projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: move code related to KVM_SET_BOOT_CPU_ID to x86
[deliverable/linux.git] / arch / x86 / kvm / x86.c
CommitLineData
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		 80static
81u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
50a37eb4 82#else
1260edbe 83static 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 89static void update_cr8_intercept(struct kvm_vcpu *vcpu);
7460fb4a 90static void process_nmi(struct kvm_vcpu *vcpu);
6addfc42 91static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
674eea0f 92
97896d04 93struct kvm_x86_ops *kvm_x86_ops;
5fdbf976 94EXPORT_SYMBOL_GPL(kvm_x86_ops);
97896d04 95
476bc001
RR		 96static bool ignore_msrs = 0;
97module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
ed85c068 98
9ed96e87
MT		 99unsigned int min_timer_period_us = 500;
100module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
101
630994b3
MT		 102static bool __read_mostly kvmclock_periodic_sync = true;
103module_param(kvmclock_periodic_sync, bool, S_IRUGO);
104
92a1f12d
JR		 105bool kvm_has_tsc_control;
106EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
107u32 kvm_max_guest_tsc_khz;
108EXPORT_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 */
111static u32 tsc_tolerance_ppm = 250;
112module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
113
d0659d94
MT		 114/* lapic timer advance (tscdeadline mode only) in nanoseconds */
115unsigned int lapic_timer_advance_ns = 0;
116module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
117
16a96021
MT		 118static bool backwards_tsc_observed = false;
119
18863bdd
AK		 120#define KVM_NR_SHARED_MSRS 16
121
122struct kvm_shared_msrs_global {
123 int nr;
2bf78fa7 124 u32 msrs[KVM_NR_SHARED_MSRS];
18863bdd
AK		 125};
126
127struct 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
136static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
013f6a5d 137static struct kvm_shared_msrs __percpu *shared_msrs;
18863bdd 138
417bc304 139struct 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		 176u64 __read_mostly host_xcr0;
177
b6785def 178static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
d6aa1000 179
af585b92
GN		 180static 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		 187static 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 205static 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
222void 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}
231EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
232
233static 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 241int 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}
261EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
262
13a34e06 263static 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		 272u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
273{
8a5a87d9 274 return vcpu->arch.apic_base;
6866b83e
CO		 275}
276EXPORT_SYMBOL_GPL(kvm_get_apic_base);
277
58cb628d
JK		 278int 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}
299EXPORT_SYMBOL_GPL(kvm_set_apic_base);
300
2605fc21 301asmlinkage __visible void kvm_spurious_fault(void)
e3ba45b8
GL		 302{
303 /* Fault while not rebooting. We want the trace. */
304 BUG();
305}
306EXPORT_SYMBOL_GPL(kvm_spurious_fault);
307
3fd28fce
ED		 308#define EXCPT_BENIGN 0
309#define EXCPT_CONTRIBUTORY 1
310#define EXCPT_PF 2
311
312static 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
334static 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 354static 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		 398void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
399{
ce7ddec4 400 kvm_multiple_exception(vcpu, nr, false, 0, false);
298101da
AK		 401}
402EXPORT_SYMBOL_GPL(kvm_queue_exception);
403
ce7ddec4
JR		 404void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
405{
406 kvm_multiple_exception(vcpu, nr, false, 0, true);
407}
408EXPORT_SYMBOL_GPL(kvm_requeue_exception);
409
db8fcefa 410void 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}
417EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
8df25a32 418
6389ee94 419void 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 425EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
c3c91fee 426
ef54bcfe 427static 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		 437void 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}
442EXPORT_SYMBOL_GPL(kvm_inject_nmi);
443
298101da
AK		 444void 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}
448EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
449
ce7ddec4
JR		 450void 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}
454EXPORT_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 */
460bool 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 467EXPORT_SYMBOL_GPL(kvm_require_cpl);
298101da 468
16f8a6f9
NA		 469bool 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}
477EXPORT_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 */
484int 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}
501EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
502
69b0049a 503static 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 513int 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 542out:
a03490ed
CO		 543
544 return ret;
545}
cc4b6871 546EXPORT_SYMBOL_GPL(load_pdptrs);
a03490ed 547
d835dfec
AK		 548static 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 570out:
d835dfec
AK		 571
572 return changed;
573}
574
49a9b07e 575int 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 630EXPORT_SYMBOL_GPL(kvm_set_cr0);
a03490ed 631
2d3ad1f4 632void 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 636EXPORT_SYMBOL_GPL(kvm_lmsw);
a03490ed 637
42bdf991
MT		 638static 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
648static 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 657static 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
697int 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}
706EXPORT_SYMBOL_GPL(kvm_set_xcr);
707
a83b29c6 708int 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 759EXPORT_SYMBOL_GPL(kvm_set_cr4);
a03490ed 760
2390218b 761int 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 785EXPORT_SYMBOL_GPL(kvm_set_cr3);
a03490ed 786
eea1cff9 787int 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 797EXPORT_SYMBOL_GPL(kvm_set_cr8);
a03490ed 798
2d3ad1f4 799unsigned 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 806EXPORT_SYMBOL_GPL(kvm_get_cr8);
a03490ed 807
ae561ede
NA		 808static 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		 819static 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		 825static 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		 839static 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 848static 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
877int 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		 885EXPORT_SYMBOL_GPL(kvm_set_dr);
886
16f8a6f9 887int 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		 909EXPORT_SYMBOL_GPL(kvm_get_dr);
910
022cd0e8
AK		 911bool 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}
924EXPORT_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		 936static 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
946static unsigned num_msrs_to_save;
947
62ef68bb
PB		 948static 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,
e7d9513b
AS		 953 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
954 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
62ef68bb
PB		 955 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
956 MSR_KVM_PV_EOI_EN,
957
ba904635 958 MSR_IA32_TSC_ADJUST,
a3e06bbe 959 MSR_IA32_TSCDEADLINE,
043405e1 960 MSR_IA32_MISC_ENABLE,
908e75f3
AK		 961 MSR_IA32_MCG_STATUS,
962 MSR_IA32_MCG_CTL,
64d60670 963 MSR_IA32_SMBASE,
043405e1
CO		 964};
965
62ef68bb
PB		 966static unsigned num_emulated_msrs;
967
384bb783 968bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
15c4a640 969{
b69e8cae 970 if (efer & efer_reserved_bits)
384bb783 971 return false;
15c4a640 972
1b2fd70c
AG		 973 if (efer & EFER_FFXSR) {
974 struct kvm_cpuid_entry2 *feat;
975
976 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
b69e8cae 977 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
384bb783 978 return false;
1b2fd70c
AG		 979 }
980
d8017474
AG		 981 if (efer & EFER_SVME) {
982 struct kvm_cpuid_entry2 *feat;
983
984 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
b69e8cae 985 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
384bb783 986 return false;
d8017474
AG		 987 }
988
384bb783
JK		 989 return true;
990}
991EXPORT_SYMBOL_GPL(kvm_valid_efer);
992
993static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
994{
995 u64 old_efer = vcpu->arch.efer;
996
997 if (!kvm_valid_efer(vcpu, efer))
998 return 1;
999
1000 if (is_paging(vcpu)
1001 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1002 return 1;
1003
15c4a640 1004 efer &= ~EFER_LMA;
f6801dff 1005 efer |= vcpu->arch.efer & EFER_LMA;
15c4a640 1006
a3d204e2
SY		 1007 kvm_x86_ops->set_efer(vcpu, efer);
1008
aad82703
SY		 1009 /* Update reserved bits */
1010 if ((efer ^ old_efer) & EFER_NX)
1011 kvm_mmu_reset_context(vcpu);
1012
b69e8cae 1013 return 0;
15c4a640
CO		 1014}
1015
f2b4b7dd
JR		 1016void kvm_enable_efer_bits(u64 mask)
1017{
1018 efer_reserved_bits &= ~mask;
1019}
1020EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1021
15c4a640
CO		 1022/*
1023 * Writes msr value into into the appropriate "register".
1024 * Returns 0 on success, non-0 otherwise.
1025 * Assumes vcpu_load() was already called.
1026 */
8fe8ab46 1027int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
15c4a640 1028{
854e8bb1
NA		 1029 switch (msr->index) {
1030 case MSR_FS_BASE:
1031 case MSR_GS_BASE:
1032 case MSR_KERNEL_GS_BASE:
1033 case MSR_CSTAR:
1034 case MSR_LSTAR:
1035 if (is_noncanonical_address(msr->data))
1036 return 1;
1037 break;
1038 case MSR_IA32_SYSENTER_EIP:
1039 case MSR_IA32_SYSENTER_ESP:
1040 /*
1041 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1042 * non-canonical address is written on Intel but not on
1043 * AMD (which ignores the top 32-bits, because it does
1044 * not implement 64-bit SYSENTER).
1045 *
1046 * 64-bit code should hence be able to write a non-canonical
1047 * value on AMD. Making the address canonical ensures that
1048 * vmentry does not fail on Intel after writing a non-canonical
1049 * value, and that something deterministic happens if the guest
1050 * invokes 64-bit SYSENTER.
1051 */
1052 msr->data = get_canonical(msr->data);
1053 }
8fe8ab46 1054 return kvm_x86_ops->set_msr(vcpu, msr);
15c4a640 1055}
854e8bb1 1056EXPORT_SYMBOL_GPL(kvm_set_msr);
15c4a640 1057
313a3dc7
CO		 1058/*
1059 * Adapt set_msr() to msr_io()'s calling convention
1060 */
609e36d3
PB		 1061static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1062{
1063 struct msr_data msr;
1064 int r;
1065
1066 msr.index = index;
1067 msr.host_initiated = true;
1068 r = kvm_get_msr(vcpu, &msr);
1069 if (r)
1070 return r;
1071
1072 *data = msr.data;
1073 return 0;
1074}
1075
313a3dc7
CO		 1076static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1077{
8fe8ab46
WA		 1078 struct msr_data msr;
1079
1080 msr.data = *data;
1081 msr.index = index;
1082 msr.host_initiated = true;
1083 return kvm_set_msr(vcpu, &msr);
313a3dc7
CO		 1084}
1085
16e8d74d
MT		 1086#ifdef CONFIG_X86_64
1087struct pvclock_gtod_data {
1088 seqcount_t seq;
1089
1090 struct { /* extract of a clocksource struct */
1091 int vclock_mode;
1092 cycle_t cycle_last;
1093 cycle_t mask;
1094 u32 mult;
1095 u32 shift;
1096 } clock;
1097
cbcf2dd3
TG		 1098 u64 boot_ns;
1099 u64 nsec_base;
16e8d74d
MT		 1100};
1101
1102static struct pvclock_gtod_data pvclock_gtod_data;
1103
1104static void update_pvclock_gtod(struct timekeeper *tk)
1105{
1106 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
cbcf2dd3
TG		 1107 u64 boot_ns;
1108
876e7881 1109 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
16e8d74d
MT		 1110
1111 write_seqcount_begin(&vdata->seq);
1112
1113 /* copy pvclock gtod data */
876e7881
PZ		 1114 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1115 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1116 vdata->clock.mask = tk->tkr_mono.mask;
1117 vdata->clock.mult = tk->tkr_mono.mult;
1118 vdata->clock.shift = tk->tkr_mono.shift;
16e8d74d 1119
cbcf2dd3 1120 vdata->boot_ns = boot_ns;
876e7881 1121 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
16e8d74d
MT		 1122
1123 write_seqcount_end(&vdata->seq);
1124}
1125#endif
1126
bab5bb39
NK		 1127void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1128{
1129 /*
1130 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1131 * vcpu_enter_guest. This function is only called from
1132 * the physical CPU that is running vcpu.
1133 */
1134 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1135}
16e8d74d 1136
18068523
GOC		 1137static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1138{
9ed3c444
AK		 1139 int version;
1140 int r;
50d0a0f9 1141 struct pvclock_wall_clock wc;
923de3cf 1142 struct timespec boot;
18068523
GOC		 1143
1144 if (!wall_clock)
1145 return;
1146
9ed3c444
AK		 1147 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1148 if (r)
1149 return;
1150
1151 if (version & 1)
1152 ++version; /* first time write, random junk */
1153
1154 ++version;
18068523 1155
18068523
GOC		 1156 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1157
50d0a0f9
GH		 1158 /*
1159 * The guest calculates current wall clock time by adding
34c238a1 1160 * system time (updated by kvm_guest_time_update below) to the
50d0a0f9
GH		 1161 * wall clock specified here. guest system time equals host
1162 * system time for us, thus we must fill in host boot time here.
1163 */
923de3cf 1164 getboottime(&boot);
50d0a0f9 1165
4b648665
BR		 1166 if (kvm->arch.kvmclock_offset) {
1167 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1168 boot = timespec_sub(boot, ts);
1169 }
50d0a0f9
GH		 1170 wc.sec = boot.tv_sec;
1171 wc.nsec = boot.tv_nsec;
1172 wc.version = version;
18068523
GOC		 1173
1174 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1175
1176 version++;
1177 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
18068523
GOC		 1178}
1179
50d0a0f9
GH		 1180static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1181{
1182 uint32_t quotient, remainder;
1183
1184 /* Don't try to replace with do_div(), this one calculates
1185 * "(dividend << 32) / divisor" */
1186 __asm__ ( "divl %4"
1187 : "=a" (quotient), "=d" (remainder)
1188 : "0" (0), "1" (dividend), "r" (divisor) );
1189 return quotient;
1190}
1191
5f4e3f88
ZA		 1192static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1193 s8 *pshift, u32 *pmultiplier)
50d0a0f9 1194{
5f4e3f88 1195 uint64_t scaled64;
50d0a0f9
GH		 1196 int32_t shift = 0;
1197 uint64_t tps64;
1198 uint32_t tps32;
1199
5f4e3f88
ZA		 1200 tps64 = base_khz * 1000LL;
1201 scaled64 = scaled_khz * 1000LL;
50933623 1202 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
50d0a0f9
GH		 1203 tps64 >>= 1;
1204 shift--;
1205 }
1206
1207 tps32 = (uint32_t)tps64;
50933623
JK		 1208 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1209 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
5f4e3f88
ZA		 1210 scaled64 >>= 1;
1211 else
1212 tps32 <<= 1;
50d0a0f9
GH		 1213 shift++;
1214 }
1215
5f4e3f88
ZA		 1216 *pshift = shift;
1217 *pmultiplier = div_frac(scaled64, tps32);
50d0a0f9 1218
5f4e3f88
ZA		 1219 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1220 __func__, base_khz, scaled_khz, shift, *pmultiplier);
50d0a0f9
GH		 1221}
1222
d828199e 1223#ifdef CONFIG_X86_64
16e8d74d 1224static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
d828199e 1225#endif
16e8d74d 1226
c8076604 1227static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
69b0049a 1228static unsigned long max_tsc_khz;
c8076604 1229
cc578287 1230static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
8cfdc000 1231{
cc578287
ZA		 1232 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1233 vcpu->arch.virtual_tsc_shift);
8cfdc000
ZA		 1234}
1235
cc578287 1236static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1e993611 1237{
cc578287
ZA		 1238 u64 v = (u64)khz * (1000000 + ppm);
1239 do_div(v, 1000000);
1240 return v;
1e993611
JR		 1241}
1242
cc578287 1243static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
759379dd 1244{
cc578287
ZA		 1245 u32 thresh_lo, thresh_hi;
1246 int use_scaling = 0;
217fc9cf 1247
03ba32ca
MT		 1248 /* tsc_khz can be zero if TSC calibration fails */
1249 if (this_tsc_khz == 0)
1250 return;
1251
c285545f
ZA		 1252 /* Compute a scale to convert nanoseconds in TSC cycles */
1253 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
cc578287
ZA		 1254 &vcpu->arch.virtual_tsc_shift,
1255 &vcpu->arch.virtual_tsc_mult);
1256 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1257
1258 /*
1259 * Compute the variation in TSC rate which is acceptable
1260 * within the range of tolerance and decide if the
1261 * rate being applied is within that bounds of the hardware
1262 * rate. If so, no scaling or compensation need be done.
1263 */
1264 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1265 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1266 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1267 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1268 use_scaling = 1;
1269 }
1270 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
c285545f
ZA		 1271}
1272
1273static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1274{
e26101b1 1275 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
cc578287
ZA		 1276 vcpu->arch.virtual_tsc_mult,
1277 vcpu->arch.virtual_tsc_shift);
e26101b1 1278 tsc += vcpu->arch.this_tsc_write;
c285545f
ZA		 1279 return tsc;
1280}
1281
69b0049a 1282static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
b48aa97e
MT		 1283{
1284#ifdef CONFIG_X86_64
1285 bool vcpus_matched;
b48aa97e
MT		 1286 struct kvm_arch *ka = &vcpu->kvm->arch;
1287 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1288
1289 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1290 atomic_read(&vcpu->kvm->online_vcpus));
1291
7f187922
MT		 1292 /*
1293 * Once the masterclock is enabled, always perform request in
1294 * order to update it.
1295 *
1296 * In order to enable masterclock, the host clocksource must be TSC
1297 * and the vcpus need to have matched TSCs. When that happens,
1298 * perform request to enable masterclock.
1299 */
1300 if (ka->use_master_clock ||
1301 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
b48aa97e
MT		 1302 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1303
1304 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1305 atomic_read(&vcpu->kvm->online_vcpus),
1306 ka->use_master_clock, gtod->clock.vclock_mode);
1307#endif
1308}
1309
ba904635
WA		 1310static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1311{
1312 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1313 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1314}
1315
8fe8ab46 1316void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
99e3e30a
ZA		 1317{
1318 struct kvm *kvm = vcpu->kvm;
f38e098f 1319 u64 offset, ns, elapsed;
99e3e30a 1320 unsigned long flags;
02626b6a 1321 s64 usdiff;
b48aa97e 1322 bool matched;
0d3da0d2 1323 bool already_matched;
8fe8ab46 1324 u64 data = msr->data;
99e3e30a 1325
038f8c11 1326 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
857e4099 1327 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
759379dd 1328 ns = get_kernel_ns();
f38e098f 1329 elapsed = ns - kvm->arch.last_tsc_nsec;
5d3cb0f6 1330
03ba32ca 1331 if (vcpu->arch.virtual_tsc_khz) {
8915aa27
MT		 1332 int faulted = 0;
1333
03ba32ca
MT		 1334 /* n.b - signed multiplication and division required */
1335 usdiff = data - kvm->arch.last_tsc_write;
5d3cb0f6 1336#ifdef CONFIG_X86_64
03ba32ca 1337 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
5d3cb0f6 1338#else
03ba32ca 1339 /* do_div() only does unsigned */
8915aa27
MT		 1340 asm("1: idivl %[divisor]\n"
1341 "2: xor %%edx, %%edx\n"
1342 " movl $0, %[faulted]\n"
1343 "3:\n"
1344 ".section .fixup,\"ax\"\n"
1345 "4: movl $1, %[faulted]\n"
1346 " jmp 3b\n"
1347 ".previous\n"
1348
1349 _ASM_EXTABLE(1b, 4b)
1350
1351 : "=A"(usdiff), [faulted] "=r" (faulted)
1352 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1353
5d3cb0f6 1354#endif
03ba32ca
MT		 1355 do_div(elapsed, 1000);
1356 usdiff -= elapsed;
1357 if (usdiff < 0)
1358 usdiff = -usdiff;
8915aa27
MT		 1359
1360 /* idivl overflow => difference is larger than USEC_PER_SEC */
1361 if (faulted)
1362 usdiff = USEC_PER_SEC;
03ba32ca
MT		 1363 } else
1364 usdiff = USEC_PER_SEC; /* disable TSC match window below */
f38e098f
ZA		 1365
1366 /*
5d3cb0f6
ZA		 1367 * Special case: TSC write with a small delta (1 second) of virtual
1368 * cycle time against real time is interpreted as an attempt to
1369 * synchronize the CPU.
1370 *
1371 * For a reliable TSC, we can match TSC offsets, and for an unstable
1372 * TSC, we add elapsed time in this computation. We could let the
1373 * compensation code attempt to catch up if we fall behind, but
1374 * it's better to try to match offsets from the beginning.
1375 */
02626b6a 1376 if (usdiff < USEC_PER_SEC &&
5d3cb0f6 1377 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
f38e098f 1378 if (!check_tsc_unstable()) {
e26101b1 1379 offset = kvm->arch.cur_tsc_offset;
f38e098f
ZA		 1380 pr_debug("kvm: matched tsc offset for %llu\n", data);
1381 } else {
857e4099 1382 u64 delta = nsec_to_cycles(vcpu, elapsed);
5d3cb0f6
ZA		 1383 data += delta;
1384 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
759379dd 1385 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
f38e098f 1386 }
b48aa97e 1387 matched = true;
0d3da0d2 1388 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
e26101b1
ZA		 1389 } else {
1390 /*
1391 * We split periods of matched TSC writes into generations.
1392 * For each generation, we track the original measured
1393 * nanosecond time, offset, and write, so if TSCs are in
1394 * sync, we can match exact offset, and if not, we can match
4a969980 1395 * exact software computation in compute_guest_tsc()
e26101b1
ZA		 1396 *
1397 * These values are tracked in kvm->arch.cur_xxx variables.
1398 */
1399 kvm->arch.cur_tsc_generation++;
1400 kvm->arch.cur_tsc_nsec = ns;
1401 kvm->arch.cur_tsc_write = data;
1402 kvm->arch.cur_tsc_offset = offset;
b48aa97e 1403 matched = false;
0d3da0d2 1404 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
e26101b1 1405 kvm->arch.cur_tsc_generation, data);
f38e098f 1406 }
e26101b1
ZA		 1407
1408 /*
1409 * We also track th most recent recorded KHZ, write and time to
1410 * allow the matching interval to be extended at each write.
1411 */
f38e098f
ZA		 1412 kvm->arch.last_tsc_nsec = ns;
1413 kvm->arch.last_tsc_write = data;
5d3cb0f6 1414 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
99e3e30a 1415
b183aa58 1416 vcpu->arch.last_guest_tsc = data;
e26101b1
ZA		 1417
1418 /* Keep track of which generation this VCPU has synchronized to */
1419 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1420 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1421 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1422
ba904635
WA		 1423 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1424 update_ia32_tsc_adjust_msr(vcpu, offset);
e26101b1
ZA		 1425 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1426 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
b48aa97e
MT		 1427
1428 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
0d3da0d2 1429 if (!matched) {
b48aa97e 1430 kvm->arch.nr_vcpus_matched_tsc = 0;
0d3da0d2
TG		 1431 } else if (!already_matched) {
1432 kvm->arch.nr_vcpus_matched_tsc++;
1433 }
b48aa97e
MT		 1434
1435 kvm_track_tsc_matching(vcpu);
1436 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
99e3e30a 1437}
e26101b1 1438
99e3e30a
ZA		 1439EXPORT_SYMBOL_GPL(kvm_write_tsc);
1440
d828199e
MT		 1441#ifdef CONFIG_X86_64
1442
1443static cycle_t read_tsc(void)
1444{
1445 cycle_t ret;
1446 u64 last;
1447
1448 /*
1449 * Empirically, a fence (of type that depends on the CPU)
1450 * before rdtsc is enough to ensure that rdtsc is ordered
1451 * with respect to loads. The various CPU manuals are unclear
1452 * as to whether rdtsc can be reordered with later loads,
1453 * but no one has ever seen it happen.
1454 */
1455 rdtsc_barrier();
1456 ret = (cycle_t)vget_cycles();
1457
1458 last = pvclock_gtod_data.clock.cycle_last;
1459
1460 if (likely(ret >= last))
1461 return ret;
1462
1463 /*
1464 * GCC likes to generate cmov here, but this branch is extremely
1465 * predictable (it's just a funciton of time and the likely is
1466 * very likely) and there's a data dependence, so force GCC
1467 * to generate a branch instead. I don't barrier() because
1468 * we don't actually need a barrier, and if this function
1469 * ever gets inlined it will generate worse code.
1470 */
1471 asm volatile ("");
1472 return last;
1473}
1474
1475static inline u64 vgettsc(cycle_t *cycle_now)
1476{
1477 long v;
1478 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1479
1480 *cycle_now = read_tsc();
1481
1482 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1483 return v * gtod->clock.mult;
1484}
1485
cbcf2dd3 1486static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
d828199e 1487{
cbcf2dd3 1488 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
d828199e 1489 unsigned long seq;
d828199e 1490 int mode;
cbcf2dd3 1491 u64 ns;
d828199e 1492
d828199e
MT		 1493 do {
1494 seq = read_seqcount_begin(&gtod->seq);
1495 mode = gtod->clock.vclock_mode;
cbcf2dd3 1496 ns = gtod->nsec_base;
d828199e
MT		 1497 ns += vgettsc(cycle_now);
1498 ns >>= gtod->clock.shift;
cbcf2dd3 1499 ns += gtod->boot_ns;
d828199e 1500 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
cbcf2dd3 1501 *t = ns;
d828199e
MT		 1502
1503 return mode;
1504}
1505
1506/* returns true if host is using tsc clocksource */
1507static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1508{
d828199e
MT		 1509 /* checked again under seqlock below */
1510 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1511 return false;
1512
cbcf2dd3 1513 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
d828199e
MT		 1514}
1515#endif
1516
1517/*
1518 *
b48aa97e
MT		 1519 * Assuming a stable TSC across physical CPUS, and a stable TSC
1520 * across virtual CPUs, the following condition is possible.
1521 * Each numbered line represents an event visible to both
d828199e
MT		 1522 * CPUs at the next numbered event.
1523 *
1524 * "timespecX" represents host monotonic time. "tscX" represents
1525 * RDTSC value.
1526 *
1527 * VCPU0 on CPU0 | VCPU1 on CPU1
1528 *
1529 * 1. read timespec0,tsc0
1530 * 2. | timespec1 = timespec0 + N
1531 * | tsc1 = tsc0 + M
1532 * 3. transition to guest | transition to guest
1533 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1534 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1535 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1536 *
1537 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1538 *
1539 * - ret0 < ret1
1540 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1541 * ...
1542 * - 0 < N - M => M < N
1543 *
1544 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1545 * always the case (the difference between two distinct xtime instances
1546 * might be smaller then the difference between corresponding TSC reads,
1547 * when updating guest vcpus pvclock areas).
1548 *
1549 * To avoid that problem, do not allow visibility of distinct
1550 * system_timestamp/tsc_timestamp values simultaneously: use a master
1551 * copy of host monotonic time values. Update that master copy
1552 * in lockstep.
1553 *
b48aa97e 1554 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
d828199e
MT		 1555 *
1556 */
1557
1558static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1559{
1560#ifdef CONFIG_X86_64
1561 struct kvm_arch *ka = &kvm->arch;
1562 int vclock_mode;
b48aa97e
MT		 1563 bool host_tsc_clocksource, vcpus_matched;
1564
1565 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1566 atomic_read(&kvm->online_vcpus));
d828199e
MT		 1567
1568 /*
1569 * If the host uses TSC clock, then passthrough TSC as stable
1570 * to the guest.
1571 */
b48aa97e 1572 host_tsc_clocksource = kvm_get_time_and_clockread(
d828199e
MT		 1573 &ka->master_kernel_ns,
1574 &ka->master_cycle_now);
1575
16a96021 1576 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
54750f2c
MT		 1577 && !backwards_tsc_observed
1578 && !ka->boot_vcpu_runs_old_kvmclock;
b48aa97e 1579
d828199e
MT		 1580 if (ka->use_master_clock)
1581 atomic_set(&kvm_guest_has_master_clock, 1);
1582
1583 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
b48aa97e
MT		 1584 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1585 vcpus_matched);
d828199e
MT		 1586#endif
1587}
1588
2e762ff7
MT		 1589static void kvm_gen_update_masterclock(struct kvm *kvm)
1590{
1591#ifdef CONFIG_X86_64
1592 int i;
1593 struct kvm_vcpu *vcpu;
1594 struct kvm_arch *ka = &kvm->arch;
1595
1596 spin_lock(&ka->pvclock_gtod_sync_lock);
1597 kvm_make_mclock_inprogress_request(kvm);
1598 /* no guest entries from this point */
1599 pvclock_update_vm_gtod_copy(kvm);
1600
1601 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 1602 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2e762ff7
MT		 1603
1604 /* guest entries allowed */
1605 kvm_for_each_vcpu(i, vcpu, kvm)
1606 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1607
1608 spin_unlock(&ka->pvclock_gtod_sync_lock);
1609#endif
1610}
1611
34c238a1 1612static int kvm_guest_time_update(struct kvm_vcpu *v)
18068523 1613{
d828199e 1614 unsigned long flags, this_tsc_khz;
18068523 1615 struct kvm_vcpu_arch *vcpu = &v->arch;
d828199e 1616 struct kvm_arch *ka = &v->kvm->arch;
f25e656d 1617 s64 kernel_ns;
d828199e 1618 u64 tsc_timestamp, host_tsc;
0b79459b 1619 struct pvclock_vcpu_time_info guest_hv_clock;
51d59c6b 1620 u8 pvclock_flags;
d828199e
MT		 1621 bool use_master_clock;
1622
1623 kernel_ns = 0;
1624 host_tsc = 0;
18068523 1625
d828199e
MT		 1626 /*
1627 * If the host uses TSC clock, then passthrough TSC as stable
1628 * to the guest.
1629 */
1630 spin_lock(&ka->pvclock_gtod_sync_lock);
1631 use_master_clock = ka->use_master_clock;
1632 if (use_master_clock) {
1633 host_tsc = ka->master_cycle_now;
1634 kernel_ns = ka->master_kernel_ns;
1635 }
1636 spin_unlock(&ka->pvclock_gtod_sync_lock);
c09664bb
MT		 1637
1638 /* Keep irq disabled to prevent changes to the clock */
1639 local_irq_save(flags);
89cbc767 1640 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
c09664bb
MT		 1641 if (unlikely(this_tsc_khz == 0)) {
1642 local_irq_restore(flags);
1643 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1644 return 1;
1645 }
d828199e
MT		 1646 if (!use_master_clock) {
1647 host_tsc = native_read_tsc();
1648 kernel_ns = get_kernel_ns();
1649 }
1650
1651 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1652
c285545f
ZA		 1653 /*
1654 * We may have to catch up the TSC to match elapsed wall clock
1655 * time for two reasons, even if kvmclock is used.
1656 * 1) CPU could have been running below the maximum TSC rate
1657 * 2) Broken TSC compensation resets the base at each VCPU
1658 * entry to avoid unknown leaps of TSC even when running
1659 * again on the same CPU. This may cause apparent elapsed
1660 * time to disappear, and the guest to stand still or run
1661 * very slowly.
1662 */
1663 if (vcpu->tsc_catchup) {
1664 u64 tsc = compute_guest_tsc(v, kernel_ns);
1665 if (tsc > tsc_timestamp) {
f1e2b260 1666 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
c285545f
ZA		 1667 tsc_timestamp = tsc;
1668 }
50d0a0f9
GH		 1669 }
1670
18068523
GOC		 1671 local_irq_restore(flags);
1672
0b79459b 1673 if (!vcpu->pv_time_enabled)
c285545f 1674 return 0;
18068523 1675
e48672fa 1676 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
5f4e3f88
ZA		 1677 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1678 &vcpu->hv_clock.tsc_shift,
1679 &vcpu->hv_clock.tsc_to_system_mul);
e48672fa 1680 vcpu->hw_tsc_khz = this_tsc_khz;
8cfdc000
ZA		 1681 }
1682
1683 /* With all the info we got, fill in the values */
1d5f066e 1684 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
759379dd 1685 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
28e4639a 1686 vcpu->last_guest_tsc = tsc_timestamp;
51d59c6b 1687
09a0c3f1
OH		 1688 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1689 &guest_hv_clock, sizeof(guest_hv_clock))))
1690 return 0;
1691
5dca0d91
RK		 1692 /* This VCPU is paused, but it's legal for a guest to read another
1693 * VCPU's kvmclock, so we really have to follow the specification where
1694 * it says that version is odd if data is being modified, and even after
1695 * it is consistent.
