smbase = smbase; | |
4731 | } | |
4732 | ||
67f4d428 NA |
4733 | static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt, |
4734 | u32 pmc) | |
4735 | { | |
c6702c9d | 4736 | return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc); |
67f4d428 NA |
4737 | } |
4738 | ||
222d21aa AK |
4739 | static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt, |
4740 | u32 pmc, u64 *pdata) | |
4741 | { | |
c6702c9d | 4742 | return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata); |
222d21aa AK |
4743 | } |
4744 | ||
6c3287f7 AK |
4745 | static void emulator_halt(struct x86_emulate_ctxt *ctxt) |
4746 | { | |
4747 | emul_to_vcpu(ctxt)->arch.halt_request = 1; | |
4748 | } | |
4749 | ||
5037f6f3 AK |
4750 | static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt) |
4751 | { | |
4752 | preempt_disable(); | |
5197b808 | 4753 | kvm_load_guest_fpu(emul_to_vcpu(ctxt)); |
5037f6f3 AK |
4754 | /* |
4755 | * CR0.TS may reference the host fpu state, not the guest fpu state, | |
4756 | * so it may be clear at this point. | |
4757 | */ | |
4758 | clts(); | |
4759 | } | |
4760 | ||
4761 | static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt) | |
4762 | { | |
4763 | preempt_enable(); | |
4764 | } | |
4765 | ||
2953538e | 4766 | static int emulator_intercept(struct x86_emulate_ctxt *ctxt, |
8a76d7f2 | 4767 | struct x86_instruction_info *info, |
c4f035c6 AK |
4768 | enum x86_intercept_stage stage) |
4769 | { | |
2953538e | 4770 | return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage); |
c4f035c6 AK |
4771 | } |
4772 | ||
0017f93a | 4773 | static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt, |
bdb42f5a SB |
4774 | u32 *eax, u32 *ebx, u32 *ecx, u32 *edx) |
4775 | { | |
0017f93a | 4776 | kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx); |
bdb42f5a SB |
4777 | } |
4778 | ||
dd856efa AK |
4779 | static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg) |
4780 | { | |
4781 | return kvm_register_read(emul_to_vcpu(ctxt), reg); | |
4782 | } | |
4783 | ||
4784 | static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val) | |
4785 | { | |
4786 | kvm_register_write(emul_to_vcpu(ctxt), reg, val); | |
4787 | } | |
4788 | ||
801806d9 NA |
4789 | static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked) |
4790 | { | |
4791 | kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked); | |
4792 | } | |
4793 | ||
0225fb50 | 4794 | static const struct x86_emulate_ops emulate_ops = { |
dd856efa AK |
4795 | .read_gpr = emulator_read_gpr, |
4796 | .write_gpr = emulator_write_gpr, | |
1871c602 | 4797 | .read_std = kvm_read_guest_virt_system, |
2dafc6c2 | 4798 | .write_std = kvm_write_guest_virt_system, |
1871c602 | 4799 | .fetch = kvm_fetch_guest_virt, |
bbd9b64e CO |
4800 | .read_emulated = emulator_read_emulated, |
4801 | .write_emulated = emulator_write_emulated, | |
4802 | .cmpxchg_emulated = emulator_cmpxchg_emulated, | |
3cb16fe7 | 4803 | .invlpg = emulator_invlpg, |
cf8f70bf GN |
4804 | .pio_in_emulated = emulator_pio_in_emulated, |
4805 | .pio_out_emulated = emulator_pio_out_emulated, | |
1aa36616 AK |
4806 | .get_segment = emulator_get_segment, |
4807 | .set_segment = emulator_set_segment, | |
5951c442 | 4808 | .get_cached_segment_base = emulator_get_cached_segment_base, |
2dafc6c2 | 4809 | .get_gdt = emulator_get_gdt, |
160ce1f1 | 4810 | .get_idt = emulator_get_idt, |
1ac9d0cf AK |
4811 | .set_gdt = emulator_set_gdt, |
4812 | .set_idt = emulator_set_idt, | |
52a46617 GN |
4813 | .get_cr = emulator_get_cr, |
4814 | .set_cr = emulator_set_cr, | |
9c537244 | 4815 | .cpl = emulator_get_cpl, |
35aa5375 GN |
4816 | .get_dr = emulator_get_dr, |
4817 | .set_dr = emulator_set_dr, | |
64d60670 PB |
4818 | .get_smbase = emulator_get_smbase, |
4819 | .set_smbase = emulator_set_smbase, | |
717746e3 AK |
4820 | .set_msr = emulator_set_msr, |
4821 | .get_msr = emulator_get_msr, | |
67f4d428 | 4822 | .check_pmc = emulator_check_pmc, |
222d21aa | 4823 | .read_pmc = emulator_read_pmc, |
6c3287f7 | 4824 | .halt = emulator_halt, |
bcaf5cc5 | 4825 | .wbinvd = emulator_wbinvd, |
d6aa1000 | 4826 | .fix_hypercall = emulator_fix_hypercall, |
5037f6f3 AK |
4827 | .get_fpu = emulator_get_fpu, |
4828 | .put_fpu = emulator_put_fpu, | |
c4f035c6 | 4829 | .intercept = emulator_intercept, |
bdb42f5a | 4830 | .get_cpuid = emulator_get_cpuid, |
801806d9 | 4831 | .set_nmi_mask = emulator_set_nmi_mask, |
bbd9b64e CO |
4832 | }; |
4833 | ||
95cb2295 GN |
4834 | static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask) |
4835 | { | |
37ccdcbe | 4836 | u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu); |
95cb2295 GN |
4837 | /* |
4838 | * an sti; sti; sequence only disable interrupts for the first | |
4839 | * instruction. So, if the last instruction, be it emulated or | |
4840 | * not, left the system with the INT_STI flag enabled, it | |
4841 | * means that the last instruction is an sti. We should not | |
4842 | * leave the flag on in this case. The same goes for mov ss | |
4843 | */ | |
37ccdcbe PB |
4844 | if (int_shadow & mask) |
4845 | mask = 0; | |
6addfc42 | 4846 | if (unlikely(int_shadow || mask)) { |
95cb2295 | 4847 | kvm_x86_ops->set_interrupt_shadow(vcpu, mask); |
6addfc42 PB |
4848 | if (!mask) |
4849 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
4850 | } | |
95cb2295 GN |
4851 | } |
4852 | ||
ef54bcfe | 4853 | static bool inject_emulated_exception(struct kvm_vcpu *vcpu) |
54b8486f GN |
4854 | { |
4855 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; | |
da9cb575 | 4856 | if (ctxt->exception.vector == PF_VECTOR) |
ef54bcfe PB |
4857 | return kvm_propagate_fault(vcpu, &ctxt->exception); |
4858 | ||
4859 | if (ctxt->exception.error_code_valid) | |
da9cb575 AK |
4860 | kvm_queue_exception_e(vcpu, ctxt->exception.vector, |
4861 | ctxt->exception.error_code); | |
54b8486f | 4862 | else |
da9cb575 | 4863 | kvm_queue_exception(vcpu, ctxt->exception.vector); |
ef54bcfe | 4864 | return false; |
54b8486f GN |
4865 | } |
4866 | ||
8ec4722d MG |
4867 | static void init_emulate_ctxt(struct kvm_vcpu *vcpu) |
4868 | { | |
adf52235 | 4869 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
8ec4722d MG |
4870 | int cs_db, cs_l; |
4871 | ||
8ec4722d MG |
4872 | kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l); |
4873 | ||
adf52235 TY |
4874 | ctxt->eflags = kvm_get_rflags(vcpu); |
4875 | ctxt->eip = kvm_rip_read(vcpu); | |
4876 | ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL : | |
4877 | (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 : | |
42bf549f | 4878 | (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 : |
adf52235 TY |
4879 | cs_db ? X86EMUL_MODE_PROT32 : |
4880 | X86EMUL_MODE_PROT16; | |
a584539b | 4881 | BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK); |
64d60670 PB |
4882 | BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK); |
4883 | BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK); | |
a584539b | 4884 | ctxt->emul_flags = vcpu->arch.hflags; |
adf52235 | 4885 | |
dd856efa | 4886 | init_decode_cache(ctxt); |
7ae441ea | 4887 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; |
8ec4722d MG |
4888 | } |
4889 | ||
71f9833b | 4890 | int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip) |
63995653 | 4891 | { |
9d74191a | 4892 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
63995653 MG |
4893 | int ret; |
4894 | ||
4895 | init_emulate_ctxt(vcpu); | |
4896 | ||
9dac77fa AK |
4897 | ctxt->op_bytes = 2; |
4898 | ctxt->ad_bytes = 2; | |
4899 | ctxt->_eip = ctxt->eip + inc_eip; | |
9d74191a | 4900 | ret = emulate_int_real(ctxt, irq); |
63995653 MG |
4901 | |
4902 | if (ret != X86EMUL_CONTINUE) | |
4903 | return EMULATE_FAIL; | |
4904 | ||
9dac77fa | 4905 | ctxt->eip = ctxt->_eip; |
9d74191a TY |
4906 | kvm_rip_write(vcpu, ctxt->eip); |
4907 | kvm_set_rflags(vcpu, ctxt->eflags); | |
63995653 MG |
4908 | |
4909 | if (irq == NMI_VECTOR) | |
7460fb4a | 4910 | vcpu->arch.nmi_pending = 0; |
63995653 MG |
4911 | else |
4912 | vcpu->arch.interrupt.pending = false; | |
4913 | ||
4914 | return EMULATE_DONE; | |
4915 | } | |
4916 | EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt); | |
4917 | ||
6d77dbfc GN |
4918 | static int handle_emulation_failure(struct kvm_vcpu *vcpu) |
4919 | { | |
fc3a9157 JR |
4920 | int r = EMULATE_DONE; |
4921 | ||
6d77dbfc GN |
4922 | ++vcpu->stat.insn_emulation_fail; |
4923 | trace_kvm_emulate_insn_failed(vcpu); | |
a2b9e6c1 | 4924 | if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) { |
fc3a9157 JR |
4925 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; |
4926 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
4927 | vcpu->run->internal.ndata = 0; | |
4928 | r = EMULATE_FAIL; | |
4929 | } | |
6d77dbfc | 4930 | kvm_queue_exception(vcpu, UD_VECTOR); |
fc3a9157 JR |
4931 | |
4932 | return r; | |
6d77dbfc GN |
4933 | } |
4934 | ||
93c05d3e | 4935 | static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2, |
991eebf9 GN |
4936 | bool write_fault_to_shadow_pgtable, |
4937 | int emulation_type) | |
a6f177ef | 4938 | { |
95b3cf69 | 4939 | gpa_t gpa = cr2; |
8e3d9d06 | 4940 | pfn_t pfn; |
a6f177ef | 4941 | |
991eebf9 GN |
4942 | if (emulation_type & EMULTYPE_NO_REEXECUTE) |
4943 | return false; | |
4944 | ||
95b3cf69 XG |
4945 | if (!vcpu->arch.mmu.direct_map) { |
4946 | /* | |
4947 | * Write permission should be allowed since only | |
4948 | * write access need to be emulated. | |
4949 | */ | |
4950 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL); | |
a6f177ef | 4951 | |
95b3cf69 XG |
4952 | /* |
4953 | * If the mapping is invalid in guest, let cpu retry | |
4954 | * it to generate fault. | |
4955 | */ | |
4956 | if (gpa == UNMAPPED_GVA) | |
4957 | return true; | |
4958 | } | |
a6f177ef | 4959 | |
8e3d9d06 XG |
4960 | /* |
4961 | * Do not retry the unhandleable instruction if it faults on the | |
4962 | * readonly host memory, otherwise it will goto a infinite loop: | |
4963 | * retry instruction -> write #PF -> emulation fail -> retry | |
4964 | * instruction -> ... | |
4965 | */ | |
4966 | pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa)); | |
95b3cf69 XG |
4967 | |
4968 | /* | |
4969 | * If the instruction failed on the error pfn, it can not be fixed, | |
4970 | * report the error to userspace. | |
4971 | */ | |
4972 | if (is_error_noslot_pfn(pfn)) | |
4973 | return false; | |
4974 | ||
4975 | kvm_release_pfn_clean(pfn); | |
4976 | ||
4977 | /* The instructions are well-emulated on direct mmu. */ | |
4978 | if (vcpu->arch.mmu.direct_map) { | |
4979 | unsigned int indirect_shadow_pages; | |
4980 | ||
4981 | spin_lock(&vcpu->kvm->mmu_lock); | |
4982 | indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages; | |
4983 | spin_unlock(&vcpu->kvm->mmu_lock); | |
4984 | ||
4985 | if (indirect_shadow_pages) | |
4986 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
4987 | ||
a6f177ef | 4988 | return true; |
8e3d9d06 | 4989 | } |
a6f177ef | 4990 | |
95b3cf69 XG |
4991 | /* |
4992 | * if emulation was due to access to shadowed page table | |
4993 | * and it failed try to unshadow page and re-enter the | |
4994 | * guest to let CPU execute the instruction. | |
4995 | */ | |
4996 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
93c05d3e XG |
4997 | |
4998 | /* | |
4999 | * If the access faults on its page table, it can not | |
5000 | * be fixed by unprotecting shadow page and it should | |
5001 | * be reported to userspace. | |
5002 | */ | |
5003 | return !write_fault_to_shadow_pgtable; | |
a6f177ef GN |
5004 | } |
5005 | ||
1cb3f3ae XG |
5006 | static bool retry_instruction(struct x86_emulate_ctxt *ctxt, |
5007 | unsigned long cr2, int emulation_type) | |
5008 | { | |
5009 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
5010 | unsigned long last_retry_eip, last_retry_addr, gpa = cr2; | |
5011 | ||
5012 | last_retry_eip = vcpu->arch.last_retry_eip; | |
5013 | last_retry_addr = vcpu->arch.last_retry_addr; | |
5014 | ||
5015 | /* | |
5016 | * If the emulation is caused by #PF and it is non-page_table | |
5017 | * writing instruction, it means the VM-EXIT is caused by shadow | |
5018 | * page protected, we can zap the shadow page and retry this | |
5019 | * instruction directly. | |
5020 | * | |
5021 | * Note: if the guest uses a non-page-table modifying instruction | |
5022 | * on the PDE that points to the instruction, then we will unmap | |
5023 | * the instruction and go to an infinite loop. So, we cache the | |
5024 | * last retried eip and the last fault address, if we meet the eip | |
5025 | * and the address again, we can break out of the potential infinite | |
5026 | * loop. | |
5027 | */ | |
5028 | vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0; | |
5029 | ||
5030 | if (!(emulation_type & EMULTYPE_RETRY)) | |
5031 | return false; | |
5032 | ||
5033 | if (x86_page_table_writing_insn(ctxt)) | |
5034 | return false; | |
5035 | ||
5036 | if (ctxt->eip == last_retry_eip && last_retry_addr == cr2) | |
5037 | return false; | |
5038 | ||
5039 | vcpu->arch.last_retry_eip = ctxt->eip; | |
5040 | vcpu->arch.last_retry_addr = cr2; | |
5041 | ||
5042 | if (!vcpu->arch.mmu.direct_map) | |
5043 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL); | |
5044 | ||
22368028 | 5045 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); |
1cb3f3ae XG |
5046 | |
5047 | return true; | |
5048 | } | |
5049 | ||
716d51ab GN |
5050 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu); |
5051 | static int complete_emulated_pio(struct kvm_vcpu *vcpu); | |
5052 | ||
64d60670 | 5053 | static void kvm_smm_changed(struct kvm_vcpu *vcpu) |
a584539b | 5054 | { |
64d60670 | 5055 | if (!(vcpu->arch.hflags & HF_SMM_MASK)) { |
660a5d51 PB |
5056 | /* This is a good place to trace that we are exiting SMM. */ |
5057 | trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false); | |
5058 | ||
64d60670 PB |
5059 | if (unlikely(vcpu->arch.smi_pending)) { |
5060 | kvm_make_request(KVM_REQ_SMI, vcpu); | |
5061 | vcpu->arch.smi_pending = 0; | |
cd7764fe PB |
5062 | } else { |
5063 | /* Process a latched INIT, if any. */ | |
5064 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
64d60670 PB |
5065 | } |
5066 | } | |
699023e2 PB |
5067 | |
5068 | kvm_mmu_reset_context(vcpu); | |
64d60670 PB |
5069 | } |
5070 | ||
5071 | static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags) | |
5072 | { | |
5073 | unsigned changed = vcpu->arch.hflags ^ emul_flags; | |
5074 | ||
a584539b | 5075 | vcpu->arch.hflags = emul_flags; |
64d60670 PB |
5076 | |
5077 | if (changed & HF_SMM_MASK) | |
5078 | kvm_smm_changed(vcpu); | |
a584539b PB |
5079 | } |
5080 | ||
4a1e10d5 PB |
5081 | static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7, |
5082 | unsigned long *db) | |
5083 | { | |
5084 | u32 dr6 = 0; | |
5085 | int i; | |
5086 | u32 enable, rwlen; | |
5087 | ||
5088 | enable = dr7; | |
5089 | rwlen = dr7 >> 16; | |
5090 | for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4) | |
5091 | if ((enable & 3) && (rwlen & 15) == type && db[i] == addr) | |
5092 | dr6 |= (1 << i); | |
5093 | return dr6; | |
5094 | } | |
5095 | ||
6addfc42 | 5096 | static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r) |
663f4c61 PB |
5097 | { |
5098 | struct kvm_run *kvm_run = vcpu->run; | |
5099 | ||
5100 | /* | |
6addfc42 PB |
5101 | * rflags is the old, "raw" value of the flags. The new value has |
5102 | * not been saved yet. | |
663f4c61 PB |
5103 | * |
5104 | * This is correct even for TF set by the guest, because "the | |
5105 | * processor will not generate this exception after the instruction | |
5106 | * that sets the TF flag". | |
5107 | */ | |
663f4c61 PB |
5108 | if (unlikely(rflags & X86_EFLAGS_TF)) { |
5109 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) { | |
6f43ed01 NA |
5110 | kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | |
5111 | DR6_RTM; | |
663f4c61 PB |
5112 | kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip; |
5113 | kvm_run->debug.arch.exception = DB_VECTOR; | |
5114 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
5115 | *r = EMULATE_USER_EXIT; | |
5116 | } else { | |
5117 | vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF; | |
5118 | /* | |
5119 | * "Certain debug exceptions may clear bit 0-3. The | |
5120 | * remaining contents of the DR6 register are never | |
5121 | * cleared by the processor". | |
5122 | */ | |
5123 | vcpu->arch.dr6 &= ~15; | |
6f43ed01 | 5124 | vcpu->arch.dr6 |= DR6_BS | DR6_RTM; |
663f4c61 PB |
5125 | kvm_queue_exception(vcpu, DB_VECTOR); |
5126 | } | |
5127 | } | |
5128 | } | |
5129 | ||
4a1e10d5 PB |
5130 | static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r) |
5131 | { | |
4a1e10d5 PB |
