]> 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: latch INITs while in system management mode
[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"
313a3dc7 31
18068523 32#include <linux/clocksource.h>
4d5c5d0f 33#include <linux/interrupt.h>
313a3dc7
CO		 34#include <linux/kvm.h>
35#include <linux/fs.h>
36#include <linux/vmalloc.h>
5fb76f9b 37#include <linux/module.h>
0de10343 38#include <linux/mman.h>
2bacc55c 39#include <linux/highmem.h>
19de40a8 40#include <linux/iommu.h>
62c476c7 41#include <linux/intel-iommu.h>
c8076604 42#include <linux/cpufreq.h>
18863bdd 43#include <linux/user-return-notifier.h>
a983fb23 44#include <linux/srcu.h>
5a0e3ad6 45#include <linux/slab.h>
ff9d07a0 46#include <linux/perf_event.h>
7bee342a 47#include <linux/uaccess.h>
af585b92 48#include <linux/hash.h>
a1b60c1c 49#include <linux/pci.h>
16e8d74d
MT		 50#include <linux/timekeeper_internal.h>
51#include <linux/pvclock_gtod.h>
aec51dc4 52#include <trace/events/kvm.h>
2ed152af 53
229456fc
MT		 54#define CREATE_TRACE_POINTS
55#include "trace.h"
043405e1 56
24f1e32c 57#include <asm/debugreg.h>
d825ed0a 58#include <asm/msr.h>
a5f61300 59#include <asm/desc.h>
0bed3b56 60#include <asm/mtrr.h>
890ca9ae 61#include <asm/mce.h>
7cf30855 62#include <asm/i387.h>
1361b83a 63#include <asm/fpu-internal.h> /* Ugh! */
98918833 64#include <asm/xcr.h>
1d5f066e 65#include <asm/pvclock.h>
217fc9cf 66#include <asm/div64.h>
043405e1 67
313a3dc7 68#define MAX_IO_MSRS 256
890ca9ae 69#define KVM_MAX_MCE_BANKS 32
5854dbca 70#define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
890ca9ae 71
0f65dd70
AK		 72#define emul_to_vcpu(ctxt) \
73 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
74
50a37eb4
JR		 75/* EFER defaults:
76 * - enable syscall per default because its emulated by KVM
77 * - enable LME and LMA per default on 64 bit KVM
78 */
79#ifdef CONFIG_X86_64
1260edbe
LJ		 80static
81u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
50a37eb4 82#else
1260edbe 83static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
50a37eb4 84#endif
313a3dc7 85
ba1389b7
AK		 86#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
87#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
417bc304 88
cb142eb7 89static void update_cr8_intercept(struct kvm_vcpu *vcpu);
7460fb4a 90static void process_nmi(struct kvm_vcpu *vcpu);
6addfc42 91static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
674eea0f 92
97896d04 93struct kvm_x86_ops *kvm_x86_ops;
5fdbf976 94EXPORT_SYMBOL_GPL(kvm_x86_ops);
97896d04 95
476bc001
RR		 96static bool ignore_msrs = 0;
97module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
ed85c068 98
9ed96e87
MT		 99unsigned int min_timer_period_us = 500;
100module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
101
630994b3
MT		 102static bool __read_mostly kvmclock_periodic_sync = true;
103module_param(kvmclock_periodic_sync, bool, S_IRUGO);
104
92a1f12d
JR		 105bool kvm_has_tsc_control;
106EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
107u32 kvm_max_guest_tsc_khz;
108EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
109
cc578287
ZA		 110/* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
111static u32 tsc_tolerance_ppm = 250;
112module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
113
d0659d94
MT		 114/* lapic timer advance (tscdeadline mode only) in nanoseconds */
115unsigned int lapic_timer_advance_ns = 0;
116module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
117
16a96021
MT		 118static bool backwards_tsc_observed = false;
119
18863bdd
AK		 120#define KVM_NR_SHARED_MSRS 16
121
122struct kvm_shared_msrs_global {
123 int nr;
2bf78fa7 124 u32 msrs[KVM_NR_SHARED_MSRS];
18863bdd
AK		 125};
126
127struct kvm_shared_msrs {
128 struct user_return_notifier urn;
129 bool registered;
2bf78fa7
SY		 130 struct kvm_shared_msr_values {
131 u64 host;
132 u64 curr;
133 } values[KVM_NR_SHARED_MSRS];
18863bdd
AK		 134};
135
136static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
013f6a5d 137static struct kvm_shared_msrs __percpu *shared_msrs;
18863bdd 138
417bc304 139struct kvm_stats_debugfs_item debugfs_entries[] = {
ba1389b7
AK		 140 { "pf_fixed", VCPU_STAT(pf_fixed) },
141 { "pf_guest", VCPU_STAT(pf_guest) },
142 { "tlb_flush", VCPU_STAT(tlb_flush) },
143 { "invlpg", VCPU_STAT(invlpg) },
144 { "exits", VCPU_STAT(exits) },
145 { "io_exits", VCPU_STAT(io_exits) },
146 { "mmio_exits", VCPU_STAT(mmio_exits) },
147 { "signal_exits", VCPU_STAT(signal_exits) },
148 { "irq_window", VCPU_STAT(irq_window_exits) },
f08864b4 149 { "nmi_window", VCPU_STAT(nmi_window_exits) },
ba1389b7 150 { "halt_exits", VCPU_STAT(halt_exits) },
f7819512 151 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
ba1389b7 152 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
f11c3a8d 153 { "hypercalls", VCPU_STAT(hypercalls) },
ba1389b7
AK		 154 { "request_irq", VCPU_STAT(request_irq_exits) },
155 { "irq_exits", VCPU_STAT(irq_exits) },
156 { "host_state_reload", VCPU_STAT(host_state_reload) },
157 { "efer_reload", VCPU_STAT(efer_reload) },
158 { "fpu_reload", VCPU_STAT(fpu_reload) },
159 { "insn_emulation", VCPU_STAT(insn_emulation) },
160 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
fa89a817 161 { "irq_injections", VCPU_STAT(irq_injections) },
c4abb7c9 162 { "nmi_injections", VCPU_STAT(nmi_injections) },
4cee5764
AK		 163 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
164 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
165 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
166 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
167 { "mmu_flooded", VM_STAT(mmu_flooded) },
168 { "mmu_recycled", VM_STAT(mmu_recycled) },
dfc5aa00 169 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
4731d4c7 170 { "mmu_unsync", VM_STAT(mmu_unsync) },
0f74a24c 171 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
05da4558 172 { "largepages", VM_STAT(lpages) },
417bc304
HB		 173 { NULL }
174};
175
2acf923e
DC		 176u64 __read_mostly host_xcr0;
177
b6785def 178static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
d6aa1000 179
af585b92
GN		 180static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
181{
182 int i;
183 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
184 vcpu->arch.apf.gfns[i] = ~0;
185}
186
18863bdd
AK		 187static void kvm_on_user_return(struct user_return_notifier *urn)
188{
189 unsigned slot;
18863bdd
AK		 190 struct kvm_shared_msrs *locals
191 = container_of(urn, struct kvm_shared_msrs, urn);
2bf78fa7 192 struct kvm_shared_msr_values *values;
18863bdd
AK		 193
194 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
2bf78fa7
SY		 195 values = &locals->values[slot];
196 if (values->host != values->curr) {
197 wrmsrl(shared_msrs_global.msrs[slot], values->host);
198 values->curr = values->host;
18863bdd
AK		 199 }
200 }
201 locals->registered = false;
202 user_return_notifier_unregister(urn);
203}
204
2bf78fa7 205static void shared_msr_update(unsigned slot, u32 msr)
18863bdd 206{
18863bdd 207 u64 value;
013f6a5d
MT		 208 unsigned int cpu = smp_processor_id();
209 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
18863bdd 210
2bf78fa7
SY		 211 /* only read, and nobody should modify it at this time,
212 * so don't need lock */
213 if (slot >= shared_msrs_global.nr) {
214 printk(KERN_ERR "kvm: invalid MSR slot!");
215 return;
216 }
217 rdmsrl_safe(msr, &value);
218 smsr->values[slot].host = value;
219 smsr->values[slot].curr = value;
220}
221
222void kvm_define_shared_msr(unsigned slot, u32 msr)
223{
0123be42 224 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
18863bdd
AK		 225 if (slot >= shared_msrs_global.nr)
226 shared_msrs_global.nr = slot + 1;
2bf78fa7
SY		 227 shared_msrs_global.msrs[slot] = msr;
228 /* we need ensured the shared_msr_global have been updated */
229 smp_wmb();
18863bdd
AK		 230}
231EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
232
233static void kvm_shared_msr_cpu_online(void)
234{
235 unsigned i;
18863bdd
AK		 236
237 for (i = 0; i < shared_msrs_global.nr; ++i)
2bf78fa7 238 shared_msr_update(i, shared_msrs_global.msrs[i]);
18863bdd
AK		 239}
240
8b3c3104 241int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
18863bdd 242{
013f6a5d
MT		 243 unsigned int cpu = smp_processor_id();
244 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
8b3c3104 245 int err;
18863bdd 246
2bf78fa7 247 if (((value ^ smsr->values[slot].curr) & mask) == 0)
8b3c3104 248 return 0;
2bf78fa7 249 smsr->values[slot].curr = value;
8b3c3104
AH		 250 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
251 if (err)
252 return 1;
253
18863bdd
AK		 254 if (!smsr->registered) {
255 smsr->urn.on_user_return = kvm_on_user_return;
256 user_return_notifier_register(&smsr->urn);
257 smsr->registered = true;
258 }
8b3c3104 259 return 0;
18863bdd
AK		 260}
261EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
262
13a34e06 263static void drop_user_return_notifiers(void)
3548bab5 264{
013f6a5d
MT		 265 unsigned int cpu = smp_processor_id();
266 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
3548bab5
AK		 267
268 if (smsr->registered)
269 kvm_on_user_return(&smsr->urn);
270}
271
6866b83e
CO		 272u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
273{
8a5a87d9 274 return vcpu->arch.apic_base;
6866b83e
CO		 275}
276EXPORT_SYMBOL_GPL(kvm_get_apic_base);
277
58cb628d
JK		 278int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
279{
280 u64 old_state = vcpu->arch.apic_base &
281 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
282 u64 new_state = msr_info->data &
283 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
284 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
285 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
286
287 if (!msr_info->host_initiated &&
288 ((msr_info->data & reserved_bits) != 0 ||
289 new_state == X2APIC_ENABLE ||
290 (new_state == MSR_IA32_APICBASE_ENABLE &&
291 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
292 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
293 old_state == 0)))
294 return 1;
295
296 kvm_lapic_set_base(vcpu, msr_info->data);
297 return 0;
6866b83e
CO		 298}
299EXPORT_SYMBOL_GPL(kvm_set_apic_base);
300
2605fc21 301asmlinkage __visible void kvm_spurious_fault(void)
e3ba45b8
GL		 302{
303 /* Fault while not rebooting. We want the trace. */
304 BUG();
305}
306EXPORT_SYMBOL_GPL(kvm_spurious_fault);
307
3fd28fce
ED		 308#define EXCPT_BENIGN 0
309#define EXCPT_CONTRIBUTORY 1
310#define EXCPT_PF 2
311
312static int exception_class(int vector)
313{
314 switch (vector) {
315 case PF_VECTOR:
316 return EXCPT_PF;
317 case DE_VECTOR:
318 case TS_VECTOR:
319 case NP_VECTOR:
320 case SS_VECTOR:
321 case GP_VECTOR:
322 return EXCPT_CONTRIBUTORY;
323 default:
324 break;
325 }
326 return EXCPT_BENIGN;
327}
328
d6e8c854
NA		 329#define EXCPT_FAULT 0
330#define EXCPT_TRAP 1
331#define EXCPT_ABORT 2
332#define EXCPT_INTERRUPT 3
333
334static int exception_type(int vector)
335{
336 unsigned int mask;
337
338 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
339 return EXCPT_INTERRUPT;
340
341 mask = 1 << vector;
342
343 /* #DB is trap, as instruction watchpoints are handled elsewhere */
344 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
345 return EXCPT_TRAP;
346
347 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
348 return EXCPT_ABORT;
349
350 /* Reserved exceptions will result in fault */
351 return EXCPT_FAULT;
352}
353
3fd28fce 354static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
ce7ddec4
JR		 355 unsigned nr, bool has_error, u32 error_code,
356 bool reinject)
3fd28fce
ED		 357{
358 u32 prev_nr;
359 int class1, class2;
360
3842d135
AK		 361 kvm_make_request(KVM_REQ_EVENT, vcpu);
362
3fd28fce
ED		 363 if (!vcpu->arch.exception.pending) {
364 queue:
3ffb2468
NA		 365 if (has_error && !is_protmode(vcpu))
366 has_error = false;
3fd28fce
ED		 367 vcpu->arch.exception.pending = true;
368 vcpu->arch.exception.has_error_code = has_error;
369 vcpu->arch.exception.nr = nr;
370 vcpu->arch.exception.error_code = error_code;
3f0fd292 371 vcpu->arch.exception.reinject = reinject;
3fd28fce
ED		 372 return;
373 }
374
375 /* to check exception */
376 prev_nr = vcpu->arch.exception.nr;
377 if (prev_nr == DF_VECTOR) {
378 /* triple fault -> shutdown */
a8eeb04a 379 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3fd28fce
ED		 380 return;
381 }
382 class1 = exception_class(prev_nr);
383 class2 = exception_class(nr);
384 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
385 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
386 /* generate double fault per SDM Table 5-5 */
387 vcpu->arch.exception.pending = true;
388 vcpu->arch.exception.has_error_code = true;
389 vcpu->arch.exception.nr = DF_VECTOR;
390 vcpu->arch.exception.error_code = 0;
391 } else
392 /* replace previous exception with a new one in a hope
393 that instruction re-execution will regenerate lost
394 exception */
395 goto queue;
396}
397
298101da
AK		 398void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
399{
ce7ddec4 400 kvm_multiple_exception(vcpu, nr, false, 0, false);
298101da
AK		 401}
402EXPORT_SYMBOL_GPL(kvm_queue_exception);
403
ce7ddec4
JR		 404void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
405{
406 kvm_multiple_exception(vcpu, nr, false, 0, true);
407}
408EXPORT_SYMBOL_GPL(kvm_requeue_exception);
409
db8fcefa 410void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
c3c91fee 411{
db8fcefa
AP		 412 if (err)
413 kvm_inject_gp(vcpu, 0);
414 else
415 kvm_x86_ops->skip_emulated_instruction(vcpu);
416}
417EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
8df25a32 418
6389ee94 419void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
c3c91fee
AK		 420{
421 ++vcpu->stat.pf_guest;
6389ee94
AK		 422 vcpu->arch.cr2 = fault->address;
423 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
c3c91fee 424}
27d6c865 425EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
c3c91fee 426
ef54bcfe 427static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
d4f8cf66 428{
6389ee94
AK		 429 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
430 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
d4f8cf66 431 else
6389ee94 432 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
ef54bcfe
PB		 433
434 return fault->nested_page_fault;
d4f8cf66
JR		 435}
436
3419ffc8
SY		 437void kvm_inject_nmi(struct kvm_vcpu *vcpu)
438{
7460fb4a
AK		 439 atomic_inc(&vcpu->arch.nmi_queued);
440 kvm_make_request(KVM_REQ_NMI, vcpu);
3419ffc8
SY		 441}
442EXPORT_SYMBOL_GPL(kvm_inject_nmi);
443
298101da
AK		 444void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
445{
ce7ddec4 446 kvm_multiple_exception(vcpu, nr, true, error_code, false);
298101da
AK		 447}
448EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
449
ce7ddec4
JR		 450void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
451{
452 kvm_multiple_exception(vcpu, nr, true, error_code, true);
453}
454EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
455
0a79b009
AK		 456/*
457 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
458 * a #GP and return false.