1696 *
1697 * Version field updates must be kept separate. This is because
1698 * kvm_write_guest_cached might use a "rep movs" instruction, and
1699 * writes within a string instruction are weakly ordered. So there
1700 * are three writes overall.
1701 *
1702 * As a small optimization, only write the version field in the first
1703 * and third write. The vcpu->pv_time cache is still valid, because the
1704 * version field is the first in the struct.
18068523 1705 */
5dca0d91
RK		 1706 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1707
1708 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1709 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1710 &vcpu->hv_clock,
1711 sizeof(vcpu->hv_clock.version));
1712
1713 smp_wmb();
78c0337a
MT		 1714
1715 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
0b79459b 1716 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
78c0337a
MT		 1717
1718 if (vcpu->pvclock_set_guest_stopped_request) {
1719 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1720 vcpu->pvclock_set_guest_stopped_request = false;
1721 }
1722
b7e60c5a
MT		 1723 pvclock_flags |= PVCLOCK_COUNTS_FROM_ZERO;
1724
d828199e
MT		 1725 /* If the host uses TSC clocksource, then it is stable */
1726 if (use_master_clock)
1727 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1728
78c0337a
MT		 1729 vcpu->hv_clock.flags = pvclock_flags;
1730
ce1a5e60
DM		 1731 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1732
0b79459b
AH		 1733 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1734 &vcpu->hv_clock,
1735 sizeof(vcpu->hv_clock));
5dca0d91
RK		 1736
1737 smp_wmb();
1738
1739 vcpu->hv_clock.version++;
1740 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1741 &vcpu->hv_clock,
1742 sizeof(vcpu->hv_clock.version));
8cfdc000 1743 return 0;
c8076604
GH		 1744}
1745
0061d53d
MT		 1746/*
1747 * kvmclock updates which are isolated to a given vcpu, such as
1748 * vcpu->cpu migration, should not allow system_timestamp from
1749 * the rest of the vcpus to remain static. Otherwise ntp frequency
1750 * correction applies to one vcpu's system_timestamp but not
1751 * the others.
1752 *
1753 * So in those cases, request a kvmclock update for all vcpus.
7e44e449
AJ		 1754 * We need to rate-limit these requests though, as they can
1755 * considerably slow guests that have a large number of vcpus.
1756 * The time for a remote vcpu to update its kvmclock is bound
1757 * by the delay we use to rate-limit the updates.
0061d53d
MT		 1758 */
1759
7e44e449
AJ		 1760#define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1761
1762static void kvmclock_update_fn(struct work_struct *work)
0061d53d
MT		 1763{
1764 int i;
7e44e449
AJ		 1765 struct delayed_work *dwork = to_delayed_work(work);
1766 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1767 kvmclock_update_work);
1768 struct kvm *kvm = container_of(ka, struct kvm, arch);
0061d53d
MT		 1769 struct kvm_vcpu *vcpu;
1770
1771 kvm_for_each_vcpu(i, vcpu, kvm) {
105b21bb 1772 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0061d53d
MT		 1773 kvm_vcpu_kick(vcpu);
1774 }
1775}
1776
7e44e449
AJ		 1777static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1778{
1779 struct kvm *kvm = v->kvm;
1780
105b21bb 1781 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
7e44e449
AJ		 1782 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1783 KVMCLOCK_UPDATE_DELAY);
1784}
1785
332967a3
AJ		 1786#define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1787
1788static void kvmclock_sync_fn(struct work_struct *work)
1789{
1790 struct delayed_work *dwork = to_delayed_work(work);
1791 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1792 kvmclock_sync_work);
1793 struct kvm *kvm = container_of(ka, struct kvm, arch);
1794
630994b3
MT		 1795 if (!kvmclock_periodic_sync)
1796 return;
1797
332967a3
AJ		 1798 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1799 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1800 KVMCLOCK_SYNC_PERIOD);
1801}
1802
890ca9ae 1803static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
15c4a640 1804{
890ca9ae
HY		 1805 u64 mcg_cap = vcpu->arch.mcg_cap;
1806 unsigned bank_num = mcg_cap & 0xff;
1807
15c4a640 1808 switch (msr) {
15c4a640 1809 case MSR_IA32_MCG_STATUS:
890ca9ae 1810 vcpu->arch.mcg_status = data;
15c4a640 1811 break;
c7ac679c 1812 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 1813 if (!(mcg_cap & MCG_CTL_P))
1814 return 1;
1815 if (data != 0 && data != ~(u64)0)
1816 return -1;
1817 vcpu->arch.mcg_ctl = data;
1818 break;
1819 default:
1820 if (msr >= MSR_IA32_MC0_CTL &&
81760dcc 1821 msr < MSR_IA32_MCx_CTL(bank_num)) {
890ca9ae 1822 u32 offset = msr - MSR_IA32_MC0_CTL;
114be429
AP		 1823 /* only 0 or all 1s can be written to IA32_MCi_CTL
1824 * some Linux kernels though clear bit 10 in bank 4 to
1825 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1826 * this to avoid an uncatched #GP in the guest
1827 */
890ca9ae 1828 if ((offset & 0x3) == 0 &&
114be429 1829 data != 0 && (data | (1 << 10)) != ~(u64)0)
890ca9ae
HY		 1830 return -1;
1831 vcpu->arch.mce_banks[offset] = data;
1832 break;
1833 }
1834 return 1;
1835 }
1836 return 0;
1837}
1838
ffde22ac
ES		 1839static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1840{
1841 struct kvm *kvm = vcpu->kvm;
1842 int lm = is_long_mode(vcpu);
1843 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1844 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1845 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1846 : kvm->arch.xen_hvm_config.blob_size_32;
1847 u32 page_num = data & ~PAGE_MASK;
1848 u64 page_addr = data & PAGE_MASK;
1849 u8 *page;
1850 int r;
1851
1852 r = -E2BIG;
1853 if (page_num >= blob_size)
1854 goto out;
1855 r = -ENOMEM;
ff5c2c03
SL		 1856 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1857 if (IS_ERR(page)) {
1858 r = PTR_ERR(page);
ffde22ac 1859 goto out;
ff5c2c03 1860 }
54bf36aa 1861 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
ffde22ac
ES		 1862 goto out_free;
1863 r = 0;
1864out_free:
1865 kfree(page);
1866out:
1867 return r;
1868}
1869
344d9588
GN		 1870static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1871{
1872 gpa_t gpa = data & ~0x3f;
1873
4a969980 1874 /* Bits 2:5 are reserved, Should be zero */
6adba527 1875 if (data & 0x3c)
344d9588
GN		 1876 return 1;
1877
1878 vcpu->arch.apf.msr_val = data;
1879
1880 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1881 kvm_clear_async_pf_completion_queue(vcpu);
1882 kvm_async_pf_hash_reset(vcpu);
1883 return 0;
1884 }
1885
8f964525
AH		 1886 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1887 sizeof(u32)))
344d9588
GN		 1888 return 1;
1889
6adba527 1890 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
344d9588
GN		 1891 kvm_async_pf_wakeup_all(vcpu);
1892 return 0;
1893}
1894
12f9a48f
GC		 1895static void kvmclock_reset(struct kvm_vcpu *vcpu)
1896{
0b79459b 1897 vcpu->arch.pv_time_enabled = false;
12f9a48f
GC		 1898}
1899
c9aaa895
GC		 1900static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1901{
1902 u64 delta;
1903
1904 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1905 return;
1906
1907 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1908 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1909 vcpu->arch.st.accum_steal = delta;
1910}
1911
1912static void record_steal_time(struct kvm_vcpu *vcpu)
1913{
1914 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1915 return;
1916
1917 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1918 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1919 return;
1920
1921 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1922 vcpu->arch.st.steal.version += 2;
1923 vcpu->arch.st.accum_steal = 0;
1924
1925 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1926 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1927}
1928
8fe8ab46 1929int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
15c4a640 1930{
5753785f 1931 bool pr = false;
8fe8ab46
WA		 1932 u32 msr = msr_info->index;
1933 u64 data = msr_info->data;
5753785f 1934
15c4a640 1935 switch (msr) {
2e32b719
BP		 1936 case MSR_AMD64_NB_CFG:
1937 case MSR_IA32_UCODE_REV:
1938 case MSR_IA32_UCODE_WRITE:
1939 case MSR_VM_HSAVE_PA:
1940 case MSR_AMD64_PATCH_LOADER:
1941 case MSR_AMD64_BU_CFG2:
1942 break;
1943
15c4a640 1944 case MSR_EFER:
b69e8cae 1945 return set_efer(vcpu, data);
8f1589d9
AP		 1946 case MSR_K7_HWCR:
1947 data &= ~(u64)0x40; /* ignore flush filter disable */
82494028 1948 data &= ~(u64)0x100; /* ignore ignne emulation enable */
a223c313 1949 data &= ~(u64)0x8; /* ignore TLB cache disable */
22d48b2d 1950 data &= ~(u64)0x40000; /* ignore Mc status write enable */
8f1589d9 1951 if (data != 0) {
a737f256
CD		 1952 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1953 data);
8f1589d9
AP		 1954 return 1;
1955 }
15c4a640 1956 break;
f7c6d140
AP		 1957 case MSR_FAM10H_MMIO_CONF_BASE:
1958 if (data != 0) {
a737f256
CD		 1959 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1960 "0x%llx\n", data);
f7c6d140
AP		 1961 return 1;
1962 }
15c4a640 1963 break;
b5e2fec0
AG		 1964 case MSR_IA32_DEBUGCTLMSR:
1965 if (!data) {
1966 /* We support the non-activated case already */
1967 break;
1968 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1969 /* Values other than LBR and BTF are vendor-specific,
1970 thus reserved and should throw a #GP */
1971 return 1;
1972 }
a737f256
CD		 1973 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1974 __func__, data);
b5e2fec0 1975 break;
9ba075a6 1976 case 0x200 ... 0x2ff:
ff53604b 1977 return kvm_mtrr_set_msr(vcpu, msr, data);
15c4a640 1978 case MSR_IA32_APICBASE:
58cb628d 1979 return kvm_set_apic_base(vcpu, msr_info);
0105d1a5
GN		 1980 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1981 return kvm_x2apic_msr_write(vcpu, msr, data);
a3e06bbe
LJ		 1982 case MSR_IA32_TSCDEADLINE:
1983 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1984 break;
ba904635
WA		 1985 case MSR_IA32_TSC_ADJUST:
1986 if (guest_cpuid_has_tsc_adjust(vcpu)) {
1987 if (!msr_info->host_initiated) {
d913b904 1988 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
ba904635
WA		 1989 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
1990 }
1991 vcpu->arch.ia32_tsc_adjust_msr = data;
1992 }
1993 break;
15c4a640 1994 case MSR_IA32_MISC_ENABLE:
ad312c7c 1995 vcpu->arch.ia32_misc_enable_msr = data;
15c4a640 1996 break;
64d60670
PB		 1997 case MSR_IA32_SMBASE:
1998 if (!msr_info->host_initiated)
1999 return 1;
2000 vcpu->arch.smbase = data;
2001 break;
11c6bffa 2002 case MSR_KVM_WALL_CLOCK_NEW:
18068523
GOC		 2003 case MSR_KVM_WALL_CLOCK:
2004 vcpu->kvm->arch.wall_clock = data;
2005 kvm_write_wall_clock(vcpu->kvm, data);
2006 break;
11c6bffa 2007 case MSR_KVM_SYSTEM_TIME_NEW:
18068523 2008 case MSR_KVM_SYSTEM_TIME: {
0b79459b 2009 u64 gpa_offset;
54750f2c
MT		 2010 struct kvm_arch *ka = &vcpu->kvm->arch;
2011
12f9a48f 2012 kvmclock_reset(vcpu);
18068523 2013
54750f2c
MT		 2014 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2015 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2016
2017 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2018 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2019 &vcpu->requests);
2020
2021 ka->boot_vcpu_runs_old_kvmclock = tmp;
b7e60c5a
MT		 2022
2023 ka->kvmclock_offset = -get_kernel_ns();
54750f2c
MT		 2024 }
2025
18068523 2026 vcpu->arch.time = data;
0061d53d 2027 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
18068523
GOC		 2028
2029 /* we verify if the enable bit is set... */
2030 if (!(data & 1))
2031 break;
2032
0b79459b 2033 gpa_offset = data & ~(PAGE_MASK | 1);
18068523 2034
0b79459b 2035 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
8f964525
AH		 2036 &vcpu->arch.pv_time, data & ~1ULL,
2037 sizeof(struct pvclock_vcpu_time_info)))
0b79459b
AH		 2038 vcpu->arch.pv_time_enabled = false;
2039 else
2040 vcpu->arch.pv_time_enabled = true;
32cad84f 2041
18068523
GOC		 2042 break;
2043 }
344d9588
GN		 2044 case MSR_KVM_ASYNC_PF_EN:
2045 if (kvm_pv_enable_async_pf(vcpu, data))
2046 return 1;
2047 break;
c9aaa895
GC		 2048 case MSR_KVM_STEAL_TIME:
2049
2050 if (unlikely(!sched_info_on()))
2051 return 1;
2052
2053 if (data & KVM_STEAL_RESERVED_MASK)
2054 return 1;
2055
2056 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
8f964525
AH		 2057 data & KVM_STEAL_VALID_BITS,
2058 sizeof(struct kvm_steal_time)))
c9aaa895
GC		 2059 return 1;
2060
2061 vcpu->arch.st.msr_val = data;
2062
2063 if (!(data & KVM_MSR_ENABLED))
2064 break;
2065
2066 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2067
2068 preempt_disable();
2069 accumulate_steal_time(vcpu);
2070 preempt_enable();
2071
2072 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2073
2074 break;
ae7a2a3f
MT		 2075 case MSR_KVM_PV_EOI_EN:
2076 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2077 return 1;
2078 break;
c9aaa895 2079
890ca9ae
HY		 2080 case MSR_IA32_MCG_CTL:
2081 case MSR_IA32_MCG_STATUS:
81760dcc 2082 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
890ca9ae 2083 return set_msr_mce(vcpu, msr, data);
71db6023 2084
6912ac32
WH		 2085 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2086 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2087 pr = true; /* fall through */
2088 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2089 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
c6702c9d 2090 if (kvm_pmu_is_valid_msr(vcpu, msr))
afd80d85 2091 return kvm_pmu_set_msr(vcpu, msr_info);
5753785f
GN		 2092
2093 if (pr || data != 0)
a737f256
CD		 2094 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2095 "0x%x data 0x%llx\n", msr, data);
5753785f 2096 break;
84e0cefa
JS		 2097 case MSR_K7_CLK_CTL:
2098 /*
2099 * Ignore all writes to this no longer documented MSR.
2100 * Writes are only relevant for old K7 processors,
2101 * all pre-dating SVM, but a recommended workaround from
4a969980 2102 * AMD for these chips. It is possible to specify the
84e0cefa
JS		 2103 * affected processor models on the command line, hence
2104 * the need to ignore the workaround.
2105 */
2106 break;
55cd8e5a 2107 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
e7d9513b
AS		 2108 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2109 case HV_X64_MSR_CRASH_CTL:
2110 return kvm_hv_set_msr_common(vcpu, msr, data,
2111 msr_info->host_initiated);
91c9c3ed 2112 case MSR_IA32_BBL_CR_CTL3:
2113 /* Drop writes to this legacy MSR -- see rdmsr
2114 * counterpart for further detail.
2115 */
a737f256 2116 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
91c9c3ed 2117 break;
2b036c6b
BO		 2118 case MSR_AMD64_OSVW_ID_LENGTH:
2119 if (!guest_cpuid_has_osvw(vcpu))
2120 return 1;
2121 vcpu->arch.osvw.length = data;
2122 break;
2123 case MSR_AMD64_OSVW_STATUS:
2124 if (!guest_cpuid_has_osvw(vcpu))
2125 return 1;
2126 vcpu->arch.osvw.status = data;
2127 break;
15c4a640 2128 default:
ffde22ac
ES		 2129 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2130 return xen_hvm_config(vcpu, data);
c6702c9d 2131 if (kvm_pmu_is_valid_msr(vcpu, msr))
afd80d85 2132 return kvm_pmu_set_msr(vcpu, msr_info);
ed85c068 2133 if (!ignore_msrs) {
a737f256
CD		 2134 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2135 msr, data);
ed85c068
AP		 2136 return 1;
2137 } else {
a737f256
CD		 2138 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2139 msr, data);
ed85c068
AP		 2140 break;
2141 }
15c4a640
CO		 2142 }
2143 return 0;
2144}
2145EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2146
2147
2148/*
2149 * Reads an msr value (of 'msr_index') into 'pdata'.
2150 * Returns 0 on success, non-0 otherwise.
2151 * Assumes vcpu_load() was already called.
2152 */
609e36d3 2153int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
15c4a640 2154{
609e36d3 2155 return kvm_x86_ops->get_msr(vcpu, msr);
15c4a640 2156}
ff651cb6 2157EXPORT_SYMBOL_GPL(kvm_get_msr);
15c4a640 2158
890ca9ae 2159static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
15c4a640
CO		 2160{
2161 u64 data;
890ca9ae
HY		 2162 u64 mcg_cap = vcpu->arch.mcg_cap;
2163 unsigned bank_num = mcg_cap & 0xff;
15c4a640
CO		 2164
2165 switch (msr) {
15c4a640
CO		 2166 case MSR_IA32_P5_MC_ADDR:
2167 case MSR_IA32_P5_MC_TYPE:
890ca9ae
HY		 2168 data = 0;
2169 break;
15c4a640 2170 case MSR_IA32_MCG_CAP:
890ca9ae
HY		 2171 data = vcpu->arch.mcg_cap;
2172 break;
c7ac679c 2173 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 2174 if (!(mcg_cap & MCG_CTL_P))
2175 return 1;
2176 data = vcpu->arch.mcg_ctl;
2177 break;
2178 case MSR_IA32_MCG_STATUS:
2179 data = vcpu->arch.mcg_status;
2180 break;
2181 default:
2182 if (msr >= MSR_IA32_MC0_CTL &&
81760dcc 2183 msr < MSR_IA32_MCx_CTL(bank_num)) {
890ca9ae
HY		 2184 u32 offset = msr - MSR_IA32_MC0_CTL;
2185 data = vcpu->arch.mce_banks[offset];
2186 break;
2187 }
2188 return 1;
2189 }
2190 *pdata = data;
2191 return 0;
2192}
2193
609e36d3 2194int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
890ca9ae 2195{
609e36d3 2196 switch (msr_info->index) {
890ca9ae 2197 case MSR_IA32_PLATFORM_ID:
15c4a640 2198 case MSR_IA32_EBL_CR_POWERON:
b5e2fec0
AG		 2199 case MSR_IA32_DEBUGCTLMSR:
2200 case MSR_IA32_LASTBRANCHFROMIP:
2201 case MSR_IA32_LASTBRANCHTOIP:
2202 case MSR_IA32_LASTINTFROMIP:
2203 case MSR_IA32_LASTINTTOIP:
60af2ecd
JSR		 2204 case MSR_K8_SYSCFG:
2205 case MSR_K7_HWCR:
61a6bd67 2206 case MSR_VM_HSAVE_PA:
1fdbd48c 2207 case MSR_K8_INT_PENDING_MSG:
c323c0e5 2208 case MSR_AMD64_NB_CFG:
f7c6d140 2209 case MSR_FAM10H_MMIO_CONF_BASE:
2e32b719 2210 case MSR_AMD64_BU_CFG2:
609e36d3 2211 msr_info->data = 0;
15c4a640 2212 break;
6912ac32
WH		 2213 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2214 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2215 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2216 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
c6702c9d 2217 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
609e36d3
PB		 2218 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2219 msr_info->data = 0;
5753785f 2220 break;
742bc670 2221 case MSR_IA32_UCODE_REV:
609e36d3 2222 msr_info->data = 0x100000000ULL;
742bc670 2223 break;
9ba075a6 2224 case MSR_MTRRcap:
9ba075a6 2225 case 0x200 ... 0x2ff:
ff53604b 2226 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
15c4a640 2227 case 0xcd: /* fsb frequency */
609e36d3 2228 msr_info->data = 3;
15c4a640 2229 break;
7b914098
JS		 2230 /*
2231 * MSR_EBC_FREQUENCY_ID
2232 * Conservative value valid for even the basic CPU models.
2233 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2234 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2235 * and 266MHz for model 3, or 4. Set Core Clock
2236 * Frequency to System Bus Frequency Ratio to 1 (bits
2237 * 31:24) even though these are only valid for CPU
2238 * models > 2, however guests may end up dividing or
2239 * multiplying by zero otherwise.
2240 */
2241 case MSR_EBC_FREQUENCY_ID:
609e36d3 2242 msr_info->data = 1 << 24;
7b914098 2243 break;
15c4a640 2244 case MSR_IA32_APICBASE:
609e36d3 2245 msr_info->data = kvm_get_apic_base(vcpu);
15c4a640 2246 break;
0105d1a5 2247 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
609e36d3 2248 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
0105d1a5 2249 break;
a3e06bbe 2250 case MSR_IA32_TSCDEADLINE:
609e36d3 2251 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
a3e06bbe 2252 break;
ba904635 2253 case MSR_IA32_TSC_ADJUST:
609e36d3 2254 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
ba904635 2255 break;
15c4a640 2256 case MSR_IA32_MISC_ENABLE:
609e36d3 2257 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
15c4a640 2258 break;
64d60670
PB		 2259 case MSR_IA32_SMBASE:
2260 if (!msr_info->host_initiated)
2261 return 1;
2262 msr_info->data = vcpu->arch.smbase;
15c4a640 2263 break;
847f0ad8
AG		 2264 case MSR_IA32_PERF_STATUS:
2265 /* TSC increment by tick */
609e36d3 2266 msr_info->data = 1000ULL;
847f0ad8 2267 /* CPU multiplier */
b0996ae4 2268 msr_info->data |= (((uint64_t)4ULL) << 40);
847f0ad8 2269 break;
15c4a640 2270 case MSR_EFER:
609e36d3 2271 msr_info->data = vcpu->arch.efer;
15c4a640 2272 break;
18068523 2273 case MSR_KVM_WALL_CLOCK:
11c6bffa 2274 case MSR_KVM_WALL_CLOCK_NEW:
609e36d3 2275 msr_info->data = vcpu->kvm->arch.wall_clock;
18068523
GOC		 2276 break;
2277 case MSR_KVM_SYSTEM_TIME:
11c6bffa 2278 case MSR_KVM_SYSTEM_TIME_NEW:
609e36d3 2279 msr_info->data = vcpu->arch.time;
18068523 2280 break;
344d9588 2281 case MSR_KVM_ASYNC_PF_EN:
609e36d3 2282 msr_info->data = vcpu->arch.apf.msr_val;
344d9588 2283 break;
c9aaa895 2284 case MSR_KVM_STEAL_TIME:
609e36d3 2285 msr_info->data = vcpu->arch.st.msr_val;
c9aaa895 2286 break;
1d92128f 2287 case MSR_KVM_PV_EOI_EN:
609e36d3 2288 msr_info->data = vcpu->arch.pv_eoi.msr_val;
1d92128f 2289 break;
890ca9ae
HY		 2290 case MSR_IA32_P5_MC_ADDR:
2291 case MSR_IA32_P5_MC_TYPE:
2292 case MSR_IA32_MCG_CAP:
2293 case MSR_IA32_MCG_CTL:
2294 case MSR_IA32_MCG_STATUS:
81760dcc 2295 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
609e36d3 2296 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
84e0cefa
JS		 2297 case MSR_K7_CLK_CTL:
2298 /*
2299 * Provide expected ramp-up count for K7. All other
2300 * are set to zero, indicating minimum divisors for
2301 * every field.
2302 *
2303 * This prevents guest kernels on AMD host with CPU
2304 * type 6, model 8 and higher from exploding due to
2305 * the rdmsr failing.
2306 */
609e36d3 2307 msr_info->data = 0x20000000;
84e0cefa 2308 break;
55cd8e5a 2309 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
e7d9513b
AS		 2310 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2311 case HV_X64_MSR_CRASH_CTL:
e83d5887
AS		 2312 return kvm_hv_get_msr_common(vcpu,
2313 msr_info->index, &msr_info->data);
55cd8e5a 2314 break;
91c9c3ed 2315 case MSR_IA32_BBL_CR_CTL3:
2316 /* This legacy MSR exists but isn't fully documented in current
2317 * silicon. It is however accessed by winxp in very narrow
2318 * scenarios where it sets bit #19, itself documented as
2319 * a "reserved" bit. Best effort attempt to source coherent
2320 * read data here should the balance of the register be
2321 * interpreted by the guest:
2322 *
2323 * L2 cache control register 3: 64GB range, 256KB size,
2324 * enabled, latency 0x1, configured
2325 */
609e36d3 2326 msr_info->data = 0xbe702111;
91c9c3ed 2327 break;
2b036c6b
BO		 2328 case MSR_AMD64_OSVW_ID_LENGTH:
2329 if (!guest_cpuid_has_osvw(vcpu))
2330 return 1;
609e36d3 2331 msr_info->data = vcpu->arch.osvw.length;
2b036c6b
BO		 2332 break;
2333 case MSR_AMD64_OSVW_STATUS:
2334 if (!guest_cpuid_has_osvw(vcpu))
2335 return 1;
609e36d3 2336 msr_info->data = vcpu->arch.osvw.status;
2b036c6b 2337 break;
15c4a640 2338 default:
c6702c9d 2339 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
609e36d3 2340 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
ed85c068 2341 if (!ignore_msrs) {
609e36d3 2342 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
ed85c068
AP		 2343 return 1;
2344 } else {
609e36d3
PB		 2345 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2346 msr_info->data = 0;
ed85c068
AP		 2347 }
2348 break;
15c4a640 2349 }
15c4a640
CO		 2350 return 0;
2351}
2352EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2353
313a3dc7
CO		 2354/*
2355 * Read or write a bunch of msrs. All parameters are kernel addresses.
2356 *
2357 * @return number of msrs set successfully.
2358 */
2359static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2360 struct kvm_msr_entry *entries,
2361 int (*do_msr)(struct kvm_vcpu *vcpu,
2362 unsigned index, u64 *data))
2363{
f656ce01 2364 int i, idx;
313a3dc7 2365
f656ce01 2366 idx = srcu_read_lock(&vcpu->kvm->srcu);
313a3dc7
CO		 2367 for (i = 0; i < msrs->nmsrs; ++i)
2368 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2369 break;
f656ce01 2370 srcu_read_unlock(&vcpu->kvm->srcu, idx);
313a3dc7 2371
313a3dc7
CO		 2372 return i;
2373}
2374
2375/*
2376 * Read or write a bunch of msrs. Parameters are user addresses.
2377 *
2378 * @return number of msrs set successfully.
2379 */
2380static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2381 int (*do_msr)(struct kvm_vcpu *vcpu,
2382 unsigned index, u64 *data),
2383 int writeback)
2384{
2385 struct kvm_msrs msrs;
2386 struct kvm_msr_entry *entries;
2387 int r, n;
2388 unsigned size;
2389
2390 r = -EFAULT;
2391 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2392 goto out;
2393
2394 r = -E2BIG;
2395 if (msrs.nmsrs >= MAX_IO_MSRS)
2396 goto out;
2397
313a3dc7 2398 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
ff5c2c03
SL		 2399 entries = memdup_user(user_msrs->entries, size);
2400 if (IS_ERR(entries)) {
2401 r = PTR_ERR(entries);
313a3dc7 2402 goto out;
ff5c2c03 2403 }
313a3dc7
CO		 2404
2405 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2406 if (r < 0)
2407 goto out_free;
2408
2409 r = -EFAULT;
2410 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2411 goto out_free;
2412
2413 r = n;
2414
2415out_free:
7a73c028 2416 kfree(entries);
313a3dc7
CO		 2417out:
2418 return r;
2419}
2420
784aa3d7 2421int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
018d00d2
ZX		 2422{
2423 int r;
2424
2425 switch (ext) {
2426 case KVM_CAP_IRQCHIP:
2427 case KVM_CAP_HLT:
2428 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
018d00d2 2429 case KVM_CAP_SET_TSS_ADDR:
07716717 2430 case KVM_CAP_EXT_CPUID:
9c15bb1d 2431 case KVM_CAP_EXT_EMUL_CPUID:
c8076604 2432 case KVM_CAP_CLOCKSOURCE:
7837699f 2433 case KVM_CAP_PIT:
a28e4f5a 2434 case KVM_CAP_NOP_IO_DELAY:
62d9f0db 2435 case KVM_CAP_MP_STATE:
ed848624 2436 case KVM_CAP_SYNC_MMU:
a355c85c 2437 case KVM_CAP_USER_NMI:
52d939a0 2438 case KVM_CAP_REINJECT_CONTROL:
4925663a 2439 case KVM_CAP_IRQ_INJECT_STATUS:
d34e6b17 2440 case KVM_CAP_IOEVENTFD:
f848a5a8 2441 case KVM_CAP_IOEVENTFD_NO_LENGTH:
c5ff41ce 2442 case KVM_CAP_PIT2:
e9f42757 2443 case KVM_CAP_PIT_STATE2:
b927a3ce 2444 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
ffde22ac 2445 case KVM_CAP_XEN_HVM:
afbcf7ab 2446 case KVM_CAP_ADJUST_CLOCK:
3cfc3092 2447 case KVM_CAP_VCPU_EVENTS:
55cd8e5a 2448 case KVM_CAP_HYPERV:
10388a07 2449 case KVM_CAP_HYPERV_VAPIC:
c25bc163 2450 case KVM_CAP_HYPERV_SPIN:
ab9f4ecb 2451 case KVM_CAP_PCI_SEGMENT:
a1efbe77 2452 case KVM_CAP_DEBUGREGS:
d2be1651 2453 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2d5b5a66 2454 case KVM_CAP_XSAVE:
344d9588 2455 case KVM_CAP_ASYNC_PF:
92a1f12d 2456 case KVM_CAP_GET_TSC_KHZ:
1c0b28c2 2457 case KVM_CAP_KVMCLOCK_CTRL:
4d8b81ab 2458 case KVM_CAP_READONLY_MEM:
5f66b620 2459 case KVM_CAP_HYPERV_TIME:
100943c5 2460 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
defcf51f 2461 case KVM_CAP_TSC_DEADLINE_TIMER:
90de4a18
NA		 2462 case KVM_CAP_ENABLE_CAP_VM:
2463 case KVM_CAP_DISABLE_QUIRKS:
d71ba788 2464 case KVM_CAP_SET_BOOT_CPU_ID:
2a5bab10
AW		 2465#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2466 case KVM_CAP_ASSIGN_DEV_IRQ:
2467 case KVM_CAP_PCI_2_3:
2468#endif
018d00d2
ZX		 2469 r = 1;
2470 break;
6d396b55
PB		 2471 case KVM_CAP_X86_SMM:
2472 /* SMBASE is usually relocated above 1M on modern chipsets,
2473 * and SMM handlers might indeed rely on 4G segment limits,
2474 * so do not report SMM to be available if real mode is
2475 * emulated via vm86 mode. Still, do not go to great lengths
2476 * to avoid userspace's usage of the feature, because it is a
2477 * fringe case that is not enabled except via specific settings
2478 * of the module parameters.