5132 | if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) && |
5133 | (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) { | |
82b32774 NA |
5134 | struct kvm_run *kvm_run = vcpu->run; |
5135 | unsigned long eip = kvm_get_linear_rip(vcpu); | |
5136 | u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
4a1e10d5 PB |
5137 | vcpu->arch.guest_debug_dr7, |
5138 | vcpu->arch.eff_db); | |
5139 | ||
5140 | if (dr6 != 0) { | |
6f43ed01 | 5141 | kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM; |
82b32774 | 5142 | kvm_run->debug.arch.pc = eip; |
4a1e10d5 PB |
5143 | kvm_run->debug.arch.exception = DB_VECTOR; |
5144 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
5145 | *r = EMULATE_USER_EXIT; | |
5146 | return true; | |
5147 | } | |
5148 | } | |
5149 | ||
4161a569 NA |
5150 | if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) && |
5151 | !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) { | |
82b32774 NA |
5152 | unsigned long eip = kvm_get_linear_rip(vcpu); |
5153 | u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
4a1e10d5 PB |
5154 | vcpu->arch.dr7, |
5155 | vcpu->arch.db); | |
5156 | ||
5157 | if (dr6 != 0) { | |
5158 | vcpu->arch.dr6 &= ~15; | |
6f43ed01 | 5159 | vcpu->arch.dr6 |= dr6 | DR6_RTM; |
4a1e10d5 PB |
5160 | kvm_queue_exception(vcpu, DB_VECTOR); |
5161 | *r = EMULATE_DONE; | |
5162 | return true; | |
5163 | } | |
5164 | } | |
5165 | ||
5166 | return false; | |
5167 | } | |
5168 | ||
51d8b661 AP |
5169 | int x86_emulate_instruction(struct kvm_vcpu *vcpu, |
5170 | unsigned long cr2, | |
dc25e89e AP |
5171 | int emulation_type, |
5172 | void *insn, | |
5173 | int insn_len) | |
bbd9b64e | 5174 | { |
95cb2295 | 5175 | int r; |
9d74191a | 5176 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
7ae441ea | 5177 | bool writeback = true; |
93c05d3e | 5178 | bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable; |
bbd9b64e | 5179 | |
93c05d3e XG |
5180 | /* |
5181 | * Clear write_fault_to_shadow_pgtable here to ensure it is | |
5182 | * never reused. | |
5183 | */ | |
5184 | vcpu->arch.write_fault_to_shadow_pgtable = false; | |
26eef70c | 5185 | kvm_clear_exception_queue(vcpu); |
8d7d8102 | 5186 | |
571008da | 5187 | if (!(emulation_type & EMULTYPE_NO_DECODE)) { |
8ec4722d | 5188 | init_emulate_ctxt(vcpu); |
4a1e10d5 PB |
5189 | |
5190 | /* | |
5191 | * We will reenter on the same instruction since | |
5192 | * we do not set complete_userspace_io. This does not | |
5193 | * handle watchpoints yet, those would be handled in | |
5194 | * the emulate_ops. | |
5195 | */ | |
5196 | if (kvm_vcpu_check_breakpoint(vcpu, &r)) | |
5197 | return r; | |
5198 | ||
9d74191a TY |
5199 | ctxt->interruptibility = 0; |
5200 | ctxt->have_exception = false; | |
e0ad0b47 | 5201 | ctxt->exception.vector = -1; |
9d74191a | 5202 | ctxt->perm_ok = false; |
bbd9b64e | 5203 | |
b51e974f | 5204 | ctxt->ud = emulation_type & EMULTYPE_TRAP_UD; |
4005996e | 5205 | |
9d74191a | 5206 | r = x86_decode_insn(ctxt, insn, insn_len); |
bbd9b64e | 5207 | |
e46479f8 | 5208 | trace_kvm_emulate_insn_start(vcpu); |
f2b5756b | 5209 | ++vcpu->stat.insn_emulation; |
1d2887e2 | 5210 | if (r != EMULATION_OK) { |
4005996e AK |
5211 | if (emulation_type & EMULTYPE_TRAP_UD) |
5212 | return EMULATE_FAIL; | |
991eebf9 GN |
5213 | if (reexecute_instruction(vcpu, cr2, write_fault_to_spt, |
5214 | emulation_type)) | |
bbd9b64e | 5215 | return EMULATE_DONE; |
6d77dbfc GN |
5216 | if (emulation_type & EMULTYPE_SKIP) |
5217 | return EMULATE_FAIL; | |
5218 | return handle_emulation_failure(vcpu); | |
bbd9b64e CO |
5219 | } |
5220 | } | |
5221 | ||
ba8afb6b | 5222 | if (emulation_type & EMULTYPE_SKIP) { |
9dac77fa | 5223 | kvm_rip_write(vcpu, ctxt->_eip); |
bb663c7a NA |
5224 | if (ctxt->eflags & X86_EFLAGS_RF) |
5225 | kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF); | |
ba8afb6b GN |
5226 | return EMULATE_DONE; |
5227 | } | |
5228 | ||
1cb3f3ae XG |
5229 | if (retry_instruction(ctxt, cr2, emulation_type)) |
5230 | return EMULATE_DONE; | |
5231 | ||
7ae441ea | 5232 | /* this is needed for vmware backdoor interface to work since it |
4d2179e1 | 5233 | changes registers values during IO operation */ |
7ae441ea GN |
5234 | if (vcpu->arch.emulate_regs_need_sync_from_vcpu) { |
5235 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; | |
dd856efa | 5236 | emulator_invalidate_register_cache(ctxt); |
7ae441ea | 5237 | } |
4d2179e1 | 5238 | |
5cd21917 | 5239 | restart: |
9d74191a | 5240 | r = x86_emulate_insn(ctxt); |
bbd9b64e | 5241 | |
775fde86 JR |
5242 | if (r == EMULATION_INTERCEPTED) |
5243 | return EMULATE_DONE; | |
5244 | ||
d2ddd1c4 | 5245 | if (r == EMULATION_FAILED) { |
991eebf9 GN |
5246 | if (reexecute_instruction(vcpu, cr2, write_fault_to_spt, |
5247 | emulation_type)) | |
c3cd7ffa GN |
5248 | return EMULATE_DONE; |
5249 | ||
6d77dbfc | 5250 | return handle_emulation_failure(vcpu); |
bbd9b64e CO |
5251 | } |
5252 | ||
9d74191a | 5253 | if (ctxt->have_exception) { |
d2ddd1c4 | 5254 | r = EMULATE_DONE; |
ef54bcfe PB |
5255 | if (inject_emulated_exception(vcpu)) |
5256 | return r; | |
d2ddd1c4 | 5257 | } else if (vcpu->arch.pio.count) { |
0912c977 PB |
5258 | if (!vcpu->arch.pio.in) { |
5259 | /* FIXME: return into emulator if single-stepping. */ | |
3457e419 | 5260 | vcpu->arch.pio.count = 0; |
0912c977 | 5261 | } else { |
7ae441ea | 5262 | writeback = false; |
716d51ab GN |
5263 | vcpu->arch.complete_userspace_io = complete_emulated_pio; |
5264 | } | |
ac0a48c3 | 5265 | r = EMULATE_USER_EXIT; |
7ae441ea GN |
5266 | } else if (vcpu->mmio_needed) { |
5267 | if (!vcpu->mmio_is_write) | |
5268 | writeback = false; | |
ac0a48c3 | 5269 | r = EMULATE_USER_EXIT; |
716d51ab | 5270 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; |
7ae441ea | 5271 | } else if (r == EMULATION_RESTART) |
5cd21917 | 5272 | goto restart; |
d2ddd1c4 GN |
5273 | else |
5274 | r = EMULATE_DONE; | |
f850e2e6 | 5275 | |
7ae441ea | 5276 | if (writeback) { |
6addfc42 | 5277 | unsigned long rflags = kvm_x86_ops->get_rflags(vcpu); |
9d74191a | 5278 | toggle_interruptibility(vcpu, ctxt->interruptibility); |
7ae441ea | 5279 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
a584539b PB |
5280 | if (vcpu->arch.hflags != ctxt->emul_flags) |
5281 | kvm_set_hflags(vcpu, ctxt->emul_flags); | |
9d74191a | 5282 | kvm_rip_write(vcpu, ctxt->eip); |
663f4c61 | 5283 | if (r == EMULATE_DONE) |
6addfc42 | 5284 | kvm_vcpu_check_singlestep(vcpu, rflags, &r); |
38827dbd NA |
5285 | if (!ctxt->have_exception || |
5286 | exception_type(ctxt->exception.vector) == EXCPT_TRAP) | |
5287 | __kvm_set_rflags(vcpu, ctxt->eflags); | |
6addfc42 PB |
5288 | |
5289 | /* | |
5290 | * For STI, interrupts are shadowed; so KVM_REQ_EVENT will | |
5291 | * do nothing, and it will be requested again as soon as | |
5292 | * the shadow expires. But we still need to check here, | |
5293 | * because POPF has no interrupt shadow. | |
5294 | */ | |
5295 | if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF)) | |
5296 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
7ae441ea GN |
5297 | } else |
5298 | vcpu->arch.emulate_regs_need_sync_to_vcpu = true; | |
e85d28f8 GN |
5299 | |
5300 | return r; | |
de7d789a | 5301 | } |
51d8b661 | 5302 | EXPORT_SYMBOL_GPL(x86_emulate_instruction); |
de7d789a | 5303 | |
cf8f70bf | 5304 | int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port) |
de7d789a | 5305 | { |
cf8f70bf | 5306 | unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX); |
ca1d4a9e AK |
5307 | int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt, |
5308 | size, port, &val, 1); | |
cf8f70bf | 5309 | /* do not return to emulator after return from userspace */ |
7972995b | 5310 | vcpu->arch.pio.count = 0; |
de7d789a CO |
5311 | return ret; |
5312 | } | |
cf8f70bf | 5313 | EXPORT_SYMBOL_GPL(kvm_fast_pio_out); |
de7d789a | 5314 | |
8cfdc000 ZA |
5315 | static void tsc_bad(void *info) |
5316 | { | |
0a3aee0d | 5317 | __this_cpu_write(cpu_tsc_khz, 0); |
8cfdc000 ZA |
5318 | } |
5319 | ||
5320 | static void tsc_khz_changed(void *data) | |
c8076604 | 5321 | { |
8cfdc000 ZA |
5322 | struct cpufreq_freqs *freq = data; |
5323 | unsigned long khz = 0; | |
5324 | ||
5325 | if (data) | |
5326 | khz = freq->new; | |
5327 | else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) | |
5328 | khz = cpufreq_quick_get(raw_smp_processor_id()); | |
5329 | if (!khz) | |
5330 | khz = tsc_khz; | |
0a3aee0d | 5331 | __this_cpu_write(cpu_tsc_khz, khz); |
c8076604 GH |
5332 | } |
5333 | ||
c8076604 GH |
5334 | static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val, |
5335 | void *data) | |
5336 | { | |
5337 | struct cpufreq_freqs *freq = data; | |
5338 | struct kvm *kvm; | |
5339 | struct kvm_vcpu *vcpu; | |
5340 | int i, send_ipi = 0; | |
5341 | ||
8cfdc000 ZA |
5342 | /* |
5343 | * We allow guests to temporarily run on slowing clocks, | |
5344 | * provided we notify them after, or to run on accelerating | |
5345 | * clocks, provided we notify them before. Thus time never | |
5346 | * goes backwards. | |
5347 | * | |
5348 | * However, we have a problem. We can't atomically update | |
5349 | * the frequency of a given CPU from this function; it is | |
5350 | * merely a notifier, which can be called from any CPU. | |
5351 | * Changing the TSC frequency at arbitrary points in time | |
5352 | * requires a recomputation of local variables related to | |
5353 | * the TSC for each VCPU. We must flag these local variables | |
5354 | * to be updated and be sure the update takes place with the | |
5355 | * new frequency before any guests proceed. | |
5356 | * | |
5357 | * Unfortunately, the combination of hotplug CPU and frequency | |
5358 | * change creates an intractable locking scenario; the order | |
5359 | * of when these callouts happen is undefined with respect to | |
5360 | * CPU hotplug, and they can race with each other. As such, | |
5361 | * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is | |
5362 | * undefined; you can actually have a CPU frequency change take | |
5363 | * place in between the computation of X and the setting of the | |
5364 | * variable. To protect against this problem, all updates of | |
5365 | * the per_cpu tsc_khz variable are done in an interrupt | |
5366 | * protected IPI, and all callers wishing to update the value | |
5367 | * must wait for a synchronous IPI to complete (which is trivial | |
5368 | * if the caller is on the CPU already). This establishes the | |
5369 | * necessary total order on variable updates. | |
5370 | * | |
5371 | * Note that because a guest time update may take place | |
5372 | * anytime after the setting of the VCPU's request bit, the | |
5373 | * correct TSC value must be set before the request. However, | |
5374 | * to ensure the update actually makes it to any guest which | |
5375 | * starts running in hardware virtualization between the set | |
5376 | * and the acquisition of the spinlock, we must also ping the | |
5377 | * CPU after setting the request bit. | |
5378 | * | |
5379 | */ | |
5380 | ||
c8076604 GH |
5381 | if (val == CPUFREQ_PRECHANGE && freq->old > freq->new) |
5382 | return 0; | |
5383 | if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new) | |
5384 | return 0; | |
8cfdc000 ZA |
5385 | |
5386 | smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1); | |
c8076604 | 5387 | |
2f303b74 | 5388 | spin_lock(&kvm_lock); |
c8076604 | 5389 | list_for_each_entry(kvm, &vm_list, vm_list) { |
988a2cae | 5390 | kvm_for_each_vcpu(i, vcpu, kvm) { |
c8076604 GH |
5391 | if (vcpu->cpu != freq->cpu) |
5392 | continue; | |
c285545f | 5393 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
c8076604 | 5394 | if (vcpu->cpu != smp_processor_id()) |
8cfdc000 | 5395 | send_ipi = 1; |
c8076604 GH |
5396 | } |
5397 | } | |
2f303b74 | 5398 | spin_unlock(&kvm_lock); |
c8076604 GH |
5399 | |
5400 | if (freq->old < freq->new && send_ipi) { | |
5401 | /* | |
5402 | * We upscale the frequency. Must make the guest | |
5403 | * doesn't see old kvmclock values while running with | |
5404 | * the new frequency, otherwise we risk the guest sees | |
5405 | * time go backwards. | |
5406 | * | |
5407 | * In case we update the frequency for another cpu | |
5408 | * (which might be in guest context) send an interrupt | |
5409 | * to kick the cpu out of guest context. Next time | |
5410 | * guest context is entered kvmclock will be updated, | |
5411 | * so the guest will not see stale values. | |
5412 | */ | |
8cfdc000 | 5413 | smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1); |
c8076604 GH |
5414 | } |
5415 | return 0; | |
5416 | } | |
5417 | ||
5418 | static struct notifier_block kvmclock_cpufreq_notifier_block = { | |
8cfdc000 ZA |
5419 | .notifier_call = kvmclock_cpufreq_notifier |
5420 | }; | |
5421 | ||
5422 | static int kvmclock_cpu_notifier(struct notifier_block *nfb, | |
5423 | unsigned long action, void *hcpu) | |
5424 | { | |
5425 | unsigned int cpu = (unsigned long)hcpu; | |
5426 | ||
5427 | switch (action) { | |
5428 | case CPU_ONLINE: | |
5429 | case CPU_DOWN_FAILED: | |
5430 | smp_call_function_single(cpu, tsc_khz_changed, NULL, 1); | |
5431 | break; | |
5432 | case CPU_DOWN_PREPARE: | |
5433 | smp_call_function_single(cpu, tsc_bad, NULL, 1); | |
5434 | break; | |
5435 | } | |
5436 | return NOTIFY_OK; | |
5437 | } | |
5438 | ||
5439 | static struct notifier_block kvmclock_cpu_notifier_block = { | |
5440 | .notifier_call = kvmclock_cpu_notifier, | |
5441 | .priority = -INT_MAX | |
c8076604 GH |
5442 | }; |
5443 | ||
b820cc0c ZA |
5444 | static void kvm_timer_init(void) |
5445 | { | |
5446 | int cpu; | |
5447 | ||
c285545f | 5448 | max_tsc_khz = tsc_khz; |
460dd42e SB |
5449 | |
5450 | cpu_notifier_register_begin(); | |
b820cc0c | 5451 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) { |
c285545f ZA |
5452 | #ifdef CONFIG_CPU_FREQ |
5453 | struct cpufreq_policy policy; | |
5454 | memset(&policy, 0, sizeof(policy)); | |
3e26f230 AK |
5455 | cpu = get_cpu(); |
5456 | cpufreq_get_policy(&policy, cpu); | |
c285545f ZA |
5457 | if (policy.cpuinfo.max_freq) |
5458 | max_tsc_khz = policy.cpuinfo.max_freq; | |
3e26f230 | 5459 | put_cpu(); |
c285545f | 5460 | #endif |
b820cc0c ZA |
5461 | cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block, |
5462 | CPUFREQ_TRANSITION_NOTIFIER); | |
5463 | } | |
c285545f | 5464 | pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz); |
8cfdc000 ZA |
5465 | for_each_online_cpu(cpu) |
5466 | smp_call_function_single(cpu, tsc_khz_changed, NULL, 1); | |
460dd42e SB |
5467 | |
5468 | __register_hotcpu_notifier(&kvmclock_cpu_notifier_block); | |
5469 | cpu_notifier_register_done(); | |
5470 | ||
b820cc0c ZA |
5471 | } |
5472 | ||
ff9d07a0 ZY |
5473 | static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu); |
5474 | ||
f5132b01 | 5475 | int kvm_is_in_guest(void) |
ff9d07a0 | 5476 | { |
086c9855 | 5477 | return __this_cpu_read(current_vcpu) != NULL; |
ff9d07a0 ZY |
5478 | } |
5479 | ||
5480 | static int kvm_is_user_mode(void) | |
5481 | { | |
5482 | int user_mode = 3; | |
dcf46b94 | 5483 | |
086c9855 AS |
5484 | if (__this_cpu_read(current_vcpu)) |
5485 | user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu)); | |
dcf46b94 | 5486 | |
ff9d07a0 ZY |
5487 | return user_mode != 0; |
5488 | } | |
5489 | ||
5490 | static unsigned long kvm_get_guest_ip(void) | |
5491 | { | |
5492 | unsigned long ip = 0; | |
dcf46b94 | 5493 | |
086c9855 AS |
5494 | if (__this_cpu_read(current_vcpu)) |
5495 | ip = kvm_rip_read(__this_cpu_read(current_vcpu)); | |
dcf46b94 | 5496 | |
ff9d07a0 ZY |
5497 | return ip; |
5498 | } | |
5499 | ||
5500 | static struct perf_guest_info_callbacks kvm_guest_cbs = { | |
5501 | .is_in_guest = kvm_is_in_guest, | |
5502 | .is_user_mode = kvm_is_user_mode, | |
5503 | .get_guest_ip = kvm_get_guest_ip, | |
5504 | }; | |
5505 | ||
5506 | void kvm_before_handle_nmi(struct kvm_vcpu *vcpu) | |
5507 | { | |
086c9855 | 5508 | __this_cpu_write(current_vcpu, vcpu); |
ff9d07a0 ZY |
5509 | } |
5510 | EXPORT_SYMBOL_GPL(kvm_before_handle_nmi); | |
5511 | ||
5512 | void kvm_after_handle_nmi(struct kvm_vcpu *vcpu) | |
5513 | { | |
086c9855 | 5514 | __this_cpu_write(current_vcpu, NULL); |
ff9d07a0 ZY |
5515 | } |
5516 | EXPORT_SYMBOL_GPL(kvm_after_handle_nmi); | |
5517 | ||
ce88decf XG |
5518 | static void kvm_set_mmio_spte_mask(void) |
5519 | { | |
5520 | u64 mask; | |
5521 | int maxphyaddr = boot_cpu_data.x86_phys_bits; | |
5522 | ||
5523 | /* | |
5524 | * Set the reserved bits and the present bit of an paging-structure | |
5525 | * entry to generate page fault with PFER.RSV = 1. | |
5526 | */ | |
885032b9 | 5527 | /* Mask the reserved physical address bits. */ |
d1431483 | 5528 | mask = rsvd_bits(maxphyaddr, 51); |
