]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - arch/x86/kvm/x86.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: x86: unify handling of interrupt window
[mirror_ubuntu-bionic-kernel.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) },
62bea5bf 152 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
ba1389b7 153 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
f11c3a8d 154 { "hypercalls", VCPU_STAT(hypercalls) },
ba1389b7
AK		 155 { "request_irq", VCPU_STAT(request_irq_exits) },
156 { "irq_exits", VCPU_STAT(irq_exits) },
157 { "host_state_reload", VCPU_STAT(host_state_reload) },
158 { "efer_reload", VCPU_STAT(efer_reload) },
159 { "fpu_reload", VCPU_STAT(fpu_reload) },
160 { "insn_emulation", VCPU_STAT(insn_emulation) },
161 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
fa89a817 162 { "irq_injections", VCPU_STAT(irq_injections) },
c4abb7c9 163 { "nmi_injections", VCPU_STAT(nmi_injections) },
4cee5764
AK		 164 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
165 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
166 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
167 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
168 { "mmu_flooded", VM_STAT(mmu_flooded) },
169 { "mmu_recycled", VM_STAT(mmu_recycled) },
dfc5aa00 170 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
4731d4c7 171 { "mmu_unsync", VM_STAT(mmu_unsync) },
0f74a24c 172 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
05da4558 173 { "largepages", VM_STAT(lpages) },
417bc304
HB		 174 { NULL }
175};
176
2acf923e
DC		 177u64 __read_mostly host_xcr0;
178
b6785def 179static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
d6aa1000 180
af585b92
GN		 181static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
182{
183 int i;
184 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
185 vcpu->arch.apf.gfns[i] = ~0;
186}
187
18863bdd
AK		 188static void kvm_on_user_return(struct user_return_notifier *urn)
189{
190 unsigned slot;
18863bdd
AK		 191 struct kvm_shared_msrs *locals
192 = container_of(urn, struct kvm_shared_msrs, urn);
2bf78fa7 193 struct kvm_shared_msr_values *values;
18863bdd
AK		 194
195 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
2bf78fa7
SY		 196 values = &locals->values[slot];
197 if (values->host != values->curr) {
198 wrmsrl(shared_msrs_global.msrs[slot], values->host);
199 values->curr = values->host;
18863bdd
AK		 200 }
201 }
202 locals->registered = false;
203 user_return_notifier_unregister(urn);
204}
205
2bf78fa7 206static void shared_msr_update(unsigned slot, u32 msr)
18863bdd 207{
18863bdd 208 u64 value;
013f6a5d
MT		 209 unsigned int cpu = smp_processor_id();
210 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
18863bdd 211
2bf78fa7
SY		 212 /* only read, and nobody should modify it at this time,
213 * so don't need lock */
214 if (slot >= shared_msrs_global.nr) {
215 printk(KERN_ERR "kvm: invalid MSR slot!");
216 return;
217 }
218 rdmsrl_safe(msr, &value);
219 smsr->values[slot].host = value;
220 smsr->values[slot].curr = value;
221}
222
223void kvm_define_shared_msr(unsigned slot, u32 msr)
224{
0123be42 225 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
c847fe88 226 shared_msrs_global.msrs[slot] = msr;
18863bdd
AK		 227 if (slot >= shared_msrs_global.nr)
228 shared_msrs_global.nr = slot + 1;
18863bdd
AK		 229}
230EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
231
232static void kvm_shared_msr_cpu_online(void)
233{
234 unsigned i;
18863bdd
AK		 235
236 for (i = 0; i < shared_msrs_global.nr; ++i)
2bf78fa7 237 shared_msr_update(i, shared_msrs_global.msrs[i]);
18863bdd
AK		 238}
239
8b3c3104 240int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
18863bdd 241{
013f6a5d
MT		 242 unsigned int cpu = smp_processor_id();
243 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
8b3c3104 244 int err;
18863bdd 245
2bf78fa7 246 if (((value ^ smsr->values[slot].curr) & mask) == 0)
8b3c3104 247 return 0;
2bf78fa7 248 smsr->values[slot].curr = value;
8b3c3104
AH		 249 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
250 if (err)
251 return 1;
252
18863bdd
AK		 253 if (!smsr->registered) {
254 smsr->urn.on_user_return = kvm_on_user_return;
255 user_return_notifier_register(&smsr->urn);
256 smsr->registered = true;
257 }
8b3c3104 258 return 0;
18863bdd
AK		 259}
260EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
261
13a34e06 262static void drop_user_return_notifiers(void)
3548bab5 263{
013f6a5d
MT		 264 unsigned int cpu = smp_processor_id();
265 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
3548bab5
AK		 266
267 if (smsr->registered)
268 kvm_on_user_return(&smsr->urn);
269}
270
6866b83e
CO		 271u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
272{
8a5a87d9 273 return vcpu->arch.apic_base;
6866b83e
CO		 274}
275EXPORT_SYMBOL_GPL(kvm_get_apic_base);
276
58cb628d
JK		 277int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
278{
279 u64 old_state = vcpu->arch.apic_base &
280 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
281 u64 new_state = msr_info->data &
282 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
283 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
284 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
285
286 if (!msr_info->host_initiated &&
287 ((msr_info->data & reserved_bits) != 0 ||
288 new_state == X2APIC_ENABLE ||
289 (new_state == MSR_IA32_APICBASE_ENABLE &&
290 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
291 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
292 old_state == 0)))
293 return 1;
294
295 kvm_lapic_set_base(vcpu, msr_info->data);
296 return 0;
6866b83e
CO		 297}
298EXPORT_SYMBOL_GPL(kvm_set_apic_base);
299
2605fc21 300asmlinkage __visible void kvm_spurious_fault(void)
e3ba45b8
GL		 301{
302 /* Fault while not rebooting. We want the trace. */
303 BUG();
304}
305EXPORT_SYMBOL_GPL(kvm_spurious_fault);
306
3fd28fce
ED		 307#define EXCPT_BENIGN 0
308#define EXCPT_CONTRIBUTORY 1
309#define EXCPT_PF 2
310
311static int exception_class(int vector)
312{
313 switch (vector) {
314 case PF_VECTOR:
315 return EXCPT_PF;
316 case DE_VECTOR:
317 case TS_VECTOR:
318 case NP_VECTOR:
319 case SS_VECTOR:
320 case GP_VECTOR:
321 return EXCPT_CONTRIBUTORY;
322 default:
323 break;
324 }
325 return EXCPT_BENIGN;
326}
327
d6e8c854
NA		 328#define EXCPT_FAULT 0
329#define EXCPT_TRAP 1
330#define EXCPT_ABORT 2
331#define EXCPT_INTERRUPT 3
332
333static int exception_type(int vector)
334{
335 unsigned int mask;
336
337 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
338 return EXCPT_INTERRUPT;
339
340 mask = 1 << vector;
341
342 /* #DB is trap, as instruction watchpoints are handled elsewhere */
343 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
344 return EXCPT_TRAP;
345
346 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
347 return EXCPT_ABORT;
348
349 /* Reserved exceptions will result in fault */
350 return EXCPT_FAULT;
351}
352
3fd28fce 353static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
ce7ddec4
JR		 354 unsigned nr, bool has_error, u32 error_code,
355 bool reinject)
3fd28fce
ED		 356{
357 u32 prev_nr;
358 int class1, class2;
359
3842d135
AK		 360 kvm_make_request(KVM_REQ_EVENT, vcpu);
361
3fd28fce
ED		 362 if (!vcpu->arch.exception.pending) {
363 queue:
3ffb2468
NA		 364 if (has_error && !is_protmode(vcpu))
365 has_error = false;
3fd28fce
ED		 366 vcpu->arch.exception.pending = true;
367 vcpu->arch.exception.has_error_code = has_error;
368 vcpu->arch.exception.nr = nr;
369 vcpu->arch.exception.error_code = error_code;
3f0fd292 370 vcpu->arch.exception.reinject = reinject;
3fd28fce
ED		 371 return;
372 }
373
374 /* to check exception */
375 prev_nr = vcpu->arch.exception.nr;
376 if (prev_nr == DF_VECTOR) {
377 /* triple fault -> shutdown */
a8eeb04a 378 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3fd28fce
ED		 379 return;
380 }
381 class1 = exception_class(prev_nr);
382 class2 = exception_class(nr);
383 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
384 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
385 /* generate double fault per SDM Table 5-5 */
386 vcpu->arch.exception.pending = true;
387 vcpu->arch.exception.has_error_code = true;
388 vcpu->arch.exception.nr = DF_VECTOR;
389 vcpu->arch.exception.error_code = 0;
390 } else
391 /* replace previous exception with a new one in a hope
392 that instruction re-execution will regenerate lost
393 exception */
394 goto queue;
395}
396
298101da
AK		 397void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
398{
ce7ddec4 399 kvm_multiple_exception(vcpu, nr, false, 0, false);
298101da
AK		 400}
401EXPORT_SYMBOL_GPL(kvm_queue_exception);
402
ce7ddec4
JR		 403void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
404{
405 kvm_multiple_exception(vcpu, nr, false, 0, true);
406}
407EXPORT_SYMBOL_GPL(kvm_requeue_exception);
408
db8fcefa 409void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
c3c91fee 410{
db8fcefa
AP		 411 if (err)
412 kvm_inject_gp(vcpu, 0);
413 else
414 kvm_x86_ops->skip_emulated_instruction(vcpu);
415}
416EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
8df25a32 417
6389ee94 418void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
c3c91fee
AK		 419{
420 ++vcpu->stat.pf_guest;
6389ee94
AK		 421 vcpu->arch.cr2 = fault->address;
422 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
c3c91fee 423}
27d6c865 424EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
c3c91fee 425
ef54bcfe 426static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
d4f8cf66 427{
6389ee94
AK		 428 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
429 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
d4f8cf66 430 else
6389ee94 431 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
ef54bcfe
PB		 432
433 return fault->nested_page_fault;
d4f8cf66
JR		 434}
435
3419ffc8
SY		 436void kvm_inject_nmi(struct kvm_vcpu *vcpu)
437{
7460fb4a
AK		 438 atomic_inc(&vcpu->arch.nmi_queued);
439 kvm_make_request(KVM_REQ_NMI, vcpu);
3419ffc8
SY		 440}
441EXPORT_SYMBOL_GPL(kvm_inject_nmi);
442
298101da
AK		 443void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
444{
ce7ddec4 445 kvm_multiple_exception(vcpu, nr, true, error_code, false);
298101da
AK		 446}
447EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
448
ce7ddec4
JR		 449void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
450{
451 kvm_multiple_exception(vcpu, nr, true, error_code, true);
452}
453EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
454
0a79b009
AK		 455/*
456 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
457 * a #GP and return false.
458 */
459bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
298101da 460{
0a79b009
AK		 461 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
462 return true;
463 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
464 return false;
298101da 465}
0a79b009 466EXPORT_SYMBOL_GPL(kvm_require_cpl);
298101da 467
16f8a6f9
NA		 468bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
469{
470 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
471 return true;
472
473 kvm_queue_exception(vcpu, UD_VECTOR);
474 return false;
475}
476EXPORT_SYMBOL_GPL(kvm_require_dr);
477
ec92fe44
JR		 478/*
479 * This function will be used to read from the physical memory of the currently
54bf36aa 480 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
ec92fe44
JR		 481 * can read from guest physical or from the guest's guest physical memory.
482 */
483int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
484 gfn_t ngfn, void *data, int offset, int len,
485 u32 access)
486{
54987b7a 487 struct x86_exception exception;
ec92fe44
JR		 488 gfn_t real_gfn;
489 gpa_t ngpa;
490
491 ngpa = gfn_to_gpa(ngfn);
54987b7a 492 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
ec92fe44
JR		 493 if (real_gfn == UNMAPPED_GVA)
494 return -EFAULT;
495
496 real_gfn = gpa_to_gfn(real_gfn);
497
54bf36aa 498 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
ec92fe44
JR		 499}
500EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
501
69b0049a 502static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3d06b8bf
JR		 503 void *data, int offset, int len, u32 access)
504{
505 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
506 data, offset, len, access);
507}
508
a03490ed
CO		 509/*
510 * Load the pae pdptrs. Return true is they are all valid.
511 */
ff03a073 512int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
a03490ed
CO		 513{
514 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
515 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
516 int i;
517 int ret;
ff03a073 518 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
a03490ed 519
ff03a073
JR		 520 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
521 offset * sizeof(u64), sizeof(pdpte),
522 PFERR_USER_MASK|PFERR_WRITE_MASK);
a03490ed
CO		 523 if (ret < 0) {
524 ret = 0;
525 goto out;
526 }
527 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
43a3795a 528 if (is_present_gpte(pdpte[i]) &&
a0a64f50
XG		 529 (pdpte[i] &
530 vcpu->arch.mmu.guest_rsvd_check.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;
35754c98 791 if (lapic_in_kernel(vcpu))
a03490ed
CO		 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 800{
35754c98 801 if (lapic_in_kernel(vcpu))
a03490ed
CO		 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{
03b9730b
AL		 1445 cycle_t ret = (cycle_t)rdtsc_ordered();
1446 u64 last = pvclock_gtod_data.clock.cycle_last;
d828199e
MT		 1447
1448 if (likely(ret >= last))
1449 return ret;
1450
1451 /*
1452 * GCC likes to generate cmov here, but this branch is extremely
1453 * predictable (it's just a funciton of time and the likely is
1454 * very likely) and there's a data dependence, so force GCC
1455 * to generate a branch instead. I don't barrier() because
1456 * we don't actually need a barrier, and if this function
1457 * ever gets inlined it will generate worse code.
1458 */
1459 asm volatile ("");
1460 return last;
1461}
1462
1463static inline u64 vgettsc(cycle_t *cycle_now)
1464{
1465 long v;
1466 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1467
1468 *cycle_now = read_tsc();
1469
1470 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1471 return v * gtod->clock.mult;
1472}
1473
cbcf2dd3 1474static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
d828199e 1475{
cbcf2dd3 1476 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
d828199e 1477 unsigned long seq;
d828199e 1478 int mode;
cbcf2dd3 1479 u64 ns;
d828199e 1480
d828199e
MT		 1481 do {
1482 seq = read_seqcount_begin(&gtod->seq);
1483 mode = gtod->clock.vclock_mode;
cbcf2dd3 1484 ns = gtod->nsec_base;
d828199e
MT		 1485 ns += vgettsc(cycle_now);
1486 ns >>= gtod->clock.shift;
cbcf2dd3 1487 ns += gtod->boot_ns;
d828199e 1488 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
cbcf2dd3 1489 *t = ns;
d828199e
MT		 1490
1491 return mode;
1492}
1493
1494/* returns true if host is using tsc clocksource */
1495static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1496{
d828199e
MT		 1497 /* checked again under seqlock below */
1498 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1499 return false;
1500
cbcf2dd3 1501 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
d828199e
MT		 1502}
1503#endif
1504
1505/*
1506 *
b48aa97e
MT		 1507 * Assuming a stable TSC across physical CPUS, and a stable TSC
1508 * across virtual CPUs, the following condition is possible.
1509 * Each numbered line represents an event visible to both
d828199e
MT		 1510 * CPUs at the next numbered event.
1511 *
1512 * "timespecX" represents host monotonic time. "tscX" represents
1513 * RDTSC value.
1514 *
1515 * VCPU0 on CPU0 | VCPU1 on CPU1
1516 *
1517 * 1. read timespec0,tsc0
1518 * 2. | timespec1 = timespec0 + N
1519 * | tsc1 = tsc0 + M
1520 * 3. transition to guest | transition to guest
1521 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1522 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1523 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1524 *
1525 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1526 *
1527 * - ret0 < ret1
1528 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1529 * ...
1530 * - 0 < N - M => M < N
1531 *
1532 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1533 * always the case (the difference between two distinct xtime instances
1534 * might be smaller then the difference between corresponding TSC reads,
1535 * when updating guest vcpus pvclock areas).
1536 *
1537 * To avoid that problem, do not allow visibility of distinct
1538 * system_timestamp/tsc_timestamp values simultaneously: use a master
1539 * copy of host monotonic time values. Update that master copy
1540 * in lockstep.
1541 *
b48aa97e 1542 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
d828199e
MT		 1543 *
1544 */
1545
1546static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1547{
1548#ifdef CONFIG_X86_64
1549 struct kvm_arch *ka = &kvm->arch;
1550 int vclock_mode;
b48aa97e
MT		 1551 bool host_tsc_clocksource, vcpus_matched;
1552
1553 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1554 atomic_read(&kvm->online_vcpus));
d828199e
MT		 1555
1556 /*
1557 * If the host uses TSC clock, then passthrough TSC as stable
1558 * to the guest.
1559 */
b48aa97e 1560 host_tsc_clocksource = kvm_get_time_and_clockread(
d828199e
MT		 1561 &ka->master_kernel_ns,
1562 &ka->master_cycle_now);
1563
16a96021 1564 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
54750f2c
MT		 1565 && !backwards_tsc_observed
1566 && !ka->boot_vcpu_runs_old_kvmclock;
b48aa97e 1567
d828199e
MT		 1568 if (ka->use_master_clock)
1569 atomic_set(&kvm_guest_has_master_clock, 1);
1570
1571 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
b48aa97e
MT		 1572 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1573 vcpus_matched);
d828199e
MT		 1574#endif
1575}
1576
2e762ff7
MT		 1577static void kvm_gen_update_masterclock(struct kvm *kvm)
1578{
1579#ifdef CONFIG_X86_64
1580 int i;
1581 struct kvm_vcpu *vcpu;
1582 struct kvm_arch *ka = &kvm->arch;
1583
1584 spin_lock(&ka->pvclock_gtod_sync_lock);
1585 kvm_make_mclock_inprogress_request(kvm);
1586 /* no guest entries from this point */
1587 pvclock_update_vm_gtod_copy(kvm);
1588
1589 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 1590 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2e762ff7
MT		 1591
1592 /* guest entries allowed */
1593 kvm_for_each_vcpu(i, vcpu, kvm)
1594 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1595
1596 spin_unlock(&ka->pvclock_gtod_sync_lock);
1597#endif
1598}
1599
34c238a1 1600static int kvm_guest_time_update(struct kvm_vcpu *v)
18068523 1601{
d828199e 1602 unsigned long flags, this_tsc_khz;
18068523 1603 struct kvm_vcpu_arch *vcpu = &v->arch;
d828199e 1604 struct kvm_arch *ka = &v->kvm->arch;
f25e656d 1605 s64 kernel_ns;
d828199e 1606 u64 tsc_timestamp, host_tsc;
0b79459b 1607 struct pvclock_vcpu_time_info guest_hv_clock;
51d59c6b 1608 u8 pvclock_flags;
d828199e
MT		 1609 bool use_master_clock;
1610
1611 kernel_ns = 0;
1612 host_tsc = 0;
18068523 1613
d828199e
MT		 1614 /*
1615 * If the host uses TSC clock, then passthrough TSC as stable
1616 * to the guest.
1617 */
1618 spin_lock(&ka->pvclock_gtod_sync_lock);
1619 use_master_clock = ka->use_master_clock;
1620 if (use_master_clock) {
1621 host_tsc = ka->master_cycle_now;
1622 kernel_ns = ka->master_kernel_ns;
1623 }
1624 spin_unlock(&ka->pvclock_gtod_sync_lock);
c09664bb
MT		 1625
1626 /* Keep irq disabled to prevent changes to the clock */
1627 local_irq_save(flags);
89cbc767 1628 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
c09664bb
MT		 1629 if (unlikely(this_tsc_khz == 0)) {
1630 local_irq_restore(flags);
1631 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1632 return 1;
1633 }
d828199e 1634 if (!use_master_clock) {
4ea1636b 1635 host_tsc = rdtsc();
d828199e
MT		 1636 kernel_ns = get_kernel_ns();
1637 }
1638
1639 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1640
c285545f
ZA		 1641 /*
1642 * We may have to catch up the TSC to match elapsed wall clock
1643 * time for two reasons, even if kvmclock is used.
1644 * 1) CPU could have been running below the maximum TSC rate
1645 * 2) Broken TSC compensation resets the base at each VCPU
1646 * entry to avoid unknown leaps of TSC even when running
1647 * again on the same CPU. This may cause apparent elapsed
1648 * time to disappear, and the guest to stand still or run
1649 * very slowly.
1650 */
1651 if (vcpu->tsc_catchup) {
1652 u64 tsc = compute_guest_tsc(v, kernel_ns);
1653 if (tsc > tsc_timestamp) {
f1e2b260 1654 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
c285545f
ZA		 1655 tsc_timestamp = tsc;
1656 }
50d0a0f9
GH		 1657 }
1658
18068523
GOC		 1659 local_irq_restore(flags);
1660
0b79459b 1661 if (!vcpu->pv_time_enabled)
c285545f 1662 return 0;
18068523 1663
e48672fa 1664 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
5f4e3f88
ZA		 1665 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1666 &vcpu->hv_clock.tsc_shift,
1667 &vcpu->hv_clock.tsc_to_system_mul);
e48672fa 1668 vcpu->hw_tsc_khz = this_tsc_khz;
8cfdc000
ZA		 1669 }
1670
1671 /* With all the info we got, fill in the values */
1d5f066e 1672 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
759379dd 1673 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
28e4639a 1674 vcpu->last_guest_tsc = tsc_timestamp;
51d59c6b 1675
09a0c3f1
OH		 1676 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1677 &guest_hv_clock, sizeof(guest_hv_clock))))
1678 return 0;
1679
5dca0d91
RK		 1680 /* This VCPU is paused, but it's legal for a guest to read another
1681 * VCPU's kvmclock, so we really have to follow the specification where
1682 * it says that version is odd if data is being modified, and even after
1683 * it is consistent.
1684 *
1685 * Version field updates must be kept separate. This is because
1686 * kvm_write_guest_cached might use a "rep movs" instruction, and
1687 * writes within a string instruction are weakly ordered. So there
1688 * are three writes overall.
1689 *
1690 * As a small optimization, only write the version field in the first
1691 * and third write. The vcpu->pv_time cache is still valid, because the
1692 * version field is the first in the struct.
18068523 1693 */
5dca0d91
RK		 1694 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1695
1696 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1697 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1698 &vcpu->hv_clock,
1699 sizeof(vcpu->hv_clock.version));
1700
1701 smp_wmb();
78c0337a
MT		 1702
1703 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
0b79459b 1704 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
78c0337a
MT		 1705
1706 if (vcpu->pvclock_set_guest_stopped_request) {
1707 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1708 vcpu->pvclock_set_guest_stopped_request = false;
1709 }
1710
d828199e
MT		 1711 /* If the host uses TSC clocksource, then it is stable */
1712 if (use_master_clock)
1713 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1714
78c0337a
MT		 1715 vcpu->hv_clock.flags = pvclock_flags;
1716
ce1a5e60
DM		 1717 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1718
0b79459b
AH		 1719 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1720 &vcpu->hv_clock,
1721 sizeof(vcpu->hv_clock));
5dca0d91
RK		 1722
1723 smp_wmb();
1724
1725 vcpu->hv_clock.version++;
1726 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1727 &vcpu->hv_clock,
1728 sizeof(vcpu->hv_clock.version));
8cfdc000 1729 return 0;
c8076604
GH		 1730}
1731
0061d53d
MT		 1732/*
1733 * kvmclock updates which are isolated to a given vcpu, such as
1734 * vcpu->cpu migration, should not allow system_timestamp from
1735 * the rest of the vcpus to remain static. Otherwise ntp frequency
1736 * correction applies to one vcpu's system_timestamp but not
1737 * the others.
1738 *
1739 * So in those cases, request a kvmclock update for all vcpus.
7e44e449
AJ		 1740 * We need to rate-limit these requests though, as they can
1741 * considerably slow guests that have a large number of vcpus.
1742 * The time for a remote vcpu to update its kvmclock is bound
1743 * by the delay we use to rate-limit the updates.
0061d53d
MT		 1744 */
1745
7e44e449
AJ		 1746#define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1747
1748static void kvmclock_update_fn(struct work_struct *work)
0061d53d
MT		 1749{
1750 int i;
7e44e449
AJ		 1751 struct delayed_work *dwork = to_delayed_work(work);
1752 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1753 kvmclock_update_work);
1754 struct kvm *kvm = container_of(ka, struct kvm, arch);
0061d53d
MT		 1755 struct kvm_vcpu *vcpu;
1756
1757 kvm_for_each_vcpu(i, vcpu, kvm) {
105b21bb 1758 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0061d53d
MT		 1759 kvm_vcpu_kick(vcpu);
1760 }
1761}
1762
7e44e449
AJ		 1763static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1764{
1765 struct kvm *kvm = v->kvm;
1766
105b21bb 1767 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
7e44e449
AJ		 1768 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1769 KVMCLOCK_UPDATE_DELAY);
1770}
1771
332967a3
AJ		 1772#define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1773
1774static void kvmclock_sync_fn(struct work_struct *work)
1775{
1776 struct delayed_work *dwork = to_delayed_work(work);
1777 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1778 kvmclock_sync_work);
1779 struct kvm *kvm = container_of(ka, struct kvm, arch);
1780
630994b3
MT		 1781 if (!kvmclock_periodic_sync)
1782 return;
1783
332967a3
AJ		 1784 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1785 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1786 KVMCLOCK_SYNC_PERIOD);
1787}
1788
890ca9ae 1789static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
15c4a640 1790{
890ca9ae
HY		 1791 u64 mcg_cap = vcpu->arch.mcg_cap;
1792 unsigned bank_num = mcg_cap & 0xff;
1793
15c4a640 1794 switch (msr) {
15c4a640 1795 case MSR_IA32_MCG_STATUS:
890ca9ae 1796 vcpu->arch.mcg_status = data;
15c4a640 1797 break;
c7ac679c 1798 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 1799 if (!(mcg_cap & MCG_CTL_P))
1800 return 1;
1801 if (data != 0 && data != ~(u64)0)
1802 return -1;
1803 vcpu->arch.mcg_ctl = data;
1804 break;
1805 default:
1806 if (msr >= MSR_IA32_MC0_CTL &&
81760dcc 1807 msr < MSR_IA32_MCx_CTL(bank_num)) {
890ca9ae 1808 u32 offset = msr - MSR_IA32_MC0_CTL;
114be429
AP		 1809 /* only 0 or all 1s can be written to IA32_MCi_CTL
1810 * some Linux kernels though clear bit 10 in bank 4 to
1811 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1812 * this to avoid an uncatched #GP in the guest
1813 */
890ca9ae 1814 if ((offset & 0x3) == 0 &&
114be429 1815 data != 0 && (data | (1 << 10)) != ~(u64)0)
890ca9ae
HY		 1816 return -1;
1817 vcpu->arch.mce_banks[offset] = data;
1818 break;
1819 }
1820 return 1;
1821 }
1822 return 0;
1823}
1824
ffde22ac
ES		 1825static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1826{
1827 struct kvm *kvm = vcpu->kvm;
1828 int lm = is_long_mode(vcpu);
1829 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1830 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1831 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1832 : kvm->arch.xen_hvm_config.blob_size_32;
1833 u32 page_num = data & ~PAGE_MASK;
1834 u64 page_addr = data & PAGE_MASK;
1835 u8 *page;
1836 int r;
1837
1838 r = -E2BIG;
1839 if (page_num >= blob_size)
1840 goto out;
1841 r = -ENOMEM;
ff5c2c03
SL		 1842 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1843 if (IS_ERR(page)) {
1844 r = PTR_ERR(page);
ffde22ac 1845 goto out;
ff5c2c03 1846 }
54bf36aa 1847 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
ffde22ac
ES		 1848 goto out_free;
1849 r = 0;
1850out_free:
1851 kfree(page);
1852out:
1853 return r;
1854}
1855
344d9588
GN		 1856static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1857{
1858 gpa_t gpa = data & ~0x3f;
1859
4a969980 1860 /* Bits 2:5 are reserved, Should be zero */
6adba527 1861 if (data & 0x3c)
344d9588
GN		 1862 return 1;
1863
1864 vcpu->arch.apf.msr_val = data;
1865
1866 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1867 kvm_clear_async_pf_completion_queue(vcpu);
1868 kvm_async_pf_hash_reset(vcpu);
1869 return 0;
1870 }
1871
8f964525
AH		 1872 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1873 sizeof(u32)))
344d9588
GN		 1874 return 1;
1875
6adba527 1876 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
344d9588
GN		 1877 kvm_async_pf_wakeup_all(vcpu);
1878 return 0;
1879}
1880
12f9a48f
GC		 1881static void kvmclock_reset(struct kvm_vcpu *vcpu)
1882{
0b79459b 1883 vcpu->arch.pv_time_enabled = false;
12f9a48f
GC		 1884}
1885
c9aaa895
GC		 1886static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1887{
1888 u64 delta;
1889
1890 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1891 return;
1892
1893 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1894 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1895 vcpu->arch.st.accum_steal = delta;
1896}
1897
1898static void record_steal_time(struct kvm_vcpu *vcpu)
1899{
1900 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1901 return;
1902
1903 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1904 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1905 return;
1906
1907 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1908 vcpu->arch.st.steal.version += 2;
1909 vcpu->arch.st.accum_steal = 0;
1910
1911 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1912 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1913}
1914
8fe8ab46 1915int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
15c4a640 1916{
5753785f 1917 bool pr = false;
8fe8ab46
WA		 1918 u32 msr = msr_info->index;
1919 u64 data = msr_info->data;
5753785f 1920
15c4a640 1921 switch (msr) {
2e32b719
BP		 1922 case MSR_AMD64_NB_CFG:
1923 case MSR_IA32_UCODE_REV:
1924 case MSR_IA32_UCODE_WRITE:
1925 case MSR_VM_HSAVE_PA:
1926 case MSR_AMD64_PATCH_LOADER:
1927 case MSR_AMD64_BU_CFG2:
1928 break;
1929
15c4a640 1930 case MSR_EFER:
b69e8cae 1931 return set_efer(vcpu, data);
8f1589d9
AP		 1932 case MSR_K7_HWCR:
1933 data &= ~(u64)0x40; /* ignore flush filter disable */
82494028 1934 data &= ~(u64)0x100; /* ignore ignne emulation enable */
a223c313 1935 data &= ~(u64)0x8; /* ignore TLB cache disable */
22d48b2d 1936 data &= ~(u64)0x40000; /* ignore Mc status write enable */
8f1589d9 1937 if (data != 0) {
a737f256
CD		 1938 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1939 data);
8f1589d9
AP		 1940 return 1;
1941 }
15c4a640 1942 break;
f7c6d140
AP		 1943 case MSR_FAM10H_MMIO_CONF_BASE:
1944 if (data != 0) {
a737f256
CD		 1945 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1946 "0x%llx\n", data);
f7c6d140
AP		 1947 return 1;
1948 }
15c4a640 1949 break;
b5e2fec0
AG		 1950 case MSR_IA32_DEBUGCTLMSR:
1951 if (!data) {
1952 /* We support the non-activated case already */
1953 break;
1954 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1955 /* Values other than LBR and BTF are vendor-specific,
1956 thus reserved and should throw a #GP */
1957 return 1;
1958 }
a737f256
CD		 1959 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1960 __func__, data);
b5e2fec0 1961 break;
9ba075a6 1962 case 0x200 ... 0x2ff:
ff53604b 1963 return kvm_mtrr_set_msr(vcpu, msr, data);
15c4a640 1964 case MSR_IA32_APICBASE:
58cb628d 1965 return kvm_set_apic_base(vcpu, msr_info);
0105d1a5
GN		 1966 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1967 return kvm_x2apic_msr_write(vcpu, msr, data);
a3e06bbe
LJ		 1968 case MSR_IA32_TSCDEADLINE:
1969 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1970 break;
ba904635
WA		 1971 case MSR_IA32_TSC_ADJUST:
1972 if (guest_cpuid_has_tsc_adjust(vcpu)) {
1973 if (!msr_info->host_initiated) {
d913b904 1974 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
d7add054 1975 adjust_tsc_offset_guest(vcpu, adj);
ba904635
WA		 1976 }
1977 vcpu->arch.ia32_tsc_adjust_msr = data;
1978 }
1979 break;
15c4a640 1980 case MSR_IA32_MISC_ENABLE:
ad312c7c 1981 vcpu->arch.ia32_misc_enable_msr = data;
15c4a640 1982 break;
64d60670
PB		 1983 case MSR_IA32_SMBASE:
1984 if (!msr_info->host_initiated)
1985 return 1;
1986 vcpu->arch.smbase = data;
1987 break;
11c6bffa 1988 case MSR_KVM_WALL_CLOCK_NEW:
18068523
GOC		 1989 case MSR_KVM_WALL_CLOCK:
1990 vcpu->kvm->arch.wall_clock = data;
1991 kvm_write_wall_clock(vcpu->kvm, data);
1992 break;
11c6bffa 1993 case MSR_KVM_SYSTEM_TIME_NEW:
18068523 1994 case MSR_KVM_SYSTEM_TIME: {
0b79459b 1995 u64 gpa_offset;
54750f2c
MT		 1996 struct kvm_arch *ka = &vcpu->kvm->arch;
1997
12f9a48f 1998 kvmclock_reset(vcpu);
18068523 1999
54750f2c
MT		 2000 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2001 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2002
2003 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2004 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2005 &vcpu->requests);
2006
2007 ka->boot_vcpu_runs_old_kvmclock = tmp;
2008 }
2009
18068523 2010 vcpu->arch.time = data;
0061d53d 2011 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
18068523
GOC		 2012
2013 /* we verify if the enable bit is set... */
2014 if (!(data & 1))
2015 break;
2016
0b79459b 2017 gpa_offset = data & ~(PAGE_MASK | 1);
18068523 2018
0b79459b 2019 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
8f964525
AH		 2020 &vcpu->arch.pv_time, data & ~1ULL,
2021 sizeof(struct pvclock_vcpu_time_info)))
0b79459b
AH		 2022 vcpu->arch.pv_time_enabled = false;
2023 else
2024 vcpu->arch.pv_time_enabled = true;
32cad84f 2025
18068523
GOC		 2026 break;
2027 }
344d9588
GN		 2028 case MSR_KVM_ASYNC_PF_EN:
2029 if (kvm_pv_enable_async_pf(vcpu, data))
2030 return 1;
2031 break;
c9aaa895
GC		 2032 case MSR_KVM_STEAL_TIME:
2033
2034 if (unlikely(!sched_info_on()))
2035 return 1;
2036
2037 if (data & KVM_STEAL_RESERVED_MASK)
2038 return 1;
2039
2040 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
8f964525
AH		 2041 data & KVM_STEAL_VALID_BITS,
2042 sizeof(struct kvm_steal_time)))
c9aaa895
GC		 2043 return 1;
2044
2045 vcpu->arch.st.msr_val = data;
2046
2047 if (!(data & KVM_MSR_ENABLED))
2048 break;
2049
2050 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2051
2052 preempt_disable();
2053 accumulate_steal_time(vcpu);
2054 preempt_enable();
2055
2056 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2057
2058 break;
ae7a2a3f
MT		 2059 case MSR_KVM_PV_EOI_EN:
2060 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2061 return 1;
2062 break;
c9aaa895 2063
890ca9ae
HY		 2064 case MSR_IA32_MCG_CTL:
2065 case MSR_IA32_MCG_STATUS:
81760dcc 2066 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
890ca9ae 2067 return set_msr_mce(vcpu, msr, data);
71db6023 2068
6912ac32
WH		 2069 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2070 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2071 pr = true; /* fall through */
2072 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2073 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
c6702c9d 2074 if (kvm_pmu_is_valid_msr(vcpu, msr))
afd80d85 2075 return kvm_pmu_set_msr(vcpu, msr_info);
5753785f
GN		 2076
2077 if (pr || data != 0)
a737f256
CD		 2078 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2079 "0x%x data 0x%llx\n", msr, data);
5753785f 2080 break;
84e0cefa
JS		 2081 case MSR_K7_CLK_CTL:
2082 /*
2083 * Ignore all writes to this no longer documented MSR.