459 */
460bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
298101da 461{
0a79b009
AK		 462 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
463 return true;
464 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
465 return false;
298101da 466}
0a79b009 467EXPORT_SYMBOL_GPL(kvm_require_cpl);
298101da 468
16f8a6f9
NA		 469bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
470{
471 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
472 return true;
473
474 kvm_queue_exception(vcpu, UD_VECTOR);
475 return false;
476}
477EXPORT_SYMBOL_GPL(kvm_require_dr);
478
ec92fe44
JR		 479/*
480 * This function will be used to read from the physical memory of the currently
481 * running guest. The difference to kvm_read_guest_page is that this function
482 * can read from guest physical or from the guest's guest physical memory.
483 */
484int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
485 gfn_t ngfn, void *data, int offset, int len,
486 u32 access)
487{
54987b7a 488 struct x86_exception exception;
ec92fe44
JR		 489 gfn_t real_gfn;
490 gpa_t ngpa;
491
492 ngpa = gfn_to_gpa(ngfn);
54987b7a 493 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
ec92fe44
JR		 494 if (real_gfn == UNMAPPED_GVA)
495 return -EFAULT;
496
497 real_gfn = gpa_to_gfn(real_gfn);
498
499 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
500}
501EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
502
69b0049a 503static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3d06b8bf
JR		 504 void *data, int offset, int len, u32 access)
505{
506 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
507 data, offset, len, access);
508}
509
a03490ed
CO		 510/*
511 * Load the pae pdptrs. Return true is they are all valid.
512 */
ff03a073 513int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
a03490ed
CO		 514{
515 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
516 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
517 int i;
518 int ret;
ff03a073 519 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
a03490ed 520
ff03a073
JR		 521 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
522 offset * sizeof(u64), sizeof(pdpte),
523 PFERR_USER_MASK|PFERR_WRITE_MASK);
a03490ed
CO		 524 if (ret < 0) {
525 ret = 0;
526 goto out;
527 }
528 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
43a3795a 529 if (is_present_gpte(pdpte[i]) &&
20c466b5 530 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
a03490ed
CO		 531 ret = 0;
532 goto out;
533 }
534 }
535 ret = 1;
536
ff03a073 537 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
6de4f3ad
AK		 538 __set_bit(VCPU_EXREG_PDPTR,
539 (unsigned long *)&vcpu->arch.regs_avail);
540 __set_bit(VCPU_EXREG_PDPTR,
541 (unsigned long *)&vcpu->arch.regs_dirty);
a03490ed 542out:
a03490ed
CO		 543
544 return ret;
545}
cc4b6871 546EXPORT_SYMBOL_GPL(load_pdptrs);
a03490ed 547
d835dfec
AK		 548static bool pdptrs_changed(struct kvm_vcpu *vcpu)
549{
ff03a073 550 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
d835dfec 551 bool changed = true;
3d06b8bf
JR		 552 int offset;
553 gfn_t gfn;
d835dfec
AK		 554 int r;
555
556 if (is_long_mode(vcpu) || !is_pae(vcpu))
557 return false;
558
6de4f3ad
AK		 559 if (!test_bit(VCPU_EXREG_PDPTR,
560 (unsigned long *)&vcpu->arch.regs_avail))
561 return true;
562
9f8fe504
AK		 563 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
564 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
3d06b8bf
JR		 565 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
566 PFERR_USER_MASK | PFERR_WRITE_MASK);
d835dfec
AK		 567 if (r < 0)
568 goto out;
ff03a073 569 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
d835dfec 570out:
d835dfec
AK		 571
572 return changed;
573}
574
49a9b07e 575int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
a03490ed 576{
aad82703 577 unsigned long old_cr0 = kvm_read_cr0(vcpu);
d81135a5 578 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
aad82703 579
f9a48e6a
AK		 580 cr0 |= X86_CR0_ET;
581
ab344828 582#ifdef CONFIG_X86_64
0f12244f
GN		 583 if (cr0 & 0xffffffff00000000UL)
584 return 1;
ab344828
GN		 585#endif
586
587 cr0 &= ~CR0_RESERVED_BITS;
a03490ed 588
0f12244f
GN		 589 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
590 return 1;
a03490ed 591
0f12244f
GN		 592 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
593 return 1;
a03490ed
CO		 594
595 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
596#ifdef CONFIG_X86_64
f6801dff 597 if ((vcpu->arch.efer & EFER_LME)) {
a03490ed
CO		 598 int cs_db, cs_l;
599
0f12244f
GN		 600 if (!is_pae(vcpu))
601 return 1;
a03490ed 602 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
0f12244f
GN		 603 if (cs_l)
604 return 1;
a03490ed
CO		 605 } else
606#endif
ff03a073 607 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
9f8fe504 608 kvm_read_cr3(vcpu)))
0f12244f 609 return 1;
a03490ed
CO		 610 }
611
ad756a16
MJ		 612 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
613 return 1;
614
a03490ed 615 kvm_x86_ops->set_cr0(vcpu, cr0);
a03490ed 616
d170c419 617 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
e5f3f027 618 kvm_clear_async_pf_completion_queue(vcpu);
d170c419
LJ		 619 kvm_async_pf_hash_reset(vcpu);
620 }
e5f3f027 621
aad82703
SY		 622 if ((cr0 ^ old_cr0) & update_bits)
623 kvm_mmu_reset_context(vcpu);
0f12244f
GN		 624 return 0;
625}
2d3ad1f4 626EXPORT_SYMBOL_GPL(kvm_set_cr0);
a03490ed 627
2d3ad1f4 628void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
a03490ed 629{
49a9b07e 630 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
a03490ed 631}
2d3ad1f4 632EXPORT_SYMBOL_GPL(kvm_lmsw);
a03490ed 633
42bdf991
MT		 634static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
635{
636 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
637 !vcpu->guest_xcr0_loaded) {
638 /* kvm_set_xcr() also depends on this */
639 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
640 vcpu->guest_xcr0_loaded = 1;
641 }
642}
643
644static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
645{
646 if (vcpu->guest_xcr0_loaded) {
647 if (vcpu->arch.xcr0 != host_xcr0)
648 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
649 vcpu->guest_xcr0_loaded = 0;
650 }
651}
652
69b0049a 653static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
2acf923e 654{
56c103ec
LJ		 655 u64 xcr0 = xcr;
656 u64 old_xcr0 = vcpu->arch.xcr0;
46c34cb0 657 u64 valid_bits;
2acf923e
DC		 658
659 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
660 if (index != XCR_XFEATURE_ENABLED_MASK)
661 return 1;
2acf923e
DC		 662 if (!(xcr0 & XSTATE_FP))
663 return 1;
664 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
665 return 1;
46c34cb0
PB		 666
667 /*
668 * Do not allow the guest to set bits that we do not support
669 * saving. However, xcr0 bit 0 is always set, even if the
670 * emulated CPU does not support XSAVE (see fx_init).