885032b9 XG |
5529 | |
5530 | /* Bit 62 is always reserved for 32bit host. */ | |
5531 | mask |= 0x3ull << 62; | |
5532 | ||
5533 | /* Set the present bit. */ | |
ce88decf XG |
5534 | mask |= 1ull; |
5535 | ||
5536 | #ifdef CONFIG_X86_64 | |
5537 | /* | |
5538 | * If reserved bit is not supported, clear the present bit to disable | |
5539 | * mmio page fault. | |
5540 | */ | |
5541 | if (maxphyaddr == 52) | |
5542 | mask &= ~1ull; | |
5543 | #endif | |
5544 | ||
5545 | kvm_mmu_set_mmio_spte_mask(mask); | |
5546 | } | |
5547 | ||
16e8d74d MT |
5548 | #ifdef CONFIG_X86_64 |
5549 | static void pvclock_gtod_update_fn(struct work_struct *work) | |
5550 | { | |
d828199e MT |
5551 | struct kvm *kvm; |
5552 | ||
5553 | struct kvm_vcpu *vcpu; | |
5554 | int i; | |
5555 | ||
2f303b74 | 5556 | spin_lock(&kvm_lock); |
d828199e MT |
5557 | list_for_each_entry(kvm, &vm_list, vm_list) |
5558 | kvm_for_each_vcpu(i, vcpu, kvm) | |
105b21bb | 5559 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
d828199e | 5560 | atomic_set(&kvm_guest_has_master_clock, 0); |
2f303b74 | 5561 | spin_unlock(&kvm_lock); |
16e8d74d MT |
5562 | } |
5563 | ||
5564 | static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn); | |
5565 | ||
5566 | /* | |
5567 | * Notification about pvclock gtod data update. | |
5568 | */ | |
5569 | static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused, | |
5570 | void *priv) | |
5571 | { | |
5572 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
5573 | struct timekeeper *tk = priv; | |
5574 | ||
5575 | update_pvclock_gtod(tk); | |
5576 | ||
5577 | /* disable master clock if host does not trust, or does not | |
5578 | * use, TSC clocksource | |
5579 | */ | |
5580 | if (gtod->clock.vclock_mode != VCLOCK_TSC && | |
5581 | atomic_read(&kvm_guest_has_master_clock) != 0) | |
5582 | queue_work(system_long_wq, &pvclock_gtod_work); | |
5583 | ||
5584 | return 0; | |
5585 | } | |
5586 | ||
5587 | static struct notifier_block pvclock_gtod_notifier = { | |
5588 | .notifier_call = pvclock_gtod_notify, | |
5589 | }; | |
5590 | #endif | |
5591 | ||
f8c16bba | 5592 | int kvm_arch_init(void *opaque) |
043405e1 | 5593 | { |
b820cc0c | 5594 | int r; |
6b61edf7 | 5595 | struct kvm_x86_ops *ops = opaque; |
f8c16bba | 5596 | |
f8c16bba ZX |
5597 | if (kvm_x86_ops) { |
5598 | printk(KERN_ERR "kvm: already loaded the other module\n"); | |
56c6d28a ZX |
5599 | r = -EEXIST; |
5600 | goto out; | |
f8c16bba ZX |
5601 | } |
5602 | ||
5603 | if (!ops->cpu_has_kvm_support()) { | |
5604 | printk(KERN_ERR "kvm: no hardware support\n"); | |
56c6d28a ZX |
5605 | r = -EOPNOTSUPP; |
5606 | goto out; | |
f8c16bba ZX |
5607 | } |
5608 | if (ops->disabled_by_bios()) { | |
5609 | printk(KERN_ERR "kvm: disabled by bios\n"); | |
56c6d28a ZX |
5610 | r = -EOPNOTSUPP; |
5611 | goto out; | |
f8c16bba ZX |
5612 | } |
5613 | ||
013f6a5d MT |
5614 | r = -ENOMEM; |
5615 | shared_msrs = alloc_percpu(struct kvm_shared_msrs); | |
5616 | if (!shared_msrs) { | |
5617 | printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n"); | |
5618 | goto out; | |
5619 | } | |
5620 | ||
97db56ce AK |
5621 | r = kvm_mmu_module_init(); |
5622 | if (r) | |
013f6a5d | 5623 | goto out_free_percpu; |
97db56ce | 5624 | |
ce88decf | 5625 | kvm_set_mmio_spte_mask(); |
97db56ce | 5626 | |
f8c16bba | 5627 | kvm_x86_ops = ops; |
920c8377 | 5628 | |
7b52345e | 5629 | kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK, |
4b12f0de | 5630 | PT_DIRTY_MASK, PT64_NX_MASK, 0); |
c8076604 | 5631 | |
b820cc0c | 5632 | kvm_timer_init(); |
c8076604 | 5633 | |
ff9d07a0 ZY |
5634 | perf_register_guest_info_callbacks(&kvm_guest_cbs); |
5635 | ||
2acf923e DC |
5636 | if (cpu_has_xsave) |
5637 | host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK); | |
5638 | ||
c5cc421b | 5639 | kvm_lapic_init(); |
16e8d74d MT |
5640 | #ifdef CONFIG_X86_64 |
5641 | pvclock_gtod_register_notifier(&pvclock_gtod_notifier); | |
5642 | #endif | |
5643 | ||
f8c16bba | 5644 | return 0; |
56c6d28a | 5645 | |
013f6a5d MT |
5646 | out_free_percpu: |
5647 | free_percpu(shared_msrs); | |
56c6d28a | 5648 | out: |
56c6d28a | 5649 | return r; |
043405e1 | 5650 | } |
8776e519 | 5651 | |
f8c16bba ZX |
5652 | void kvm_arch_exit(void) |
5653 | { | |
ff9d07a0 ZY |
5654 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); |
5655 | ||
888d256e JK |
5656 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) |
5657 | cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block, | |
5658 | CPUFREQ_TRANSITION_NOTIFIER); | |
8cfdc000 | 5659 | unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block); |
16e8d74d MT |
5660 | #ifdef CONFIG_X86_64 |
5661 | pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier); | |
5662 | #endif | |
f8c16bba | 5663 | kvm_x86_ops = NULL; |
56c6d28a | 5664 | kvm_mmu_module_exit(); |
013f6a5d | 5665 | free_percpu(shared_msrs); |
56c6d28a | 5666 | } |
f8c16bba | 5667 | |
5cb56059 | 5668 | int kvm_vcpu_halt(struct kvm_vcpu *vcpu) |
8776e519 HB |
5669 | { |
5670 | ++vcpu->stat.halt_exits; | |
5671 | if (irqchip_in_kernel(vcpu->kvm)) { | |
a4535290 | 5672 | vcpu->arch.mp_state = KVM_MP_STATE_HALTED; |
8776e519 HB |
5673 | return 1; |
5674 | } else { | |
5675 | vcpu->run->exit_reason = KVM_EXIT_HLT; | |
5676 | return 0; | |
5677 | } | |
5678 | } | |
5cb56059 JS |
5679 | EXPORT_SYMBOL_GPL(kvm_vcpu_halt); |
5680 | ||
5681 | int kvm_emulate_halt(struct kvm_vcpu *vcpu) | |
5682 | { | |
5683 | kvm_x86_ops->skip_emulated_instruction(vcpu); | |
5684 | return kvm_vcpu_halt(vcpu); | |
5685 | } | |
8776e519 HB |
5686 | EXPORT_SYMBOL_GPL(kvm_emulate_halt); |
5687 | ||
6aef266c SV |
5688 | /* |
5689 | * kvm_pv_kick_cpu_op: Kick a vcpu. | |
5690 | * | |
5691 | * @apicid - apicid of vcpu to be kicked. | |
5692 | */ | |
5693 | static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid) | |
5694 | { | |
24d2166b | 5695 | struct kvm_lapic_irq lapic_irq; |
6aef266c | 5696 | |
24d2166b R |
5697 | lapic_irq.shorthand = 0; |
5698 | lapic_irq.dest_mode = 0; | |
5699 | lapic_irq.dest_id = apicid; | |
93bbf0b8 | 5700 | lapic_irq.msi_redir_hint = false; |
6aef266c | 5701 | |
24d2166b | 5702 | lapic_irq.delivery_mode = APIC_DM_REMRD; |
795a149e | 5703 | kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL); |
6aef266c SV |
5704 | } |
5705 | ||
8776e519 HB |
5706 | int kvm_emulate_hypercall(struct kvm_vcpu *vcpu) |
5707 | { | |
5708 | unsigned long nr, a0, a1, a2, a3, ret; | |
a449c7aa | 5709 | int op_64_bit, r = 1; |
8776e519 | 5710 | |
5cb56059 JS |
5711 | kvm_x86_ops->skip_emulated_instruction(vcpu); |
5712 | ||
55cd8e5a GN |
5713 | if (kvm_hv_hypercall_enabled(vcpu->kvm)) |
5714 | return kvm_hv_hypercall(vcpu); | |
5715 | ||
5fdbf976 MT |
5716 | nr = kvm_register_read(vcpu, VCPU_REGS_RAX); |
5717 | a0 = kvm_register_read(vcpu, VCPU_REGS_RBX); | |
5718 | a1 = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
5719 | a2 = kvm_register_read(vcpu, VCPU_REGS_RDX); | |
5720 | a3 = kvm_register_read(vcpu, VCPU_REGS_RSI); | |
8776e519 | 5721 | |
229456fc | 5722 | trace_kvm_hypercall(nr, a0, a1, a2, a3); |
2714d1d3 | 5723 | |
a449c7aa NA |
5724 | op_64_bit = is_64_bit_mode(vcpu); |
5725 | if (!op_64_bit) { | |
8776e519 HB |
5726 | nr &= 0xFFFFFFFF; |
5727 | a0 &= 0xFFFFFFFF; | |
5728 | a1 &= 0xFFFFFFFF; | |
5729 | a2 &= 0xFFFFFFFF; | |
5730 | a3 &= 0xFFFFFFFF; | |
5731 | } | |
5732 | ||
07708c4a JK |
5733 | if (kvm_x86_ops->get_cpl(vcpu) != 0) { |
5734 | ret = -KVM_EPERM; | |
5735 | goto out; | |
5736 | } | |
5737 | ||
8776e519 | 5738 | switch (nr) { |
b93463aa AK |
5739 | case KVM_HC_VAPIC_POLL_IRQ: |
5740 | ret = 0; | |
5741 | break; | |
6aef266c SV |
5742 | case KVM_HC_KICK_CPU: |
5743 | kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1); | |
5744 | ret = 0; | |
5745 | break; | |
8776e519 HB |
5746 | default: |
5747 | ret = -KVM_ENOSYS; | |
5748 | break; | |
5749 | } | |
07708c4a | 5750 | out: |
a449c7aa NA |
5751 | if (!op_64_bit) |
5752 | ret = (u32)ret; | |
5fdbf976 | 5753 | kvm_register_write(vcpu, VCPU_REGS_RAX, ret); |
f11c3a8d | 5754 | ++vcpu->stat.hypercalls; |
2f333bcb | 5755 | return r; |
8776e519 HB |
5756 | } |
5757 | EXPORT_SYMBOL_GPL(kvm_emulate_hypercall); | |
5758 | ||
b6785def | 5759 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt) |
8776e519 | 5760 | { |
d6aa1000 | 5761 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
8776e519 | 5762 | char instruction[3]; |
5fdbf976 | 5763 | unsigned long rip = kvm_rip_read(vcpu); |
8776e519 | 5764 | |
8776e519 | 5765 | kvm_x86_ops->patch_hypercall(vcpu, instruction); |
8776e519 | 5766 | |
9d74191a | 5767 | return emulator_write_emulated(ctxt, rip, instruction, 3, NULL); |
8776e519 HB |
5768 | } |
5769 | ||
b6c7a5dc HB |
5770 | /* |
5771 | * Check if userspace requested an interrupt window, and that the | |
5772 | * interrupt window is open. | |
5773 | * | |
5774 | * No need to exit to userspace if we already have an interrupt queued. | |
5775 | */ | |
851ba692 | 5776 | static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu) |
b6c7a5dc | 5777 | { |
8061823a | 5778 | return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) && |
851ba692 | 5779 | vcpu->run->request_interrupt_window && |
5df56646 | 5780 | kvm_arch_interrupt_allowed(vcpu)); |
b6c7a5dc HB |
5781 | } |
5782 | ||
851ba692 | 5783 | static void post_kvm_run_save(struct kvm_vcpu *vcpu) |
b6c7a5dc | 5784 | { |
851ba692 AK |
5785 | struct kvm_run *kvm_run = vcpu->run; |
5786 | ||
91586a3b | 5787 | kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0; |
f077825a | 5788 | kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0; |
2d3ad1f4 | 5789 | kvm_run->cr8 = kvm_get_cr8(vcpu); |
b6c7a5dc | 5790 | kvm_run->apic_base = kvm_get_apic_base(vcpu); |
4531220b | 5791 | if (irqchip_in_kernel(vcpu->kvm)) |
b6c7a5dc | 5792 | kvm_run->ready_for_interrupt_injection = 1; |
4531220b | 5793 | else |
b6c7a5dc | 5794 | kvm_run->ready_for_interrupt_injection = |
fa9726b0 GN |
5795 | kvm_arch_interrupt_allowed(vcpu) && |
5796 | !kvm_cpu_has_interrupt(vcpu) && | |
5797 | !kvm_event_needs_reinjection(vcpu); | |
b6c7a5dc HB |
5798 | } |
5799 | ||
95ba8273 GN |
5800 | static void update_cr8_intercept(struct kvm_vcpu *vcpu) |
5801 | { | |
5802 | int max_irr, tpr; | |
5803 | ||
5804 | if (!kvm_x86_ops->update_cr8_intercept) | |
5805 | return; | |
5806 | ||
88c808fd AK |
5807 | if (!vcpu->arch.apic) |
5808 | return; | |
5809 | ||
8db3baa2 GN |
5810 | if (!vcpu->arch.apic->vapic_addr) |
5811 | max_irr = kvm_lapic_find_highest_irr(vcpu); | |
5812 | else | |
5813 | max_irr = -1; | |
95ba8273 GN |
5814 | |
5815 | if (max_irr != -1) | |
5816 | max_irr >>= 4; | |
5817 | ||
5818 | tpr = kvm_lapic_get_cr8(vcpu); | |
5819 | ||
5820 | kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr); | |
5821 | } | |
5822 | ||
b6b8a145 | 5823 | static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win) |
95ba8273 | 5824 | { |
b6b8a145 JK |
5825 | int r; |
5826 | ||
95ba8273 | 5827 | /* try to reinject previous events if any */ |
b59bb7bd | 5828 | if (vcpu->arch.exception.pending) { |
5c1c85d0 AK |
5829 | trace_kvm_inj_exception(vcpu->arch.exception.nr, |
5830 | vcpu->arch.exception.has_error_code, | |
5831 | vcpu->arch.exception.error_code); | |
d6e8c854 NA |
5832 | |
5833 | if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT) | |
5834 | __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) | | |
5835 | X86_EFLAGS_RF); | |
5836 | ||
6bdf0662 NA |
5837 | if (vcpu->arch.exception.nr == DB_VECTOR && |
5838 | (vcpu->arch.dr7 & DR7_GD)) { | |
5839 | vcpu->arch.dr7 &= ~DR7_GD; | |
5840 | kvm_update_dr7(vcpu); | |
5841 | } | |
5842 | ||
b59bb7bd GN |
5843 | kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr, |
5844 | vcpu->arch.exception.has_error_code, | |
ce7ddec4 JR |
5845 | vcpu->arch.exception.error_code, |
5846 | vcpu->arch.exception.reinject); | |
b6b8a145 | 5847 | return 0; |
b59bb7bd GN |
5848 | } |
5849 | ||
95ba8273 GN |
5850 | if (vcpu->arch.nmi_injected) { |
5851 | kvm_x86_ops->set_nmi(vcpu); | |
b6b8a145 | 5852 | return 0; |
95ba8273 GN |
5853 | } |
5854 | ||
5855 | if (vcpu->arch.interrupt.pending) { | |
66fd3f7f | 5856 | kvm_x86_ops->set_irq(vcpu); |
b6b8a145 JK |
5857 | return 0; |
5858 | } | |
5859 | ||
5860 | if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) { | |
5861 | r = kvm_x86_ops->check_nested_events(vcpu, req_int_win); | |
5862 | if (r != 0) | |
5863 | return r; | |
95ba8273 GN |
5864 | } |
5865 | ||
5866 | /* try to inject new event if pending */ | |
5867 | if (vcpu->arch.nmi_pending) { | |
5868 | if (kvm_x86_ops->nmi_allowed(vcpu)) { | |
7460fb4a | 5869 | --vcpu->arch.nmi_pending; |
95ba8273 GN |
5870 | vcpu->arch.nmi_injected = true; |
5871 | kvm_x86_ops->set_nmi(vcpu); | |
5872 | } | |
c7c9c56c | 5873 | } else if (kvm_cpu_has_injectable_intr(vcpu)) { |
9242b5b6 BD |
5874 | /* |
5875 | * Because interrupts can be injected asynchronously, we are | |
5876 | * calling check_nested_events again here to avoid a race condition. | |
5877 | * See https://lkml.org/lkml/2014/7/2/60 for discussion about this | |
5878 | * proposal and current concerns. Perhaps we should be setting | |
5879 | * KVM_REQ_EVENT only on certain events and not unconditionally? | |
5880 | */ | |
5881 | if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) { | |
5882 | r = kvm_x86_ops->check_nested_events(vcpu, req_int_win); | |
5883 | if (r != 0) | |
5884 | return r; | |
5885 | } | |
95ba8273 | 5886 | if (kvm_x86_ops->interrupt_allowed(vcpu)) { |
66fd3f7f GN |
5887 | kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), |
5888 | false); | |
5889 | kvm_x86_ops->set_irq(vcpu); | |
95ba8273 GN |
5890 | } |
5891 | } | |
b6b8a145 | 5892 | return 0; |
95ba8273 GN |
5893 | } |
5894 | ||
7460fb4a AK |
5895 | static void process_nmi(struct kvm_vcpu *vcpu) |
5896 | { | |
5897 | unsigned limit = 2; | |
5898 | ||
5899 | /* | |
5900 | * x86 is limited to one NMI running, and one NMI pending after it. | |
5901 | * If an NMI is already in progress, limit further NMIs to just one. | |
5902 | * Otherwise, allow two (and we'll inject the first one immediately). | |
5903 | */ | |
5904 | if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected) | |
5905 | limit = 1; | |
5906 | ||
5907 | vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0); | |
5908 | vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit); | |
5909 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
5910 | } | |
5911 | ||
660a5d51 PB |
5912 | #define put_smstate(type, buf, offset, val) \ |
5913 | *(type *)((buf) + (offset) - 0x7e00) = val | |
5914 | ||
5915 | static u32 process_smi_get_segment_flags(struct kvm_segment *seg) | |
5916 | { | |
5917 | u32 flags = 0; | |
5918 | flags |= seg->g << 23; | |
5919 | flags |= seg->db << 22; | |
5920 | flags |= seg->l << 21; | |
5921 | flags |= seg->avl << 20; | |
5922 | flags |= seg->present << 15; | |
5923 | flags |= seg->dpl << 13; | |
5924 | flags |= seg->s << 12; | |
5925 | flags |= seg->type << 8; | |
5926 | return flags; | |
5927 | } | |
5928 | ||
5929 | static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n) | |
5930 | { | |
5931 | struct kvm_segment seg; | |
5932 | int offset; | |
5933 | ||
5934 | kvm_get_segment(vcpu, &seg, n); | |
5935 | put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector); | |
5936 | ||
5937 | if (n < 3) | |
5938 | offset = 0x7f84 + n * 12; | |
5939 | else | |
5940 | offset = 0x7f2c + (n - 3) * 12; | |
5941 | ||
5942 | put_smstate(u32, buf, offset + 8, seg.base); | |
5943 | put_smstate(u32, buf, offset + 4, seg.limit); | |
5944 | put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg)); | |
5945 | } | |
5946 | ||
5947 | static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n) | |
5948 | { | |
5949 | struct kvm_segment seg; | |
5950 | int offset; | |
5951 | u16 flags; | |
5952 | ||
5953 | kvm_get_segment(vcpu, &seg, n); | |
5954 | offset = 0x7e00 + n * 16; | |
5955 | ||
5956 | flags = process_smi_get_segment_flags(&seg) >> 8; | |
5957 | put_smstate(u16, buf, offset, seg.selector); | |
5958 | put_smstate(u16, buf, offset + 2, flags); | |
5959 | put_smstate(u32, buf, offset + 4, seg.limit); | |
5960 | put_smstate(u64, buf, offset + 8, seg.base); | |
5961 | } | |
5962 | ||
5963 | static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf) | |
5964 | { | |
5965 | struct desc_ptr dt; | |
5966 | struct kvm_segment seg; | |
5967 | unsigned long val; | |
5968 | int i; | |
5969 | ||
5970 | put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu)); | |
5971 | put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu)); | |
5972 | put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu)); | |
5973 | put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu)); | |