2084 * Writes are only relevant for old K7 processors,
2085 * all pre-dating SVM, but a recommended workaround from
4a969980 2086 * AMD for these chips. It is possible to specify the
84e0cefa
JS		 2087 * affected processor models on the command line, hence
2088 * the need to ignore the workaround.
2089 */
2090 break;
55cd8e5a 2091 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
e7d9513b
AS		 2092 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2093 case HV_X64_MSR_CRASH_CTL:
2094 return kvm_hv_set_msr_common(vcpu, msr, data,
2095 msr_info->host_initiated);
91c9c3ed 2096 case MSR_IA32_BBL_CR_CTL3:
2097 /* Drop writes to this legacy MSR -- see rdmsr
2098 * counterpart for further detail.
2099 */
a737f256 2100 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
91c9c3ed 2101 break;
2b036c6b
BO		 2102 case MSR_AMD64_OSVW_ID_LENGTH:
2103 if (!guest_cpuid_has_osvw(vcpu))
2104 return 1;
2105 vcpu->arch.osvw.length = data;
2106 break;
2107 case MSR_AMD64_OSVW_STATUS:
2108 if (!guest_cpuid_has_osvw(vcpu))
2109 return 1;
2110 vcpu->arch.osvw.status = data;
2111 break;
15c4a640 2112 default:
ffde22ac
ES		 2113 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2114 return xen_hvm_config(vcpu, data);
c6702c9d 2115 if (kvm_pmu_is_valid_msr(vcpu, msr))
afd80d85 2116 return kvm_pmu_set_msr(vcpu, msr_info);
ed85c068 2117 if (!ignore_msrs) {
a737f256
CD		 2118 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2119 msr, data);
ed85c068
AP		 2120 return 1;
2121 } else {
a737f256
CD		 2122 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2123 msr, data);
ed85c068
AP		 2124 break;
2125 }
15c4a640
CO		 2126 }
2127 return 0;
2128}
2129EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2130
2131
2132/*
2133 * Reads an msr value (of 'msr_index') into 'pdata'.
2134 * Returns 0 on success, non-0 otherwise.
2135 * Assumes vcpu_load() was already called.
2136 */
609e36d3 2137int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
15c4a640 2138{
609e36d3 2139 return kvm_x86_ops->get_msr(vcpu, msr);
15c4a640 2140}
ff651cb6 2141EXPORT_SYMBOL_GPL(kvm_get_msr);
15c4a640 2142
890ca9ae 2143static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
15c4a640
CO		 2144{
2145 u64 data;
890ca9ae
HY		 2146 u64 mcg_cap = vcpu->arch.mcg_cap;
2147 unsigned bank_num = mcg_cap & 0xff;
15c4a640
CO		 2148
2149 switch (msr) {
15c4a640
CO		 2150 case MSR_IA32_P5_MC_ADDR:
2151 case MSR_IA32_P5_MC_TYPE:
890ca9ae
HY		 2152 data = 0;
2153 break;
15c4a640 2154 case MSR_IA32_MCG_CAP:
890ca9ae
HY		 2155 data = vcpu->arch.mcg_cap;
2156 break;
c7ac679c 2157 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 2158 if (!(mcg_cap & MCG_CTL_P))
2159 return 1;
2160 data = vcpu->arch.mcg_ctl;
2161 break;
2162 case MSR_IA32_MCG_STATUS:
2163 data = vcpu->arch.mcg_status;
2164 break;
2165 default:
2166 if (msr >= MSR_IA32_MC0_CTL &&
81760dcc 2167 msr < MSR_IA32_MCx_CTL(bank_num)) {
890ca9ae
HY		 2168 u32 offset = msr - MSR_IA32_MC0_CTL;
2169 data = vcpu->arch.mce_banks[offset];
2170 break;
2171 }
2172 return 1;
2173 }
2174 *pdata = data;
2175 return 0;
2176}
2177
609e36d3 2178int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
890ca9ae 2179{
609e36d3 2180 switch (msr_info->index) {
890ca9ae 2181 case MSR_IA32_PLATFORM_ID:
15c4a640 2182 case MSR_IA32_EBL_CR_POWERON:
b5e2fec0
AG		 2183 case MSR_IA32_DEBUGCTLMSR:
2184 case MSR_IA32_LASTBRANCHFROMIP:
2185 case MSR_IA32_LASTBRANCHTOIP:
2186 case MSR_IA32_LASTINTFROMIP:
2187 case MSR_IA32_LASTINTTOIP:
60af2ecd 2188 case MSR_K8_SYSCFG:
3afb1121
PB		 2189 case MSR_K8_TSEG_ADDR:
2190 case MSR_K8_TSEG_MASK:
60af2ecd 2191 case MSR_K7_HWCR:
61a6bd67 2192 case MSR_VM_HSAVE_PA:
1fdbd48c 2193 case MSR_K8_INT_PENDING_MSG:
c323c0e5 2194 case MSR_AMD64_NB_CFG:
f7c6d140 2195 case MSR_FAM10H_MMIO_CONF_BASE:
2e32b719 2196 case MSR_AMD64_BU_CFG2:
609e36d3 2197 msr_info->data = 0;
15c4a640 2198 break;
6912ac32
WH		 2199 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2200 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2201 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2202 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
c6702c9d 2203 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
609e36d3
PB		 2204 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2205 msr_info->data = 0;
5753785f 2206 break;
742bc670 2207 case MSR_IA32_UCODE_REV:
609e36d3 2208 msr_info->data = 0x100000000ULL;
742bc670 2209 break;
9ba075a6 2210 case MSR_MTRRcap:
9ba075a6 2211 case 0x200 ... 0x2ff:
ff53604b 2212 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
15c4a640 2213 case 0xcd: /* fsb frequency */
609e36d3 2214 msr_info->data = 3;
15c4a640 2215 break;
7b914098
JS		 2216 /*
2217 * MSR_EBC_FREQUENCY_ID
2218 * Conservative value valid for even the basic CPU models.
2219 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2220 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2221 * and 266MHz for model 3, or 4. Set Core Clock
2222 * Frequency to System Bus Frequency Ratio to 1 (bits
2223 * 31:24) even though these are only valid for CPU
2224 * models > 2, however guests may end up dividing or
2225 * multiplying by zero otherwise.
2226 */
2227 case MSR_EBC_FREQUENCY_ID:
609e36d3 2228 msr_info->data = 1 << 24;
7b914098 2229 break;
15c4a640 2230 case MSR_IA32_APICBASE:
609e36d3 2231 msr_info->data = kvm_get_apic_base(vcpu);
15c4a640 2232 break;
0105d1a5 2233 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
609e36d3 2234 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
0105d1a5 2235 break;
a3e06bbe 2236 case MSR_IA32_TSCDEADLINE:
609e36d3 2237 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
a3e06bbe 2238 break;
ba904635 2239 case MSR_IA32_TSC_ADJUST:
609e36d3 2240 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
ba904635 2241 break;
15c4a640 2242 case MSR_IA32_MISC_ENABLE:
609e36d3 2243 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
15c4a640 2244 break;
64d60670
PB		 2245 case MSR_IA32_SMBASE:
2246 if (!msr_info->host_initiated)
2247 return 1;
2248 msr_info->data = vcpu->arch.smbase;
15c4a640 2249 break;
847f0ad8
AG		 2250 case MSR_IA32_PERF_STATUS:
2251 /* TSC increment by tick */
609e36d3 2252 msr_info->data = 1000ULL;
847f0ad8 2253 /* CPU multiplier */
b0996ae4 2254 msr_info->data |= (((uint64_t)4ULL) << 40);
847f0ad8 2255 break;
15c4a640 2256 case MSR_EFER:
609e36d3 2257 msr_info->data = vcpu->arch.efer;
15c4a640 2258 break;
18068523 2259 case MSR_KVM_WALL_CLOCK:
11c6bffa 2260 case MSR_KVM_WALL_CLOCK_NEW:
609e36d3 2261 msr_info->data = vcpu->kvm->arch.wall_clock;
18068523
GOC		 2262 break;
2263 case MSR_KVM_SYSTEM_TIME:
11c6bffa 2264 case MSR_KVM_SYSTEM_TIME_NEW:
609e36d3 2265 msr_info->data = vcpu->arch.time;
18068523 2266 break;
344d9588 2267 case MSR_KVM_ASYNC_PF_EN:
609e36d3 2268 msr_info->data = vcpu->arch.apf.msr_val;
344d9588 2269 break;
c9aaa895 2270 case MSR_KVM_STEAL_TIME:
609e36d3 2271 msr_info->data = vcpu->arch.st.msr_val;
c9aaa895 2272 break;
1d92128f 2273 case MSR_KVM_PV_EOI_EN:
609e36d3 2274 msr_info->data = vcpu->arch.pv_eoi.msr_val;
1d92128f 2275 break;
890ca9ae
HY		 2276 case MSR_IA32_P5_MC_ADDR:
2277 case MSR_IA32_P5_MC_TYPE:
2278 case MSR_IA32_MCG_CAP:
2279 case MSR_IA32_MCG_CTL:
2280 case MSR_IA32_MCG_STATUS:
81760dcc 2281 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
609e36d3 2282 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
84e0cefa
JS		 2283 case MSR_K7_CLK_CTL:
2284 /*
2285 * Provide expected ramp-up count for K7. All other
2286 * are set to zero, indicating minimum divisors for
2287 * every field.
2288 *
2289 * This prevents guest kernels on AMD host with CPU
2290 * type 6, model 8 and higher from exploding due to
2291 * the rdmsr failing.
2292 */
609e36d3 2293 msr_info->data = 0x20000000;
84e0cefa 2294 break;
55cd8e5a 2295 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
e7d9513b
AS		 2296 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2297 case HV_X64_MSR_CRASH_CTL:
e83d5887
AS		 2298 return kvm_hv_get_msr_common(vcpu,
2299 msr_info->index, &msr_info->data);
55cd8e5a 2300 break;
91c9c3ed 2301 case MSR_IA32_BBL_CR_CTL3:
2302 /* This legacy MSR exists but isn't fully documented in current
2303 * silicon. It is however accessed by winxp in very narrow
2304 * scenarios where it sets bit #19, itself documented as
2305 * a "reserved" bit. Best effort attempt to source coherent
2306 * read data here should the balance of the register be
2307 * interpreted by the guest:
2308 *
2309 * L2 cache control register 3: 64GB range, 256KB size,
2310 * enabled, latency 0x1, configured
2311 */
609e36d3 2312 msr_info->data = 0xbe702111;
91c9c3ed 2313 break;
2b036c6b
BO		 2314 case MSR_AMD64_OSVW_ID_LENGTH:
2315 if (!guest_cpuid_has_osvw(vcpu))
2316 return 1;
609e36d3 2317 msr_info->data = vcpu->arch.osvw.length;
2b036c6b
BO		 2318 break;
2319 case MSR_AMD64_OSVW_STATUS:
2320 if (!guest_cpuid_has_osvw(vcpu))
2321 return 1;
609e36d3 2322 msr_info->data = vcpu->arch.osvw.status;
2b036c6b 2323 break;
15c4a640 2324 default:
c6702c9d 2325 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
609e36d3 2326 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
ed85c068 2327 if (!ignore_msrs) {
609e36d3 2328 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
ed85c068
AP		 2329 return 1;
2330 } else {
609e36d3
PB		 2331 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2332 msr_info->data = 0;
ed85c068
AP		 2333 }
2334 break;
15c4a640 2335 }
15c4a640
CO		 2336 return 0;
2337}
2338EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2339
313a3dc7
CO		 2340/*
2341 * Read or write a bunch of msrs. All parameters are kernel addresses.
2342 *
2343 * @return number of msrs set successfully.
2344 */
2345static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2346 struct kvm_msr_entry *entries,
2347 int (*do_msr)(struct kvm_vcpu *vcpu,
2348 unsigned index, u64 *data))
2349{
f656ce01 2350 int i, idx;
313a3dc7 2351
f656ce01 2352 idx = srcu_read_lock(&vcpu->kvm->srcu);
313a3dc7
CO		 2353 for (i = 0; i < msrs->nmsrs; ++i)
2354 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2355 break;
f656ce01 2356 srcu_read_unlock(&vcpu->kvm->srcu, idx);
313a3dc7 2357
313a3dc7
CO		 2358 return i;
2359}
2360
2361/*
2362 * Read or write a bunch of msrs. Parameters are user addresses.
2363 *
2364 * @return number of msrs set successfully.
2365 */
2366static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2367 int (*do_msr)(struct kvm_vcpu *vcpu,
2368 unsigned index, u64 *data),
2369 int writeback)
2370{
2371 struct kvm_msrs msrs;
2372 struct kvm_msr_entry *entries;
2373 int r, n;
2374 unsigned size;
2375
2376 r = -EFAULT;
2377 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2378 goto out;
2379
2380 r = -E2BIG;
2381 if (msrs.nmsrs >= MAX_IO_MSRS)
2382 goto out;
2383
313a3dc7 2384 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
ff5c2c03
SL		 2385 entries = memdup_user(user_msrs->entries, size);
2386 if (IS_ERR(entries)) {
2387 r = PTR_ERR(entries);
313a3dc7 2388 goto out;
ff5c2c03 2389 }
313a3dc7
CO		 2390
2391 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2392 if (r < 0)
2393 goto out_free;
2394
2395 r = -EFAULT;
2396 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2397 goto out_free;
2398
2399 r = n;
2400
2401out_free:
7a73c028 2402 kfree(entries);
313a3dc7
CO		 2403out:
2404 return r;
2405}
2406
784aa3d7 2407int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
018d00d2
ZX		 2408{
2409 int r;
2410
2411 switch (ext) {
2412 case KVM_CAP_IRQCHIP:
2413 case KVM_CAP_HLT:
2414 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
018d00d2 2415 case KVM_CAP_SET_TSS_ADDR:
07716717 2416 case KVM_CAP_EXT_CPUID:
9c15bb1d 2417 case KVM_CAP_EXT_EMUL_CPUID:
c8076604 2418 case KVM_CAP_CLOCKSOURCE:
7837699f 2419 case KVM_CAP_PIT:
a28e4f5a 2420 case KVM_CAP_NOP_IO_DELAY:
62d9f0db 2421 case KVM_CAP_MP_STATE:
ed848624 2422 case KVM_CAP_SYNC_MMU:
a355c85c 2423 case KVM_CAP_USER_NMI:
52d939a0 2424 case KVM_CAP_REINJECT_CONTROL:
4925663a 2425 case KVM_CAP_IRQ_INJECT_STATUS:
d34e6b17 2426 case KVM_CAP_IOEVENTFD:
f848a5a8 2427 case KVM_CAP_IOEVENTFD_NO_LENGTH:
c5ff41ce 2428 case KVM_CAP_PIT2:
e9f42757 2429 case KVM_CAP_PIT_STATE2:
b927a3ce 2430 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
ffde22ac 2431 case KVM_CAP_XEN_HVM:
afbcf7ab 2432 case KVM_CAP_ADJUST_CLOCK:
3cfc3092 2433 case KVM_CAP_VCPU_EVENTS:
55cd8e5a 2434 case KVM_CAP_HYPERV:
10388a07 2435 case KVM_CAP_HYPERV_VAPIC:
c25bc163 2436 case KVM_CAP_HYPERV_SPIN:
ab9f4ecb 2437 case KVM_CAP_PCI_SEGMENT:
a1efbe77 2438 case KVM_CAP_DEBUGREGS:
d2be1651 2439 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2d5b5a66 2440 case KVM_CAP_XSAVE:
344d9588 2441 case KVM_CAP_ASYNC_PF:
92a1f12d 2442 case KVM_CAP_GET_TSC_KHZ:
1c0b28c2 2443 case KVM_CAP_KVMCLOCK_CTRL:
4d8b81ab 2444 case KVM_CAP_READONLY_MEM:
5f66b620 2445 case KVM_CAP_HYPERV_TIME:
100943c5 2446 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
defcf51f 2447 case KVM_CAP_TSC_DEADLINE_TIMER:
90de4a18
NA		 2448 case KVM_CAP_ENABLE_CAP_VM:
2449 case KVM_CAP_DISABLE_QUIRKS:
d71ba788 2450 case KVM_CAP_SET_BOOT_CPU_ID:
2a5bab10
AW		 2451#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2452 case KVM_CAP_ASSIGN_DEV_IRQ:
2453 case KVM_CAP_PCI_2_3:
2454#endif
018d00d2
ZX		 2455 r = 1;
2456 break;
6d396b55
PB		 2457 case KVM_CAP_X86_SMM:
2458 /* SMBASE is usually relocated above 1M on modern chipsets,
2459 * and SMM handlers might indeed rely on 4G segment limits,
2460 * so do not report SMM to be available if real mode is
2461 * emulated via vm86 mode. Still, do not go to great lengths
2462 * to avoid userspace's usage of the feature, because it is a
2463 * fringe case that is not enabled except via specific settings
2464 * of the module parameters.
2465 */
2466 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2467 break;
542472b5
LV		 2468 case KVM_CAP_COALESCED_MMIO:
2469 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2470 break;
774ead3a
AK		 2471 case KVM_CAP_VAPIC:
2472 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2473 break;
f725230a 2474 case KVM_CAP_NR_VCPUS:
8c3ba334
SL		 2475 r = KVM_SOFT_MAX_VCPUS;
2476 break;
2477 case KVM_CAP_MAX_VCPUS:
f725230a
AK		 2478 r = KVM_MAX_VCPUS;
2479 break;
a988b910 2480 case KVM_CAP_NR_MEMSLOTS:
bbacc0c1 2481 r = KVM_USER_MEM_SLOTS;
a988b910 2482 break;
a68a6a72
MT		 2483 case KVM_CAP_PV_MMU: /* obsolete */
2484 r = 0;
2f333bcb 2485 break;
4cee4b72 2486#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
62c476c7 2487 case KVM_CAP_IOMMU:
a1b60c1c 2488 r = iommu_present(&pci_bus_type);
62c476c7 2489 break;
4cee4b72 2490#endif
890ca9ae
HY		 2491 case KVM_CAP_MCE:
2492 r = KVM_MAX_MCE_BANKS;
2493 break;
2d5b5a66
SY		 2494 case KVM_CAP_XCRS:
2495 r = cpu_has_xsave;
2496 break;
92a1f12d
JR		 2497 case KVM_CAP_TSC_CONTROL:
2498 r = kvm_has_tsc_control;
2499 break;
018d00d2
ZX		 2500 default:
2501 r = 0;
2502 break;
2503 }
2504 return r;
2505
2506}
2507
043405e1
CO		 2508long kvm_arch_dev_ioctl(struct file *filp,
2509 unsigned int ioctl, unsigned long arg)
2510{
2511 void __user *argp = (void __user *)arg;
2512 long r;
2513
2514 switch (ioctl) {
2515 case KVM_GET_MSR_INDEX_LIST: {
2516 struct kvm_msr_list __user *user_msr_list = argp;
2517 struct kvm_msr_list msr_list;
2518 unsigned n;
2519
2520 r = -EFAULT;
2521 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2522 goto out;
2523 n = msr_list.nmsrs;
62ef68bb 2524 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
043405e1
CO		 2525 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2526 goto out;
2527 r = -E2BIG;
e125e7b6 2528 if (n < msr_list.nmsrs)
043405e1
CO		 2529 goto out;
2530 r = -EFAULT;
2531 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2532 num_msrs_to_save * sizeof(u32)))
2533 goto out;
e125e7b6 2534 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
043405e1 2535 &emulated_msrs,
62ef68bb 2536 num_emulated_msrs * sizeof(u32)))
043405e1
CO		 2537 goto out;
2538 r = 0;
2539 break;
2540 }
9c15bb1d
BP		 2541 case KVM_GET_SUPPORTED_CPUID:
2542 case KVM_GET_EMULATED_CPUID: {
674eea0f
AK		 2543 struct kvm_cpuid2 __user *cpuid_arg = argp;
2544 struct kvm_cpuid2 cpuid;
2545
2546 r = -EFAULT;
2547 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2548 goto out;
9c15bb1d
BP		 2549
2550 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2551 ioctl);
674eea0f
AK		 2552 if (r)
2553 goto out;
2554
2555 r = -EFAULT;
2556 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2557 goto out;
2558 r = 0;
2559 break;
2560 }
890ca9ae
HY		 2561 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2562 u64 mce_cap;
2563
2564 mce_cap = KVM_MCE_CAP_SUPPORTED;
2565 r = -EFAULT;
2566 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2567 goto out;
2568 r = 0;
2569 break;
2570 }
043405e1
CO		 2571 default:
2572 r = -EINVAL;
2573 }
2574out:
2575 return r;
2576}
2577
f5f48ee1
SY		 2578static void wbinvd_ipi(void *garbage)
2579{
2580 wbinvd();
2581}
2582
2583static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2584{
e0f0bbc5 2585 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
f5f48ee1
SY		 2586}
2587
313a3dc7
CO		 2588void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2589{
f5f48ee1
SY		 2590 /* Address WBINVD may be executed by guest */
2591 if (need_emulate_wbinvd(vcpu)) {
2592 if (kvm_x86_ops->has_wbinvd_exit())
2593 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2594 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2595 smp_call_function_single(vcpu->cpu,
2596 wbinvd_ipi, NULL, 1);
2597 }
2598
313a3dc7 2599 kvm_x86_ops->vcpu_load(vcpu, cpu);
8f6055cb 2600
0dd6a6ed
ZA		 2601 /* Apply any externally detected TSC adjustments (due to suspend) */
2602 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2603 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2604 vcpu->arch.tsc_offset_adjustment = 0;
105b21bb 2605 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed 2606 }
8f6055cb 2607
48434c20 2608 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
6f526ec5 2609 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4ea1636b 2610 rdtsc() - vcpu->arch.last_host_tsc;
e48672fa
ZA		 2611 if (tsc_delta < 0)
2612 mark_tsc_unstable("KVM discovered backwards TSC");
c285545f 2613 if (check_tsc_unstable()) {
b183aa58
ZA		 2614 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2615 vcpu->arch.last_guest_tsc);
2616 kvm_x86_ops->write_tsc_offset(vcpu, offset);
c285545f 2617 vcpu->arch.tsc_catchup = 1;
c285545f 2618 }
d98d07ca
MT		 2619 /*
2620 * On a host with synchronized TSC, there is no need to update
2621 * kvmclock on vcpu->cpu migration
2622 */
2623 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
0061d53d 2624 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
c285545f
ZA		 2625 if (vcpu->cpu != cpu)
2626 kvm_migrate_timers(vcpu);
e48672fa 2627 vcpu->cpu = cpu;
6b7d7e76 2628 }
c9aaa895
GC		 2629
2630 accumulate_steal_time(vcpu);
2631 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
313a3dc7
CO		 2632}
2633
2634void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2635{
02daab21 2636 kvm_x86_ops->vcpu_put(vcpu);
1c11e713 2637 kvm_put_guest_fpu(vcpu);
4ea1636b 2638 vcpu->arch.last_host_tsc = rdtsc();
313a3dc7
CO		 2639}
2640
313a3dc7
CO		 2641static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2642 struct kvm_lapic_state *s)
2643{
5a71785d 2644 kvm_x86_ops->sync_pir_to_irr(vcpu);
ad312c7c 2645 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
313a3dc7
CO		 2646
2647 return 0;
2648}
2649
2650static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2651 struct kvm_lapic_state *s)
2652{
64eb0620 2653 kvm_apic_post_state_restore(vcpu, s);
cb142eb7 2654 update_cr8_intercept(vcpu);
313a3dc7
CO		 2655
2656 return 0;
2657}
2658
f77bc6a4
ZX		 2659static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2660 struct kvm_interrupt *irq)
2661{
02cdb50f 2662 if (irq->irq >= KVM_NR_INTERRUPTS)
f77bc6a4
ZX		 2663 return -EINVAL;
2664 if (irqchip_in_kernel(vcpu->kvm))
2665 return -ENXIO;
f77bc6a4 2666
66fd3f7f 2667 kvm_queue_interrupt(vcpu, irq->irq, false);
3842d135 2668 kvm_make_request(KVM_REQ_EVENT, vcpu);
f77bc6a4 2669
f77bc6a4
ZX		 2670 return 0;
2671}
2672
c4abb7c9
JK		 2673static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2674{
c4abb7c9 2675 kvm_inject_nmi(vcpu);
c4abb7c9
JK		 2676
2677 return 0;
2678}
2679
f077825a
PB		 2680static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2681{
64d60670
PB		 2682 kvm_make_request(KVM_REQ_SMI, vcpu);
2683
f077825a
PB		 2684 return 0;
2685}
2686
b209749f
AK		 2687static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2688 struct kvm_tpr_access_ctl *tac)
2689{
2690 if (tac->flags)
2691 return -EINVAL;
2692 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2693 return 0;
2694}
2695
890ca9ae
HY		 2696static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2697 u64 mcg_cap)
2698{
2699 int r;
2700 unsigned bank_num = mcg_cap & 0xff, bank;
2701
2702 r = -EINVAL;
a9e38c3e 2703 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
890ca9ae
HY		 2704 goto out;
2705 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2706 goto out;
2707 r = 0;
2708 vcpu->arch.mcg_cap = mcg_cap;
2709 /* Init IA32_MCG_CTL to all 1s */
2710 if (mcg_cap & MCG_CTL_P)
2711 vcpu->arch.mcg_ctl = ~(u64)0;
2712 /* Init IA32_MCi_CTL to all 1s */
2713 for (bank = 0; bank < bank_num; bank++)
2714 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2715out:
2716 return r;
2717}
2718
2719static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2720 struct kvm_x86_mce *mce)
2721{
2722 u64 mcg_cap = vcpu->arch.mcg_cap;
2723 unsigned bank_num = mcg_cap & 0xff;
2724 u64 *banks = vcpu->arch.mce_banks;
2725
2726 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2727 return -EINVAL;
2728 /*
2729 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2730 * reporting is disabled
2731 */
2732 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2733 vcpu->arch.mcg_ctl != ~(u64)0)
2734 return 0;
2735 banks += 4 * mce->bank;
2736 /*
2737 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2738 * reporting is disabled for the bank
2739 */
2740 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2741 return 0;
2742 if (mce->status & MCI_STATUS_UC) {
2743 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
fc78f519 2744 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
a8eeb04a 2745 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
890ca9ae
HY		 2746 return 0;
2747 }
2748 if (banks[1] & MCI_STATUS_VAL)
2749 mce->status |= MCI_STATUS_OVER;
2750 banks[2] = mce->addr;
2751 banks[3] = mce->misc;
2752 vcpu->arch.mcg_status = mce->mcg_status;
2753 banks[1] = mce->status;
2754 kvm_queue_exception(vcpu, MC_VECTOR);
2755 } else if (!(banks[1] & MCI_STATUS_VAL)
2756 || !(banks[1] & MCI_STATUS_UC)) {
2757 if (banks[1] & MCI_STATUS_VAL)
2758 mce->status |= MCI_STATUS_OVER;
2759 banks[2] = mce->addr;
2760 banks[3] = mce->misc;
2761 banks[1] = mce->status;
2762 } else
2763 banks[1] |= MCI_STATUS_OVER;
2764 return 0;
2765}
2766
3cfc3092
JK		 2767static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2768 struct kvm_vcpu_events *events)
2769{
7460fb4a 2770 process_nmi(vcpu);
03b82a30
JK		 2771 events->exception.injected =
2772 vcpu->arch.exception.pending &&
2773 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3cfc3092
JK		 2774 events->exception.nr = vcpu->arch.exception.nr;
2775 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
97e69aa6 2776 events->exception.pad = 0;
3cfc3092
JK		 2777 events->exception.error_code = vcpu->arch.exception.error_code;
2778
03b82a30
JK		 2779 events->interrupt.injected =
2780 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3cfc3092 2781 events->interrupt.nr = vcpu->arch.interrupt.nr;
03b82a30 2782 events->interrupt.soft = 0;
37ccdcbe 2783 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3cfc3092
JK		 2784
2785 events->nmi.injected = vcpu->arch.nmi_injected;
7460fb4a 2786 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3cfc3092 2787 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
97e69aa6 2788 events->nmi.pad = 0;
3cfc3092 2789
66450a21 2790 events->sipi_vector = 0; /* never valid when reporting to user space */
3cfc3092 2791
f077825a
PB		 2792 events->smi.smm = is_smm(vcpu);
2793 events->smi.pending = vcpu->arch.smi_pending;
2794 events->smi.smm_inside_nmi =
2795 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2796 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2797
dab4b911 2798 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
f077825a
PB		 2799 | KVM_VCPUEVENT_VALID_SHADOW
2800 | KVM_VCPUEVENT_VALID_SMM);
97e69aa6 2801 memset(&events->reserved, 0, sizeof(events->reserved));
3cfc3092
JK		 2802}
2803
2804static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2805 struct kvm_vcpu_events *events)
2806{
dab4b911 2807 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64 2808 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
f077825a
PB		 2809 | KVM_VCPUEVENT_VALID_SHADOW
2810 | KVM_VCPUEVENT_VALID_SMM))
3cfc3092
JK		 2811 return -EINVAL;
2812
7460fb4a 2813 process_nmi(vcpu);
3cfc3092
JK		 2814 vcpu->arch.exception.pending = events->exception.injected;
2815 vcpu->arch.exception.nr = events->exception.nr;
2816 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2817 vcpu->arch.exception.error_code = events->exception.error_code;
2818
2819 vcpu->arch.interrupt.pending = events->interrupt.injected;
2820 vcpu->arch.interrupt.nr = events->interrupt.nr;
2821 vcpu->arch.interrupt.soft = events->interrupt.soft;
48005f64
JK		 2822 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2823 kvm_x86_ops->set_interrupt_shadow(vcpu,
2824 events->interrupt.shadow);
3cfc3092
JK		 2825
2826 vcpu->arch.nmi_injected = events->nmi.injected;
dab4b911
JK		 2827 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2828 vcpu->arch.nmi_pending = events->nmi.pending;
3cfc3092
JK		 2829 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2830
66450a21
JK		 2831 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2832 kvm_vcpu_has_lapic(vcpu))
2833 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3cfc3092 2834
f077825a
PB		 2835 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
2836 if (events->smi.smm)
2837 vcpu->arch.hflags |= HF_SMM_MASK;
2838 else
2839 vcpu->arch.hflags &= ~HF_SMM_MASK;
2840 vcpu->arch.smi_pending = events->smi.pending;
2841 if (events->smi.smm_inside_nmi)
2842 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
2843 else
2844 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
2845 if (kvm_vcpu_has_lapic(vcpu)) {
2846 if (events->smi.latched_init)
2847 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2848 else
2849 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2850 }
2851 }
2852
3842d135
AK		 2853 kvm_make_request(KVM_REQ_EVENT, vcpu);
2854
3cfc3092
JK		 2855 return 0;
2856}
2857
a1efbe77
JK		 2858static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2859 struct kvm_debugregs *dbgregs)
2860{
73aaf249
JK		 2861 unsigned long val;
2862
a1efbe77 2863 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
16f8a6f9 2864 kvm_get_dr(vcpu, 6, &val);
73aaf249 2865 dbgregs->dr6 = val;
a1efbe77
JK		 2866 dbgregs->dr7 = vcpu->arch.dr7;
2867 dbgregs->flags = 0;
97e69aa6 2868 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
a1efbe77
JK		 2869}
2870
2871static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2872 struct kvm_debugregs *dbgregs)
2873{
2874 if (dbgregs->flags)
2875 return -EINVAL;
2876
a1efbe77 2877 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
ae561ede 2878 kvm_update_dr0123(vcpu);
a1efbe77 2879 vcpu->arch.dr6 = dbgregs->dr6;
73aaf249 2880 kvm_update_dr6(vcpu);
a1efbe77 2881 vcpu->arch.dr7 = dbgregs->dr7;
9926c9fd 2882 kvm_update_dr7(vcpu);
a1efbe77 2883
a1efbe77
JK		 2884 return 0;
2885}
2886
df1daba7
PB		 2887#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
2888
2889static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
2890{
c47ada30 2891 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
400e4b20 2892 u64 xstate_bv = xsave->header.xfeatures;
df1daba7
PB		 2893 u64 valid;
2894
2895 /*
2896 * Copy legacy XSAVE area, to avoid complications with CPUID
2897 * leaves 0 and 1 in the loop below.