671 */
672 valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
673 if (xcr0 & ~valid_bits)
2acf923e 674 return 1;
46c34cb0 675
390bd528
LJ		 676 if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
677 return 1;
678
612263b3
CP		 679 if (xcr0 & XSTATE_AVX512) {
680 if (!(xcr0 & XSTATE_YMM))
681 return 1;
682 if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512)
683 return 1;
684 }
42bdf991 685 kvm_put_guest_xcr0(vcpu);
2acf923e 686 vcpu->arch.xcr0 = xcr0;
56c103ec
LJ		 687
688 if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
689 kvm_update_cpuid(vcpu);
2acf923e
DC		 690 return 0;
691}
692
693int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
694{
764bcbc5
Z		 695 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
696 __kvm_set_xcr(vcpu, index, xcr)) {
2acf923e
DC		 697 kvm_inject_gp(vcpu, 0);
698 return 1;
699 }
700 return 0;
701}
702EXPORT_SYMBOL_GPL(kvm_set_xcr);
703
a83b29c6 704int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
a03490ed 705{
fc78f519 706 unsigned long old_cr4 = kvm_read_cr4(vcpu);
edc90b7d
XG		 707 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
708 X86_CR4_SMEP | X86_CR4_SMAP;
709
0f12244f
GN		 710 if (cr4 & CR4_RESERVED_BITS)
711 return 1;
a03490ed 712
2acf923e
DC		 713 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
714 return 1;
715
c68b734f
YW		 716 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
717 return 1;
718
97ec8c06
FW		 719 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
720 return 1;
721
afcbf13f 722 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
74dc2b4f
YW		 723 return 1;
724
a03490ed 725 if (is_long_mode(vcpu)) {
0f12244f
GN		 726 if (!(cr4 & X86_CR4_PAE))
727 return 1;
a2edf57f
AK		 728 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
729 && ((cr4 ^ old_cr4) & pdptr_bits)
9f8fe504
AK		 730 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
731 kvm_read_cr3(vcpu)))
0f12244f
GN		 732 return 1;
733
ad756a16
MJ		 734 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
735 if (!guest_cpuid_has_pcid(vcpu))
736 return 1;
737
738 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
739 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
740 return 1;
741 }
742
5e1746d6 743 if (kvm_x86_ops->set_cr4(vcpu, cr4))
0f12244f 744 return 1;
a03490ed 745
ad756a16
MJ		 746 if (((cr4 ^ old_cr4) & pdptr_bits) ||
747 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
aad82703 748 kvm_mmu_reset_context(vcpu);
0f12244f 749
2acf923e 750 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
00b27a3e 751 kvm_update_cpuid(vcpu);
2acf923e 752
0f12244f
GN		 753 return 0;
754}
2d3ad1f4 755EXPORT_SYMBOL_GPL(kvm_set_cr4);
a03490ed 756
2390218b 757int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
a03490ed 758{
ac146235 759#ifdef CONFIG_X86_64
9d88fca7 760 cr3 &= ~CR3_PCID_INVD;
ac146235 761#endif
9d88fca7 762
9f8fe504 763 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
0ba73cda 764 kvm_mmu_sync_roots(vcpu);
77c3913b 765 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
0f12244f 766 return 0;
d835dfec
AK		 767 }
768
a03490ed 769 if (is_long_mode(vcpu)) {
d9f89b88
JK		 770 if (cr3 & CR3_L_MODE_RESERVED_BITS)
771 return 1;
772 } else if (is_pae(vcpu) && is_paging(vcpu) &&
773 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
346874c9 774 return 1;
a03490ed 775
0f12244f 776 vcpu->arch.cr3 = cr3;
aff48baa 777 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
d8d173da 778 kvm_mmu_new_cr3(vcpu);
0f12244f
GN		 779 return 0;
780}
2d3ad1f4 781EXPORT_SYMBOL_GPL(kvm_set_cr3);
a03490ed 782
eea1cff9 783int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
a03490ed 784{
0f12244f
GN		 785 if (cr8 & CR8_RESERVED_BITS)
786 return 1;
a03490ed
CO		 787 if (irqchip_in_kernel(vcpu->kvm))
788 kvm_lapic_set_tpr(vcpu, cr8);
789 else
ad312c7c 790 vcpu->arch.cr8 = cr8;
0f12244f
GN		 791 return 0;
792}
2d3ad1f4 793EXPORT_SYMBOL_GPL(kvm_set_cr8);
a03490ed 794
2d3ad1f4 795unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
a03490ed
CO		 796{
797 if (irqchip_in_kernel(vcpu->kvm))
798 return kvm_lapic_get_cr8(vcpu);
799 else
ad312c7c 800 return vcpu->arch.cr8;
a03490ed 801}
2d3ad1f4 802EXPORT_SYMBOL_GPL(kvm_get_cr8);
a03490ed 803
ae561ede
NA		 804static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
805{
806 int i;
807
808 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
809 for (i = 0; i < KVM_NR_DB_REGS; i++)
810 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
811 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
812 }
813}
814
73aaf249
JK		 815static void kvm_update_dr6(struct kvm_vcpu *vcpu)
816{
817 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
818 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
819}
820
c8639010
JK		 821static void kvm_update_dr7(struct kvm_vcpu *vcpu)
822{
823 unsigned long dr7;
824
825 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
826 dr7 = vcpu->arch.guest_debug_dr7;
827 else
828 dr7 = vcpu->arch.dr7;
829 kvm_x86_ops->set_dr7(vcpu, dr7);
360b948d
PB		 830 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
831 if (dr7 & DR7_BP_EN_MASK)
832 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
c8639010
JK		 833}
834
6f43ed01
NA		 835static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
836{
837 u64 fixed = DR6_FIXED_1;
838
839 if (!guest_cpuid_has_rtm(vcpu))
840 fixed |= DR6_RTM;
841 return fixed;
842}
843
338dbc97 844static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
020df079
GN		 845{
846 switch (dr) {
847 case 0 ... 3:
848 vcpu->arch.db[dr] = val;
849 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
850 vcpu->arch.eff_db[dr] = val;
851 break;
852 case 4:
020df079
GN		 853 /* fall through */
854 case 6:
338dbc97
GN		 855 if (val & 0xffffffff00000000ULL)
856 return -1; /* #GP */
6f43ed01 857 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
73aaf249 858 kvm_update_dr6(vcpu);
020df079
GN		 859 break;
860 case 5:
020df079
GN		 861 /* fall through */
862 default: /* 7 */
338dbc97
GN		 863 if (val & 0xffffffff00000000ULL)
864 return -1; /* #GP */
020df079 865 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
c8639010 866 kvm_update_dr7(vcpu);
020df079
GN		 867 break;
868 }
869
870 return 0;
871}
338dbc97
GN		 872
873int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
874{
16f8a6f9 875 if (__kvm_set_dr(vcpu, dr, val)) {
338dbc97 876 kvm_inject_gp(vcpu, 0);
16f8a6f9
NA		 877 return 1;
878 }
879 return 0;
338dbc97 880}
020df079
GN		 881EXPORT_SYMBOL_GPL(kvm_set_dr);
882
16f8a6f9 883int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
020df079
GN		 884{
885 switch (dr) {
886 case 0 ... 3:
887 *val = vcpu->arch.db[dr];
888 break;
889 case 4:
020df079
GN		 890 /* fall through */
891 case 6:
73aaf249
JK		 892 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
893 *val = vcpu->arch.dr6;
894 else
895 *val = kvm_x86_ops->get_dr6(vcpu);
020df079
GN		 896 break;
897 case 5:
020df079
GN		 898 /* fall through */
899 default: /* 7 */
900 *val = vcpu->arch.dr7;
901 break;
902 }
338dbc97
GN		 903 return 0;
904}
020df079
GN		 905EXPORT_SYMBOL_GPL(kvm_get_dr);
906
022cd0e8
AK		 907bool kvm_rdpmc(struct kvm_vcpu *vcpu)
908{
909 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
910 u64 data;
911 int err;
912
913 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
914 if (err)
915 return err;
916 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
917 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
918 return err;
919}
920EXPORT_SYMBOL_GPL(kvm_rdpmc);
921
043405e1
CO		 922/*
923 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
924 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
925 *
926 * This list is modified at module load time to reflect the
e3267cbb 927 * capabilities of the host cpu. This capabilities test skips MSRs that are
62ef68bb
PB		 928 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
929 * may depend on host virtualization features rather than host cpu features.
043405e1 930 */
e3267cbb 931
043405e1
CO		 932static u32 msrs_to_save[] = {
933 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
8c06585d 934 MSR_STAR,
043405e1
CO		 935#ifdef CONFIG_X86_64
936 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
937#endif
b3897a49 938 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
0dd376e7 939 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
043405e1
CO		 940};
941
942static unsigned num_msrs_to_save;
943
62ef68bb
PB		 944static u32 emulated_msrs[] = {
945 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
946 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
947 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
948 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
949 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
950 MSR_KVM_PV_EOI_EN,
951
ba904635 952 MSR_IA32_TSC_ADJUST,
a3e06bbe 953 MSR_IA32_TSCDEADLINE,
043405e1 954 MSR_IA32_MISC_ENABLE,
908e75f3
AK		 955 MSR_IA32_MCG_STATUS,
956 MSR_IA32_MCG_CTL,
64d60670 957 MSR_IA32_SMBASE,
043405e1
CO		 958};
959
62ef68bb
PB		 960static unsigned num_emulated_msrs;
961
384bb783 962bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
15c4a640 963{
b69e8cae 964 if (efer & efer_reserved_bits)
384bb783 965 return false;
15c4a640 966
1b2fd70c
AG		 967 if (efer & EFER_FFXSR) {
968 struct kvm_cpuid_entry2 *feat;
969
970 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
b69e8cae 971 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
384bb783 972 return false;
1b2fd70c
AG		 973 }
974
d8017474
AG		 975 if (efer & EFER_SVME) {
976 struct kvm_cpuid_entry2 *feat;
977
978 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
b69e8cae 979 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
384bb783 980 return false;
d8017474
AG		 981 }
982
384bb783
JK		 983 return true;
984}
985EXPORT_SYMBOL_GPL(kvm_valid_efer);
986
987static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
988{
989 u64 old_efer = vcpu->arch.efer;
990
991 if (!kvm_valid_efer(vcpu, efer))
992 return 1;
993
994 if (is_paging(vcpu)
995 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
996 return 1;
997
15c4a640 998 efer &= ~EFER_LMA;
f6801dff 999 efer |= vcpu->arch.efer & EFER_LMA;
15c4a640 1000
a3d204e2
SY		 1001 kvm_x86_ops->set_efer(vcpu, efer);
1002
aad82703
SY		 1003 /* Update reserved bits */
1004 if ((efer ^ old_efer) & EFER_NX)
1005 kvm_mmu_reset_context(vcpu);
1006
b69e8cae 1007 return 0;
15c4a640
CO		 1008}
1009
f2b4b7dd
JR		 1010void kvm_enable_efer_bits(u64 mask)
1011{
1012 efer_reserved_bits &= ~mask;
1013}
1014EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1015
15c4a640
CO		 1016/*
1017 * Writes msr value into into the appropriate "register".
1018 * Returns 0 on success, non-0 otherwise.
1019 * Assumes vcpu_load() was already called.
1020 */
8fe8ab46 1021int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
15c4a640 1022{
854e8bb1
NA		 1023 switch (msr->index) {
1024 case MSR_FS_BASE:
1025 case MSR_GS_BASE:
1026 case MSR_KERNEL_GS_BASE:
1027 case MSR_CSTAR:
1028 case MSR_LSTAR:
1029 if (is_noncanonical_address(msr->data))
1030 return 1;
1031 break;
1032 case MSR_IA32_SYSENTER_EIP:
1033 case MSR_IA32_SYSENTER_ESP:
1034 /*
1035 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1036 * non-canonical address is written on Intel but not on
1037 * AMD (which ignores the top 32-bits, because it does
1038 * not implement 64-bit SYSENTER).
1039 *
1040 * 64-bit code should hence be able to write a non-canonical
1041 * value on AMD. Making the address canonical ensures that
1042 * vmentry does not fail on Intel after writing a non-canonical
1043 * value, and that something deterministic happens if the guest
1044 * invokes 64-bit SYSENTER.
1045 */
1046 msr->data = get_canonical(msr->data);
1047 }
8fe8ab46 1048 return kvm_x86_ops->set_msr(vcpu, msr);
15c4a640 1049}
854e8bb1 1050EXPORT_SYMBOL_GPL(kvm_set_msr);
15c4a640 1051
313a3dc7
CO		 1052/*
1053 * Adapt set_msr() to msr_io()'s calling convention
1054 */
609e36d3
PB		 1055static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1056{
1057 struct msr_data msr;
1058 int r;
1059
1060 msr.index = index;
1061 msr.host_initiated = true;
1062 r = kvm_get_msr(vcpu, &msr);
1063 if (r)
1064 return r;
1065
1066 *data = msr.data;
1067 return 0;
1068}
1069
313a3dc7
CO		 1070static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1071{
8fe8ab46
WA		 1072 struct msr_data msr;
1073
1074 msr.data = *data;
1075 msr.index = index;
1076 msr.host_initiated = true;
1077 return kvm_set_msr(vcpu, &msr);
313a3dc7
CO		 1078}
1079
16e8d74d
MT		 1080#ifdef CONFIG_X86_64
1081struct pvclock_gtod_data {
1082 seqcount_t seq;
1083
1084 struct { /* extract of a clocksource struct */
1085 int vclock_mode;
1086 cycle_t cycle_last;
1087 cycle_t mask;
1088 u32 mult;
1089 u32 shift;
1090 } clock;
1091
cbcf2dd3
TG		 1092 u64 boot_ns;
1093 u64 nsec_base;
16e8d74d
MT		 1094};
1095
1096static struct pvclock_gtod_data pvclock_gtod_data;
1097
1098static void update_pvclock_gtod(struct timekeeper *tk)
1099{
1100 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
cbcf2dd3
TG		 1101 u64 boot_ns;
1102
876e7881 1103 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
16e8d74d
MT		 1104
1105 write_seqcount_begin(&vdata->seq);
1106
1107 /* copy pvclock gtod data */
876e7881
PZ		 1108 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1109 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1110 vdata->clock.mask = tk->tkr_mono.mask;
1111 vdata->clock.mult = tk->tkr_mono.mult;
1112 vdata->clock.shift = tk->tkr_mono.shift;
16e8d74d 1113
cbcf2dd3 1114 vdata->boot_ns = boot_ns;
876e7881 1115 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
16e8d74d
MT		 1116
1117 write_seqcount_end(&vdata->seq);
1118}
1119#endif
1120
bab5bb39
NK		 1121void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1122{
1123 /*
1124 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1125 * vcpu_enter_guest. This function is only called from
1126 * the physical CPU that is running vcpu.
1127 */
1128 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1129}
16e8d74d 1130
18068523
GOC		 1131static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1132{
9ed3c444
AK		 1133 int version;
1134 int r;
50d0a0f9 1135 struct pvclock_wall_clock wc;
923de3cf 1136 struct timespec boot;
18068523
GOC		 1137
1138 if (!wall_clock)
1139 return;
1140
9ed3c444
AK		 1141 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1142 if (r)
1143 return;
1144
1145 if (version & 1)
1146 ++version; /* first time write, random junk */
1147
1148 ++version;
18068523 1149
18068523
GOC		 1150 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1151
50d0a0f9
GH		 1152 /*
1153 * The guest calculates current wall clock time by adding
34c238a1 1154 * system time (updated by kvm_guest_time_update below) to the
50d0a0f9
GH		 1155 * wall clock specified here. guest system time equals host
1156 * system time for us, thus we must fill in host boot time here.