5974 | ||
5975 | for (i = 0; i < 8; i++) | |
5976 | put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i)); | |
5977 | ||
5978 | kvm_get_dr(vcpu, 6, &val); | |
5979 | put_smstate(u32, buf, 0x7fcc, (u32)val); | |
5980 | kvm_get_dr(vcpu, 7, &val); | |
5981 | put_smstate(u32, buf, 0x7fc8, (u32)val); | |
5982 | ||
5983 | kvm_get_segment(vcpu, &seg, VCPU_SREG_TR); | |
5984 | put_smstate(u32, buf, 0x7fc4, seg.selector); | |
5985 | put_smstate(u32, buf, 0x7f64, seg.base); | |
5986 | put_smstate(u32, buf, 0x7f60, seg.limit); | |
5987 | put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg)); | |
5988 | ||
5989 | kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR); | |
5990 | put_smstate(u32, buf, 0x7fc0, seg.selector); | |
5991 | put_smstate(u32, buf, 0x7f80, seg.base); | |
5992 | put_smstate(u32, buf, 0x7f7c, seg.limit); | |
5993 | put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg)); | |
5994 | ||
5995 | kvm_x86_ops->get_gdt(vcpu, &dt); | |
5996 | put_smstate(u32, buf, 0x7f74, dt.address); | |
5997 | put_smstate(u32, buf, 0x7f70, dt.size); | |
5998 | ||
5999 | kvm_x86_ops->get_idt(vcpu, &dt); | |
6000 | put_smstate(u32, buf, 0x7f58, dt.address); | |
6001 | put_smstate(u32, buf, 0x7f54, dt.size); | |
6002 | ||
6003 | for (i = 0; i < 6; i++) | |
6004 | process_smi_save_seg_32(vcpu, buf, i); | |
6005 | ||
6006 | put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu)); | |
6007 | ||
6008 | /* revision id */ | |
6009 | put_smstate(u32, buf, 0x7efc, 0x00020000); | |
6010 | put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase); | |
6011 | } | |
6012 | ||
6013 | static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf) | |
6014 | { | |
6015 | #ifdef CONFIG_X86_64 | |
6016 | struct desc_ptr dt; | |
6017 | struct kvm_segment seg; | |
6018 | unsigned long val; | |
6019 | int i; | |
6020 | ||
6021 | for (i = 0; i < 16; i++) | |
6022 | put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i)); | |
6023 | ||
6024 | put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu)); | |
6025 | put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu)); | |
6026 | ||
6027 | kvm_get_dr(vcpu, 6, &val); | |
6028 | put_smstate(u64, buf, 0x7f68, val); | |
6029 | kvm_get_dr(vcpu, 7, &val); | |
6030 | put_smstate(u64, buf, 0x7f60, val); | |
6031 | ||
6032 | put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu)); | |
6033 | put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu)); | |
6034 | put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu)); | |
6035 | ||
6036 | put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase); | |
6037 | ||
6038 | /* revision id */ | |
6039 | put_smstate(u32, buf, 0x7efc, 0x00020064); | |
6040 | ||
6041 | put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer); | |
6042 | ||
6043 | kvm_get_segment(vcpu, &seg, VCPU_SREG_TR); | |
6044 | put_smstate(u16, buf, 0x7e90, seg.selector); | |
6045 | put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8); | |
6046 | put_smstate(u32, buf, 0x7e94, seg.limit); | |
6047 | put_smstate(u64, buf, 0x7e98, seg.base); | |
6048 | ||
6049 | kvm_x86_ops->get_idt(vcpu, &dt); | |
6050 | put_smstate(u32, buf, 0x7e84, dt.size); | |
6051 | put_smstate(u64, buf, 0x7e88, dt.address); | |
6052 | ||
6053 | kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR); | |
6054 | put_smstate(u16, buf, 0x7e70, seg.selector); | |
6055 | put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8); | |
6056 | put_smstate(u32, buf, 0x7e74, seg.limit); | |
6057 | put_smstate(u64, buf, 0x7e78, seg.base); | |
6058 | ||
6059 | kvm_x86_ops->get_gdt(vcpu, &dt); | |
6060 | put_smstate(u32, buf, 0x7e64, dt.size); | |
6061 | put_smstate(u64, buf, 0x7e68, dt.address); | |
6062 | ||
6063 | for (i = 0; i < 6; i++) | |
6064 | process_smi_save_seg_64(vcpu, buf, i); | |
6065 | #else | |
6066 | WARN_ON_ONCE(1); | |
6067 | #endif | |
6068 | } | |
6069 | ||
64d60670 PB |
6070 | static void process_smi(struct kvm_vcpu *vcpu) |
6071 | { | |
660a5d51 PB |
6072 | struct kvm_segment cs, ds; |
6073 | char buf[512]; | |
6074 | u32 cr0; | |
6075 | ||
64d60670 PB |
6076 | if (is_smm(vcpu)) { |
6077 | vcpu->arch.smi_pending = true; | |
6078 | return; | |
6079 | } | |
6080 | ||
660a5d51 PB |
6081 | trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true); |
6082 | vcpu->arch.hflags |= HF_SMM_MASK; | |
6083 | memset(buf, 0, 512); | |
6084 | if (guest_cpuid_has_longmode(vcpu)) | |
6085 | process_smi_save_state_64(vcpu, buf); | |
6086 | else | |
6087 | process_smi_save_state_32(vcpu, buf); | |
6088 | ||
54bf36aa | 6089 | kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf)); |
660a5d51 PB |
6090 | |
6091 | if (kvm_x86_ops->get_nmi_mask(vcpu)) | |
6092 | vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK; | |
6093 | else | |
6094 | kvm_x86_ops->set_nmi_mask(vcpu, true); | |
6095 | ||
6096 | kvm_set_rflags(vcpu, X86_EFLAGS_FIXED); | |
6097 | kvm_rip_write(vcpu, 0x8000); | |
6098 | ||
6099 | cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG); | |
6100 | kvm_x86_ops->set_cr0(vcpu, cr0); | |
6101 | vcpu->arch.cr0 = cr0; | |
6102 | ||
6103 | kvm_x86_ops->set_cr4(vcpu, 0); | |
6104 | ||
6105 | __kvm_set_dr(vcpu, 7, DR7_FIXED_1); | |
6106 | ||
6107 | cs.selector = (vcpu->arch.smbase >> 4) & 0xffff; | |
6108 | cs.base = vcpu->arch.smbase; | |
6109 | ||
6110 | ds.selector = 0; | |
6111 | ds.base = 0; | |
6112 | ||
6113 | cs.limit = ds.limit = 0xffffffff; | |
6114 | cs.type = ds.type = 0x3; | |
6115 | cs.dpl = ds.dpl = 0; | |
6116 | cs.db = ds.db = 0; | |
6117 | cs.s = ds.s = 1; | |
6118 | cs.l = ds.l = 0; | |
6119 | cs.g = ds.g = 1; | |
6120 | cs.avl = ds.avl = 0; | |
6121 | cs.present = ds.present = 1; | |
6122 | cs.unusable = ds.unusable = 0; | |
6123 | cs.padding = ds.padding = 0; | |
6124 | ||
6125 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
6126 | kvm_set_segment(vcpu, &ds, VCPU_SREG_DS); | |
6127 | kvm_set_segment(vcpu, &ds, VCPU_SREG_ES); | |
6128 | kvm_set_segment(vcpu, &ds, VCPU_SREG_FS); | |
6129 | kvm_set_segment(vcpu, &ds, VCPU_SREG_GS); | |
6130 | kvm_set_segment(vcpu, &ds, VCPU_SREG_SS); | |
6131 | ||
6132 | if (guest_cpuid_has_longmode(vcpu)) | |
6133 | kvm_x86_ops->set_efer(vcpu, 0); | |
6134 | ||
6135 | kvm_update_cpuid(vcpu); | |
6136 | kvm_mmu_reset_context(vcpu); | |
64d60670 PB |
6137 | } |
6138 | ||
3d81bc7e | 6139 | static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu) |
c7c9c56c YZ |
6140 | { |
6141 | u64 eoi_exit_bitmap[4]; | |
cf9e65b7 | 6142 | u32 tmr[8]; |
c7c9c56c | 6143 | |
3d81bc7e YZ |
6144 | if (!kvm_apic_hw_enabled(vcpu->arch.apic)) |
6145 | return; | |
c7c9c56c YZ |
6146 | |
6147 | memset(eoi_exit_bitmap, 0, 32); | |
cf9e65b7 | 6148 | memset(tmr, 0, 32); |
c7c9c56c | 6149 | |
cf9e65b7 | 6150 | kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr); |
c7c9c56c | 6151 | kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap); |
cf9e65b7 | 6152 | kvm_apic_update_tmr(vcpu, tmr); |
c7c9c56c YZ |
6153 | } |
6154 | ||
a70656b6 RK |
6155 | static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu) |
6156 | { | |
6157 | ++vcpu->stat.tlb_flush; | |
6158 | kvm_x86_ops->tlb_flush(vcpu); | |
6159 | } | |
6160 | ||
4256f43f TC |
6161 | void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu) |
6162 | { | |
c24ae0dc TC |
6163 | struct page *page = NULL; |
6164 | ||
f439ed27 PB |
6165 | if (!irqchip_in_kernel(vcpu->kvm)) |
6166 | return; | |
6167 | ||
4256f43f TC |
6168 | if (!kvm_x86_ops->set_apic_access_page_addr) |
6169 | return; | |
6170 | ||
c24ae0dc | 6171 | page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT); |
e8fd5e9e AA |
6172 | if (is_error_page(page)) |
6173 | return; | |
c24ae0dc TC |
6174 | kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page)); |
6175 | ||
6176 | /* | |
6177 | * Do not pin apic access page in memory, the MMU notifier | |
6178 | * will call us again if it is migrated or swapped out. | |
6179 | */ | |
6180 | put_page(page); | |
4256f43f TC |
6181 | } |
6182 | EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page); | |
6183 | ||
fe71557a TC |
6184 | void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm, |
6185 | unsigned long address) | |
6186 | { | |
c24ae0dc TC |
6187 | /* |
6188 | * The physical address of apic access page is stored in the VMCS. | |
6189 | * Update it when it becomes invalid. | |
6190 | */ | |
6191 | if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT)) | |
6192 | kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD); | |
fe71557a TC |
6193 | } |
6194 | ||
9357d939 | 6195 | /* |
362c698f | 6196 | * Returns 1 to let vcpu_run() continue the guest execution loop without |
9357d939 TY |
6197 | * exiting to the userspace. Otherwise, the value will be returned to the |
6198 | * userspace. | |
6199 | */ | |
851ba692 | 6200 | static int vcpu_enter_guest(struct kvm_vcpu *vcpu) |
b6c7a5dc HB |
6201 | { |
6202 | int r; | |
6a8b1d13 | 6203 | bool req_int_win = !irqchip_in_kernel(vcpu->kvm) && |
851ba692 | 6204 | vcpu->run->request_interrupt_window; |
730dca42 | 6205 | bool req_immediate_exit = false; |
b6c7a5dc | 6206 | |
3e007509 | 6207 | if (vcpu->requests) { |
a8eeb04a | 6208 | if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) |
2e53d63a | 6209 | kvm_mmu_unload(vcpu); |
a8eeb04a | 6210 | if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu)) |
2f599714 | 6211 | __kvm_migrate_timers(vcpu); |
d828199e MT |
6212 | if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu)) |
6213 | kvm_gen_update_masterclock(vcpu->kvm); | |
0061d53d MT |
6214 | if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu)) |
6215 | kvm_gen_kvmclock_update(vcpu); | |
34c238a1 ZA |
6216 | if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) { |
6217 | r = kvm_guest_time_update(vcpu); | |
8cfdc000 ZA |
6218 | if (unlikely(r)) |
6219 | goto out; | |
6220 | } | |
a8eeb04a | 6221 | if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu)) |
4731d4c7 | 6222 | kvm_mmu_sync_roots(vcpu); |
a8eeb04a | 6223 | if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) |
a70656b6 | 6224 | kvm_vcpu_flush_tlb(vcpu); |
a8eeb04a | 6225 | if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) { |
851ba692 | 6226 | vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS; |
b93463aa AK |
6227 | r = 0; |
6228 | goto out; | |
6229 | } | |
a8eeb04a | 6230 | if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) { |
851ba692 | 6231 | vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN; |
71c4dfaf JR |
6232 | r = 0; |
6233 | goto out; | |
6234 | } | |
a8eeb04a | 6235 | if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) { |
02daab21 AK |
6236 | vcpu->fpu_active = 0; |
6237 | kvm_x86_ops->fpu_deactivate(vcpu); | |
6238 | } | |
af585b92 GN |
6239 | if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) { |
6240 | /* Page is swapped out. Do synthetic halt */ | |
6241 | vcpu->arch.apf.halted = true; | |
6242 | r = 1; | |
6243 | goto out; | |
6244 | } | |
c9aaa895 GC |
6245 | if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu)) |
6246 | record_steal_time(vcpu); | |
64d60670 PB |
6247 | if (kvm_check_request(KVM_REQ_SMI, vcpu)) |
6248 | process_smi(vcpu); | |
7460fb4a AK |
6249 | if (kvm_check_request(KVM_REQ_NMI, vcpu)) |
6250 | process_nmi(vcpu); | |
f5132b01 | 6251 | if (kvm_check_request(KVM_REQ_PMU, vcpu)) |
c6702c9d | 6252 | kvm_pmu_handle_event(vcpu); |
f5132b01 | 6253 | if (kvm_check_request(KVM_REQ_PMI, vcpu)) |
c6702c9d | 6254 | kvm_pmu_deliver_pmi(vcpu); |
3d81bc7e YZ |
6255 | if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu)) |
6256 | vcpu_scan_ioapic(vcpu); | |
4256f43f TC |
6257 | if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu)) |
6258 | kvm_vcpu_reload_apic_access_page(vcpu); | |
2f52d58c | 6259 | } |
b93463aa | 6260 | |
b463a6f7 | 6261 | if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) { |
66450a21 JK |
6262 | kvm_apic_accept_events(vcpu); |
6263 | if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) { | |
6264 | r = 1; | |
6265 | goto out; | |
6266 | } | |
6267 | ||
b6b8a145 JK |
6268 | if (inject_pending_event(vcpu, req_int_win) != 0) |
6269 | req_immediate_exit = true; | |
b463a6f7 | 6270 | /* enable NMI/IRQ window open exits if needed */ |
b6b8a145 | 6271 | else if (vcpu->arch.nmi_pending) |
c9a7953f | 6272 | kvm_x86_ops->enable_nmi_window(vcpu); |
c7c9c56c | 6273 | else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win) |
c9a7953f | 6274 | kvm_x86_ops->enable_irq_window(vcpu); |
b463a6f7 AK |
6275 | |
6276 | if (kvm_lapic_enabled(vcpu)) { | |
c7c9c56c YZ |
6277 | /* |
6278 | * Update architecture specific hints for APIC | |
6279 | * virtual interrupt delivery. | |
6280 | */ | |
6281 | if (kvm_x86_ops->hwapic_irr_update) | |
6282 | kvm_x86_ops->hwapic_irr_update(vcpu, | |
6283 | kvm_lapic_find_highest_irr(vcpu)); | |
b463a6f7 AK |
6284 | update_cr8_intercept(vcpu); |
6285 | kvm_lapic_sync_to_vapic(vcpu); | |
6286 | } | |
6287 | } | |
6288 | ||
d8368af8 AK |
6289 | r = kvm_mmu_reload(vcpu); |
6290 | if (unlikely(r)) { | |
d905c069 | 6291 | goto cancel_injection; |
d8368af8 AK |
6292 | } |
6293 | ||
b6c7a5dc HB |
6294 | preempt_disable(); |
6295 | ||
6296 | kvm_x86_ops->prepare_guest_switch(vcpu); | |
2608d7a1 AK |
6297 | if (vcpu->fpu_active) |
6298 | kvm_load_guest_fpu(vcpu); | |
2acf923e | 6299 | kvm_load_guest_xcr0(vcpu); |
b6c7a5dc | 6300 | |
6b7e2d09 XG |
6301 | vcpu->mode = IN_GUEST_MODE; |
6302 | ||
01b71917 MT |
6303 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); |
6304 | ||
6b7e2d09 XG |
6305 | /* We should set ->mode before check ->requests, |
6306 | * see the comment in make_all_cpus_request. | |
6307 | */ | |
01b71917 | 6308 | smp_mb__after_srcu_read_unlock(); |
b6c7a5dc | 6309 | |
d94e1dc9 | 6310 | local_irq_disable(); |
32f88400 | 6311 | |
6b7e2d09 | 6312 | if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests |
d94e1dc9 | 6313 | || need_resched() || signal_pending(current)) { |
6b7e2d09 | 6314 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 6315 | smp_wmb(); |
6c142801 AK |
6316 | local_irq_enable(); |
6317 | preempt_enable(); | |
01b71917 | 6318 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
6c142801 | 6319 | r = 1; |
d905c069 | 6320 | goto cancel_injection; |
6c142801 AK |
6321 | } |
6322 | ||
d6185f20 NHE |
6323 | if (req_immediate_exit) |
6324 | smp_send_reschedule(vcpu->cpu); | |
6325 | ||
ccf73aaf | 6326 | __kvm_guest_enter(); |
b6c7a5dc | 6327 | |
42dbaa5a | 6328 | if (unlikely(vcpu->arch.switch_db_regs)) { |
42dbaa5a JK |
6329 | set_debugreg(0, 7); |
6330 | set_debugreg(vcpu->arch.eff_db[0], 0); | |
6331 | set_debugreg(vcpu->arch.eff_db[1], 1); | |
6332 | set_debugreg(vcpu->arch.eff_db[2], 2); | |
6333 | set_debugreg(vcpu->arch.eff_db[3], 3); | |
c77fb5fe | 6334 | set_debugreg(vcpu->arch.dr6, 6); |
ae561ede | 6335 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD; |
42dbaa5a | 6336 | } |
b6c7a5dc | 6337 | |
229456fc | 6338 | trace_kvm_entry(vcpu->vcpu_id); |
d0659d94 | 6339 | wait_lapic_expire(vcpu); |
851ba692 | 6340 | kvm_x86_ops->run(vcpu); |
b6c7a5dc | 6341 | |
c77fb5fe PB |
6342 | /* |
6343 | * Do this here before restoring debug registers on the host. And | |
6344 | * since we do this before handling the vmexit, a DR access vmexit | |
6345 | * can (a) read the correct value of the debug registers, (b) set | |
6346 | * KVM_DEBUGREG_WONT_EXIT again. | |
6347 | */ | |
6348 | if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) { | |
6349 | int i; | |
6350 | ||
6351 | WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP); | |
6352 | kvm_x86_ops->sync_dirty_debug_regs(vcpu); | |
6353 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
6354 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
6355 | } | |
6356 | ||
24f1e32c FW |
6357 | /* |
6358 | * If the guest has used debug registers, at least dr7 | |
6359 | * will be disabled while returning to the host. | |
6360 | * If we don't have active breakpoints in the host, we don't | |
6361 | * care about the messed up debug address registers. But if | |
6362 | * we have some of them active, restore the old state. | |
6363 | */ | |
59d8eb53 | 6364 | if (hw_breakpoint_active()) |
24f1e32c | 6365 | hw_breakpoint_restore(); |
42dbaa5a | 6366 | |
886b470c MT |
6367 | vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu, |
6368 | native_read_tsc()); | |
1d5f066e | 6369 | |
6b7e2d09 | 6370 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 6371 | smp_wmb(); |
a547c6db YZ |
6372 | |
6373 | /* Interrupt is enabled by handle_external_intr() */ | |
6374 | kvm_x86_ops->handle_external_intr(vcpu); | |
b6c7a5dc HB |
6375 | |
6376 | ++vcpu->stat.exits; | |
6377 | ||
6378 | /* | |