2479 */
2480 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2481 break;
542472b5
LV		 2482 case KVM_CAP_COALESCED_MMIO:
2483 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2484 break;
774ead3a
AK		 2485 case KVM_CAP_VAPIC:
2486 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2487 break;
f725230a 2488 case KVM_CAP_NR_VCPUS:
8c3ba334
SL		 2489 r = KVM_SOFT_MAX_VCPUS;
2490 break;
2491 case KVM_CAP_MAX_VCPUS:
f725230a
AK		 2492 r = KVM_MAX_VCPUS;
2493 break;
a988b910 2494 case KVM_CAP_NR_MEMSLOTS:
bbacc0c1 2495 r = KVM_USER_MEM_SLOTS;
a988b910 2496 break;
a68a6a72
MT		 2497 case KVM_CAP_PV_MMU: /* obsolete */
2498 r = 0;
2f333bcb 2499 break;
4cee4b72 2500#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
62c476c7 2501 case KVM_CAP_IOMMU:
a1b60c1c 2502 r = iommu_present(&pci_bus_type);
62c476c7 2503 break;
4cee4b72 2504#endif
890ca9ae
HY		 2505 case KVM_CAP_MCE:
2506 r = KVM_MAX_MCE_BANKS;
2507 break;
2d5b5a66
SY		 2508 case KVM_CAP_XCRS:
2509 r = cpu_has_xsave;
2510 break;
92a1f12d
JR		 2511 case KVM_CAP_TSC_CONTROL:
2512 r = kvm_has_tsc_control;
2513 break;
018d00d2
ZX		 2514 default:
2515 r = 0;
2516 break;
2517 }
2518 return r;
2519
2520}
2521
043405e1
CO		 2522long kvm_arch_dev_ioctl(struct file *filp,
2523 unsigned int ioctl, unsigned long arg)
2524{
2525 void __user *argp = (void __user *)arg;
2526 long r;
2527
2528 switch (ioctl) {
2529 case KVM_GET_MSR_INDEX_LIST: {
2530 struct kvm_msr_list __user *user_msr_list = argp;
2531 struct kvm_msr_list msr_list;
2532 unsigned n;
2533
2534 r = -EFAULT;
2535 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2536 goto out;
2537 n = msr_list.nmsrs;
62ef68bb 2538 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
043405e1
CO		 2539 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2540 goto out;
2541 r = -E2BIG;
e125e7b6 2542 if (n < msr_list.nmsrs)
043405e1
CO		 2543 goto out;
2544 r = -EFAULT;
2545 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2546 num_msrs_to_save * sizeof(u32)))
2547 goto out;
e125e7b6 2548 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
043405e1 2549 &emulated_msrs,
62ef68bb 2550 num_emulated_msrs * sizeof(u32)))
043405e1
CO		 2551 goto out;
2552 r = 0;
2553 break;
2554 }
9c15bb1d
BP		 2555 case KVM_GET_SUPPORTED_CPUID:
2556 case KVM_GET_EMULATED_CPUID: {
674eea0f
AK		 2557 struct kvm_cpuid2 __user *cpuid_arg = argp;
2558 struct kvm_cpuid2 cpuid;
2559
2560 r = -EFAULT;
2561 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2562 goto out;
9c15bb1d
BP		 2563
2564 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2565 ioctl);
674eea0f
AK		 2566 if (r)
2567 goto out;
2568
2569 r = -EFAULT;
2570 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2571 goto out;
2572 r = 0;
2573 break;
2574 }
890ca9ae
HY		 2575 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2576 u64 mce_cap;
2577
2578 mce_cap = KVM_MCE_CAP_SUPPORTED;
2579 r = -EFAULT;
2580 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2581 goto out;
2582 r = 0;
2583 break;
2584 }
043405e1
CO		 2585 default:
2586 r = -EINVAL;
2587 }
2588out:
2589 return r;
2590}
2591
f5f48ee1
SY		 2592static void wbinvd_ipi(void *garbage)
2593{
2594 wbinvd();
2595}
2596
2597static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2598{
e0f0bbc5 2599 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
f5f48ee1
SY		 2600}
2601
313a3dc7
CO		 2602void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2603{
f5f48ee1
SY		 2604 /* Address WBINVD may be executed by guest */
2605 if (need_emulate_wbinvd(vcpu)) {
2606 if (kvm_x86_ops->has_wbinvd_exit())
2607 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2608 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2609 smp_call_function_single(vcpu->cpu,
2610 wbinvd_ipi, NULL, 1);
2611 }
2612
313a3dc7 2613 kvm_x86_ops->vcpu_load(vcpu, cpu);
8f6055cb 2614
0dd6a6ed
ZA		 2615 /* Apply any externally detected TSC adjustments (due to suspend) */
2616 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2617 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2618 vcpu->arch.tsc_offset_adjustment = 0;
105b21bb 2619 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed 2620 }
8f6055cb 2621
48434c20 2622 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
6f526ec5
ZA		 2623 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2624 native_read_tsc() - vcpu->arch.last_host_tsc;
e48672fa
ZA		 2625 if (tsc_delta < 0)
2626 mark_tsc_unstable("KVM discovered backwards TSC");
c285545f 2627 if (check_tsc_unstable()) {
b183aa58
ZA		 2628 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2629 vcpu->arch.last_guest_tsc);
2630 kvm_x86_ops->write_tsc_offset(vcpu, offset);
c285545f 2631 vcpu->arch.tsc_catchup = 1;
c285545f 2632 }
d98d07ca
MT		 2633 /*
2634 * On a host with synchronized TSC, there is no need to update
2635 * kvmclock on vcpu->cpu migration
2636 */
2637 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
0061d53d 2638 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
c285545f
ZA		 2639 if (vcpu->cpu != cpu)
2640 kvm_migrate_timers(vcpu);
e48672fa 2641 vcpu->cpu = cpu;
6b7d7e76 2642 }
c9aaa895
GC		 2643
2644 accumulate_steal_time(vcpu);
2645 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
313a3dc7
CO		 2646}
2647
2648void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2649{
02daab21 2650 kvm_x86_ops->vcpu_put(vcpu);
1c11e713 2651 kvm_put_guest_fpu(vcpu);
6f526ec5 2652 vcpu->arch.last_host_tsc = native_read_tsc();
313a3dc7
CO		 2653}
2654
313a3dc7
CO		 2655static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2656 struct kvm_lapic_state *s)
2657{
5a71785d 2658 kvm_x86_ops->sync_pir_to_irr(vcpu);
ad312c7c 2659 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
313a3dc7
CO		 2660
2661 return 0;
2662}
2663
2664static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2665 struct kvm_lapic_state *s)
2666{
64eb0620 2667 kvm_apic_post_state_restore(vcpu, s);
cb142eb7 2668 update_cr8_intercept(vcpu);
313a3dc7
CO		 2669
2670 return 0;
2671}
2672
f77bc6a4
ZX		 2673static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2674 struct kvm_interrupt *irq)
2675{
02cdb50f 2676 if (irq->irq >= KVM_NR_INTERRUPTS)
f77bc6a4
ZX		 2677 return -EINVAL;
2678 if (irqchip_in_kernel(vcpu->kvm))
2679 return -ENXIO;
f77bc6a4 2680
66fd3f7f 2681 kvm_queue_interrupt(vcpu, irq->irq, false);
3842d135 2682 kvm_make_request(KVM_REQ_EVENT, vcpu);
f77bc6a4 2683
f77bc6a4
ZX		 2684 return 0;
2685}
2686
c4abb7c9
JK		 2687static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2688{
c4abb7c9 2689 kvm_inject_nmi(vcpu);
c4abb7c9
JK		 2690
2691 return 0;
2692}
2693
f077825a
PB		 2694static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2695{
64d60670
PB		 2696 kvm_make_request(KVM_REQ_SMI, vcpu);
2697
f077825a
PB		 2698 return 0;
2699}
2700
b209749f
AK		 2701static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2702 struct kvm_tpr_access_ctl *tac)
2703{
2704 if (tac->flags)
2705 return -EINVAL;
2706 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2707 return 0;
2708}
2709
890ca9ae
HY		 2710static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2711 u64 mcg_cap)
2712{
2713 int r;
2714 unsigned bank_num = mcg_cap & 0xff, bank;
2715
2716 r = -EINVAL;
a9e38c3e 2717 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
890ca9ae
HY		 2718 goto out;
2719 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2720 goto out;
2721 r = 0;
2722 vcpu->arch.mcg_cap = mcg_cap;
2723 /* Init IA32_MCG_CTL to all 1s */
2724 if (mcg_cap & MCG_CTL_P)
2725 vcpu->arch.mcg_ctl = ~(u64)0;
2726 /* Init IA32_MCi_CTL to all 1s */
2727 for (bank = 0; bank < bank_num; bank++)
2728 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2729out:
2730 return r;
2731}
2732
2733static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2734 struct kvm_x86_mce *mce)
2735{
2736 u64 mcg_cap = vcpu->arch.mcg_cap;
2737 unsigned bank_num = mcg_cap & 0xff;
2738 u64 *banks = vcpu->arch.mce_banks;
2739
2740 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2741 return -EINVAL;
2742 /*
2743 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2744 * reporting is disabled
2745 */
2746 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2747 vcpu->arch.mcg_ctl != ~(u64)0)
2748 return 0;
2749 banks += 4 * mce->bank;
2750 /*
2751 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2752 * reporting is disabled for the bank
2753 */
2754 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2755 return 0;
2756 if (mce->status & MCI_STATUS_UC) {
2757 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
fc78f519 2758 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
a8eeb04a 2759 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
890ca9ae
HY		 2760 return 0;
2761 }
2762 if (banks[1] & MCI_STATUS_VAL)
2763 mce->status |= MCI_STATUS_OVER;
2764 banks[2] = mce->addr;
2765 banks[3] = mce->misc;
2766 vcpu->arch.mcg_status = mce->mcg_status;
2767 banks[1] = mce->status;
2768 kvm_queue_exception(vcpu, MC_VECTOR);
2769 } else if (!(banks[1] & MCI_STATUS_VAL)
2770 || !(banks[1] & MCI_STATUS_UC)) {
2771 if (banks[1] & MCI_STATUS_VAL)
2772 mce->status |= MCI_STATUS_OVER;
2773 banks[2] = mce->addr;
2774 banks[3] = mce->misc;
2775 banks[1] = mce->status;
2776 } else
2777 banks[1] |= MCI_STATUS_OVER;
2778 return 0;
2779}
2780
3cfc3092
JK		 2781static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2782 struct kvm_vcpu_events *events)
2783{
7460fb4a 2784 process_nmi(vcpu);
03b82a30
JK		 2785 events->exception.injected =
2786 vcpu->arch.exception.pending &&
2787 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3cfc3092
JK		 2788 events->exception.nr = vcpu->arch.exception.nr;
2789 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
97e69aa6 2790 events->exception.pad = 0;
3cfc3092
JK		 2791 events->exception.error_code = vcpu->arch.exception.error_code;
2792
03b82a30
JK		 2793 events->interrupt.injected =
2794 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3cfc3092 2795 events->interrupt.nr = vcpu->arch.interrupt.nr;
03b82a30 2796 events->interrupt.soft = 0;
37ccdcbe 2797 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3cfc3092
JK		 2798
2799 events->nmi.injected = vcpu->arch.nmi_injected;
7460fb4a 2800 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3cfc3092 2801 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
97e69aa6 2802 events->nmi.pad = 0;
3cfc3092 2803
66450a21 2804 events->sipi_vector = 0; /* never valid when reporting to user space */
3cfc3092 2805
f077825a
PB		 2806 events->smi.smm = is_smm(vcpu);
2807 events->smi.pending = vcpu->arch.smi_pending;
2808 events->smi.smm_inside_nmi =
2809 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2810 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2811
dab4b911 2812 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
f077825a
PB		 2813 | KVM_VCPUEVENT_VALID_SHADOW
2814 | KVM_VCPUEVENT_VALID_SMM);
97e69aa6 2815 memset(&events->reserved, 0, sizeof(events->reserved));
3cfc3092
JK		 2816}
2817
2818static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2819 struct kvm_vcpu_events *events)
2820{
dab4b911 2821 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64 2822 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
f077825a
PB		 2823 | KVM_VCPUEVENT_VALID_SHADOW
2824 | KVM_VCPUEVENT_VALID_SMM))
3cfc3092
JK		 2825 return -EINVAL;
2826
7460fb4a 2827 process_nmi(vcpu);
3cfc3092
JK		 2828 vcpu->arch.exception.pending = events->exception.injected;
2829 vcpu->arch.exception.nr = events->exception.nr;
2830 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2831 vcpu->arch.exception.error_code = events->exception.error_code;
2832
2833 vcpu->arch.interrupt.pending = events->interrupt.injected;
2834 vcpu->arch.interrupt.nr = events->interrupt.nr;
2835 vcpu->arch.interrupt.soft = events->interrupt.soft;
48005f64
JK		 2836 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2837 kvm_x86_ops->set_interrupt_shadow(vcpu,
2838 events->interrupt.shadow);
3cfc3092
JK		 2839
2840 vcpu->arch.nmi_injected = events->nmi.injected;
dab4b911
JK		 2841 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2842 vcpu->arch.nmi_pending = events->nmi.pending;
3cfc3092
JK		 2843 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2844
66450a21
JK		 2845 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2846 kvm_vcpu_has_lapic(vcpu))
2847 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3cfc3092 2848
f077825a
PB		 2849 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
2850 if (events->smi.smm)
2851 vcpu->arch.hflags |= HF_SMM_MASK;
2852 else
2853 vcpu->arch.hflags &= ~HF_SMM_MASK;
2854 vcpu->arch.smi_pending = events->smi.pending;
2855 if (events->smi.smm_inside_nmi)
2856 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
2857 else
2858 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
2859 if (kvm_vcpu_has_lapic(vcpu)) {
2860 if (events->smi.latched_init)
2861 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2862 else
2863 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2864 }
2865 }
2866
3842d135
AK		 2867 kvm_make_request(KVM_REQ_EVENT, vcpu);
2868
3cfc3092
JK		 2869 return 0;
2870}
2871
a1efbe77
JK		 2872static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2873 struct kvm_debugregs *dbgregs)
2874{
73aaf249
JK		 2875 unsigned long val;
2876
a1efbe77 2877 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
16f8a6f9 2878 kvm_get_dr(vcpu, 6, &val);
73aaf249 2879 dbgregs->dr6 = val;
a1efbe77
JK		 2880 dbgregs->dr7 = vcpu->arch.dr7;
2881 dbgregs->flags = 0;
97e69aa6 2882 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
a1efbe77
JK		 2883}
2884
2885static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2886 struct kvm_debugregs *dbgregs)
2887{
2888 if (dbgregs->flags)
2889 return -EINVAL;
2890
a1efbe77 2891 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
ae561ede 2892 kvm_update_dr0123(vcpu);
a1efbe77 2893 vcpu->arch.dr6 = dbgregs->dr6;
73aaf249 2894 kvm_update_dr6(vcpu);
a1efbe77 2895 vcpu->arch.dr7 = dbgregs->dr7;
9926c9fd 2896 kvm_update_dr7(vcpu);
a1efbe77 2897
a1efbe77
JK		 2898 return 0;
2899}
2900
df1daba7
PB		 2901#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
2902
2903static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
2904{
c47ada30 2905 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
400e4b20 2906 u64 xstate_bv = xsave->header.xfeatures;
df1daba7
PB		 2907 u64 valid;
2908
2909 /*
2910 * Copy legacy XSAVE area, to avoid complications with CPUID
2911 * leaves 0 and 1 in the loop below.
2912 */
2913 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
2914
2915 /* Set XSTATE_BV */
2916 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
2917
2918 /*
2919 * Copy each region from the possibly compacted offset to the
2920 * non-compacted offset.
2921 */
2922 valid = xstate_bv & ~XSTATE_FPSSE;
2923 while (valid) {
2924 u64 feature = valid & -valid;
2925 int index = fls64(feature) - 1;
2926 void *src = get_xsave_addr(xsave, feature);
2927
2928 if (src) {
2929 u32 size, offset, ecx, edx;
2930 cpuid_count(XSTATE_CPUID, index,
2931 &size, &offset, &ecx, &edx);
2932 memcpy(dest + offset, src, size);
2933 }
2934
2935 valid -= feature;
2936 }
2937}
2938
2939static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
2940{
c47ada30 2941 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
df1daba7
PB		 2942 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
2943 u64 valid;
2944
2945 /*
2946 * Copy legacy XSAVE area, to avoid complications with CPUID
2947 * leaves 0 and 1 in the loop below.
2948 */
2949 memcpy(xsave, src, XSAVE_HDR_OFFSET);
2950
2951 /* Set XSTATE_BV and possibly XCOMP_BV. */
400e4b20 2952 xsave->header.xfeatures = xstate_bv;
df1daba7 2953 if (cpu_has_xsaves)
3a54450b 2954 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
df1daba7
PB		 2955
2956 /*
2957 * Copy each region from the non-compacted offset to the
2958 * possibly compacted offset.
2959 */
2960 valid = xstate_bv & ~XSTATE_FPSSE;
2961 while (valid) {
2962 u64 feature = valid & -valid;
2963 int index = fls64(feature) - 1;
2964 void *dest = get_xsave_addr(xsave, feature);
2965
2966 if (dest) {
2967 u32 size, offset, ecx, edx;
2968 cpuid_count(XSTATE_CPUID, index,
2969 &size, &offset, &ecx, &edx);
2970 memcpy(dest, src + offset, size);
ee4100da 2971 }
df1daba7
PB		 2972
2973 valid -= feature;
2974 }
2975}
2976
2d5b5a66
SY		 2977static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2978 struct kvm_xsave *guest_xsave)
2979{
4344ee98 2980 if (cpu_has_xsave) {
df1daba7
PB		 2981 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
2982 fill_xsave((u8 *) guest_xsave->region, vcpu);
4344ee98 2983 } else {
2d5b5a66 2984 memcpy(guest_xsave->region,
7366ed77 2985 &vcpu->arch.guest_fpu.state.fxsave,
c47ada30 2986 sizeof(struct fxregs_state));
2d5b5a66
SY		 2987 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2988 XSTATE_FPSSE;
2989 }
2990}
2991
2992static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2993 struct kvm_xsave *guest_xsave)
2994{
2995 u64 xstate_bv =
2996 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2997
d7876f1b
PB		 2998 if (cpu_has_xsave) {
2999 /*
3000 * Here we allow setting states that are not present in
3001 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3002 * with old userspace.
3003 */
4ff41732 3004 if (xstate_bv & ~kvm_supported_xcr0())
d7876f1b 3005 return -EINVAL;
df1daba7 3006 load_xsave(vcpu, (u8 *)guest_xsave->region);
d7876f1b 3007 } else {
2d5b5a66
SY		 3008 if (xstate_bv & ~XSTATE_FPSSE)
3009 return -EINVAL;
7366ed77 3010 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
c47ada30 3011 guest_xsave->region, sizeof(struct fxregs_state));
2d5b5a66
SY		 3012 }
3013 return 0;
3014}
3015
3016static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3017 struct kvm_xcrs *guest_xcrs)
3018{
3019 if (!cpu_has_xsave) {
3020 guest_xcrs->nr_xcrs = 0;
3021 return;
3022 }
3023
3024 guest_xcrs->nr_xcrs = 1;
3025 guest_xcrs->flags = 0;
3026 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3027 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3028}
3029
3030static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3031 struct kvm_xcrs *guest_xcrs)
3032{
3033 int i, r = 0;
3034
3035 if (!cpu_has_xsave)
3036 return -EINVAL;
3037
3038 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3039 return -EINVAL;
3040
3041 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3042 /* Only support XCR0 currently */
c67a04cb 3043 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
2d5b5a66 3044 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
c67a04cb 3045 guest_xcrs->xcrs[i].value);
2d5b5a66
SY		 3046 break;
3047 }
3048 if (r)
3049 r = -EINVAL;
3050 return r;
3051}
3052
1c0b28c2
EM		 3053/*
3054 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3055 * stopped by the hypervisor. This function will be called from the host only.
3056 * EINVAL is returned when the host attempts to set the flag for a guest that
3057 * does not support pv clocks.
3058 */
3059static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3060{
0b79459b 3061 if (!vcpu->arch.pv_time_enabled)
1c0b28c2 3062 return -EINVAL;
51d59c6b 3063 vcpu->arch.pvclock_set_guest_stopped_request = true;
1c0b28c2
EM		 3064 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3065 return 0;
3066}
3067
313a3dc7
CO		 3068long kvm_arch_vcpu_ioctl(struct file *filp,
3069 unsigned int ioctl, unsigned long arg)
3070{
3071 struct kvm_vcpu *vcpu = filp->private_data;
3072 void __user *argp = (void __user *)arg;
3073 int r;
d1ac91d8
AK		 3074 union {
3075 struct kvm_lapic_state *lapic;
3076 struct kvm_xsave *xsave;
3077 struct kvm_xcrs *xcrs;
3078 void *buffer;
3079 } u;
3080
3081 u.buffer = NULL;
313a3dc7
CO		 3082 switch (ioctl) {
3083 case KVM_GET_LAPIC: {
2204ae3c
MT		 3084 r = -EINVAL;
3085 if (!vcpu->arch.apic)
3086 goto out;
d1ac91d8 3087 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
313a3dc7 3088
b772ff36 3089 r = -ENOMEM;
d1ac91d8 3090 if (!u.lapic)
b772ff36 3091 goto out;
d1ac91d8 3092 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
313a3dc7
CO		 3093 if (r)
3094 goto out;
3095 r = -EFAULT;
d1ac91d8 3096 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
313a3dc7
CO		 3097 goto out;
3098 r = 0;
3099 break;
3100 }
3101 case KVM_SET_LAPIC: {
2204ae3c
MT		 3102 r = -EINVAL;
3103 if (!vcpu->arch.apic)
3104 goto out;
ff5c2c03 3105 u.lapic = memdup_user(argp, sizeof(*u.lapic));
18595411
GC		 3106 if (IS_ERR(u.lapic))
3107 return PTR_ERR(u.lapic);
ff5c2c03 3108
d1ac91d8 3109 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
313a3dc7
CO		 3110 break;
3111 }
f77bc6a4
ZX		 3112 case KVM_INTERRUPT: {
3113 struct kvm_interrupt irq;
3114
3115 r = -EFAULT;
3116 if (copy_from_user(&irq, argp, sizeof irq))
3117 goto out;
3118 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
f77bc6a4
ZX		 3119 break;
3120 }
c4abb7c9
JK		 3121 case KVM_NMI: {
3122 r = kvm_vcpu_ioctl_nmi(vcpu);
c4abb7c9
JK		 3123 break;
3124 }
f077825a
PB		 3125 case KVM_SMI: {
3126 r = kvm_vcpu_ioctl_smi(vcpu);
3127 break;
3128 }
313a3dc7
CO		 3129 case KVM_SET_CPUID: {
3130 struct kvm_cpuid __user *cpuid_arg = argp;
3131 struct kvm_cpuid cpuid;
3132
3133 r = -EFAULT;
3134 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3135 goto out;
3136 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
313a3dc7
CO		 3137 break;
3138 }
07716717
DK		 3139 case KVM_SET_CPUID2: {
3140 struct kvm_cpuid2 __user *cpuid_arg = argp;
3141 struct kvm_cpuid2 cpuid;
3142
3143 r = -EFAULT;
3144 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3145 goto out;
3146 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
19355475 3147 cpuid_arg->entries);
07716717
DK		 3148 break;
3149 }
3150 case KVM_GET_CPUID2: {
3151 struct kvm_cpuid2 __user *cpuid_arg = argp;
3152 struct kvm_cpuid2 cpuid;
3153
3154 r = -EFAULT;
3155 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3156 goto out;
3157 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
19355475 3158 cpuid_arg->entries);
07716717
DK		 3159 if (r)
3160 goto out;
3161 r = -EFAULT;
3162 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3163 goto out;
3164 r = 0;
3165 break;
3166 }
313a3dc7 3167 case KVM_GET_MSRS:
609e36d3 3168 r = msr_io(vcpu, argp, do_get_msr, 1);
313a3dc7
CO		 3169 break;
3170 case KVM_SET_MSRS:
3171 r = msr_io(vcpu, argp, do_set_msr, 0);
3172 break;
b209749f
AK		 3173 case KVM_TPR_ACCESS_REPORTING: {
3174 struct kvm_tpr_access_ctl tac;
3175
3176 r = -EFAULT;
3177 if (copy_from_user(&tac, argp, sizeof tac))
3178 goto out;
3179 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3180 if (r)
3181 goto out;
3182 r = -EFAULT;
3183 if (copy_to_user(argp, &tac, sizeof tac))
3184 goto out;
3185 r = 0;
3186 break;
3187 };
b93463aa
AK		 3188 case KVM_SET_VAPIC_ADDR: {
3189 struct kvm_vapic_addr va;
3190
3191 r = -EINVAL;
3192 if (!irqchip_in_kernel(vcpu->kvm))
3193 goto out;
3194 r = -EFAULT;
3195 if (copy_from_user(&va, argp, sizeof va))
3196 goto out;
fda4e2e8 3197 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
b93463aa
AK		 3198 break;
3199 }
890ca9ae
HY		 3200 case KVM_X86_SETUP_MCE: {
3201 u64 mcg_cap;
3202
3203 r = -EFAULT;
3204 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3205 goto out;
3206 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3207 break;
3208 }
3209 case KVM_X86_SET_MCE: {
3210 struct kvm_x86_mce mce;
3211
3212 r = -EFAULT;
3213 if (copy_from_user(&mce, argp, sizeof mce))
3214 goto out;
3215 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3216 break;
3217 }
3cfc3092
JK		 3218 case KVM_GET_VCPU_EVENTS: {
3219 struct kvm_vcpu_events events;
3220
3221 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3222
3223 r = -EFAULT;
3224 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3225 break;
3226 r = 0;
3227 break;
3228 }
3229 case KVM_SET_VCPU_EVENTS: {
3230 struct kvm_vcpu_events events;
3231
3232 r = -EFAULT;
3233 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3234 break;
3235
3236 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3237 break;
3238 }
a1efbe77
JK		 3239 case KVM_GET_DEBUGREGS: {
3240 struct kvm_debugregs dbgregs;
3241
3242 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3243
3244 r = -EFAULT;
3245 if (copy_to_user(argp, &dbgregs,
3246 sizeof(struct kvm_debugregs)))
3247 break;
3248 r = 0;
3249 break;
3250 }
3251 case KVM_SET_DEBUGREGS: {
3252 struct kvm_debugregs dbgregs;
3253
3254 r = -EFAULT;
3255 if (copy_from_user(&dbgregs, argp,
3256 sizeof(struct kvm_debugregs)))
3257 break;
3258
3259 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3260 break;
3261 }
2d5b5a66 3262 case KVM_GET_XSAVE: {
d1ac91d8 3263 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2d5b5a66 3264 r = -ENOMEM;
d1ac91d8 3265 if (!u.xsave)
2d5b5a66
SY		 3266 break;
3267
d1ac91d8 3268 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3269
3270 r = -EFAULT;
d1ac91d8 3271 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2d5b5a66
SY		 3272 break;
3273 r = 0;
3274 break;
3275 }
3276 case KVM_SET_XSAVE: {
ff5c2c03 3277 u.xsave = memdup_user(argp, sizeof(*u.xsave));
18595411
GC		 3278 if (IS_ERR(u.xsave))
3279 return PTR_ERR(u.xsave);
2d5b5a66 3280
d1ac91d8 3281 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3282 break;
3283 }
3284 case KVM_GET_XCRS: {
d1ac91d8 3285 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2d5b5a66 3286 r = -ENOMEM;
d1ac91d8 3287 if (!u.xcrs)
2d5b5a66
SY		 3288 break;
3289
d1ac91d8 3290 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3291
3292 r = -EFAULT;
d1ac91d8 3293 if (copy_to_user(argp, u.xcrs,
2d5b5a66
SY		 3294 sizeof(struct kvm_xcrs)))
3295 break;
3296 r = 0;
3297 break;
3298 }
3299 case KVM_SET_XCRS: {
ff5c2c03 3300 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
18595411
GC		 3301 if (IS_ERR(u.xcrs))
3302 return PTR_ERR(u.xcrs);
2d5b5a66 3303
d1ac91d8 3304 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3305 break;
3306 }
92a1f12d
JR		 3307 case KVM_SET_TSC_KHZ: {
3308 u32 user_tsc_khz;
3309
3310 r = -EINVAL;
92a1f12d
JR		 3311 user_tsc_khz = (u32)arg;
3312
3313 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3314 goto out;
3315
cc578287
ZA		 3316 if (user_tsc_khz == 0)
3317 user_tsc_khz = tsc_khz;
3318
3319 kvm_set_tsc_khz(vcpu, user_tsc_khz);
92a1f12d
JR		 3320
3321 r = 0;
3322 goto out;
3323 }
3324 case KVM_GET_TSC_KHZ: {
cc578287 3325 r = vcpu->arch.virtual_tsc_khz;
92a1f12d
JR		 3326 goto out;
3327 }
1c0b28c2
EM		 3328 case KVM_KVMCLOCK_CTRL: {
3329 r = kvm_set_guest_paused(vcpu);
3330 goto out;
3331 }
313a3dc7
CO		 3332 default:
3333 r = -EINVAL;
3334 }
3335out:
d1ac91d8 3336 kfree(u.buffer);
313a3dc7
CO		 3337 return r;
3338}
3339
5b1c1493
CO		 3340int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3341{
3342 return VM_FAULT_SIGBUS;
3343}
3344
1fe779f8
CO		 3345static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3346{
3347 int ret;
3348
3349 if (addr > (unsigned int)(-3 * PAGE_SIZE))
951179ce 3350 return -EINVAL;
1fe779f8
CO		 3351 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3352 return ret;
3353}
3354
b927a3ce
SY		 3355static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3356 u64 ident_addr)
3357{
3358 kvm->arch.ept_identity_map_addr = ident_addr;
3359 return 0;
3360}
3361
1fe779f8
CO		 3362static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3363 u32 kvm_nr_mmu_pages)
3364{
3365 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3366 return -EINVAL;
3367
79fac95e 3368 mutex_lock(&kvm->slots_lock);
1fe779f8
CO		 3369
3370 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
f05e70ac 3371 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1fe779f8 3372
79fac95e 3373 mutex_unlock(&kvm->slots_lock);
1fe779f8
CO		 3374 return 0;
3375}
3376
3377static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3378{
39de71ec 3379 return kvm->arch.n_max_mmu_pages;
1fe779f8
CO		 3380}
3381
1fe779f8
CO		 3382static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3383{
3384 int r;
3385
3386 r = 0;
3387 switch (chip->chip_id) {
3388 case KVM_IRQCHIP_PIC_MASTER:
3389 memcpy(&chip->chip.pic,
3390 &pic_irqchip(kvm)->pics[0],
3391 sizeof(struct kvm_pic_state));
3392 break;
3393 case KVM_IRQCHIP_PIC_SLAVE:
3394 memcpy(&chip->chip.pic,
3395 &pic_irqchip(kvm)->pics[1],
3396 sizeof(struct kvm_pic_state));
3397 break;
3398 case KVM_IRQCHIP_IOAPIC:
eba0226b 3399 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3400 break;
3401 default:
3402 r = -EINVAL;
3403 break;
3404 }
3405 return r;
3406}
3407
3408static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3409{
3410 int r;
3411
3412 r = 0;
3413 switch (chip->chip_id) {
3414 case KVM_IRQCHIP_PIC_MASTER:
f4f51050 3415 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3416 memcpy(&pic_irqchip(kvm)->pics[0],
3417 &chip->chip.pic,
3418 sizeof(struct kvm_pic_state));
f4f51050 3419 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3420 break;
3421 case KVM_IRQCHIP_PIC_SLAVE:
f4f51050 3422 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3423 memcpy(&pic_irqchip(kvm)->pics[1],
3424 &chip->chip.pic,
3425 sizeof(struct kvm_pic_state));
f4f51050 3426 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3427 break;
3428 case KVM_IRQCHIP_IOAPIC:
eba0226b 3429 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3430 break;
3431 default:
3432 r = -EINVAL;
3433 break;
3434 }
3435 kvm_pic_update_irq(pic_irqchip(kvm));
3436 return r;
3437}
3438
e0f63cb9
SY		 3439static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3440{
3441 int r = 0;
3442
894a9c55 3443 mutex_lock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9 3444 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
894a9c55 3445 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9
SY		 3446 return r;
3447}
3448
3449static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3450{
3451 int r = 0;
3452
894a9c55 3453 mutex_lock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9 3454 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
e9f42757
BK		 3455 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3456 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3457 return r;
3458}
3459
3460static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3461{
3462 int r = 0;
3463
3464 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3465 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3466 sizeof(ps->channels));
3467 ps->flags = kvm->arch.vpit->pit_state.flags;
3468 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
97e69aa6 3469 memset(&ps->reserved, 0, sizeof(ps->reserved));
e9f42757
BK		 3470 return r;
3471}
3472
3473static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3474{
3475 int r = 0, start = 0;
3476 u32 prev_legacy, cur_legacy;
3477 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3478 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3479 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3480 if (!prev_legacy && cur_legacy)
3481 start = 1;
3482 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3483 sizeof(kvm->arch.vpit->pit_state.channels));
3484 kvm->arch.vpit->pit_state.flags = ps->flags;
3485 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
894a9c55 3486 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9
SY		 3487 return r;
3488}
3489
52d939a0
MT		 3490static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3491 struct kvm_reinject_control *control)
3492{
3493 if (!kvm->arch.vpit)
3494 return -ENXIO;
894a9c55 3495 mutex_lock(&kvm->arch.vpit->pit_state.lock);
26ef1924 3496 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
894a9c55 3497 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
52d939a0
MT		 3498 return 0;
3499}
3500
95d4c16c 3501/**
60c34612
TY		 3502 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3503 * @kvm: kvm instance
3504 * @log: slot id and address to which we copy the log
95d4c16c 3505 *
e108ff2f
PB		 3506 * Steps 1-4 below provide general overview of dirty page logging. See
3507 * kvm_get_dirty_log_protect() function description for additional details.