1157 */
923de3cf 1158 getboottime(&boot);
50d0a0f9 1159
4b648665
BR		 1160 if (kvm->arch.kvmclock_offset) {
1161 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1162 boot = timespec_sub(boot, ts);
1163 }
50d0a0f9
GH		 1164 wc.sec = boot.tv_sec;
1165 wc.nsec = boot.tv_nsec;
1166 wc.version = version;
18068523
GOC		 1167
1168 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1169
1170 version++;
1171 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
18068523
GOC		 1172}
1173
50d0a0f9
GH		 1174static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1175{
1176 uint32_t quotient, remainder;
1177
1178 /* Don't try to replace with do_div(), this one calculates
1179 * "(dividend << 32) / divisor" */
1180 __asm__ ( "divl %4"
1181 : "=a" (quotient), "=d" (remainder)
1182 : "0" (0), "1" (dividend), "r" (divisor) );
1183 return quotient;
1184}
1185
5f4e3f88
ZA		 1186static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1187 s8 *pshift, u32 *pmultiplier)
50d0a0f9 1188{
5f4e3f88 1189 uint64_t scaled64;
50d0a0f9
GH		 1190 int32_t shift = 0;
1191 uint64_t tps64;
1192 uint32_t tps32;
1193
5f4e3f88
ZA		 1194 tps64 = base_khz * 1000LL;
1195 scaled64 = scaled_khz * 1000LL;
50933623 1196 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
50d0a0f9
GH		 1197 tps64 >>= 1;
1198 shift--;
1199 }
1200
1201 tps32 = (uint32_t)tps64;
50933623
JK		 1202 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1203 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
5f4e3f88
ZA		 1204 scaled64 >>= 1;
1205 else
1206 tps32 <<= 1;
50d0a0f9
GH		 1207 shift++;
1208 }
1209
5f4e3f88
ZA		 1210 *pshift = shift;
1211 *pmultiplier = div_frac(scaled64, tps32);
50d0a0f9 1212
5f4e3f88
ZA		 1213 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1214 __func__, base_khz, scaled_khz, shift, *pmultiplier);
50d0a0f9
GH		 1215}
1216
759379dd
ZA		 1217static inline u64 get_kernel_ns(void)
1218{
bb0b5812 1219 return ktime_get_boot_ns();
50d0a0f9
GH		 1220}
1221
d828199e 1222#ifdef CONFIG_X86_64
16e8d74d 1223static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
d828199e 1224#endif
16e8d74d 1225
c8076604 1226static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
69b0049a 1227static unsigned long max_tsc_khz;
c8076604 1228
cc578287 1229static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
8cfdc000 1230{
cc578287
ZA		 1231 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1232 vcpu->arch.virtual_tsc_shift);
8cfdc000
ZA		 1233}
1234
cc578287 1235static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1e993611 1236{
cc578287
ZA		 1237 u64 v = (u64)khz * (1000000 + ppm);
1238 do_div(v, 1000000);
1239 return v;
1e993611
JR		 1240}
1241
cc578287 1242static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
759379dd 1243{
cc578287
ZA		 1244 u32 thresh_lo, thresh_hi;
1245 int use_scaling = 0;
217fc9cf 1246
03ba32ca
MT		 1247 /* tsc_khz can be zero if TSC calibration fails */
1248 if (this_tsc_khz == 0)
1249 return;
1250
c285545f
ZA		 1251 /* Compute a scale to convert nanoseconds in TSC cycles */
1252 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
cc578287
ZA		 1253 &vcpu->arch.virtual_tsc_shift,
1254 &vcpu->arch.virtual_tsc_mult);
1255 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1256
1257 /*
1258 * Compute the variation in TSC rate which is acceptable
1259 * within the range of tolerance and decide if the
1260 * rate being applied is within that bounds of the hardware
1261 * rate. If so, no scaling or compensation need be done.
1262 */
1263 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1264 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1265 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1266 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1267 use_scaling = 1;
1268 }
1269 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
c285545f
ZA		 1270}
1271
1272static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1273{
e26101b1 1274 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
cc578287
ZA		 1275 vcpu->arch.virtual_tsc_mult,
1276 vcpu->arch.virtual_tsc_shift);
e26101b1 1277 tsc += vcpu->arch.this_tsc_write;
c285545f
ZA		 1278 return tsc;
1279}
1280
69b0049a 1281static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
b48aa97e
MT		 1282{
1283#ifdef CONFIG_X86_64
1284 bool vcpus_matched;
b48aa97e
MT		 1285 struct kvm_arch *ka = &vcpu->kvm->arch;
1286 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1287
1288 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1289 atomic_read(&vcpu->kvm->online_vcpus));
1290
7f187922
MT		 1291 /*
1292 * Once the masterclock is enabled, always perform request in
1293 * order to update it.
1294 *
1295 * In order to enable masterclock, the host clocksource must be TSC
1296 * and the vcpus need to have matched TSCs. When that happens,
1297 * perform request to enable masterclock.
1298 */
1299 if (ka->use_master_clock ||
1300 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
b48aa97e
MT		 1301 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1302
1303 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1304 atomic_read(&vcpu->kvm->online_vcpus),
1305 ka->use_master_clock, gtod->clock.vclock_mode);
1306#endif
1307}
1308
ba904635
WA		 1309static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1310{
1311 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1312 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1313}
1314
8fe8ab46 1315void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
99e3e30a
ZA		 1316{
1317 struct kvm *kvm = vcpu->kvm;
f38e098f 1318 u64 offset, ns, elapsed;
99e3e30a 1319 unsigned long flags;
02626b6a 1320 s64 usdiff;
b48aa97e 1321 bool matched;
0d3da0d2 1322 bool already_matched;
8fe8ab46 1323 u64 data = msr->data;
99e3e30a 1324
038f8c11 1325 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
857e4099 1326 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
759379dd 1327 ns = get_kernel_ns();
f38e098f 1328 elapsed = ns - kvm->arch.last_tsc_nsec;
5d3cb0f6 1329
03ba32ca 1330 if (vcpu->arch.virtual_tsc_khz) {
8915aa27
MT		 1331 int faulted = 0;
1332
03ba32ca
MT		 1333 /* n.b - signed multiplication and division required */
1334 usdiff = data - kvm->arch.last_tsc_write;
5d3cb0f6 1335#ifdef CONFIG_X86_64
03ba32ca 1336 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
5d3cb0f6 1337#else
03ba32ca 1338 /* do_div() only does unsigned */
8915aa27
MT		 1339 asm("1: idivl %[divisor]\n"
1340 "2: xor %%edx, %%edx\n"
1341 " movl $0, %[faulted]\n"
1342 "3:\n"
1343 ".section .fixup,\"ax\"\n"
1344 "4: movl $1, %[faulted]\n"
1345 " jmp 3b\n"
1346 ".previous\n"
1347
1348 _ASM_EXTABLE(1b, 4b)
1349
1350 : "=A"(usdiff), [faulted] "=r" (faulted)
1351 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1352
5d3cb0f6 1353#endif
03ba32ca
MT		 1354 do_div(elapsed, 1000);
1355 usdiff -= elapsed;
1356 if (usdiff < 0)
1357 usdiff = -usdiff;
8915aa27
MT		 1358
1359 /* idivl overflow => difference is larger than USEC_PER_SEC */
1360 if (faulted)
1361 usdiff = USEC_PER_SEC;
03ba32ca
MT		 1362 } else
1363 usdiff = USEC_PER_SEC; /* disable TSC match window below */
f38e098f
ZA		 1364
1365 /*
5d3cb0f6
ZA		 1366 * Special case: TSC write with a small delta (1 second) of virtual
1367 * cycle time against real time is interpreted as an attempt to
1368 * synchronize the CPU.
1369 *
1370 * For a reliable TSC, we can match TSC offsets, and for an unstable
1371 * TSC, we add elapsed time in this computation. We could let the
1372 * compensation code attempt to catch up if we fall behind, but
1373 * it's better to try to match offsets from the beginning.
1374 */
02626b6a 1375 if (usdiff < USEC_PER_SEC &&
5d3cb0f6 1376 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
f38e098f 1377 if (!check_tsc_unstable()) {
e26101b1 1378 offset = kvm->arch.cur_tsc_offset;
f38e098f
ZA		 1379 pr_debug("kvm: matched tsc offset for %llu\n", data);
1380 } else {
857e4099 1381 u64 delta = nsec_to_cycles(vcpu, elapsed);
5d3cb0f6
ZA		 1382 data += delta;
1383 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
759379dd 1384 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
f38e098f 1385 }
b48aa97e 1386 matched = true;
0d3da0d2 1387 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
e26101b1
ZA		 1388 } else {
1389 /*
1390 * We split periods of matched TSC writes into generations.
1391 * For each generation, we track the original measured
1392 * nanosecond time, offset, and write, so if TSCs are in
1393 * sync, we can match exact offset, and if not, we can match
4a969980 1394 * exact software computation in compute_guest_tsc()
e26101b1
ZA		 1395 *
1396 * These values are tracked in kvm->arch.cur_xxx variables.
1397 */
1398 kvm->arch.cur_tsc_generation++;
1399 kvm->arch.cur_tsc_nsec = ns;
1400 kvm->arch.cur_tsc_write = data;
1401 kvm->arch.cur_tsc_offset = offset;
b48aa97e 1402 matched = false;
0d3da0d2 1403 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
e26101b1 1404 kvm->arch.cur_tsc_generation, data);
f38e098f 1405 }
e26101b1
ZA		 1406
1407 /*
1408 * We also track th most recent recorded KHZ, write and time to
1409 * allow the matching interval to be extended at each write.
1410 */
f38e098f
ZA		 1411 kvm->arch.last_tsc_nsec = ns;
1412 kvm->arch.last_tsc_write = data;
5d3cb0f6 1413 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
99e3e30a 1414
b183aa58 1415 vcpu->arch.last_guest_tsc = data;
e26101b1
ZA		 1416
1417 /* Keep track of which generation this VCPU has synchronized to */
1418 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1419 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1420 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1421
ba904635
WA		 1422 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1423 update_ia32_tsc_adjust_msr(vcpu, offset);
e26101b1
ZA		 1424 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1425 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
b48aa97e
MT		 1426
1427 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
0d3da0d2 1428 if (!matched) {
b48aa97e 1429 kvm->arch.nr_vcpus_matched_tsc = 0;
0d3da0d2
TG		 1430 } else if (!already_matched) {
1431 kvm->arch.nr_vcpus_matched_tsc++;
1432 }
b48aa97e
MT		 1433
1434 kvm_track_tsc_matching(vcpu);
1435 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
99e3e30a 1436}
e26101b1 1437
99e3e30a
ZA		 1438EXPORT_SYMBOL_GPL(kvm_write_tsc);
1439
d828199e
MT		 1440#ifdef CONFIG_X86_64
1441
1442static cycle_t read_tsc(void)
1443{
1444 cycle_t ret;
1445 u64 last;
1446
1447 /*
1448 * Empirically, a fence (of type that depends on the CPU)
1449 * before rdtsc is enough to ensure that rdtsc is ordered
1450 * with respect to loads. The various CPU manuals are unclear
1451 * as to whether rdtsc can be reordered with later loads,
1452 * but no one has ever seen it happen.
1453 */
1454 rdtsc_barrier();
1455 ret = (cycle_t)vget_cycles();
1456
1457 last = pvclock_gtod_data.clock.cycle_last;
1458
1459 if (likely(ret >= last))
1460 return ret;
1461
1462 /*
1463 * GCC likes to generate cmov here, but this branch is extremely
1464 * predictable (it's just a funciton of time and the likely is
1465 * very likely) and there's a data dependence, so force GCC
1466 * to generate a branch instead. I don't barrier() because
1467 * we don't actually need a barrier, and if this function
1468 * ever gets inlined it will generate worse code.
1469 */
1470 asm volatile ("");
1471 return last;
1472}
1473
1474static inline u64 vgettsc(cycle_t *cycle_now)
1475{
1476 long v;
1477 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1478
1479 *cycle_now = read_tsc();
1480
1481 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1482 return v * gtod->clock.mult;
1483}
1484
cbcf2dd3 1485static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
d828199e 1486{
cbcf2dd3 1487 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
d828199e 1488 unsigned long seq;
d828199e 1489 int mode;
cbcf2dd3 1490 u64 ns;
d828199e 1491
d828199e
MT		 1492 do {
1493 seq = read_seqcount_begin(&gtod->seq);
1494 mode = gtod->clock.vclock_mode;
cbcf2dd3 1495 ns = gtod->nsec_base;
d828199e
MT		 1496 ns += vgettsc(cycle_now);
1497 ns >>= gtod->clock.shift;
cbcf2dd3 1498 ns += gtod->boot_ns;
d828199e 1499 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
cbcf2dd3 1500 *t = ns;
d828199e
MT		 1501
1502 return mode;
1503}
1504
1505/* returns true if host is using tsc clocksource */
1506static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1507{
d828199e
MT		 1508 /* checked again under seqlock below */
1509 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1510 return false;
1511
cbcf2dd3 1512 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
d828199e
MT		 1513}
1514#endif
1515
1516/*
1517 *
b48aa97e
MT		 1518 * Assuming a stable TSC across physical CPUS, and a stable TSC
1519 * across virtual CPUs, the following condition is possible.
1520 * Each numbered line represents an event visible to both
d828199e
MT		 1521 * CPUs at the next numbered event.
1522 *
1523 * "timespecX" represents host monotonic time. "tscX" represents
1524 * RDTSC value.
1525 *
1526 * VCPU0 on CPU0 | VCPU1 on CPU1
1527 *
1528 * 1. read timespec0,tsc0
1529 * 2. | timespec1 = timespec0 + N
1530 * | tsc1 = tsc0 + M
1531 * 3. transition to guest | transition to guest
1532 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1533 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1534 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1535 *
1536 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1537 *
1538 * - ret0 < ret1
1539 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1540 * ...
1541 * - 0 < N - M => M < N
1542 *
1543 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1544 * always the case (the difference between two distinct xtime instances
1545 * might be smaller then the difference between corresponding TSC reads,
1546 * when updating guest vcpus pvclock areas).
1547 *
1548 * To avoid that problem, do not allow visibility of distinct
1549 * system_timestamp/tsc_timestamp values simultaneously: use a master
1550 * copy of host monotonic time values. Update that master copy
1551 * in lockstep.