6379 | * We must have an instruction between local_irq_enable() and | |
6380 | * kvm_guest_exit(), so the timer interrupt isn't delayed by | |
6381 | * the interrupt shadow. The stat.exits increment will do nicely. | |
6382 | * But we need to prevent reordering, hence this barrier(): | |
6383 | */ | |
6384 | barrier(); | |
6385 | ||
6386 | kvm_guest_exit(); | |
6387 | ||
6388 | preempt_enable(); | |
6389 | ||
f656ce01 | 6390 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
3200f405 | 6391 | |
b6c7a5dc HB |
6392 | /* |
6393 | * Profile KVM exit RIPs: | |
6394 | */ | |
6395 | if (unlikely(prof_on == KVM_PROFILING)) { | |
5fdbf976 MT |
6396 | unsigned long rip = kvm_rip_read(vcpu); |
6397 | profile_hit(KVM_PROFILING, (void *)rip); | |
b6c7a5dc HB |
6398 | } |
6399 | ||
cc578287 ZA |
6400 | if (unlikely(vcpu->arch.tsc_always_catchup)) |
6401 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); | |
298101da | 6402 | |
5cfb1d5a MT |
6403 | if (vcpu->arch.apic_attention) |
6404 | kvm_lapic_sync_from_vapic(vcpu); | |
b93463aa | 6405 | |
851ba692 | 6406 | r = kvm_x86_ops->handle_exit(vcpu); |
d905c069 MT |
6407 | return r; |
6408 | ||
6409 | cancel_injection: | |
6410 | kvm_x86_ops->cancel_injection(vcpu); | |
ae7a2a3f MT |
6411 | if (unlikely(vcpu->arch.apic_attention)) |
6412 | kvm_lapic_sync_from_vapic(vcpu); | |
d7690175 MT |
6413 | out: |
6414 | return r; | |
6415 | } | |
b6c7a5dc | 6416 | |
362c698f PB |
6417 | static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu) |
6418 | { | |
9c8fd1ba PB |
6419 | if (!kvm_arch_vcpu_runnable(vcpu)) { |
6420 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); | |
6421 | kvm_vcpu_block(vcpu); | |
6422 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); | |
6423 | if (!kvm_check_request(KVM_REQ_UNHALT, vcpu)) | |
6424 | return 1; | |
6425 | } | |
362c698f PB |
6426 | |
6427 | kvm_apic_accept_events(vcpu); | |
6428 | switch(vcpu->arch.mp_state) { | |
6429 | case KVM_MP_STATE_HALTED: | |
6430 | vcpu->arch.pv.pv_unhalted = false; | |
6431 | vcpu->arch.mp_state = | |
6432 | KVM_MP_STATE_RUNNABLE; | |
6433 | case KVM_MP_STATE_RUNNABLE: | |
6434 | vcpu->arch.apf.halted = false; | |
6435 | break; | |
6436 | case KVM_MP_STATE_INIT_RECEIVED: | |
6437 | break; | |
6438 | default: | |
6439 | return -EINTR; | |
6440 | break; | |
6441 | } | |
6442 | return 1; | |
6443 | } | |
09cec754 | 6444 | |
362c698f | 6445 | static int vcpu_run(struct kvm_vcpu *vcpu) |
d7690175 MT |
6446 | { |
6447 | int r; | |
f656ce01 | 6448 | struct kvm *kvm = vcpu->kvm; |
d7690175 | 6449 | |
f656ce01 | 6450 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
d7690175 | 6451 | |
362c698f | 6452 | for (;;) { |
af585b92 GN |
6453 | if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE && |
6454 | !vcpu->arch.apf.halted) | |
851ba692 | 6455 | r = vcpu_enter_guest(vcpu); |
362c698f PB |
6456 | else |
6457 | r = vcpu_block(kvm, vcpu); | |
09cec754 GN |
6458 | if (r <= 0) |
6459 | break; | |
6460 | ||
6461 | clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests); | |
6462 | if (kvm_cpu_has_pending_timer(vcpu)) | |
6463 | kvm_inject_pending_timer_irqs(vcpu); | |
6464 | ||
851ba692 | 6465 | if (dm_request_for_irq_injection(vcpu)) { |
09cec754 | 6466 | r = -EINTR; |
851ba692 | 6467 | vcpu->run->exit_reason = KVM_EXIT_INTR; |
09cec754 | 6468 | ++vcpu->stat.request_irq_exits; |
362c698f | 6469 | break; |
09cec754 | 6470 | } |
af585b92 GN |
6471 | |
6472 | kvm_check_async_pf_completion(vcpu); | |
6473 | ||
09cec754 GN |
6474 | if (signal_pending(current)) { |
6475 | r = -EINTR; | |
851ba692 | 6476 | vcpu->run->exit_reason = KVM_EXIT_INTR; |
09cec754 | 6477 | ++vcpu->stat.signal_exits; |
362c698f | 6478 | break; |
09cec754 GN |
6479 | } |
6480 | if (need_resched()) { | |
f656ce01 | 6481 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
c08ac06a | 6482 | cond_resched(); |
f656ce01 | 6483 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
d7690175 | 6484 | } |
b6c7a5dc HB |
6485 | } |
6486 | ||
f656ce01 | 6487 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
b6c7a5dc HB |
6488 | |
6489 | return r; | |
6490 | } | |
6491 | ||
716d51ab GN |
6492 | static inline int complete_emulated_io(struct kvm_vcpu *vcpu) |
6493 | { | |
6494 | int r; | |
6495 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); | |
6496 | r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE); | |
6497 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); | |
6498 | if (r != EMULATE_DONE) | |
6499 | return 0; | |
6500 | return 1; | |
6501 | } | |
6502 | ||
6503 | static int complete_emulated_pio(struct kvm_vcpu *vcpu) | |
6504 | { | |
6505 | BUG_ON(!vcpu->arch.pio.count); | |
6506 | ||
6507 | return complete_emulated_io(vcpu); | |
6508 | } | |
6509 | ||
f78146b0 AK |
6510 | /* |
6511 | * Implements the following, as a state machine: | |
6512 | * | |
6513 | * read: | |
6514 | * for each fragment | |
87da7e66 XG |
6515 | * for each mmio piece in the fragment |
6516 | * write gpa, len | |
6517 | * exit | |
6518 | * copy data | |
f78146b0 AK |
6519 | * execute insn |
6520 | * | |
6521 | * write: | |
6522 | * for each fragment | |
87da7e66 XG |
6523 | * for each mmio piece in the fragment |
6524 | * write gpa, len | |
6525 | * copy data | |
6526 | * exit | |
f78146b0 | 6527 | */ |
716d51ab | 6528 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu) |
5287f194 AK |
6529 | { |
6530 | struct kvm_run *run = vcpu->run; | |
f78146b0 | 6531 | struct kvm_mmio_fragment *frag; |
87da7e66 | 6532 | unsigned len; |
5287f194 | 6533 | |
716d51ab | 6534 | BUG_ON(!vcpu->mmio_needed); |
5287f194 | 6535 | |
716d51ab | 6536 | /* Complete previous fragment */ |
87da7e66 XG |
6537 | frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment]; |
6538 | len = min(8u, frag->len); | |
716d51ab | 6539 | if (!vcpu->mmio_is_write) |
87da7e66 XG |
6540 | memcpy(frag->data, run->mmio.data, len); |
6541 | ||
6542 | if (frag->len <= 8) { | |
6543 | /* Switch to the next fragment. */ | |
6544 | frag++; | |
6545 | vcpu->mmio_cur_fragment++; | |
6546 | } else { | |
6547 | /* Go forward to the next mmio piece. */ | |
6548 | frag->data += len; | |
6549 | frag->gpa += len; | |
6550 | frag->len -= len; | |
6551 | } | |
6552 | ||
a08d3b3b | 6553 | if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) { |
716d51ab | 6554 | vcpu->mmio_needed = 0; |
0912c977 PB |
6555 | |
6556 | /* FIXME: return into emulator if single-stepping. */ | |
cef4dea0 | 6557 | if (vcpu->mmio_is_write) |
716d51ab GN |
6558 | return 1; |
6559 | vcpu->mmio_read_completed = 1; | |
6560 | return complete_emulated_io(vcpu); | |
6561 | } | |
87da7e66 | 6562 | |
716d51ab GN |
6563 | run->exit_reason = KVM_EXIT_MMIO; |
6564 | run->mmio.phys_addr = frag->gpa; | |
6565 | if (vcpu->mmio_is_write) | |
87da7e66 XG |
6566 | memcpy(run->mmio.data, frag->data, min(8u, frag->len)); |
6567 | run->mmio.len = min(8u, frag->len); | |
716d51ab GN |
6568 | run->mmio.is_write = vcpu->mmio_is_write; |
6569 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; | |
6570 | return 0; | |
5287f194 AK |
6571 | } |
6572 | ||
716d51ab | 6573 | |
b6c7a5dc HB |
6574 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) |
6575 | { | |
c5bedc68 | 6576 | struct fpu *fpu = ¤t->thread.fpu; |
b6c7a5dc HB |
6577 | int r; |
6578 | sigset_t sigsaved; | |
6579 | ||
c4d72e2d | 6580 | fpu__activate_curr(fpu); |
e5c30142 | 6581 | |
ac9f6dc0 AK |
6582 | if (vcpu->sigset_active) |
6583 | sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); | |
6584 | ||
a4535290 | 6585 | if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) { |
b6c7a5dc | 6586 | kvm_vcpu_block(vcpu); |
66450a21 | 6587 | kvm_apic_accept_events(vcpu); |
d7690175 | 6588 | clear_bit(KVM_REQ_UNHALT, &vcpu->requests); |
ac9f6dc0 AK |
6589 | r = -EAGAIN; |
6590 | goto out; | |
b6c7a5dc HB |
6591 | } |
6592 | ||
b6c7a5dc | 6593 | /* re-sync apic's tpr */ |
eea1cff9 AP |
6594 | if (!irqchip_in_kernel(vcpu->kvm)) { |
6595 | if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) { | |
6596 | r = -EINVAL; | |
6597 | goto out; | |
6598 | } | |
6599 | } | |
b6c7a5dc | 6600 | |
716d51ab GN |
6601 | if (unlikely(vcpu->arch.complete_userspace_io)) { |
6602 | int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io; | |
6603 | vcpu->arch.complete_userspace_io = NULL; | |
6604 | r = cui(vcpu); | |
6605 | if (r <= 0) | |
6606 | goto out; | |
6607 | } else | |
6608 | WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed); | |
5287f194 | 6609 | |
362c698f | 6610 | r = vcpu_run(vcpu); |
b6c7a5dc HB |
6611 | |
6612 | out: | |
f1d86e46 | 6613 | post_kvm_run_save(vcpu); |
b6c7a5dc HB |
6614 | if (vcpu->sigset_active) |
6615 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); | |
6616 | ||
b6c7a5dc HB |
6617 | return r; |
6618 | } | |
6619 | ||
6620 | int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) | |
6621 | { | |
7ae441ea GN |
6622 | if (vcpu->arch.emulate_regs_need_sync_to_vcpu) { |
6623 | /* | |
6624 | * We are here if userspace calls get_regs() in the middle of | |
6625 | * instruction emulation. Registers state needs to be copied | |
4a969980 | 6626 | * back from emulation context to vcpu. Userspace shouldn't do |
7ae441ea GN |
6627 | * that usually, but some bad designed PV devices (vmware |
6628 | * backdoor interface) need this to work | |
6629 | */ | |
dd856efa | 6630 | emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt); |
7ae441ea GN |
6631 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
6632 | } | |
5fdbf976 MT |
6633 | regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX); |
6634 | regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX); | |
6635 | regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX); | |
6636 | regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX); | |
6637 | regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI); | |
6638 | regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI); | |
6639 | regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP); | |
6640 | regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP); | |
b6c7a5dc | 6641 | #ifdef CONFIG_X86_64 |
5fdbf976 MT |
6642 | regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8); |
6643 | regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9); | |
6644 | regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10); | |
6645 | regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11); | |
6646 | regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12); | |
6647 | regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13); | |
6648 | regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14); | |
6649 | regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15); | |
b6c7a5dc HB |
6650 | #endif |
6651 | ||
5fdbf976 | 6652 | regs->rip = kvm_rip_read(vcpu); |
91586a3b | 6653 | regs->rflags = kvm_get_rflags(vcpu); |
b6c7a5dc | 6654 | |
b6c7a5dc HB |
6655 | return 0; |
6656 | } | |
6657 | ||
6658 | int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) | |
6659 | { | |
7ae441ea GN |
6660 | vcpu->arch.emulate_regs_need_sync_from_vcpu = true; |
6661 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; | |
6662 | ||
5fdbf976 MT |
6663 | kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax); |
6664 | kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx); | |
6665 | kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx); | |
6666 | kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx); | |
6667 | kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi); | |
6668 | kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi); | |
6669 | kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp); | |
6670 | kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp); | |
b6c7a5dc | 6671 | #ifdef CONFIG_X86_64 |
5fdbf976 MT |
6672 | kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8); |
6673 | kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9); | |
6674 | kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10); | |
6675 | kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11); | |
6676 | kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12); | |
6677 | kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13); | |
6678 | kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14); | |
6679 | kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15); | |
b6c7a5dc HB |
6680 | #endif |
6681 | ||
5fdbf976 | 6682 | kvm_rip_write(vcpu, regs->rip); |
91586a3b | 6683 | kvm_set_rflags(vcpu, regs->rflags); |
b6c7a5dc | 6684 | |
b4f14abd JK |
6685 | vcpu->arch.exception.pending = false; |
6686 | ||
3842d135 AK |
6687 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
6688 | ||
b6c7a5dc HB |
6689 | return 0; |
6690 | } | |
6691 | ||
b6c7a5dc HB |
6692 | void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l) |
6693 | { | |
6694 | struct kvm_segment cs; | |
6695 | ||
3e6e0aab | 6696 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); |
b6c7a5dc HB |
6697 | *db = cs.db; |
6698 | *l = cs.l; | |
6699 | } | |
6700 | EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits); | |
6701 | ||
6702 | int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, | |
6703 | struct kvm_sregs *sregs) | |
6704 | { | |
89a27f4d | 6705 | struct desc_ptr dt; |
b6c7a5dc | 6706 | |
3e6e0aab GT |
6707 | kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
6708 | kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
6709 | kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
6710 | kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
6711 | kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
6712 | kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 6713 | |
3e6e0aab GT |
6714 | kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
6715 | kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc HB |
6716 | |
6717 | kvm_x86_ops->get_idt(vcpu, &dt); | |
89a27f4d GN |
6718 | sregs->idt.limit = dt.size; |
6719 | sregs->idt.base = dt.address; | |
b6c7a5dc | 6720 | kvm_x86_ops->get_gdt(vcpu, &dt); |
89a27f4d GN |
6721 | sregs->gdt.limit = dt.size; |
6722 | sregs->gdt.base = dt.address; | |
b6c7a5dc | 6723 | |
4d4ec087 | 6724 | sregs->cr0 = kvm_read_cr0(vcpu); |
ad312c7c | 6725 | sregs->cr2 = vcpu->arch.cr2; |
9f8fe504 | 6726 | sregs->cr3 = kvm_read_cr3(vcpu); |
fc78f519 | 6727 | sregs->cr4 = kvm_read_cr4(vcpu); |
2d3ad1f4 | 6728 | sregs->cr8 = kvm_get_cr8(vcpu); |
f6801dff | 6729 | sregs->efer = vcpu->arch.efer; |
b6c7a5dc HB |
6730 | sregs->apic_base = kvm_get_apic_base(vcpu); |
6731 | ||
923c61bb | 6732 | memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap); |
b6c7a5dc | 6733 | |
36752c9b | 6734 | if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft) |
14d0bc1f GN |
6735 | set_bit(vcpu->arch.interrupt.nr, |
6736 | (unsigned long *)sregs->interrupt_bitmap); | |
16d7a191 | 6737 | |
b6c7a5dc HB |
6738 | return 0; |
6739 | } | |
6740 | ||
62d9f0db MT |
6741 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
6742 | struct kvm_mp_state *mp_state) | |
6743 | { | |
66450a21 | 6744 | kvm_apic_accept_events(vcpu); |
6aef266c SV |
6745 | if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED && |
6746 | vcpu->arch.pv.pv_unhalted) | |
6747 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
6748 | else | |
6749 | mp_state->mp_state = vcpu->arch.mp_state; | |
6750 | ||
62d9f0db MT |
6751 | return 0; |
6752 | } | |
6753 | ||
6754 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
6755 | struct kvm_mp_state *mp_state) | |
6756 | { | |
66450a21 JK |
6757 | if (!kvm_vcpu_has_lapic(vcpu) && |
6758 | mp_state->mp_state != KVM_MP_STATE_RUNNABLE) | |
6759 | return -EINVAL; | |
6760 | ||
6761 | if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) { | |
6762 | vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED; | |
6763 | set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events); | |
6764 | } else | |
6765 | vcpu->arch.mp_state = mp_state->mp_state; | |
3842d135 | 6766 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
62d9f0db MT |
6767 | return 0; |
6768 | } | |
6769 | ||
7f3d35fd KW |
6770 | int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index, |
6771 | int reason, bool has_error_code, u32 error_code) | |
b6c7a5dc | 6772 | { |
9d74191a | 6773 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
8ec4722d | 6774 | int ret; |
e01c2426 | 6775 | |
8ec4722d | 6776 | init_emulate_ctxt(vcpu); |
c697518a | 6777 | |
7f3d35fd | 6778 | ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason, |
9d74191a | 6779 | has_error_code, error_code); |
c697518a | 6780 | |
c697518a | 6781 | if (ret) |
19d04437 | 6782 | return EMULATE_FAIL; |
37817f29 | 6783 | |
9d74191a TY |
6784 | kvm_rip_write(vcpu, ctxt->eip); |