3508 *
3509 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3510 * always flush the TLB (step 4) even if previous step failed and the dirty
3511 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3512 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3513 * writes will be marked dirty for next log read.
95d4c16c 3514 *
60c34612
TY		 3515 * 1. Take a snapshot of the bit and clear it if needed.
3516 * 2. Write protect the corresponding page.
e108ff2f
PB		 3517 * 3. Copy the snapshot to the userspace.
3518 * 4. Flush TLB's if needed.
5bb064dc 3519 */
60c34612 3520int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
5bb064dc 3521{
60c34612 3522 bool is_dirty = false;
e108ff2f 3523 int r;
5bb064dc 3524
79fac95e 3525 mutex_lock(&kvm->slots_lock);
5bb064dc 3526
88178fd4
KH		 3527 /*
3528 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3529 */
3530 if (kvm_x86_ops->flush_log_dirty)
3531 kvm_x86_ops->flush_log_dirty(kvm);
3532
e108ff2f 3533 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
198c74f4
XG		 3534
3535 /*
3536 * All the TLBs can be flushed out of mmu lock, see the comments in
3537 * kvm_mmu_slot_remove_write_access().
3538 */
e108ff2f 3539 lockdep_assert_held(&kvm->slots_lock);
198c74f4
XG		 3540 if (is_dirty)
3541 kvm_flush_remote_tlbs(kvm);
3542
79fac95e 3543 mutex_unlock(&kvm->slots_lock);
5bb064dc
ZX		 3544 return r;
3545}
3546
aa2fbe6d
YZ		 3547int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3548 bool line_status)
23d43cf9
CD		 3549{
3550 if (!irqchip_in_kernel(kvm))
3551 return -ENXIO;
3552
3553 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
aa2fbe6d
YZ		 3554 irq_event->irq, irq_event->level,
3555 line_status);
23d43cf9
CD		 3556 return 0;
3557}
3558
90de4a18
NA		 3559static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3560 struct kvm_enable_cap *cap)
3561{
3562 int r;
3563
3564 if (cap->flags)
3565 return -EINVAL;
3566
3567 switch (cap->cap) {
3568 case KVM_CAP_DISABLE_QUIRKS:
3569 kvm->arch.disabled_quirks = cap->args[0];
3570 r = 0;
3571 break;
3572 default:
3573 r = -EINVAL;
3574 break;
3575 }
3576 return r;
3577}
3578
1fe779f8
CO		 3579long kvm_arch_vm_ioctl(struct file *filp,
3580 unsigned int ioctl, unsigned long arg)
3581{
3582 struct kvm *kvm = filp->private_data;
3583 void __user *argp = (void __user *)arg;
367e1319 3584 int r = -ENOTTY;
f0d66275
DH		 3585 /*
3586 * This union makes it completely explicit to gcc-3.x
3587 * that these two variables' stack usage should be
3588 * combined, not added together.
3589 */
3590 union {
3591 struct kvm_pit_state ps;
e9f42757 3592 struct kvm_pit_state2 ps2;
c5ff41ce 3593 struct kvm_pit_config pit_config;
f0d66275 3594 } u;
1fe779f8
CO		 3595
3596 switch (ioctl) {
3597 case KVM_SET_TSS_ADDR:
3598 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1fe779f8 3599 break;
b927a3ce
SY		 3600 case KVM_SET_IDENTITY_MAP_ADDR: {
3601 u64 ident_addr;
3602
3603 r = -EFAULT;
3604 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3605 goto out;
3606 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
b927a3ce
SY		 3607 break;
3608 }
1fe779f8
CO		 3609 case KVM_SET_NR_MMU_PAGES:
3610 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1fe779f8
CO		 3611 break;
3612 case KVM_GET_NR_MMU_PAGES:
3613 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3614 break;
3ddea128
MT		 3615 case KVM_CREATE_IRQCHIP: {
3616 struct kvm_pic *vpic;
3617
3618 mutex_lock(&kvm->lock);
3619 r = -EEXIST;
3620 if (kvm->arch.vpic)
3621 goto create_irqchip_unlock;
3e515705
AK		 3622 r = -EINVAL;
3623 if (atomic_read(&kvm->online_vcpus))
3624 goto create_irqchip_unlock;
1fe779f8 3625 r = -ENOMEM;
3ddea128
MT		 3626 vpic = kvm_create_pic(kvm);
3627 if (vpic) {
1fe779f8
CO		 3628 r = kvm_ioapic_init(kvm);
3629 if (r) {
175504cd 3630 mutex_lock(&kvm->slots_lock);
72bb2fcd 3631 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
743eeb0b
SL		 3632 &vpic->dev_master);
3633 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3634 &vpic->dev_slave);
3635 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3636 &vpic->dev_eclr);
175504cd 3637 mutex_unlock(&kvm->slots_lock);
3ddea128
MT		 3638 kfree(vpic);
3639 goto create_irqchip_unlock;
1fe779f8
CO		 3640 }
3641 } else
3ddea128
MT		 3642 goto create_irqchip_unlock;
3643 smp_wmb();
3644 kvm->arch.vpic = vpic;
3645 smp_wmb();
399ec807
AK		 3646 r = kvm_setup_default_irq_routing(kvm);
3647 if (r) {
175504cd 3648 mutex_lock(&kvm->slots_lock);
3ddea128 3649 mutex_lock(&kvm->irq_lock);
72bb2fcd
WY		 3650 kvm_ioapic_destroy(kvm);
3651 kvm_destroy_pic(kvm);
3ddea128 3652 mutex_unlock(&kvm->irq_lock);
175504cd 3653 mutex_unlock(&kvm->slots_lock);
399ec807 3654 }
3ddea128
MT		 3655 create_irqchip_unlock:
3656 mutex_unlock(&kvm->lock);
1fe779f8 3657 break;
3ddea128 3658 }
7837699f 3659 case KVM_CREATE_PIT:
c5ff41ce
JK		 3660 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3661 goto create_pit;
3662 case KVM_CREATE_PIT2:
3663 r = -EFAULT;
3664 if (copy_from_user(&u.pit_config, argp,
3665 sizeof(struct kvm_pit_config)))
3666 goto out;
3667 create_pit:
79fac95e 3668 mutex_lock(&kvm->slots_lock);
269e05e4
AK		 3669 r = -EEXIST;
3670 if (kvm->arch.vpit)
3671 goto create_pit_unlock;
7837699f 3672 r = -ENOMEM;
c5ff41ce 3673 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
7837699f
SY		 3674 if (kvm->arch.vpit)
3675 r = 0;
269e05e4 3676 create_pit_unlock:
79fac95e 3677 mutex_unlock(&kvm->slots_lock);
7837699f 3678 break;
1fe779f8
CO		 3679 case KVM_GET_IRQCHIP: {
3680 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 3681 struct kvm_irqchip *chip;
1fe779f8 3682
ff5c2c03
SL		 3683 chip = memdup_user(argp, sizeof(*chip));
3684 if (IS_ERR(chip)) {
3685 r = PTR_ERR(chip);
1fe779f8 3686 goto out;
ff5c2c03
SL		 3687 }
3688
1fe779f8
CO		 3689 r = -ENXIO;
3690 if (!irqchip_in_kernel(kvm))
f0d66275
DH		 3691 goto get_irqchip_out;
3692 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1fe779f8 3693 if (r)
f0d66275 3694 goto get_irqchip_out;
1fe779f8 3695 r = -EFAULT;
f0d66275
DH		 3696 if (copy_to_user(argp, chip, sizeof *chip))
3697 goto get_irqchip_out;
1fe779f8 3698 r = 0;
f0d66275
DH		 3699 get_irqchip_out:
3700 kfree(chip);
1fe779f8
CO		 3701 break;
3702 }
3703 case KVM_SET_IRQCHIP: {
3704 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 3705 struct kvm_irqchip *chip;
1fe779f8 3706
ff5c2c03
SL		 3707 chip = memdup_user(argp, sizeof(*chip));
3708 if (IS_ERR(chip)) {
3709 r = PTR_ERR(chip);
1fe779f8 3710 goto out;
ff5c2c03
SL		 3711 }
3712
1fe779f8
CO		 3713 r = -ENXIO;
3714 if (!irqchip_in_kernel(kvm))
f0d66275
DH		 3715 goto set_irqchip_out;
3716 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1fe779f8 3717 if (r)
f0d66275 3718 goto set_irqchip_out;
1fe779f8 3719 r = 0;
f0d66275
DH		 3720 set_irqchip_out:
3721 kfree(chip);
1fe779f8
CO		 3722 break;
3723 }
e0f63cb9 3724 case KVM_GET_PIT: {
e0f63cb9 3725 r = -EFAULT;
f0d66275 3726 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 3727 goto out;
3728 r = -ENXIO;
3729 if (!kvm->arch.vpit)
3730 goto out;
f0d66275 3731 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
e0f63cb9
SY		 3732 if (r)
3733 goto out;
3734 r = -EFAULT;
f0d66275 3735 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 3736 goto out;
3737 r = 0;
3738 break;
3739 }
3740 case KVM_SET_PIT: {
e0f63cb9 3741 r = -EFAULT;
f0d66275 3742 if (copy_from_user(&u.ps, argp, sizeof u.ps))
e0f63cb9
SY		 3743 goto out;
3744 r = -ENXIO;
3745 if (!kvm->arch.vpit)
3746 goto out;
f0d66275 3747 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
e0f63cb9
SY		 3748 break;
3749 }
e9f42757
BK		 3750 case KVM_GET_PIT2: {
3751 r = -ENXIO;
3752 if (!kvm->arch.vpit)
3753 goto out;
3754 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3755 if (r)
3756 goto out;
3757 r = -EFAULT;
3758 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3759 goto out;
3760 r = 0;
3761 break;
3762 }
3763 case KVM_SET_PIT2: {
3764 r = -EFAULT;
3765 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3766 goto out;
3767 r = -ENXIO;
3768 if (!kvm->arch.vpit)
3769 goto out;
3770 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
e9f42757
BK		 3771 break;
3772 }
52d939a0
MT		 3773 case KVM_REINJECT_CONTROL: {
3774 struct kvm_reinject_control control;
3775 r = -EFAULT;
3776 if (copy_from_user(&control, argp, sizeof(control)))
3777 goto out;
3778 r = kvm_vm_ioctl_reinject(kvm, &control);
52d939a0
MT		 3779 break;
3780 }
d71ba788
PB		 3781 case KVM_SET_BOOT_CPU_ID:
3782 r = 0;
3783 mutex_lock(&kvm->lock);
3784 if (atomic_read(&kvm->online_vcpus) != 0)
3785 r = -EBUSY;
3786 else
3787 kvm->arch.bsp_vcpu_id = arg;
3788 mutex_unlock(&kvm->lock);
3789 break;
ffde22ac
ES		 3790 case KVM_XEN_HVM_CONFIG: {
3791 r = -EFAULT;
3792 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3793 sizeof(struct kvm_xen_hvm_config)))
3794 goto out;
3795 r = -EINVAL;
3796 if (kvm->arch.xen_hvm_config.flags)
3797 goto out;
3798 r = 0;
3799 break;
3800 }
afbcf7ab 3801 case KVM_SET_CLOCK: {
afbcf7ab
GC		 3802 struct kvm_clock_data user_ns;
3803 u64 now_ns;
3804 s64 delta;
3805
3806 r = -EFAULT;
3807 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3808 goto out;
3809
3810 r = -EINVAL;
3811 if (user_ns.flags)
3812 goto out;
3813
3814 r = 0;
395c6b0a 3815 local_irq_disable();
759379dd 3816 now_ns = get_kernel_ns();
afbcf7ab 3817 delta = user_ns.clock - now_ns;
395c6b0a 3818 local_irq_enable();
afbcf7ab 3819 kvm->arch.kvmclock_offset = delta;
2e762ff7 3820 kvm_gen_update_masterclock(kvm);
afbcf7ab
GC		 3821 break;
3822 }
3823 case KVM_GET_CLOCK: {
afbcf7ab
GC		 3824 struct kvm_clock_data user_ns;
3825 u64 now_ns;
3826
395c6b0a 3827 local_irq_disable();
759379dd 3828 now_ns = get_kernel_ns();
afbcf7ab 3829 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
395c6b0a 3830 local_irq_enable();
afbcf7ab 3831 user_ns.flags = 0;
97e69aa6 3832 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
afbcf7ab
GC		 3833
3834 r = -EFAULT;
3835 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3836 goto out;
3837 r = 0;
3838 break;
3839 }
90de4a18
NA		 3840 case KVM_ENABLE_CAP: {
3841 struct kvm_enable_cap cap;
afbcf7ab 3842
90de4a18
NA		 3843 r = -EFAULT;
3844 if (copy_from_user(&cap, argp, sizeof(cap)))
3845 goto out;
3846 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
3847 break;
3848 }
1fe779f8 3849 default:
c274e03a 3850 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1fe779f8
CO		 3851 }
3852out:
3853 return r;
3854}
3855
a16b043c 3856static void kvm_init_msr_list(void)
043405e1
CO		 3857{
3858 u32 dummy[2];
3859 unsigned i, j;
3860
62ef68bb 3861 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
043405e1
CO		 3862 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3863 continue;
93c4adc7
PB		 3864
3865 /*
3866 * Even MSRs that are valid in the host may not be exposed
3867 * to the guests in some cases. We could work around this
3868 * in VMX with the generic MSR save/load machinery, but it
3869 * is not really worthwhile since it will really only
3870 * happen with nested virtualization.
3871 */
3872 switch (msrs_to_save[i]) {
3873 case MSR_IA32_BNDCFGS:
3874 if (!kvm_x86_ops->mpx_supported())
3875 continue;
3876 break;
3877 default:
3878 break;
3879 }
3880
043405e1
CO		 3881 if (j < i)
3882 msrs_to_save[j] = msrs_to_save[i];
3883 j++;
3884 }
3885 num_msrs_to_save = j;
62ef68bb
PB		 3886
3887 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
3888 switch (emulated_msrs[i]) {
6d396b55
PB		 3889 case MSR_IA32_SMBASE:
3890 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
3891 continue;
3892 break;
62ef68bb
PB		 3893 default:
3894 break;
3895 }
3896
3897 if (j < i)
3898 emulated_msrs[j] = emulated_msrs[i];
3899 j++;
3900 }
3901 num_emulated_msrs = j;
043405e1
CO		 3902}
3903
bda9020e
MT		 3904static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3905 const void *v)
bbd9b64e 3906{
70252a10
AK		 3907 int handled = 0;
3908 int n;
3909
3910 do {
3911 n = min(len, 8);
3912 if (!(vcpu->arch.apic &&
e32edf4f
NN		 3913 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
3914 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10
AK		 3915 break;
3916 handled += n;
3917 addr += n;
3918 len -= n;
3919 v += n;
3920 } while (len);
bbd9b64e 3921
70252a10 3922 return handled;
bbd9b64e
CO		 3923}
3924
bda9020e 3925static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
bbd9b64e 3926{
70252a10
AK		 3927 int handled = 0;
3928 int n;
3929
3930 do {
3931 n = min(len, 8);
3932 if (!(vcpu->arch.apic &&
e32edf4f
NN		 3933 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
3934 addr, n, v))
3935 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10
AK		 3936 break;
3937 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3938 handled += n;
3939 addr += n;
3940 len -= n;
3941 v += n;
3942 } while (len);
bbd9b64e 3943
70252a10 3944 return handled;
bbd9b64e
CO		 3945}
3946
2dafc6c2
GN		 3947static void kvm_set_segment(struct kvm_vcpu *vcpu,
3948 struct kvm_segment *var, int seg)
3949{
3950 kvm_x86_ops->set_segment(vcpu, var, seg);
3951}
3952
3953void kvm_get_segment(struct kvm_vcpu *vcpu,
3954 struct kvm_segment *var, int seg)
3955{
3956 kvm_x86_ops->get_segment(vcpu, var, seg);
3957}
3958
54987b7a
PB		 3959gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
3960 struct x86_exception *exception)
02f59dc9
JR		 3961{
3962 gpa_t t_gpa;
02f59dc9
JR		 3963
3964 BUG_ON(!mmu_is_nested(vcpu));
3965
3966 /* NPT walks are always user-walks */
3967 access |= PFERR_USER_MASK;
54987b7a 3968 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
02f59dc9
JR		 3969
3970 return t_gpa;
3971}
3972
ab9ae313
AK		 3973gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3974 struct x86_exception *exception)
1871c602
GN		 3975{
3976 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
ab9ae313 3977 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 3978}
3979
ab9ae313
AK		 3980 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3981 struct x86_exception *exception)
1871c602
GN		 3982{
3983 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3984 access |= PFERR_FETCH_MASK;
ab9ae313 3985 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 3986}
3987
ab9ae313
AK		 3988gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3989 struct x86_exception *exception)
1871c602
GN		 3990{
3991 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3992 access |= PFERR_WRITE_MASK;
ab9ae313 3993 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 3994}
3995
3996/* uses this to access any guest's mapped memory without checking CPL */
ab9ae313
AK		 3997gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3998 struct x86_exception *exception)
1871c602 3999{
ab9ae313 4000 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
1871c602
GN		 4001}
4002
4003static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4004 struct kvm_vcpu *vcpu, u32 access,
bcc55cba 4005 struct x86_exception *exception)
bbd9b64e
CO		 4006{
4007 void *data = val;
10589a46 4008 int r = X86EMUL_CONTINUE;
bbd9b64e
CO		 4009
4010 while (bytes) {
14dfe855 4011 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
ab9ae313 4012 exception);
bbd9b64e 4013 unsigned offset = addr & (PAGE_SIZE-1);
77c2002e 4014 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
bbd9b64e
CO		 4015 int ret;
4016
bcc55cba 4017 if (gpa == UNMAPPED_GVA)
ab9ae313 4018 return X86EMUL_PROPAGATE_FAULT;
54bf36aa
PB		 4019 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4020 offset, toread);
10589a46 4021 if (ret < 0) {
c3cd7ffa 4022 r = X86EMUL_IO_NEEDED;
10589a46
MT		 4023 goto out;
4024 }
bbd9b64e 4025
77c2002e
IE		 4026 bytes -= toread;
4027 data += toread;
4028 addr += toread;
bbd9b64e 4029 }
10589a46 4030out:
10589a46 4031 return r;
bbd9b64e 4032}
77c2002e 4033
1871c602 4034/* used for instruction fetching */
0f65dd70
AK		 4035static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4036 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4037 struct x86_exception *exception)
1871c602 4038{
0f65dd70 4039 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4040 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
44583cba
PB		 4041 unsigned offset;
4042 int ret;
0f65dd70 4043
44583cba
PB		 4044 /* Inline kvm_read_guest_virt_helper for speed. */
4045 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4046 exception);
4047 if (unlikely(gpa == UNMAPPED_GVA))
4048 return X86EMUL_PROPAGATE_FAULT;
4049
4050 offset = addr & (PAGE_SIZE-1);
4051 if (WARN_ON(offset + bytes > PAGE_SIZE))
4052 bytes = (unsigned)PAGE_SIZE - offset;
54bf36aa
PB		 4053 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4054 offset, bytes);
44583cba
PB		 4055 if (unlikely(ret < 0))
4056 return X86EMUL_IO_NEEDED;
4057
4058 return X86EMUL_CONTINUE;
1871c602
GN		 4059}
4060
064aea77 4061int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
0f65dd70 4062 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4063 struct x86_exception *exception)
1871c602 4064{
0f65dd70 4065 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4066 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
0f65dd70 4067
1871c602 4068 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
bcc55cba 4069 exception);
1871c602 4070}
064aea77 4071EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
1871c602 4072
0f65dd70
AK		 4073static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4074 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4075 struct x86_exception *exception)
1871c602 4076{
0f65dd70 4077 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
bcc55cba 4078 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
1871c602
GN		 4079}
4080
6a4d7550 4081int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
0f65dd70 4082 gva_t addr, void *val,
2dafc6c2 4083 unsigned int bytes,
bcc55cba 4084 struct x86_exception *exception)
77c2002e 4085{
0f65dd70 4086 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
77c2002e
IE		 4087 void *data = val;
4088 int r = X86EMUL_CONTINUE;
4089
4090 while (bytes) {
14dfe855
JR		 4091 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4092 PFERR_WRITE_MASK,
ab9ae313 4093 exception);
77c2002e
IE		 4094 unsigned offset = addr & (PAGE_SIZE-1);
4095 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4096 int ret;
4097
bcc55cba 4098 if (gpa == UNMAPPED_GVA)
ab9ae313 4099 return X86EMUL_PROPAGATE_FAULT;
54bf36aa 4100 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
77c2002e 4101 if (ret < 0) {
c3cd7ffa 4102 r = X86EMUL_IO_NEEDED;
77c2002e
IE		 4103 goto out;
4104 }
4105
4106 bytes -= towrite;
4107 data += towrite;
4108 addr += towrite;
4109 }
4110out:
4111 return r;
4112}
6a4d7550 4113EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
77c2002e 4114
af7cc7d1
XG		 4115static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4116 gpa_t *gpa, struct x86_exception *exception,
4117 bool write)
4118{
97d64b78
AK		 4119 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4120 | (write ? PFERR_WRITE_MASK : 0);
af7cc7d1 4121
97d64b78 4122 if (vcpu_match_mmio_gva(vcpu, gva)
97ec8c06
FW		 4123 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4124 vcpu->arch.access, access)) {
bebb106a
XG		 4125 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4126 (gva & (PAGE_SIZE - 1));
4f022648 4127 trace_vcpu_match_mmio(gva, *gpa, write, false);
bebb106a
XG		 4128 return 1;
4129 }
4130
af7cc7d1
XG		 4131 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4132
4133 if (*gpa == UNMAPPED_GVA)
4134 return -1;
4135
4136 /* For APIC access vmexit */
4137 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4138 return 1;
4139
4f022648
XG		 4140 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4141 trace_vcpu_match_mmio(gva, *gpa, write, true);
bebb106a 4142 return 1;
4f022648 4143 }
bebb106a 4144
af7cc7d1
XG		 4145 return 0;
4146}
4147
3200f405 4148int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
bcc55cba 4149 const void *val, int bytes)
bbd9b64e
CO		 4150{
4151 int ret;
4152
54bf36aa 4153 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
9f811285 4154 if (ret < 0)
bbd9b64e 4155 return 0;
f57f2ef5 4156 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
bbd9b64e
CO		 4157 return 1;
4158}
4159
77d197b2
XG		 4160struct read_write_emulator_ops {
4161 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4162 int bytes);
4163 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4164 void *val, int bytes);
4165 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4166 int bytes, void *val);
4167 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4168 void *val, int bytes);
4169 bool write;
4170};
4171
4172static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4173{
4174 if (vcpu->mmio_read_completed) {
77d197b2 4175 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
f78146b0 4176 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
77d197b2
XG		 4177 vcpu->mmio_read_completed = 0;
4178 return 1;
4179 }
4180
4181 return 0;
4182}
4183
4184static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4185 void *val, int bytes)
4186{
54bf36aa 4187 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
77d197b2
XG		 4188}
4189
4190static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4191 void *val, int bytes)
4192{
4193 return emulator_write_phys(vcpu, gpa, val, bytes);
4194}
4195
4196static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4197{
4198 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4199 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4200}
4201
4202static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4203 void *val, int bytes)
4204{
4205 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4206 return X86EMUL_IO_NEEDED;
4207}
4208
4209static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4210 void *val, int bytes)
4211{
f78146b0
AK		 4212 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4213
87da7e66 4214 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
77d197b2
XG		 4215 return X86EMUL_CONTINUE;
4216}
4217
0fbe9b0b 4218static const struct read_write_emulator_ops read_emultor = {
77d197b2
XG		 4219 .read_write_prepare = read_prepare,
4220 .read_write_emulate = read_emulate,
4221 .read_write_mmio = vcpu_mmio_read,
4222 .read_write_exit_mmio = read_exit_mmio,
4223};
4224
0fbe9b0b 4225static const struct read_write_emulator_ops write_emultor = {
77d197b2
XG		 4226 .read_write_emulate = write_emulate,
4227 .read_write_mmio = write_mmio,
4228 .read_write_exit_mmio = write_exit_mmio,
4229 .write = true,
4230};
4231
22388a3c
XG		 4232static int emulator_read_write_onepage(unsigned long addr, void *val,
4233 unsigned int bytes,
4234 struct x86_exception *exception,
4235 struct kvm_vcpu *vcpu,
0fbe9b0b 4236 const struct read_write_emulator_ops *ops)
bbd9b64e 4237{
af7cc7d1
XG		 4238 gpa_t gpa;
4239 int handled, ret;
22388a3c 4240 bool write = ops->write;
f78146b0 4241 struct kvm_mmio_fragment *frag;
10589a46 4242
22388a3c 4243 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
bbd9b64e 4244
af7cc7d1 4245 if (ret < 0)
bbd9b64e 4246 return X86EMUL_PROPAGATE_FAULT;
bbd9b64e
CO		 4247
4248 /* For APIC access vmexit */
af7cc7d1 4249 if (ret)
bbd9b64e
CO		 4250 goto mmio;
4251
22388a3c 4252 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
bbd9b64e
CO		 4253 return X86EMUL_CONTINUE;
4254
4255mmio:
4256 /*
4257 * Is this MMIO handled locally?