1552 *
b48aa97e 1553 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
d828199e
MT		 1554 *
1555 */
1556
1557static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1558{
1559#ifdef CONFIG_X86_64
1560 struct kvm_arch *ka = &kvm->arch;
1561 int vclock_mode;
b48aa97e
MT		 1562 bool host_tsc_clocksource, vcpus_matched;
1563
1564 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1565 atomic_read(&kvm->online_vcpus));
d828199e
MT		 1566
1567 /*
1568 * If the host uses TSC clock, then passthrough TSC as stable
1569 * to the guest.
1570 */
b48aa97e 1571 host_tsc_clocksource = kvm_get_time_and_clockread(
d828199e
MT		 1572 &ka->master_kernel_ns,
1573 &ka->master_cycle_now);
1574
16a96021 1575 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
54750f2c
MT		 1576 && !backwards_tsc_observed
1577 && !ka->boot_vcpu_runs_old_kvmclock;
b48aa97e 1578
d828199e
MT		 1579 if (ka->use_master_clock)
1580 atomic_set(&kvm_guest_has_master_clock, 1);
1581
1582 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
b48aa97e
MT		 1583 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1584 vcpus_matched);
d828199e
MT		 1585#endif
1586}
1587
2e762ff7
MT		 1588static void kvm_gen_update_masterclock(struct kvm *kvm)
1589{
1590#ifdef CONFIG_X86_64
1591 int i;
1592 struct kvm_vcpu *vcpu;
1593 struct kvm_arch *ka = &kvm->arch;
1594
1595 spin_lock(&ka->pvclock_gtod_sync_lock);
1596 kvm_make_mclock_inprogress_request(kvm);
1597 /* no guest entries from this point */
1598 pvclock_update_vm_gtod_copy(kvm);
1599
1600 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 1601 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2e762ff7
MT		 1602
1603 /* guest entries allowed */
1604 kvm_for_each_vcpu(i, vcpu, kvm)
1605 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1606
1607 spin_unlock(&ka->pvclock_gtod_sync_lock);
1608#endif
1609}
1610
34c238a1 1611static int kvm_guest_time_update(struct kvm_vcpu *v)
18068523 1612{
d828199e 1613 unsigned long flags, this_tsc_khz;
18068523 1614 struct kvm_vcpu_arch *vcpu = &v->arch;
d828199e 1615 struct kvm_arch *ka = &v->kvm->arch;
f25e656d 1616 s64 kernel_ns;
d828199e 1617 u64 tsc_timestamp, host_tsc;
0b79459b 1618 struct pvclock_vcpu_time_info guest_hv_clock;
51d59c6b 1619 u8 pvclock_flags;
d828199e
MT		 1620 bool use_master_clock;
1621
1622 kernel_ns = 0;
1623 host_tsc = 0;
18068523 1624
d828199e
MT		 1625 /*
1626 * If the host uses TSC clock, then passthrough TSC as stable
1627 * to the guest.
1628 */
1629 spin_lock(&ka->pvclock_gtod_sync_lock);
1630 use_master_clock = ka->use_master_clock;
1631 if (use_master_clock) {
1632 host_tsc = ka->master_cycle_now;
1633 kernel_ns = ka->master_kernel_ns;
1634 }
1635 spin_unlock(&ka->pvclock_gtod_sync_lock);
c09664bb
MT		 1636
1637 /* Keep irq disabled to prevent changes to the clock */
1638 local_irq_save(flags);
89cbc767 1639 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
c09664bb
MT		 1640 if (unlikely(this_tsc_khz == 0)) {
1641 local_irq_restore(flags);
1642 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1643 return 1;
1644 }
d828199e
MT		 1645 if (!use_master_clock) {
1646 host_tsc = native_read_tsc();
1647 kernel_ns = get_kernel_ns();
1648 }
1649
1650 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1651
c285545f
ZA		 1652 /*
1653 * We may have to catch up the TSC to match elapsed wall clock
1654 * time for two reasons, even if kvmclock is used.
1655 * 1) CPU could have been running below the maximum TSC rate
1656 * 2) Broken TSC compensation resets the base at each VCPU
1657 * entry to avoid unknown leaps of TSC even when running
1658 * again on the same CPU. This may cause apparent elapsed
1659 * time to disappear, and the guest to stand still or run
1660 * very slowly.
1661 */
1662 if (vcpu->tsc_catchup) {
1663 u64 tsc = compute_guest_tsc(v, kernel_ns);
1664 if (tsc > tsc_timestamp) {
f1e2b260 1665 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
c285545f
ZA		 1666 tsc_timestamp = tsc;
1667 }
50d0a0f9
GH		 1668 }
1669
18068523
GOC		 1670 local_irq_restore(flags);
1671
0b79459b 1672 if (!vcpu->pv_time_enabled)
c285545f 1673 return 0;
18068523 1674
e48672fa 1675 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
5f4e3f88
ZA		 1676 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1677 &vcpu->hv_clock.tsc_shift,
1678 &vcpu->hv_clock.tsc_to_system_mul);
e48672fa 1679 vcpu->hw_tsc_khz = this_tsc_khz;
8cfdc000
ZA		 1680 }
1681
1682 /* With all the info we got, fill in the values */
1d5f066e 1683 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
759379dd 1684 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
28e4639a 1685 vcpu->last_guest_tsc = tsc_timestamp;
51d59c6b 1686
09a0c3f1
OH		 1687 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1688 &guest_hv_clock, sizeof(guest_hv_clock))))
1689 return 0;
1690
5dca0d91
RK		 1691 /* This VCPU is paused, but it's legal for a guest to read another
1692 * VCPU's kvmclock, so we really have to follow the specification where
1693 * it says that version is odd if data is being modified, and even after
1694 * it is consistent.
1695 *
1696 * Version field updates must be kept separate. This is because
1697 * kvm_write_guest_cached might use a "rep movs" instruction, and
1698 * writes within a string instruction are weakly ordered. So there
1699 * are three writes overall.
1700 *
1701 * As a small optimization, only write the version field in the first
1702 * and third write. The vcpu->pv_time cache is still valid, because the
1703 * version field is the first in the struct.
18068523 1704 */
5dca0d91
RK		 1705 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1706
1707 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1708 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1709 &vcpu->hv_clock,
1710 sizeof(vcpu->hv_clock.version));
1711
1712 smp_wmb();
78c0337a
MT		 1713
1714 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
0b79459b 1715 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
78c0337a
MT		 1716
1717 if (vcpu->pvclock_set_guest_stopped_request) {
1718 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1719 vcpu->pvclock_set_guest_stopped_request = false;
1720 }
1721
b7e60c5a
MT		 1722 pvclock_flags |= PVCLOCK_COUNTS_FROM_ZERO;
1723
d828199e
MT		 1724 /* If the host uses TSC clocksource, then it is stable */
1725 if (use_master_clock)
1726 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1727
78c0337a
MT		 1728 vcpu->hv_clock.flags = pvclock_flags;
1729
ce1a5e60
DM		 1730 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1731
0b79459b
AH		 1732 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1733 &vcpu->hv_clock,
1734 sizeof(vcpu->hv_clock));
5dca0d91
RK		 1735
1736 smp_wmb();
1737
1738 vcpu->hv_clock.version++;
1739 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1740 &vcpu->hv_clock,
1741 sizeof(vcpu->hv_clock.version));
8cfdc000 1742 return 0;
c8076604
GH		 1743}
1744
0061d53d
MT		 1745/*
1746 * kvmclock updates which are isolated to a given vcpu, such as
1747 * vcpu->cpu migration, should not allow system_timestamp from
1748 * the rest of the vcpus to remain static. Otherwise ntp frequency
1749 * correction applies to one vcpu's system_timestamp but not
1750 * the others.
1751 *
1752 * So in those cases, request a kvmclock update for all vcpus.
7e44e449
AJ		 1753 * We need to rate-limit these requests though, as they can
1754 * considerably slow guests that have a large number of vcpus.
1755 * The time for a remote vcpu to update its kvmclock is bound
1756 * by the delay we use to rate-limit the updates.
0061d53d
MT		 1757 */
1758
7e44e449
AJ		 1759#define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1760
1761static void kvmclock_update_fn(struct work_struct *work)
0061d53d
MT		 1762{
1763 int i;
7e44e449
AJ		 1764 struct delayed_work *dwork = to_delayed_work(work);
1765 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1766 kvmclock_update_work);
1767 struct kvm *kvm = container_of(ka, struct kvm, arch);
0061d53d
MT		 1768 struct kvm_vcpu *vcpu;
1769
1770 kvm_for_each_vcpu(i, vcpu, kvm) {
105b21bb 1771 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0061d53d
MT		 1772 kvm_vcpu_kick(vcpu);
1773 }
1774}
1775
7e44e449
AJ		 1776static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1777{
1778 struct kvm *kvm = v->kvm;
1779
105b21bb 1780 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
7e44e449
AJ		 1781 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1782 KVMCLOCK_UPDATE_DELAY);
1783}
1784
332967a3
AJ		 1785#define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1786
1787static void kvmclock_sync_fn(struct work_struct *work)
1788{
1789 struct delayed_work *dwork = to_delayed_work(work);
1790 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1791 kvmclock_sync_work);
1792 struct kvm *kvm = container_of(ka, struct kvm, arch);
1793
630994b3
MT		 1794 if (!kvmclock_periodic_sync)
1795 return;
1796
332967a3
AJ		 1797 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1798 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1799 KVMCLOCK_SYNC_PERIOD);
1800}
1801
9ba075a6
AK		 1802static bool msr_mtrr_valid(unsigned msr)
1803{
1804 switch (msr) {
1805 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1806 case MSR_MTRRfix64K_00000:
1807 case MSR_MTRRfix16K_80000:
1808 case MSR_MTRRfix16K_A0000:
1809 case MSR_MTRRfix4K_C0000:
1810 case MSR_MTRRfix4K_C8000:
1811 case MSR_MTRRfix4K_D0000:
1812 case MSR_MTRRfix4K_D8000:
1813 case MSR_MTRRfix4K_E0000:
1814 case MSR_MTRRfix4K_E8000:
1815 case MSR_MTRRfix4K_F0000:
1816 case MSR_MTRRfix4K_F8000:
1817 case MSR_MTRRdefType:
1818 case MSR_IA32_CR_PAT:
1819 return true;
1820 case 0x2f8:
1821 return true;
1822 }
1823 return false;
1824}
1825
d6289b93
MT		 1826static bool valid_pat_type(unsigned t)
1827{
1828 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1829}
1830
1831static bool valid_mtrr_type(unsigned t)
1832{
1833 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1834}
1835
4566654b 1836bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
d6289b93
MT		 1837{
1838 int i;
fd275235 1839 u64 mask;
d6289b93
MT		 1840
1841 if (!msr_mtrr_valid(msr))
1842 return false;
1843
1844 if (msr == MSR_IA32_CR_PAT) {
1845 for (i = 0; i < 8; i++)
1846 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1847 return false;
1848 return true;
1849 } else if (msr == MSR_MTRRdefType) {
1850 if (data & ~0xcff)
1851 return false;
1852 return valid_mtrr_type(data & 0xff);
1853 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1854 for (i = 0; i < 8 ; i++)
1855 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1856 return false;
1857 return true;
1858 }
1859
1860 /* variable MTRRs */
adfb5d27
WL		 1861 WARN_ON(!(msr >= 0x200 && msr < 0x200 + 2 * KVM_NR_VAR_MTRR));
1862
fd275235 1863 mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
d7a2a246 1864 if ((msr & 1) == 0) {
adfb5d27 1865 /* MTRR base */
d7a2a246
WL		 1866 if (!valid_mtrr_type(data & 0xff))
1867 return false;
1868 mask |= 0xf00;
1869 } else
1870 /* MTRR mask */
1871 mask |= 0x7ff;
1872 if (data & mask) {
1873 kvm_inject_gp(vcpu, 0);
1874 return false;
1875 }
1876
adfb5d27 1877 return true;
d6289b93 1878}
4566654b 1879EXPORT_SYMBOL_GPL(kvm_mtrr_valid);
d6289b93 1880
efdfe536
XG		 1881static void update_mtrr(struct kvm_vcpu *vcpu, u32 msr)
1882{
1883 struct mtrr_state_type *mtrr_state = &vcpu->arch.mtrr_state;
1884 unsigned char mtrr_enabled = mtrr_state->enabled;
1885 gfn_t start, end, mask;
1886 int index;
1887 bool is_fixed = true;
1888
1889 if (msr == MSR_IA32_CR_PAT || !tdp_enabled ||
1890 !kvm_arch_has_noncoherent_dma(vcpu->kvm))
1891 return;
1892
1893 if (!(mtrr_enabled & 0x2) && msr != MSR_MTRRdefType)
1894 return;
1895
1896 switch (msr) {
1897 case MSR_MTRRfix64K_00000:
1898 start = 0x0;
1899 end = 0x80000;
1900 break;
1901 case MSR_MTRRfix16K_80000:
1902 start = 0x80000;
1903 end = 0xa0000;
1904 break;
1905 case MSR_MTRRfix16K_A0000:
1906 start = 0xa0000;
1907 end = 0xc0000;
1908 break;
1909 case MSR_MTRRfix4K_C0000 ... MSR_MTRRfix4K_F8000:
1910 index = msr - MSR_MTRRfix4K_C0000;
1911 start = 0xc0000 + index * (32 << 10);
1912 end = start + (32 << 10);
1913 break;
1914 case MSR_MTRRdefType:
1915 is_fixed = false;
1916 start = 0x0;
1917 end = ~0ULL;
1918 break;
1919 default:
1920 /* variable range MTRRs. */
1921 is_fixed = false;
1922 index = (msr - 0x200) / 2;
1923 start = (((u64)mtrr_state->var_ranges[index].base_hi) << 32) +
1924 (mtrr_state->var_ranges[index].base_lo & PAGE_MASK);
1925 mask = (((u64)mtrr_state->var_ranges[index].mask_hi) << 32) +
1926 (mtrr_state->var_ranges[index].mask_lo & PAGE_MASK);
1927 mask |= ~0ULL << cpuid_maxphyaddr(vcpu);
1928
1929 end = ((start & mask) | ~mask) + 1;
1930 }
1931
1932 if (is_fixed && !(mtrr_enabled & 0x1))
1933 return;
1934
1935 kvm_zap_gfn_range(vcpu->kvm, gpa_to_gfn(start), gpa_to_gfn(end));
1936}
1937
9ba075a6
AK		 1938static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1939{
0bed3b56
SY		 1940 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1941
4566654b 1942 if (!kvm_mtrr_valid(vcpu, msr, data))
9ba075a6
AK		 1943 return 1;
1944
0bed3b56
SY		 1945 if (msr == MSR_MTRRdefType) {
1946 vcpu->arch.mtrr_state.def_type = data;
1947 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1948 } else if (msr == MSR_MTRRfix64K_00000)
1949 p[0] = data;
1950 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1951 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1952 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1953 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1954 else if (msr == MSR_IA32_CR_PAT)