6785 | kvm_set_rflags(vcpu, ctxt->eflags); | |
3842d135 | 6786 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
19d04437 | 6787 | return EMULATE_DONE; |
37817f29 IE |
6788 | } |
6789 | EXPORT_SYMBOL_GPL(kvm_task_switch); | |
6790 | ||
b6c7a5dc HB |
6791 | int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, |
6792 | struct kvm_sregs *sregs) | |
6793 | { | |
58cb628d | 6794 | struct msr_data apic_base_msr; |
b6c7a5dc | 6795 | int mmu_reset_needed = 0; |
63f42e02 | 6796 | int pending_vec, max_bits, idx; |
89a27f4d | 6797 | struct desc_ptr dt; |
b6c7a5dc | 6798 | |
6d1068b3 PM |
6799 | if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE)) |
6800 | return -EINVAL; | |
6801 | ||
89a27f4d GN |
6802 | dt.size = sregs->idt.limit; |
6803 | dt.address = sregs->idt.base; | |
b6c7a5dc | 6804 | kvm_x86_ops->set_idt(vcpu, &dt); |
89a27f4d GN |
6805 | dt.size = sregs->gdt.limit; |
6806 | dt.address = sregs->gdt.base; | |
b6c7a5dc HB |
6807 | kvm_x86_ops->set_gdt(vcpu, &dt); |
6808 | ||
ad312c7c | 6809 | vcpu->arch.cr2 = sregs->cr2; |
9f8fe504 | 6810 | mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3; |
dc7e795e | 6811 | vcpu->arch.cr3 = sregs->cr3; |
aff48baa | 6812 | __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail); |
b6c7a5dc | 6813 | |
2d3ad1f4 | 6814 | kvm_set_cr8(vcpu, sregs->cr8); |
b6c7a5dc | 6815 | |
f6801dff | 6816 | mmu_reset_needed |= vcpu->arch.efer != sregs->efer; |
b6c7a5dc | 6817 | kvm_x86_ops->set_efer(vcpu, sregs->efer); |
58cb628d JK |
6818 | apic_base_msr.data = sregs->apic_base; |
6819 | apic_base_msr.host_initiated = true; | |
6820 | kvm_set_apic_base(vcpu, &apic_base_msr); | |
b6c7a5dc | 6821 | |
4d4ec087 | 6822 | mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0; |
b6c7a5dc | 6823 | kvm_x86_ops->set_cr0(vcpu, sregs->cr0); |
d7306163 | 6824 | vcpu->arch.cr0 = sregs->cr0; |
b6c7a5dc | 6825 | |
fc78f519 | 6826 | mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4; |
b6c7a5dc | 6827 | kvm_x86_ops->set_cr4(vcpu, sregs->cr4); |
3ea3aa8c | 6828 | if (sregs->cr4 & X86_CR4_OSXSAVE) |
00b27a3e | 6829 | kvm_update_cpuid(vcpu); |
63f42e02 XG |
6830 | |
6831 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
7c93be44 | 6832 | if (!is_long_mode(vcpu) && is_pae(vcpu)) { |
9f8fe504 | 6833 | load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)); |
7c93be44 MT |
6834 | mmu_reset_needed = 1; |
6835 | } | |
63f42e02 | 6836 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
b6c7a5dc HB |
6837 | |
6838 | if (mmu_reset_needed) | |
6839 | kvm_mmu_reset_context(vcpu); | |
6840 | ||
a50abc3b | 6841 | max_bits = KVM_NR_INTERRUPTS; |
923c61bb GN |
6842 | pending_vec = find_first_bit( |
6843 | (const unsigned long *)sregs->interrupt_bitmap, max_bits); | |
6844 | if (pending_vec < max_bits) { | |
66fd3f7f | 6845 | kvm_queue_interrupt(vcpu, pending_vec, false); |
923c61bb | 6846 | pr_debug("Set back pending irq %d\n", pending_vec); |
b6c7a5dc HB |
6847 | } |
6848 | ||
3e6e0aab GT |
6849 | kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
6850 | kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
6851 | kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
6852 | kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
6853 | kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
6854 | kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 6855 | |
3e6e0aab GT |
6856 | kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
6857 | kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc | 6858 | |
5f0269f5 ME |
6859 | update_cr8_intercept(vcpu); |
6860 | ||
9c3e4aab | 6861 | /* Older userspace won't unhalt the vcpu on reset. */ |
c5af89b6 | 6862 | if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 && |
9c3e4aab | 6863 | sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 && |
3eeb3288 | 6864 | !is_protmode(vcpu)) |
9c3e4aab MT |
6865 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
6866 | ||
3842d135 AK |
6867 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
6868 | ||
b6c7a5dc HB |
6869 | return 0; |
6870 | } | |
6871 | ||
d0bfb940 JK |
6872 | int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, |
6873 | struct kvm_guest_debug *dbg) | |
b6c7a5dc | 6874 | { |
355be0b9 | 6875 | unsigned long rflags; |
ae675ef0 | 6876 | int i, r; |
b6c7a5dc | 6877 | |
4f926bf2 JK |
6878 | if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) { |
6879 | r = -EBUSY; | |
6880 | if (vcpu->arch.exception.pending) | |
2122ff5e | 6881 | goto out; |
4f926bf2 JK |
6882 | if (dbg->control & KVM_GUESTDBG_INJECT_DB) |
6883 | kvm_queue_exception(vcpu, DB_VECTOR); | |
6884 | else | |
6885 | kvm_queue_exception(vcpu, BP_VECTOR); | |
6886 | } | |
6887 | ||
91586a3b JK |
6888 | /* |
6889 | * Read rflags as long as potentially injected trace flags are still | |
6890 | * filtered out. | |
6891 | */ | |
6892 | rflags = kvm_get_rflags(vcpu); | |
355be0b9 JK |
6893 | |
6894 | vcpu->guest_debug = dbg->control; | |
6895 | if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE)) | |
6896 | vcpu->guest_debug = 0; | |
6897 | ||
6898 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) { | |
ae675ef0 JK |
6899 | for (i = 0; i < KVM_NR_DB_REGS; ++i) |
6900 | vcpu->arch.eff_db[i] = dbg->arch.debugreg[i]; | |
c8639010 | 6901 | vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7]; |
ae675ef0 JK |
6902 | } else { |
6903 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
6904 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
ae675ef0 | 6905 | } |
c8639010 | 6906 | kvm_update_dr7(vcpu); |
ae675ef0 | 6907 | |
f92653ee JK |
6908 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) |
6909 | vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) + | |
6910 | get_segment_base(vcpu, VCPU_SREG_CS); | |
94fe45da | 6911 | |
91586a3b JK |
6912 | /* |
6913 | * Trigger an rflags update that will inject or remove the trace | |
6914 | * flags. | |
6915 | */ | |
6916 | kvm_set_rflags(vcpu, rflags); | |
b6c7a5dc | 6917 | |
c8639010 | 6918 | kvm_x86_ops->update_db_bp_intercept(vcpu); |
b6c7a5dc | 6919 | |
4f926bf2 | 6920 | r = 0; |
d0bfb940 | 6921 | |
2122ff5e | 6922 | out: |
b6c7a5dc HB |
6923 | |
6924 | return r; | |
6925 | } | |
6926 | ||
8b006791 ZX |
6927 | /* |
6928 | * Translate a guest virtual address to a guest physical address. | |
6929 | */ | |
6930 | int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, | |
6931 | struct kvm_translation *tr) | |
6932 | { | |
6933 | unsigned long vaddr = tr->linear_address; | |
6934 | gpa_t gpa; | |
f656ce01 | 6935 | int idx; |
8b006791 | 6936 | |
f656ce01 | 6937 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
1871c602 | 6938 | gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL); |
f656ce01 | 6939 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
8b006791 ZX |
6940 | tr->physical_address = gpa; |
6941 | tr->valid = gpa != UNMAPPED_GVA; | |
6942 | tr->writeable = 1; | |
6943 | tr->usermode = 0; | |
8b006791 ZX |
6944 | |
6945 | return 0; | |
6946 | } | |
6947 | ||
d0752060 HB |
6948 | int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) |
6949 | { | |
c47ada30 | 6950 | struct fxregs_state *fxsave = |
7366ed77 | 6951 | &vcpu->arch.guest_fpu.state.fxsave; |
d0752060 | 6952 | |
d0752060 HB |
6953 | memcpy(fpu->fpr, fxsave->st_space, 128); |
6954 | fpu->fcw = fxsave->cwd; | |
6955 | fpu->fsw = fxsave->swd; | |
6956 | fpu->ftwx = fxsave->twd; | |
6957 | fpu->last_opcode = fxsave->fop; | |
6958 | fpu->last_ip = fxsave->rip; | |
6959 | fpu->last_dp = fxsave->rdp; | |
6960 | memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space); | |
6961 | ||
d0752060 HB |
6962 | return 0; |
6963 | } | |
6964 | ||
6965 | int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) | |
6966 | { | |
c47ada30 | 6967 | struct fxregs_state *fxsave = |
7366ed77 | 6968 | &vcpu->arch.guest_fpu.state.fxsave; |
d0752060 | 6969 | |
d0752060 HB |
6970 | memcpy(fxsave->st_space, fpu->fpr, 128); |
6971 | fxsave->cwd = fpu->fcw; | |
6972 | fxsave->swd = fpu->fsw; | |
6973 | fxsave->twd = fpu->ftwx; | |
6974 | fxsave->fop = fpu->last_opcode; | |
6975 | fxsave->rip = fpu->last_ip; | |
6976 | fxsave->rdp = fpu->last_dp; | |
6977 | memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space); | |
6978 | ||
d0752060 HB |
6979 | return 0; |
6980 | } | |
6981 | ||
0ee6a517 | 6982 | static void fx_init(struct kvm_vcpu *vcpu) |
d0752060 | 6983 | { |
bf935b0b | 6984 | fpstate_init(&vcpu->arch.guest_fpu.state); |
df1daba7 | 6985 | if (cpu_has_xsaves) |
7366ed77 | 6986 | vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv = |
df1daba7 | 6987 | host_xcr0 | XSTATE_COMPACTION_ENABLED; |
d0752060 | 6988 | |
2acf923e DC |
6989 | /* |
6990 | * Ensure guest xcr0 is valid for loading | |
6991 | */ | |
6992 | vcpu->arch.xcr0 = XSTATE_FP; | |
6993 | ||
ad312c7c | 6994 | vcpu->arch.cr0 |= X86_CR0_ET; |
d0752060 | 6995 | } |
d0752060 HB |
6996 | |
6997 | void kvm_load_guest_fpu(struct kvm_vcpu *vcpu) | |
6998 | { | |
2608d7a1 | 6999 | if (vcpu->guest_fpu_loaded) |
d0752060 HB |
7000 | return; |
7001 | ||
2acf923e DC |
7002 | /* |
7003 | * Restore all possible states in the guest, | |
7004 | * and assume host would use all available bits. | |
7005 | * Guest xcr0 would be loaded later. | |
7006 | */ | |
7007 | kvm_put_guest_xcr0(vcpu); | |
d0752060 | 7008 | vcpu->guest_fpu_loaded = 1; |
b1a74bf8 | 7009 | __kernel_fpu_begin(); |
003e2e8b | 7010 | __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state); |
0c04851c | 7011 | trace_kvm_fpu(1); |
d0752060 | 7012 | } |
d0752060 HB |
7013 | |
7014 | void kvm_put_guest_fpu(struct kvm_vcpu *vcpu) | |
7015 | { | |
2acf923e DC |
7016 | kvm_put_guest_xcr0(vcpu); |
7017 | ||
653f52c3 RR |
7018 | if (!vcpu->guest_fpu_loaded) { |
7019 | vcpu->fpu_counter = 0; | |
d0752060 | 7020 | return; |
653f52c3 | 7021 | } |
d0752060 HB |
7022 | |
7023 | vcpu->guest_fpu_loaded = 0; | |
4f836347 | 7024 | copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu); |
b1a74bf8 | 7025 | __kernel_fpu_end(); |
f096ed85 | 7026 | ++vcpu->stat.fpu_reload; |
653f52c3 RR |
7027 | /* |
7028 | * If using eager FPU mode, or if the guest is a frequent user | |
7029 | * of the FPU, just leave the FPU active for next time. | |
7030 | * Every 255 times fpu_counter rolls over to 0; a guest that uses | |
7031 | * the FPU in bursts will revert to loading it on demand. | |
7032 | */ | |
a9b4fb7e | 7033 | if (!vcpu->arch.eager_fpu) { |
653f52c3 RR |
7034 | if (++vcpu->fpu_counter < 5) |
7035 | kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu); | |
7036 | } | |
0c04851c | 7037 | trace_kvm_fpu(0); |
d0752060 | 7038 | } |
e9b11c17 ZX |
7039 | |
7040 | void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) | |
7041 | { | |
12f9a48f | 7042 | kvmclock_reset(vcpu); |
7f1ea208 | 7043 | |
f5f48ee1 | 7044 | free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); |
e9b11c17 ZX |
7045 | kvm_x86_ops->vcpu_free(vcpu); |
7046 | } | |
7047 | ||
7048 | struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, | |
7049 | unsigned int id) | |
7050 | { | |
c447e76b LL |
7051 | struct kvm_vcpu *vcpu; |
7052 | ||
6755bae8 ZA |
7053 | if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0) |
7054 | printk_once(KERN_WARNING | |
7055 | "kvm: SMP vm created on host with unstable TSC; " | |
7056 | "guest TSC will not be reliable\n"); | |
c447e76b LL |
7057 | |
7058 | vcpu = kvm_x86_ops->vcpu_create(kvm, id); | |
7059 | ||
c447e76b | 7060 | return vcpu; |
26e5215f | 7061 | } |
e9b11c17 | 7062 | |
26e5215f AK |
7063 | int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu) |
7064 | { | |
7065 | int r; | |
e9b11c17 | 7066 | |
19efffa2 | 7067 | kvm_vcpu_mtrr_init(vcpu); |
9fc77441 MT |
7068 | r = vcpu_load(vcpu); |
7069 | if (r) | |
7070 | return r; | |
d28bc9dd | 7071 | kvm_vcpu_reset(vcpu, false); |
8a3c1a33 | 7072 | kvm_mmu_setup(vcpu); |
e9b11c17 | 7073 | vcpu_put(vcpu); |
26e5215f | 7074 | return r; |
e9b11c17 ZX |
7075 | } |
7076 | ||
31928aa5 | 7077 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
42897d86 | 7078 | { |
8fe8ab46 | 7079 | struct msr_data msr; |
332967a3 | 7080 | struct kvm *kvm = vcpu->kvm; |
42897d86 | 7081 | |
31928aa5 DD |
7082 | if (vcpu_load(vcpu)) |
7083 | return; | |
8fe8ab46 WA |
7084 | msr.data = 0x0; |
7085 | msr.index = MSR_IA32_TSC; | |
7086 | msr.host_initiated = true; | |
7087 | kvm_write_tsc(vcpu, &msr); | |
42897d86 MT |
7088 | vcpu_put(vcpu); |
7089 | ||
630994b3 MT |
7090 | if (!kvmclock_periodic_sync) |
7091 | return; | |
7092 | ||
332967a3 AJ |
7093 | schedule_delayed_work(&kvm->arch.kvmclock_sync_work, |
7094 | KVMCLOCK_SYNC_PERIOD); | |
42897d86 MT |
7095 | } |
7096 | ||
d40ccc62 | 7097 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) |
e9b11c17 | 7098 | { |
9fc77441 | 7099 | int r; |
344d9588 GN |
7100 | vcpu->arch.apf.msr_val = 0; |
7101 | ||
9fc77441 MT |
7102 | r = vcpu_load(vcpu); |
7103 | BUG_ON(r); | |
e9b11c17 ZX |
7104 | kvm_mmu_unload(vcpu); |
7105 | vcpu_put(vcpu); | |
7106 | ||
7107 | kvm_x86_ops->vcpu_free(vcpu); | |
7108 | } | |
7109 | ||
d28bc9dd | 7110 | void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) |
e9b11c17 | 7111 | { |
e69fab5d PB |
7112 | vcpu->arch.hflags = 0; |
7113 | ||
7460fb4a AK |
7114 | atomic_set(&vcpu->arch.nmi_queued, 0); |
7115 | vcpu->arch.nmi_pending = 0; | |
448fa4a9 | 7116 | vcpu->arch.nmi_injected = false; |
5f7552d4 NA |
7117 | kvm_clear_interrupt_queue(vcpu); |
7118 | kvm_clear_exception_queue(vcpu); | |
448fa4a9 | 7119 | |
42dbaa5a | 7120 | memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db)); |
ae561ede | 7121 | kvm_update_dr0123(vcpu); |
6f43ed01 | 7122 | vcpu->arch.dr6 = DR6_INIT; |
73aaf249 | 7123 | kvm_update_dr6(vcpu); |
42dbaa5a | 7124 | vcpu->arch.dr7 = DR7_FIXED_1; |
c8639010 | 7125 | kvm_update_dr7(vcpu); |
42dbaa5a | 7126 | |
1119022c NA |
7127 | vcpu->arch.cr2 = 0; |
7128 | ||
3842d135 | 7129 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
344d9588 | 7130 | vcpu->arch.apf.msr_val = 0; |
c9aaa895 | 7131 | vcpu->arch.st.msr_val = 0; |
3842d135 | 7132 | |
12f9a48f GC |
7133 | kvmclock_reset(vcpu); |
7134 | ||
af585b92 GN |
7135 | kvm_clear_async_pf_completion_queue(vcpu); |
7136 | kvm_async_pf_hash_reset(vcpu); | |
7137 | vcpu->arch.apf.halted = false; | |
3842d135 | 7138 | |
64d60670 | 7139 | if (!init_event) { |
d28bc9dd | 7140 | kvm_pmu_reset(vcpu); |
64d60670 PB |
7141 | vcpu->arch.smbase = 0x30000; |
7142 | } | |
f5132b01 | 7143 | |
66f7b72e JS |
7144 | memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs)); |
7145 | vcpu->arch.regs_avail = ~0; | |
7146 | vcpu->arch.regs_dirty = ~0; | |
7147 | ||
d28bc9dd | 7148 | kvm_x86_ops->vcpu_reset(vcpu, init_event); |
e9b11c17 ZX |
7149 | } |
7150 | ||
2b4a273b | 7151 | void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector) |
66450a21 JK |
7152 | { |
7153 | struct kvm_segment cs; | |
7154 | ||
7155 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
7156 | cs.selector = vector << 8; | |
7157 | cs.base = vector << 12; | |
7158 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
7159 | kvm_rip_write(vcpu, 0); | |
e9b11c17 ZX |
7160 | } |
7161 | ||
13a34e06 | 7162 | int kvm_arch_hardware_enable(void) |
e9b11c17 | 7163 | { |
ca84d1a2 ZA |
7164 | struct kvm *kvm; |
7165 | struct kvm_vcpu *vcpu; | |
7166 | int i; | |
0dd6a6ed ZA |
7167 | int ret; |
7168 | u64 local_tsc; | |
7169 | u64 max_tsc = 0; | |
7170 | bool stable, backwards_tsc = false; | |
18863bdd AK |
7171 | |
7172 | kvm_shared_msr_cpu_online(); | |
13a34e06 | 7173 | ret = kvm_x86_ops->hardware_enable(); |
0dd6a6ed ZA |
7174 | if (ret != 0) |
7175 | return ret; | |
7176 | ||
7177 | local_tsc = native_read_tsc(); | |
7178 | stable = !check_tsc_unstable(); | |
7179 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
7180 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
7181 | if (!stable && vcpu->cpu == smp_processor_id()) | |
105b21bb | 7182 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0dd6a6ed ZA |
7183 | if (stable && vcpu->arch.last_host_tsc > local_tsc) { |
7184 | backwards_tsc = true; | |
7185 | if (vcpu->arch.last_host_tsc > max_tsc) | |
7186 | max_tsc = vcpu->arch.last_host_tsc; | |
7187 | } | |
7188 | } | |
7189 | } | |
7190 | ||
7191 | /* | |
7192 | * Sometimes, even reliable TSCs go backwards. This happens on | |
7193 | * platforms that reset TSC during suspend or hibernate actions, but | |
7194 | * maintain synchronization. We must compensate. Fortunately, we can | |
7195 | * detect that condition here, which happens early in CPU bringup, | |
7196 | * before any KVM threads can be running. Unfortunately, we can't | |
7197 | * bring the TSCs fully up to date with real time, as we aren't yet far | |