2898 */
2899 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
2900
2901 /* Set XSTATE_BV */
2902 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
2903
2904 /*
2905 * Copy each region from the possibly compacted offset to the
2906 * non-compacted offset.
2907 */
2908 valid = xstate_bv & ~XSTATE_FPSSE;
2909 while (valid) {
2910 u64 feature = valid & -valid;
2911 int index = fls64(feature) - 1;
2912 void *src = get_xsave_addr(xsave, feature);
2913
2914 if (src) {
2915 u32 size, offset, ecx, edx;
2916 cpuid_count(XSTATE_CPUID, index,
2917 &size, &offset, &ecx, &edx);
2918 memcpy(dest + offset, src, size);
2919 }
2920
2921 valid -= feature;
2922 }
2923}
2924
2925static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
2926{
c47ada30 2927 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
df1daba7
PB		 2928 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
2929 u64 valid;
2930
2931 /*
2932 * Copy legacy XSAVE area, to avoid complications with CPUID
2933 * leaves 0 and 1 in the loop below.
2934 */
2935 memcpy(xsave, src, XSAVE_HDR_OFFSET);
2936
2937 /* Set XSTATE_BV and possibly XCOMP_BV. */
400e4b20 2938 xsave->header.xfeatures = xstate_bv;
df1daba7 2939 if (cpu_has_xsaves)
3a54450b 2940 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
df1daba7
PB		 2941
2942 /*
2943 * Copy each region from the non-compacted offset to the
2944 * possibly compacted offset.
2945 */
2946 valid = xstate_bv & ~XSTATE_FPSSE;
2947 while (valid) {
2948 u64 feature = valid & -valid;
2949 int index = fls64(feature) - 1;
2950 void *dest = get_xsave_addr(xsave, feature);
2951
2952 if (dest) {
2953 u32 size, offset, ecx, edx;
2954 cpuid_count(XSTATE_CPUID, index,
2955 &size, &offset, &ecx, &edx);
2956 memcpy(dest, src + offset, size);
ee4100da 2957 }
df1daba7
PB		 2958
2959 valid -= feature;
2960 }
2961}
2962
2d5b5a66
SY		 2963static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2964 struct kvm_xsave *guest_xsave)
2965{
4344ee98 2966 if (cpu_has_xsave) {
df1daba7
PB		 2967 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
2968 fill_xsave((u8 *) guest_xsave->region, vcpu);
4344ee98 2969 } else {
2d5b5a66 2970 memcpy(guest_xsave->region,
7366ed77 2971 &vcpu->arch.guest_fpu.state.fxsave,
c47ada30 2972 sizeof(struct fxregs_state));
2d5b5a66
SY		 2973 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2974 XSTATE_FPSSE;
2975 }
2976}
2977
2978static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2979 struct kvm_xsave *guest_xsave)
2980{
2981 u64 xstate_bv =
2982 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2983
d7876f1b
PB		 2984 if (cpu_has_xsave) {
2985 /*
2986 * Here we allow setting states that are not present in
2987 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
2988 * with old userspace.
2989 */
4ff41732 2990 if (xstate_bv & ~kvm_supported_xcr0())
d7876f1b 2991 return -EINVAL;
df1daba7 2992 load_xsave(vcpu, (u8 *)guest_xsave->region);
d7876f1b 2993 } else {
2d5b5a66
SY		 2994 if (xstate_bv & ~XSTATE_FPSSE)
2995 return -EINVAL;
7366ed77 2996 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
c47ada30 2997 guest_xsave->region, sizeof(struct fxregs_state));
2d5b5a66
SY		 2998 }
2999 return 0;
3000}
3001
3002static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3003 struct kvm_xcrs *guest_xcrs)
3004{
3005 if (!cpu_has_xsave) {
3006 guest_xcrs->nr_xcrs = 0;
3007 return;
3008 }
3009
3010 guest_xcrs->nr_xcrs = 1;
3011 guest_xcrs->flags = 0;
3012 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3013 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3014}
3015
3016static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3017 struct kvm_xcrs *guest_xcrs)
3018{
3019 int i, r = 0;
3020
3021 if (!cpu_has_xsave)
3022 return -EINVAL;
3023
3024 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3025 return -EINVAL;
3026
3027 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3028 /* Only support XCR0 currently */
c67a04cb 3029 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
2d5b5a66 3030 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
c67a04cb 3031 guest_xcrs->xcrs[i].value);
2d5b5a66
SY		 3032 break;
3033 }
3034 if (r)
3035 r = -EINVAL;
3036 return r;
3037}
3038
1c0b28c2
EM		 3039/*
3040 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3041 * stopped by the hypervisor. This function will be called from the host only.
3042 * EINVAL is returned when the host attempts to set the flag for a guest that
3043 * does not support pv clocks.
3044 */
3045static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3046{
0b79459b 3047 if (!vcpu->arch.pv_time_enabled)
1c0b28c2 3048 return -EINVAL;
51d59c6b 3049 vcpu->arch.pvclock_set_guest_stopped_request = true;
1c0b28c2
EM		 3050 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3051 return 0;
3052}
3053
313a3dc7
CO		 3054long kvm_arch_vcpu_ioctl(struct file *filp,
3055 unsigned int ioctl, unsigned long arg)
3056{
3057 struct kvm_vcpu *vcpu = filp->private_data;
3058 void __user *argp = (void __user *)arg;
3059 int r;
d1ac91d8
AK		 3060 union {
3061 struct kvm_lapic_state *lapic;
3062 struct kvm_xsave *xsave;
3063 struct kvm_xcrs *xcrs;
3064 void *buffer;
3065 } u;
3066
3067 u.buffer = NULL;
313a3dc7
CO		 3068 switch (ioctl) {
3069 case KVM_GET_LAPIC: {
2204ae3c
MT		 3070 r = -EINVAL;
3071 if (!vcpu->arch.apic)
3072 goto out;
d1ac91d8 3073 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
313a3dc7 3074
b772ff36 3075 r = -ENOMEM;
d1ac91d8 3076 if (!u.lapic)
b772ff36 3077 goto out;
d1ac91d8 3078 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
313a3dc7
CO		 3079 if (r)
3080 goto out;
3081 r = -EFAULT;
d1ac91d8 3082 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
313a3dc7
CO		 3083 goto out;
3084 r = 0;
3085 break;
3086 }
3087 case KVM_SET_LAPIC: {
2204ae3c
MT		 3088 r = -EINVAL;
3089 if (!vcpu->arch.apic)
3090 goto out;
ff5c2c03 3091 u.lapic = memdup_user(argp, sizeof(*u.lapic));
18595411
GC		 3092 if (IS_ERR(u.lapic))
3093 return PTR_ERR(u.lapic);
ff5c2c03 3094
d1ac91d8 3095 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
313a3dc7
CO		 3096 break;
3097 }
f77bc6a4
ZX		 3098 case KVM_INTERRUPT: {
3099 struct kvm_interrupt irq;
3100
3101 r = -EFAULT;
3102 if (copy_from_user(&irq, argp, sizeof irq))
3103 goto out;
3104 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
f77bc6a4
ZX		 3105 break;
3106 }
c4abb7c9
JK		 3107 case KVM_NMI: {
3108 r = kvm_vcpu_ioctl_nmi(vcpu);
c4abb7c9
JK		 3109 break;
3110 }
f077825a
PB		 3111 case KVM_SMI: {
3112 r = kvm_vcpu_ioctl_smi(vcpu);
3113 break;
3114 }
313a3dc7
CO		 3115 case KVM_SET_CPUID: {
3116 struct kvm_cpuid __user *cpuid_arg = argp;
3117 struct kvm_cpuid cpuid;
3118
3119 r = -EFAULT;
3120 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3121 goto out;
3122 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
313a3dc7
CO		 3123 break;
3124 }
07716717
DK		 3125 case KVM_SET_CPUID2: {
3126 struct kvm_cpuid2 __user *cpuid_arg = argp;
3127 struct kvm_cpuid2 cpuid;
3128
3129 r = -EFAULT;
3130 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3131 goto out;
3132 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
19355475 3133 cpuid_arg->entries);
07716717
DK		 3134 break;
3135 }
3136 case KVM_GET_CPUID2: {
3137 struct kvm_cpuid2 __user *cpuid_arg = argp;
3138 struct kvm_cpuid2 cpuid;
3139
3140 r = -EFAULT;
3141 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3142 goto out;
3143 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
19355475 3144 cpuid_arg->entries);
07716717
DK		 3145 if (r)
3146 goto out;
3147 r = -EFAULT;
3148 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3149 goto out;
3150 r = 0;
3151 break;
3152 }
313a3dc7 3153 case KVM_GET_MSRS:
609e36d3 3154 r = msr_io(vcpu, argp, do_get_msr, 1);
313a3dc7
CO		 3155 break;
3156 case KVM_SET_MSRS:
3157 r = msr_io(vcpu, argp, do_set_msr, 0);
3158 break;
b209749f
AK		 3159 case KVM_TPR_ACCESS_REPORTING: {
3160 struct kvm_tpr_access_ctl tac;
3161
3162 r = -EFAULT;
3163 if (copy_from_user(&tac, argp, sizeof tac))
3164 goto out;
3165 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3166 if (r)
3167 goto out;
3168 r = -EFAULT;
3169 if (copy_to_user(argp, &tac, sizeof tac))
3170 goto out;
3171 r = 0;
3172 break;
3173 };
b93463aa
AK		 3174 case KVM_SET_VAPIC_ADDR: {
3175 struct kvm_vapic_addr va;
3176
3177 r = -EINVAL;
35754c98 3178 if (!lapic_in_kernel(vcpu))
b93463aa
AK		 3179 goto out;
3180 r = -EFAULT;
3181 if (copy_from_user(&va, argp, sizeof va))
3182 goto out;
fda4e2e8 3183 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
b93463aa
AK		 3184 break;
3185 }
890ca9ae
HY		 3186 case KVM_X86_SETUP_MCE: {
3187 u64 mcg_cap;
3188
3189 r = -EFAULT;
3190 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3191 goto out;
3192 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3193 break;
3194 }
3195 case KVM_X86_SET_MCE: {
3196 struct kvm_x86_mce mce;
3197
3198 r = -EFAULT;
3199 if (copy_from_user(&mce, argp, sizeof mce))
3200 goto out;
3201 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3202 break;
3203 }
3cfc3092
JK		 3204 case KVM_GET_VCPU_EVENTS: {
3205 struct kvm_vcpu_events events;
3206
3207 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3208
3209 r = -EFAULT;
3210 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3211 break;
3212 r = 0;
3213 break;
3214 }
3215 case KVM_SET_VCPU_EVENTS: {
3216 struct kvm_vcpu_events events;
3217
3218 r = -EFAULT;
3219 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3220 break;
3221
3222 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3223 break;
3224 }
a1efbe77
JK		 3225 case KVM_GET_DEBUGREGS: {
3226 struct kvm_debugregs dbgregs;
3227
3228 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3229
3230 r = -EFAULT;
3231 if (copy_to_user(argp, &dbgregs,
3232 sizeof(struct kvm_debugregs)))
3233 break;
3234 r = 0;
3235 break;
3236 }
3237 case KVM_SET_DEBUGREGS: {
3238 struct kvm_debugregs dbgregs;
3239
3240 r = -EFAULT;
3241 if (copy_from_user(&dbgregs, argp,
3242 sizeof(struct kvm_debugregs)))
3243 break;
3244
3245 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3246 break;
3247 }
2d5b5a66 3248 case KVM_GET_XSAVE: {
d1ac91d8 3249 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2d5b5a66 3250 r = -ENOMEM;
d1ac91d8 3251 if (!u.xsave)
2d5b5a66
SY		 3252 break;
3253
d1ac91d8 3254 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3255
3256 r = -EFAULT;
d1ac91d8 3257 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2d5b5a66
SY		 3258 break;
3259 r = 0;
3260 break;
3261 }
3262 case KVM_SET_XSAVE: {
ff5c2c03 3263 u.xsave = memdup_user(argp, sizeof(*u.xsave));
18595411
GC		 3264 if (IS_ERR(u.xsave))
3265 return PTR_ERR(u.xsave);
2d5b5a66 3266
d1ac91d8 3267 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3268 break;
3269 }
3270 case KVM_GET_XCRS: {
d1ac91d8 3271 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2d5b5a66 3272 r = -ENOMEM;
d1ac91d8 3273 if (!u.xcrs)
2d5b5a66
SY		 3274 break;
3275
d1ac91d8 3276 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3277
3278 r = -EFAULT;
d1ac91d8 3279 if (copy_to_user(argp, u.xcrs,
2d5b5a66
SY		 3280 sizeof(struct kvm_xcrs)))
3281 break;
3282 r = 0;
3283 break;
3284 }
3285 case KVM_SET_XCRS: {
ff5c2c03 3286 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
18595411
GC		 3287 if (IS_ERR(u.xcrs))
3288 return PTR_ERR(u.xcrs);
2d5b5a66 3289
d1ac91d8 3290 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3291 break;
3292 }
92a1f12d
JR		 3293 case KVM_SET_TSC_KHZ: {
3294 u32 user_tsc_khz;
3295
3296 r = -EINVAL;
92a1f12d
JR		 3297 user_tsc_khz = (u32)arg;
3298
3299 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3300 goto out;
3301
cc578287
ZA		 3302 if (user_tsc_khz == 0)
3303 user_tsc_khz = tsc_khz;
3304
3305 kvm_set_tsc_khz(vcpu, user_tsc_khz);
92a1f12d
JR		 3306
3307 r = 0;
3308 goto out;
3309 }
3310 case KVM_GET_TSC_KHZ: {
cc578287 3311 r = vcpu->arch.virtual_tsc_khz;
92a1f12d
JR		 3312 goto out;
3313 }
1c0b28c2
EM		 3314 case KVM_KVMCLOCK_CTRL: {
3315 r = kvm_set_guest_paused(vcpu);
3316 goto out;
3317 }
313a3dc7
CO		 3318 default:
3319 r = -EINVAL;
3320 }
3321out:
d1ac91d8 3322 kfree(u.buffer);
313a3dc7
CO		 3323 return r;
3324}
3325
5b1c1493
CO		 3326int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3327{
3328 return VM_FAULT_SIGBUS;
3329}
3330
1fe779f8
CO		 3331static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3332{
3333 int ret;
3334
3335 if (addr > (unsigned int)(-3 * PAGE_SIZE))
951179ce 3336 return -EINVAL;
1fe779f8
CO		 3337 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3338 return ret;
3339}
3340
b927a3ce
SY		 3341static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3342 u64 ident_addr)
3343{
3344 kvm->arch.ept_identity_map_addr = ident_addr;
3345 return 0;
3346}
3347
1fe779f8
CO		 3348static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3349 u32 kvm_nr_mmu_pages)
3350{
3351 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3352 return -EINVAL;
3353
79fac95e 3354 mutex_lock(&kvm->slots_lock);
1fe779f8
CO		 3355
3356 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
f05e70ac 3357 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1fe779f8 3358
79fac95e 3359 mutex_unlock(&kvm->slots_lock);
1fe779f8
CO		 3360 return 0;
3361}
3362
3363static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3364{
39de71ec 3365 return kvm->arch.n_max_mmu_pages;
1fe779f8
CO		 3366}
3367
1fe779f8
CO		 3368static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3369{
3370 int r;
3371
3372 r = 0;
3373 switch (chip->chip_id) {
3374 case KVM_IRQCHIP_PIC_MASTER:
3375 memcpy(&chip->chip.pic,
3376 &pic_irqchip(kvm)->pics[0],
3377 sizeof(struct kvm_pic_state));
3378 break;
3379 case KVM_IRQCHIP_PIC_SLAVE:
3380 memcpy(&chip->chip.pic,
3381 &pic_irqchip(kvm)->pics[1],
3382 sizeof(struct kvm_pic_state));
3383 break;
3384 case KVM_IRQCHIP_IOAPIC:
eba0226b 3385 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3386 break;
3387 default:
3388 r = -EINVAL;
3389 break;
3390 }
3391 return r;
3392}
3393
3394static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3395{
3396 int r;
3397
3398 r = 0;
3399 switch (chip->chip_id) {
3400 case KVM_IRQCHIP_PIC_MASTER:
f4f51050 3401 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3402 memcpy(&pic_irqchip(kvm)->pics[0],
3403 &chip->chip.pic,
3404 sizeof(struct kvm_pic_state));
f4f51050 3405 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3406 break;
3407 case KVM_IRQCHIP_PIC_SLAVE:
f4f51050 3408 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3409 memcpy(&pic_irqchip(kvm)->pics[1],
3410 &chip->chip.pic,
3411 sizeof(struct kvm_pic_state));
f4f51050 3412 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3413 break;
3414 case KVM_IRQCHIP_IOAPIC:
eba0226b 3415 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3416 break;
3417 default:
3418 r = -EINVAL;
3419 break;
3420 }
3421 kvm_pic_update_irq(pic_irqchip(kvm));
3422 return r;
3423}
3424
e0f63cb9
SY		 3425static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3426{
3427 int r = 0;
3428
894a9c55 3429 mutex_lock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9 3430 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
894a9c55 3431 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9
SY		 3432 return r;
3433}
3434
3435static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3436{
3437 int r = 0;
3438
894a9c55 3439 mutex_lock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9 3440 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
e9f42757
BK		 3441 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3442 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3443 return r;
3444}
3445
3446static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3447{
3448 int r = 0;
3449
3450 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3451 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3452 sizeof(ps->channels));
3453 ps->flags = kvm->arch.vpit->pit_state.flags;
3454 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
97e69aa6 3455 memset(&ps->reserved, 0, sizeof(ps->reserved));
e9f42757
BK		 3456 return r;
3457}
3458
3459static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3460{
3461 int r = 0, start = 0;
3462 u32 prev_legacy, cur_legacy;
3463 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3464 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3465 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3466 if (!prev_legacy && cur_legacy)
3467 start = 1;
3468 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3469 sizeof(kvm->arch.vpit->pit_state.channels));
3470 kvm->arch.vpit->pit_state.flags = ps->flags;
3471 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
894a9c55 3472 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9
SY		 3473 return r;
3474}
3475
52d939a0
MT		 3476static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3477 struct kvm_reinject_control *control)
3478{
3479 if (!kvm->arch.vpit)
3480 return -ENXIO;
894a9c55 3481 mutex_lock(&kvm->arch.vpit->pit_state.lock);
26ef1924 3482 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
894a9c55 3483 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
52d939a0
MT		 3484 return 0;
3485}
3486
95d4c16c 3487/**
60c34612
TY		 3488 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3489 * @kvm: kvm instance
3490 * @log: slot id and address to which we copy the log
95d4c16c 3491 *
e108ff2f
PB		 3492 * Steps 1-4 below provide general overview of dirty page logging. See
3493 * kvm_get_dirty_log_protect() function description for additional details.
3494 *
3495 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3496 * always flush the TLB (step 4) even if previous step failed and the dirty
3497 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3498 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3499 * writes will be marked dirty for next log read.
95d4c16c 3500 *
60c34612
TY		 3501 * 1. Take a snapshot of the bit and clear it if needed.
3502 * 2. Write protect the corresponding page.
e108ff2f
PB		 3503 * 3. Copy the snapshot to the userspace.
3504 * 4. Flush TLB's if needed.
5bb064dc 3505 */
60c34612 3506int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
5bb064dc 3507{
60c34612 3508 bool is_dirty = false;
e108ff2f 3509 int r;
5bb064dc 3510
79fac95e 3511 mutex_lock(&kvm->slots_lock);
5bb064dc 3512
88178fd4
KH		 3513 /*
3514 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3515 */
3516 if (kvm_x86_ops->flush_log_dirty)
3517 kvm_x86_ops->flush_log_dirty(kvm);
3518
e108ff2f 3519 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
198c74f4
XG		 3520
3521 /*
3522 * All the TLBs can be flushed out of mmu lock, see the comments in
3523 * kvm_mmu_slot_remove_write_access().
3524 */
e108ff2f 3525 lockdep_assert_held(&kvm->slots_lock);
198c74f4
XG		 3526 if (is_dirty)
3527 kvm_flush_remote_tlbs(kvm);
3528
79fac95e 3529 mutex_unlock(&kvm->slots_lock);
5bb064dc
ZX		 3530 return r;
3531}
3532
aa2fbe6d
YZ		 3533int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3534 bool line_status)
23d43cf9
CD		 3535{
3536 if (!irqchip_in_kernel(kvm))
3537 return -ENXIO;
3538
3539 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
aa2fbe6d
YZ		 3540 irq_event->irq, irq_event->level,
3541 line_status);
23d43cf9
CD		 3542 return 0;
3543}
3544
90de4a18
NA		 3545static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3546 struct kvm_enable_cap *cap)
3547{
3548 int r;
3549
3550 if (cap->flags)
3551 return -EINVAL;
3552
3553 switch (cap->cap) {
3554 case KVM_CAP_DISABLE_QUIRKS:
3555 kvm->arch.disabled_quirks = cap->args[0];
3556 r = 0;
3557 break;
3558 default:
3559 r = -EINVAL;
3560 break;
3561 }
3562 return r;
3563}
3564
1fe779f8
CO		 3565long kvm_arch_vm_ioctl(struct file *filp,
3566 unsigned int ioctl, unsigned long arg)
3567{
3568 struct kvm *kvm = filp->private_data;
3569 void __user *argp = (void __user *)arg;
367e1319 3570 int r = -ENOTTY;
f0d66275
DH		 3571 /*
3572 * This union makes it completely explicit to gcc-3.x
3573 * that these two variables' stack usage should be
3574 * combined, not added together.