1955 vcpu->arch.pat = data;
1956 else { /* Variable MTRRs */
1957 int idx, is_mtrr_mask;
1958 u64 *pt;
1959
1960 idx = (msr - 0x200) / 2;
1961 is_mtrr_mask = msr - 0x200 - 2 * idx;
1962 if (!is_mtrr_mask)
1963 pt =
1964 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1965 else
1966 pt =
1967 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1968 *pt = data;
1969 }
1970
efdfe536 1971 update_mtrr(vcpu, msr);
9ba075a6
AK		 1972 return 0;
1973}
15c4a640 1974
890ca9ae 1975static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
15c4a640 1976{
890ca9ae
HY		 1977 u64 mcg_cap = vcpu->arch.mcg_cap;
1978 unsigned bank_num = mcg_cap & 0xff;
1979
15c4a640 1980 switch (msr) {
15c4a640 1981 case MSR_IA32_MCG_STATUS:
890ca9ae 1982 vcpu->arch.mcg_status = data;
15c4a640 1983 break;
c7ac679c 1984 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 1985 if (!(mcg_cap & MCG_CTL_P))
1986 return 1;
1987 if (data != 0 && data != ~(u64)0)
1988 return -1;
1989 vcpu->arch.mcg_ctl = data;
1990 break;
1991 default:
1992 if (msr >= MSR_IA32_MC0_CTL &&
81760dcc 1993 msr < MSR_IA32_MCx_CTL(bank_num)) {
890ca9ae 1994 u32 offset = msr - MSR_IA32_MC0_CTL;
114be429
AP		 1995 /* only 0 or all 1s can be written to IA32_MCi_CTL
1996 * some Linux kernels though clear bit 10 in bank 4 to
1997 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1998 * this to avoid an uncatched #GP in the guest
1999 */
890ca9ae 2000 if ((offset & 0x3) == 0 &&
114be429 2001 data != 0 && (data | (1 << 10)) != ~(u64)0)
890ca9ae
HY		 2002 return -1;
2003 vcpu->arch.mce_banks[offset] = data;
2004 break;
2005 }
2006 return 1;
2007 }
2008 return 0;
2009}
2010
ffde22ac
ES		 2011static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2012{
2013 struct kvm *kvm = vcpu->kvm;
2014 int lm = is_long_mode(vcpu);
2015 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2016 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2017 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2018 : kvm->arch.xen_hvm_config.blob_size_32;
2019 u32 page_num = data & ~PAGE_MASK;
2020 u64 page_addr = data & PAGE_MASK;
2021 u8 *page;
2022 int r;
2023
2024 r = -E2BIG;
2025 if (page_num >= blob_size)
2026 goto out;
2027 r = -ENOMEM;
ff5c2c03
SL		 2028 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2029 if (IS_ERR(page)) {
2030 r = PTR_ERR(page);
ffde22ac 2031 goto out;
ff5c2c03 2032 }
ffde22ac
ES		 2033 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
2034 goto out_free;
2035 r = 0;
2036out_free:
2037 kfree(page);
2038out:
2039 return r;
2040}
2041
55cd8e5a
GN		 2042static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
2043{
2044 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
2045}
2046
2047static bool kvm_hv_msr_partition_wide(u32 msr)
2048{
2049 bool r = false;
2050 switch (msr) {
2051 case HV_X64_MSR_GUEST_OS_ID:
2052 case HV_X64_MSR_HYPERCALL:
e984097b
VR		 2053 case HV_X64_MSR_REFERENCE_TSC:
2054 case HV_X64_MSR_TIME_REF_COUNT:
55cd8e5a
GN		 2055 r = true;
2056 break;
2057 }
2058
2059 return r;
2060}
2061
2062static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2063{
2064 struct kvm *kvm = vcpu->kvm;
2065
2066 switch (msr) {
2067 case HV_X64_MSR_GUEST_OS_ID:
2068 kvm->arch.hv_guest_os_id = data;
2069 /* setting guest os id to zero disables hypercall page */
2070 if (!kvm->arch.hv_guest_os_id)
2071 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
2072 break;
2073 case HV_X64_MSR_HYPERCALL: {
2074 u64 gfn;
2075 unsigned long addr;
2076 u8 instructions[4];
2077
2078 /* if guest os id is not set hypercall should remain disabled */
2079 if (!kvm->arch.hv_guest_os_id)
2080 break;
2081 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
2082 kvm->arch.hv_hypercall = data;
2083 break;
2084 }
2085 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
2086 addr = gfn_to_hva(kvm, gfn);
2087 if (kvm_is_error_hva(addr))
2088 return 1;
2089 kvm_x86_ops->patch_hypercall(vcpu, instructions);
2090 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
8b0cedff 2091 if (__copy_to_user((void __user *)addr, instructions, 4))
55cd8e5a
GN		 2092 return 1;
2093 kvm->arch.hv_hypercall = data;
b94b64c9 2094 mark_page_dirty(kvm, gfn);
55cd8e5a
GN		 2095 break;
2096 }
e984097b
VR		 2097 case HV_X64_MSR_REFERENCE_TSC: {
2098 u64 gfn;
2099 HV_REFERENCE_TSC_PAGE tsc_ref;
2100 memset(&tsc_ref, 0, sizeof(tsc_ref));
2101 kvm->arch.hv_tsc_page = data;
2102 if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
2103 break;
2104 gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
e1fa108d 2105 if (kvm_write_guest(kvm, gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
e984097b
VR		 2106 &tsc_ref, sizeof(tsc_ref)))
2107 return 1;
2108 mark_page_dirty(kvm, gfn);
2109 break;
2110 }
55cd8e5a 2111 default:
a737f256
CD		 2112 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2113 "data 0x%llx\n", msr, data);
55cd8e5a
GN		 2114 return 1;
2115 }
2116 return 0;
2117}
2118
2119static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2120{
10388a07
GN		 2121 switch (msr) {
2122 case HV_X64_MSR_APIC_ASSIST_PAGE: {
b3af1e88 2123 u64 gfn;
10388a07 2124 unsigned long addr;
55cd8e5a 2125
10388a07
GN		 2126 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
2127 vcpu->arch.hv_vapic = data;
b63cf42f
MT		 2128 if (kvm_lapic_enable_pv_eoi(vcpu, 0))
2129 return 1;
10388a07
GN		 2130 break;
2131 }
b3af1e88
VR		 2132 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
2133 addr = gfn_to_hva(vcpu->kvm, gfn);
10388a07
GN		 2134 if (kvm_is_error_hva(addr))
2135 return 1;
8b0cedff 2136 if (__clear_user((void __user *)addr, PAGE_SIZE))
10388a07
GN		 2137 return 1;
2138 vcpu->arch.hv_vapic = data;
b3af1e88 2139 mark_page_dirty(vcpu->kvm, gfn);
b63cf42f
MT		 2140 if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
2141 return 1;
10388a07
GN		 2142 break;
2143 }
2144 case HV_X64_MSR_EOI:
2145 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
2146 case HV_X64_MSR_ICR:
2147 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
2148 case HV_X64_MSR_TPR:
2149 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
2150 default:
a737f256
CD		 2151 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2152 "data 0x%llx\n", msr, data);
10388a07
GN		 2153 return 1;
2154 }
2155
2156 return 0;
55cd8e5a
GN		 2157}
2158
344d9588
GN		 2159static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2160{
2161 gpa_t gpa = data & ~0x3f;
2162
4a969980 2163 /* Bits 2:5 are reserved, Should be zero */
6adba527 2164 if (data & 0x3c)
344d9588
GN		 2165 return 1;
2166
2167 vcpu->arch.apf.msr_val = data;
2168
2169 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2170 kvm_clear_async_pf_completion_queue(vcpu);
2171 kvm_async_pf_hash_reset(vcpu);
2172 return 0;
2173 }
2174
8f964525
AH		 2175 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2176 sizeof(u32)))
344d9588
GN		 2177 return 1;
2178
6adba527 2179 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
344d9588
GN		 2180 kvm_async_pf_wakeup_all(vcpu);
2181 return 0;
2182}
2183
12f9a48f
GC		 2184static void kvmclock_reset(struct kvm_vcpu *vcpu)
2185{
0b79459b 2186 vcpu->arch.pv_time_enabled = false;
12f9a48f
GC		 2187}
2188
c9aaa895
GC		 2189static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2190{
2191 u64 delta;
2192
2193 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2194 return;
2195
2196 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2197 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2198 vcpu->arch.st.accum_steal = delta;
2199}
2200
2201static void record_steal_time(struct kvm_vcpu *vcpu)
2202{
2203 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2204 return;
2205
2206 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2207 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2208 return;
2209
2210 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2211 vcpu->arch.st.steal.version += 2;
2212 vcpu->arch.st.accum_steal = 0;
2213
2214 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2215 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2216}
2217
8fe8ab46 2218int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
15c4a640 2219{
5753785f 2220 bool pr = false;
8fe8ab46
WA		 2221 u32 msr = msr_info->index;
2222 u64 data = msr_info->data;
5753785f 2223
15c4a640 2224 switch (msr) {
2e32b719
BP		 2225 case MSR_AMD64_NB_CFG:
2226 case MSR_IA32_UCODE_REV:
2227 case MSR_IA32_UCODE_WRITE:
2228 case MSR_VM_HSAVE_PA:
2229 case MSR_AMD64_PATCH_LOADER:
2230 case MSR_AMD64_BU_CFG2:
2231 break;
2232
15c4a640 2233 case MSR_EFER:
b69e8cae 2234 return set_efer(vcpu, data);
8f1589d9
AP		 2235 case MSR_K7_HWCR:
2236 data &= ~(u64)0x40; /* ignore flush filter disable */
82494028 2237 data &= ~(u64)0x100; /* ignore ignne emulation enable */
a223c313 2238 data &= ~(u64)0x8; /* ignore TLB cache disable */
22d48b2d 2239 data &= ~(u64)0x40000; /* ignore Mc status write enable */
8f1589d9 2240 if (data != 0) {
a737f256
CD		 2241 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2242 data);
8f1589d9
AP		 2243 return 1;
2244 }
15c4a640 2245 break;
f7c6d140
AP		 2246 case MSR_FAM10H_MMIO_CONF_BASE:
2247 if (data != 0) {
a737f256
CD		 2248 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2249 "0x%llx\n", data);
f7c6d140
AP		 2250 return 1;
2251 }
15c4a640 2252 break;
b5e2fec0
AG		 2253 case MSR_IA32_DEBUGCTLMSR:
2254 if (!data) {
2255 /* We support the non-activated case already */
2256 break;
2257 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2258 /* Values other than LBR and BTF are vendor-specific,
2259 thus reserved and should throw a #GP */
2260 return 1;
2261 }
a737f256
CD		 2262 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2263 __func__, data);
b5e2fec0 2264 break;
9ba075a6
AK		 2265 case 0x200 ... 0x2ff:
2266 return set_msr_mtrr(vcpu, msr, data);
15c4a640 2267 case MSR_IA32_APICBASE:
58cb628d 2268 return kvm_set_apic_base(vcpu, msr_info);
0105d1a5
GN		 2269 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2270 return kvm_x2apic_msr_write(vcpu, msr, data);
a3e06bbe
LJ		 2271 case MSR_IA32_TSCDEADLINE:
2272 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2273 break;
ba904635
WA		 2274 case MSR_IA32_TSC_ADJUST:
2275 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2276 if (!msr_info->host_initiated) {
d913b904 2277 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
ba904635
WA		 2278 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2279 }
2280 vcpu->arch.ia32_tsc_adjust_msr = data;
2281 }
2282 break;
15c4a640 2283 case MSR_IA32_MISC_ENABLE:
ad312c7c 2284 vcpu->arch.ia32_misc_enable_msr = data;
15c4a640 2285 break;
64d60670
PB		 2286 case MSR_IA32_SMBASE:
2287 if (!msr_info->host_initiated)
2288 return 1;
2289 vcpu->arch.smbase = data;
2290 break;
11c6bffa 2291 case MSR_KVM_WALL_CLOCK_NEW:
18068523
GOC		 2292 case MSR_KVM_WALL_CLOCK:
2293 vcpu->kvm->arch.wall_clock = data;
2294 kvm_write_wall_clock(vcpu->kvm, data);
2295 break;
11c6bffa 2296 case MSR_KVM_SYSTEM_TIME_NEW:
18068523 2297 case MSR_KVM_SYSTEM_TIME: {
0b79459b 2298 u64 gpa_offset;
54750f2c
MT		 2299 struct kvm_arch *ka = &vcpu->kvm->arch;
2300
12f9a48f 2301 kvmclock_reset(vcpu);
18068523 2302
54750f2c
MT		 2303 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2304 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2305
2306 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2307 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2308 &vcpu->requests);
2309
2310 ka->boot_vcpu_runs_old_kvmclock = tmp;
b7e60c5a
MT		 2311
2312 ka->kvmclock_offset = -get_kernel_ns();
54750f2c
MT		 2313 }
2314
18068523 2315 vcpu->arch.time = data;
0061d53d 2316 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
18068523
GOC		 2317
2318 /* we verify if the enable bit is set... */
2319 if (!(data & 1))
2320 break;
2321
0b79459b 2322 gpa_offset = data & ~(PAGE_MASK | 1);
18068523 2323
0b79459b 2324 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
8f964525
AH		 2325 &vcpu->arch.pv_time, data & ~1ULL,
2326 sizeof(struct pvclock_vcpu_time_info)))
0b79459b
AH		 2327 vcpu->arch.pv_time_enabled = false;
2328 else
2329 vcpu->arch.pv_time_enabled = true;
32cad84f 2330
18068523
GOC		 2331 break;
2332 }
344d9588
GN		 2333 case MSR_KVM_ASYNC_PF_EN:
2334 if (kvm_pv_enable_async_pf(vcpu, data))
2335 return 1;
2336 break;
c9aaa895
GC		 2337 case MSR_KVM_STEAL_TIME:
2338
2339 if (unlikely(!sched_info_on()))
2340 return 1;
2341
2342 if (data & KVM_STEAL_RESERVED_MASK)
2343 return 1;
2344
2345 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
8f964525
AH		 2346 data & KVM_STEAL_VALID_BITS,
2347 sizeof(struct kvm_steal_time)))
c9aaa895
GC		 2348 return 1;
2349
2350 vcpu->arch.st.msr_val = data;
2351
2352 if (!(data & KVM_MSR_ENABLED))
2353 break;
2354
2355 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2356
2357 preempt_disable();
2358 accumulate_steal_time(vcpu);
2359 preempt_enable();
2360
2361 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2362
2363 break;
ae7a2a3f
MT		 2364 case MSR_KVM_PV_EOI_EN:
2365 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2366 return 1;
2367 break;
c9aaa895 2368
890ca9ae
HY		 2369 case MSR_IA32_MCG_CTL:
2370 case MSR_IA32_MCG_STATUS:
81760dcc 2371 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
890ca9ae 2372 return set_msr_mce(vcpu, msr, data);
71db6023
AP		 2373
2374 /* Performance counters are not protected by a CPUID bit,
2375 * so we should check all of them in the generic path for the sake of
2376 * cross vendor migration.