4258 */
22388a3c 4259 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
70252a10 4260 if (handled == bytes)
bbd9b64e 4261 return X86EMUL_CONTINUE;
bbd9b64e 4262
70252a10
AK		 4263 gpa += handled;
4264 bytes -= handled;
4265 val += handled;
4266
87da7e66
XG		 4267 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4268 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4269 frag->gpa = gpa;
4270 frag->data = val;
4271 frag->len = bytes;
f78146b0 4272 return X86EMUL_CONTINUE;
bbd9b64e
CO		 4273}
4274
52eb5a6d
XL		 4275static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4276 unsigned long addr,
22388a3c
XG		 4277 void *val, unsigned int bytes,
4278 struct x86_exception *exception,
0fbe9b0b 4279 const struct read_write_emulator_ops *ops)
bbd9b64e 4280{
0f65dd70 4281 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
f78146b0
AK		 4282 gpa_t gpa;
4283 int rc;
4284
4285 if (ops->read_write_prepare &&
4286 ops->read_write_prepare(vcpu, val, bytes))
4287 return X86EMUL_CONTINUE;
4288
4289 vcpu->mmio_nr_fragments = 0;
0f65dd70 4290
bbd9b64e
CO		 4291 /* Crossing a page boundary? */
4292 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
f78146b0 4293 int now;
bbd9b64e
CO		 4294
4295 now = -addr & ~PAGE_MASK;
22388a3c
XG		 4296 rc = emulator_read_write_onepage(addr, val, now, exception,
4297 vcpu, ops);
4298
bbd9b64e
CO		 4299 if (rc != X86EMUL_CONTINUE)
4300 return rc;
4301 addr += now;
bac15531
NA		 4302 if (ctxt->mode != X86EMUL_MODE_PROT64)
4303 addr = (u32)addr;
bbd9b64e
CO		 4304 val += now;
4305 bytes -= now;
4306 }
22388a3c 4307
f78146b0
AK		 4308 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4309 vcpu, ops);
4310 if (rc != X86EMUL_CONTINUE)
4311 return rc;
4312
4313 if (!vcpu->mmio_nr_fragments)
4314 return rc;
4315
4316 gpa = vcpu->mmio_fragments[0].gpa;
4317
4318 vcpu->mmio_needed = 1;
4319 vcpu->mmio_cur_fragment = 0;
4320
87da7e66 4321 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
f78146b0
AK		 4322 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4323 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4324 vcpu->run->mmio.phys_addr = gpa;
4325
4326 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
22388a3c
XG		 4327}
4328
4329static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4330 unsigned long addr,
4331 void *val,
4332 unsigned int bytes,
4333 struct x86_exception *exception)
4334{
4335 return emulator_read_write(ctxt, addr, val, bytes,
4336 exception, &read_emultor);
4337}
4338
52eb5a6d 4339static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
22388a3c
XG		 4340 unsigned long addr,
4341 const void *val,
4342 unsigned int bytes,
4343 struct x86_exception *exception)
4344{
4345 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4346 exception, &write_emultor);
bbd9b64e 4347}
bbd9b64e 4348
daea3e73
AK		 4349#define CMPXCHG_TYPE(t, ptr, old, new) \
4350 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4351
4352#ifdef CONFIG_X86_64
4353# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4354#else
4355# define CMPXCHG64(ptr, old, new) \
9749a6c0 4356 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
daea3e73
AK		 4357#endif
4358
0f65dd70
AK		 4359static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4360 unsigned long addr,
bbd9b64e
CO		 4361 const void *old,
4362 const void *new,
4363 unsigned int bytes,
0f65dd70 4364 struct x86_exception *exception)
bbd9b64e 4365{
0f65dd70 4366 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
daea3e73
AK		 4367 gpa_t gpa;
4368 struct page *page;
4369 char *kaddr;
4370 bool exchanged;
2bacc55c 4371
daea3e73
AK		 4372 /* guests cmpxchg8b have to be emulated atomically */
4373 if (bytes > 8 || (bytes & (bytes - 1)))
4374 goto emul_write;
10589a46 4375
daea3e73 4376 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
2bacc55c 4377
daea3e73
AK		 4378 if (gpa == UNMAPPED_GVA ||
4379 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4380 goto emul_write;
2bacc55c 4381
daea3e73
AK		 4382 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4383 goto emul_write;
72dc67a6 4384
54bf36aa 4385 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
32cad84f 4386 if (is_error_page(page))
c19b8bd6 4387 goto emul_write;
72dc67a6 4388
8fd75e12 4389 kaddr = kmap_atomic(page);
daea3e73
AK		 4390 kaddr += offset_in_page(gpa);
4391 switch (bytes) {
4392 case 1:
4393 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4394 break;
4395 case 2:
4396 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4397 break;
4398 case 4:
4399 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4400 break;
4401 case 8:
4402 exchanged = CMPXCHG64(kaddr, old, new);
4403 break;
4404 default:
4405 BUG();
2bacc55c 4406 }
8fd75e12 4407 kunmap_atomic(kaddr);
daea3e73
AK		 4408 kvm_release_page_dirty(page);
4409
4410 if (!exchanged)
4411 return X86EMUL_CMPXCHG_FAILED;
4412
54bf36aa 4413 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
f57f2ef5 4414 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
8f6abd06
GN		 4415
4416 return X86EMUL_CONTINUE;
4a5f48f6 4417
3200f405 4418emul_write:
daea3e73 4419 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2bacc55c 4420
0f65dd70 4421 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
bbd9b64e
CO		 4422}
4423
cf8f70bf
GN		 4424static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4425{
4426 /* TODO: String I/O for in kernel device */
4427 int r;
4428
4429 if (vcpu->arch.pio.in)
e32edf4f 4430 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
cf8f70bf
GN		 4431 vcpu->arch.pio.size, pd);
4432 else
e32edf4f 4433 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
cf8f70bf
GN		 4434 vcpu->arch.pio.port, vcpu->arch.pio.size,
4435 pd);
4436 return r;
4437}
4438
6f6fbe98
XG		 4439static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4440 unsigned short port, void *val,
4441 unsigned int count, bool in)
cf8f70bf 4442{
cf8f70bf 4443 vcpu->arch.pio.port = port;
6f6fbe98 4444 vcpu->arch.pio.in = in;
7972995b 4445 vcpu->arch.pio.count = count;
cf8f70bf
GN		 4446 vcpu->arch.pio.size = size;
4447
4448 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
7972995b 4449 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4450 return 1;
4451 }
4452
4453 vcpu->run->exit_reason = KVM_EXIT_IO;
6f6fbe98 4454 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
cf8f70bf
GN		 4455 vcpu->run->io.size = size;
4456 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4457 vcpu->run->io.count = count;
4458 vcpu->run->io.port = port;
4459
4460 return 0;
4461}
4462
6f6fbe98
XG		 4463static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4464 int size, unsigned short port, void *val,
4465 unsigned int count)
cf8f70bf 4466{
ca1d4a9e 4467 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6f6fbe98 4468 int ret;
ca1d4a9e 4469
6f6fbe98
XG		 4470 if (vcpu->arch.pio.count)
4471 goto data_avail;
cf8f70bf 4472
6f6fbe98
XG		 4473 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4474 if (ret) {
4475data_avail:
4476 memcpy(val, vcpu->arch.pio_data, size * count);
1171903d 4477 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
7972995b 4478 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4479 return 1;
4480 }
4481
cf8f70bf
GN		 4482 return 0;
4483}
4484
6f6fbe98
XG		 4485static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4486 int size, unsigned short port,
4487 const void *val, unsigned int count)
4488{
4489 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4490
4491 memcpy(vcpu->arch.pio_data, val, size * count);
1171903d 4492 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6f6fbe98
XG		 4493 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4494}
4495
bbd9b64e
CO		 4496static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4497{
4498 return kvm_x86_ops->get_segment_base(vcpu, seg);
4499}
4500
3cb16fe7 4501static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
bbd9b64e 4502{
3cb16fe7 4503 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
bbd9b64e
CO		 4504}
4505
5cb56059 4506int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
f5f48ee1
SY		 4507{
4508 if (!need_emulate_wbinvd(vcpu))
4509 return X86EMUL_CONTINUE;
4510
4511 if (kvm_x86_ops->has_wbinvd_exit()) {
2eec7343
JK		 4512 int cpu = get_cpu();
4513
4514 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
f5f48ee1
SY		 4515 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4516 wbinvd_ipi, NULL, 1);
2eec7343 4517 put_cpu();
f5f48ee1 4518 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
2eec7343
JK		 4519 } else
4520 wbinvd();
f5f48ee1
SY		 4521 return X86EMUL_CONTINUE;
4522}
5cb56059
JS		 4523
4524int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4525{
4526 kvm_x86_ops->skip_emulated_instruction(vcpu);
4527 return kvm_emulate_wbinvd_noskip(vcpu);
4528}
f5f48ee1
SY		 4529EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4530
5cb56059
JS		 4531
4532
bcaf5cc5
AK		 4533static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4534{
5cb56059 4535 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
bcaf5cc5
AK		 4536}
4537
52eb5a6d
XL		 4538static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4539 unsigned long *dest)
bbd9b64e 4540{
16f8a6f9 4541 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
bbd9b64e
CO		 4542}
4543
52eb5a6d
XL		 4544static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4545 unsigned long value)
bbd9b64e 4546{
338dbc97 4547
717746e3 4548 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
bbd9b64e
CO		 4549}
4550
52a46617 4551static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5fdbf976 4552{
52a46617 4553 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5fdbf976
MT		 4554}
4555
717746e3 4556static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
bbd9b64e 4557{
717746e3 4558 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
52a46617
GN		 4559 unsigned long value;
4560
4561 switch (cr) {
4562 case 0:
4563 value = kvm_read_cr0(vcpu);
4564 break;
4565 case 2:
4566 value = vcpu->arch.cr2;
4567 break;
4568 case 3:
9f8fe504 4569 value = kvm_read_cr3(vcpu);
52a46617
GN		 4570 break;
4571 case 4:
4572 value = kvm_read_cr4(vcpu);
4573 break;
4574 case 8:
4575 value = kvm_get_cr8(vcpu);
4576 break;
4577 default:
a737f256 4578 kvm_err("%s: unexpected cr %u\n", __func__, cr);
52a46617
GN		 4579 return 0;
4580 }
4581
4582 return value;
4583}
4584
717746e3 4585static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
52a46617 4586{
717746e3 4587 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
0f12244f
GN		 4588 int res = 0;
4589
52a46617
GN		 4590 switch (cr) {
4591 case 0:
49a9b07e 4592 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
52a46617
GN		 4593 break;
4594 case 2:
4595 vcpu->arch.cr2 = val;
4596 break;
4597 case 3:
2390218b 4598 res = kvm_set_cr3(vcpu, val);
52a46617
GN		 4599 break;
4600 case 4:
a83b29c6 4601 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
52a46617
GN		 4602 break;
4603 case 8:
eea1cff9 4604 res = kvm_set_cr8(vcpu, val);
52a46617
GN		 4605 break;
4606 default:
a737f256 4607 kvm_err("%s: unexpected cr %u\n", __func__, cr);
0f12244f 4608 res = -1;
52a46617 4609 }
0f12244f
GN		 4610
4611 return res;
52a46617
GN		 4612}
4613
717746e3 4614static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
9c537244 4615{
717746e3 4616 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
9c537244
GN		 4617}
4618
4bff1e86 4619static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
2dafc6c2 4620{
4bff1e86 4621 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
2dafc6c2
GN		 4622}
4623
4bff1e86 4624static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
160ce1f1 4625{
4bff1e86 4626 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
160ce1f1
MG		 4627}
4628
1ac9d0cf
AK		 4629static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4630{
4631 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4632}
4633
4634static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4635{
4636 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4637}
4638
4bff1e86
AK		 4639static unsigned long emulator_get_cached_segment_base(
4640 struct x86_emulate_ctxt *ctxt, int seg)
5951c442 4641{
4bff1e86 4642 return get_segment_base(emul_to_vcpu(ctxt), seg);
5951c442
GN		 4643}
4644
1aa36616
AK		 4645static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4646 struct desc_struct *desc, u32 *base3,
4647 int seg)
2dafc6c2
GN		 4648{
4649 struct kvm_segment var;
4650
4bff1e86 4651 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
1aa36616 4652 *selector = var.selector;
2dafc6c2 4653
378a8b09
GN		 4654 if (var.unusable) {
4655 memset(desc, 0, sizeof(*desc));
2dafc6c2 4656 return false;
378a8b09 4657 }
2dafc6c2
GN		 4658
4659 if (var.g)
4660 var.limit >>= 12;
4661 set_desc_limit(desc, var.limit);
4662 set_desc_base(desc, (unsigned long)var.base);
5601d05b
GN		 4663#ifdef CONFIG_X86_64
4664 if (base3)
4665 *base3 = var.base >> 32;
4666#endif
2dafc6c2
GN		 4667 desc->type = var.type;
4668 desc->s = var.s;
4669 desc->dpl = var.dpl;
4670 desc->p = var.present;
4671 desc->avl = var.avl;
4672 desc->l = var.l;
4673 desc->d = var.db;
4674 desc->g = var.g;
4675
4676 return true;
4677}
4678
1aa36616
AK		 4679static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4680 struct desc_struct *desc, u32 base3,
4681 int seg)
2dafc6c2 4682{
4bff1e86 4683 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
2dafc6c2
GN		 4684 struct kvm_segment var;
4685
1aa36616 4686 var.selector = selector;
2dafc6c2 4687 var.base = get_desc_base(desc);
5601d05b
GN		 4688#ifdef CONFIG_X86_64
4689 var.base |= ((u64)base3) << 32;
4690#endif
2dafc6c2
GN		 4691 var.limit = get_desc_limit(desc);
4692 if (desc->g)
4693 var.limit = (var.limit << 12) | 0xfff;
4694 var.type = desc->type;
2dafc6c2
GN		 4695 var.dpl = desc->dpl;
4696 var.db = desc->d;
4697 var.s = desc->s;
4698 var.l = desc->l;
4699 var.g = desc->g;
4700 var.avl = desc->avl;
4701 var.present = desc->p;
4702 var.unusable = !var.present;
4703 var.padding = 0;
4704
4705 kvm_set_segment(vcpu, &var, seg);
4706 return;
4707}
4708
717746e3
AK		 4709static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4710 u32 msr_index, u64 *pdata)
4711{
609e36d3
PB		 4712 struct msr_data msr;
4713 int r;
4714
4715 msr.index = msr_index;
4716 msr.host_initiated = false;
4717 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4718 if (r)
4719 return r;
4720
4721 *pdata = msr.data;
4722 return 0;
717746e3
AK		 4723}
4724
4725static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4726 u32 msr_index, u64 data)
4727{
8fe8ab46
WA		 4728 struct msr_data msr;
4729
4730 msr.data = data;
4731 msr.index = msr_index;
4732 msr.host_initiated = false;
4733 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
717746e3
AK		 4734}
4735
64d60670
PB		 4736static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4737{
4738 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4739
4740 return vcpu->arch.smbase;
4741}
4742
4743static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4744{
4745 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4746
4747 vcpu->arch.smbase = smbase;
4748}
4749
67f4d428
NA		 4750static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4751 u32 pmc)
4752{
c6702c9d 4753 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
67f4d428
NA		 4754}
4755
222d21aa
AK		 4756static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4757 u32 pmc, u64 *pdata)
4758{
c6702c9d 4759 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
222d21aa
AK		 4760}
4761
6c3287f7
AK		 4762static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4763{
4764 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4765}
4766
5037f6f3
AK		 4767static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4768{
4769 preempt_disable();
5197b808 4770 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5037f6f3
AK		 4771 /*
4772 * CR0.TS may reference the host fpu state, not the guest fpu state,
4773 * so it may be clear at this point.
4774 */
4775 clts();
4776}
4777
4778static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4779{
4780 preempt_enable();
4781}
4782
2953538e 4783static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
8a76d7f2 4784 struct x86_instruction_info *info,
c4f035c6
AK		 4785 enum x86_intercept_stage stage)
4786{
2953538e 4787 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
c4f035c6
AK		 4788}
4789
0017f93a 4790static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
bdb42f5a
SB		 4791 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4792{
0017f93a 4793 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
bdb42f5a
SB		 4794}
4795
dd856efa
AK		 4796static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4797{
4798 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4799}
4800
4801static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4802{
4803 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4804}
4805
801806d9
NA		 4806static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
4807{
4808 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
4809}
4810
0225fb50 4811static const struct x86_emulate_ops emulate_ops = {
dd856efa
AK		 4812 .read_gpr = emulator_read_gpr,
4813 .write_gpr = emulator_write_gpr,
1871c602 4814 .read_std = kvm_read_guest_virt_system,
2dafc6c2 4815 .write_std = kvm_write_guest_virt_system,
1871c602 4816 .fetch = kvm_fetch_guest_virt,
bbd9b64e
CO		 4817 .read_emulated = emulator_read_emulated,
4818 .write_emulated = emulator_write_emulated,
4819 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3cb16fe7 4820 .invlpg = emulator_invlpg,
cf8f70bf
GN		 4821 .pio_in_emulated = emulator_pio_in_emulated,
4822 .pio_out_emulated = emulator_pio_out_emulated,
1aa36616
AK		 4823 .get_segment = emulator_get_segment,
4824 .set_segment = emulator_set_segment,
5951c442 4825 .get_cached_segment_base = emulator_get_cached_segment_base,
2dafc6c2 4826 .get_gdt = emulator_get_gdt,
160ce1f1 4827 .get_idt = emulator_get_idt,
1ac9d0cf
AK		 4828 .set_gdt = emulator_set_gdt,
4829 .set_idt = emulator_set_idt,
52a46617
GN		 4830 .get_cr = emulator_get_cr,
4831 .set_cr = emulator_set_cr,
9c537244 4832 .cpl = emulator_get_cpl,
35aa5375
GN		 4833 .get_dr = emulator_get_dr,
4834 .set_dr = emulator_set_dr,
64d60670
PB		 4835 .get_smbase = emulator_get_smbase,
4836 .set_smbase = emulator_set_smbase,
717746e3
AK		 4837 .set_msr = emulator_set_msr,
4838 .get_msr = emulator_get_msr,
67f4d428 4839 .check_pmc = emulator_check_pmc,
222d21aa 4840 .read_pmc = emulator_read_pmc,
6c3287f7 4841 .halt = emulator_halt,
bcaf5cc5 4842 .wbinvd = emulator_wbinvd,
d6aa1000 4843 .fix_hypercall = emulator_fix_hypercall,
5037f6f3
AK		 4844 .get_fpu = emulator_get_fpu,
4845 .put_fpu = emulator_put_fpu,
c4f035c6 4846 .intercept = emulator_intercept,
bdb42f5a 4847 .get_cpuid = emulator_get_cpuid,
801806d9 4848 .set_nmi_mask = emulator_set_nmi_mask,
bbd9b64e
CO		 4849};
4850
95cb2295
GN		 4851static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4852{
37ccdcbe 4853 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
95cb2295
GN		 4854 /*
4855 * an sti; sti; sequence only disable interrupts for the first
4856 * instruction. So, if the last instruction, be it emulated or
4857 * not, left the system with the INT_STI flag enabled, it
4858 * means that the last instruction is an sti. We should not
4859 * leave the flag on in this case. The same goes for mov ss
4860 */
37ccdcbe
PB		 4861 if (int_shadow & mask)
4862 mask = 0;
6addfc42 4863 if (unlikely(int_shadow || mask)) {
95cb2295 4864 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6addfc42
PB		 4865 if (!mask)
4866 kvm_make_request(KVM_REQ_EVENT, vcpu);
4867 }
95cb2295
GN		 4868}
4869
ef54bcfe 4870static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
54b8486f
GN		 4871{
4872 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
da9cb575 4873 if (ctxt->exception.vector == PF_VECTOR)
ef54bcfe
PB		 4874 return kvm_propagate_fault(vcpu, &ctxt->exception);
4875
4876 if (ctxt->exception.error_code_valid)
da9cb575
AK		 4877 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4878 ctxt->exception.error_code);
54b8486f 4879 else
da9cb575 4880 kvm_queue_exception(vcpu, ctxt->exception.vector);
ef54bcfe 4881 return false;
54b8486f
GN		 4882}
4883
8ec4722d
MG		 4884static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4885{
adf52235 4886 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d
MG		 4887 int cs_db, cs_l;
4888
8ec4722d
MG		 4889 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4890
adf52235
TY		 4891 ctxt->eflags = kvm_get_rflags(vcpu);
4892 ctxt->eip = kvm_rip_read(vcpu);
4893 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4894 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
42bf549f 4895 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
adf52235
TY		 4896 cs_db ? X86EMUL_MODE_PROT32 :
4897 X86EMUL_MODE_PROT16;
a584539b 4898 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
64d60670
PB		 4899 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
4900 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
a584539b 4901 ctxt->emul_flags = vcpu->arch.hflags;
adf52235 4902
dd856efa 4903 init_decode_cache(ctxt);
7ae441ea 4904 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8ec4722d
MG		 4905}
4906
71f9833b 4907int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
63995653 4908{
9d74191a 4909 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
63995653
MG		 4910 int ret;
4911
4912 init_emulate_ctxt(vcpu);
4913
9dac77fa
AK		 4914 ctxt->op_bytes = 2;
4915 ctxt->ad_bytes = 2;
4916 ctxt->_eip = ctxt->eip + inc_eip;
9d74191a 4917 ret = emulate_int_real(ctxt, irq);
63995653
MG		 4918
4919 if (ret != X86EMUL_CONTINUE)
4920 return EMULATE_FAIL;
4921
9dac77fa 4922 ctxt->eip = ctxt->_eip;
9d74191a
TY		 4923 kvm_rip_write(vcpu, ctxt->eip);
4924 kvm_set_rflags(vcpu, ctxt->eflags);
63995653
MG		 4925
4926 if (irq == NMI_VECTOR)
7460fb4a 4927 vcpu->arch.nmi_pending = 0;
63995653
MG		 4928 else
4929 vcpu->arch.interrupt.pending = false;
4930
4931 return EMULATE_DONE;
4932}
4933EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4934
6d77dbfc
GN		 4935static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4936{
fc3a9157
JR		 4937 int r = EMULATE_DONE;
4938
6d77dbfc
GN		 4939 ++vcpu->stat.insn_emulation_fail;
4940 trace_kvm_emulate_insn_failed(vcpu);
a2b9e6c1 4941 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
fc3a9157
JR		 4942 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4943 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4944 vcpu->run->internal.ndata = 0;
4945 r = EMULATE_FAIL;
4946 }
6d77dbfc 4947 kvm_queue_exception(vcpu, UD_VECTOR);
fc3a9157
JR		 4948
4949 return r;
6d77dbfc
GN		 4950}
4951
93c05d3e 4952static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
991eebf9
GN		 4953 bool write_fault_to_shadow_pgtable,
4954 int emulation_type)
a6f177ef 4955{
95b3cf69 4956 gpa_t gpa = cr2;
8e3d9d06 4957 pfn_t pfn;
a6f177ef 4958
991eebf9
GN		 4959 if (emulation_type & EMULTYPE_NO_REEXECUTE)
4960 return false;
4961
95b3cf69
XG		 4962 if (!vcpu->arch.mmu.direct_map) {
4963 /*
4964 * Write permission should be allowed since only
4965 * write access need to be emulated.
4966 */
4967 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
a6f177ef 4968
95b3cf69
XG		 4969 /*
4970 * If the mapping is invalid in guest, let cpu retry
4971 * it to generate fault.
4972 */
4973 if (gpa == UNMAPPED_GVA)
4974 return true;
4975 }
a6f177ef 4976
8e3d9d06
XG		 4977 /*
4978 * Do not retry the unhandleable instruction if it faults on the
4979 * readonly host memory, otherwise it will goto a infinite loop:
4980 * retry instruction -> write #PF -> emulation fail -> retry
4981 * instruction -> ...
4982 */
4983 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
95b3cf69
XG		 4984
4985 /*
4986 * If the instruction failed on the error pfn, it can not be fixed,
4987 * report the error to userspace.
4988 */
4989 if (is_error_noslot_pfn(pfn))
4990 return false;
4991
4992 kvm_release_pfn_clean(pfn);
4993
4994 /* The instructions are well-emulated on direct mmu. */
4995 if (vcpu->arch.mmu.direct_map) {
4996 unsigned int indirect_shadow_pages;
4997
4998 spin_lock(&vcpu->kvm->mmu_lock);
4999 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5000 spin_unlock(&vcpu->kvm->mmu_lock);
5001
5002 if (indirect_shadow_pages)
5003 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5004
a6f177ef 5005 return true;
8e3d9d06 5006 }
a6f177ef 5007
95b3cf69
XG		 5008 /*
5009 * if emulation was due to access to shadowed page table
5010 * and it failed try to unshadow page and re-enter the
5011 * guest to let CPU execute the instruction.
5012 */
5013 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
93c05d3e
XG		 5014
5015 /*
5016 * If the access faults on its page table, it can not
5017 * be fixed by unprotecting shadow page and it should
5018 * be reported to userspace.
5019 */
5020 return !write_fault_to_shadow_pgtable;
a6f177ef
GN		 5021}
5022
1cb3f3ae
XG		 5023static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5024 unsigned long cr2, int emulation_type)
5025{
5026 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5027 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5028
5029 last_retry_eip = vcpu->arch.last_retry_eip;
5030 last_retry_addr = vcpu->arch.last_retry_addr;
5031
5032 /*
5033 * If the emulation is caused by #PF and it is non-page_table
5034 * writing instruction, it means the VM-EXIT is caused by shadow
5035 * page protected, we can zap the shadow page and retry this
5036 * instruction directly.
5037 *
5038 * Note: if the guest uses a non-page-table modifying instruction
5039 * on the PDE that points to the instruction, then we will unmap
5040 * the instruction and go to an infinite loop. So, we cache the
5041 * last retried eip and the last fault address, if we meet the eip
5042 * and the address again, we can break out of the potential infinite
5043 * loop.
5044 */
5045 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5046
5047 if (!(emulation_type & EMULTYPE_RETRY))
5048 return false;
5049
5050 if (x86_page_table_writing_insn(ctxt))
5051 return false;
5052
5053 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5054 return false;
5055
5056 vcpu->arch.last_retry_eip = ctxt->eip;
5057 vcpu->arch.last_retry_addr = cr2;
5058
5059 if (!vcpu->arch.mmu.direct_map)
5060 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5061
22368028 5062 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
1cb3f3ae
XG		 5063
5064 return true;
5065}
5066
716d51ab
GN		 5067static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5068static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5069
64d60670 5070static void kvm_smm_changed(struct kvm_vcpu *vcpu)
a584539b 5071{
64d60670 5072 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
660a5d51
PB		 5073 /* This is a good place to trace that we are exiting SMM. */
5074 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5075
64d60670
PB		 5076 if (unlikely(vcpu->arch.smi_pending)) {
5077 kvm_make_request(KVM_REQ_SMI, vcpu);
5078 vcpu->arch.smi_pending = 0;
cd7764fe
PB		 5079 } else {
5080 /* Process a latched INIT, if any. */
5081 kvm_make_request(KVM_REQ_EVENT, vcpu);
64d60670
PB		 5082 }
5083 }
699023e2
PB		 5084
5085 kvm_mmu_reset_context(vcpu);
64d60670
PB		 5086}
5087
5088static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5089{
5090 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5091
a584539b 5092 vcpu->arch.hflags = emul_flags;
64d60670
PB		 5093
5094 if (changed & HF_SMM_MASK)
5095 kvm_smm_changed(vcpu);
a584539b
PB		 5096}
5097
4a1e10d5
PB		 5098static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5099 unsigned long *db)
5100{
5101 u32 dr6 = 0;
5102 int i;
5103 u32 enable, rwlen;
5104
5105 enable = dr7;
5106 rwlen = dr7 >> 16;
5107 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5108 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5109 dr6 |= (1 << i);
5110 return dr6;
5111}
5112
6addfc42 5113static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
663f4c61
PB		 5114{
5115 struct kvm_run *kvm_run = vcpu->run;
5116
5117 /*
6addfc42
PB		 5118 * rflags is the old, "raw" value of the flags. The new value has
5119 * not been saved yet.
663f4c61
PB		 5120 *
5121 * This is correct even for TF set by the guest, because "the
5122 * processor will not generate this exception after the instruction
5123 * that sets the TF flag".
5124 */
663f4c61
PB		 5125 if (unlikely(rflags & X86_EFLAGS_TF)) {
5126 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6f43ed01
NA		 5127 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5128 DR6_RTM;
663f4c61
PB		 5129 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5130 kvm_run->debug.arch.exception = DB_VECTOR;
5131 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5132 *r = EMULATE_USER_EXIT;
5133 } else {
5134 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5135 /*
5136 * "Certain debug exceptions may clear bit 0-3. The
5137 * remaining contents of the DR6 register are never
5138 * cleared by the processor".
5139 */
5140 vcpu->arch.dr6 &= ~15;
6f43ed01 5141 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
663f4c61
PB		 5142 kvm_queue_exception(vcpu, DB_VECTOR);
5143 }
5144 }
5145}
5146
4a1e10d5
PB		 5147static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5148{
4a1e10d5
PB		 5149 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5150 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
82b32774
NA		 5151 struct kvm_run *kvm_run = vcpu->run;
5152 unsigned long eip = kvm_get_linear_rip(vcpu);
5153 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5154 vcpu->arch.guest_debug_dr7,
5155 vcpu->arch.eff_db);
5156
5157 if (dr6 != 0) {
6f43ed01 5158 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
82b32774 5159 kvm_run->debug.arch.pc = eip;
4a1e10d5
PB		 5160 kvm_run->debug.arch.exception = DB_VECTOR;
5161 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5162 *r = EMULATE_USER_EXIT;
5163 return true;
5164 }
5165 }
5166
4161a569
NA		 5167 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5168 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
82b32774
NA		 5169 unsigned long eip = kvm_get_linear_rip(vcpu);
5170 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5171 vcpu->arch.dr7,
5172 vcpu->arch.db);
5173
5174 if (dr6 != 0) {
5175 vcpu->arch.dr6 &= ~15;
6f43ed01 5176 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4a1e10d5
PB		 5177 kvm_queue_exception(vcpu, DB_VECTOR);
5178 *r = EMULATE_DONE;
5179 return true;
5180 }
5181 }
5182
5183 return false;
5184}
5185
51d8b661
AP		 5186int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5187 unsigned long cr2,
dc25e89e
AP		 5188 int emulation_type,
5189 void *insn,
5190 int insn_len)
bbd9b64e 5191{
95cb2295 5192 int r;
9d74191a 5193 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7ae441ea 5194 bool writeback = true;
93c05d3e 5195 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
bbd9b64e 5196
93c05d3e
XG		 5197 /*
5198 * Clear write_fault_to_shadow_pgtable here to ensure it is
5199 * never reused.
5200 */
5201 vcpu->arch.write_fault_to_shadow_pgtable = false;
26eef70c 5202 kvm_clear_exception_queue(vcpu);
8d7d8102 5203
571008da 5204 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8ec4722d 5205 init_emulate_ctxt(vcpu);
4a1e10d5
PB		 5206
5207 /*
5208 * We will reenter on the same instruction since
5209 * we do not set complete_userspace_io. This does not
5210 * handle watchpoints yet, those would be handled in
5211 * the emulate_ops.
5212 */
5213 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5214 return r;
5215
9d74191a
TY		 5216 ctxt->interruptibility = 0;
5217 ctxt->have_exception = false;
e0ad0b47 5218 ctxt->exception.vector = -1;
9d74191a 5219 ctxt->perm_ok = false;
bbd9b64e 5220
b51e974f 5221 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
4005996e 5222
9d74191a 5223 r = x86_decode_insn(ctxt, insn, insn_len);
bbd9b64e 5224
e46479f8 5225 trace_kvm_emulate_insn_start(vcpu);
f2b5756b 5226 ++vcpu->stat.insn_emulation;
1d2887e2 5227 if (r != EMULATION_OK) {
4005996e
AK		 5228 if (emulation_type & EMULTYPE_TRAP_UD)
5229 return EMULATE_FAIL;
991eebf9
GN		 5230 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5231 emulation_type))
bbd9b64e 5232 return EMULATE_DONE;
6d77dbfc
GN		 5233 if (emulation_type & EMULTYPE_SKIP)
5234 return EMULATE_FAIL;
5235 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 5236 }
5237 }
5238
ba8afb6b 5239 if (emulation_type & EMULTYPE_SKIP) {
9dac77fa 5240 kvm_rip_write(vcpu, ctxt->_eip);
bb663c7a
NA		 5241 if (ctxt->eflags & X86_EFLAGS_RF)
5242 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
ba8afb6b
GN		 5243 return EMULATE_DONE;
5244 }
5245
1cb3f3ae
XG		 5246 if (retry_instruction(ctxt, cr2, emulation_type))
5247 return EMULATE_DONE;
5248
7ae441ea 5249 /* this is needed for vmware backdoor interface to work since it
4d2179e1 5250 changes registers values during IO operation */
7ae441ea
GN		 5251 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5252 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
dd856efa 5253 emulator_invalidate_register_cache(ctxt);
7ae441ea 5254 }
4d2179e1 5255
5cd21917 5256restart:
9d74191a 5257 r = x86_emulate_insn(ctxt);
bbd9b64e 5258
775fde86
JR		 5259 if (r == EMULATION_INTERCEPTED)
5260 return EMULATE_DONE;
5261
d2ddd1c4 5262 if (r == EMULATION_FAILED) {
991eebf9
GN		 5263 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5264 emulation_type))
c3cd7ffa
GN		 5265 return EMULATE_DONE;
5266
6d77dbfc 5267 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 5268 }
5269
9d74191a 5270 if (ctxt->have_exception) {
d2ddd1c4 5271 r = EMULATE_DONE;
ef54bcfe
PB		 5272 if (inject_emulated_exception(vcpu))
5273 return r;
d2ddd1c4 5274 } else if (vcpu->arch.pio.count) {
0912c977
PB		 5275 if (!vcpu->arch.pio.in) {
5276 /* FIXME: return into emulator if single-stepping. */
3457e419 5277 vcpu->arch.pio.count = 0;
0912c977 5278 } else {
7ae441ea 5279 writeback = false;
716d51ab
GN		 5280 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5281 }
ac0a48c3 5282 r = EMULATE_USER_EXIT;
7ae441ea
GN		 5283 } else if (vcpu->mmio_needed) {
5284 if (!vcpu->mmio_is_write)
5285 writeback = false;
ac0a48c3 5286 r = EMULATE_USER_EXIT;
716d51ab 5287 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7ae441ea 5288 } else if (r == EMULATION_RESTART)
5cd21917 5289 goto restart;
d2ddd1c4
GN		 5290 else
5291 r = EMULATE_DONE;
f850e2e6 5292
7ae441ea 5293 if (writeback) {
6addfc42 5294 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
9d74191a 5295 toggle_interruptibility(vcpu, ctxt->interruptibility);
7ae441ea 5296 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
a584539b
PB		 5297 if (vcpu->arch.hflags != ctxt->emul_flags)
5298 kvm_set_hflags(vcpu, ctxt->emul_flags);
9d74191a 5299 kvm_rip_write(vcpu, ctxt->eip);
663f4c61 5300 if (r == EMULATE_DONE)
6addfc42 5301 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
38827dbd
NA		 5302 if (!ctxt->have_exception ||
5303 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5304 __kvm_set_rflags(vcpu, ctxt->eflags);
6addfc42
PB		 5305
5306 /*
5307 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5308 * do nothing, and it will be requested again as soon as
5309 * the shadow expires. But we still need to check here,
5310 * because POPF has no interrupt shadow.
5311 */
5312 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5313 kvm_make_request(KVM_REQ_EVENT, vcpu);
7ae441ea
GN		 5314 } else
5315 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
e85d28f8
GN		 5316
5317 return r;
de7d789a 5318}
51d8b661 5319EXPORT_SYMBOL_GPL(x86_emulate_instruction);
de7d789a 5320
cf8f70bf 5321int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
de7d789a 5322{
cf8f70bf 5323 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
ca1d4a9e
AK		 5324 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5325 size, port, &val, 1);
cf8f70bf 5326 /* do not return to emulator after return from userspace */
7972995b 5327 vcpu->arch.pio.count = 0;
de7d789a
CO		 5328 return ret;
5329}
cf8f70bf 5330EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
de7d789a 5331
8cfdc000
ZA		 5332static void tsc_bad(void *info)
5333{
0a3aee0d 5334 __this_cpu_write(cpu_tsc_khz, 0);
8cfdc000
ZA		 5335}
5336
5337static void tsc_khz_changed(void *data)
c8076604 5338{
8cfdc000
ZA		 5339 struct cpufreq_freqs *freq = data;
5340 unsigned long khz = 0;
5341
5342 if (data)
5343 khz = freq->new;
5344 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5345 khz = cpufreq_quick_get(raw_smp_processor_id());
5346 if (!khz)
5347 khz = tsc_khz;
0a3aee0d 5348 __this_cpu_write(cpu_tsc_khz, khz);
c8076604
GH		 5349}
5350
c8076604
GH		 5351static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5352 void *data)
5353{
5354 struct cpufreq_freqs *freq = data;
5355 struct kvm *kvm;
5356 struct kvm_vcpu *vcpu;
5357 int i, send_ipi = 0;
5358
8cfdc000
ZA		 5359 /*
5360 * We allow guests to temporarily run on slowing clocks,
5361 * provided we notify them after, or to run on accelerating
5362 * clocks, provided we notify them before. Thus time never
5363 * goes backwards.