3575 */
3576 union {
3577 struct kvm_pit_state ps;
e9f42757 3578 struct kvm_pit_state2 ps2;
c5ff41ce 3579 struct kvm_pit_config pit_config;
f0d66275 3580 } u;
1fe779f8
CO		 3581
3582 switch (ioctl) {
3583 case KVM_SET_TSS_ADDR:
3584 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1fe779f8 3585 break;
b927a3ce
SY		 3586 case KVM_SET_IDENTITY_MAP_ADDR: {
3587 u64 ident_addr;
3588
3589 r = -EFAULT;
3590 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3591 goto out;
3592 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
b927a3ce
SY		 3593 break;
3594 }
1fe779f8
CO		 3595 case KVM_SET_NR_MMU_PAGES:
3596 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1fe779f8
CO		 3597 break;
3598 case KVM_GET_NR_MMU_PAGES:
3599 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3600 break;
3ddea128
MT		 3601 case KVM_CREATE_IRQCHIP: {
3602 struct kvm_pic *vpic;
3603
3604 mutex_lock(&kvm->lock);
3605 r = -EEXIST;
3606 if (kvm->arch.vpic)
3607 goto create_irqchip_unlock;
3e515705
AK		 3608 r = -EINVAL;
3609 if (atomic_read(&kvm->online_vcpus))
3610 goto create_irqchip_unlock;
1fe779f8 3611 r = -ENOMEM;
3ddea128
MT		 3612 vpic = kvm_create_pic(kvm);
3613 if (vpic) {
1fe779f8
CO		 3614 r = kvm_ioapic_init(kvm);
3615 if (r) {
175504cd 3616 mutex_lock(&kvm->slots_lock);
71ba994c 3617 kvm_destroy_pic(vpic);
175504cd 3618 mutex_unlock(&kvm->slots_lock);
3ddea128 3619 goto create_irqchip_unlock;
1fe779f8
CO		 3620 }
3621 } else
3ddea128 3622 goto create_irqchip_unlock;
399ec807
AK		 3623 r = kvm_setup_default_irq_routing(kvm);
3624 if (r) {
175504cd 3625 mutex_lock(&kvm->slots_lock);
3ddea128 3626 mutex_lock(&kvm->irq_lock);
72bb2fcd 3627 kvm_ioapic_destroy(kvm);
71ba994c 3628 kvm_destroy_pic(vpic);
3ddea128 3629 mutex_unlock(&kvm->irq_lock);
175504cd 3630 mutex_unlock(&kvm->slots_lock);
71ba994c 3631 goto create_irqchip_unlock;
399ec807 3632 }
71ba994c
PB		 3633 /* Write kvm->irq_routing before kvm->arch.vpic. */
3634 smp_wmb();
3635 kvm->arch.vpic = vpic;
3ddea128
MT		 3636 create_irqchip_unlock:
3637 mutex_unlock(&kvm->lock);
1fe779f8 3638 break;
3ddea128 3639 }
7837699f 3640 case KVM_CREATE_PIT:
c5ff41ce
JK		 3641 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3642 goto create_pit;
3643 case KVM_CREATE_PIT2:
3644 r = -EFAULT;
3645 if (copy_from_user(&u.pit_config, argp,
3646 sizeof(struct kvm_pit_config)))
3647 goto out;
3648 create_pit:
79fac95e 3649 mutex_lock(&kvm->slots_lock);
269e05e4
AK		 3650 r = -EEXIST;
3651 if (kvm->arch.vpit)
3652 goto create_pit_unlock;
7837699f 3653 r = -ENOMEM;
c5ff41ce 3654 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
7837699f
SY		 3655 if (kvm->arch.vpit)
3656 r = 0;
269e05e4 3657 create_pit_unlock:
79fac95e 3658 mutex_unlock(&kvm->slots_lock);
7837699f 3659 break;
1fe779f8
CO		 3660 case KVM_GET_IRQCHIP: {
3661 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 3662 struct kvm_irqchip *chip;
1fe779f8 3663
ff5c2c03
SL		 3664 chip = memdup_user(argp, sizeof(*chip));
3665 if (IS_ERR(chip)) {
3666 r = PTR_ERR(chip);
1fe779f8 3667 goto out;
ff5c2c03
SL		 3668 }
3669
1fe779f8
CO		 3670 r = -ENXIO;
3671 if (!irqchip_in_kernel(kvm))
f0d66275
DH		 3672 goto get_irqchip_out;
3673 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1fe779f8 3674 if (r)
f0d66275 3675 goto get_irqchip_out;
1fe779f8 3676 r = -EFAULT;
f0d66275
DH		 3677 if (copy_to_user(argp, chip, sizeof *chip))
3678 goto get_irqchip_out;
1fe779f8 3679 r = 0;
f0d66275
DH		 3680 get_irqchip_out:
3681 kfree(chip);
1fe779f8
CO		 3682 break;
3683 }
3684 case KVM_SET_IRQCHIP: {
3685 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 3686 struct kvm_irqchip *chip;
1fe779f8 3687
ff5c2c03
SL		 3688 chip = memdup_user(argp, sizeof(*chip));
3689 if (IS_ERR(chip)) {
3690 r = PTR_ERR(chip);
1fe779f8 3691 goto out;
ff5c2c03
SL		 3692 }
3693
1fe779f8
CO		 3694 r = -ENXIO;
3695 if (!irqchip_in_kernel(kvm))
f0d66275
DH		 3696 goto set_irqchip_out;
3697 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1fe779f8 3698 if (r)
f0d66275 3699 goto set_irqchip_out;
1fe779f8 3700 r = 0;
f0d66275
DH		 3701 set_irqchip_out:
3702 kfree(chip);
1fe779f8
CO		 3703 break;
3704 }
e0f63cb9 3705 case KVM_GET_PIT: {
e0f63cb9 3706 r = -EFAULT;
f0d66275 3707 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 3708 goto out;
3709 r = -ENXIO;
3710 if (!kvm->arch.vpit)
3711 goto out;
f0d66275 3712 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
e0f63cb9
SY		 3713 if (r)
3714 goto out;
3715 r = -EFAULT;
f0d66275 3716 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 3717 goto out;
3718 r = 0;
3719 break;
3720 }
3721 case KVM_SET_PIT: {
e0f63cb9 3722 r = -EFAULT;
f0d66275 3723 if (copy_from_user(&u.ps, argp, sizeof u.ps))
e0f63cb9
SY		 3724 goto out;
3725 r = -ENXIO;
3726 if (!kvm->arch.vpit)
3727 goto out;
f0d66275 3728 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
e0f63cb9
SY		 3729 break;
3730 }
e9f42757
BK		 3731 case KVM_GET_PIT2: {
3732 r = -ENXIO;
3733 if (!kvm->arch.vpit)
3734 goto out;
3735 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3736 if (r)
3737 goto out;
3738 r = -EFAULT;
3739 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3740 goto out;
3741 r = 0;
3742 break;
3743 }
3744 case KVM_SET_PIT2: {
3745 r = -EFAULT;
3746 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3747 goto out;
3748 r = -ENXIO;
3749 if (!kvm->arch.vpit)
3750 goto out;
3751 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
e9f42757
BK		 3752 break;
3753 }
52d939a0
MT		 3754 case KVM_REINJECT_CONTROL: {
3755 struct kvm_reinject_control control;
3756 r = -EFAULT;
3757 if (copy_from_user(&control, argp, sizeof(control)))
3758 goto out;
3759 r = kvm_vm_ioctl_reinject(kvm, &control);
52d939a0
MT		 3760 break;
3761 }
d71ba788
PB		 3762 case KVM_SET_BOOT_CPU_ID:
3763 r = 0;
3764 mutex_lock(&kvm->lock);
3765 if (atomic_read(&kvm->online_vcpus) != 0)
3766 r = -EBUSY;
3767 else
3768 kvm->arch.bsp_vcpu_id = arg;
3769 mutex_unlock(&kvm->lock);
3770 break;
ffde22ac
ES		 3771 case KVM_XEN_HVM_CONFIG: {
3772 r = -EFAULT;
3773 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3774 sizeof(struct kvm_xen_hvm_config)))
3775 goto out;
3776 r = -EINVAL;
3777 if (kvm->arch.xen_hvm_config.flags)
3778 goto out;
3779 r = 0;
3780 break;
3781 }
afbcf7ab 3782 case KVM_SET_CLOCK: {
afbcf7ab
GC		 3783 struct kvm_clock_data user_ns;
3784 u64 now_ns;
3785 s64 delta;
3786
3787 r = -EFAULT;
3788 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3789 goto out;
3790
3791 r = -EINVAL;
3792 if (user_ns.flags)
3793 goto out;
3794
3795 r = 0;
395c6b0a 3796 local_irq_disable();
759379dd 3797 now_ns = get_kernel_ns();
afbcf7ab 3798 delta = user_ns.clock - now_ns;
395c6b0a 3799 local_irq_enable();
afbcf7ab 3800 kvm->arch.kvmclock_offset = delta;
2e762ff7 3801 kvm_gen_update_masterclock(kvm);
afbcf7ab
GC		 3802 break;
3803 }
3804 case KVM_GET_CLOCK: {
afbcf7ab
GC		 3805 struct kvm_clock_data user_ns;
3806 u64 now_ns;
3807
395c6b0a 3808 local_irq_disable();
759379dd 3809 now_ns = get_kernel_ns();
afbcf7ab 3810 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
395c6b0a 3811 local_irq_enable();
afbcf7ab 3812 user_ns.flags = 0;
97e69aa6 3813 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
afbcf7ab
GC		 3814
3815 r = -EFAULT;
3816 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3817 goto out;
3818 r = 0;
3819 break;
3820 }
90de4a18
NA		 3821 case KVM_ENABLE_CAP: {
3822 struct kvm_enable_cap cap;
afbcf7ab 3823
90de4a18
NA		 3824 r = -EFAULT;
3825 if (copy_from_user(&cap, argp, sizeof(cap)))
3826 goto out;
3827 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
3828 break;
3829 }
1fe779f8 3830 default:
c274e03a 3831 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1fe779f8
CO		 3832 }
3833out:
3834 return r;
3835}
3836
a16b043c 3837static void kvm_init_msr_list(void)
043405e1
CO		 3838{
3839 u32 dummy[2];
3840 unsigned i, j;
3841
62ef68bb 3842 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
043405e1
CO		 3843 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3844 continue;
93c4adc7
PB		 3845
3846 /*
3847 * Even MSRs that are valid in the host may not be exposed
3848 * to the guests in some cases. We could work around this
3849 * in VMX with the generic MSR save/load machinery, but it
3850 * is not really worthwhile since it will really only
3851 * happen with nested virtualization.
3852 */
3853 switch (msrs_to_save[i]) {
3854 case MSR_IA32_BNDCFGS:
3855 if (!kvm_x86_ops->mpx_supported())
3856 continue;
3857 break;
3858 default:
3859 break;
3860 }
3861
043405e1
CO		 3862 if (j < i)
3863 msrs_to_save[j] = msrs_to_save[i];
3864 j++;
3865 }
3866 num_msrs_to_save = j;
62ef68bb
PB		 3867
3868 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
3869 switch (emulated_msrs[i]) {
6d396b55
PB		 3870 case MSR_IA32_SMBASE:
3871 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
3872 continue;
3873 break;
62ef68bb
PB		 3874 default:
3875 break;
3876 }
3877
3878 if (j < i)
3879 emulated_msrs[j] = emulated_msrs[i];
3880 j++;
3881 }
3882 num_emulated_msrs = j;
043405e1
CO		 3883}
3884
bda9020e
MT		 3885static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3886 const void *v)
bbd9b64e 3887{
70252a10
AK		 3888 int handled = 0;
3889 int n;
3890
3891 do {
3892 n = min(len, 8);
3893 if (!(vcpu->arch.apic &&
e32edf4f
NN		 3894 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
3895 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10
AK		 3896 break;
3897 handled += n;
3898 addr += n;
3899 len -= n;
3900 v += n;
3901 } while (len);
bbd9b64e 3902
70252a10 3903 return handled;
bbd9b64e
CO		 3904}
3905
bda9020e 3906static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
bbd9b64e 3907{
70252a10
AK		 3908 int handled = 0;
3909 int n;
3910
3911 do {
3912 n = min(len, 8);
3913 if (!(vcpu->arch.apic &&
e32edf4f
NN		 3914 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
3915 addr, n, v))
3916 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10
AK		 3917 break;
3918 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3919 handled += n;
3920 addr += n;
3921 len -= n;
3922 v += n;
3923 } while (len);
bbd9b64e 3924
70252a10 3925 return handled;
bbd9b64e
CO		 3926}
3927
2dafc6c2
GN		 3928static void kvm_set_segment(struct kvm_vcpu *vcpu,
3929 struct kvm_segment *var, int seg)
3930{
3931 kvm_x86_ops->set_segment(vcpu, var, seg);
3932}
3933
3934void kvm_get_segment(struct kvm_vcpu *vcpu,
3935 struct kvm_segment *var, int seg)
3936{
3937 kvm_x86_ops->get_segment(vcpu, var, seg);
3938}
3939
54987b7a
PB		 3940gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
3941 struct x86_exception *exception)
02f59dc9
JR		 3942{
3943 gpa_t t_gpa;
02f59dc9
JR		 3944
3945 BUG_ON(!mmu_is_nested(vcpu));
3946
3947 /* NPT walks are always user-walks */
3948 access |= PFERR_USER_MASK;
54987b7a 3949 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
02f59dc9
JR		 3950
3951 return t_gpa;
3952}
3953
ab9ae313
AK		 3954gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3955 struct x86_exception *exception)
1871c602
GN		 3956{
3957 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
ab9ae313 3958 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 3959}
3960
ab9ae313
AK		 3961 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3962 struct x86_exception *exception)
1871c602
GN		 3963{
3964 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3965 access |= PFERR_FETCH_MASK;
ab9ae313 3966 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 3967}
3968
ab9ae313
AK		 3969gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3970 struct x86_exception *exception)
1871c602
GN		 3971{
3972 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3973 access |= PFERR_WRITE_MASK;
ab9ae313 3974 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 3975}
3976
3977/* uses this to access any guest's mapped memory without checking CPL */
ab9ae313
AK		 3978gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3979 struct x86_exception *exception)
1871c602 3980{
ab9ae313 3981 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
1871c602
GN		 3982}
3983
3984static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3985 struct kvm_vcpu *vcpu, u32 access,
bcc55cba 3986 struct x86_exception *exception)
bbd9b64e
CO		 3987{
3988 void *data = val;
10589a46 3989 int r = X86EMUL_CONTINUE;
bbd9b64e
CO		 3990
3991 while (bytes) {
14dfe855 3992 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
ab9ae313 3993 exception);
bbd9b64e 3994 unsigned offset = addr & (PAGE_SIZE-1);
77c2002e 3995 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
bbd9b64e
CO		 3996 int ret;
3997
bcc55cba 3998 if (gpa == UNMAPPED_GVA)
ab9ae313 3999 return X86EMUL_PROPAGATE_FAULT;
54bf36aa
PB		 4000 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4001 offset, toread);
10589a46 4002 if (ret < 0) {
c3cd7ffa 4003 r = X86EMUL_IO_NEEDED;
10589a46
MT		 4004 goto out;
4005 }
bbd9b64e 4006
77c2002e
IE		 4007 bytes -= toread;
4008 data += toread;
4009 addr += toread;
bbd9b64e 4010 }
10589a46 4011out:
10589a46 4012 return r;
bbd9b64e 4013}
77c2002e 4014
1871c602 4015/* used for instruction fetching */
0f65dd70
AK		 4016static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4017 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4018 struct x86_exception *exception)
1871c602 4019{
0f65dd70 4020 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4021 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
44583cba
PB		 4022 unsigned offset;
4023 int ret;
0f65dd70 4024
44583cba
PB		 4025 /* Inline kvm_read_guest_virt_helper for speed. */
4026 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4027 exception);
4028 if (unlikely(gpa == UNMAPPED_GVA))
4029 return X86EMUL_PROPAGATE_FAULT;
4030
4031 offset = addr & (PAGE_SIZE-1);
4032 if (WARN_ON(offset + bytes > PAGE_SIZE))
4033 bytes = (unsigned)PAGE_SIZE - offset;
54bf36aa
PB		 4034 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4035 offset, bytes);
44583cba
PB		 4036 if (unlikely(ret < 0))
4037 return X86EMUL_IO_NEEDED;
4038
4039 return X86EMUL_CONTINUE;
1871c602
GN		 4040}
4041
064aea77 4042int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
0f65dd70 4043 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4044 struct x86_exception *exception)
1871c602 4045{
0f65dd70 4046 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4047 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
0f65dd70 4048
1871c602 4049 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
bcc55cba 4050 exception);
1871c602 4051}
064aea77 4052EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
1871c602 4053
0f65dd70
AK		 4054static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4055 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4056 struct x86_exception *exception)
1871c602 4057{
0f65dd70 4058 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
bcc55cba 4059 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
1871c602
GN		 4060}
4061
6a4d7550 4062int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
0f65dd70 4063 gva_t addr, void *val,
2dafc6c2 4064 unsigned int bytes,
bcc55cba 4065 struct x86_exception *exception)
77c2002e 4066{
0f65dd70 4067 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
77c2002e
IE		 4068 void *data = val;
4069 int r = X86EMUL_CONTINUE;
4070
4071 while (bytes) {
14dfe855
JR		 4072 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4073 PFERR_WRITE_MASK,
ab9ae313 4074 exception);
77c2002e
IE		 4075 unsigned offset = addr & (PAGE_SIZE-1);
4076 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4077 int ret;
4078
bcc55cba 4079 if (gpa == UNMAPPED_GVA)
ab9ae313 4080 return X86EMUL_PROPAGATE_FAULT;
54bf36aa 4081 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
77c2002e 4082 if (ret < 0) {
c3cd7ffa 4083 r = X86EMUL_IO_NEEDED;
77c2002e
IE		 4084 goto out;
4085 }
4086
4087 bytes -= towrite;
4088 data += towrite;
4089 addr += towrite;
4090 }
4091out:
4092 return r;
4093}
6a4d7550 4094EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
77c2002e 4095
af7cc7d1
XG		 4096static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4097 gpa_t *gpa, struct x86_exception *exception,
4098 bool write)
4099{
97d64b78
AK		 4100 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4101 | (write ? PFERR_WRITE_MASK : 0);
af7cc7d1 4102
97d64b78 4103 if (vcpu_match_mmio_gva(vcpu, gva)
97ec8c06
FW		 4104 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4105 vcpu->arch.access, access)) {
bebb106a
XG		 4106 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4107 (gva & (PAGE_SIZE - 1));
4f022648 4108 trace_vcpu_match_mmio(gva, *gpa, write, false);
bebb106a
XG		 4109 return 1;
4110 }
4111
af7cc7d1
XG		 4112 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4113
4114 if (*gpa == UNMAPPED_GVA)
4115 return -1;
4116
4117 /* For APIC access vmexit */
4118 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4119 return 1;
4120
4f022648
XG		 4121 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4122 trace_vcpu_match_mmio(gva, *gpa, write, true);
bebb106a 4123 return 1;
4f022648 4124 }
bebb106a 4125
af7cc7d1
XG		 4126 return 0;
4127}
4128
3200f405 4129int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
bcc55cba 4130 const void *val, int bytes)
bbd9b64e
CO		 4131{
4132 int ret;
4133
54bf36aa 4134 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
9f811285 4135 if (ret < 0)
bbd9b64e 4136 return 0;
f57f2ef5 4137 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
bbd9b64e
CO		 4138 return 1;
4139}
4140
77d197b2
XG		 4141struct read_write_emulator_ops {
4142 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4143 int bytes);
4144 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4145 void *val, int bytes);
4146 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4147 int bytes, void *val);
4148 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4149 void *val, int bytes);
4150 bool write;
4151};
4152
4153static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4154{
4155 if (vcpu->mmio_read_completed) {
77d197b2 4156 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
f78146b0 4157 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
77d197b2
XG		 4158 vcpu->mmio_read_completed = 0;
4159 return 1;
4160 }
4161
4162 return 0;
4163}
4164
4165static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4166 void *val, int bytes)
4167{
54bf36aa 4168 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
77d197b2
XG		 4169}
4170
4171static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4172 void *val, int bytes)
4173{
4174 return emulator_write_phys(vcpu, gpa, val, bytes);
4175}
4176
4177static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4178{
4179 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4180 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4181}
4182
4183static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4184 void *val, int bytes)
4185{
4186 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4187 return X86EMUL_IO_NEEDED;
4188}
4189
4190static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4191 void *val, int bytes)
4192{
f78146b0
AK		 4193 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4194
87da7e66 4195 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
77d197b2
XG		 4196 return X86EMUL_CONTINUE;
4197}
4198
0fbe9b0b 4199static const struct read_write_emulator_ops read_emultor = {
77d197b2
XG		 4200 .read_write_prepare = read_prepare,
4201 .read_write_emulate = read_emulate,
4202 .read_write_mmio = vcpu_mmio_read,
4203 .read_write_exit_mmio = read_exit_mmio,
4204};
4205
0fbe9b0b 4206static const struct read_write_emulator_ops write_emultor = {
77d197b2
XG		 4207 .read_write_emulate = write_emulate,
4208 .read_write_mmio = write_mmio,
4209 .read_write_exit_mmio = write_exit_mmio,
4210 .write = true,
4211};
4212
22388a3c
XG		 4213static int emulator_read_write_onepage(unsigned long addr, void *val,
4214 unsigned int bytes,
4215 struct x86_exception *exception,
4216 struct kvm_vcpu *vcpu,
0fbe9b0b 4217 const struct read_write_emulator_ops *ops)
bbd9b64e 4218{
af7cc7d1
XG		 4219 gpa_t gpa;
4220 int handled, ret;
22388a3c 4221 bool write = ops->write;
f78146b0 4222 struct kvm_mmio_fragment *frag;
10589a46 4223
22388a3c 4224 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
bbd9b64e 4225
af7cc7d1 4226 if (ret < 0)
bbd9b64e 4227 return X86EMUL_PROPAGATE_FAULT;
bbd9b64e
CO		 4228
4229 /* For APIC access vmexit */
af7cc7d1 4230 if (ret)
bbd9b64e
CO		 4231 goto mmio;
4232
22388a3c 4233 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
bbd9b64e
CO		 4234 return X86EMUL_CONTINUE;
4235
4236mmio:
4237 /*
4238 * Is this MMIO handled locally?
4239 */
22388a3c 4240 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
70252a10 4241 if (handled == bytes)
bbd9b64e 4242 return X86EMUL_CONTINUE;
bbd9b64e 4243
70252a10
AK		 4244 gpa += handled;
4245 bytes -= handled;
4246 val += handled;
4247
87da7e66
XG		 4248 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4249 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4250 frag->gpa = gpa;
4251 frag->data = val;
4252 frag->len = bytes;
f78146b0 4253 return X86EMUL_CONTINUE;
bbd9b64e
CO		 4254}
4255
52eb5a6d
XL		 4256static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4257 unsigned long addr,
22388a3c
XG		 4258 void *val, unsigned int bytes,
4259 struct x86_exception *exception,
0fbe9b0b 4260 const struct read_write_emulator_ops *ops)
bbd9b64e 4261{
0f65dd70 4262 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
f78146b0
AK		 4263 gpa_t gpa;
4264 int rc;
4265
4266 if (ops->read_write_prepare &&
4267 ops->read_write_prepare(vcpu, val, bytes))
4268 return X86EMUL_CONTINUE;
4269
4270 vcpu->mmio_nr_fragments = 0;
0f65dd70 4271
bbd9b64e
CO		 4272 /* Crossing a page boundary? */
4273 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
f78146b0 4274 int now;
bbd9b64e
CO		 4275
4276 now = -addr & ~PAGE_MASK;
22388a3c
XG		 4277 rc = emulator_read_write_onepage(addr, val, now, exception,
4278 vcpu, ops);
4279
bbd9b64e
CO		 4280 if (rc != X86EMUL_CONTINUE)
4281 return rc;
4282 addr += now;
bac15531
NA		 4283 if (ctxt->mode != X86EMUL_MODE_PROT64)
4284 addr = (u32)addr;
bbd9b64e
CO		 4285 val += now;
4286 bytes -= now;
4287 }
22388a3c 4288
f78146b0
AK		 4289 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4290 vcpu, ops);
4291 if (rc != X86EMUL_CONTINUE)
4292 return rc;
4293
4294 if (!vcpu->mmio_nr_fragments)
4295 return rc;
4296
4297 gpa = vcpu->mmio_fragments[0].gpa;
4298
4299 vcpu->mmio_needed = 1;
4300 vcpu->mmio_cur_fragment = 0;
4301
87da7e66 4302 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
f78146b0
AK		 4303 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4304 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4305 vcpu->run->mmio.phys_addr = gpa;
4306
4307 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
22388a3c
XG		 4308}
4309
4310static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4311 unsigned long addr,
4312 void *val,
4313 unsigned int bytes,
4314 struct x86_exception *exception)
4315{
4316 return emulator_read_write(ctxt, addr, val, bytes,
4317 exception, &read_emultor);
4318}
4319
52eb5a6d 4320static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
22388a3c
XG		 4321 unsigned long addr,
4322 const void *val,
4323 unsigned int bytes,
4324 struct x86_exception *exception)
4325{
4326 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4327 exception, &write_emultor);
bbd9b64e 4328}
bbd9b64e 4329
daea3e73
AK		 4330#define CMPXCHG_TYPE(t, ptr, old, new) \
4331 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4332
4333#ifdef CONFIG_X86_64
4334# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4335#else
4336# define CMPXCHG64(ptr, old, new) \
9749a6c0 4337 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
daea3e73
AK		 4338#endif
4339
0f65dd70
AK		 4340static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4341 unsigned long addr,
bbd9b64e
CO		 4342 const void *old,
4343 const void *new,
4344 unsigned int bytes,
0f65dd70 4345 struct x86_exception *exception)
bbd9b64e 4346{
0f65dd70 4347 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
daea3e73
AK		 4348 gpa_t gpa;
4349 struct page *page;
4350 char *kaddr;
4351 bool exchanged;
2bacc55c 4352
daea3e73
AK		 4353 /* guests cmpxchg8b have to be emulated atomically */
4354 if (bytes > 8 || (bytes & (bytes - 1)))
4355 goto emul_write;
10589a46 4356
daea3e73 4357 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
2bacc55c 4358
daea3e73
AK		 4359 if (gpa == UNMAPPED_GVA ||
4360 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4361 goto emul_write;
2bacc55c 4362
daea3e73
AK		 4363 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4364 goto emul_write;
72dc67a6 4365
54bf36aa 4366 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
32cad84f 4367 if (is_error_page(page))
c19b8bd6 4368 goto emul_write;
72dc67a6 4369
8fd75e12 4370 kaddr = kmap_atomic(page);
daea3e73
AK		 4371 kaddr += offset_in_page(gpa);
4372 switch (bytes) {
4373 case 1:
4374 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4375 break;
4376 case 2:
4377 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4378 break;
4379 case 4:
4380 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4381 break;
4382 case 8:
4383 exchanged = CMPXCHG64(kaddr, old, new);
4384 break;
4385 default:
4386 BUG();
2bacc55c 4387 }
8fd75e12 4388 kunmap_atomic(kaddr);
daea3e73
AK		 4389 kvm_release_page_dirty(page);
4390
4391 if (!exchanged)
4392 return X86EMUL_CMPXCHG_FAILED;
4393
54bf36aa 4394 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
f57f2ef5 4395 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
8f6abd06
GN		 4396
4397 return X86EMUL_CONTINUE;
4a5f48f6 4398
3200f405 4399emul_write:
daea3e73 4400 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2bacc55c 4401
0f65dd70 4402 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
bbd9b64e
CO		 4403}
4404
cf8f70bf
GN		 4405static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4406{
4407 /* TODO: String I/O for in kernel device */
4408 int r;
4409
4410 if (vcpu->arch.pio.in)
e32edf4f 4411 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
cf8f70bf
GN		 4412 vcpu->arch.pio.size, pd);
4413 else
e32edf4f 4414 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
cf8f70bf
GN		 4415 vcpu->arch.pio.port, vcpu->arch.pio.size,
4416 pd);
4417 return r;
4418}
4419
6f6fbe98
XG		 4420static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4421 unsigned short port, void *val,
4422 unsigned int count, bool in)
cf8f70bf 4423{
cf8f70bf 4424 vcpu->arch.pio.port = port;
6f6fbe98 4425 vcpu->arch.pio.in = in;
7972995b 4426 vcpu->arch.pio.count = count;
cf8f70bf
GN		 4427 vcpu->arch.pio.size = size;
4428
4429 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
7972995b 4430 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4431 return 1;
4432 }
4433
4434 vcpu->run->exit_reason = KVM_EXIT_IO;
6f6fbe98 4435 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
cf8f70bf
GN		 4436 vcpu->run->io.size = size;
4437 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4438 vcpu->run->io.count = count;
4439 vcpu->run->io.port = port;
4440
4441 return 0;
4442}
4443
6f6fbe98
XG		 4444static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4445 int size, unsigned short port, void *val,
4446 unsigned int count)
cf8f70bf 4447{
ca1d4a9e 4448 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6f6fbe98 4449 int ret;
ca1d4a9e 4450
6f6fbe98
XG		 4451 if (vcpu->arch.pio.count)
4452 goto data_avail;
cf8f70bf 4453
6f6fbe98
XG		 4454 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4455 if (ret) {
4456data_avail:
4457 memcpy(val, vcpu->arch.pio_data, size * count);
1171903d 4458 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
7972995b 4459 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4460 return 1;
4461 }
4462
cf8f70bf
GN		 4463 return 0;
4464}
4465
6f6fbe98
XG		 4466static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4467 int size, unsigned short port,
4468 const void *val, unsigned int count)
4469{
4470 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4471
4472 memcpy(vcpu->arch.pio_data, val, size * count);
1171903d 4473 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6f6fbe98
XG		 4474 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4475}
4476
bbd9b64e
CO		 4477static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4478{
4479 return kvm_x86_ops->get_segment_base(vcpu, seg);
4480}
4481
3cb16fe7 4482static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
bbd9b64e 4483{
3cb16fe7 4484 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
bbd9b64e
CO		 4485}
4486
5cb56059 4487int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
f5f48ee1
SY		 4488{
4489 if (!need_emulate_wbinvd(vcpu))
4490 return X86EMUL_CONTINUE;
4491
4492 if (kvm_x86_ops->has_wbinvd_exit()) {
2eec7343
JK		 4493 int cpu = get_cpu();
4494
4495 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
f5f48ee1
SY		 4496 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4497 wbinvd_ipi, NULL, 1);
2eec7343 4498 put_cpu();
f5f48ee1 4499 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
2eec7343
JK		 4500 } else
4501 wbinvd();
f5f48ee1
SY		 4502 return X86EMUL_CONTINUE;
4503}
5cb56059
JS		 4504
4505int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4506{
4507 kvm_x86_ops->skip_emulated_instruction(vcpu);
4508 return kvm_emulate_wbinvd_noskip(vcpu);
4509}
f5f48ee1
SY		 4510EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4511
5cb56059
JS		 4512
4513
bcaf5cc5
AK		 4514static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4515{
5cb56059 4516 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
bcaf5cc5
AK		 4517}
4518
52eb5a6d
XL		 4519static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4520 unsigned long *dest)
bbd9b64e 4521{
16f8a6f9 4522 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
bbd9b64e
CO		 4523}
4524
52eb5a6d
XL		 4525static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4526 unsigned long value)
bbd9b64e 4527{
338dbc97 4528
717746e3 4529 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
bbd9b64e
CO		 4530}
4531
52a46617 4532static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5fdbf976 4533{
52a46617 4534 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5fdbf976
MT		 4535}
4536
717746e3 4537static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
bbd9b64e 4538{
717746e3 4539 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
52a46617
GN		 4540 unsigned long value;
4541
4542 switch (cr) {
4543 case 0:
4544 value = kvm_read_cr0(vcpu);
4545 break;
4546 case 2:
4547 value = vcpu->arch.cr2;
4548 break;
4549 case 3:
9f8fe504 4550 value = kvm_read_cr3(vcpu);
52a46617
GN		 4551 break;
4552 case 4:
4553 value = kvm_read_cr4(vcpu);
4554 break;
4555 case 8:
4556 value = kvm_get_cr8(vcpu);
4557 break;
4558 default:
a737f256 4559 kvm_err("%s: unexpected cr %u\n", __func__, cr);
52a46617
GN		 4560 return 0;
4561 }
4562
4563 return value;
4564}
4565
717746e3 4566static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
52a46617 4567{
717746e3 4568 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
0f12244f
GN		 4569 int res = 0;
4570
52a46617
GN		 4571 switch (cr) {
4572 case 0:
49a9b07e 4573 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
52a46617
GN		 4574 break;
4575 case 2:
4576 vcpu->arch.cr2 = val;
4577 break;
4578 case 3:
2390218b 4579 res = kvm_set_cr3(vcpu, val);
52a46617
GN		 4580 break;
4581 case 4:
a83b29c6 4582 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
52a46617
GN		 4583 break;
4584 case 8:
eea1cff9 4585 res = kvm_set_cr8(vcpu, val);
52a46617
GN		 4586 break;
4587 default:
a737f256 4588 kvm_err("%s: unexpected cr %u\n", __func__, cr);
0f12244f 4589 res = -1;
52a46617 4590 }
0f12244f
GN		 4591
4592 return res;
52a46617
GN		 4593}
4594
717746e3 4595static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
9c537244 4596{
717746e3 4597 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
9c537244
GN		 4598}
4599
4bff1e86 4600static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
2dafc6c2 4601{
4bff1e86 4602 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
2dafc6c2
GN		 4603}
4604
4bff1e86 4605static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
160ce1f1 4606{
4bff1e86 4607 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
160ce1f1
MG		 4608}
4609
1ac9d0cf
AK		 4610static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4611{
4612 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4613}
4614
4615static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4616{
4617 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4618}
4619
4bff1e86
AK		 4620static unsigned long emulator_get_cached_segment_base(
4621 struct x86_emulate_ctxt *ctxt, int seg)
5951c442 4622{
4bff1e86 4623 return get_segment_base(emul_to_vcpu(ctxt), seg);
5951c442
GN		 4624}
4625
1aa36616
AK		 4626static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4627 struct desc_struct *desc, u32 *base3,
4628 int seg)
2dafc6c2
GN		 4629{
4630 struct kvm_segment var;
4631
4bff1e86 4632 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
1aa36616 4633 *selector = var.selector;
2dafc6c2 4634
378a8b09
GN		 4635 if (var.unusable) {
4636 memset(desc, 0, sizeof(*desc));
2dafc6c2 4637 return false;
378a8b09 4638 }
2dafc6c2
GN		 4639
4640 if (var.g)
4641 var.limit >>= 12;
4642 set_desc_limit(desc, var.limit);
4643 set_desc_base(desc, (unsigned long)var.base);
5601d05b
GN		 4644#ifdef CONFIG_X86_64
4645 if (base3)
4646 *base3 = var.base >> 32;
4647#endif
2dafc6c2
GN		 4648 desc->type = var.type;
4649 desc->s = var.s;
4650 desc->dpl = var.dpl;
4651 desc->p = var.present;
4652 desc->avl = var.avl;
4653 desc->l = var.l;
4654 desc->d = var.db;
4655 desc->g = var.g;
4656
4657 return true;
4658}
4659
1aa36616
AK		 4660static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4661 struct desc_struct *desc, u32 base3,
4662 int seg)
2dafc6c2 4663{
4bff1e86 4664 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
2dafc6c2
GN		 4665 struct kvm_segment var;
4666
1aa36616 4667 var.selector = selector;
2dafc6c2 4668 var.base = get_desc_base(desc);
5601d05b
GN		 4669#ifdef CONFIG_X86_64
4670 var.base |= ((u64)base3) << 32;
4671#endif
2dafc6c2
GN		 4672 var.limit = get_desc_limit(desc);
4673 if (desc->g)
4674 var.limit = (var.limit << 12) | 0xfff;
4675 var.type = desc->type;
2dafc6c2
GN		 4676 var.dpl = desc->dpl;
4677 var.db = desc->d;
4678 var.s = desc->s;
4679 var.l = desc->l;
4680 var.g = desc->g;
4681 var.avl = desc->avl;
4682 var.present = desc->p;
4683 var.unusable = !var.present;
4684 var.padding = 0;
4685
4686 kvm_set_segment(vcpu, &var, seg);
4687 return;
4688}
4689
717746e3
AK		 4690static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4691 u32 msr_index, u64 *pdata)
4692{
609e36d3
PB		 4693 struct msr_data msr;
4694 int r;
4695
4696 msr.index = msr_index;
4697 msr.host_initiated = false;
4698 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4699 if (r)
4700 return r;
4701
4702 *pdata = msr.data;
4703 return 0;
717746e3
AK		 4704}
4705
4706static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4707 u32 msr_index, u64 data)
4708{
8fe8ab46
WA		 4709 struct msr_data msr;
4710
4711 msr.data = data;
4712 msr.index = msr_index;
4713 msr.host_initiated = false;
4714 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
717746e3
AK		 4715}
4716
64d60670
PB		 4717static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4718{
4719 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4720
4721 return vcpu->arch.smbase;
4722}
4723
4724static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4725{
4726 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4727
4728 vcpu->arch.smbase = smbase;
4729}
4730
67f4d428
NA		 4731static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4732 u32 pmc)
4733{
c6702c9d 4734 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
67f4d428
NA		 4735}
4736
222d21aa
AK		 4737static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4738 u32 pmc, u64 *pdata)
4739{
c6702c9d 4740 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
222d21aa
AK		 4741}
4742
6c3287f7
AK		 4743static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4744{
4745 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4746}
4747
5037f6f3
AK		 4748static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4749{
4750 preempt_disable();
5197b808 4751 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5037f6f3
AK		 4752 /*
4753 * CR0.TS may reference the host fpu state, not the guest fpu state,
4754 * so it may be clear at this point.