2377 * Writing a zero into the event select MSRs disables them,
2378 * which we perfectly emulate ;-). Any other value should be at least
2379 * reported, some guests depend on them.
2380 */
71db6023
AP		 2381 case MSR_K7_EVNTSEL0:
2382 case MSR_K7_EVNTSEL1:
2383 case MSR_K7_EVNTSEL2:
2384 case MSR_K7_EVNTSEL3:
2385 if (data != 0)
a737f256
CD		 2386 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2387 "0x%x data 0x%llx\n", msr, data);
71db6023
AP		 2388 break;
2389 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2390 * so we ignore writes to make it happy.
2391 */
71db6023
AP		 2392 case MSR_K7_PERFCTR0:
2393 case MSR_K7_PERFCTR1:
2394 case MSR_K7_PERFCTR2:
2395 case MSR_K7_PERFCTR3:
a737f256
CD		 2396 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2397 "0x%x data 0x%llx\n", msr, data);
71db6023 2398 break;
5753785f
GN		 2399 case MSR_P6_PERFCTR0:
2400 case MSR_P6_PERFCTR1:
2401 pr = true;
2402 case MSR_P6_EVNTSEL0:
2403 case MSR_P6_EVNTSEL1:
2404 if (kvm_pmu_msr(vcpu, msr))
afd80d85 2405 return kvm_pmu_set_msr(vcpu, msr_info);
5753785f
GN		 2406
2407 if (pr || data != 0)
a737f256
CD		 2408 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2409 "0x%x data 0x%llx\n", msr, data);
5753785f 2410 break;
84e0cefa
JS		 2411 case MSR_K7_CLK_CTL:
2412 /*
2413 * Ignore all writes to this no longer documented MSR.
2414 * Writes are only relevant for old K7 processors,
2415 * all pre-dating SVM, but a recommended workaround from
4a969980 2416 * AMD for these chips. It is possible to specify the
84e0cefa
JS		 2417 * affected processor models on the command line, hence
2418 * the need to ignore the workaround.
2419 */
2420 break;
55cd8e5a
GN		 2421 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2422 if (kvm_hv_msr_partition_wide(msr)) {
2423 int r;
2424 mutex_lock(&vcpu->kvm->lock);
2425 r = set_msr_hyperv_pw(vcpu, msr, data);
2426 mutex_unlock(&vcpu->kvm->lock);
2427 return r;
2428 } else
2429 return set_msr_hyperv(vcpu, msr, data);
2430 break;
91c9c3ed 2431 case MSR_IA32_BBL_CR_CTL3:
2432 /* Drop writes to this legacy MSR -- see rdmsr
2433 * counterpart for further detail.
2434 */
a737f256 2435 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
91c9c3ed 2436 break;
2b036c6b
BO		 2437 case MSR_AMD64_OSVW_ID_LENGTH:
2438 if (!guest_cpuid_has_osvw(vcpu))
2439 return 1;
2440 vcpu->arch.osvw.length = data;
2441 break;
2442 case MSR_AMD64_OSVW_STATUS:
2443 if (!guest_cpuid_has_osvw(vcpu))
2444 return 1;
2445 vcpu->arch.osvw.status = data;
2446 break;
15c4a640 2447 default:
ffde22ac
ES		 2448 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2449 return xen_hvm_config(vcpu, data);
f5132b01 2450 if (kvm_pmu_msr(vcpu, msr))
afd80d85 2451 return kvm_pmu_set_msr(vcpu, msr_info);
ed85c068 2452 if (!ignore_msrs) {
a737f256
CD		 2453 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2454 msr, data);
ed85c068
AP		 2455 return 1;
2456 } else {
a737f256
CD		 2457 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2458 msr, data);
ed85c068
AP		 2459 break;
2460 }
15c4a640
CO		 2461 }
2462 return 0;
2463}
2464EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2465
2466
2467/*
2468 * Reads an msr value (of 'msr_index') into 'pdata'.
2469 * Returns 0 on success, non-0 otherwise.
2470 * Assumes vcpu_load() was already called.
2471 */
609e36d3 2472int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
15c4a640 2473{
609e36d3 2474 return kvm_x86_ops->get_msr(vcpu, msr);
15c4a640 2475}
ff651cb6 2476EXPORT_SYMBOL_GPL(kvm_get_msr);
15c4a640 2477
9ba075a6
AK		 2478static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2479{
0bed3b56
SY		 2480 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2481
9ba075a6
AK		 2482 if (!msr_mtrr_valid(msr))
2483 return 1;
2484
0bed3b56
SY		 2485 if (msr == MSR_MTRRdefType)
2486 *pdata = vcpu->arch.mtrr_state.def_type +
2487 (vcpu->arch.mtrr_state.enabled << 10);
2488 else if (msr == MSR_MTRRfix64K_00000)
2489 *pdata = p[0];
2490 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2491 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2492 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2493 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2494 else if (msr == MSR_IA32_CR_PAT)
2495 *pdata = vcpu->arch.pat;
2496 else { /* Variable MTRRs */
2497 int idx, is_mtrr_mask;
2498 u64 *pt;
2499
2500 idx = (msr - 0x200) / 2;
2501 is_mtrr_mask = msr - 0x200 - 2 * idx;
2502 if (!is_mtrr_mask)
2503 pt =
2504 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2505 else
2506 pt =
2507 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2508 *pdata = *pt;
2509 }
2510
9ba075a6
AK		 2511 return 0;
2512}
2513
890ca9ae 2514static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
15c4a640
CO		 2515{
2516 u64 data;
890ca9ae
HY		 2517 u64 mcg_cap = vcpu->arch.mcg_cap;
2518 unsigned bank_num = mcg_cap & 0xff;
15c4a640
CO		 2519
2520 switch (msr) {
15c4a640
CO		 2521 case MSR_IA32_P5_MC_ADDR:
2522 case MSR_IA32_P5_MC_TYPE:
890ca9ae
HY		 2523 data = 0;
2524 break;
15c4a640 2525 case MSR_IA32_MCG_CAP:
890ca9ae
HY		 2526 data = vcpu->arch.mcg_cap;
2527 break;
c7ac679c 2528 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 2529 if (!(mcg_cap & MCG_CTL_P))
2530 return 1;
2531 data = vcpu->arch.mcg_ctl;
2532 break;
2533 case MSR_IA32_MCG_STATUS:
2534 data = vcpu->arch.mcg_status;
2535 break;
2536 default:
2537 if (msr >= MSR_IA32_MC0_CTL &&
81760dcc 2538 msr < MSR_IA32_MCx_CTL(bank_num)) {
890ca9ae
HY		 2539 u32 offset = msr - MSR_IA32_MC0_CTL;
2540 data = vcpu->arch.mce_banks[offset];
2541 break;
2542 }
2543 return 1;
2544 }
2545 *pdata = data;
2546 return 0;
2547}
2548
55cd8e5a
GN		 2549static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2550{
2551 u64 data = 0;
2552 struct kvm *kvm = vcpu->kvm;
2553
2554 switch (msr) {
2555 case HV_X64_MSR_GUEST_OS_ID:
2556 data = kvm->arch.hv_guest_os_id;
2557 break;
2558 case HV_X64_MSR_HYPERCALL:
2559 data = kvm->arch.hv_hypercall;
2560 break;
e984097b
VR		 2561 case HV_X64_MSR_TIME_REF_COUNT: {
2562 data =
2563 div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2564 break;
2565 }
2566 case HV_X64_MSR_REFERENCE_TSC:
2567 data = kvm->arch.hv_tsc_page;
2568 break;
55cd8e5a 2569 default:
a737f256 2570 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
55cd8e5a
GN		 2571 return 1;
2572 }
2573
2574 *pdata = data;
2575 return 0;
2576}
2577
2578static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2579{
2580 u64 data = 0;
2581
2582 switch (msr) {
2583 case HV_X64_MSR_VP_INDEX: {
2584 int r;
2585 struct kvm_vcpu *v;
684851a1
TY		 2586 kvm_for_each_vcpu(r, v, vcpu->kvm) {
2587 if (v == vcpu) {
55cd8e5a 2588 data = r;
684851a1
TY		 2589 break;
2590 }
2591 }
55cd8e5a
GN		 2592 break;
2593 }
10388a07
GN		 2594 case HV_X64_MSR_EOI:
2595 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2596 case HV_X64_MSR_ICR:
2597 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2598 case HV_X64_MSR_TPR:
2599 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
14fa67ee 2600 case HV_X64_MSR_APIC_ASSIST_PAGE:
d1613ad5
MW		 2601 data = vcpu->arch.hv_vapic;
2602 break;
55cd8e5a 2603 default:
a737f256 2604 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
55cd8e5a
GN		 2605 return 1;
2606 }
2607 *pdata = data;
2608 return 0;
2609}
2610
609e36d3 2611int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
890ca9ae
HY		 2612{
2613 u64 data;
2614
609e36d3 2615 switch (msr_info->index) {
890ca9ae 2616 case MSR_IA32_PLATFORM_ID:
15c4a640 2617 case MSR_IA32_EBL_CR_POWERON:
b5e2fec0
AG		 2618 case MSR_IA32_DEBUGCTLMSR:
2619 case MSR_IA32_LASTBRANCHFROMIP:
2620 case MSR_IA32_LASTBRANCHTOIP:
2621 case MSR_IA32_LASTINTFROMIP:
2622 case MSR_IA32_LASTINTTOIP:
60af2ecd
JSR		 2623 case MSR_K8_SYSCFG:
2624 case MSR_K7_HWCR:
61a6bd67 2625 case MSR_VM_HSAVE_PA:
9e699624 2626 case MSR_K7_EVNTSEL0:
dc9b2d93
WH		 2627 case MSR_K7_EVNTSEL1:
2628 case MSR_K7_EVNTSEL2:
2629 case MSR_K7_EVNTSEL3:
1f3ee616 2630 case MSR_K7_PERFCTR0:
dc9b2d93
WH		 2631 case MSR_K7_PERFCTR1:
2632 case MSR_K7_PERFCTR2:
2633 case MSR_K7_PERFCTR3:
1fdbd48c 2634 case MSR_K8_INT_PENDING_MSG:
c323c0e5 2635 case MSR_AMD64_NB_CFG:
f7c6d140 2636 case MSR_FAM10H_MMIO_CONF_BASE:
2e32b719 2637 case MSR_AMD64_BU_CFG2:
609e36d3 2638 msr_info->data = 0;
15c4a640 2639 break;
5753785f
GN		 2640 case MSR_P6_PERFCTR0:
2641 case MSR_P6_PERFCTR1:
2642 case MSR_P6_EVNTSEL0:
2643 case MSR_P6_EVNTSEL1:
609e36d3
PB		 2644 if (kvm_pmu_msr(vcpu, msr_info->index))
2645 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2646 msr_info->data = 0;
5753785f 2647 break;
742bc670 2648 case MSR_IA32_UCODE_REV:
609e36d3 2649 msr_info->data = 0x100000000ULL;
742bc670 2650 break;
9ba075a6 2651 case MSR_MTRRcap:
609e36d3 2652 msr_info->data = 0x500 | KVM_NR_VAR_MTRR;
9ba075a6
AK		 2653 break;
2654 case 0x200 ... 0x2ff:
609e36d3 2655 return get_msr_mtrr(vcpu, msr_info->index, &msr_info->data);
15c4a640 2656 case 0xcd: /* fsb frequency */
609e36d3 2657 msr_info->data = 3;
15c4a640 2658 break;
7b914098
JS		 2659 /*
2660 * MSR_EBC_FREQUENCY_ID
2661 * Conservative value valid for even the basic CPU models.