5364 *
5365 * However, we have a problem. We can't atomically update
5366 * the frequency of a given CPU from this function; it is
5367 * merely a notifier, which can be called from any CPU.
5368 * Changing the TSC frequency at arbitrary points in time
5369 * requires a recomputation of local variables related to
5370 * the TSC for each VCPU. We must flag these local variables
5371 * to be updated and be sure the update takes place with the
5372 * new frequency before any guests proceed.
5373 *
5374 * Unfortunately, the combination of hotplug CPU and frequency
5375 * change creates an intractable locking scenario; the order
5376 * of when these callouts happen is undefined with respect to
5377 * CPU hotplug, and they can race with each other. As such,
5378 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5379 * undefined; you can actually have a CPU frequency change take
5380 * place in between the computation of X and the setting of the
5381 * variable. To protect against this problem, all updates of
5382 * the per_cpu tsc_khz variable are done in an interrupt
5383 * protected IPI, and all callers wishing to update the value
5384 * must wait for a synchronous IPI to complete (which is trivial
5385 * if the caller is on the CPU already). This establishes the
5386 * necessary total order on variable updates.
5387 *
5388 * Note that because a guest time update may take place
5389 * anytime after the setting of the VCPU's request bit, the
5390 * correct TSC value must be set before the request. However,
5391 * to ensure the update actually makes it to any guest which
5392 * starts running in hardware virtualization between the set
5393 * and the acquisition of the spinlock, we must also ping the
5394 * CPU after setting the request bit.
5395 *
5396 */
5397
c8076604
GH		 5398 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5399 return 0;
5400 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5401 return 0;
8cfdc000
ZA		 5402
5403 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604 5404
2f303b74 5405 spin_lock(&kvm_lock);
c8076604 5406 list_for_each_entry(kvm, &vm_list, vm_list) {
988a2cae 5407 kvm_for_each_vcpu(i, vcpu, kvm) {
c8076604
GH		 5408 if (vcpu->cpu != freq->cpu)
5409 continue;
c285545f 5410 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
c8076604 5411 if (vcpu->cpu != smp_processor_id())
8cfdc000 5412 send_ipi = 1;
c8076604
GH		 5413 }
5414 }
2f303b74 5415 spin_unlock(&kvm_lock);
c8076604
GH		 5416
5417 if (freq->old < freq->new && send_ipi) {
5418 /*
5419 * We upscale the frequency. Must make the guest
5420 * doesn't see old kvmclock values while running with
5421 * the new frequency, otherwise we risk the guest sees
5422 * time go backwards.
5423 *
5424 * In case we update the frequency for another cpu
5425 * (which might be in guest context) send an interrupt
5426 * to kick the cpu out of guest context. Next time
5427 * guest context is entered kvmclock will be updated,
5428 * so the guest will not see stale values.
5429 */
8cfdc000 5430 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604
GH		 5431 }
5432 return 0;
5433}
5434
5435static struct notifier_block kvmclock_cpufreq_notifier_block = {
8cfdc000
ZA		 5436 .notifier_call = kvmclock_cpufreq_notifier
5437};
5438
5439static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5440 unsigned long action, void *hcpu)
5441{
5442 unsigned int cpu = (unsigned long)hcpu;
5443
5444 switch (action) {
5445 case CPU_ONLINE:
5446 case CPU_DOWN_FAILED:
5447 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5448 break;
5449 case CPU_DOWN_PREPARE:
5450 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5451 break;
5452 }
5453 return NOTIFY_OK;
5454}
5455
5456static struct notifier_block kvmclock_cpu_notifier_block = {
5457 .notifier_call = kvmclock_cpu_notifier,
5458 .priority = -INT_MAX
c8076604
GH		 5459};
5460
b820cc0c
ZA		 5461static void kvm_timer_init(void)
5462{
5463 int cpu;
5464
c285545f 5465 max_tsc_khz = tsc_khz;
460dd42e
SB		 5466
5467 cpu_notifier_register_begin();
b820cc0c 5468 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
c285545f
ZA		 5469#ifdef CONFIG_CPU_FREQ
5470 struct cpufreq_policy policy;
5471 memset(&policy, 0, sizeof(policy));
3e26f230
AK		 5472 cpu = get_cpu();
5473 cpufreq_get_policy(&policy, cpu);
c285545f
ZA		 5474 if (policy.cpuinfo.max_freq)
5475 max_tsc_khz = policy.cpuinfo.max_freq;
3e26f230 5476 put_cpu();
c285545f 5477#endif
b820cc0c
ZA		 5478 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5479 CPUFREQ_TRANSITION_NOTIFIER);
5480 }
c285545f 5481 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
8cfdc000
ZA		 5482 for_each_online_cpu(cpu)
5483 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
460dd42e
SB		 5484
5485 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5486 cpu_notifier_register_done();
5487
b820cc0c
ZA		 5488}
5489
ff9d07a0
ZY		 5490static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5491
f5132b01 5492int kvm_is_in_guest(void)
ff9d07a0 5493{
086c9855 5494 return __this_cpu_read(current_vcpu) != NULL;
ff9d07a0
ZY		 5495}
5496
5497static int kvm_is_user_mode(void)
5498{
5499 int user_mode = 3;
dcf46b94 5500
086c9855
AS		 5501 if (__this_cpu_read(current_vcpu))
5502 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
dcf46b94 5503
ff9d07a0
ZY		 5504 return user_mode != 0;
5505}
5506
5507static unsigned long kvm_get_guest_ip(void)
5508{
5509 unsigned long ip = 0;
dcf46b94 5510
086c9855
AS		 5511 if (__this_cpu_read(current_vcpu))
5512 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
dcf46b94 5513
ff9d07a0
ZY		 5514 return ip;
5515}
5516
5517static struct perf_guest_info_callbacks kvm_guest_cbs = {
5518 .is_in_guest = kvm_is_in_guest,
5519 .is_user_mode = kvm_is_user_mode,
5520 .get_guest_ip = kvm_get_guest_ip,
5521};
5522
5523void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5524{
086c9855 5525 __this_cpu_write(current_vcpu, vcpu);
ff9d07a0
ZY		 5526}
5527EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5528
5529void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5530{
086c9855 5531 __this_cpu_write(current_vcpu, NULL);
ff9d07a0
ZY		 5532}
5533EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5534
ce88decf
XG		 5535static void kvm_set_mmio_spte_mask(void)
5536{
5537 u64 mask;
5538 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5539
5540 /*
5541 * Set the reserved bits and the present bit of an paging-structure
5542 * entry to generate page fault with PFER.RSV = 1.
5543 */
885032b9 5544 /* Mask the reserved physical address bits. */
d1431483 5545 mask = rsvd_bits(maxphyaddr, 51);
885032b9
XG		 5546
5547 /* Bit 62 is always reserved for 32bit host. */
5548 mask |= 0x3ull << 62;
5549
5550 /* Set the present bit. */
ce88decf
XG		 5551 mask |= 1ull;
5552
5553#ifdef CONFIG_X86_64
5554 /*
5555 * If reserved bit is not supported, clear the present bit to disable
5556 * mmio page fault.
5557 */
5558 if (maxphyaddr == 52)
5559 mask &= ~1ull;
5560#endif
5561
5562 kvm_mmu_set_mmio_spte_mask(mask);
5563}
5564
16e8d74d
MT		 5565#ifdef CONFIG_X86_64
5566static void pvclock_gtod_update_fn(struct work_struct *work)
5567{
d828199e
MT		 5568 struct kvm *kvm;
5569
5570 struct kvm_vcpu *vcpu;
5571 int i;
5572
2f303b74 5573 spin_lock(&kvm_lock);
d828199e
MT		 5574 list_for_each_entry(kvm, &vm_list, vm_list)
5575 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 5576 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
d828199e 5577 atomic_set(&kvm_guest_has_master_clock, 0);
2f303b74 5578 spin_unlock(&kvm_lock);
16e8d74d
MT		 5579}
5580
5581static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5582
5583/*
5584 * Notification about pvclock gtod data update.
5585 */
5586static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5587 void *priv)
5588{
5589 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5590 struct timekeeper *tk = priv;
5591
5592 update_pvclock_gtod(tk);
5593
5594 /* disable master clock if host does not trust, or does not
5595 * use, TSC clocksource
5596 */
5597 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5598 atomic_read(&kvm_guest_has_master_clock) != 0)
5599 queue_work(system_long_wq, &pvclock_gtod_work);
5600
5601 return 0;
5602}
5603
5604static struct notifier_block pvclock_gtod_notifier = {
5605 .notifier_call = pvclock_gtod_notify,
5606};
5607#endif
5608
f8c16bba 5609int kvm_arch_init(void *opaque)
043405e1 5610{
b820cc0c 5611 int r;
6b61edf7 5612 struct kvm_x86_ops *ops = opaque;
f8c16bba 5613
f8c16bba
ZX		 5614 if (kvm_x86_ops) {
5615 printk(KERN_ERR "kvm: already loaded the other module\n");
56c6d28a
ZX		 5616 r = -EEXIST;
5617 goto out;
f8c16bba
ZX		 5618 }
5619
5620 if (!ops->cpu_has_kvm_support()) {
5621 printk(KERN_ERR "kvm: no hardware support\n");
56c6d28a
ZX		 5622 r = -EOPNOTSUPP;
5623 goto out;
f8c16bba
ZX		 5624 }
5625 if (ops->disabled_by_bios()) {
5626 printk(KERN_ERR "kvm: disabled by bios\n");
56c6d28a
ZX		 5627 r = -EOPNOTSUPP;
5628 goto out;
f8c16bba
ZX		 5629 }
5630
013f6a5d
MT		 5631 r = -ENOMEM;
5632 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5633 if (!shared_msrs) {
5634 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5635 goto out;
5636 }
5637
97db56ce
AK		 5638 r = kvm_mmu_module_init();
5639 if (r)
013f6a5d 5640 goto out_free_percpu;
97db56ce 5641
ce88decf 5642 kvm_set_mmio_spte_mask();
97db56ce 5643
f8c16bba 5644 kvm_x86_ops = ops;
920c8377 5645
7b52345e 5646 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4b12f0de 5647 PT_DIRTY_MASK, PT64_NX_MASK, 0);
c8076604 5648
b820cc0c 5649 kvm_timer_init();
c8076604 5650
ff9d07a0
ZY		 5651 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5652
2acf923e
DC		 5653 if (cpu_has_xsave)
5654 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5655
c5cc421b 5656 kvm_lapic_init();
16e8d74d
MT		 5657#ifdef CONFIG_X86_64
5658 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5659#endif
5660
f8c16bba 5661 return 0;
56c6d28a 5662
013f6a5d
MT		 5663out_free_percpu:
5664 free_percpu(shared_msrs);
56c6d28a 5665out:
56c6d28a 5666 return r;
043405e1 5667}
8776e519 5668
f8c16bba
ZX		 5669void kvm_arch_exit(void)
5670{
ff9d07a0
ZY		 5671 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5672
888d256e
JK		 5673 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5674 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5675 CPUFREQ_TRANSITION_NOTIFIER);
8cfdc000 5676 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
16e8d74d
MT		 5677#ifdef CONFIG_X86_64
5678 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5679#endif
f8c16bba 5680 kvm_x86_ops = NULL;
56c6d28a 5681 kvm_mmu_module_exit();
013f6a5d 5682 free_percpu(shared_msrs);
56c6d28a 5683}
f8c16bba 5684
5cb56059 5685int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8776e519
HB		 5686{
5687 ++vcpu->stat.halt_exits;
5688 if (irqchip_in_kernel(vcpu->kvm)) {
a4535290 5689 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
8776e519
HB		 5690 return 1;
5691 } else {
5692 vcpu->run->exit_reason = KVM_EXIT_HLT;
5693 return 0;
5694 }
5695}
5cb56059
JS		 5696EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5697
5698int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5699{
5700 kvm_x86_ops->skip_emulated_instruction(vcpu);
5701 return kvm_vcpu_halt(vcpu);
5702}
8776e519
HB		 5703EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5704
6aef266c
SV		 5705/*
5706 * kvm_pv_kick_cpu_op: Kick a vcpu.
5707 *
5708 * @apicid - apicid of vcpu to be kicked.
5709 */
5710static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5711{
24d2166b 5712 struct kvm_lapic_irq lapic_irq;
6aef266c 5713
24d2166b
R		 5714 lapic_irq.shorthand = 0;
5715 lapic_irq.dest_mode = 0;
5716 lapic_irq.dest_id = apicid;
93bbf0b8 5717 lapic_irq.msi_redir_hint = false;
6aef266c 5718
24d2166b 5719 lapic_irq.delivery_mode = APIC_DM_REMRD;
795a149e 5720 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6aef266c
SV		 5721}
5722
8776e519
HB		 5723int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5724{
5725 unsigned long nr, a0, a1, a2, a3, ret;
a449c7aa 5726 int op_64_bit, r = 1;
8776e519 5727
5cb56059
JS		 5728 kvm_x86_ops->skip_emulated_instruction(vcpu);
5729
55cd8e5a
GN		 5730 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5731 return kvm_hv_hypercall(vcpu);
5732
5fdbf976
MT		 5733 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5734 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5735 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5736 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5737 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
8776e519 5738
229456fc 5739 trace_kvm_hypercall(nr, a0, a1, a2, a3);
2714d1d3 5740
a449c7aa
NA		 5741 op_64_bit = is_64_bit_mode(vcpu);
5742 if (!op_64_bit) {
8776e519
HB		 5743 nr &= 0xFFFFFFFF;
5744 a0 &= 0xFFFFFFFF;
5745 a1 &= 0xFFFFFFFF;
5746 a2 &= 0xFFFFFFFF;
5747 a3 &= 0xFFFFFFFF;
5748 }
5749
07708c4a
JK		 5750 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5751 ret = -KVM_EPERM;
5752 goto out;
5753 }
5754
8776e519 5755 switch (nr) {
b93463aa
AK		 5756 case KVM_HC_VAPIC_POLL_IRQ:
5757 ret = 0;
5758 break;
6aef266c
SV		 5759 case KVM_HC_KICK_CPU:
5760 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5761 ret = 0;
5762 break;
8776e519
HB		 5763 default:
5764 ret = -KVM_ENOSYS;
5765 break;
5766 }
07708c4a 5767out:
a449c7aa
NA		 5768 if (!op_64_bit)
5769 ret = (u32)ret;
5fdbf976 5770 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
f11c3a8d 5771 ++vcpu->stat.hypercalls;
2f333bcb 5772 return r;
8776e519
HB		 5773}
5774EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5775
b6785def 5776static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8776e519 5777{
d6aa1000 5778 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8776e519 5779 char instruction[3];
5fdbf976 5780 unsigned long rip = kvm_rip_read(vcpu);
8776e519 5781
8776e519 5782 kvm_x86_ops->patch_hypercall(vcpu, instruction);
8776e519 5783
9d74191a 5784 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
8776e519
HB		 5785}
5786
b6c7a5dc
HB		 5787/*
5788 * Check if userspace requested an interrupt window, and that the
5789 * interrupt window is open.
5790 *
5791 * No need to exit to userspace if we already have an interrupt queued.
5792 */
851ba692 5793static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
b6c7a5dc 5794{
8061823a 5795 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
851ba692 5796 vcpu->run->request_interrupt_window &&
5df56646 5797 kvm_arch_interrupt_allowed(vcpu));
b6c7a5dc
HB		 5798}
5799
851ba692 5800static void post_kvm_run_save(struct kvm_vcpu *vcpu)
b6c7a5dc 5801{
851ba692
AK		 5802 struct kvm_run *kvm_run = vcpu->run;
5803
91586a3b 5804 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
f077825a 5805 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
2d3ad1f4 5806 kvm_run->cr8 = kvm_get_cr8(vcpu);
b6c7a5dc 5807 kvm_run->apic_base = kvm_get_apic_base(vcpu);
4531220b 5808 if (irqchip_in_kernel(vcpu->kvm))
b6c7a5dc 5809 kvm_run->ready_for_interrupt_injection = 1;
4531220b 5810 else
b6c7a5dc 5811 kvm_run->ready_for_interrupt_injection =
fa9726b0
GN		 5812 kvm_arch_interrupt_allowed(vcpu) &&
5813 !kvm_cpu_has_interrupt(vcpu) &&
5814 !kvm_event_needs_reinjection(vcpu);
b6c7a5dc
HB		 5815}
5816
95ba8273
GN		 5817static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5818{
5819 int max_irr, tpr;
5820
5821 if (!kvm_x86_ops->update_cr8_intercept)
5822 return;
5823
88c808fd
AK		 5824 if (!vcpu->arch.apic)
5825 return;
5826
8db3baa2
GN		 5827 if (!vcpu->arch.apic->vapic_addr)
5828 max_irr = kvm_lapic_find_highest_irr(vcpu);
5829 else
5830 max_irr = -1;
95ba8273
GN		 5831
5832 if (max_irr != -1)
5833 max_irr >>= 4;
5834
5835 tpr = kvm_lapic_get_cr8(vcpu);
5836
5837 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5838}
5839
b6b8a145 5840static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
95ba8273 5841{
b6b8a145
JK		 5842 int r;
5843
95ba8273 5844 /* try to reinject previous events if any */
b59bb7bd 5845 if (vcpu->arch.exception.pending) {
5c1c85d0
AK		 5846 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5847 vcpu->arch.exception.has_error_code,
5848 vcpu->arch.exception.error_code);
d6e8c854
NA		 5849
5850 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
5851 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
5852 X86_EFLAGS_RF);
5853
6bdf0662
NA		 5854 if (vcpu->arch.exception.nr == DB_VECTOR &&
5855 (vcpu->arch.dr7 & DR7_GD)) {
5856 vcpu->arch.dr7 &= ~DR7_GD;
5857 kvm_update_dr7(vcpu);
5858 }
5859
b59bb7bd
GN		 5860 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5861 vcpu->arch.exception.has_error_code,
ce7ddec4
JR		 5862 vcpu->arch.exception.error_code,
5863 vcpu->arch.exception.reinject);
b6b8a145 5864 return 0;
b59bb7bd
GN		 5865 }
5866
95ba8273
GN		 5867 if (vcpu->arch.nmi_injected) {
5868 kvm_x86_ops->set_nmi(vcpu);
b6b8a145 5869 return 0;
95ba8273
GN		 5870 }
5871
5872 if (vcpu->arch.interrupt.pending) {
66fd3f7f 5873 kvm_x86_ops->set_irq(vcpu);
b6b8a145
JK		 5874 return 0;
5875 }
5876
5877 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5878 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5879 if (r != 0)
5880 return r;
95ba8273
GN		 5881 }
5882
5883 /* try to inject new event if pending */
5884 if (vcpu->arch.nmi_pending) {
5885 if (kvm_x86_ops->nmi_allowed(vcpu)) {
7460fb4a 5886 --vcpu->arch.nmi_pending;
95ba8273
GN		 5887 vcpu->arch.nmi_injected = true;
5888 kvm_x86_ops->set_nmi(vcpu);
5889 }
c7c9c56c 5890 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
9242b5b6
BD		 5891 /*
5892 * Because interrupts can be injected asynchronously, we are
5893 * calling check_nested_events again here to avoid a race condition.
5894 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
5895 * proposal and current concerns. Perhaps we should be setting
5896 * KVM_REQ_EVENT only on certain events and not unconditionally?
5897 */
5898 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5899 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5900 if (r != 0)
5901 return r;
5902 }
95ba8273 5903 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
66fd3f7f
GN		 5904 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5905 false);
5906 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 5907 }
5908 }
b6b8a145 5909 return 0;
95ba8273
GN		 5910}
5911
7460fb4a
AK		 5912static void process_nmi(struct kvm_vcpu *vcpu)
5913{
5914 unsigned limit = 2;
5915
5916 /*
5917 * x86 is limited to one NMI running, and one NMI pending after it.
5918 * If an NMI is already in progress, limit further NMIs to just one.
5919 * Otherwise, allow two (and we'll inject the first one immediately).
5920 */
5921 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5922 limit = 1;
5923
5924 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5925 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5926 kvm_make_request(KVM_REQ_EVENT, vcpu);
5927}
5928
660a5d51
PB		 5929#define put_smstate(type, buf, offset, val) \
5930 *(type *)((buf) + (offset) - 0x7e00) = val
5931
5932static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
5933{
5934 u32 flags = 0;
5935 flags |= seg->g << 23;
5936 flags |= seg->db << 22;
5937 flags |= seg->l << 21;
5938 flags |= seg->avl << 20;
5939 flags |= seg->present << 15;
5940 flags |= seg->dpl << 13;
5941 flags |= seg->s << 12;
5942 flags |= seg->type << 8;
5943 return flags;
5944}
5945
5946static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
5947{
5948 struct kvm_segment seg;
5949 int offset;
5950
5951 kvm_get_segment(vcpu, &seg, n);
5952 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
5953
5954 if (n < 3)
5955 offset = 0x7f84 + n * 12;
5956 else
5957 offset = 0x7f2c + (n - 3) * 12;
5958
5959 put_smstate(u32, buf, offset + 8, seg.base);
5960 put_smstate(u32, buf, offset + 4, seg.limit);
5961 put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
5962}
5963
5964static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
5965{
5966 struct kvm_segment seg;
5967 int offset;
5968 u16 flags;
5969
5970 kvm_get_segment(vcpu, &seg, n);
5971 offset = 0x7e00 + n * 16;
5972
5973 flags = process_smi_get_segment_flags(&seg) >> 8;
5974 put_smstate(u16, buf, offset, seg.selector);
5975 put_smstate(u16, buf, offset + 2, flags);
5976 put_smstate(u32, buf, offset + 4, seg.limit);
5977 put_smstate(u64, buf, offset + 8, seg.base);
5978}
5979
5980static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
5981{
5982 struct desc_ptr dt;
5983 struct kvm_segment seg;
5984 unsigned long val;
5985 int i;
5986
5987 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
5988 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
5989 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
5990 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
5991
5992 for (i = 0; i < 8; i++)
5993 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
5994
5995 kvm_get_dr(vcpu, 6, &val);
5996 put_smstate(u32, buf, 0x7fcc, (u32)val);
5997 kvm_get_dr(vcpu, 7, &val);
5998 put_smstate(u32, buf, 0x7fc8, (u32)val);
5999
6000 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6001 put_smstate(u32, buf, 0x7fc4, seg.selector);
6002 put_smstate(u32, buf, 0x7f64, seg.base);
6003 put_smstate(u32, buf, 0x7f60, seg.limit);
6004 put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
6005
6006 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6007 put_smstate(u32, buf, 0x7fc0, seg.selector);
6008 put_smstate(u32, buf, 0x7f80, seg.base);
6009 put_smstate(u32, buf, 0x7f7c, seg.limit);
6010 put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6011
6012 kvm_x86_ops->get_gdt(vcpu, &dt);
6013 put_smstate(u32, buf, 0x7f74, dt.address);
6014 put_smstate(u32, buf, 0x7f70, dt.size);
6015
6016 kvm_x86_ops->get_idt(vcpu, &dt);
6017 put_smstate(u32, buf, 0x7f58, dt.address);
6018 put_smstate(u32, buf, 0x7f54, dt.size);
6019
6020 for (i = 0; i < 6; i++)
6021 process_smi_save_seg_32(vcpu, buf, i);
6022
6023 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6024
6025 /* revision id */
6026 put_smstate(u32, buf, 0x7efc, 0x00020000);
6027 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6028}
6029
6030static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6031{
6032#ifdef CONFIG_X86_64
6033 struct desc_ptr dt;
6034 struct kvm_segment seg;
6035 unsigned long val;
6036 int i;
6037
6038 for (i = 0; i < 16; i++)
6039 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6040
6041 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6042 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6043
6044 kvm_get_dr(vcpu, 6, &val);
6045 put_smstate(u64, buf, 0x7f68, val);
6046 kvm_get_dr(vcpu, 7, &val);
6047 put_smstate(u64, buf, 0x7f60, val);
6048
6049 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6050 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6051 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6052
6053 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6054
6055 /* revision id */
6056 put_smstate(u32, buf, 0x7efc, 0x00020064);
6057
6058 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6059
6060 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6061 put_smstate(u16, buf, 0x7e90, seg.selector);
6062 put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6063 put_smstate(u32, buf, 0x7e94, seg.limit);
6064 put_smstate(u64, buf, 0x7e98, seg.base);
6065
6066 kvm_x86_ops->get_idt(vcpu, &dt);
6067 put_smstate(u32, buf, 0x7e84, dt.size);
6068 put_smstate(u64, buf, 0x7e88, dt.address);
6069
6070 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6071 put_smstate(u16, buf, 0x7e70, seg.selector);
6072 put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6073 put_smstate(u32, buf, 0x7e74, seg.limit);
6074 put_smstate(u64, buf, 0x7e78, seg.base);
6075
6076 kvm_x86_ops->get_gdt(vcpu, &dt);
6077 put_smstate(u32, buf, 0x7e64, dt.size);
6078 put_smstate(u64, buf, 0x7e68, dt.address);
6079
6080 for (i = 0; i < 6; i++)
6081 process_smi_save_seg_64(vcpu, buf, i);
6082#else
6083 WARN_ON_ONCE(1);
6084#endif
6085}
6086
64d60670
PB		 6087static void process_smi(struct kvm_vcpu *vcpu)
6088{
660a5d51
PB		 6089 struct kvm_segment cs, ds;
6090 char buf[512];
6091 u32 cr0;
6092
64d60670
PB		 6093 if (is_smm(vcpu)) {
6094 vcpu->arch.smi_pending = true;
6095 return;
6096 }
6097
660a5d51
PB		 6098 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6099 vcpu->arch.hflags |= HF_SMM_MASK;
6100 memset(buf, 0, 512);
6101 if (guest_cpuid_has_longmode(vcpu))
6102 process_smi_save_state_64(vcpu, buf);
6103 else
6104 process_smi_save_state_32(vcpu, buf);
6105
54bf36aa 6106 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
660a5d51
PB		 6107
6108 if (kvm_x86_ops->get_nmi_mask(vcpu))
6109 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6110 else
6111 kvm_x86_ops->set_nmi_mask(vcpu, true);
6112
6113 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6114 kvm_rip_write(vcpu, 0x8000);
6115
6116 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6117 kvm_x86_ops->set_cr0(vcpu, cr0);
6118 vcpu->arch.cr0 = cr0;
6119
6120 kvm_x86_ops->set_cr4(vcpu, 0);
6121
6122 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6123
6124 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6125 cs.base = vcpu->arch.smbase;
6126
6127 ds.selector = 0;
6128 ds.base = 0;
6129
6130 cs.limit = ds.limit = 0xffffffff;
6131 cs.type = ds.type = 0x3;
6132 cs.dpl = ds.dpl = 0;
6133 cs.db = ds.db = 0;
6134 cs.s = ds.s = 1;
6135 cs.l = ds.l = 0;
6136 cs.g = ds.g = 1;
6137 cs.avl = ds.avl = 0;
6138 cs.present = ds.present = 1;
6139 cs.unusable = ds.unusable = 0;
6140 cs.padding = ds.padding = 0;
6141
6142 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6143 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6144 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6145 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6146 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6147 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6148
6149 if (guest_cpuid_has_longmode(vcpu))
6150 kvm_x86_ops->set_efer(vcpu, 0);
6151
6152 kvm_update_cpuid(vcpu);
6153 kvm_mmu_reset_context(vcpu);
64d60670
PB		 6154}
6155
3d81bc7e 6156static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
c7c9c56c
YZ		 6157{
6158 u64 eoi_exit_bitmap[4];
cf9e65b7 6159 u32 tmr[8];
c7c9c56c 6160
3d81bc7e
YZ		 6161 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6162 return;
c7c9c56c
YZ		 6163
6164 memset(eoi_exit_bitmap, 0, 32);
cf9e65b7 6165 memset(tmr, 0, 32);
c7c9c56c 6166
cf9e65b7 6167 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
c7c9c56c 6168 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
cf9e65b7 6169 kvm_apic_update_tmr(vcpu, tmr);
c7c9c56c
YZ		 6170}
6171
a70656b6
RK		 6172static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6173{
6174 ++vcpu->stat.tlb_flush;
6175 kvm_x86_ops->tlb_flush(vcpu);
6176}
6177
4256f43f
TC		 6178void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6179{
c24ae0dc
TC		 6180 struct page *page = NULL;
6181
f439ed27
PB		 6182 if (!irqchip_in_kernel(vcpu->kvm))
6183 return;
6184
4256f43f
TC		 6185 if (!kvm_x86_ops->set_apic_access_page_addr)
6186 return;
6187
c24ae0dc 6188 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
e8fd5e9e
AA		 6189 if (is_error_page(page))
6190 return;
c24ae0dc
TC		 6191 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6192
6193 /*
6194 * Do not pin apic access page in memory, the MMU notifier
6195 * will call us again if it is migrated or swapped out.
6196 */
6197 put_page(page);
4256f43f
TC		 6198}
6199EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6200
fe71557a
TC		 6201void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6202 unsigned long address)
6203{
c24ae0dc
TC		 6204 /*
6205 * The physical address of apic access page is stored in the VMCS.
6206 * Update it when it becomes invalid.
6207 */
6208 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6209 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
fe71557a
TC		 6210}
6211
9357d939 6212/*
362c698f 6213 * Returns 1 to let vcpu_run() continue the guest execution loop without
9357d939
TY		 6214 * exiting to the userspace. Otherwise, the value will be returned to the
6215 * userspace.
6216 */
851ba692 6217static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
b6c7a5dc
HB		 6218{
6219 int r;
6a8b1d13 6220 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
851ba692 6221 vcpu->run->request_interrupt_window;
730dca42 6222 bool req_immediate_exit = false;
b6c7a5dc 6223
3e007509 6224 if (vcpu->requests) {
a8eeb04a 6225 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
2e53d63a 6226 kvm_mmu_unload(vcpu);
a8eeb04a 6227 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
2f599714 6228 __kvm_migrate_timers(vcpu);
d828199e
MT		 6229 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6230 kvm_gen_update_masterclock(vcpu->kvm);
0061d53d
MT		 6231 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6232 kvm_gen_kvmclock_update(vcpu);
34c238a1
ZA		 6233 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6234 r = kvm_guest_time_update(vcpu);
8cfdc000
ZA		 6235 if (unlikely(r))
6236 goto out;
6237 }
a8eeb04a 6238 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
4731d4c7 6239 kvm_mmu_sync_roots(vcpu);
a8eeb04a 6240 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
a70656b6 6241 kvm_vcpu_flush_tlb(vcpu);
a8eeb04a 6242 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
851ba692 6243 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
b93463aa
AK		 6244 r = 0;
6245 goto out;
6246 }
a8eeb04a 6247 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
851ba692 6248 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
71c4dfaf
JR		 6249 r = 0;
6250 goto out;
6251 }
a8eeb04a 6252 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
02daab21
AK		 6253 vcpu->fpu_active = 0;
6254 kvm_x86_ops->fpu_deactivate(vcpu);
6255 }
af585b92
GN		 6256 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6257 /* Page is swapped out. Do synthetic halt */
6258 vcpu->arch.apf.halted = true;
6259 r = 1;
6260 goto out;
6261 }
c9aaa895
GC		 6262 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6263 record_steal_time(vcpu);
64d60670
PB		 6264 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6265 process_smi(vcpu);
7460fb4a
AK		 6266 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6267 process_nmi(vcpu);
f5132b01 6268 if (kvm_check_request(KVM_REQ_PMU, vcpu))
c6702c9d 6269 kvm_pmu_handle_event(vcpu);
f5132b01 6270 if (kvm_check_request(KVM_REQ_PMI, vcpu))
c6702c9d 6271 kvm_pmu_deliver_pmi(vcpu);
3d81bc7e
YZ		 6272 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6273 vcpu_scan_ioapic(vcpu);
4256f43f
TC		 6274 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6275 kvm_vcpu_reload_apic_access_page(vcpu);
2ce79189
AS		 6276 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6277 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6278 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6279 r = 0;
6280 goto out;
6281 }
2f52d58c 6282 }
b93463aa 6283
b463a6f7 6284 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
66450a21
JK		 6285 kvm_apic_accept_events(vcpu);
6286 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6287 r = 1;
6288 goto out;
6289 }
6290
b6b8a145
JK		 6291 if (inject_pending_event(vcpu, req_int_win) != 0)
6292 req_immediate_exit = true;
b463a6f7 6293 /* enable NMI/IRQ window open exits if needed */
b6b8a145 6294 else if (vcpu->arch.nmi_pending)
c9a7953f 6295 kvm_x86_ops->enable_nmi_window(vcpu);
c7c9c56c 6296 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
c9a7953f 6297 kvm_x86_ops->enable_irq_window(vcpu);
b463a6f7
AK		 6298
6299 if (kvm_lapic_enabled(vcpu)) {
c7c9c56c
YZ		 6300 /*
6301 * Update architecture specific hints for APIC
6302 * virtual interrupt delivery.