4755 */
4756 clts();
4757}
4758
4759static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4760{
4761 preempt_enable();
4762}
4763
2953538e 4764static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
8a76d7f2 4765 struct x86_instruction_info *info,
c4f035c6
AK		 4766 enum x86_intercept_stage stage)
4767{
2953538e 4768 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
c4f035c6
AK		 4769}
4770
0017f93a 4771static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
bdb42f5a
SB		 4772 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4773{
0017f93a 4774 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
bdb42f5a
SB		 4775}
4776
dd856efa
AK		 4777static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4778{
4779 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4780}
4781
4782static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4783{
4784 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4785}
4786
801806d9
NA		 4787static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
4788{
4789 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
4790}
4791
0225fb50 4792static const struct x86_emulate_ops emulate_ops = {
dd856efa
AK		 4793 .read_gpr = emulator_read_gpr,
4794 .write_gpr = emulator_write_gpr,
1871c602 4795 .read_std = kvm_read_guest_virt_system,
2dafc6c2 4796 .write_std = kvm_write_guest_virt_system,
1871c602 4797 .fetch = kvm_fetch_guest_virt,
bbd9b64e
CO		 4798 .read_emulated = emulator_read_emulated,
4799 .write_emulated = emulator_write_emulated,
4800 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3cb16fe7 4801 .invlpg = emulator_invlpg,
cf8f70bf
GN		 4802 .pio_in_emulated = emulator_pio_in_emulated,
4803 .pio_out_emulated = emulator_pio_out_emulated,
1aa36616
AK		 4804 .get_segment = emulator_get_segment,
4805 .set_segment = emulator_set_segment,
5951c442 4806 .get_cached_segment_base = emulator_get_cached_segment_base,
2dafc6c2 4807 .get_gdt = emulator_get_gdt,
160ce1f1 4808 .get_idt = emulator_get_idt,
1ac9d0cf
AK		 4809 .set_gdt = emulator_set_gdt,
4810 .set_idt = emulator_set_idt,
52a46617
GN		 4811 .get_cr = emulator_get_cr,
4812 .set_cr = emulator_set_cr,
9c537244 4813 .cpl = emulator_get_cpl,
35aa5375
GN		 4814 .get_dr = emulator_get_dr,
4815 .set_dr = emulator_set_dr,
64d60670
PB		 4816 .get_smbase = emulator_get_smbase,
4817 .set_smbase = emulator_set_smbase,
717746e3
AK		 4818 .set_msr = emulator_set_msr,
4819 .get_msr = emulator_get_msr,
67f4d428 4820 .check_pmc = emulator_check_pmc,
222d21aa 4821 .read_pmc = emulator_read_pmc,
6c3287f7 4822 .halt = emulator_halt,
bcaf5cc5 4823 .wbinvd = emulator_wbinvd,
d6aa1000 4824 .fix_hypercall = emulator_fix_hypercall,
5037f6f3
AK		 4825 .get_fpu = emulator_get_fpu,
4826 .put_fpu = emulator_put_fpu,
c4f035c6 4827 .intercept = emulator_intercept,
bdb42f5a 4828 .get_cpuid = emulator_get_cpuid,
801806d9 4829 .set_nmi_mask = emulator_set_nmi_mask,
bbd9b64e
CO		 4830};
4831
95cb2295
GN		 4832static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4833{
37ccdcbe 4834 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
95cb2295
GN		 4835 /*
4836 * an sti; sti; sequence only disable interrupts for the first
4837 * instruction. So, if the last instruction, be it emulated or
4838 * not, left the system with the INT_STI flag enabled, it
4839 * means that the last instruction is an sti. We should not
4840 * leave the flag on in this case. The same goes for mov ss
4841 */
37ccdcbe
PB		 4842 if (int_shadow & mask)
4843 mask = 0;
6addfc42 4844 if (unlikely(int_shadow || mask)) {
95cb2295 4845 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6addfc42
PB		 4846 if (!mask)
4847 kvm_make_request(KVM_REQ_EVENT, vcpu);
4848 }
95cb2295
GN		 4849}
4850
ef54bcfe 4851static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
54b8486f
GN		 4852{
4853 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
da9cb575 4854 if (ctxt->exception.vector == PF_VECTOR)
ef54bcfe
PB		 4855 return kvm_propagate_fault(vcpu, &ctxt->exception);
4856
4857 if (ctxt->exception.error_code_valid)
da9cb575
AK		 4858 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4859 ctxt->exception.error_code);
54b8486f 4860 else
da9cb575 4861 kvm_queue_exception(vcpu, ctxt->exception.vector);
ef54bcfe 4862 return false;
54b8486f
GN		 4863}
4864
8ec4722d
MG		 4865static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4866{
adf52235 4867 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d
MG		 4868 int cs_db, cs_l;
4869
8ec4722d
MG		 4870 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4871
adf52235
TY		 4872 ctxt->eflags = kvm_get_rflags(vcpu);
4873 ctxt->eip = kvm_rip_read(vcpu);
4874 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4875 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
42bf549f 4876 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
adf52235
TY		 4877 cs_db ? X86EMUL_MODE_PROT32 :
4878 X86EMUL_MODE_PROT16;
a584539b 4879 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
64d60670
PB		 4880 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
4881 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
a584539b 4882 ctxt->emul_flags = vcpu->arch.hflags;
adf52235 4883
dd856efa 4884 init_decode_cache(ctxt);
7ae441ea 4885 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8ec4722d
MG		 4886}
4887
71f9833b 4888int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
63995653 4889{
9d74191a 4890 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
63995653
MG		 4891 int ret;
4892
4893 init_emulate_ctxt(vcpu);
4894
9dac77fa
AK		 4895 ctxt->op_bytes = 2;
4896 ctxt->ad_bytes = 2;
4897 ctxt->_eip = ctxt->eip + inc_eip;
9d74191a 4898 ret = emulate_int_real(ctxt, irq);
63995653
MG		 4899
4900 if (ret != X86EMUL_CONTINUE)
4901 return EMULATE_FAIL;
4902
9dac77fa 4903 ctxt->eip = ctxt->_eip;
9d74191a
TY		 4904 kvm_rip_write(vcpu, ctxt->eip);
4905 kvm_set_rflags(vcpu, ctxt->eflags);
63995653
MG		 4906
4907 if (irq == NMI_VECTOR)
7460fb4a 4908 vcpu->arch.nmi_pending = 0;
63995653
MG		 4909 else
4910 vcpu->arch.interrupt.pending = false;
4911
4912 return EMULATE_DONE;
4913}
4914EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4915
6d77dbfc
GN		 4916static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4917{
fc3a9157
JR		 4918 int r = EMULATE_DONE;
4919
6d77dbfc
GN		 4920 ++vcpu->stat.insn_emulation_fail;
4921 trace_kvm_emulate_insn_failed(vcpu);
a2b9e6c1 4922 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
fc3a9157
JR		 4923 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4924 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4925 vcpu->run->internal.ndata = 0;
4926 r = EMULATE_FAIL;
4927 }
6d77dbfc 4928 kvm_queue_exception(vcpu, UD_VECTOR);
fc3a9157
JR		 4929
4930 return r;
6d77dbfc
GN		 4931}
4932
93c05d3e 4933static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
991eebf9
GN		 4934 bool write_fault_to_shadow_pgtable,
4935 int emulation_type)
a6f177ef 4936{
95b3cf69 4937 gpa_t gpa = cr2;
8e3d9d06 4938 pfn_t pfn;
a6f177ef 4939
991eebf9
GN		 4940 if (emulation_type & EMULTYPE_NO_REEXECUTE)
4941 return false;
4942
95b3cf69
XG		 4943 if (!vcpu->arch.mmu.direct_map) {
4944 /*
4945 * Write permission should be allowed since only
4946 * write access need to be emulated.
4947 */
4948 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
a6f177ef 4949
95b3cf69
XG		 4950 /*
4951 * If the mapping is invalid in guest, let cpu retry
4952 * it to generate fault.
4953 */
4954 if (gpa == UNMAPPED_GVA)
4955 return true;
4956 }
a6f177ef 4957
8e3d9d06
XG		 4958 /*
4959 * Do not retry the unhandleable instruction if it faults on the
4960 * readonly host memory, otherwise it will goto a infinite loop:
4961 * retry instruction -> write #PF -> emulation fail -> retry
4962 * instruction -> ...
4963 */
4964 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
95b3cf69
XG		 4965
4966 /*
4967 * If the instruction failed on the error pfn, it can not be fixed,
4968 * report the error to userspace.
4969 */
4970 if (is_error_noslot_pfn(pfn))
4971 return false;
4972
4973 kvm_release_pfn_clean(pfn);
4974
4975 /* The instructions are well-emulated on direct mmu. */
4976 if (vcpu->arch.mmu.direct_map) {
4977 unsigned int indirect_shadow_pages;
4978
4979 spin_lock(&vcpu->kvm->mmu_lock);
4980 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
4981 spin_unlock(&vcpu->kvm->mmu_lock);
4982
4983 if (indirect_shadow_pages)
4984 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4985
a6f177ef 4986 return true;
8e3d9d06 4987 }
a6f177ef 4988
95b3cf69
XG		 4989 /*
4990 * if emulation was due to access to shadowed page table
4991 * and it failed try to unshadow page and re-enter the
4992 * guest to let CPU execute the instruction.
4993 */
4994 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
93c05d3e
XG		 4995
4996 /*
4997 * If the access faults on its page table, it can not
4998 * be fixed by unprotecting shadow page and it should
4999 * be reported to userspace.
5000 */
5001 return !write_fault_to_shadow_pgtable;
a6f177ef
GN		 5002}
5003
1cb3f3ae
XG		 5004static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5005 unsigned long cr2, int emulation_type)
5006{
5007 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5008 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5009
5010 last_retry_eip = vcpu->arch.last_retry_eip;
5011 last_retry_addr = vcpu->arch.last_retry_addr;
5012
5013 /*
5014 * If the emulation is caused by #PF and it is non-page_table
5015 * writing instruction, it means the VM-EXIT is caused by shadow
5016 * page protected, we can zap the shadow page and retry this
5017 * instruction directly.
5018 *
5019 * Note: if the guest uses a non-page-table modifying instruction
5020 * on the PDE that points to the instruction, then we will unmap
5021 * the instruction and go to an infinite loop. So, we cache the
5022 * last retried eip and the last fault address, if we meet the eip
5023 * and the address again, we can break out of the potential infinite
5024 * loop.
5025 */
5026 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5027
5028 if (!(emulation_type & EMULTYPE_RETRY))
5029 return false;
5030
5031 if (x86_page_table_writing_insn(ctxt))
5032 return false;
5033
5034 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5035 return false;
5036
5037 vcpu->arch.last_retry_eip = ctxt->eip;
5038 vcpu->arch.last_retry_addr = cr2;
5039
5040 if (!vcpu->arch.mmu.direct_map)
5041 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5042
22368028 5043 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
1cb3f3ae
XG		 5044
5045 return true;
5046}
5047
716d51ab
GN		 5048static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5049static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5050
64d60670 5051static void kvm_smm_changed(struct kvm_vcpu *vcpu)
a584539b 5052{
64d60670 5053 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
660a5d51
PB		 5054 /* This is a good place to trace that we are exiting SMM. */
5055 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5056
64d60670
PB		 5057 if (unlikely(vcpu->arch.smi_pending)) {
5058 kvm_make_request(KVM_REQ_SMI, vcpu);
5059 vcpu->arch.smi_pending = 0;
cd7764fe
PB		 5060 } else {
5061 /* Process a latched INIT, if any. */
5062 kvm_make_request(KVM_REQ_EVENT, vcpu);
64d60670
PB		 5063 }
5064 }
699023e2
PB		 5065
5066 kvm_mmu_reset_context(vcpu);
64d60670
PB		 5067}
5068
5069static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5070{
5071 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5072
a584539b 5073 vcpu->arch.hflags = emul_flags;
64d60670
PB		 5074
5075 if (changed & HF_SMM_MASK)
5076 kvm_smm_changed(vcpu);
a584539b
PB		 5077}
5078
4a1e10d5
PB		 5079static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5080 unsigned long *db)
5081{
5082 u32 dr6 = 0;
5083 int i;
5084 u32 enable, rwlen;
5085
5086 enable = dr7;
5087 rwlen = dr7 >> 16;
5088 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5089 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5090 dr6 |= (1 << i);
5091 return dr6;
5092}
5093
6addfc42 5094static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
663f4c61
PB		 5095{
5096 struct kvm_run *kvm_run = vcpu->run;
5097
5098 /*
6addfc42
PB		 5099 * rflags is the old, "raw" value of the flags. The new value has
5100 * not been saved yet.
663f4c61
PB		 5101 *
5102 * This is correct even for TF set by the guest, because "the
5103 * processor will not generate this exception after the instruction
5104 * that sets the TF flag".
5105 */
663f4c61
PB		 5106 if (unlikely(rflags & X86_EFLAGS_TF)) {
5107 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6f43ed01
NA		 5108 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5109 DR6_RTM;
663f4c61
PB		 5110 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5111 kvm_run->debug.arch.exception = DB_VECTOR;
5112 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5113 *r = EMULATE_USER_EXIT;
5114 } else {
5115 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5116 /*
5117 * "Certain debug exceptions may clear bit 0-3. The
5118 * remaining contents of the DR6 register are never
5119 * cleared by the processor".
5120 */
5121 vcpu->arch.dr6 &= ~15;
6f43ed01 5122 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
663f4c61
PB		 5123 kvm_queue_exception(vcpu, DB_VECTOR);
5124 }
5125 }
5126}
5127
4a1e10d5
PB		 5128static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5129{
4a1e10d5
PB		 5130 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5131 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
82b32774
NA		 5132 struct kvm_run *kvm_run = vcpu->run;
5133 unsigned long eip = kvm_get_linear_rip(vcpu);
5134 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5135 vcpu->arch.guest_debug_dr7,
5136 vcpu->arch.eff_db);
5137
5138 if (dr6 != 0) {
6f43ed01 5139 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
82b32774 5140 kvm_run->debug.arch.pc = eip;
4a1e10d5
PB		 5141 kvm_run->debug.arch.exception = DB_VECTOR;
5142 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5143 *r = EMULATE_USER_EXIT;
5144 return true;
5145 }
5146 }
5147
4161a569
NA		 5148 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5149 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
82b32774
NA		 5150 unsigned long eip = kvm_get_linear_rip(vcpu);
5151 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5152 vcpu->arch.dr7,
5153 vcpu->arch.db);
5154
5155 if (dr6 != 0) {
5156 vcpu->arch.dr6 &= ~15;
6f43ed01 5157 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4a1e10d5
PB		 5158 kvm_queue_exception(vcpu, DB_VECTOR);
5159 *r = EMULATE_DONE;
5160 return true;
5161 }
5162 }
5163
5164 return false;
5165}
5166
51d8b661
AP		 5167int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5168 unsigned long cr2,
dc25e89e
AP		 5169 int emulation_type,
5170 void *insn,
5171 int insn_len)
bbd9b64e 5172{
95cb2295 5173 int r;
9d74191a 5174 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7ae441ea 5175 bool writeback = true;
93c05d3e 5176 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
bbd9b64e 5177
93c05d3e
XG		 5178 /*
5179 * Clear write_fault_to_shadow_pgtable here to ensure it is
5180 * never reused.
5181 */
5182 vcpu->arch.write_fault_to_shadow_pgtable = false;
26eef70c 5183 kvm_clear_exception_queue(vcpu);
8d7d8102 5184
571008da 5185 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8ec4722d 5186 init_emulate_ctxt(vcpu);
4a1e10d5
PB		 5187
5188 /*
5189 * We will reenter on the same instruction since
5190 * we do not set complete_userspace_io. This does not
5191 * handle watchpoints yet, those would be handled in
5192 * the emulate_ops.
5193 */
5194 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5195 return r;
5196
9d74191a
TY		 5197 ctxt->interruptibility = 0;
5198 ctxt->have_exception = false;
e0ad0b47 5199 ctxt->exception.vector = -1;
9d74191a 5200 ctxt->perm_ok = false;
bbd9b64e 5201
b51e974f 5202 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
4005996e 5203
9d74191a 5204 r = x86_decode_insn(ctxt, insn, insn_len);
bbd9b64e 5205
e46479f8 5206 trace_kvm_emulate_insn_start(vcpu);
f2b5756b 5207 ++vcpu->stat.insn_emulation;
1d2887e2 5208 if (r != EMULATION_OK) {
4005996e
AK		 5209 if (emulation_type & EMULTYPE_TRAP_UD)
5210 return EMULATE_FAIL;
991eebf9
GN		 5211 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5212 emulation_type))
bbd9b64e 5213 return EMULATE_DONE;
6d77dbfc
GN		 5214 if (emulation_type & EMULTYPE_SKIP)
5215 return EMULATE_FAIL;
5216 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 5217 }
5218 }
5219
ba8afb6b 5220 if (emulation_type & EMULTYPE_SKIP) {
9dac77fa 5221 kvm_rip_write(vcpu, ctxt->_eip);
bb663c7a
NA		 5222 if (ctxt->eflags & X86_EFLAGS_RF)
5223 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
ba8afb6b
GN		 5224 return EMULATE_DONE;
5225 }
5226
1cb3f3ae
XG		 5227 if (retry_instruction(ctxt, cr2, emulation_type))
5228 return EMULATE_DONE;
5229
7ae441ea 5230 /* this is needed for vmware backdoor interface to work since it
4d2179e1 5231 changes registers values during IO operation */
7ae441ea
GN		 5232 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5233 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
dd856efa 5234 emulator_invalidate_register_cache(ctxt);
7ae441ea 5235 }
4d2179e1 5236
5cd21917 5237restart:
9d74191a 5238 r = x86_emulate_insn(ctxt);
bbd9b64e 5239
775fde86
JR		 5240 if (r == EMULATION_INTERCEPTED)
5241 return EMULATE_DONE;
5242
d2ddd1c4 5243 if (r == EMULATION_FAILED) {
991eebf9
GN		 5244 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5245 emulation_type))
c3cd7ffa
GN		 5246 return EMULATE_DONE;
5247
6d77dbfc 5248 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 5249 }
5250
9d74191a 5251 if (ctxt->have_exception) {
d2ddd1c4 5252 r = EMULATE_DONE;
ef54bcfe
PB		 5253 if (inject_emulated_exception(vcpu))
5254 return r;
d2ddd1c4 5255 } else if (vcpu->arch.pio.count) {
0912c977
PB		 5256 if (!vcpu->arch.pio.in) {
5257 /* FIXME: return into emulator if single-stepping. */
3457e419 5258 vcpu->arch.pio.count = 0;
0912c977 5259 } else {
7ae441ea 5260 writeback = false;
716d51ab
GN		 5261 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5262 }
ac0a48c3 5263 r = EMULATE_USER_EXIT;
7ae441ea
GN		 5264 } else if (vcpu->mmio_needed) {
5265 if (!vcpu->mmio_is_write)
5266 writeback = false;
ac0a48c3 5267 r = EMULATE_USER_EXIT;
716d51ab 5268 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7ae441ea 5269 } else if (r == EMULATION_RESTART)
5cd21917 5270 goto restart;
d2ddd1c4
GN		 5271 else
5272 r = EMULATE_DONE;
f850e2e6 5273
7ae441ea 5274 if (writeback) {
6addfc42 5275 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
9d74191a 5276 toggle_interruptibility(vcpu, ctxt->interruptibility);
7ae441ea 5277 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
a584539b
PB		 5278 if (vcpu->arch.hflags != ctxt->emul_flags)
5279 kvm_set_hflags(vcpu, ctxt->emul_flags);
9d74191a 5280 kvm_rip_write(vcpu, ctxt->eip);
663f4c61 5281 if (r == EMULATE_DONE)
6addfc42 5282 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
38827dbd
NA		 5283 if (!ctxt->have_exception ||
5284 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5285 __kvm_set_rflags(vcpu, ctxt->eflags);
6addfc42
PB		 5286
5287 /*
5288 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5289 * do nothing, and it will be requested again as soon as
5290 * the shadow expires. But we still need to check here,
5291 * because POPF has no interrupt shadow.
5292 */
5293 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5294 kvm_make_request(KVM_REQ_EVENT, vcpu);
7ae441ea
GN		 5295 } else
5296 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
e85d28f8
GN		 5297
5298 return r;
de7d789a 5299}
51d8b661 5300EXPORT_SYMBOL_GPL(x86_emulate_instruction);
de7d789a 5301
cf8f70bf 5302int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
de7d789a 5303{
cf8f70bf 5304 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
ca1d4a9e
AK		 5305 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5306 size, port, &val, 1);
cf8f70bf 5307 /* do not return to emulator after return from userspace */
7972995b 5308 vcpu->arch.pio.count = 0;
de7d789a
CO		 5309 return ret;
5310}
cf8f70bf 5311EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
de7d789a 5312
8cfdc000
ZA		 5313static void tsc_bad(void *info)
5314{
0a3aee0d 5315 __this_cpu_write(cpu_tsc_khz, 0);
8cfdc000
ZA		 5316}
5317
5318static void tsc_khz_changed(void *data)
c8076604 5319{
8cfdc000
ZA		 5320 struct cpufreq_freqs *freq = data;
5321 unsigned long khz = 0;
5322
5323 if (data)
5324 khz = freq->new;
5325 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5326 khz = cpufreq_quick_get(raw_smp_processor_id());
5327 if (!khz)
5328 khz = tsc_khz;
0a3aee0d 5329 __this_cpu_write(cpu_tsc_khz, khz);
c8076604
GH		 5330}
5331
c8076604
GH		 5332static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5333 void *data)
5334{
5335 struct cpufreq_freqs *freq = data;
5336 struct kvm *kvm;
5337 struct kvm_vcpu *vcpu;
5338 int i, send_ipi = 0;
5339
8cfdc000
ZA		 5340 /*
5341 * We allow guests to temporarily run on slowing clocks,
5342 * provided we notify them after, or to run on accelerating
5343 * clocks, provided we notify them before. Thus time never
5344 * goes backwards.
5345 *
5346 * However, we have a problem. We can't atomically update
5347 * the frequency of a given CPU from this function; it is
5348 * merely a notifier, which can be called from any CPU.
5349 * Changing the TSC frequency at arbitrary points in time
5350 * requires a recomputation of local variables related to
5351 * the TSC for each VCPU. We must flag these local variables
5352 * to be updated and be sure the update takes place with the
5353 * new frequency before any guests proceed.
5354 *
5355 * Unfortunately, the combination of hotplug CPU and frequency
5356 * change creates an intractable locking scenario; the order
5357 * of when these callouts happen is undefined with respect to
5358 * CPU hotplug, and they can race with each other. As such,
5359 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5360 * undefined; you can actually have a CPU frequency change take
5361 * place in between the computation of X and the setting of the
5362 * variable. To protect against this problem, all updates of
5363 * the per_cpu tsc_khz variable are done in an interrupt
5364 * protected IPI, and all callers wishing to update the value
5365 * must wait for a synchronous IPI to complete (which is trivial
5366 * if the caller is on the CPU already). This establishes the
5367 * necessary total order on variable updates.
5368 *
5369 * Note that because a guest time update may take place
5370 * anytime after the setting of the VCPU's request bit, the
5371 * correct TSC value must be set before the request. However,
5372 * to ensure the update actually makes it to any guest which
5373 * starts running in hardware virtualization between the set
5374 * and the acquisition of the spinlock, we must also ping the
5375 * CPU after setting the request bit.
5376 *
5377 */
5378
c8076604
GH		 5379 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5380 return 0;
5381 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5382 return 0;
8cfdc000
ZA		 5383
5384 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604 5385
2f303b74 5386 spin_lock(&kvm_lock);
c8076604 5387 list_for_each_entry(kvm, &vm_list, vm_list) {
988a2cae 5388 kvm_for_each_vcpu(i, vcpu, kvm) {
c8076604
GH		 5389 if (vcpu->cpu != freq->cpu)
5390 continue;
c285545f 5391 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
c8076604 5392 if (vcpu->cpu != smp_processor_id())
8cfdc000 5393 send_ipi = 1;
c8076604
GH		 5394 }
5395 }
2f303b74 5396 spin_unlock(&kvm_lock);
c8076604
GH		 5397
5398 if (freq->old < freq->new && send_ipi) {
5399 /*
5400 * We upscale the frequency. Must make the guest
5401 * doesn't see old kvmclock values while running with
5402 * the new frequency, otherwise we risk the guest sees
5403 * time go backwards.
5404 *
5405 * In case we update the frequency for another cpu
5406 * (which might be in guest context) send an interrupt
5407 * to kick the cpu out of guest context. Next time
5408 * guest context is entered kvmclock will be updated,
5409 * so the guest will not see stale values.