7198 | * enough into CPU bringup that we know how much real time has actually | |
7199 | * elapsed; our helper function, get_kernel_ns() will be using boot | |
7200 | * variables that haven't been updated yet. | |
7201 | * | |
7202 | * So we simply find the maximum observed TSC above, then record the | |
7203 | * adjustment to TSC in each VCPU. When the VCPU later gets loaded, | |
7204 | * the adjustment will be applied. Note that we accumulate | |
7205 | * adjustments, in case multiple suspend cycles happen before some VCPU | |
7206 | * gets a chance to run again. In the event that no KVM threads get a | |
7207 | * chance to run, we will miss the entire elapsed period, as we'll have | |
7208 | * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may | |
7209 | * loose cycle time. This isn't too big a deal, since the loss will be | |
7210 | * uniform across all VCPUs (not to mention the scenario is extremely | |
7211 | * unlikely). It is possible that a second hibernate recovery happens | |
7212 | * much faster than a first, causing the observed TSC here to be | |
7213 | * smaller; this would require additional padding adjustment, which is | |
7214 | * why we set last_host_tsc to the local tsc observed here. | |
7215 | * | |
7216 | * N.B. - this code below runs only on platforms with reliable TSC, | |
7217 | * as that is the only way backwards_tsc is set above. Also note | |
7218 | * that this runs for ALL vcpus, which is not a bug; all VCPUs should | |
7219 | * have the same delta_cyc adjustment applied if backwards_tsc | |
7220 | * is detected. Note further, this adjustment is only done once, | |
7221 | * as we reset last_host_tsc on all VCPUs to stop this from being | |
7222 | * called multiple times (one for each physical CPU bringup). | |
7223 | * | |
4a969980 | 7224 | * Platforms with unreliable TSCs don't have to deal with this, they |
0dd6a6ed ZA |
7225 | * will be compensated by the logic in vcpu_load, which sets the TSC to |
7226 | * catchup mode. This will catchup all VCPUs to real time, but cannot | |
7227 | * guarantee that they stay in perfect synchronization. | |
7228 | */ | |
7229 | if (backwards_tsc) { | |
7230 | u64 delta_cyc = max_tsc - local_tsc; | |
16a96021 | 7231 | backwards_tsc_observed = true; |
0dd6a6ed ZA |
7232 | list_for_each_entry(kvm, &vm_list, vm_list) { |
7233 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
7234 | vcpu->arch.tsc_offset_adjustment += delta_cyc; | |
7235 | vcpu->arch.last_host_tsc = local_tsc; | |
105b21bb | 7236 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
0dd6a6ed ZA |
7237 | } |
7238 | ||
7239 | /* | |
7240 | * We have to disable TSC offset matching.. if you were | |
7241 | * booting a VM while issuing an S4 host suspend.... | |
7242 | * you may have some problem. Solving this issue is | |
7243 | * left as an exercise to the reader. | |
7244 | */ | |
7245 | kvm->arch.last_tsc_nsec = 0; | |
7246 | kvm->arch.last_tsc_write = 0; | |
7247 | } | |
7248 | ||
7249 | } | |
7250 | return 0; | |
e9b11c17 ZX |
7251 | } |
7252 | ||
13a34e06 | 7253 | void kvm_arch_hardware_disable(void) |
e9b11c17 | 7254 | { |
13a34e06 RK |
7255 | kvm_x86_ops->hardware_disable(); |
7256 | drop_user_return_notifiers(); | |
e9b11c17 ZX |
7257 | } |
7258 | ||
7259 | int kvm_arch_hardware_setup(void) | |
7260 | { | |
9e9c3fe4 NA |
7261 | int r; |
7262 | ||
7263 | r = kvm_x86_ops->hardware_setup(); | |
7264 | if (r != 0) | |
7265 | return r; | |
7266 | ||
7267 | kvm_init_msr_list(); | |
7268 | return 0; | |
e9b11c17 ZX |
7269 | } |
7270 | ||
7271 | void kvm_arch_hardware_unsetup(void) | |
7272 | { | |
7273 | kvm_x86_ops->hardware_unsetup(); | |
7274 | } | |
7275 | ||
7276 | void kvm_arch_check_processor_compat(void *rtn) | |
7277 | { | |
7278 | kvm_x86_ops->check_processor_compatibility(rtn); | |
7279 | } | |
7280 | ||
3e515705 AK |
7281 | bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu) |
7282 | { | |
7283 | return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL); | |
7284 | } | |
7285 | ||
54e9818f GN |
7286 | struct static_key kvm_no_apic_vcpu __read_mostly; |
7287 | ||
e9b11c17 ZX |
7288 | int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) |
7289 | { | |
7290 | struct page *page; | |
7291 | struct kvm *kvm; | |
7292 | int r; | |
7293 | ||
7294 | BUG_ON(vcpu->kvm == NULL); | |
7295 | kvm = vcpu->kvm; | |
7296 | ||
6aef266c | 7297 | vcpu->arch.pv.pv_unhalted = false; |
9aabc88f | 7298 | vcpu->arch.emulate_ctxt.ops = &emulate_ops; |
58d269d8 | 7299 | if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu)) |
a4535290 | 7300 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
e9b11c17 | 7301 | else |
a4535290 | 7302 | vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED; |
e9b11c17 ZX |
7303 | |
7304 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
7305 | if (!page) { | |
7306 | r = -ENOMEM; | |
7307 | goto fail; | |
7308 | } | |
ad312c7c | 7309 | vcpu->arch.pio_data = page_address(page); |
e9b11c17 | 7310 | |
cc578287 | 7311 | kvm_set_tsc_khz(vcpu, max_tsc_khz); |
c285545f | 7312 | |
e9b11c17 ZX |
7313 | r = kvm_mmu_create(vcpu); |
7314 | if (r < 0) | |
7315 | goto fail_free_pio_data; | |
7316 | ||
7317 | if (irqchip_in_kernel(kvm)) { | |
7318 | r = kvm_create_lapic(vcpu); | |
7319 | if (r < 0) | |
7320 | goto fail_mmu_destroy; | |
54e9818f GN |
7321 | } else |
7322 | static_key_slow_inc(&kvm_no_apic_vcpu); | |
e9b11c17 | 7323 | |
890ca9ae HY |
7324 | vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4, |
7325 | GFP_KERNEL); | |
7326 | if (!vcpu->arch.mce_banks) { | |
7327 | r = -ENOMEM; | |
443c39bc | 7328 | goto fail_free_lapic; |
890ca9ae HY |
7329 | } |
7330 | vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS; | |
7331 | ||
f1797359 WY |
7332 | if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) { |
7333 | r = -ENOMEM; | |
f5f48ee1 | 7334 | goto fail_free_mce_banks; |
f1797359 | 7335 | } |
f5f48ee1 | 7336 | |
0ee6a517 | 7337 | fx_init(vcpu); |
66f7b72e | 7338 | |
ba904635 | 7339 | vcpu->arch.ia32_tsc_adjust_msr = 0x0; |
0b79459b | 7340 | vcpu->arch.pv_time_enabled = false; |
d7876f1b PB |
7341 | |
7342 | vcpu->arch.guest_supported_xcr0 = 0; | |
4344ee98 | 7343 | vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; |
d7876f1b | 7344 | |
5a4f55cd EK |
7345 | vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu); |
7346 | ||
74545705 RK |
7347 | vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT; |
7348 | ||
af585b92 | 7349 | kvm_async_pf_hash_reset(vcpu); |
f5132b01 | 7350 | kvm_pmu_init(vcpu); |
af585b92 | 7351 | |
e9b11c17 | 7352 | return 0; |
0ee6a517 | 7353 | |
f5f48ee1 SY |
7354 | fail_free_mce_banks: |
7355 | kfree(vcpu->arch.mce_banks); | |
443c39bc WY |
7356 | fail_free_lapic: |
7357 | kvm_free_lapic(vcpu); | |
e9b11c17 ZX |
7358 | fail_mmu_destroy: |
7359 | kvm_mmu_destroy(vcpu); | |
7360 | fail_free_pio_data: | |
ad312c7c | 7361 | free_page((unsigned long)vcpu->arch.pio_data); |
e9b11c17 ZX |
7362 | fail: |
7363 | return r; | |
7364 | } | |
7365 | ||
7366 | void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) | |
7367 | { | |
f656ce01 MT |
7368 | int idx; |
7369 | ||
f5132b01 | 7370 | kvm_pmu_destroy(vcpu); |
36cb93fd | 7371 | kfree(vcpu->arch.mce_banks); |
e9b11c17 | 7372 | kvm_free_lapic(vcpu); |
f656ce01 | 7373 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
e9b11c17 | 7374 | kvm_mmu_destroy(vcpu); |
f656ce01 | 7375 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
ad312c7c | 7376 | free_page((unsigned long)vcpu->arch.pio_data); |
54e9818f GN |
7377 | if (!irqchip_in_kernel(vcpu->kvm)) |
7378 | static_key_slow_dec(&kvm_no_apic_vcpu); | |
e9b11c17 | 7379 | } |
d19a9cd2 | 7380 | |
e790d9ef RK |
7381 | void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) |
7382 | { | |
ae97a3b8 | 7383 | kvm_x86_ops->sched_in(vcpu, cpu); |
e790d9ef RK |
7384 | } |
7385 | ||
e08b9637 | 7386 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
d19a9cd2 | 7387 | { |
e08b9637 CO |
7388 | if (type) |
7389 | return -EINVAL; | |
7390 | ||
6ef768fa | 7391 | INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list); |
f05e70ac | 7392 | INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); |
365c8868 | 7393 | INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages); |
4d5c5d0f | 7394 | INIT_LIST_HEAD(&kvm->arch.assigned_dev_head); |
e0f0bbc5 | 7395 | atomic_set(&kvm->arch.noncoherent_dma_count, 0); |
d19a9cd2 | 7396 | |
5550af4d SY |
7397 | /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */ |
7398 | set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap); | |
7a84428a AW |
7399 | /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */ |
7400 | set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, | |
7401 | &kvm->arch.irq_sources_bitmap); | |
5550af4d | 7402 | |
038f8c11 | 7403 | raw_spin_lock_init(&kvm->arch.tsc_write_lock); |
1e08ec4a | 7404 | mutex_init(&kvm->arch.apic_map_lock); |
d828199e MT |
7405 | spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock); |
7406 | ||
7407 | pvclock_update_vm_gtod_copy(kvm); | |
53f658b3 | 7408 | |
7e44e449 | 7409 | INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn); |
332967a3 | 7410 | INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn); |
7e44e449 | 7411 | |
d89f5eff | 7412 | return 0; |
d19a9cd2 ZX |
7413 | } |
7414 | ||
7415 | static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu) | |
7416 | { | |
9fc77441 MT |
7417 | int r; |
7418 | r = vcpu_load(vcpu); | |
7419 | BUG_ON(r); | |
d19a9cd2 ZX |
7420 | kvm_mmu_unload(vcpu); |
7421 | vcpu_put(vcpu); | |
7422 | } | |
7423 | ||
7424 | static void kvm_free_vcpus(struct kvm *kvm) | |
7425 | { | |
7426 | unsigned int i; | |
988a2cae | 7427 | struct kvm_vcpu *vcpu; |
d19a9cd2 ZX |
7428 | |
7429 | /* | |
7430 | * Unpin any mmu pages first. | |
7431 | */ | |
af585b92 GN |
7432 | kvm_for_each_vcpu(i, vcpu, kvm) { |
7433 | kvm_clear_async_pf_completion_queue(vcpu); | |
988a2cae | 7434 | kvm_unload_vcpu_mmu(vcpu); |
af585b92 | 7435 | } |
988a2cae GN |
7436 | kvm_for_each_vcpu(i, vcpu, kvm) |
7437 | kvm_arch_vcpu_free(vcpu); | |
7438 | ||
7439 | mutex_lock(&kvm->lock); | |
7440 | for (i = 0; i < atomic_read(&kvm->online_vcpus); i++) | |
7441 | kvm->vcpus[i] = NULL; | |
d19a9cd2 | 7442 | |
988a2cae GN |
7443 | atomic_set(&kvm->online_vcpus, 0); |
7444 | mutex_unlock(&kvm->lock); | |
d19a9cd2 ZX |
7445 | } |
7446 | ||
ad8ba2cd SY |
7447 | void kvm_arch_sync_events(struct kvm *kvm) |
7448 | { | |
332967a3 | 7449 | cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work); |
7e44e449 | 7450 | cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work); |
ba4cef31 | 7451 | kvm_free_all_assigned_devices(kvm); |
aea924f6 | 7452 | kvm_free_pit(kvm); |
ad8ba2cd SY |
7453 | } |
7454 | ||
9da0e4d5 PB |
7455 | int __x86_set_memory_region(struct kvm *kvm, |
7456 | const struct kvm_userspace_memory_region *mem) | |
7457 | { | |
7458 | int i, r; | |
7459 | ||
7460 | /* Called with kvm->slots_lock held. */ | |
7461 | BUG_ON(mem->slot >= KVM_MEM_SLOTS_NUM); | |
7462 | ||
7463 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
7464 | struct kvm_userspace_memory_region m = *mem; | |
7465 | ||
7466 | m.slot |= i << 16; | |
7467 | r = __kvm_set_memory_region(kvm, &m); | |
7468 | if (r < 0) | |
7469 | return r; | |
7470 | } | |
7471 | ||
7472 | return 0; | |
7473 | } | |
7474 | EXPORT_SYMBOL_GPL(__x86_set_memory_region); | |
7475 | ||
7476 | int x86_set_memory_region(struct kvm *kvm, | |
7477 | const struct kvm_userspace_memory_region *mem) | |
7478 | { | |
7479 | int r; | |
7480 | ||
7481 | mutex_lock(&kvm->slots_lock); | |
7482 | r = __x86_set_memory_region(kvm, mem); | |
7483 | mutex_unlock(&kvm->slots_lock); | |
7484 | ||
7485 | return r; | |
7486 | } | |
7487 | EXPORT_SYMBOL_GPL(x86_set_memory_region); | |
7488 | ||
d19a9cd2 ZX |
7489 | void kvm_arch_destroy_vm(struct kvm *kvm) |
7490 | { | |
27469d29 AH |
7491 | if (current->mm == kvm->mm) { |
7492 | /* | |
7493 | * Free memory regions allocated on behalf of userspace, | |
7494 | * unless the the memory map has changed due to process exit | |
7495 | * or fd copying. | |
7496 | */ | |
7497 | struct kvm_userspace_memory_region mem; | |
7498 | memset(&mem, 0, sizeof(mem)); | |
7499 | mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT; | |
9da0e4d5 | 7500 | x86_set_memory_region(kvm, &mem); |
27469d29 AH |
7501 | |
7502 | mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT; | |
9da0e4d5 | 7503 | x86_set_memory_region(kvm, &mem); |
27469d29 AH |
7504 | |
7505 | mem.slot = TSS_PRIVATE_MEMSLOT; | |
9da0e4d5 | 7506 | x86_set_memory_region(kvm, &mem); |
27469d29 | 7507 | } |
6eb55818 | 7508 | kvm_iommu_unmap_guest(kvm); |
d7deeeb0 ZX |
7509 | kfree(kvm->arch.vpic); |
7510 | kfree(kvm->arch.vioapic); | |
d19a9cd2 | 7511 | kvm_free_vcpus(kvm); |
1e08ec4a | 7512 | kfree(rcu_dereference_check(kvm->arch.apic_map, 1)); |
d19a9cd2 | 7513 | } |
0de10343 | 7514 | |
5587027c | 7515 | void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, |
db3fe4eb TY |
7516 | struct kvm_memory_slot *dont) |
7517 | { | |
7518 | int i; | |
7519 | ||
d89cc617 TY |
7520 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
7521 | if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) { | |
548ef284 | 7522 | kvfree(free->arch.rmap[i]); |
d89cc617 | 7523 | free->arch.rmap[i] = NULL; |
77d11309 | 7524 | } |
d89cc617 TY |
7525 | if (i == 0) |
7526 | continue; | |
7527 | ||
7528 | if (!dont || free->arch.lpage_info[i - 1] != | |
7529 | dont->arch.lpage_info[i - 1]) { | |
548ef284 | 7530 | kvfree(free->arch.lpage_info[i - 1]); |
d89cc617 | 7531 | free->arch.lpage_info[i - 1] = NULL; |
db3fe4eb TY |
7532 | } |
7533 | } | |
7534 | } | |
7535 | ||
5587027c AK |
7536 | int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot, |
7537 | unsigned long npages) | |
db3fe4eb TY |
7538 | { |
7539 | int i; | |
7540 | ||
d89cc617 | 7541 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
db3fe4eb TY |
7542 | unsigned long ugfn; |
7543 | int lpages; | |
d89cc617 | 7544 | int level = i + 1; |
db3fe4eb TY |
7545 | |
7546 | lpages = gfn_to_index(slot->base_gfn + npages - 1, | |
7547 | slot->base_gfn, level) + 1; | |
7548 | ||
d89cc617 TY |
7549 | slot->arch.rmap[i] = |
7550 | kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i])); | |
7551 | if (!slot->arch.rmap[i]) | |
77d11309 | 7552 | goto out_free; |
d89cc617 TY |
7553 | if (i == 0) |
7554 | continue; | |
77d11309 | 7555 | |
d89cc617 TY |
7556 | slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages * |
7557 | sizeof(*slot->arch.lpage_info[i - 1])); | |
7558 | if (!slot->arch.lpage_info[i - 1]) | |
db3fe4eb TY |
7559 | goto out_free; |
7560 | ||
7561 | if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1)) | |
d89cc617 | 7562 | slot->arch.lpage_info[i - 1][0].write_count = 1; |
db3fe4eb | 7563 | if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1)) |
d89cc617 | 7564 | slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1; |
db3fe4eb TY |
7565 | ugfn = slot->userspace_addr >> PAGE_SHIFT; |
7566 | /* | |
7567 | * If the gfn and userspace address are not aligned wrt each | |
7568 | * other, or if explicitly asked to, disable large page | |
7569 | * support for this slot | |
7570 | */ | |
7571 | if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) || | |
7572 | !kvm_largepages_enabled()) { | |
7573 | unsigned long j; | |
7574 | ||
7575 | for (j = 0; j < lpages; ++j) | |
d89cc617 | 7576 | slot->arch.lpage_info[i - 1][j].write_count = 1; |
db3fe4eb TY |
7577 | } |
7578 | } | |
7579 | ||
7580 | return 0; | |
7581 | ||
7582 | out_free: | |
d89cc617 | 7583 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
548ef284 | 7584 | kvfree(slot->arch.rmap[i]); |
d89cc617 TY |
7585 | slot->arch.rmap[i] = NULL; |
7586 | if (i == 0) | |
7587 | continue; | |
7588 | ||
548ef284 | 7589 | kvfree(slot->arch.lpage_info[i - 1]); |
d89cc617 | 7590 | slot->arch.lpage_info[i - 1] = NULL; |
db3fe4eb TY |
7591 | } |
7592 | return -ENOMEM; | |
7593 | } | |
7594 | ||
15f46015 | 7595 | void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots) |
e59dbe09 | 7596 | { |
e6dff7d1 TY |
7597 | /* |
7598 | * memslots->generation has been incremented. | |
7599 | * mmio generation may have reached its maximum value. | |
7600 | */ | |
54bf36aa | 7601 | kvm_mmu_invalidate_mmio_sptes(kvm, slots); |
e59dbe09 TY |
7602 | } |
7603 | ||
f7784b8e MT |
7604 | int kvm_arch_prepare_memory_region(struct kvm *kvm, |
7605 | struct kvm_memory_slot *memslot, | |
09170a49 | 7606 | const struct kvm_userspace_memory_region *mem, |
7b6195a9 | 7607 | enum kvm_mr_change change) |
0de10343 | 7608 | { |
7a905b14 TY |
7609 | /* |
7610 | * Only private memory slots need to be mapped here since | |