6303 */
6304 if (kvm_x86_ops->hwapic_irr_update)
6305 kvm_x86_ops->hwapic_irr_update(vcpu,
6306 kvm_lapic_find_highest_irr(vcpu));
b463a6f7
AK		 6307 update_cr8_intercept(vcpu);
6308 kvm_lapic_sync_to_vapic(vcpu);
6309 }
6310 }
6311
d8368af8
AK		 6312 r = kvm_mmu_reload(vcpu);
6313 if (unlikely(r)) {
d905c069 6314 goto cancel_injection;
d8368af8
AK		 6315 }
6316
b6c7a5dc
HB		 6317 preempt_disable();
6318
6319 kvm_x86_ops->prepare_guest_switch(vcpu);
2608d7a1
AK		 6320 if (vcpu->fpu_active)
6321 kvm_load_guest_fpu(vcpu);
2acf923e 6322 kvm_load_guest_xcr0(vcpu);
b6c7a5dc 6323
6b7e2d09
XG		 6324 vcpu->mode = IN_GUEST_MODE;
6325
01b71917
MT		 6326 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6327
6b7e2d09
XG		 6328 /* We should set ->mode before check ->requests,
6329 * see the comment in make_all_cpus_request.
6330 */
01b71917 6331 smp_mb__after_srcu_read_unlock();
b6c7a5dc 6332
d94e1dc9 6333 local_irq_disable();
32f88400 6334
6b7e2d09 6335 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
d94e1dc9 6336 || need_resched() || signal_pending(current)) {
6b7e2d09 6337 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 6338 smp_wmb();
6c142801
AK		 6339 local_irq_enable();
6340 preempt_enable();
01b71917 6341 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6c142801 6342 r = 1;
d905c069 6343 goto cancel_injection;
6c142801
AK		 6344 }
6345
d6185f20
NHE		 6346 if (req_immediate_exit)
6347 smp_send_reschedule(vcpu->cpu);
6348
ccf73aaf 6349 __kvm_guest_enter();
b6c7a5dc 6350
42dbaa5a 6351 if (unlikely(vcpu->arch.switch_db_regs)) {
42dbaa5a
JK		 6352 set_debugreg(0, 7);
6353 set_debugreg(vcpu->arch.eff_db[0], 0);
6354 set_debugreg(vcpu->arch.eff_db[1], 1);
6355 set_debugreg(vcpu->arch.eff_db[2], 2);
6356 set_debugreg(vcpu->arch.eff_db[3], 3);
c77fb5fe 6357 set_debugreg(vcpu->arch.dr6, 6);
ae561ede 6358 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
42dbaa5a 6359 }
b6c7a5dc 6360
229456fc 6361 trace_kvm_entry(vcpu->vcpu_id);
d0659d94 6362 wait_lapic_expire(vcpu);
851ba692 6363 kvm_x86_ops->run(vcpu);
b6c7a5dc 6364
c77fb5fe
PB		 6365 /*
6366 * Do this here before restoring debug registers on the host. And
6367 * since we do this before handling the vmexit, a DR access vmexit
6368 * can (a) read the correct value of the debug registers, (b) set
6369 * KVM_DEBUGREG_WONT_EXIT again.
6370 */
6371 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6372 int i;
6373
6374 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6375 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6376 for (i = 0; i < KVM_NR_DB_REGS; i++)
6377 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6378 }
6379
24f1e32c
FW		 6380 /*
6381 * If the guest has used debug registers, at least dr7
6382 * will be disabled while returning to the host.
6383 * If we don't have active breakpoints in the host, we don't
6384 * care about the messed up debug address registers. But if
6385 * we have some of them active, restore the old state.
6386 */
59d8eb53 6387 if (hw_breakpoint_active())
24f1e32c 6388 hw_breakpoint_restore();
42dbaa5a 6389
886b470c
MT		 6390 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6391 native_read_tsc());
1d5f066e 6392
6b7e2d09 6393 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 6394 smp_wmb();
a547c6db
YZ		 6395
6396 /* Interrupt is enabled by handle_external_intr() */
6397 kvm_x86_ops->handle_external_intr(vcpu);
b6c7a5dc
HB		 6398
6399 ++vcpu->stat.exits;
6400
6401 /*
6402 * We must have an instruction between local_irq_enable() and
6403 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6404 * the interrupt shadow. The stat.exits increment will do nicely.
6405 * But we need to prevent reordering, hence this barrier():
6406 */
6407 barrier();
6408
6409 kvm_guest_exit();
6410
6411 preempt_enable();
6412
f656ce01 6413 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3200f405 6414
b6c7a5dc
HB		 6415 /*
6416 * Profile KVM exit RIPs:
6417 */
6418 if (unlikely(prof_on == KVM_PROFILING)) {
5fdbf976
MT		 6419 unsigned long rip = kvm_rip_read(vcpu);
6420 profile_hit(KVM_PROFILING, (void *)rip);
b6c7a5dc
HB		 6421 }
6422
cc578287
ZA		 6423 if (unlikely(vcpu->arch.tsc_always_catchup))
6424 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
298101da 6425
5cfb1d5a
MT		 6426 if (vcpu->arch.apic_attention)
6427 kvm_lapic_sync_from_vapic(vcpu);
b93463aa 6428
851ba692 6429 r = kvm_x86_ops->handle_exit(vcpu);
d905c069
MT		 6430 return r;
6431
6432cancel_injection:
6433 kvm_x86_ops->cancel_injection(vcpu);
ae7a2a3f
MT		 6434 if (unlikely(vcpu->arch.apic_attention))
6435 kvm_lapic_sync_from_vapic(vcpu);
d7690175
MT		 6436out:
6437 return r;
6438}
b6c7a5dc 6439
362c698f
PB		 6440static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6441{
9c8fd1ba
PB		 6442 if (!kvm_arch_vcpu_runnable(vcpu)) {
6443 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6444 kvm_vcpu_block(vcpu);
6445 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6446 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6447 return 1;
6448 }
362c698f
PB		 6449
6450 kvm_apic_accept_events(vcpu);
6451 switch(vcpu->arch.mp_state) {
6452 case KVM_MP_STATE_HALTED:
6453 vcpu->arch.pv.pv_unhalted = false;
6454 vcpu->arch.mp_state =
6455 KVM_MP_STATE_RUNNABLE;
6456 case KVM_MP_STATE_RUNNABLE:
6457 vcpu->arch.apf.halted = false;
6458 break;
6459 case KVM_MP_STATE_INIT_RECEIVED:
6460 break;
6461 default:
6462 return -EINTR;
6463 break;
6464 }
6465 return 1;
6466}
09cec754 6467
362c698f 6468static int vcpu_run(struct kvm_vcpu *vcpu)
d7690175
MT		 6469{
6470 int r;
f656ce01 6471 struct kvm *kvm = vcpu->kvm;
d7690175 6472
f656ce01 6473 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 6474
362c698f 6475 for (;;) {
af585b92
GN		 6476 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6477 !vcpu->arch.apf.halted)
851ba692 6478 r = vcpu_enter_guest(vcpu);
362c698f
PB		 6479 else
6480 r = vcpu_block(kvm, vcpu);
09cec754
GN		 6481 if (r <= 0)
6482 break;
6483
6484 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6485 if (kvm_cpu_has_pending_timer(vcpu))
6486 kvm_inject_pending_timer_irqs(vcpu);
6487
851ba692 6488 if (dm_request_for_irq_injection(vcpu)) {
09cec754 6489 r = -EINTR;
851ba692 6490 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754 6491 ++vcpu->stat.request_irq_exits;
362c698f 6492 break;
09cec754 6493 }
af585b92
GN		 6494
6495 kvm_check_async_pf_completion(vcpu);
6496
09cec754
GN		 6497 if (signal_pending(current)) {
6498 r = -EINTR;
851ba692 6499 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754 6500 ++vcpu->stat.signal_exits;
362c698f 6501 break;
09cec754
GN		 6502 }
6503 if (need_resched()) {
f656ce01 6504 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
c08ac06a 6505 cond_resched();
f656ce01 6506 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 6507 }
b6c7a5dc
HB		 6508 }
6509
f656ce01 6510 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
b6c7a5dc
HB		 6511
6512 return r;
6513}
6514
716d51ab
GN		 6515static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6516{
6517 int r;
6518 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6519 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6520 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6521 if (r != EMULATE_DONE)
6522 return 0;
6523 return 1;
6524}
6525
6526static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6527{
6528 BUG_ON(!vcpu->arch.pio.count);
6529
6530 return complete_emulated_io(vcpu);
6531}
6532
f78146b0
AK		 6533/*
6534 * Implements the following, as a state machine:
6535 *
6536 * read:
6537 * for each fragment
87da7e66
XG		 6538 * for each mmio piece in the fragment
6539 * write gpa, len
6540 * exit
6541 * copy data
f78146b0
AK		 6542 * execute insn
6543 *
6544 * write:
6545 * for each fragment
87da7e66
XG		 6546 * for each mmio piece in the fragment
6547 * write gpa, len
6548 * copy data
6549 * exit
f78146b0 6550 */
716d51ab 6551static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
5287f194
AK		 6552{
6553 struct kvm_run *run = vcpu->run;
f78146b0 6554 struct kvm_mmio_fragment *frag;
87da7e66 6555 unsigned len;
5287f194 6556
716d51ab 6557 BUG_ON(!vcpu->mmio_needed);
5287f194 6558
716d51ab 6559 /* Complete previous fragment */
87da7e66
XG		 6560 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6561 len = min(8u, frag->len);
716d51ab 6562 if (!vcpu->mmio_is_write)
87da7e66
XG		 6563 memcpy(frag->data, run->mmio.data, len);
6564
6565 if (frag->len <= 8) {
6566 /* Switch to the next fragment. */
6567 frag++;
6568 vcpu->mmio_cur_fragment++;
6569 } else {
6570 /* Go forward to the next mmio piece. */
6571 frag->data += len;
6572 frag->gpa += len;
6573 frag->len -= len;
6574 }
6575
a08d3b3b 6576 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
716d51ab 6577 vcpu->mmio_needed = 0;
0912c977
PB		 6578
6579 /* FIXME: return into emulator if single-stepping. */
cef4dea0 6580 if (vcpu->mmio_is_write)
716d51ab
GN		 6581 return 1;
6582 vcpu->mmio_read_completed = 1;
6583 return complete_emulated_io(vcpu);
6584 }
87da7e66 6585
716d51ab
GN		 6586 run->exit_reason = KVM_EXIT_MMIO;
6587 run->mmio.phys_addr = frag->gpa;
6588 if (vcpu->mmio_is_write)
87da7e66
XG		 6589 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6590 run->mmio.len = min(8u, frag->len);
716d51ab
GN		 6591 run->mmio.is_write = vcpu->mmio_is_write;
6592 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6593 return 0;
5287f194
AK		 6594}
6595
716d51ab 6596
b6c7a5dc
HB		 6597int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6598{
c5bedc68 6599 struct fpu *fpu = &current->thread.fpu;
b6c7a5dc
HB		 6600 int r;
6601 sigset_t sigsaved;
6602
c4d72e2d 6603 fpu__activate_curr(fpu);
e5c30142 6604
ac9f6dc0
AK		 6605 if (vcpu->sigset_active)
6606 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6607
a4535290 6608 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
b6c7a5dc 6609 kvm_vcpu_block(vcpu);
66450a21 6610 kvm_apic_accept_events(vcpu);
d7690175 6611 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
ac9f6dc0
AK		 6612 r = -EAGAIN;
6613 goto out;
b6c7a5dc
HB		 6614 }
6615
b6c7a5dc 6616 /* re-sync apic's tpr */
eea1cff9
AP		 6617 if (!irqchip_in_kernel(vcpu->kvm)) {
6618 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6619 r = -EINVAL;
6620 goto out;
6621 }
6622 }
b6c7a5dc 6623
716d51ab
GN		 6624 if (unlikely(vcpu->arch.complete_userspace_io)) {
6625 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6626 vcpu->arch.complete_userspace_io = NULL;
6627 r = cui(vcpu);
6628 if (r <= 0)
6629 goto out;
6630 } else
6631 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
5287f194 6632
362c698f 6633 r = vcpu_run(vcpu);
b6c7a5dc
HB		 6634
6635out:
f1d86e46 6636 post_kvm_run_save(vcpu);
b6c7a5dc
HB		 6637 if (vcpu->sigset_active)
6638 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6639
b6c7a5dc
HB		 6640 return r;
6641}
6642
6643int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6644{
7ae441ea
GN		 6645 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6646 /*
6647 * We are here if userspace calls get_regs() in the middle of
6648 * instruction emulation. Registers state needs to be copied
4a969980 6649 * back from emulation context to vcpu. Userspace shouldn't do
7ae441ea
GN		 6650 * that usually, but some bad designed PV devices (vmware
6651 * backdoor interface) need this to work
6652 */
dd856efa 6653 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7ae441ea
GN		 6654 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6655 }
5fdbf976
MT		 6656 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6657 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6658 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6659 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6660 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6661 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6662 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6663 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
b6c7a5dc 6664#ifdef CONFIG_X86_64
5fdbf976
MT		 6665 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6666 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6667 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6668 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6669 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6670 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6671 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6672 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
b6c7a5dc
HB		 6673#endif
6674
5fdbf976 6675 regs->rip = kvm_rip_read(vcpu);
91586a3b 6676 regs->rflags = kvm_get_rflags(vcpu);
b6c7a5dc 6677
b6c7a5dc
HB		 6678 return 0;
6679}
6680
6681int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6682{
7ae441ea
GN		 6683 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6684 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6685
5fdbf976
MT		 6686 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6687 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6688 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6689 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6690 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6691 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6692 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6693 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
b6c7a5dc 6694#ifdef CONFIG_X86_64
5fdbf976
MT		 6695 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6696 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6697 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6698 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6699 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6700 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6701 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6702 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
b6c7a5dc
HB		 6703#endif
6704
5fdbf976 6705 kvm_rip_write(vcpu, regs->rip);
91586a3b 6706 kvm_set_rflags(vcpu, regs->rflags);
b6c7a5dc 6707
b4f14abd
JK		 6708 vcpu->arch.exception.pending = false;
6709
3842d135
AK		 6710 kvm_make_request(KVM_REQ_EVENT, vcpu);
6711
b6c7a5dc
HB		 6712 return 0;
6713}
6714
b6c7a5dc
HB		 6715void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6716{
6717 struct kvm_segment cs;
6718
3e6e0aab 6719 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
b6c7a5dc
HB		 6720 *db = cs.db;
6721 *l = cs.l;
6722}
6723EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6724
6725int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6726 struct kvm_sregs *sregs)
6727{
89a27f4d 6728 struct desc_ptr dt;
b6c7a5dc 6729
3e6e0aab
GT		 6730 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6731 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6732 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6733 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6734 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6735 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 6736
3e6e0aab
GT		 6737 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6738 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc
HB		 6739
6740 kvm_x86_ops->get_idt(vcpu, &dt);
89a27f4d
GN		 6741 sregs->idt.limit = dt.size;
6742 sregs->idt.base = dt.address;
b6c7a5dc 6743 kvm_x86_ops->get_gdt(vcpu, &dt);
89a27f4d
GN		 6744 sregs->gdt.limit = dt.size;
6745 sregs->gdt.base = dt.address;
b6c7a5dc 6746
4d4ec087 6747 sregs->cr0 = kvm_read_cr0(vcpu);
ad312c7c 6748 sregs->cr2 = vcpu->arch.cr2;
9f8fe504 6749 sregs->cr3 = kvm_read_cr3(vcpu);
fc78f519 6750 sregs->cr4 = kvm_read_cr4(vcpu);
2d3ad1f4 6751 sregs->cr8 = kvm_get_cr8(vcpu);
f6801dff 6752 sregs->efer = vcpu->arch.efer;
b6c7a5dc
HB		 6753 sregs->apic_base = kvm_get_apic_base(vcpu);
6754
923c61bb 6755 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
b6c7a5dc 6756
36752c9b 6757 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
14d0bc1f
GN		 6758 set_bit(vcpu->arch.interrupt.nr,
6759 (unsigned long *)sregs->interrupt_bitmap);
16d7a191 6760
b6c7a5dc
HB		 6761 return 0;
6762}
6763
62d9f0db
MT		 6764int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6765 struct kvm_mp_state *mp_state)
6766{
66450a21 6767 kvm_apic_accept_events(vcpu);
6aef266c
SV		 6768 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6769 vcpu->arch.pv.pv_unhalted)
6770 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6771 else
6772 mp_state->mp_state = vcpu->arch.mp_state;
6773
62d9f0db
MT		 6774 return 0;
6775}
6776
6777int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6778 struct kvm_mp_state *mp_state)
6779{
66450a21
JK		 6780 if (!kvm_vcpu_has_lapic(vcpu) &&
6781 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6782 return -EINVAL;
6783
6784 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6785 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6786 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6787 } else
6788 vcpu->arch.mp_state = mp_state->mp_state;
3842d135 6789 kvm_make_request(KVM_REQ_EVENT, vcpu);
62d9f0db
MT		 6790 return 0;
6791}
6792
7f3d35fd
KW		 6793int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6794 int reason, bool has_error_code, u32 error_code)
b6c7a5dc 6795{
9d74191a 6796 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d 6797 int ret;
e01c2426 6798
8ec4722d 6799 init_emulate_ctxt(vcpu);
c697518a 6800
7f3d35fd 6801 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9d74191a 6802 has_error_code, error_code);
c697518a 6803
c697518a 6804 if (ret)
19d04437 6805 return EMULATE_FAIL;
37817f29 6806
9d74191a
TY		 6807 kvm_rip_write(vcpu, ctxt->eip);
6808 kvm_set_rflags(vcpu, ctxt->eflags);
3842d135 6809 kvm_make_request(KVM_REQ_EVENT, vcpu);
19d04437 6810 return EMULATE_DONE;
37817f29
IE		 6811}
6812EXPORT_SYMBOL_GPL(kvm_task_switch);
6813
b6c7a5dc
HB		 6814int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6815 struct kvm_sregs *sregs)
6816{
58cb628d 6817 struct msr_data apic_base_msr;
b6c7a5dc 6818 int mmu_reset_needed = 0;
63f42e02 6819 int pending_vec, max_bits, idx;
89a27f4d 6820 struct desc_ptr dt;
b6c7a5dc 6821
6d1068b3
PM		 6822 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6823 return -EINVAL;
6824
89a27f4d
GN		 6825 dt.size = sregs->idt.limit;
6826 dt.address = sregs->idt.base;
b6c7a5dc 6827 kvm_x86_ops->set_idt(vcpu, &dt);
89a27f4d
GN		 6828 dt.size = sregs->gdt.limit;
6829 dt.address = sregs->gdt.base;
b6c7a5dc
HB		 6830 kvm_x86_ops->set_gdt(vcpu, &dt);
6831
ad312c7c 6832 vcpu->arch.cr2 = sregs->cr2;
9f8fe504 6833 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
dc7e795e 6834 vcpu->arch.cr3 = sregs->cr3;
aff48baa 6835 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
b6c7a5dc 6836
2d3ad1f4 6837 kvm_set_cr8(vcpu, sregs->cr8);
b6c7a5dc 6838
f6801dff 6839 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
b6c7a5dc 6840 kvm_x86_ops->set_efer(vcpu, sregs->efer);
58cb628d
JK		 6841 apic_base_msr.data = sregs->apic_base;
6842 apic_base_msr.host_initiated = true;
6843 kvm_set_apic_base(vcpu, &apic_base_msr);
b6c7a5dc 6844
4d4ec087 6845 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
b6c7a5dc 6846 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
d7306163 6847 vcpu->arch.cr0 = sregs->cr0;
b6c7a5dc 6848
fc78f519 6849 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
b6c7a5dc 6850 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3ea3aa8c 6851 if (sregs->cr4 & X86_CR4_OSXSAVE)
00b27a3e 6852 kvm_update_cpuid(vcpu);
63f42e02
XG		 6853
6854 idx = srcu_read_lock(&vcpu->kvm->srcu);
7c93be44 6855 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
9f8fe504 6856 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7c93be44
MT		 6857 mmu_reset_needed = 1;
6858 }
63f42e02 6859 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b6c7a5dc
HB		 6860
6861 if (mmu_reset_needed)
6862 kvm_mmu_reset_context(vcpu);
6863
a50abc3b 6864 max_bits = KVM_NR_INTERRUPTS;
923c61bb
GN		 6865 pending_vec = find_first_bit(
6866 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6867 if (pending_vec < max_bits) {
66fd3f7f 6868 kvm_queue_interrupt(vcpu, pending_vec, false);
923c61bb 6869 pr_debug("Set back pending irq %d\n", pending_vec);
b6c7a5dc
HB		 6870 }
6871
3e6e0aab
GT		 6872 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6873 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6874 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6875 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6876 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6877 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 6878
3e6e0aab
GT		 6879 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6880 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc 6881
5f0269f5
ME		 6882 update_cr8_intercept(vcpu);
6883
9c3e4aab 6884 /* Older userspace won't unhalt the vcpu on reset. */
c5af89b6 6885 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9c3e4aab 6886 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3eeb3288 6887 !is_protmode(vcpu))
9c3e4aab
MT		 6888 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6889
3842d135
AK		 6890 kvm_make_request(KVM_REQ_EVENT, vcpu);
6891
b6c7a5dc
HB		 6892 return 0;
6893}
6894
d0bfb940
JK		 6895int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6896 struct kvm_guest_debug *dbg)
b6c7a5dc 6897{
355be0b9 6898 unsigned long rflags;
ae675ef0 6899 int i, r;
b6c7a5dc 6900
4f926bf2
JK		 6901 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6902 r = -EBUSY;
6903 if (vcpu->arch.exception.pending)
2122ff5e 6904 goto out;
4f926bf2
JK		 6905 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6906 kvm_queue_exception(vcpu, DB_VECTOR);
6907 else
6908 kvm_queue_exception(vcpu, BP_VECTOR);
6909 }
6910
91586a3b
JK		 6911 /*
6912 * Read rflags as long as potentially injected trace flags are still
6913 * filtered out.
6914 */
6915 rflags = kvm_get_rflags(vcpu);
355be0b9
JK		 6916
6917 vcpu->guest_debug = dbg->control;
6918 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6919 vcpu->guest_debug = 0;
6920
6921 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
ae675ef0
JK		 6922 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6923 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
c8639010 6924 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
ae675ef0
JK		 6925 } else {
6926 for (i = 0; i < KVM_NR_DB_REGS; i++)
6927 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
ae675ef0 6928 }
c8639010 6929 kvm_update_dr7(vcpu);
ae675ef0 6930
f92653ee
JK		 6931 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6932 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6933 get_segment_base(vcpu, VCPU_SREG_CS);
94fe45da 6934
91586a3b
JK		 6935 /*
6936 * Trigger an rflags update that will inject or remove the trace
6937 * flags.
6938 */
6939 kvm_set_rflags(vcpu, rflags);
b6c7a5dc 6940
c8639010 6941 kvm_x86_ops->update_db_bp_intercept(vcpu);
b6c7a5dc 6942
4f926bf2 6943 r = 0;
d0bfb940 6944
2122ff5e 6945out:
b6c7a5dc
HB		 6946
6947 return r;
6948}
6949
8b006791
ZX		 6950/*
6951 * Translate a guest virtual address to a guest physical address.
6952 */
6953int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6954 struct kvm_translation *tr)
6955{
6956 unsigned long vaddr = tr->linear_address;
6957 gpa_t gpa;
f656ce01 6958 int idx;
8b006791 6959
f656ce01 6960 idx = srcu_read_lock(&vcpu->kvm->srcu);
1871c602 6961 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
f656ce01 6962 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8b006791
ZX		 6963 tr->physical_address = gpa;
6964 tr->valid = gpa != UNMAPPED_GVA;
6965 tr->writeable = 1;
6966 tr->usermode = 0;
8b006791
ZX		 6967
6968 return 0;
6969}
6970
d0752060
HB		 6971int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6972{
c47ada30 6973 struct fxregs_state *fxsave =
7366ed77 6974 &vcpu->arch.guest_fpu.state.fxsave;
d0752060 6975
d0752060
HB		 6976 memcpy(fpu->fpr, fxsave->st_space, 128);
6977 fpu->fcw = fxsave->cwd;
6978 fpu->fsw = fxsave->swd;
6979 fpu->ftwx = fxsave->twd;
6980 fpu->last_opcode = fxsave->fop;
6981 fpu->last_ip = fxsave->rip;
6982 fpu->last_dp = fxsave->rdp;
6983 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6984
d0752060
HB		 6985 return 0;
6986}
6987
6988int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6989{
c47ada30 6990 struct fxregs_state *fxsave =
7366ed77 6991 &vcpu->arch.guest_fpu.state.fxsave;
d0752060 6992
d0752060
HB		 6993 memcpy(fxsave->st_space, fpu->fpr, 128);
6994 fxsave->cwd = fpu->fcw;
6995 fxsave->swd = fpu->fsw;
6996 fxsave->twd = fpu->ftwx;
6997 fxsave->fop = fpu->last_opcode;
6998 fxsave->rip = fpu->last_ip;
6999 fxsave->rdp = fpu->last_dp;
7000 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7001
d0752060
HB		 7002 return 0;
7003}
7004
0ee6a517 7005static void fx_init(struct kvm_vcpu *vcpu)
d0752060 7006{
bf935b0b 7007 fpstate_init(&vcpu->arch.guest_fpu.state);
df1daba7 7008 if (cpu_has_xsaves)
7366ed77 7009 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
df1daba7 7010 host_xcr0 | XSTATE_COMPACTION_ENABLED;
d0752060 7011
2acf923e
DC		 7012 /*
7013 * Ensure guest xcr0 is valid for loading
7014 */
7015 vcpu->arch.xcr0 = XSTATE_FP;
7016
ad312c7c 7017 vcpu->arch.cr0 |= X86_CR0_ET;
d0752060 7018}
d0752060
HB		 7019
7020void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7021{
2608d7a1 7022 if (vcpu->guest_fpu_loaded)
d0752060
HB		 7023 return;
7024
2acf923e
DC		 7025 /*
7026 * Restore all possible states in the guest,
7027 * and assume host would use all available bits.
7028 * Guest xcr0 would be loaded later.
7029 */
7030 kvm_put_guest_xcr0(vcpu);
d0752060 7031 vcpu->guest_fpu_loaded = 1;
b1a74bf8 7032 __kernel_fpu_begin();
003e2e8b 7033 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
0c04851c 7034 trace_kvm_fpu(1);
d0752060 7035}
d0752060
HB		 7036
7037void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7038{
2acf923e
DC		 7039 kvm_put_guest_xcr0(vcpu);
7040
653f52c3
RR		 7041 if (!vcpu->guest_fpu_loaded) {
7042 vcpu->fpu_counter = 0;
d0752060 7043 return;
653f52c3 7044 }
d0752060
HB		 7045
7046 vcpu->guest_fpu_loaded = 0;
4f836347 7047 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
b1a74bf8 7048 __kernel_fpu_end();
f096ed85 7049 ++vcpu->stat.fpu_reload;
653f52c3
RR		 7050 /*
7051 * If using eager FPU mode, or if the guest is a frequent user
7052 * of the FPU, just leave the FPU active for next time.
7053 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7054 * the FPU in bursts will revert to loading it on demand.