5410 */
8cfdc000 5411 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604
GH		 5412 }
5413 return 0;
5414}
5415
5416static struct notifier_block kvmclock_cpufreq_notifier_block = {
8cfdc000
ZA		 5417 .notifier_call = kvmclock_cpufreq_notifier
5418};
5419
5420static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5421 unsigned long action, void *hcpu)
5422{
5423 unsigned int cpu = (unsigned long)hcpu;
5424
5425 switch (action) {
5426 case CPU_ONLINE:
5427 case CPU_DOWN_FAILED:
5428 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5429 break;
5430 case CPU_DOWN_PREPARE:
5431 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5432 break;
5433 }
5434 return NOTIFY_OK;
5435}
5436
5437static struct notifier_block kvmclock_cpu_notifier_block = {
5438 .notifier_call = kvmclock_cpu_notifier,
5439 .priority = -INT_MAX
c8076604
GH		 5440};
5441
b820cc0c
ZA		 5442static void kvm_timer_init(void)
5443{
5444 int cpu;
5445
c285545f 5446 max_tsc_khz = tsc_khz;
460dd42e
SB		 5447
5448 cpu_notifier_register_begin();
b820cc0c 5449 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
c285545f
ZA		 5450#ifdef CONFIG_CPU_FREQ
5451 struct cpufreq_policy policy;
5452 memset(&policy, 0, sizeof(policy));
3e26f230
AK		 5453 cpu = get_cpu();
5454 cpufreq_get_policy(&policy, cpu);
c285545f
ZA		 5455 if (policy.cpuinfo.max_freq)
5456 max_tsc_khz = policy.cpuinfo.max_freq;
3e26f230 5457 put_cpu();
c285545f 5458#endif
b820cc0c
ZA		 5459 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5460 CPUFREQ_TRANSITION_NOTIFIER);
5461 }
c285545f 5462 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
8cfdc000
ZA		 5463 for_each_online_cpu(cpu)
5464 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
460dd42e
SB		 5465
5466 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5467 cpu_notifier_register_done();
5468
b820cc0c
ZA		 5469}
5470
ff9d07a0
ZY		 5471static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5472
f5132b01 5473int kvm_is_in_guest(void)
ff9d07a0 5474{
086c9855 5475 return __this_cpu_read(current_vcpu) != NULL;
ff9d07a0
ZY		 5476}
5477
5478static int kvm_is_user_mode(void)
5479{
5480 int user_mode = 3;
dcf46b94 5481
086c9855
AS		 5482 if (__this_cpu_read(current_vcpu))
5483 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
dcf46b94 5484
ff9d07a0
ZY		 5485 return user_mode != 0;
5486}
5487
5488static unsigned long kvm_get_guest_ip(void)
5489{
5490 unsigned long ip = 0;
dcf46b94 5491
086c9855
AS		 5492 if (__this_cpu_read(current_vcpu))
5493 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
dcf46b94 5494
ff9d07a0
ZY		 5495 return ip;
5496}
5497
5498static struct perf_guest_info_callbacks kvm_guest_cbs = {
5499 .is_in_guest = kvm_is_in_guest,
5500 .is_user_mode = kvm_is_user_mode,
5501 .get_guest_ip = kvm_get_guest_ip,
5502};
5503
5504void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5505{
086c9855 5506 __this_cpu_write(current_vcpu, vcpu);
ff9d07a0
ZY		 5507}
5508EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5509
5510void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5511{
086c9855 5512 __this_cpu_write(current_vcpu, NULL);
ff9d07a0
ZY		 5513}
5514EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5515
ce88decf
XG		 5516static void kvm_set_mmio_spte_mask(void)
5517{
5518 u64 mask;
5519 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5520
5521 /*
5522 * Set the reserved bits and the present bit of an paging-structure
5523 * entry to generate page fault with PFER.RSV = 1.
5524 */
885032b9 5525 /* Mask the reserved physical address bits. */
d1431483 5526 mask = rsvd_bits(maxphyaddr, 51);
885032b9
XG		 5527
5528 /* Bit 62 is always reserved for 32bit host. */
5529 mask |= 0x3ull << 62;
5530
5531 /* Set the present bit. */
ce88decf
XG		 5532 mask |= 1ull;
5533
5534#ifdef CONFIG_X86_64
5535 /*
5536 * If reserved bit is not supported, clear the present bit to disable
5537 * mmio page fault.
5538 */
5539 if (maxphyaddr == 52)
5540 mask &= ~1ull;
5541#endif
5542
5543 kvm_mmu_set_mmio_spte_mask(mask);
5544}
5545
16e8d74d
MT		 5546#ifdef CONFIG_X86_64
5547static void pvclock_gtod_update_fn(struct work_struct *work)
5548{
d828199e
MT		 5549 struct kvm *kvm;
5550
5551 struct kvm_vcpu *vcpu;
5552 int i;
5553
2f303b74 5554 spin_lock(&kvm_lock);
d828199e
MT		 5555 list_for_each_entry(kvm, &vm_list, vm_list)
5556 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 5557 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
d828199e 5558 atomic_set(&kvm_guest_has_master_clock, 0);
2f303b74 5559 spin_unlock(&kvm_lock);
16e8d74d
MT		 5560}
5561
5562static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5563
5564/*
5565 * Notification about pvclock gtod data update.
5566 */
5567static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5568 void *priv)
5569{
5570 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5571 struct timekeeper *tk = priv;
5572
5573 update_pvclock_gtod(tk);
5574
5575 /* disable master clock if host does not trust, or does not
5576 * use, TSC clocksource
5577 */
5578 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5579 atomic_read(&kvm_guest_has_master_clock) != 0)
5580 queue_work(system_long_wq, &pvclock_gtod_work);
5581
5582 return 0;
5583}
5584
5585static struct notifier_block pvclock_gtod_notifier = {
5586 .notifier_call = pvclock_gtod_notify,
5587};
5588#endif
5589
f8c16bba 5590int kvm_arch_init(void *opaque)
043405e1 5591{
b820cc0c 5592 int r;
6b61edf7 5593 struct kvm_x86_ops *ops = opaque;
f8c16bba 5594
f8c16bba
ZX		 5595 if (kvm_x86_ops) {
5596 printk(KERN_ERR "kvm: already loaded the other module\n");
56c6d28a
ZX		 5597 r = -EEXIST;
5598 goto out;
f8c16bba
ZX		 5599 }
5600
5601 if (!ops->cpu_has_kvm_support()) {
5602 printk(KERN_ERR "kvm: no hardware support\n");
56c6d28a
ZX		 5603 r = -EOPNOTSUPP;
5604 goto out;
f8c16bba
ZX		 5605 }
5606 if (ops->disabled_by_bios()) {
5607 printk(KERN_ERR "kvm: disabled by bios\n");
56c6d28a
ZX		 5608 r = -EOPNOTSUPP;
5609 goto out;
f8c16bba
ZX		 5610 }
5611
013f6a5d
MT		 5612 r = -ENOMEM;
5613 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5614 if (!shared_msrs) {
5615 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5616 goto out;
5617 }
5618
97db56ce
AK		 5619 r = kvm_mmu_module_init();
5620 if (r)
013f6a5d 5621 goto out_free_percpu;
97db56ce 5622
ce88decf 5623 kvm_set_mmio_spte_mask();
97db56ce 5624
f8c16bba 5625 kvm_x86_ops = ops;
920c8377 5626
7b52345e 5627 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4b12f0de 5628 PT_DIRTY_MASK, PT64_NX_MASK, 0);
c8076604 5629
b820cc0c 5630 kvm_timer_init();
c8076604 5631
ff9d07a0
ZY		 5632 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5633
2acf923e
DC		 5634 if (cpu_has_xsave)
5635 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5636
c5cc421b 5637 kvm_lapic_init();
16e8d74d
MT		 5638#ifdef CONFIG_X86_64
5639 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5640#endif
5641
f8c16bba 5642 return 0;
56c6d28a 5643
013f6a5d
MT		 5644out_free_percpu:
5645 free_percpu(shared_msrs);
56c6d28a 5646out:
56c6d28a 5647 return r;
043405e1 5648}
8776e519 5649
f8c16bba
ZX		 5650void kvm_arch_exit(void)
5651{
ff9d07a0
ZY		 5652 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5653
888d256e
JK		 5654 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5655 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5656 CPUFREQ_TRANSITION_NOTIFIER);
8cfdc000 5657 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
16e8d74d
MT		 5658#ifdef CONFIG_X86_64
5659 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5660#endif
f8c16bba 5661 kvm_x86_ops = NULL;
56c6d28a 5662 kvm_mmu_module_exit();
013f6a5d 5663 free_percpu(shared_msrs);
56c6d28a 5664}
f8c16bba 5665
5cb56059 5666int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8776e519
HB		 5667{
5668 ++vcpu->stat.halt_exits;
35754c98 5669 if (lapic_in_kernel(vcpu)) {
a4535290 5670 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
8776e519
HB		 5671 return 1;
5672 } else {
5673 vcpu->run->exit_reason = KVM_EXIT_HLT;
5674 return 0;
5675 }
5676}
5cb56059
JS		 5677EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5678
5679int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5680{
5681 kvm_x86_ops->skip_emulated_instruction(vcpu);
5682 return kvm_vcpu_halt(vcpu);
5683}
8776e519
HB		 5684EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5685
6aef266c
SV		 5686/*
5687 * kvm_pv_kick_cpu_op: Kick a vcpu.
5688 *
5689 * @apicid - apicid of vcpu to be kicked.
5690 */
5691static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5692{
24d2166b 5693 struct kvm_lapic_irq lapic_irq;
6aef266c 5694
24d2166b
R		 5695 lapic_irq.shorthand = 0;
5696 lapic_irq.dest_mode = 0;
5697 lapic_irq.dest_id = apicid;
93bbf0b8 5698 lapic_irq.msi_redir_hint = false;
6aef266c 5699
24d2166b 5700 lapic_irq.delivery_mode = APIC_DM_REMRD;
795a149e 5701 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6aef266c
SV		 5702}
5703
8776e519
HB		 5704int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5705{
5706 unsigned long nr, a0, a1, a2, a3, ret;
a449c7aa 5707 int op_64_bit, r = 1;
8776e519 5708
5cb56059
JS		 5709 kvm_x86_ops->skip_emulated_instruction(vcpu);
5710
55cd8e5a
GN		 5711 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5712 return kvm_hv_hypercall(vcpu);
5713
5fdbf976
MT		 5714 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5715 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5716 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5717 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5718 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
8776e519 5719
229456fc 5720 trace_kvm_hypercall(nr, a0, a1, a2, a3);
2714d1d3 5721
a449c7aa
NA		 5722 op_64_bit = is_64_bit_mode(vcpu);
5723 if (!op_64_bit) {
8776e519
HB		 5724 nr &= 0xFFFFFFFF;
5725 a0 &= 0xFFFFFFFF;
5726 a1 &= 0xFFFFFFFF;
5727 a2 &= 0xFFFFFFFF;
5728 a3 &= 0xFFFFFFFF;
5729 }
5730
07708c4a
JK		 5731 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5732 ret = -KVM_EPERM;
5733 goto out;
5734 }
5735
8776e519 5736 switch (nr) {
b93463aa
AK		 5737 case KVM_HC_VAPIC_POLL_IRQ:
5738 ret = 0;
5739 break;
6aef266c
SV		 5740 case KVM_HC_KICK_CPU:
5741 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5742 ret = 0;
5743 break;
8776e519
HB		 5744 default:
5745 ret = -KVM_ENOSYS;
5746 break;
5747 }
07708c4a 5748out:
a449c7aa
NA		 5749 if (!op_64_bit)
5750 ret = (u32)ret;
5fdbf976 5751 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
f11c3a8d 5752 ++vcpu->stat.hypercalls;
2f333bcb 5753 return r;
8776e519
HB		 5754}
5755EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5756
b6785def 5757static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8776e519 5758{
d6aa1000 5759 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8776e519 5760 char instruction[3];
5fdbf976 5761 unsigned long rip = kvm_rip_read(vcpu);
8776e519 5762
8776e519 5763 kvm_x86_ops->patch_hypercall(vcpu, instruction);
8776e519 5764
9d74191a 5765 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
8776e519
HB		 5766}
5767
b6c7a5dc
HB		 5768/*
5769 * Check if userspace requested an interrupt window, and that the
5770 * interrupt window is open.
5771 *
5772 * No need to exit to userspace if we already have an interrupt queued.
5773 */
851ba692 5774static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
b6c7a5dc 5775{
8061823a 5776 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
851ba692 5777 vcpu->run->request_interrupt_window &&
5df56646 5778 kvm_arch_interrupt_allowed(vcpu));
b6c7a5dc
HB		 5779}
5780
851ba692 5781static void post_kvm_run_save(struct kvm_vcpu *vcpu)
b6c7a5dc 5782{
851ba692
AK		 5783 struct kvm_run *kvm_run = vcpu->run;
5784
91586a3b 5785 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
f077825a 5786 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
2d3ad1f4 5787 kvm_run->cr8 = kvm_get_cr8(vcpu);
b6c7a5dc 5788 kvm_run->apic_base = kvm_get_apic_base(vcpu);
4531220b 5789 if (irqchip_in_kernel(vcpu->kvm))
b6c7a5dc 5790 kvm_run->ready_for_interrupt_injection = 1;
4531220b 5791 else
b6c7a5dc 5792 kvm_run->ready_for_interrupt_injection =
fa9726b0
GN		 5793 kvm_arch_interrupt_allowed(vcpu) &&
5794 !kvm_cpu_has_interrupt(vcpu) &&
5795 !kvm_event_needs_reinjection(vcpu);
b6c7a5dc
HB		 5796}
5797
95ba8273
GN		 5798static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5799{
5800 int max_irr, tpr;
5801
5802 if (!kvm_x86_ops->update_cr8_intercept)
5803 return;
5804
88c808fd
AK		 5805 if (!vcpu->arch.apic)
5806 return;
5807
8db3baa2
GN		 5808 if (!vcpu->arch.apic->vapic_addr)
5809 max_irr = kvm_lapic_find_highest_irr(vcpu);
5810 else
5811 max_irr = -1;
95ba8273
GN		 5812
5813 if (max_irr != -1)
5814 max_irr >>= 4;
5815
5816 tpr = kvm_lapic_get_cr8(vcpu);
5817
5818 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5819}
5820
b6b8a145 5821static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
95ba8273 5822{
b6b8a145
JK		 5823 int r;
5824
95ba8273 5825 /* try to reinject previous events if any */
b59bb7bd 5826 if (vcpu->arch.exception.pending) {
5c1c85d0
AK		 5827 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5828 vcpu->arch.exception.has_error_code,
5829 vcpu->arch.exception.error_code);
d6e8c854
NA		 5830
5831 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
5832 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
5833 X86_EFLAGS_RF);
5834
6bdf0662
NA		 5835 if (vcpu->arch.exception.nr == DB_VECTOR &&
5836 (vcpu->arch.dr7 & DR7_GD)) {
5837 vcpu->arch.dr7 &= ~DR7_GD;
5838 kvm_update_dr7(vcpu);
5839 }
5840
b59bb7bd
GN		 5841 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5842 vcpu->arch.exception.has_error_code,
ce7ddec4
JR		 5843 vcpu->arch.exception.error_code,
5844 vcpu->arch.exception.reinject);
b6b8a145 5845 return 0;
b59bb7bd
GN		 5846 }
5847
95ba8273
GN		 5848 if (vcpu->arch.nmi_injected) {
5849 kvm_x86_ops->set_nmi(vcpu);
b6b8a145 5850 return 0;
95ba8273
GN		 5851 }
5852
5853 if (vcpu->arch.interrupt.pending) {
66fd3f7f 5854 kvm_x86_ops->set_irq(vcpu);
b6b8a145
JK		 5855 return 0;
5856 }
5857
5858 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5859 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5860 if (r != 0)
5861 return r;
95ba8273
GN		 5862 }
5863
5864 /* try to inject new event if pending */
5865 if (vcpu->arch.nmi_pending) {
5866 if (kvm_x86_ops->nmi_allowed(vcpu)) {
7460fb4a 5867 --vcpu->arch.nmi_pending;
95ba8273
GN		 5868 vcpu->arch.nmi_injected = true;
5869 kvm_x86_ops->set_nmi(vcpu);
5870 }
c7c9c56c 5871 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
9242b5b6
BD		 5872 /*
5873 * Because interrupts can be injected asynchronously, we are
5874 * calling check_nested_events again here to avoid a race condition.
5875 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
5876 * proposal and current concerns. Perhaps we should be setting
5877 * KVM_REQ_EVENT only on certain events and not unconditionally?
5878 */
5879 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5880 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5881 if (r != 0)
5882 return r;
5883 }
95ba8273 5884 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
66fd3f7f
GN		 5885 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5886 false);
5887 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 5888 }
5889 }
b6b8a145 5890 return 0;
95ba8273
GN		 5891}
5892
7460fb4a
AK		 5893static void process_nmi(struct kvm_vcpu *vcpu)
5894{
5895 unsigned limit = 2;
5896
5897 /*
5898 * x86 is limited to one NMI running, and one NMI pending after it.
5899 * If an NMI is already in progress, limit further NMIs to just one.
5900 * Otherwise, allow two (and we'll inject the first one immediately).
5901 */
5902 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5903 limit = 1;
5904
5905 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5906 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5907 kvm_make_request(KVM_REQ_EVENT, vcpu);
5908}
5909
660a5d51
PB		 5910#define put_smstate(type, buf, offset, val) \
5911 *(type *)((buf) + (offset) - 0x7e00) = val
5912
5913static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
5914{
5915 u32 flags = 0;
5916 flags |= seg->g << 23;
5917 flags |= seg->db << 22;
5918 flags |= seg->l << 21;
5919 flags |= seg->avl << 20;
5920 flags |= seg->present << 15;
5921 flags |= seg->dpl << 13;
5922 flags |= seg->s << 12;
5923 flags |= seg->type << 8;
5924 return flags;
5925}
5926
5927static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
5928{
5929 struct kvm_segment seg;
5930 int offset;
5931
5932 kvm_get_segment(vcpu, &seg, n);
5933 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
5934
5935 if (n < 3)
5936 offset = 0x7f84 + n * 12;
5937 else
5938 offset = 0x7f2c + (n - 3) * 12;
5939
5940 put_smstate(u32, buf, offset + 8, seg.base);
5941 put_smstate(u32, buf, offset + 4, seg.limit);
5942 put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
5943}
5944
efbb288a 5945#ifdef CONFIG_X86_64
660a5d51
PB		 5946static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
5947{
5948 struct kvm_segment seg;
5949 int offset;
5950 u16 flags;
5951
5952 kvm_get_segment(vcpu, &seg, n);
5953 offset = 0x7e00 + n * 16;
5954
5955 flags = process_smi_get_segment_flags(&seg) >> 8;
5956 put_smstate(u16, buf, offset, seg.selector);
5957 put_smstate(u16, buf, offset + 2, flags);
5958 put_smstate(u32, buf, offset + 4, seg.limit);
5959 put_smstate(u64, buf, offset + 8, seg.base);
5960}
efbb288a 5961#endif
660a5d51
PB		 5962
5963static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
5964{
5965 struct desc_ptr dt;
5966 struct kvm_segment seg;
5967 unsigned long val;
5968 int i;
5969
5970 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
5971 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
5972 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
5973 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
5974
5975 for (i = 0; i < 8; i++)
5976 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
5977
5978 kvm_get_dr(vcpu, 6, &val);
5979 put_smstate(u32, buf, 0x7fcc, (u32)val);
5980 kvm_get_dr(vcpu, 7, &val);
5981 put_smstate(u32, buf, 0x7fc8, (u32)val);
5982
5983 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
5984 put_smstate(u32, buf, 0x7fc4, seg.selector);
5985 put_smstate(u32, buf, 0x7f64, seg.base);
5986 put_smstate(u32, buf, 0x7f60, seg.limit);
5987 put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
5988
5989 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
5990 put_smstate(u32, buf, 0x7fc0, seg.selector);
5991 put_smstate(u32, buf, 0x7f80, seg.base);
5992 put_smstate(u32, buf, 0x7f7c, seg.limit);
5993 put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
5994
5995 kvm_x86_ops->get_gdt(vcpu, &dt);
5996 put_smstate(u32, buf, 0x7f74, dt.address);
5997 put_smstate(u32, buf, 0x7f70, dt.size);
5998
5999 kvm_x86_ops->get_idt(vcpu, &dt);
6000 put_smstate(u32, buf, 0x7f58, dt.address);
6001 put_smstate(u32, buf, 0x7f54, dt.size);
6002
6003 for (i = 0; i < 6; i++)
6004 process_smi_save_seg_32(vcpu, buf, i);
6005
6006 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6007
6008 /* revision id */
6009 put_smstate(u32, buf, 0x7efc, 0x00020000);
6010 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6011}
6012
6013static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6014{
6015#ifdef CONFIG_X86_64
6016 struct desc_ptr dt;
6017 struct kvm_segment seg;
6018 unsigned long val;
6019 int i;
6020
6021 for (i = 0; i < 16; i++)
6022 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6023
6024 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6025 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6026
6027 kvm_get_dr(vcpu, 6, &val);
6028 put_smstate(u64, buf, 0x7f68, val);
6029 kvm_get_dr(vcpu, 7, &val);
6030 put_smstate(u64, buf, 0x7f60, val);
6031
6032 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6033 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6034 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6035
6036 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6037
6038 /* revision id */
6039 put_smstate(u32, buf, 0x7efc, 0x00020064);
6040
6041 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6042
6043 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6044 put_smstate(u16, buf, 0x7e90, seg.selector);
6045 put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6046 put_smstate(u32, buf, 0x7e94, seg.limit);
6047 put_smstate(u64, buf, 0x7e98, seg.base);
6048
6049 kvm_x86_ops->get_idt(vcpu, &dt);
6050 put_smstate(u32, buf, 0x7e84, dt.size);
6051 put_smstate(u64, buf, 0x7e88, dt.address);
6052
6053 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6054 put_smstate(u16, buf, 0x7e70, seg.selector);
6055 put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6056 put_smstate(u32, buf, 0x7e74, seg.limit);
6057 put_smstate(u64, buf, 0x7e78, seg.base);
6058
6059 kvm_x86_ops->get_gdt(vcpu, &dt);
6060 put_smstate(u32, buf, 0x7e64, dt.size);
6061 put_smstate(u64, buf, 0x7e68, dt.address);
6062
6063 for (i = 0; i < 6; i++)
6064 process_smi_save_seg_64(vcpu, buf, i);
6065#else
6066 WARN_ON_ONCE(1);
6067#endif
6068}
6069
64d60670
PB		 6070static void process_smi(struct kvm_vcpu *vcpu)
6071{
660a5d51 6072 struct kvm_segment cs, ds;
18c3626e 6073 struct desc_ptr dt;
660a5d51
PB		 6074 char buf[512];
6075 u32 cr0;
6076
64d60670
PB		 6077 if (is_smm(vcpu)) {
6078 vcpu->arch.smi_pending = true;
6079 return;
6080 }
6081
660a5d51
PB		 6082 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6083 vcpu->arch.hflags |= HF_SMM_MASK;
6084 memset(buf, 0, 512);
6085 if (guest_cpuid_has_longmode(vcpu))
6086 process_smi_save_state_64(vcpu, buf);
6087 else
6088 process_smi_save_state_32(vcpu, buf);
6089
54bf36aa 6090 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
660a5d51
PB		 6091
6092 if (kvm_x86_ops->get_nmi_mask(vcpu))
6093 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6094 else
6095 kvm_x86_ops->set_nmi_mask(vcpu, true);
6096
6097 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6098 kvm_rip_write(vcpu, 0x8000);
6099
6100 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6101 kvm_x86_ops->set_cr0(vcpu, cr0);
6102 vcpu->arch.cr0 = cr0;
6103
6104 kvm_x86_ops->set_cr4(vcpu, 0);
6105
18c3626e
PB		 6106 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6107 dt.address = dt.size = 0;
6108 kvm_x86_ops->set_idt(vcpu, &dt);
6109
660a5d51
PB		 6110 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6111
6112 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6113 cs.base = vcpu->arch.smbase;
6114
6115 ds.selector = 0;
6116 ds.base = 0;
6117
6118 cs.limit = ds.limit = 0xffffffff;
6119 cs.type = ds.type = 0x3;
6120 cs.dpl = ds.dpl = 0;
6121 cs.db = ds.db = 0;
6122 cs.s = ds.s = 1;
6123 cs.l = ds.l = 0;
6124 cs.g = ds.g = 1;
6125 cs.avl = ds.avl = 0;
6126 cs.present = ds.present = 1;
6127 cs.unusable = ds.unusable = 0;
6128 cs.padding = ds.padding = 0;
6129
6130 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6131 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6132 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6133 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6134 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6135 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6136
6137 if (guest_cpuid_has_longmode(vcpu))
6138 kvm_x86_ops->set_efer(vcpu, 0);
6139
6140 kvm_update_cpuid(vcpu);
6141 kvm_mmu_reset_context(vcpu);
64d60670
PB		 6142}
6143
3d81bc7e 6144static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
c7c9c56c 6145{
3d81bc7e
YZ		 6146 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6147 return;
c7c9c56c 6148
3bb345f3 6149 memset(vcpu->arch.eoi_exit_bitmap, 0, 256 / 8);
c7c9c56c 6150
3bb345f3
PB		 6151 kvm_ioapic_scan_entry(vcpu, vcpu->arch.eoi_exit_bitmap);
6152 kvm_x86_ops->load_eoi_exitmap(vcpu);
c7c9c56c
YZ		 6153}
6154
a70656b6
RK		 6155static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6156{
6157 ++vcpu->stat.tlb_flush;
6158 kvm_x86_ops->tlb_flush(vcpu);
6159}
6160
4256f43f
TC		 6161void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6162{
c24ae0dc
TC		 6163 struct page *page = NULL;
6164
35754c98 6165 if (!lapic_in_kernel(vcpu))
f439ed27
PB		 6166 return;
6167
4256f43f
TC		 6168 if (!kvm_x86_ops->set_apic_access_page_addr)
6169 return;
6170
c24ae0dc 6171 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
e8fd5e9e
AA		 6172 if (is_error_page(page))
6173 return;
c24ae0dc
TC		 6174 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6175
6176 /*
6177 * Do not pin apic access page in memory, the MMU notifier
6178 * will call us again if it is migrated or swapped out.
6179 */
6180 put_page(page);
4256f43f
TC		 6181}
6182EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6183
fe71557a
TC		 6184void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6185 unsigned long address)
6186{
c24ae0dc
TC		 6187 /*
6188 * The physical address of apic access page is stored in the VMCS.
6189 * Update it when it becomes invalid.
6190 */
6191 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6192 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
fe71557a
TC		 6193}
6194
9357d939 6195/*
362c698f 6196 * Returns 1 to let vcpu_run() continue the guest execution loop without
9357d939
TY		 6197 * exiting to the userspace. Otherwise, the value will be returned to the
6198 * userspace.
6199 */
851ba692 6200static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
b6c7a5dc
HB		 6201{
6202 int r;
35754c98 6203 bool req_int_win = !lapic_in_kernel(vcpu) &&
851ba692 6204 vcpu->run->request_interrupt_window;
730dca42 6205 bool req_immediate_exit = false;
b6c7a5dc 6206
3e007509 6207 if (vcpu->requests) {
a8eeb04a 6208 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
2e53d63a 6209 kvm_mmu_unload(vcpu);
a8eeb04a 6210 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
2f599714 6211 __kvm_migrate_timers(vcpu);
d828199e
MT		 6212 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6213 kvm_gen_update_masterclock(vcpu->kvm);
0061d53d
MT		 6214 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6215 kvm_gen_kvmclock_update(vcpu);
34c238a1
ZA		 6216 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6217 r = kvm_guest_time_update(vcpu);
8cfdc000
ZA		 6218 if (unlikely(r))
6219 goto out;
6220 }
a8eeb04a 6221 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
4731d4c7 6222 kvm_mmu_sync_roots(vcpu);
a8eeb04a 6223 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
a70656b6 6224 kvm_vcpu_flush_tlb(vcpu);
a8eeb04a 6225 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
851ba692 6226 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
b93463aa
AK		 6227 r = 0;
6228 goto out;
6229 }
a8eeb04a 6230 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
851ba692 6231 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
71c4dfaf
JR		 6232 r = 0;
6233 goto out;
6234 }
a8eeb04a 6235 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
02daab21
AK		 6236 vcpu->fpu_active = 0;
6237 kvm_x86_ops->fpu_deactivate(vcpu);
6238 }
af585b92
GN		 6239 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6240 /* Page is swapped out. Do synthetic halt */
6241 vcpu->arch.apf.halted = true;
6242 r = 1;
6243 goto out;
6244 }
c9aaa895
GC		 6245 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6246 record_steal_time(vcpu);
64d60670
PB		 6247 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6248 process_smi(vcpu);
7460fb4a
AK		 6249 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6250 process_nmi(vcpu);
f5132b01 6251 if (kvm_check_request(KVM_REQ_PMU, vcpu))
c6702c9d 6252 kvm_pmu_handle_event(vcpu);
f5132b01 6253 if (kvm_check_request(KVM_REQ_PMI, vcpu))
c6702c9d 6254 kvm_pmu_deliver_pmi(vcpu);
3d81bc7e
YZ		 6255 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6256 vcpu_scan_ioapic(vcpu);
4256f43f
TC		 6257 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6258 kvm_vcpu_reload_apic_access_page(vcpu);
2ce79189
AS		 6259 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6260 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6261 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6262 r = 0;
6263 goto out;
6264 }
2f52d58c 6265 }
b93463aa 6266
b463a6f7 6267 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
66450a21
JK		 6268 kvm_apic_accept_events(vcpu);
6269 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6270 r = 1;
6271 goto out;
6272 }
6273
b6b8a145
JK		 6274 if (inject_pending_event(vcpu, req_int_win) != 0)
6275 req_immediate_exit = true;
b463a6f7 6276 /* enable NMI/IRQ window open exits if needed */
b6b8a145 6277 else if (vcpu->arch.nmi_pending)
c9a7953f 6278 kvm_x86_ops->enable_nmi_window(vcpu);
c7c9c56c 6279 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
c9a7953f 6280 kvm_x86_ops->enable_irq_window(vcpu);
b463a6f7
AK		 6281
6282 if (kvm_lapic_enabled(vcpu)) {
c7c9c56c
YZ		 6283 /*
6284 * Update architecture specific hints for APIC
6285 * virtual interrupt delivery.
6286 */
6287 if (kvm_x86_ops->hwapic_irr_update)
6288 kvm_x86_ops->hwapic_irr_update(vcpu,
6289 kvm_lapic_find_highest_irr(vcpu));
b463a6f7
AK		 6290 update_cr8_intercept(vcpu);
6291 kvm_lapic_sync_to_vapic(vcpu);
6292 }
6293 }
6294
d8368af8
AK		 6295 r = kvm_mmu_reload(vcpu);
6296 if (unlikely(r)) {
d905c069 6297 goto cancel_injection;
d8368af8
AK		 6298 }
6299
b6c7a5dc
HB		 6300 preempt_disable();
6301
6302 kvm_x86_ops->prepare_guest_switch(vcpu);
2608d7a1
AK		 6303 if (vcpu->fpu_active)
6304 kvm_load_guest_fpu(vcpu);
2acf923e 6305 kvm_load_guest_xcr0(vcpu);
b6c7a5dc 6306
6b7e2d09
XG		 6307 vcpu->mode = IN_GUEST_MODE;
6308
01b71917
MT		 6309 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6310
6b7e2d09
XG		 6311 /* We should set ->mode before check ->requests,
6312 * see the comment in make_all_cpus_request.