7611 | * KVM_SET_MEMORY_REGION ioctl is no longer supported. | |
0de10343 | 7612 | */ |
7b6195a9 | 7613 | if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) { |
7a905b14 | 7614 | unsigned long userspace_addr; |
604b38ac | 7615 | |
7a905b14 TY |
7616 | /* |
7617 | * MAP_SHARED to prevent internal slot pages from being moved | |
7618 | * by fork()/COW. | |
7619 | */ | |
7b6195a9 | 7620 | userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE, |
7a905b14 TY |
7621 | PROT_READ | PROT_WRITE, |
7622 | MAP_SHARED | MAP_ANONYMOUS, 0); | |
0de10343 | 7623 | |
7a905b14 TY |
7624 | if (IS_ERR((void *)userspace_addr)) |
7625 | return PTR_ERR((void *)userspace_addr); | |
604b38ac | 7626 | |
7a905b14 | 7627 | memslot->userspace_addr = userspace_addr; |
0de10343 ZX |
7628 | } |
7629 | ||
f7784b8e MT |
7630 | return 0; |
7631 | } | |
7632 | ||
88178fd4 KH |
7633 | static void kvm_mmu_slot_apply_flags(struct kvm *kvm, |
7634 | struct kvm_memory_slot *new) | |
7635 | { | |
7636 | /* Still write protect RO slot */ | |
7637 | if (new->flags & KVM_MEM_READONLY) { | |
7638 | kvm_mmu_slot_remove_write_access(kvm, new); | |
7639 | return; | |
7640 | } | |
7641 | ||
7642 | /* | |
7643 | * Call kvm_x86_ops dirty logging hooks when they are valid. | |
7644 | * | |
7645 | * kvm_x86_ops->slot_disable_log_dirty is called when: | |
7646 | * | |
7647 | * - KVM_MR_CREATE with dirty logging is disabled | |
7648 | * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag | |
7649 | * | |
7650 | * The reason is, in case of PML, we need to set D-bit for any slots | |
7651 | * with dirty logging disabled in order to eliminate unnecessary GPA | |
7652 | * logging in PML buffer (and potential PML buffer full VMEXT). This | |
7653 | * guarantees leaving PML enabled during guest's lifetime won't have | |
7654 | * any additonal overhead from PML when guest is running with dirty | |
7655 | * logging disabled for memory slots. | |
7656 | * | |
7657 | * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot | |
7658 | * to dirty logging mode. | |
7659 | * | |
7660 | * If kvm_x86_ops dirty logging hooks are invalid, use write protect. | |
7661 | * | |
7662 | * In case of write protect: | |
7663 | * | |
7664 | * Write protect all pages for dirty logging. | |
7665 | * | |
7666 | * All the sptes including the large sptes which point to this | |
7667 | * slot are set to readonly. We can not create any new large | |
7668 | * spte on this slot until the end of the logging. | |
7669 | * | |
7670 | * See the comments in fast_page_fault(). | |
7671 | */ | |
7672 | if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
7673 | if (kvm_x86_ops->slot_enable_log_dirty) | |
7674 | kvm_x86_ops->slot_enable_log_dirty(kvm, new); | |
7675 | else | |
7676 | kvm_mmu_slot_remove_write_access(kvm, new); | |
7677 | } else { | |
7678 | if (kvm_x86_ops->slot_disable_log_dirty) | |
7679 | kvm_x86_ops->slot_disable_log_dirty(kvm, new); | |
7680 | } | |
7681 | } | |
7682 | ||
f7784b8e | 7683 | void kvm_arch_commit_memory_region(struct kvm *kvm, |
09170a49 | 7684 | const struct kvm_userspace_memory_region *mem, |
8482644a | 7685 | const struct kvm_memory_slot *old, |
f36f3f28 | 7686 | const struct kvm_memory_slot *new, |
8482644a | 7687 | enum kvm_mr_change change) |
f7784b8e | 7688 | { |
8482644a | 7689 | int nr_mmu_pages = 0; |
f7784b8e | 7690 | |
f36f3f28 | 7691 | if (change == KVM_MR_DELETE && old->id >= KVM_USER_MEM_SLOTS) { |
f7784b8e MT |
7692 | int ret; |
7693 | ||
8482644a TY |
7694 | ret = vm_munmap(old->userspace_addr, |
7695 | old->npages * PAGE_SIZE); | |
f7784b8e MT |
7696 | if (ret < 0) |
7697 | printk(KERN_WARNING | |
7698 | "kvm_vm_ioctl_set_memory_region: " | |
7699 | "failed to munmap memory\n"); | |
7700 | } | |
7701 | ||
48c0e4e9 XG |
7702 | if (!kvm->arch.n_requested_mmu_pages) |
7703 | nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm); | |
7704 | ||
48c0e4e9 | 7705 | if (nr_mmu_pages) |
0de10343 | 7706 | kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages); |
1c91cad4 | 7707 | |
3ea3b7fa WL |
7708 | /* |
7709 | * Dirty logging tracks sptes in 4k granularity, meaning that large | |
7710 | * sptes have to be split. If live migration is successful, the guest | |
7711 | * in the source machine will be destroyed and large sptes will be | |
7712 | * created in the destination. However, if the guest continues to run | |
7713 | * in the source machine (for example if live migration fails), small | |
7714 | * sptes will remain around and cause bad performance. | |
7715 | * | |
7716 | * Scan sptes if dirty logging has been stopped, dropping those | |
7717 | * which can be collapsed into a single large-page spte. Later | |
7718 | * page faults will create the large-page sptes. | |
7719 | */ | |
7720 | if ((change != KVM_MR_DELETE) && | |
7721 | (old->flags & KVM_MEM_LOG_DIRTY_PAGES) && | |
7722 | !(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
7723 | kvm_mmu_zap_collapsible_sptes(kvm, new); | |
7724 | ||
c972f3b1 | 7725 | /* |
88178fd4 | 7726 | * Set up write protection and/or dirty logging for the new slot. |
c126d94f | 7727 | * |
88178fd4 KH |
7728 | * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have |
7729 | * been zapped so no dirty logging staff is needed for old slot. For | |
7730 | * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the | |
7731 | * new and it's also covered when dealing with the new slot. | |
f36f3f28 PB |
7732 | * |
7733 | * FIXME: const-ify all uses of struct kvm_memory_slot. | |
c972f3b1 | 7734 | */ |
88178fd4 | 7735 | if (change != KVM_MR_DELETE) |
f36f3f28 | 7736 | kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new); |
0de10343 | 7737 | } |
1d737c8a | 7738 | |
2df72e9b | 7739 | void kvm_arch_flush_shadow_all(struct kvm *kvm) |
34d4cb8f | 7740 | { |
6ca18b69 | 7741 | kvm_mmu_invalidate_zap_all_pages(kvm); |
34d4cb8f MT |
7742 | } |
7743 | ||
2df72e9b MT |
7744 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, |
7745 | struct kvm_memory_slot *slot) | |
7746 | { | |
6ca18b69 | 7747 | kvm_mmu_invalidate_zap_all_pages(kvm); |
2df72e9b MT |
7748 | } |
7749 | ||
1d737c8a ZX |
7750 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) |
7751 | { | |
b6b8a145 JK |
7752 | if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) |
7753 | kvm_x86_ops->check_nested_events(vcpu, false); | |
7754 | ||
af585b92 GN |
7755 | return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE && |
7756 | !vcpu->arch.apf.halted) | |
7757 | || !list_empty_careful(&vcpu->async_pf.done) | |
66450a21 | 7758 | || kvm_apic_has_events(vcpu) |
6aef266c | 7759 | || vcpu->arch.pv.pv_unhalted |
7460fb4a | 7760 | || atomic_read(&vcpu->arch.nmi_queued) || |
a1b37100 GN |
7761 | (kvm_arch_interrupt_allowed(vcpu) && |
7762 | kvm_cpu_has_interrupt(vcpu)); | |
1d737c8a | 7763 | } |
5736199a | 7764 | |
b6d33834 | 7765 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
5736199a | 7766 | { |
b6d33834 | 7767 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; |
5736199a | 7768 | } |
78646121 GN |
7769 | |
7770 | int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu) | |
7771 | { | |
7772 | return kvm_x86_ops->interrupt_allowed(vcpu); | |
7773 | } | |
229456fc | 7774 | |
82b32774 | 7775 | unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu) |
f92653ee | 7776 | { |
82b32774 NA |
7777 | if (is_64_bit_mode(vcpu)) |
7778 | return kvm_rip_read(vcpu); | |
7779 | return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) + | |
7780 | kvm_rip_read(vcpu)); | |
7781 | } | |
7782 | EXPORT_SYMBOL_GPL(kvm_get_linear_rip); | |
f92653ee | 7783 | |
82b32774 NA |
7784 | bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip) |
7785 | { | |
7786 | return kvm_get_linear_rip(vcpu) == linear_rip; | |
f92653ee JK |
7787 | } |
7788 | EXPORT_SYMBOL_GPL(kvm_is_linear_rip); | |
7789 | ||
94fe45da JK |
7790 | unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu) |
7791 | { | |
7792 | unsigned long rflags; | |
7793 | ||
7794 | rflags = kvm_x86_ops->get_rflags(vcpu); | |
7795 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) | |
c310bac5 | 7796 | rflags &= ~X86_EFLAGS_TF; |
94fe45da JK |
7797 | return rflags; |
7798 | } | |
7799 | EXPORT_SYMBOL_GPL(kvm_get_rflags); | |
7800 | ||
6addfc42 | 7801 | static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) |
94fe45da JK |
7802 | { |
7803 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP && | |
f92653ee | 7804 | kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip)) |
c310bac5 | 7805 | rflags |= X86_EFLAGS_TF; |
94fe45da | 7806 | kvm_x86_ops->set_rflags(vcpu, rflags); |
6addfc42 PB |
7807 | } |
7808 | ||
7809 | void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
7810 | { | |
7811 | __kvm_set_rflags(vcpu, rflags); | |
3842d135 | 7812 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
94fe45da JK |
7813 | } |
7814 | EXPORT_SYMBOL_GPL(kvm_set_rflags); | |
7815 | ||
56028d08 GN |
7816 | void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work) |
7817 | { | |
7818 | int r; | |
7819 | ||
fb67e14f | 7820 | if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) || |
f2e10669 | 7821 | work->wakeup_all) |
56028d08 GN |
7822 | return; |
7823 | ||
7824 | r = kvm_mmu_reload(vcpu); | |
7825 | if (unlikely(r)) | |
7826 | return; | |
7827 | ||
fb67e14f XG |
7828 | if (!vcpu->arch.mmu.direct_map && |
7829 | work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu)) | |
7830 | return; | |
7831 | ||
56028d08 GN |
7832 | vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true); |
7833 | } | |
7834 | ||
af585b92 GN |
7835 | static inline u32 kvm_async_pf_hash_fn(gfn_t gfn) |
7836 | { | |
7837 | return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU)); | |
7838 | } | |
7839 | ||
7840 | static inline u32 kvm_async_pf_next_probe(u32 key) | |
7841 | { | |
7842 | return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1); | |
7843 | } | |
7844 | ||
7845 | static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
7846 | { | |
7847 | u32 key = kvm_async_pf_hash_fn(gfn); | |
7848 | ||
7849 | while (vcpu->arch.apf.gfns[key] != ~0) | |
7850 | key = kvm_async_pf_next_probe(key); | |
7851 | ||
7852 | vcpu->arch.apf.gfns[key] = gfn; | |
7853 | } | |
7854 | ||
7855 | static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn) | |
7856 | { | |
7857 | int i; | |
7858 | u32 key = kvm_async_pf_hash_fn(gfn); | |
7859 | ||
7860 | for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) && | |
c7d28c24 XG |
7861 | (vcpu->arch.apf.gfns[key] != gfn && |
7862 | vcpu->arch.apf.gfns[key] != ~0); i++) | |
af585b92 GN |
7863 | key = kvm_async_pf_next_probe(key); |
7864 | ||
7865 | return key; | |
7866 | } | |
7867 | ||
7868 | bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
7869 | { | |
7870 | return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn; | |
7871 | } | |
7872 | ||
7873 | static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
7874 | { | |
7875 | u32 i, j, k; | |
7876 | ||
7877 | i = j = kvm_async_pf_gfn_slot(vcpu, gfn); | |
7878 | while (true) { | |
7879 | vcpu->arch.apf.gfns[i] = ~0; | |
7880 | do { | |
7881 | j = kvm_async_pf_next_probe(j); | |
7882 | if (vcpu->arch.apf.gfns[j] == ~0) | |
7883 | return; | |
7884 | k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]); | |
7885 | /* | |
7886 | * k lies cyclically in ]i,j] | |
7887 | * | i.k.j | | |
7888 | * |....j i.k.| or |.k..j i...| | |
7889 | */ | |
7890 | } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j)); | |
7891 | vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j]; | |
7892 | i = j; | |
7893 | } | |
7894 | } | |
7895 | ||
7c90705b GN |
7896 | static int apf_put_user(struct kvm_vcpu *vcpu, u32 val) |
7897 | { | |
7898 | ||
7899 | return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val, | |
7900 | sizeof(val)); | |
7901 | } | |
7902 | ||
af585b92 GN |
7903 | void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu, |
7904 | struct kvm_async_pf *work) | |
7905 | { | |
6389ee94 AK |
7906 | struct x86_exception fault; |
7907 | ||
7c90705b | 7908 | trace_kvm_async_pf_not_present(work->arch.token, work->gva); |
af585b92 | 7909 | kvm_add_async_pf_gfn(vcpu, work->arch.gfn); |
7c90705b GN |
7910 | |
7911 | if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) || | |
fc5f06fa GN |
7912 | (vcpu->arch.apf.send_user_only && |
7913 | kvm_x86_ops->get_cpl(vcpu) == 0)) | |
7c90705b GN |
7914 | kvm_make_request(KVM_REQ_APF_HALT, vcpu); |
7915 | else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) { | |
6389ee94 AK |
7916 | fault.vector = PF_VECTOR; |
7917 | fault.error_code_valid = true; | |
7918 | fault.error_code = 0; | |
7919 | fault.nested_page_fault = false; | |
7920 | fault.address = work->arch.token; | |
7921 | kvm_inject_page_fault(vcpu, &fault); | |
7c90705b | 7922 | } |
af585b92 GN |
7923 | } |
7924 | ||
7925 | void kvm_arch_async_page_present(struct kvm_vcpu *vcpu, | |
7926 | struct kvm_async_pf *work) | |
7927 | { | |
6389ee94 AK |
7928 | struct x86_exception fault; |
7929 | ||
7c90705b | 7930 | trace_kvm_async_pf_ready(work->arch.token, work->gva); |
f2e10669 | 7931 | if (work->wakeup_all) |
7c90705b GN |
7932 | work->arch.token = ~0; /* broadcast wakeup */ |
7933 | else | |
7934 | kvm_del_async_pf_gfn(vcpu, work->arch.gfn); | |
7935 | ||
7936 | if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) && | |
7937 | !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) { | |
6389ee94 AK |
7938 | fault.vector = PF_VECTOR; |
7939 | fault.error_code_valid = true; | |
7940 | fault.error_code = 0; | |
7941 | fault.nested_page_fault = false; | |
7942 | fault.address = work->arch.token; | |
7943 | kvm_inject_page_fault(vcpu, &fault); | |
7c90705b | 7944 | } |
e6d53e3b | 7945 | vcpu->arch.apf.halted = false; |
a4fa1635 | 7946 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
7c90705b GN |
7947 | } |
7948 | ||
7949 | bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu) | |
7950 | { | |
7951 | if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED)) | |
7952 | return true; | |
7953 | else | |
7954 | return !kvm_event_needs_reinjection(vcpu) && | |
7955 | kvm_x86_ops->interrupt_allowed(vcpu); | |
af585b92 GN |
7956 | } |
7957 | ||
5544eb9b PB |
7958 | void kvm_arch_start_assignment(struct kvm *kvm) |
7959 | { | |
7960 | atomic_inc(&kvm->arch.assigned_device_count); | |
7961 | } | |
7962 | EXPORT_SYMBOL_GPL(kvm_arch_start_assignment); | |
7963 | ||
7964 | void kvm_arch_end_assignment(struct kvm *kvm) | |
7965 | { | |
7966 | atomic_dec(&kvm->arch.assigned_device_count); | |
7967 | } | |
7968 | EXPORT_SYMBOL_GPL(kvm_arch_end_assignment); | |
7969 | ||
7970 | bool kvm_arch_has_assigned_device(struct kvm *kvm) | |
7971 | { | |
7972 | return atomic_read(&kvm->arch.assigned_device_count); | |
7973 | } | |
7974 | EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device); | |
7975 | ||
e0f0bbc5 AW |
7976 | void kvm_arch_register_noncoherent_dma(struct kvm *kvm) |
7977 | { | |
7978 | atomic_inc(&kvm->arch.noncoherent_dma_count); | |
7979 | } | |
7980 | EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma); | |
7981 | ||
7982 | void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm) | |
7983 | { | |
7984 | atomic_dec(&kvm->arch.noncoherent_dma_count); | |
7985 | } | |
7986 | EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma); | |
7987 | ||
7988 | bool kvm_arch_has_noncoherent_dma(struct kvm *kvm) | |
7989 | { | |
7990 | return atomic_read(&kvm->arch.noncoherent_dma_count); | |
7991 | } | |
7992 | EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma); | |
7993 | ||
229456fc MT |
7994 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit); |
7995 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq); | |
7996 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault); | |
7997 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr); | |
7998 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr); | |
0ac406de | 7999 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun); |
d8cabddf | 8000 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit); |
17897f36 | 8001 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject); |
236649de | 8002 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit); |
ec1ff790 | 8003 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga); |
532a46b9 | 8004 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit); |
2e554e8d | 8005 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts); |
489223ed | 8006 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset); |
7b46268d | 8007 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window); |
843e4330 | 8008 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full); |