7055 */
a9b4fb7e 7056 if (!vcpu->arch.eager_fpu) {
653f52c3
RR		 7057 if (++vcpu->fpu_counter < 5)
7058 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7059 }
0c04851c 7060 trace_kvm_fpu(0);
d0752060 7061}
e9b11c17
ZX		 7062
7063void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7064{
12f9a48f 7065 kvmclock_reset(vcpu);
7f1ea208 7066
f5f48ee1 7067 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
e9b11c17
ZX		 7068 kvm_x86_ops->vcpu_free(vcpu);
7069}
7070
7071struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7072 unsigned int id)
7073{
c447e76b
LL		 7074 struct kvm_vcpu *vcpu;
7075
6755bae8
ZA		 7076 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7077 printk_once(KERN_WARNING
7078 "kvm: SMP vm created on host with unstable TSC; "
7079 "guest TSC will not be reliable\n");
c447e76b
LL		 7080
7081 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7082
c447e76b 7083 return vcpu;
26e5215f 7084}
e9b11c17 7085
26e5215f
AK		 7086int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7087{
7088 int r;
e9b11c17 7089
19efffa2 7090 kvm_vcpu_mtrr_init(vcpu);
9fc77441
MT		 7091 r = vcpu_load(vcpu);
7092 if (r)
7093 return r;
d28bc9dd 7094 kvm_vcpu_reset(vcpu, false);
8a3c1a33 7095 kvm_mmu_setup(vcpu);
e9b11c17 7096 vcpu_put(vcpu);
26e5215f 7097 return r;
e9b11c17
ZX		 7098}
7099
31928aa5 7100void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
42897d86 7101{
8fe8ab46 7102 struct msr_data msr;
332967a3 7103 struct kvm *kvm = vcpu->kvm;
42897d86 7104
31928aa5
DD		 7105 if (vcpu_load(vcpu))
7106 return;
8fe8ab46
WA		 7107 msr.data = 0x0;
7108 msr.index = MSR_IA32_TSC;
7109 msr.host_initiated = true;
7110 kvm_write_tsc(vcpu, &msr);
42897d86
MT		 7111 vcpu_put(vcpu);
7112
630994b3
MT		 7113 if (!kvmclock_periodic_sync)
7114 return;
7115
332967a3
AJ		 7116 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7117 KVMCLOCK_SYNC_PERIOD);
42897d86
MT		 7118}
7119
d40ccc62 7120void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
e9b11c17 7121{
9fc77441 7122 int r;
344d9588
GN		 7123 vcpu->arch.apf.msr_val = 0;
7124
9fc77441
MT		 7125 r = vcpu_load(vcpu);
7126 BUG_ON(r);
e9b11c17
ZX		 7127 kvm_mmu_unload(vcpu);
7128 vcpu_put(vcpu);
7129
7130 kvm_x86_ops->vcpu_free(vcpu);
7131}
7132
d28bc9dd 7133void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
e9b11c17 7134{
e69fab5d
PB		 7135 vcpu->arch.hflags = 0;
7136
7460fb4a
AK		 7137 atomic_set(&vcpu->arch.nmi_queued, 0);
7138 vcpu->arch.nmi_pending = 0;
448fa4a9 7139 vcpu->arch.nmi_injected = false;
5f7552d4
NA		 7140 kvm_clear_interrupt_queue(vcpu);
7141 kvm_clear_exception_queue(vcpu);
448fa4a9 7142
42dbaa5a 7143 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
ae561ede 7144 kvm_update_dr0123(vcpu);
6f43ed01 7145 vcpu->arch.dr6 = DR6_INIT;
73aaf249 7146 kvm_update_dr6(vcpu);
42dbaa5a 7147 vcpu->arch.dr7 = DR7_FIXED_1;
c8639010 7148 kvm_update_dr7(vcpu);
42dbaa5a 7149
1119022c
NA		 7150 vcpu->arch.cr2 = 0;
7151
3842d135 7152 kvm_make_request(KVM_REQ_EVENT, vcpu);
344d9588 7153 vcpu->arch.apf.msr_val = 0;
c9aaa895 7154 vcpu->arch.st.msr_val = 0;
3842d135 7155
12f9a48f
GC		 7156 kvmclock_reset(vcpu);
7157
af585b92
GN		 7158 kvm_clear_async_pf_completion_queue(vcpu);
7159 kvm_async_pf_hash_reset(vcpu);
7160 vcpu->arch.apf.halted = false;
3842d135 7161
64d60670 7162 if (!init_event) {
d28bc9dd 7163 kvm_pmu_reset(vcpu);
64d60670
PB		 7164 vcpu->arch.smbase = 0x30000;
7165 }
f5132b01 7166
66f7b72e
JS		 7167 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7168 vcpu->arch.regs_avail = ~0;
7169 vcpu->arch.regs_dirty = ~0;
7170
d28bc9dd 7171 kvm_x86_ops->vcpu_reset(vcpu, init_event);
e9b11c17
ZX		 7172}
7173
2b4a273b 7174void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
66450a21
JK		 7175{
7176 struct kvm_segment cs;
7177
7178 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7179 cs.selector = vector << 8;
7180 cs.base = vector << 12;
7181 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7182 kvm_rip_write(vcpu, 0);
e9b11c17
ZX		 7183}
7184
13a34e06 7185int kvm_arch_hardware_enable(void)
e9b11c17 7186{
ca84d1a2
ZA		 7187 struct kvm *kvm;
7188 struct kvm_vcpu *vcpu;
7189 int i;
0dd6a6ed
ZA		 7190 int ret;
7191 u64 local_tsc;
7192 u64 max_tsc = 0;
7193 bool stable, backwards_tsc = false;
18863bdd
AK		 7194
7195 kvm_shared_msr_cpu_online();
13a34e06 7196 ret = kvm_x86_ops->hardware_enable();
0dd6a6ed
ZA		 7197 if (ret != 0)
7198 return ret;
7199
7200 local_tsc = native_read_tsc();
7201 stable = !check_tsc_unstable();
7202 list_for_each_entry(kvm, &vm_list, vm_list) {
7203 kvm_for_each_vcpu(i, vcpu, kvm) {
7204 if (!stable && vcpu->cpu == smp_processor_id())
105b21bb 7205 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 7206 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7207 backwards_tsc = true;
7208 if (vcpu->arch.last_host_tsc > max_tsc)
7209 max_tsc = vcpu->arch.last_host_tsc;
7210 }
7211 }
7212 }
7213
7214 /*
7215 * Sometimes, even reliable TSCs go backwards. This happens on
7216 * platforms that reset TSC during suspend or hibernate actions, but
7217 * maintain synchronization. We must compensate. Fortunately, we can
7218 * detect that condition here, which happens early in CPU bringup,
7219 * before any KVM threads can be running. Unfortunately, we can't
7220 * bring the TSCs fully up to date with real time, as we aren't yet far
7221 * enough into CPU bringup that we know how much real time has actually
7222 * elapsed; our helper function, get_kernel_ns() will be using boot
7223 * variables that haven't been updated yet.
7224 *
7225 * So we simply find the maximum observed TSC above, then record the
7226 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7227 * the adjustment will be applied. Note that we accumulate
7228 * adjustments, in case multiple suspend cycles happen before some VCPU
7229 * gets a chance to run again. In the event that no KVM threads get a
7230 * chance to run, we will miss the entire elapsed period, as we'll have
7231 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7232 * loose cycle time. This isn't too big a deal, since the loss will be
7233 * uniform across all VCPUs (not to mention the scenario is extremely
7234 * unlikely). It is possible that a second hibernate recovery happens
7235 * much faster than a first, causing the observed TSC here to be
7236 * smaller; this would require additional padding adjustment, which is
7237 * why we set last_host_tsc to the local tsc observed here.
7238 *
7239 * N.B. - this code below runs only on platforms with reliable TSC,
7240 * as that is the only way backwards_tsc is set above. Also note
7241 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7242 * have the same delta_cyc adjustment applied if backwards_tsc
7243 * is detected. Note further, this adjustment is only done once,
7244 * as we reset last_host_tsc on all VCPUs to stop this from being
7245 * called multiple times (one for each physical CPU bringup).
7246 *
4a969980 7247 * Platforms with unreliable TSCs don't have to deal with this, they
0dd6a6ed
ZA		 7248 * will be compensated by the logic in vcpu_load, which sets the TSC to
7249 * catchup mode. This will catchup all VCPUs to real time, but cannot
7250 * guarantee that they stay in perfect synchronization.
7251 */
7252 if (backwards_tsc) {
7253 u64 delta_cyc = max_tsc - local_tsc;
16a96021 7254 backwards_tsc_observed = true;
0dd6a6ed
ZA		 7255 list_for_each_entry(kvm, &vm_list, vm_list) {
7256 kvm_for_each_vcpu(i, vcpu, kvm) {
7257 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7258 vcpu->arch.last_host_tsc = local_tsc;
105b21bb 7259 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 7260 }
7261
7262 /*
7263 * We have to disable TSC offset matching.. if you were
7264 * booting a VM while issuing an S4 host suspend....
7265 * you may have some problem. Solving this issue is
7266 * left as an exercise to the reader.
7267 */
7268 kvm->arch.last_tsc_nsec = 0;
7269 kvm->arch.last_tsc_write = 0;
7270 }
7271
7272 }
7273 return 0;
e9b11c17
ZX		 7274}
7275
13a34e06 7276void kvm_arch_hardware_disable(void)
e9b11c17 7277{
13a34e06
RK		 7278 kvm_x86_ops->hardware_disable();
7279 drop_user_return_notifiers();
e9b11c17
ZX		 7280}
7281
7282int kvm_arch_hardware_setup(void)
7283{
9e9c3fe4
NA		 7284 int r;
7285
7286 r = kvm_x86_ops->hardware_setup();
7287 if (r != 0)
7288 return r;
7289
7290 kvm_init_msr_list();
7291 return 0;
e9b11c17
ZX		 7292}
7293
7294void kvm_arch_hardware_unsetup(void)
7295{
7296 kvm_x86_ops->hardware_unsetup();
7297}
7298
7299void kvm_arch_check_processor_compat(void *rtn)
7300{
7301 kvm_x86_ops->check_processor_compatibility(rtn);
d71ba788
PB		 7302}
7303
7304bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7305{
7306 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7307}
7308EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7309
7310bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7311{
7312 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
e9b11c17
ZX		 7313}
7314
3e515705
AK		 7315bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7316{
7317 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7318}
7319
54e9818f
GN		 7320struct static_key kvm_no_apic_vcpu __read_mostly;
7321
e9b11c17
ZX		 7322int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7323{
7324 struct page *page;
7325 struct kvm *kvm;
7326 int r;
7327
7328 BUG_ON(vcpu->kvm == NULL);
7329 kvm = vcpu->kvm;
7330
6aef266c 7331 vcpu->arch.pv.pv_unhalted = false;
9aabc88f 7332 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
58d269d8 7333 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
a4535290 7334 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
e9b11c17 7335 else
a4535290 7336 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
e9b11c17
ZX		 7337
7338 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7339 if (!page) {
7340 r = -ENOMEM;
7341 goto fail;
7342 }
ad312c7c 7343 vcpu->arch.pio_data = page_address(page);
e9b11c17 7344
cc578287 7345 kvm_set_tsc_khz(vcpu, max_tsc_khz);
c285545f 7346
e9b11c17
ZX		 7347 r = kvm_mmu_create(vcpu);
7348 if (r < 0)
7349 goto fail_free_pio_data;
7350
7351 if (irqchip_in_kernel(kvm)) {
7352 r = kvm_create_lapic(vcpu);
7353 if (r < 0)
7354 goto fail_mmu_destroy;
54e9818f
GN		 7355 } else
7356 static_key_slow_inc(&kvm_no_apic_vcpu);
e9b11c17 7357
890ca9ae
HY		 7358 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7359 GFP_KERNEL);
7360 if (!vcpu->arch.mce_banks) {
7361 r = -ENOMEM;
443c39bc 7362 goto fail_free_lapic;
890ca9ae
HY		 7363 }
7364 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7365
f1797359
WY		 7366 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7367 r = -ENOMEM;
f5f48ee1 7368 goto fail_free_mce_banks;
f1797359 7369 }
f5f48ee1 7370
0ee6a517 7371 fx_init(vcpu);
66f7b72e 7372
ba904635 7373 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
0b79459b 7374 vcpu->arch.pv_time_enabled = false;
d7876f1b
PB		 7375
7376 vcpu->arch.guest_supported_xcr0 = 0;
4344ee98 7377 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
d7876f1b 7378
5a4f55cd
EK		 7379 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7380
74545705
RK		 7381 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7382
af585b92 7383 kvm_async_pf_hash_reset(vcpu);
f5132b01 7384 kvm_pmu_init(vcpu);
af585b92 7385
e9b11c17 7386 return 0;
0ee6a517 7387
f5f48ee1
SY		 7388fail_free_mce_banks:
7389 kfree(vcpu->arch.mce_banks);
443c39bc
WY		 7390fail_free_lapic:
7391 kvm_free_lapic(vcpu);
e9b11c17
ZX		 7392fail_mmu_destroy:
7393 kvm_mmu_destroy(vcpu);
7394fail_free_pio_data:
ad312c7c 7395 free_page((unsigned long)vcpu->arch.pio_data);
e9b11c17
ZX		 7396fail:
7397 return r;
7398}
7399
7400void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7401{
f656ce01
MT		 7402 int idx;
7403
f5132b01 7404 kvm_pmu_destroy(vcpu);
36cb93fd 7405 kfree(vcpu->arch.mce_banks);
e9b11c17 7406 kvm_free_lapic(vcpu);
f656ce01 7407 idx = srcu_read_lock(&vcpu->kvm->srcu);
e9b11c17 7408 kvm_mmu_destroy(vcpu);
f656ce01 7409 srcu_read_unlock(&vcpu->kvm->srcu, idx);
ad312c7c 7410 free_page((unsigned long)vcpu->arch.pio_data);
54e9818f
GN		 7411 if (!irqchip_in_kernel(vcpu->kvm))
7412 static_key_slow_dec(&kvm_no_apic_vcpu);
e9b11c17 7413}
d19a9cd2 7414
e790d9ef
RK		 7415void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7416{
ae97a3b8 7417 kvm_x86_ops->sched_in(vcpu, cpu);
e790d9ef
RK		 7418}
7419
e08b9637 7420int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
d19a9cd2 7421{
e08b9637
CO		 7422 if (type)
7423 return -EINVAL;
7424
6ef768fa 7425 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
f05e70ac 7426 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
365c8868 7427 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
4d5c5d0f 7428 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
e0f0bbc5 7429 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
d19a9cd2 7430
5550af4d
SY		 7431 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7432 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7a84428a
AW		 7433 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7434 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7435 &kvm->arch.irq_sources_bitmap);
5550af4d 7436
038f8c11 7437 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
1e08ec4a 7438 mutex_init(&kvm->arch.apic_map_lock);
d828199e
MT		 7439 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7440
7441 pvclock_update_vm_gtod_copy(kvm);
53f658b3 7442
7e44e449 7443 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
332967a3 7444 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7e44e449 7445
d89f5eff 7446 return 0;
d19a9cd2
ZX		 7447}
7448
7449static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7450{
9fc77441
MT		 7451 int r;
7452 r = vcpu_load(vcpu);
7453 BUG_ON(r);
d19a9cd2
ZX		 7454 kvm_mmu_unload(vcpu);
7455 vcpu_put(vcpu);
7456}
7457
7458static void kvm_free_vcpus(struct kvm *kvm)
7459{
7460 unsigned int i;
988a2cae 7461 struct kvm_vcpu *vcpu;
d19a9cd2
ZX		 7462
7463 /*
7464 * Unpin any mmu pages first.
7465 */
af585b92
GN		 7466 kvm_for_each_vcpu(i, vcpu, kvm) {
7467 kvm_clear_async_pf_completion_queue(vcpu);
988a2cae 7468 kvm_unload_vcpu_mmu(vcpu);
af585b92 7469 }
988a2cae
GN		 7470 kvm_for_each_vcpu(i, vcpu, kvm)
7471 kvm_arch_vcpu_free(vcpu);
7472
7473 mutex_lock(&kvm->lock);
7474 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7475 kvm->vcpus[i] = NULL;
d19a9cd2 7476
988a2cae
GN		 7477 atomic_set(&kvm->online_vcpus, 0);
7478 mutex_unlock(&kvm->lock);
d19a9cd2
ZX		 7479}
7480
ad8ba2cd
SY		 7481void kvm_arch_sync_events(struct kvm *kvm)
7482{
332967a3 7483 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7e44e449 7484 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
ba4cef31 7485 kvm_free_all_assigned_devices(kvm);
aea924f6 7486 kvm_free_pit(kvm);
ad8ba2cd
SY		 7487}
7488
9da0e4d5
PB		 7489int __x86_set_memory_region(struct kvm *kvm,
7490 const struct kvm_userspace_memory_region *mem)
7491{
7492 int i, r;
7493
7494 /* Called with kvm->slots_lock held. */
7495 BUG_ON(mem->slot >= KVM_MEM_SLOTS_NUM);
7496
7497 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7498 struct kvm_userspace_memory_region m = *mem;
7499
7500 m.slot |= i << 16;
7501 r = __kvm_set_memory_region(kvm, &m);
7502 if (r < 0)
7503 return r;
7504 }
7505
7506 return 0;
7507}
7508EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7509
7510int x86_set_memory_region(struct kvm *kvm,
7511 const struct kvm_userspace_memory_region *mem)
7512{
7513 int r;
7514
7515 mutex_lock(&kvm->slots_lock);
7516 r = __x86_set_memory_region(kvm, mem);
7517 mutex_unlock(&kvm->slots_lock);
7518
7519 return r;
7520}
7521EXPORT_SYMBOL_GPL(x86_set_memory_region);
7522
d19a9cd2
ZX		 7523void kvm_arch_destroy_vm(struct kvm *kvm)
7524{
27469d29
AH		 7525 if (current->mm == kvm->mm) {
7526 /*
7527 * Free memory regions allocated on behalf of userspace,
7528 * unless the the memory map has changed due to process exit
7529 * or fd copying.
7530 */
7531 struct kvm_userspace_memory_region mem;
7532 memset(&mem, 0, sizeof(mem));
7533 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
9da0e4d5 7534 x86_set_memory_region(kvm, &mem);
27469d29
AH		 7535
7536 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
9da0e4d5 7537 x86_set_memory_region(kvm, &mem);
27469d29
AH		 7538
7539 mem.slot = TSS_PRIVATE_MEMSLOT;
9da0e4d5 7540 x86_set_memory_region(kvm, &mem);
27469d29 7541 }
6eb55818 7542 kvm_iommu_unmap_guest(kvm);
d7deeeb0
ZX		 7543 kfree(kvm->arch.vpic);
7544 kfree(kvm->arch.vioapic);
d19a9cd2 7545 kvm_free_vcpus(kvm);
1e08ec4a 7546 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
d19a9cd2 7547}
0de10343 7548
5587027c 7549void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
db3fe4eb
TY		 7550 struct kvm_memory_slot *dont)
7551{
7552 int i;
7553
d89cc617
TY		 7554 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7555 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
548ef284 7556 kvfree(free->arch.rmap[i]);
d89cc617 7557 free->arch.rmap[i] = NULL;
77d11309 7558 }
d89cc617
TY		 7559 if (i == 0)
7560 continue;
7561
7562 if (!dont || free->arch.lpage_info[i - 1] !=
7563 dont->arch.lpage_info[i - 1]) {
548ef284 7564 kvfree(free->arch.lpage_info[i - 1]);
d89cc617 7565 free->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 7566 }
7567 }
7568}
7569
5587027c
AK		 7570int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7571 unsigned long npages)
db3fe4eb
TY		 7572{
7573 int i;
7574
d89cc617 7575 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
db3fe4eb
TY		 7576 unsigned long ugfn;
7577 int lpages;
d89cc617 7578 int level = i + 1;
db3fe4eb
TY		 7579
7580 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7581 slot->base_gfn, level) + 1;
7582
d89cc617
TY		 7583 slot->arch.rmap[i] =
7584 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7585 if (!slot->arch.rmap[i])
77d11309 7586 goto out_free;
d89cc617
TY		 7587 if (i == 0)
7588 continue;
77d11309 7589
d89cc617
TY		 7590 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7591 sizeof(*slot->arch.lpage_info[i - 1]));
7592 if (!slot->arch.lpage_info[i - 1])
db3fe4eb
TY		 7593 goto out_free;
7594
7595 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
d89cc617 7596 slot->arch.lpage_info[i - 1][0].write_count = 1;
db3fe4eb 7597 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
d89cc617 7598 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
db3fe4eb
TY		 7599 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7600 /*
7601 * If the gfn and userspace address are not aligned wrt each
7602 * other, or if explicitly asked to, disable large page
7603 * support for this slot
7604 */
7605 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7606 !kvm_largepages_enabled()) {
7607 unsigned long j;
7608
7609 for (j = 0; j < lpages; ++j)
d89cc617 7610 slot->arch.lpage_info[i - 1][j].write_count = 1;
db3fe4eb
TY		 7611 }
7612 }
7613
7614 return 0;
7615
7616out_free:
d89cc617 7617 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
548ef284 7618 kvfree(slot->arch.rmap[i]);
d89cc617
TY		 7619 slot->arch.rmap[i] = NULL;
7620 if (i == 0)
7621 continue;
7622
548ef284 7623 kvfree(slot->arch.lpage_info[i - 1]);
d89cc617 7624 slot->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 7625 }
7626 return -ENOMEM;
7627}
7628
15f46015 7629void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
e59dbe09 7630{
e6dff7d1
TY		 7631 /*
7632 * memslots->generation has been incremented.
7633 * mmio generation may have reached its maximum value.
7634 */
54bf36aa 7635 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
e59dbe09
TY		 7636}
7637
f7784b8e
MT		 7638int kvm_arch_prepare_memory_region(struct kvm *kvm,
7639 struct kvm_memory_slot *memslot,
09170a49 7640 const struct kvm_userspace_memory_region *mem,
7b6195a9 7641 enum kvm_mr_change change)
0de10343 7642{
7a905b14
TY		 7643 /*
7644 * Only private memory slots need to be mapped here since
7645 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
0de10343 7646 */
7b6195a9 7647 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7a905b14 7648 unsigned long userspace_addr;
604b38ac 7649
7a905b14
TY		 7650 /*
7651 * MAP_SHARED to prevent internal slot pages from being moved
7652 * by fork()/COW.
7653 */
7b6195a9 7654 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7a905b14
TY		 7655 PROT_READ | PROT_WRITE,
7656 MAP_SHARED | MAP_ANONYMOUS, 0);
0de10343 7657
7a905b14
TY		 7658 if (IS_ERR((void *)userspace_addr))
7659 return PTR_ERR((void *)userspace_addr);
604b38ac 7660
7a905b14 7661 memslot->userspace_addr = userspace_addr;
0de10343
ZX		 7662 }
7663
f7784b8e
MT		 7664 return 0;
7665}
7666
88178fd4
KH		 7667static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7668 struct kvm_memory_slot *new)
7669{
7670 /* Still write protect RO slot */
7671 if (new->flags & KVM_MEM_READONLY) {
7672 kvm_mmu_slot_remove_write_access(kvm, new);
7673 return;
7674 }
7675
7676 /*
7677 * Call kvm_x86_ops dirty logging hooks when they are valid.
7678 *
7679 * kvm_x86_ops->slot_disable_log_dirty is called when:
7680 *
7681 * - KVM_MR_CREATE with dirty logging is disabled
7682 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7683 *
7684 * The reason is, in case of PML, we need to set D-bit for any slots
7685 * with dirty logging disabled in order to eliminate unnecessary GPA
7686 * logging in PML buffer (and potential PML buffer full VMEXT). This
7687 * guarantees leaving PML enabled during guest's lifetime won't have
7688 * any additonal overhead from PML when guest is running with dirty
7689 * logging disabled for memory slots.
7690 *
7691 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7692 * to dirty logging mode.
7693 *
7694 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7695 *
7696 * In case of write protect:
7697 *
7698 * Write protect all pages for dirty logging.
7699 *
7700 * All the sptes including the large sptes which point to this
7701 * slot are set to readonly. We can not create any new large
7702 * spte on this slot until the end of the logging.
7703 *
7704 * See the comments in fast_page_fault().
7705 */
7706 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7707 if (kvm_x86_ops->slot_enable_log_dirty)
7708 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7709 else
7710 kvm_mmu_slot_remove_write_access(kvm, new);
7711 } else {
7712 if (kvm_x86_ops->slot_disable_log_dirty)
7713 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7714 }
7715}
7716
f7784b8e 7717void kvm_arch_commit_memory_region(struct kvm *kvm,
09170a49 7718 const struct kvm_userspace_memory_region *mem,
8482644a 7719 const struct kvm_memory_slot *old,
f36f3f28 7720 const struct kvm_memory_slot *new,
8482644a 7721 enum kvm_mr_change change)
f7784b8e 7722{
8482644a 7723 int nr_mmu_pages = 0;
f7784b8e 7724
f36f3f28 7725 if (change == KVM_MR_DELETE && old->id >= KVM_USER_MEM_SLOTS) {
f7784b8e
MT		 7726 int ret;
7727
8482644a
TY		 7728 ret = vm_munmap(old->userspace_addr,
7729 old->npages * PAGE_SIZE);
f7784b8e
MT		 7730 if (ret < 0)
7731 printk(KERN_WARNING
7732 "kvm_vm_ioctl_set_memory_region: "
7733 "failed to munmap memory\n");
7734 }
7735
48c0e4e9
XG		 7736 if (!kvm->arch.n_requested_mmu_pages)
7737 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7738
48c0e4e9 7739 if (nr_mmu_pages)
0de10343 7740 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
1c91cad4 7741
3ea3b7fa
WL		 7742 /*
7743 * Dirty logging tracks sptes in 4k granularity, meaning that large
7744 * sptes have to be split. If live migration is successful, the guest
7745 * in the source machine will be destroyed and large sptes will be
7746 * created in the destination. However, if the guest continues to run
7747 * in the source machine (for example if live migration fails), small
7748 * sptes will remain around and cause bad performance.
7749 *
7750 * Scan sptes if dirty logging has been stopped, dropping those
7751 * which can be collapsed into a single large-page spte. Later
7752 * page faults will create the large-page sptes.
7753 */
7754 if ((change != KVM_MR_DELETE) &&
7755 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7756 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7757 kvm_mmu_zap_collapsible_sptes(kvm, new);
7758
c972f3b1 7759 /*
88178fd4 7760 * Set up write protection and/or dirty logging for the new slot.
c126d94f 7761 *
88178fd4
KH		 7762 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7763 * been zapped so no dirty logging staff is needed for old slot. For
7764 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7765 * new and it's also covered when dealing with the new slot.
f36f3f28
PB		 7766 *
7767 * FIXME: const-ify all uses of struct kvm_memory_slot.
c972f3b1 7768 */
88178fd4 7769 if (change != KVM_MR_DELETE)
f36f3f28 7770 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
0de10343 7771}
1d737c8a 7772
2df72e9b 7773void kvm_arch_flush_shadow_all(struct kvm *kvm)
34d4cb8f 7774{
6ca18b69 7775 kvm_mmu_invalidate_zap_all_pages(kvm);
34d4cb8f
MT		 7776}
7777
2df72e9b
MT		 7778void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7779 struct kvm_memory_slot *slot)
7780{
6ca18b69 7781 kvm_mmu_invalidate_zap_all_pages(kvm);
2df72e9b
MT		 7782}
7783
1d737c8a
ZX		 7784int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7785{
b6b8a145
JK		 7786 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7787 kvm_x86_ops->check_nested_events(vcpu, false);
7788
af585b92
GN		 7789 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7790 !vcpu->arch.apf.halted)
7791 || !list_empty_careful(&vcpu->async_pf.done)
66450a21 7792 || kvm_apic_has_events(vcpu)
6aef266c 7793 || vcpu->arch.pv.pv_unhalted
7460fb4a 7794 || atomic_read(&vcpu->arch.nmi_queued) ||
a1b37100
GN		 7795 (kvm_arch_interrupt_allowed(vcpu) &&
7796 kvm_cpu_has_interrupt(vcpu));
1d737c8a 7797}
5736199a 7798
b6d33834 7799int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
5736199a 7800{
b6d33834 7801 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
5736199a 7802}
78646121
GN		 7803
7804int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7805{
7806 return kvm_x86_ops->interrupt_allowed(vcpu);
7807}
229456fc 7808
82b32774 7809unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
f92653ee 7810{
82b32774
NA		 7811 if (is_64_bit_mode(vcpu))
7812 return kvm_rip_read(vcpu);
7813 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
7814 kvm_rip_read(vcpu));
7815}
7816EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
f92653ee 7817
82b32774
NA		 7818bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7819{
7820 return kvm_get_linear_rip(vcpu) == linear_rip;
f92653ee
JK		 7821}
7822EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7823
94fe45da
JK		 7824unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7825{
7826 unsigned long rflags;
7827
7828 rflags = kvm_x86_ops->get_rflags(vcpu);
7829 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
c310bac5 7830 rflags &= ~X86_EFLAGS_TF;
94fe45da
JK		 7831 return rflags;
7832}
7833EXPORT_SYMBOL_GPL(kvm_get_rflags);
7834
6addfc42 7835static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
94fe45da
JK		 7836{
7837 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
f92653ee 7838 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
c310bac5 7839 rflags |= X86_EFLAGS_TF;
94fe45da 7840 kvm_x86_ops->set_rflags(vcpu, rflags);
6addfc42
PB		 7841}
7842
7843void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7844{
7845 __kvm_set_rflags(vcpu, rflags);
3842d135 7846 kvm_make_request(KVM_REQ_EVENT, vcpu);
94fe45da
JK		 7847}
7848EXPORT_SYMBOL_GPL(kvm_set_rflags);
7849
56028d08
GN		 7850void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7851{
7852 int r;
7853
fb67e14f 7854 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
f2e10669 7855 work->wakeup_all)
56028d08
GN		 7856 return;
7857
7858 r = kvm_mmu_reload(vcpu);
7859 if (unlikely(r))
7860 return;
7861
fb67e14f
XG		 7862 if (!vcpu->arch.mmu.direct_map &&
7863 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7864 return;
7865
56028d08
GN		 7866 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7867}
7868
af585b92
GN		 7869static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7870{
7871 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7872}
7873
7874static inline u32 kvm_async_pf_next_probe(u32 key)
7875{
7876 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7877}
7878
7879static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7880{
7881 u32 key = kvm_async_pf_hash_fn(gfn);
7882
7883 while (vcpu->arch.apf.gfns[key] != ~0)
7884 key = kvm_async_pf_next_probe(key);
7885
7886 vcpu->arch.apf.gfns[key] = gfn;
7887}
7888
7889static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7890{
7891 int i;
7892 u32 key = kvm_async_pf_hash_fn(gfn);
7893
7894 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
c7d28c24
XG		 7895 (vcpu->arch.apf.gfns[key] != gfn &&
7896 vcpu->arch.apf.gfns[key] != ~0); i++)
af585b92
GN		 7897 key = kvm_async_pf_next_probe(key);
7898
7899 return key;
7900}
7901
7902bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7903{
7904 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7905}
7906
7907static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7908{
7909 u32 i, j, k;
7910
7911 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7912 while (true) {
7913 vcpu->arch.apf.gfns[i] = ~0;
7914 do {
7915 j = kvm_async_pf_next_probe(j);
7916 if (vcpu->arch.apf.gfns[j] == ~0)
7917 return;
7918 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7919 /*
7920 * k lies cyclically in ]i,j]
7921 * | i.k.j |
7922 * |....j i.k.| or |.k..j i...|
7923 */
7924 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7925 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7926 i = j;
7927 }
7928}
7929
7c90705b
GN		 7930static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7931{
7932
7933 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7934 sizeof(val));
7935}
7936
af585b92
GN		 7937void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7938 struct kvm_async_pf *work)
7939{
6389ee94
AK		 7940 struct x86_exception fault;
7941
7c90705b 7942 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
af585b92 7943 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7c90705b
GN		 7944
7945 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
fc5f06fa
GN		 7946 (vcpu->arch.apf.send_user_only &&
7947 kvm_x86_ops->get_cpl(vcpu) == 0))
7c90705b
GN		 7948 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7949 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6389ee94
AK		 7950 fault.vector = PF_VECTOR;
7951 fault.error_code_valid = true;
7952 fault.error_code = 0;
7953 fault.nested_page_fault = false;
7954 fault.address = work->arch.token;
7955 kvm_inject_page_fault(vcpu, &fault);
7c90705b 7956 }
af585b92
GN		 7957}
7958
7959void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7960 struct kvm_async_pf *work)
7961{
6389ee94
AK		 7962 struct x86_exception fault;
7963
7c90705b 7964 trace_kvm_async_pf_ready(work->arch.token, work->gva);
f2e10669 7965 if (work->wakeup_all)
7c90705b
GN		 7966 work->arch.token = ~0; /* broadcast wakeup */
7967 else
7968 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7969
7970 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7971 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6389ee94
AK		 7972 fault.vector = PF_VECTOR;
7973 fault.error_code_valid = true;
7974 fault.error_code = 0;
7975 fault.nested_page_fault = false;
7976 fault.address = work->arch.token;
7977 kvm_inject_page_fault(vcpu, &fault);
7c90705b 7978 }
e6d53e3b 7979 vcpu->arch.apf.halted = false;
a4fa1635 7980 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7c90705b
GN		 7981}
7982
7983bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7984{
7985 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7986 return true;
7987 else
7988 return !kvm_event_needs_reinjection(vcpu) &&
7989 kvm_x86_ops->interrupt_allowed(vcpu);
af585b92
GN		 7990}
7991
5544eb9b
PB		 7992void kvm_arch_start_assignment(struct kvm *kvm)
7993{
7994 atomic_inc(&kvm->arch.assigned_device_count);
7995}
7996EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
7997
7998void kvm_arch_end_assignment(struct kvm *kvm)
7999{
8000 atomic_dec(&kvm->arch.assigned_device_count);
8001}
8002EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8003
8004bool kvm_arch_has_assigned_device(struct kvm *kvm)
8005{
8006 return atomic_read(&kvm->arch.assigned_device_count);
8007}
8008EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8009
e0f0bbc5
AW		 8010void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8011{
8012 atomic_inc(&kvm->arch.noncoherent_dma_count);
8013}
8014EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8015
8016void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8017{
8018 atomic_dec(&kvm->arch.noncoherent_dma_count);
8019}
8020EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8021
8022bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8023{
8024 return atomic_read(&kvm->arch.noncoherent_dma_count);
8025}
8026EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8027
229456fc
MT		 8028EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8029EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8030EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8031EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8032EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
0ac406de 8033EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
d8cabddf 8034EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
17897f36 8035EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
236649de 8036EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
ec1ff790 8037EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
532a46b9 8038EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
2e554e8d 8039EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
489223ed 8040EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7b46268d 8041EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
843e4330 8042EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
Linux kernel - Deliverables
This page took 1.571966 seconds and 5 git commands to generate.