6313 */
01b71917 6314 smp_mb__after_srcu_read_unlock();
b6c7a5dc 6315
d94e1dc9 6316 local_irq_disable();
32f88400 6317
6b7e2d09 6318 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
d94e1dc9 6319 || need_resched() || signal_pending(current)) {
6b7e2d09 6320 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 6321 smp_wmb();
6c142801
AK		 6322 local_irq_enable();
6323 preempt_enable();
01b71917 6324 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6c142801 6325 r = 1;
d905c069 6326 goto cancel_injection;
6c142801
AK		 6327 }
6328
d6185f20
NHE		 6329 if (req_immediate_exit)
6330 smp_send_reschedule(vcpu->cpu);
6331
ccf73aaf 6332 __kvm_guest_enter();
b6c7a5dc 6333
42dbaa5a 6334 if (unlikely(vcpu->arch.switch_db_regs)) {
42dbaa5a
JK		 6335 set_debugreg(0, 7);
6336 set_debugreg(vcpu->arch.eff_db[0], 0);
6337 set_debugreg(vcpu->arch.eff_db[1], 1);
6338 set_debugreg(vcpu->arch.eff_db[2], 2);
6339 set_debugreg(vcpu->arch.eff_db[3], 3);
c77fb5fe 6340 set_debugreg(vcpu->arch.dr6, 6);
ae561ede 6341 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
42dbaa5a 6342 }
b6c7a5dc 6343
229456fc 6344 trace_kvm_entry(vcpu->vcpu_id);
d0659d94 6345 wait_lapic_expire(vcpu);
851ba692 6346 kvm_x86_ops->run(vcpu);
b6c7a5dc 6347
c77fb5fe
PB		 6348 /*
6349 * Do this here before restoring debug registers on the host. And
6350 * since we do this before handling the vmexit, a DR access vmexit
6351 * can (a) read the correct value of the debug registers, (b) set
6352 * KVM_DEBUGREG_WONT_EXIT again.
6353 */
6354 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6355 int i;
6356
6357 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6358 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6359 for (i = 0; i < KVM_NR_DB_REGS; i++)
6360 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6361 }
6362
24f1e32c
FW		 6363 /*
6364 * If the guest has used debug registers, at least dr7
6365 * will be disabled while returning to the host.
6366 * If we don't have active breakpoints in the host, we don't
6367 * care about the messed up debug address registers. But if
6368 * we have some of them active, restore the old state.
6369 */
59d8eb53 6370 if (hw_breakpoint_active())
24f1e32c 6371 hw_breakpoint_restore();
42dbaa5a 6372
886b470c 6373 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
4ea1636b 6374 rdtsc());
1d5f066e 6375
6b7e2d09 6376 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 6377 smp_wmb();
a547c6db
YZ		 6378
6379 /* Interrupt is enabled by handle_external_intr() */
6380 kvm_x86_ops->handle_external_intr(vcpu);
b6c7a5dc
HB		 6381
6382 ++vcpu->stat.exits;
6383
6384 /*
6385 * We must have an instruction between local_irq_enable() and
6386 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6387 * the interrupt shadow. The stat.exits increment will do nicely.
6388 * But we need to prevent reordering, hence this barrier():
6389 */
6390 barrier();
6391
6392 kvm_guest_exit();
6393
6394 preempt_enable();
6395
f656ce01 6396 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3200f405 6397
b6c7a5dc
HB		 6398 /*
6399 * Profile KVM exit RIPs:
6400 */
6401 if (unlikely(prof_on == KVM_PROFILING)) {
5fdbf976
MT		 6402 unsigned long rip = kvm_rip_read(vcpu);
6403 profile_hit(KVM_PROFILING, (void *)rip);
b6c7a5dc
HB		 6404 }
6405
cc578287
ZA		 6406 if (unlikely(vcpu->arch.tsc_always_catchup))
6407 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
298101da 6408
5cfb1d5a
MT		 6409 if (vcpu->arch.apic_attention)
6410 kvm_lapic_sync_from_vapic(vcpu);
b93463aa 6411
851ba692 6412 r = kvm_x86_ops->handle_exit(vcpu);
d905c069
MT		 6413 return r;
6414
6415cancel_injection:
6416 kvm_x86_ops->cancel_injection(vcpu);
ae7a2a3f
MT		 6417 if (unlikely(vcpu->arch.apic_attention))
6418 kvm_lapic_sync_from_vapic(vcpu);
d7690175
MT		 6419out:
6420 return r;
6421}
b6c7a5dc 6422
362c698f
PB		 6423static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6424{
9c8fd1ba
PB		 6425 if (!kvm_arch_vcpu_runnable(vcpu)) {
6426 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6427 kvm_vcpu_block(vcpu);
6428 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6429 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6430 return 1;
6431 }
362c698f
PB		 6432
6433 kvm_apic_accept_events(vcpu);
6434 switch(vcpu->arch.mp_state) {
6435 case KVM_MP_STATE_HALTED:
6436 vcpu->arch.pv.pv_unhalted = false;
6437 vcpu->arch.mp_state =
6438 KVM_MP_STATE_RUNNABLE;
6439 case KVM_MP_STATE_RUNNABLE:
6440 vcpu->arch.apf.halted = false;
6441 break;
6442 case KVM_MP_STATE_INIT_RECEIVED:
6443 break;
6444 default:
6445 return -EINTR;
6446 break;
6447 }
6448 return 1;
6449}
09cec754 6450
362c698f 6451static int vcpu_run(struct kvm_vcpu *vcpu)
d7690175
MT		 6452{
6453 int r;
f656ce01 6454 struct kvm *kvm = vcpu->kvm;
d7690175 6455
f656ce01 6456 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 6457
362c698f 6458 for (;;) {
af585b92
GN		 6459 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6460 !vcpu->arch.apf.halted)
851ba692 6461 r = vcpu_enter_guest(vcpu);
362c698f
PB		 6462 else
6463 r = vcpu_block(kvm, vcpu);
09cec754
GN		 6464 if (r <= 0)
6465 break;
6466
6467 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6468 if (kvm_cpu_has_pending_timer(vcpu))
6469 kvm_inject_pending_timer_irqs(vcpu);
6470
851ba692 6471 if (dm_request_for_irq_injection(vcpu)) {
4ca7dd8c
PB		 6472 r = 0;
6473 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
09cec754 6474 ++vcpu->stat.request_irq_exits;
362c698f 6475 break;
09cec754 6476 }
af585b92
GN		 6477
6478 kvm_check_async_pf_completion(vcpu);
6479
09cec754
GN		 6480 if (signal_pending(current)) {
6481 r = -EINTR;
851ba692 6482 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754 6483 ++vcpu->stat.signal_exits;
362c698f 6484 break;
09cec754
GN		 6485 }
6486 if (need_resched()) {
f656ce01 6487 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
c08ac06a 6488 cond_resched();
f656ce01 6489 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 6490 }
b6c7a5dc
HB		 6491 }
6492
f656ce01 6493 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
b6c7a5dc
HB		 6494
6495 return r;
6496}
6497
716d51ab
GN		 6498static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6499{
6500 int r;
6501 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6502 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6503 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6504 if (r != EMULATE_DONE)
6505 return 0;
6506 return 1;
6507}
6508
6509static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6510{
6511 BUG_ON(!vcpu->arch.pio.count);
6512
6513 return complete_emulated_io(vcpu);
6514}
6515
f78146b0
AK		 6516/*
6517 * Implements the following, as a state machine:
6518 *
6519 * read:
6520 * for each fragment
87da7e66
XG		 6521 * for each mmio piece in the fragment
6522 * write gpa, len
6523 * exit
6524 * copy data
f78146b0
AK		 6525 * execute insn
6526 *
6527 * write:
6528 * for each fragment
87da7e66
XG		 6529 * for each mmio piece in the fragment
6530 * write gpa, len
6531 * copy data
6532 * exit
f78146b0 6533 */
716d51ab 6534static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
5287f194
AK		 6535{
6536 struct kvm_run *run = vcpu->run;
f78146b0 6537 struct kvm_mmio_fragment *frag;
87da7e66 6538 unsigned len;
5287f194 6539
716d51ab 6540 BUG_ON(!vcpu->mmio_needed);
5287f194 6541
716d51ab 6542 /* Complete previous fragment */
87da7e66
XG		 6543 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6544 len = min(8u, frag->len);
716d51ab 6545 if (!vcpu->mmio_is_write)
87da7e66
XG		 6546 memcpy(frag->data, run->mmio.data, len);
6547
6548 if (frag->len <= 8) {
6549 /* Switch to the next fragment. */
6550 frag++;
6551 vcpu->mmio_cur_fragment++;
6552 } else {
6553 /* Go forward to the next mmio piece. */
6554 frag->data += len;
6555 frag->gpa += len;
6556 frag->len -= len;
6557 }
6558
a08d3b3b 6559 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
716d51ab 6560 vcpu->mmio_needed = 0;
0912c977
PB		 6561
6562 /* FIXME: return into emulator if single-stepping. */
cef4dea0 6563 if (vcpu->mmio_is_write)
716d51ab
GN		 6564 return 1;
6565 vcpu->mmio_read_completed = 1;
6566 return complete_emulated_io(vcpu);
6567 }
87da7e66 6568
716d51ab
GN		 6569 run->exit_reason = KVM_EXIT_MMIO;
6570 run->mmio.phys_addr = frag->gpa;
6571 if (vcpu->mmio_is_write)
87da7e66
XG		 6572 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6573 run->mmio.len = min(8u, frag->len);
716d51ab
GN		 6574 run->mmio.is_write = vcpu->mmio_is_write;
6575 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6576 return 0;
5287f194
AK		 6577}
6578
716d51ab 6579
b6c7a5dc
HB		 6580int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6581{
c5bedc68 6582 struct fpu *fpu = &current->thread.fpu;
b6c7a5dc
HB		 6583 int r;
6584 sigset_t sigsaved;
6585
c4d72e2d 6586 fpu__activate_curr(fpu);
e5c30142 6587
ac9f6dc0
AK		 6588 if (vcpu->sigset_active)
6589 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6590
a4535290 6591 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
b6c7a5dc 6592 kvm_vcpu_block(vcpu);
66450a21 6593 kvm_apic_accept_events(vcpu);
d7690175 6594 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
ac9f6dc0
AK		 6595 r = -EAGAIN;
6596 goto out;
b6c7a5dc
HB		 6597 }
6598
b6c7a5dc 6599 /* re-sync apic's tpr */
35754c98 6600 if (!lapic_in_kernel(vcpu)) {
eea1cff9
AP		 6601 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6602 r = -EINVAL;
6603 goto out;
6604 }
6605 }
b6c7a5dc 6606
716d51ab
GN		 6607 if (unlikely(vcpu->arch.complete_userspace_io)) {
6608 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6609 vcpu->arch.complete_userspace_io = NULL;
6610 r = cui(vcpu);
6611 if (r <= 0)
6612 goto out;
6613 } else
6614 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
5287f194 6615
362c698f 6616 r = vcpu_run(vcpu);
b6c7a5dc
HB		 6617
6618out:
f1d86e46 6619 post_kvm_run_save(vcpu);
b6c7a5dc
HB		 6620 if (vcpu->sigset_active)
6621 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6622
b6c7a5dc
HB		 6623 return r;
6624}
6625
6626int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6627{
7ae441ea
GN		 6628 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6629 /*
6630 * We are here if userspace calls get_regs() in the middle of
6631 * instruction emulation. Registers state needs to be copied
4a969980 6632 * back from emulation context to vcpu. Userspace shouldn't do
7ae441ea
GN		 6633 * that usually, but some bad designed PV devices (vmware
6634 * backdoor interface) need this to work
6635 */
dd856efa 6636 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7ae441ea
GN		 6637 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6638 }
5fdbf976
MT		 6639 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6640 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6641 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6642 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6643 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6644 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6645 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6646 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
b6c7a5dc 6647#ifdef CONFIG_X86_64
5fdbf976
MT		 6648 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6649 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6650 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6651 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6652 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6653 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6654 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6655 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
b6c7a5dc
HB		 6656#endif
6657
5fdbf976 6658 regs->rip = kvm_rip_read(vcpu);
91586a3b 6659 regs->rflags = kvm_get_rflags(vcpu);
b6c7a5dc 6660
b6c7a5dc
HB		 6661 return 0;
6662}
6663
6664int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6665{
7ae441ea
GN		 6666 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6667 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6668
5fdbf976
MT		 6669 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6670 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6671 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6672 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6673 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6674 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6675 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6676 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
b6c7a5dc 6677#ifdef CONFIG_X86_64
5fdbf976
MT		 6678 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6679 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6680 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6681 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6682 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6683 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6684 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6685 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
b6c7a5dc
HB		 6686#endif
6687
5fdbf976 6688 kvm_rip_write(vcpu, regs->rip);
91586a3b 6689 kvm_set_rflags(vcpu, regs->rflags);
b6c7a5dc 6690
b4f14abd
JK		 6691 vcpu->arch.exception.pending = false;
6692
3842d135
AK		 6693 kvm_make_request(KVM_REQ_EVENT, vcpu);
6694
b6c7a5dc
HB		 6695 return 0;
6696}
6697
b6c7a5dc
HB		 6698void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6699{
6700 struct kvm_segment cs;
6701
3e6e0aab 6702 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
b6c7a5dc
HB		 6703 *db = cs.db;
6704 *l = cs.l;
6705}
6706EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6707
6708int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6709 struct kvm_sregs *sregs)
6710{
89a27f4d 6711 struct desc_ptr dt;
b6c7a5dc 6712
3e6e0aab
GT		 6713 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6714 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6715 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6716 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6717 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6718 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 6719
3e6e0aab
GT		 6720 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6721 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc
HB		 6722
6723 kvm_x86_ops->get_idt(vcpu, &dt);
89a27f4d
GN		 6724 sregs->idt.limit = dt.size;
6725 sregs->idt.base = dt.address;
b6c7a5dc 6726 kvm_x86_ops->get_gdt(vcpu, &dt);
89a27f4d
GN		 6727 sregs->gdt.limit = dt.size;
6728 sregs->gdt.base = dt.address;
b6c7a5dc 6729
4d4ec087 6730 sregs->cr0 = kvm_read_cr0(vcpu);
ad312c7c 6731 sregs->cr2 = vcpu->arch.cr2;
9f8fe504 6732 sregs->cr3 = kvm_read_cr3(vcpu);
fc78f519 6733 sregs->cr4 = kvm_read_cr4(vcpu);
2d3ad1f4 6734 sregs->cr8 = kvm_get_cr8(vcpu);
f6801dff 6735 sregs->efer = vcpu->arch.efer;
b6c7a5dc
HB		 6736 sregs->apic_base = kvm_get_apic_base(vcpu);
6737
923c61bb 6738 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
b6c7a5dc 6739
36752c9b 6740 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
14d0bc1f
GN		 6741 set_bit(vcpu->arch.interrupt.nr,
6742 (unsigned long *)sregs->interrupt_bitmap);
16d7a191 6743
b6c7a5dc
HB		 6744 return 0;
6745}
6746
62d9f0db
MT		 6747int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6748 struct kvm_mp_state *mp_state)
6749{
66450a21 6750 kvm_apic_accept_events(vcpu);
6aef266c
SV		 6751 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6752 vcpu->arch.pv.pv_unhalted)
6753 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6754 else
6755 mp_state->mp_state = vcpu->arch.mp_state;
6756
62d9f0db
MT		 6757 return 0;
6758}
6759
6760int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6761 struct kvm_mp_state *mp_state)
6762{
66450a21
JK		 6763 if (!kvm_vcpu_has_lapic(vcpu) &&
6764 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6765 return -EINVAL;
6766
6767 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6768 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6769 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6770 } else
6771 vcpu->arch.mp_state = mp_state->mp_state;
3842d135 6772 kvm_make_request(KVM_REQ_EVENT, vcpu);
62d9f0db
MT		 6773 return 0;
6774}
6775
7f3d35fd
KW		 6776int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6777 int reason, bool has_error_code, u32 error_code)
b6c7a5dc 6778{
9d74191a 6779 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d 6780 int ret;
e01c2426 6781
8ec4722d 6782 init_emulate_ctxt(vcpu);
c697518a 6783
7f3d35fd 6784 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9d74191a 6785 has_error_code, error_code);
c697518a 6786
c697518a 6787 if (ret)
19d04437 6788 return EMULATE_FAIL;
37817f29 6789
9d74191a
TY		 6790 kvm_rip_write(vcpu, ctxt->eip);
6791 kvm_set_rflags(vcpu, ctxt->eflags);
3842d135 6792 kvm_make_request(KVM_REQ_EVENT, vcpu);
19d04437 6793 return EMULATE_DONE;
37817f29
IE		 6794}
6795EXPORT_SYMBOL_GPL(kvm_task_switch);
6796
b6c7a5dc
HB		 6797int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6798 struct kvm_sregs *sregs)
6799{
58cb628d 6800 struct msr_data apic_base_msr;
b6c7a5dc 6801 int mmu_reset_needed = 0;
63f42e02 6802 int pending_vec, max_bits, idx;
89a27f4d 6803 struct desc_ptr dt;
b6c7a5dc 6804
6d1068b3
PM		 6805 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6806 return -EINVAL;
6807
89a27f4d
GN		 6808 dt.size = sregs->idt.limit;
6809 dt.address = sregs->idt.base;
b6c7a5dc 6810 kvm_x86_ops->set_idt(vcpu, &dt);
89a27f4d
GN		 6811 dt.size = sregs->gdt.limit;
6812 dt.address = sregs->gdt.base;
b6c7a5dc
HB		 6813 kvm_x86_ops->set_gdt(vcpu, &dt);
6814
ad312c7c 6815 vcpu->arch.cr2 = sregs->cr2;
9f8fe504 6816 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
dc7e795e 6817 vcpu->arch.cr3 = sregs->cr3;
aff48baa 6818 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
b6c7a5dc 6819
2d3ad1f4 6820 kvm_set_cr8(vcpu, sregs->cr8);
b6c7a5dc 6821
f6801dff 6822 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
b6c7a5dc 6823 kvm_x86_ops->set_efer(vcpu, sregs->efer);
58cb628d
JK		 6824 apic_base_msr.data = sregs->apic_base;
6825 apic_base_msr.host_initiated = true;
6826 kvm_set_apic_base(vcpu, &apic_base_msr);
b6c7a5dc 6827
4d4ec087 6828 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
b6c7a5dc 6829 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
d7306163 6830 vcpu->arch.cr0 = sregs->cr0;
b6c7a5dc 6831
fc78f519 6832 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
b6c7a5dc 6833 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3ea3aa8c 6834 if (sregs->cr4 & X86_CR4_OSXSAVE)
00b27a3e 6835 kvm_update_cpuid(vcpu);
63f42e02
XG		 6836
6837 idx = srcu_read_lock(&vcpu->kvm->srcu);
7c93be44 6838 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
9f8fe504 6839 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7c93be44
MT		 6840 mmu_reset_needed = 1;
6841 }
63f42e02 6842 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b6c7a5dc
HB		 6843
6844 if (mmu_reset_needed)
6845 kvm_mmu_reset_context(vcpu);
6846
a50abc3b 6847 max_bits = KVM_NR_INTERRUPTS;
923c61bb
GN		 6848 pending_vec = find_first_bit(
6849 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6850 if (pending_vec < max_bits) {
66fd3f7f 6851 kvm_queue_interrupt(vcpu, pending_vec, false);
923c61bb 6852 pr_debug("Set back pending irq %d\n", pending_vec);
b6c7a5dc
HB		 6853 }
6854
3e6e0aab
GT		 6855 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6856 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6857 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6858 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6859 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6860 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 6861
3e6e0aab
GT		 6862 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6863 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc 6864
5f0269f5
ME		 6865 update_cr8_intercept(vcpu);
6866
9c3e4aab 6867 /* Older userspace won't unhalt the vcpu on reset. */
c5af89b6 6868 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9c3e4aab 6869 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3eeb3288 6870 !is_protmode(vcpu))
9c3e4aab
MT		 6871 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6872
3842d135
AK		 6873 kvm_make_request(KVM_REQ_EVENT, vcpu);
6874
b6c7a5dc
HB		 6875 return 0;
6876}
6877
d0bfb940
JK		 6878int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6879 struct kvm_guest_debug *dbg)
b6c7a5dc 6880{
355be0b9 6881 unsigned long rflags;
ae675ef0 6882 int i, r;
b6c7a5dc 6883
4f926bf2
JK		 6884 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6885 r = -EBUSY;
6886 if (vcpu->arch.exception.pending)
2122ff5e 6887 goto out;
4f926bf2
JK		 6888 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6889 kvm_queue_exception(vcpu, DB_VECTOR);
6890 else
6891 kvm_queue_exception(vcpu, BP_VECTOR);
6892 }
6893
91586a3b
JK		 6894 /*
6895 * Read rflags as long as potentially injected trace flags are still
6896 * filtered out.
6897 */
6898 rflags = kvm_get_rflags(vcpu);
355be0b9
JK		 6899
6900 vcpu->guest_debug = dbg->control;
6901 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6902 vcpu->guest_debug = 0;
6903
6904 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
ae675ef0
JK		 6905 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6906 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
c8639010 6907 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
ae675ef0
JK		 6908 } else {
6909 for (i = 0; i < KVM_NR_DB_REGS; i++)
6910 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
ae675ef0 6911 }
c8639010 6912 kvm_update_dr7(vcpu);
ae675ef0 6913
f92653ee
JK		 6914 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6915 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6916 get_segment_base(vcpu, VCPU_SREG_CS);
94fe45da 6917
91586a3b
JK		 6918 /*
6919 * Trigger an rflags update that will inject or remove the trace
6920 * flags.
6921 */
6922 kvm_set_rflags(vcpu, rflags);
b6c7a5dc 6923
c8639010 6924 kvm_x86_ops->update_db_bp_intercept(vcpu);
b6c7a5dc 6925
4f926bf2 6926 r = 0;
d0bfb940 6927
2122ff5e 6928out:
b6c7a5dc
HB		 6929
6930 return r;
6931}
6932
8b006791
ZX		 6933/*
6934 * Translate a guest virtual address to a guest physical address.
6935 */
6936int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6937 struct kvm_translation *tr)
6938{
6939 unsigned long vaddr = tr->linear_address;
6940 gpa_t gpa;
f656ce01 6941 int idx;
8b006791 6942
f656ce01 6943 idx = srcu_read_lock(&vcpu->kvm->srcu);
1871c602 6944 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
f656ce01 6945 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8b006791
ZX		 6946 tr->physical_address = gpa;
6947 tr->valid = gpa != UNMAPPED_GVA;
6948 tr->writeable = 1;
6949 tr->usermode = 0;
8b006791
ZX		 6950
6951 return 0;
6952}
6953
d0752060
HB		 6954int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6955{
c47ada30 6956 struct fxregs_state *fxsave =
7366ed77 6957 &vcpu->arch.guest_fpu.state.fxsave;
d0752060 6958
d0752060
HB		 6959 memcpy(fpu->fpr, fxsave->st_space, 128);
6960 fpu->fcw = fxsave->cwd;
6961 fpu->fsw = fxsave->swd;
6962 fpu->ftwx = fxsave->twd;
6963 fpu->last_opcode = fxsave->fop;
6964 fpu->last_ip = fxsave->rip;
6965 fpu->last_dp = fxsave->rdp;
6966 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6967
d0752060
HB		 6968 return 0;
6969}
6970
6971int kvm_arch_vcpu_ioctl_set_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(fxsave->st_space, fpu->fpr, 128);
6977 fxsave->cwd = fpu->fcw;
6978 fxsave->swd = fpu->fsw;
6979 fxsave->twd = fpu->ftwx;
6980 fxsave->fop = fpu->last_opcode;
6981 fxsave->rip = fpu->last_ip;
6982 fxsave->rdp = fpu->last_dp;
6983 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6984
d0752060
HB		 6985 return 0;
6986}
6987
0ee6a517 6988static void fx_init(struct kvm_vcpu *vcpu)
d0752060 6989{
bf935b0b 6990 fpstate_init(&vcpu->arch.guest_fpu.state);
df1daba7 6991 if (cpu_has_xsaves)
7366ed77 6992 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
df1daba7 6993 host_xcr0 | XSTATE_COMPACTION_ENABLED;
d0752060 6994
2acf923e
DC		 6995 /*
6996 * Ensure guest xcr0 is valid for loading
6997 */
6998 vcpu->arch.xcr0 = XSTATE_FP;
6999
ad312c7c 7000 vcpu->arch.cr0 |= X86_CR0_ET;
d0752060 7001}
d0752060
HB		 7002
7003void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7004{
2608d7a1 7005 if (vcpu->guest_fpu_loaded)
d0752060
HB		 7006 return;
7007
2acf923e
DC		 7008 /*
7009 * Restore all possible states in the guest,
7010 * and assume host would use all available bits.
7011 * Guest xcr0 would be loaded later.
7012 */
7013 kvm_put_guest_xcr0(vcpu);
d0752060 7014 vcpu->guest_fpu_loaded = 1;
b1a74bf8 7015 __kernel_fpu_begin();
003e2e8b 7016 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
0c04851c 7017 trace_kvm_fpu(1);
d0752060 7018}
d0752060
HB		 7019
7020void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7021{
2acf923e
DC		 7022 kvm_put_guest_xcr0(vcpu);
7023
653f52c3
RR		 7024 if (!vcpu->guest_fpu_loaded) {
7025 vcpu->fpu_counter = 0;
d0752060 7026 return;
653f52c3 7027 }
d0752060
HB		 7028
7029 vcpu->guest_fpu_loaded = 0;
4f836347 7030 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
b1a74bf8 7031 __kernel_fpu_end();
f096ed85 7032 ++vcpu->stat.fpu_reload;
653f52c3
RR		 7033 /*
7034 * If using eager FPU mode, or if the guest is a frequent user
7035 * of the FPU, just leave the FPU active for next time.
7036 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7037 * the FPU in bursts will revert to loading it on demand.
7038 */
a9b4fb7e 7039 if (!vcpu->arch.eager_fpu) {
653f52c3
RR		 7040 if (++vcpu->fpu_counter < 5)
7041 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7042 }
0c04851c 7043 trace_kvm_fpu(0);
d0752060 7044}
e9b11c17
ZX		 7045
7046void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7047{
12f9a48f 7048 kvmclock_reset(vcpu);
7f1ea208 7049
f5f48ee1 7050 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
e9b11c17
ZX		 7051 kvm_x86_ops->vcpu_free(vcpu);
7052}
7053
7054struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7055 unsigned int id)
7056{
c447e76b
LL		 7057 struct kvm_vcpu *vcpu;
7058
6755bae8
ZA		 7059 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7060 printk_once(KERN_WARNING
7061 "kvm: SMP vm created on host with unstable TSC; "
7062 "guest TSC will not be reliable\n");
c447e76b
LL		 7063
7064 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7065
c447e76b 7066 return vcpu;
26e5215f 7067}
e9b11c17 7068
26e5215f
AK		 7069int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7070{
7071 int r;
e9b11c17 7072
19efffa2 7073 kvm_vcpu_mtrr_init(vcpu);
9fc77441
MT		 7074 r = vcpu_load(vcpu);
7075 if (r)
7076 return r;
d28bc9dd 7077 kvm_vcpu_reset(vcpu, false);
8a3c1a33 7078 kvm_mmu_setup(vcpu);
e9b11c17 7079 vcpu_put(vcpu);
26e5215f 7080 return r;
e9b11c17
ZX		 7081}
7082
31928aa5 7083void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
42897d86 7084{
8fe8ab46 7085 struct msr_data msr;
332967a3 7086 struct kvm *kvm = vcpu->kvm;
42897d86 7087
31928aa5
DD		 7088 if (vcpu_load(vcpu))
7089 return;
8fe8ab46
WA		 7090 msr.data = 0x0;
7091 msr.index = MSR_IA32_TSC;
7092 msr.host_initiated = true;
7093 kvm_write_tsc(vcpu, &msr);
42897d86
MT		 7094 vcpu_put(vcpu);
7095
630994b3
MT		 7096 if (!kvmclock_periodic_sync)
7097 return;
7098
332967a3
AJ		 7099 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7100 KVMCLOCK_SYNC_PERIOD);
42897d86
MT		 7101}
7102
d40ccc62 7103void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
e9b11c17 7104{
9fc77441 7105 int r;
344d9588
GN		 7106 vcpu->arch.apf.msr_val = 0;
7107
9fc77441
MT		 7108 r = vcpu_load(vcpu);
7109 BUG_ON(r);
e9b11c17
ZX		 7110 kvm_mmu_unload(vcpu);
7111 vcpu_put(vcpu);
7112
7113 kvm_x86_ops->vcpu_free(vcpu);
7114}
7115
d28bc9dd 7116void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
e9b11c17 7117{
e69fab5d
PB		 7118 vcpu->arch.hflags = 0;
7119
7460fb4a
AK		 7120 atomic_set(&vcpu->arch.nmi_queued, 0);
7121 vcpu->arch.nmi_pending = 0;
448fa4a9 7122 vcpu->arch.nmi_injected = false;
5f7552d4
NA		 7123 kvm_clear_interrupt_queue(vcpu);
7124 kvm_clear_exception_queue(vcpu);
448fa4a9 7125
42dbaa5a 7126 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
ae561ede 7127 kvm_update_dr0123(vcpu);
6f43ed01 7128 vcpu->arch.dr6 = DR6_INIT;
73aaf249 7129 kvm_update_dr6(vcpu);
42dbaa5a 7130 vcpu->arch.dr7 = DR7_FIXED_1;
c8639010 7131 kvm_update_dr7(vcpu);
42dbaa5a 7132
1119022c
NA		 7133 vcpu->arch.cr2 = 0;
7134
3842d135 7135 kvm_make_request(KVM_REQ_EVENT, vcpu);
344d9588 7136 vcpu->arch.apf.msr_val = 0;
c9aaa895 7137 vcpu->arch.st.msr_val = 0;
3842d135 7138
12f9a48f
GC		 7139 kvmclock_reset(vcpu);
7140
af585b92
GN		 7141 kvm_clear_async_pf_completion_queue(vcpu);
7142 kvm_async_pf_hash_reset(vcpu);
7143 vcpu->arch.apf.halted = false;
3842d135 7144
64d60670 7145 if (!init_event) {
d28bc9dd 7146 kvm_pmu_reset(vcpu);
64d60670
PB		 7147 vcpu->arch.smbase = 0x30000;
7148 }
f5132b01 7149
66f7b72e
JS		 7150 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7151 vcpu->arch.regs_avail = ~0;
7152 vcpu->arch.regs_dirty = ~0;
7153
d28bc9dd 7154 kvm_x86_ops->vcpu_reset(vcpu, init_event);
e9b11c17
ZX		 7155}
7156
2b4a273b 7157void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
66450a21
JK		 7158{
7159 struct kvm_segment cs;
7160
7161 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7162 cs.selector = vector << 8;
7163 cs.base = vector << 12;
7164 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7165 kvm_rip_write(vcpu, 0);
e9b11c17
ZX		 7166}
7167
13a34e06 7168int kvm_arch_hardware_enable(void)
e9b11c17 7169{
ca84d1a2
ZA		 7170 struct kvm *kvm;
7171 struct kvm_vcpu *vcpu;
7172 int i;
0dd6a6ed
ZA		 7173 int ret;
7174 u64 local_tsc;
7175 u64 max_tsc = 0;
7176 bool stable, backwards_tsc = false;
18863bdd
AK		 7177
7178 kvm_shared_msr_cpu_online();
13a34e06 7179 ret = kvm_x86_ops->hardware_enable();
0dd6a6ed
ZA		 7180 if (ret != 0)
7181 return ret;
7182
4ea1636b 7183 local_tsc = rdtsc();
0dd6a6ed
ZA		 7184 stable = !check_tsc_unstable();
7185 list_for_each_entry(kvm, &vm_list, vm_list) {
7186 kvm_for_each_vcpu(i, vcpu, kvm) {
7187 if (!stable && vcpu->cpu == smp_processor_id())
105b21bb 7188 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 7189 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7190 backwards_tsc = true;
7191 if (vcpu->arch.last_host_tsc > max_tsc)
7192 max_tsc = vcpu->arch.last_host_tsc;
7193 }
7194 }
7195 }
7196
7197 /*
7198 * Sometimes, even reliable TSCs go backwards. This happens on
7199 * platforms that reset TSC during suspend or hibernate actions, but
7200 * maintain synchronization. We must compensate. Fortunately, we can
7201 * detect that condition here, which happens early in CPU bringup,
7202 * before any KVM threads can be running. Unfortunately, we can't
7203 * bring the TSCs fully up to date with real time, as we aren't yet far
7204 * enough into CPU bringup that we know how much real time has actually
7205 * elapsed; our helper function, get_kernel_ns() will be using boot
7206 * variables that haven't been updated yet.
7207 *
7208 * So we simply find the maximum observed TSC above, then record the
7209 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7210 * the adjustment will be applied. Note that we accumulate
7211 * adjustments, in case multiple suspend cycles happen before some VCPU
7212 * gets a chance to run again. In the event that no KVM threads get a
7213 * chance to run, we will miss the entire elapsed period, as we'll have
7214 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7215 * loose cycle time. This isn't too big a deal, since the loss will be
7216 * uniform across all VCPUs (not to mention the scenario is extremely
7217 * unlikely). It is possible that a second hibernate recovery happens
7218 * much faster than a first, causing the observed TSC here to be
7219 * smaller; this would require additional padding adjustment, which is
7220 * why we set last_host_tsc to the local tsc observed here.
7221 *
7222 * N.B. - this code below runs only on platforms with reliable TSC,
7223 * as that is the only way backwards_tsc is set above. Also note
7224 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7225 * have the same delta_cyc adjustment applied if backwards_tsc
7226 * is detected. Note further, this adjustment is only done once,
7227 * as we reset last_host_tsc on all VCPUs to stop this from being
7228 * called multiple times (one for each physical CPU bringup).
7229 *
4a969980 7230 * Platforms with unreliable TSCs don't have to deal with this, they
0dd6a6ed
ZA		 7231 * will be compensated by the logic in vcpu_load, which sets the TSC to
7232 * catchup mode. This will catchup all VCPUs to real time, but cannot
7233 * guarantee that they stay in perfect synchronization.
7234 */
7235 if (backwards_tsc) {
7236 u64 delta_cyc = max_tsc - local_tsc;
16a96021 7237 backwards_tsc_observed = true;
0dd6a6ed
ZA		 7238 list_for_each_entry(kvm, &vm_list, vm_list) {
7239 kvm_for_each_vcpu(i, vcpu, kvm) {
7240 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7241 vcpu->arch.last_host_tsc = local_tsc;
105b21bb 7242 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 7243 }
7244
7245 /*
7246 * We have to disable TSC offset matching.. if you were
7247 * booting a VM while issuing an S4 host suspend....
7248 * you may have some problem. Solving this issue is
7249 * left as an exercise to the reader.
7250 */
7251 kvm->arch.last_tsc_nsec = 0;
7252 kvm->arch.last_tsc_write = 0;
7253 }
7254
7255 }
7256 return 0;
e9b11c17
ZX		 7257}
7258
13a34e06 7259void kvm_arch_hardware_disable(void)
e9b11c17 7260{
13a34e06
RK		 7261 kvm_x86_ops->hardware_disable();
7262 drop_user_return_notifiers();
e9b11c17
ZX		 7263}
7264
7265int kvm_arch_hardware_setup(void)
7266{
9e9c3fe4
NA		 7267 int r;
7268
7269 r = kvm_x86_ops->hardware_setup();
7270 if (r != 0)
7271 return r;
7272
7273 kvm_init_msr_list();
7274 return 0;
e9b11c17
ZX		 7275}
7276
7277void kvm_arch_hardware_unsetup(void)
7278{
7279 kvm_x86_ops->hardware_unsetup();
7280}
7281
7282void kvm_arch_check_processor_compat(void *rtn)
7283{
7284 kvm_x86_ops->check_processor_compatibility(rtn);
d71ba788
PB		 7285}
7286
7287bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7288{
7289 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7290}
7291EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7292
7293bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7294{
7295 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
e9b11c17
ZX		 7296}
7297
3e515705
AK		 7298bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7299{
35754c98 7300 return irqchip_in_kernel(vcpu->kvm) == lapic_in_kernel(vcpu);
3e515705
AK		 7301}
7302
54e9818f
GN		 7303struct static_key kvm_no_apic_vcpu __read_mostly;
7304
e9b11c17
ZX		 7305int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7306{
7307 struct page *page;
7308 struct kvm *kvm;
7309 int r;
7310
7311 BUG_ON(vcpu->kvm == NULL);
7312 kvm = vcpu->kvm;
7313
6aef266c 7314 vcpu->arch.pv.pv_unhalted = false;
9aabc88f 7315 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
58d269d8 7316 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
a4535290 7317 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
e9b11c17 7318 else
a4535290 7319 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
e9b11c17
ZX		 7320
7321 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7322 if (!page) {
7323 r = -ENOMEM;
7324 goto fail;
7325 }
ad312c7c 7326 vcpu->arch.pio_data = page_address(page);
e9b11c17 7327
cc578287 7328 kvm_set_tsc_khz(vcpu, max_tsc_khz);
c285545f 7329
e9b11c17
ZX		 7330 r = kvm_mmu_create(vcpu);
7331 if (r < 0)
7332 goto fail_free_pio_data;
7333
7334 if (irqchip_in_kernel(kvm)) {
7335 r = kvm_create_lapic(vcpu);
7336 if (r < 0)
7337 goto fail_mmu_destroy;
54e9818f
GN		 7338 } else
7339 static_key_slow_inc(&kvm_no_apic_vcpu);
e9b11c17 7340
890ca9ae
HY		 7341 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7342 GFP_KERNEL);
7343 if (!vcpu->arch.mce_banks) {
7344 r = -ENOMEM;
443c39bc 7345 goto fail_free_lapic;
890ca9ae
HY		 7346 }
7347 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7348
f1797359
WY		 7349 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7350 r = -ENOMEM;
f5f48ee1 7351 goto fail_free_mce_banks;
f1797359 7352 }
f5f48ee1 7353
0ee6a517 7354 fx_init(vcpu);
66f7b72e 7355
ba904635 7356 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
0b79459b 7357 vcpu->arch.pv_time_enabled = false;
d7876f1b
PB		 7358
7359 vcpu->arch.guest_supported_xcr0 = 0;
4344ee98 7360 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
d7876f1b 7361
5a4f55cd
EK		 7362 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7363
74545705
RK		 7364 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7365
af585b92 7366 kvm_async_pf_hash_reset(vcpu);
f5132b01 7367 kvm_pmu_init(vcpu);
af585b92 7368
e9b11c17 7369 return 0;
0ee6a517 7370
f5f48ee1
SY		 7371fail_free_mce_banks:
7372 kfree(vcpu->arch.mce_banks);
443c39bc
WY		 7373fail_free_lapic:
7374 kvm_free_lapic(vcpu);
e9b11c17
ZX		 7375fail_mmu_destroy:
7376 kvm_mmu_destroy(vcpu);
7377fail_free_pio_data:
ad312c7c 7378 free_page((unsigned long)vcpu->arch.pio_data);
e9b11c17
ZX		 7379fail:
7380 return r;
7381}
7382
7383void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7384{
f656ce01
MT		 7385 int idx;
7386
f5132b01 7387 kvm_pmu_destroy(vcpu);
36cb93fd 7388 kfree(vcpu->arch.mce_banks);
e9b11c17 7389 kvm_free_lapic(vcpu);
f656ce01 7390 idx = srcu_read_lock(&vcpu->kvm->srcu);
e9b11c17 7391 kvm_mmu_destroy(vcpu);
f656ce01 7392 srcu_read_unlock(&vcpu->kvm->srcu, idx);
ad312c7c 7393 free_page((unsigned long)vcpu->arch.pio_data);
35754c98 7394 if (!lapic_in_kernel(vcpu))
54e9818f 7395 static_key_slow_dec(&kvm_no_apic_vcpu);
e9b11c17 7396}
d19a9cd2 7397
e790d9ef
RK		 7398void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7399{
ae97a3b8 7400 kvm_x86_ops->sched_in(vcpu, cpu);
e790d9ef
RK		 7401}
7402
e08b9637 7403int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
d19a9cd2 7404{
e08b9637
CO		 7405 if (type)
7406 return -EINVAL;
7407
6ef768fa 7408 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
f05e70ac 7409 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
365c8868 7410 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
4d5c5d0f 7411 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
e0f0bbc5 7412 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
d19a9cd2 7413
5550af4d
SY		 7414 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7415 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7a84428a
AW		 7416 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7417 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7418 &kvm->arch.irq_sources_bitmap);
5550af4d 7419
038f8c11 7420 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
1e08ec4a 7421 mutex_init(&kvm->arch.apic_map_lock);
d828199e
MT		 7422 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7423
7424 pvclock_update_vm_gtod_copy(kvm);
53f658b3 7425
7e44e449 7426 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
332967a3 7427 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7e44e449 7428
d89f5eff 7429 return 0;
d19a9cd2
ZX		 7430}
7431
7432static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7433{
9fc77441
MT		 7434 int r;
7435 r = vcpu_load(vcpu);
7436 BUG_ON(r);
d19a9cd2
ZX		 7437 kvm_mmu_unload(vcpu);
7438 vcpu_put(vcpu);
7439}
7440
7441static void kvm_free_vcpus(struct kvm *kvm)
7442{
7443 unsigned int i;
988a2cae 7444 struct kvm_vcpu *vcpu;
d19a9cd2
ZX		 7445
7446 /*
7447 * Unpin any mmu pages first.
7448 */
af585b92
GN		 7449 kvm_for_each_vcpu(i, vcpu, kvm) {
7450 kvm_clear_async_pf_completion_queue(vcpu);
988a2cae 7451 kvm_unload_vcpu_mmu(vcpu);
af585b92 7452 }
988a2cae
GN		 7453 kvm_for_each_vcpu(i, vcpu, kvm)
7454 kvm_arch_vcpu_free(vcpu);
7455
7456 mutex_lock(&kvm->lock);
7457 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7458 kvm->vcpus[i] = NULL;
d19a9cd2 7459
988a2cae
GN		 7460 atomic_set(&kvm->online_vcpus, 0);
7461 mutex_unlock(&kvm->lock);
d19a9cd2
ZX		 7462}
7463
ad8ba2cd
SY		 7464void kvm_arch_sync_events(struct kvm *kvm)
7465{
332967a3 7466 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7e44e449 7467 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
ba4cef31 7468 kvm_free_all_assigned_devices(kvm);
aea924f6 7469 kvm_free_pit(kvm);
ad8ba2cd
SY		 7470}
7471
9da0e4d5
PB		 7472int __x86_set_memory_region(struct kvm *kvm,
7473 const struct kvm_userspace_memory_region *mem)
7474{
7475 int i, r;
7476
7477 /* Called with kvm->slots_lock held. */
7478 BUG_ON(mem->slot >= KVM_MEM_SLOTS_NUM);
7479
7480 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7481 struct kvm_userspace_memory_region m = *mem;
7482
7483 m.slot |= i << 16;
7484 r = __kvm_set_memory_region(kvm, &m);
7485 if (r < 0)
7486 return r;
7487 }
7488
7489 return 0;
7490}
7491EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7492
7493int x86_set_memory_region(struct kvm *kvm,
7494 const struct kvm_userspace_memory_region *mem)
7495{
7496 int r;
7497
7498 mutex_lock(&kvm->slots_lock);
7499 r = __x86_set_memory_region(kvm, mem);
7500 mutex_unlock(&kvm->slots_lock);
7501
7502 return r;
7503}
7504EXPORT_SYMBOL_GPL(x86_set_memory_region);
7505
d19a9cd2
ZX		 7506void kvm_arch_destroy_vm(struct kvm *kvm)
7507{
27469d29
AH		 7508 if (current->mm == kvm->mm) {
7509 /*
7510 * Free memory regions allocated on behalf of userspace,
7511 * unless the the memory map has changed due to process exit
7512 * or fd copying.
7513 */
7514 struct kvm_userspace_memory_region mem;
7515 memset(&mem, 0, sizeof(mem));
7516 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
9da0e4d5 7517 x86_set_memory_region(kvm, &mem);
27469d29
AH		 7518
7519 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
9da0e4d5 7520 x86_set_memory_region(kvm, &mem);
27469d29
AH		 7521
7522 mem.slot = TSS_PRIVATE_MEMSLOT;
9da0e4d5 7523 x86_set_memory_region(kvm, &mem);
27469d29 7524 }
6eb55818 7525 kvm_iommu_unmap_guest(kvm);
d7deeeb0
ZX		 7526 kfree(kvm->arch.vpic);
7527 kfree(kvm->arch.vioapic);
d19a9cd2 7528 kvm_free_vcpus(kvm);
1e08ec4a 7529 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
d19a9cd2 7530}
0de10343 7531
5587027c 7532void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
db3fe4eb
TY		 7533 struct kvm_memory_slot *dont)
7534{
7535 int i;
7536
d89cc617
TY		 7537 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7538 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
548ef284 7539 kvfree(free->arch.rmap[i]);
d89cc617 7540 free->arch.rmap[i] = NULL;
77d11309 7541 }
d89cc617
TY		 7542 if (i == 0)
7543 continue;
7544
7545 if (!dont || free->arch.lpage_info[i - 1] !=
7546 dont->arch.lpage_info[i - 1]) {
548ef284 7547 kvfree(free->arch.lpage_info[i - 1]);
d89cc617 7548 free->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 7549 }
7550 }
7551}
7552
5587027c
AK		 7553int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7554 unsigned long npages)
db3fe4eb
TY		 7555{
7556 int i;
7557
d89cc617 7558 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
db3fe4eb
TY		 7559 unsigned long ugfn;
7560 int lpages;
d89cc617 7561 int level = i + 1;
db3fe4eb
TY		 7562
7563 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7564 slot->base_gfn, level) + 1;
7565
d89cc617
TY		 7566 slot->arch.rmap[i] =
7567 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7568 if (!slot->arch.rmap[i])
77d11309 7569 goto out_free;
d89cc617
TY		 7570 if (i == 0)
7571 continue;
77d11309 7572
d89cc617
TY		 7573 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7574 sizeof(*slot->arch.lpage_info[i - 1]));
7575 if (!slot->arch.lpage_info[i - 1])
db3fe4eb
TY		 7576 goto out_free;
7577
7578 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
d89cc617 7579 slot->arch.lpage_info[i - 1][0].write_count = 1;
db3fe4eb 7580 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
d89cc617 7581 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
db3fe4eb
TY		 7582 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7583 /*
7584 * If the gfn and userspace address are not aligned wrt each
7585 * other, or if explicitly asked to, disable large page
7586 * support for this slot
7587 */
7588 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7589 !kvm_largepages_enabled()) {
7590 unsigned long j;
7591
7592 for (j = 0; j < lpages; ++j)
d89cc617 7593 slot->arch.lpage_info[i - 1][j].write_count = 1;
db3fe4eb
TY		 7594 }
7595 }
7596
7597 return 0;
7598
7599out_free:
d89cc617 7600 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
548ef284 7601 kvfree(slot->arch.rmap[i]);
d89cc617
TY		 7602 slot->arch.rmap[i] = NULL;
7603 if (i == 0)
7604 continue;
7605
548ef284 7606 kvfree(slot->arch.lpage_info[i - 1]);
d89cc617 7607 slot->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 7608 }
7609 return -ENOMEM;
7610}
7611
15f46015 7612void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
e59dbe09 7613{
e6dff7d1
TY		 7614 /*
7615 * memslots->generation has been incremented.
7616 * mmio generation may have reached its maximum value.
7617 */
54bf36aa 7618 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
e59dbe09
TY		 7619}
7620
f7784b8e
MT		 7621int kvm_arch_prepare_memory_region(struct kvm *kvm,
7622 struct kvm_memory_slot *memslot,
09170a49 7623 const struct kvm_userspace_memory_region *mem,
7b6195a9 7624 enum kvm_mr_change change)
0de10343 7625{
7a905b14
TY		 7626 /*
7627 * Only private memory slots need to be mapped here since
7628 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
0de10343 7629 */
7b6195a9 7630 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7a905b14 7631 unsigned long userspace_addr;
604b38ac 7632
7a905b14
TY		 7633 /*
7634 * MAP_SHARED to prevent internal slot pages from being moved
7635 * by fork()/COW.
7636 */
7b6195a9 7637 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7a905b14
TY		 7638 PROT_READ | PROT_WRITE,
7639 MAP_SHARED | MAP_ANONYMOUS, 0);
0de10343 7640
7a905b14
TY		 7641 if (IS_ERR((void *)userspace_addr))
7642 return PTR_ERR((void *)userspace_addr);
604b38ac 7643
7a905b14 7644 memslot->userspace_addr = userspace_addr;
0de10343
ZX		 7645 }
7646
f7784b8e
MT		 7647 return 0;
7648}
7649
88178fd4
KH		 7650static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7651 struct kvm_memory_slot *new)
7652{
7653 /* Still write protect RO slot */
7654 if (new->flags & KVM_MEM_READONLY) {
7655 kvm_mmu_slot_remove_write_access(kvm, new);
7656 return;
7657 }
7658
7659 /*
7660 * Call kvm_x86_ops dirty logging hooks when they are valid.
7661 *
7662 * kvm_x86_ops->slot_disable_log_dirty is called when:
7663 *
7664 * - KVM_MR_CREATE with dirty logging is disabled
7665 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7666 *
7667 * The reason is, in case of PML, we need to set D-bit for any slots
7668 * with dirty logging disabled in order to eliminate unnecessary GPA
7669 * logging in PML buffer (and potential PML buffer full VMEXT). This
7670 * guarantees leaving PML enabled during guest's lifetime won't have
7671 * any additonal overhead from PML when guest is running with dirty
7672 * logging disabled for memory slots.
7673 *
7674 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7675 * to dirty logging mode.
7676 *
7677 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7678 *
7679 * In case of write protect:
7680 *
7681 * Write protect all pages for dirty logging.
7682 *
7683 * All the sptes including the large sptes which point to this
7684 * slot are set to readonly. We can not create any new large
7685 * spte on this slot until the end of the logging.
7686 *
7687 * See the comments in fast_page_fault().
7688 */
7689 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7690 if (kvm_x86_ops->slot_enable_log_dirty)
7691 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7692 else
7693 kvm_mmu_slot_remove_write_access(kvm, new);
7694 } else {
7695 if (kvm_x86_ops->slot_disable_log_dirty)
7696 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7697 }
7698}
7699
f7784b8e 7700void kvm_arch_commit_memory_region(struct kvm *kvm,
09170a49 7701 const struct kvm_userspace_memory_region *mem,
8482644a 7702 const struct kvm_memory_slot *old,
f36f3f28 7703 const struct kvm_memory_slot *new,
8482644a 7704 enum kvm_mr_change change)
f7784b8e 7705{
8482644a 7706 int nr_mmu_pages = 0;
f7784b8e 7707
f36f3f28 7708 if (change == KVM_MR_DELETE && old->id >= KVM_USER_MEM_SLOTS) {
f7784b8e
MT		 7709 int ret;
7710
8482644a
TY		 7711 ret = vm_munmap(old->userspace_addr,
7712 old->npages * PAGE_SIZE);
f7784b8e
MT		 7713 if (ret < 0)
7714 printk(KERN_WARNING
7715 "kvm_vm_ioctl_set_memory_region: "
7716 "failed to munmap memory\n");
7717 }
7718
48c0e4e9
XG		 7719 if (!kvm->arch.n_requested_mmu_pages)
7720 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7721
48c0e4e9 7722 if (nr_mmu_pages)
0de10343 7723 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
1c91cad4 7724
3ea3b7fa
WL		 7725 /*
7726 * Dirty logging tracks sptes in 4k granularity, meaning that large
7727 * sptes have to be split. If live migration is successful, the guest
7728 * in the source machine will be destroyed and large sptes will be
7729 * created in the destination. However, if the guest continues to run
7730 * in the source machine (for example if live migration fails), small
7731 * sptes will remain around and cause bad performance.
7732 *
7733 * Scan sptes if dirty logging has been stopped, dropping those
7734 * which can be collapsed into a single large-page spte. Later
7735 * page faults will create the large-page sptes.
7736 */
7737 if ((change != KVM_MR_DELETE) &&
7738 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7739 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7740 kvm_mmu_zap_collapsible_sptes(kvm, new);
7741
c972f3b1 7742 /*
88178fd4 7743 * Set up write protection and/or dirty logging for the new slot.
c126d94f 7744 *
88178fd4
KH		 7745 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7746 * been zapped so no dirty logging staff is needed for old slot. For
7747 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7748 * new and it's also covered when dealing with the new slot.
f36f3f28
PB		 7749 *
7750 * FIXME: const-ify all uses of struct kvm_memory_slot.
c972f3b1 7751 */
88178fd4 7752 if (change != KVM_MR_DELETE)
f36f3f28 7753 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
0de10343 7754}
1d737c8a 7755
2df72e9b 7756void kvm_arch_flush_shadow_all(struct kvm *kvm)
34d4cb8f 7757{
6ca18b69 7758 kvm_mmu_invalidate_zap_all_pages(kvm);
34d4cb8f
MT		 7759}
7760
2df72e9b
MT		 7761void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7762 struct kvm_memory_slot *slot)
7763{
6ca18b69 7764 kvm_mmu_invalidate_zap_all_pages(kvm);
2df72e9b
MT		 7765}
7766
1d737c8a
ZX		 7767int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7768{
b6b8a145
JK		 7769 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7770 kvm_x86_ops->check_nested_events(vcpu, false);
7771
af585b92
GN		 7772 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7773 !vcpu->arch.apf.halted)
7774 || !list_empty_careful(&vcpu->async_pf.done)
66450a21 7775 || kvm_apic_has_events(vcpu)
6aef266c 7776 || vcpu->arch.pv.pv_unhalted
7460fb4a 7777 || atomic_read(&vcpu->arch.nmi_queued) ||
a1b37100
GN		 7778 (kvm_arch_interrupt_allowed(vcpu) &&
7779 kvm_cpu_has_interrupt(vcpu));
1d737c8a 7780}
5736199a 7781
b6d33834 7782int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
5736199a 7783{
b6d33834 7784 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
5736199a 7785}
78646121
GN		 7786
7787int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7788{
7789 return kvm_x86_ops->interrupt_allowed(vcpu);
7790}
229456fc 7791
82b32774 7792unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
f92653ee 7793{
82b32774
NA		 7794 if (is_64_bit_mode(vcpu))
7795 return kvm_rip_read(vcpu);
7796 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
7797 kvm_rip_read(vcpu));
7798}
7799EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
f92653ee 7800
82b32774
NA		 7801bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7802{
7803 return kvm_get_linear_rip(vcpu) == linear_rip;
f92653ee
JK		 7804}
7805EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7806
94fe45da
JK		 7807unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7808{
7809 unsigned long rflags;
7810
7811 rflags = kvm_x86_ops->get_rflags(vcpu);
7812 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
c310bac5 7813 rflags &= ~X86_EFLAGS_TF;
94fe45da
JK		 7814 return rflags;
7815}
7816EXPORT_SYMBOL_GPL(kvm_get_rflags);
7817
6addfc42 7818static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
94fe45da
JK		 7819{
7820 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
f92653ee 7821 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
c310bac5 7822 rflags |= X86_EFLAGS_TF;
94fe45da 7823 kvm_x86_ops->set_rflags(vcpu, rflags);
6addfc42
PB		 7824}
7825
7826void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7827{
7828 __kvm_set_rflags(vcpu, rflags);
3842d135 7829 kvm_make_request(KVM_REQ_EVENT, vcpu);
94fe45da
JK		 7830}
7831EXPORT_SYMBOL_GPL(kvm_set_rflags);
7832
56028d08
GN		 7833void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7834{
7835 int r;
7836
fb67e14f 7837 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
f2e10669 7838 work->wakeup_all)
56028d08
GN		 7839 return;
7840
7841 r = kvm_mmu_reload(vcpu);
7842 if (unlikely(r))
7843 return;
7844
fb67e14f
XG		 7845 if (!vcpu->arch.mmu.direct_map &&
7846 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7847 return;
7848
56028d08
GN		 7849 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7850}
7851
af585b92
GN		 7852static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7853{
7854 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7855}
7856
7857static inline u32 kvm_async_pf_next_probe(u32 key)
7858{
7859 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7860}
7861
7862static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7863{
7864 u32 key = kvm_async_pf_hash_fn(gfn);
7865
7866 while (vcpu->arch.apf.gfns[key] != ~0)
7867 key = kvm_async_pf_next_probe(key);
7868
7869 vcpu->arch.apf.gfns[key] = gfn;
7870}
7871
7872static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7873{
7874 int i;
7875 u32 key = kvm_async_pf_hash_fn(gfn);
7876
7877 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
c7d28c24
XG		 7878 (vcpu->arch.apf.gfns[key] != gfn &&
7879 vcpu->arch.apf.gfns[key] != ~0); i++)
af585b92
GN		 7880 key = kvm_async_pf_next_probe(key);
7881
7882 return key;
7883}
7884
7885bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7886{
7887 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7888}
7889
7890static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7891{
7892 u32 i, j, k;
7893
7894 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7895 while (true) {
7896 vcpu->arch.apf.gfns[i] = ~0;
7897 do {
7898 j = kvm_async_pf_next_probe(j);
7899 if (vcpu->arch.apf.gfns[j] == ~0)
7900 return;
7901 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7902 /*
7903 * k lies cyclically in ]i,j]
7904 * | i.k.j |
7905 * |....j i.k.| or |.k..j i...|
7906 */
7907 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7908 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7909 i = j;
7910 }
7911}
7912
7c90705b
GN		 7913static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7914{
7915
7916 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7917 sizeof(val));
7918}
7919
af585b92
GN		 7920void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7921 struct kvm_async_pf *work)
7922{
6389ee94
AK		 7923 struct x86_exception fault;
7924
7c90705b 7925 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
af585b92 7926 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7c90705b
GN		 7927
7928 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
fc5f06fa
GN		 7929 (vcpu->arch.apf.send_user_only &&
7930 kvm_x86_ops->get_cpl(vcpu) == 0))
7c90705b
GN		 7931 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7932 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6389ee94
AK		 7933 fault.vector = PF_VECTOR;
7934 fault.error_code_valid = true;
7935 fault.error_code = 0;
7936 fault.nested_page_fault = false;
7937 fault.address = work->arch.token;
7938 kvm_inject_page_fault(vcpu, &fault);
7c90705b 7939 }
af585b92
GN		 7940}
7941
7942void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7943 struct kvm_async_pf *work)
7944{
6389ee94
AK		 7945 struct x86_exception fault;
7946
7c90705b 7947 trace_kvm_async_pf_ready(work->arch.token, work->gva);
f2e10669 7948 if (work->wakeup_all)
7c90705b
GN		 7949 work->arch.token = ~0; /* broadcast wakeup */
7950 else
7951 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7952
7953 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7954 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6389ee94
AK		 7955 fault.vector = PF_VECTOR;
7956 fault.error_code_valid = true;
7957 fault.error_code = 0;
7958 fault.nested_page_fault = false;
7959 fault.address = work->arch.token;
7960 kvm_inject_page_fault(vcpu, &fault);
7c90705b 7961 }
e6d53e3b 7962 vcpu->arch.apf.halted = false;
a4fa1635 7963 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7c90705b
GN		 7964}
7965
7966bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7967{
7968 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7969 return true;
7970 else
7971 return !kvm_event_needs_reinjection(vcpu) &&
7972 kvm_x86_ops->interrupt_allowed(vcpu);
af585b92
GN		 7973}
7974
5544eb9b
PB		 7975void kvm_arch_start_assignment(struct kvm *kvm)
7976{
7977 atomic_inc(&kvm->arch.assigned_device_count);
7978}
7979EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
7980
7981void kvm_arch_end_assignment(struct kvm *kvm)
7982{
7983 atomic_dec(&kvm->arch.assigned_device_count);
7984}
7985EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
7986
7987bool kvm_arch_has_assigned_device(struct kvm *kvm)
7988{
7989 return atomic_read(&kvm->arch.assigned_device_count);
7990}
7991EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
7992
e0f0bbc5
AW		 7993void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
7994{
7995 atomic_inc(&kvm->arch.noncoherent_dma_count);
7996}
7997EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
7998
7999void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8000{
8001 atomic_dec(&kvm->arch.noncoherent_dma_count);
8002}
8003EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8004
8005bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8006{
8007 return atomic_read(&kvm->arch.noncoherent_dma_count);
8008}
8009EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8010
229456fc
MT		 8011EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8012EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8013EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8014EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8015EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
0ac406de 8016EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
d8cabddf 8017EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
17897f36 8018EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
236649de 8019EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
ec1ff790 8020EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
532a46b9 8021EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
2e554e8d 8022EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
489223ed 8023EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7b46268d 8024EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
843e4330 8025EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
ubuntu-bionic-kernel mirror