2662 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2663 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2664 * and 266MHz for model 3, or 4. Set Core Clock
2665 * Frequency to System Bus Frequency Ratio to 1 (bits
2666 * 31:24) even though these are only valid for CPU
2667 * models > 2, however guests may end up dividing or
2668 * multiplying by zero otherwise.
2669 */
2670 case MSR_EBC_FREQUENCY_ID:
609e36d3 2671 msr_info->data = 1 << 24;
7b914098 2672 break;
15c4a640 2673 case MSR_IA32_APICBASE:
609e36d3 2674 msr_info->data = kvm_get_apic_base(vcpu);
15c4a640 2675 break;
0105d1a5 2676 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
609e36d3 2677 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
0105d1a5 2678 break;
a3e06bbe 2679 case MSR_IA32_TSCDEADLINE:
609e36d3 2680 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
a3e06bbe 2681 break;
ba904635 2682 case MSR_IA32_TSC_ADJUST:
609e36d3 2683 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
ba904635 2684 break;
15c4a640 2685 case MSR_IA32_MISC_ENABLE:
609e36d3 2686 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
15c4a640 2687 break;
64d60670
PB		 2688 case MSR_IA32_SMBASE:
2689 if (!msr_info->host_initiated)
2690 return 1;
2691 msr_info->data = vcpu->arch.smbase;
2692 break;
847f0ad8
AG		 2693 case MSR_IA32_PERF_STATUS:
2694 /* TSC increment by tick */
609e36d3 2695 msr_info->data = 1000ULL;
847f0ad8
AG		 2696 /* CPU multiplier */
2697 data |= (((uint64_t)4ULL) << 40);
2698 break;
15c4a640 2699 case MSR_EFER:
609e36d3 2700 msr_info->data = vcpu->arch.efer;
15c4a640 2701 break;
18068523 2702 case MSR_KVM_WALL_CLOCK:
11c6bffa 2703 case MSR_KVM_WALL_CLOCK_NEW:
609e36d3 2704 msr_info->data = vcpu->kvm->arch.wall_clock;
18068523
GOC		 2705 break;
2706 case MSR_KVM_SYSTEM_TIME:
11c6bffa 2707 case MSR_KVM_SYSTEM_TIME_NEW:
609e36d3 2708 msr_info->data = vcpu->arch.time;
18068523 2709 break;
344d9588 2710 case MSR_KVM_ASYNC_PF_EN:
609e36d3 2711 msr_info->data = vcpu->arch.apf.msr_val;
344d9588 2712 break;
c9aaa895 2713 case MSR_KVM_STEAL_TIME:
609e36d3 2714 msr_info->data = vcpu->arch.st.msr_val;
c9aaa895 2715 break;
1d92128f 2716 case MSR_KVM_PV_EOI_EN:
609e36d3 2717 msr_info->data = vcpu->arch.pv_eoi.msr_val;
1d92128f 2718 break;
890ca9ae
HY		 2719 case MSR_IA32_P5_MC_ADDR:
2720 case MSR_IA32_P5_MC_TYPE:
2721 case MSR_IA32_MCG_CAP:
2722 case MSR_IA32_MCG_CTL:
2723 case MSR_IA32_MCG_STATUS:
81760dcc 2724 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
609e36d3 2725 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
84e0cefa
JS		 2726 case MSR_K7_CLK_CTL:
2727 /*
2728 * Provide expected ramp-up count for K7. All other
2729 * are set to zero, indicating minimum divisors for
2730 * every field.
2731 *
2732 * This prevents guest kernels on AMD host with CPU
2733 * type 6, model 8 and higher from exploding due to
2734 * the rdmsr failing.
2735 */
609e36d3 2736 msr_info->data = 0x20000000;
84e0cefa 2737 break;
55cd8e5a 2738 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
609e36d3 2739 if (kvm_hv_msr_partition_wide(msr_info->index)) {
55cd8e5a
GN		 2740 int r;
2741 mutex_lock(&vcpu->kvm->lock);
609e36d3 2742 r = get_msr_hyperv_pw(vcpu, msr_info->index, &msr_info->data);
55cd8e5a
GN		 2743 mutex_unlock(&vcpu->kvm->lock);
2744 return r;
2745 } else
609e36d3 2746 return get_msr_hyperv(vcpu, msr_info->index, &msr_info->data);
55cd8e5a 2747 break;
91c9c3ed 2748 case MSR_IA32_BBL_CR_CTL3:
2749 /* This legacy MSR exists but isn't fully documented in current
2750 * silicon. It is however accessed by winxp in very narrow
2751 * scenarios where it sets bit #19, itself documented as
2752 * a "reserved" bit. Best effort attempt to source coherent
2753 * read data here should the balance of the register be
2754 * interpreted by the guest:
2755 *
2756 * L2 cache control register 3: 64GB range, 256KB size,
2757 * enabled, latency 0x1, configured
2758 */
609e36d3 2759 msr_info->data = 0xbe702111;
91c9c3ed 2760 break;
2b036c6b
BO		 2761 case MSR_AMD64_OSVW_ID_LENGTH:
2762 if (!guest_cpuid_has_osvw(vcpu))
2763 return 1;
609e36d3 2764 msr_info->data = vcpu->arch.osvw.length;
2b036c6b
BO		 2765 break;
2766 case MSR_AMD64_OSVW_STATUS:
2767 if (!guest_cpuid_has_osvw(vcpu))
2768 return 1;
609e36d3 2769 msr_info->data = vcpu->arch.osvw.status;
2b036c6b 2770 break;
15c4a640 2771 default:
609e36d3
PB		 2772 if (kvm_pmu_msr(vcpu, msr_info->index))
2773 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
ed85c068 2774 if (!ignore_msrs) {
609e36d3 2775 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
ed85c068
AP		 2776 return 1;
2777 } else {
609e36d3
PB		 2778 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2779 msr_info->data = 0;
ed85c068
AP		 2780 }
2781 break;
15c4a640 2782 }
15c4a640
CO		 2783 return 0;
2784}
2785EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2786
313a3dc7
CO		 2787/*
2788 * Read or write a bunch of msrs. All parameters are kernel addresses.
2789 *
2790 * @return number of msrs set successfully.
2791 */
2792static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2793 struct kvm_msr_entry *entries,
2794 int (*do_msr)(struct kvm_vcpu *vcpu,
2795 unsigned index, u64 *data))
2796{
f656ce01 2797 int i, idx;
313a3dc7 2798
f656ce01 2799 idx = srcu_read_lock(&vcpu->kvm->srcu);
313a3dc7
CO		 2800 for (i = 0; i < msrs->nmsrs; ++i)
2801 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2802 break;
f656ce01 2803 srcu_read_unlock(&vcpu->kvm->srcu, idx);
313a3dc7 2804
313a3dc7
CO		 2805 return i;
2806}
2807
2808/*
2809 * Read or write a bunch of msrs. Parameters are user addresses.
2810 *
2811 * @return number of msrs set successfully.
2812 */
2813static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2814 int (*do_msr)(struct kvm_vcpu *vcpu,
2815 unsigned index, u64 *data),
2816 int writeback)
2817{
2818 struct kvm_msrs msrs;
2819 struct kvm_msr_entry *entries;
2820 int r, n;
2821 unsigned size;
2822
2823 r = -EFAULT;
2824 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2825 goto out;
2826
2827 r = -E2BIG;
2828 if (msrs.nmsrs >= MAX_IO_MSRS)
2829 goto out;
2830
313a3dc7 2831 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
ff5c2c03
SL		 2832 entries = memdup_user(user_msrs->entries, size);
2833 if (IS_ERR(entries)) {
2834 r = PTR_ERR(entries);
313a3dc7 2835 goto out;
ff5c2c03 2836 }
313a3dc7
CO		 2837
2838 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2839 if (r < 0)
2840 goto out_free;
2841
2842 r = -EFAULT;
2843 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2844 goto out_free;
2845
2846 r = n;
2847
2848out_free:
7a73c028 2849 kfree(entries);
313a3dc7
CO		 2850out:
2851 return r;
2852}
2853
784aa3d7 2854int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
018d00d2
ZX		 2855{
2856 int r;
2857
2858 switch (ext) {
2859 case KVM_CAP_IRQCHIP:
2860 case KVM_CAP_HLT:
2861 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
018d00d2 2862 case KVM_CAP_SET_TSS_ADDR:
07716717 2863 case KVM_CAP_EXT_CPUID:
9c15bb1d 2864 case KVM_CAP_EXT_EMUL_CPUID:
c8076604 2865 case KVM_CAP_CLOCKSOURCE:
7837699f 2866 case KVM_CAP_PIT:
a28e4f5a 2867 case KVM_CAP_NOP_IO_DELAY:
62d9f0db 2868 case KVM_CAP_MP_STATE:
ed848624 2869 case KVM_CAP_SYNC_MMU:
a355c85c 2870 case KVM_CAP_USER_NMI:
52d939a0 2871 case KVM_CAP_REINJECT_CONTROL:
4925663a 2872 case KVM_CAP_IRQ_INJECT_STATUS:
d34e6b17 2873 case KVM_CAP_IOEVENTFD:
f848a5a8 2874 case KVM_CAP_IOEVENTFD_NO_LENGTH:
c5ff41ce 2875 case KVM_CAP_PIT2:
e9f42757 2876 case KVM_CAP_PIT_STATE2:
b927a3ce 2877 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
ffde22ac 2878 case KVM_CAP_XEN_HVM:
afbcf7ab 2879 case KVM_CAP_ADJUST_CLOCK:
3cfc3092 2880 case KVM_CAP_VCPU_EVENTS:
55cd8e5a 2881 case KVM_CAP_HYPERV:
10388a07 2882 case KVM_CAP_HYPERV_VAPIC:
c25bc163 2883 case KVM_CAP_HYPERV_SPIN:
ab9f4ecb 2884 case KVM_CAP_PCI_SEGMENT:
a1efbe77 2885 case KVM_CAP_DEBUGREGS:
d2be1651 2886 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2d5b5a66 2887 case KVM_CAP_XSAVE:
344d9588 2888 case KVM_CAP_ASYNC_PF:
92a1f12d 2889 case KVM_CAP_GET_TSC_KHZ:
1c0b28c2 2890 case KVM_CAP_KVMCLOCK_CTRL:
4d8b81ab 2891 case KVM_CAP_READONLY_MEM:
5f66b620 2892 case KVM_CAP_HYPERV_TIME:
100943c5 2893 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
defcf51f 2894 case KVM_CAP_TSC_DEADLINE_TIMER:
90de4a18
NA		 2895 case KVM_CAP_ENABLE_CAP_VM:
2896 case KVM_CAP_DISABLE_QUIRKS:
2a5bab10
AW		 2897#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2898 case KVM_CAP_ASSIGN_DEV_IRQ:
2899 case KVM_CAP_PCI_2_3:
2900#endif
018d00d2
ZX		 2901 r = 1;
2902 break;
542472b5
LV		 2903 case KVM_CAP_COALESCED_MMIO:
2904 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2905 break;
774ead3a
AK		 2906 case KVM_CAP_VAPIC:
2907 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2908 break;
f725230a 2909 case KVM_CAP_NR_VCPUS:
8c3ba334
SL		 2910 r = KVM_SOFT_MAX_VCPUS;
2911 break;
2912 case KVM_CAP_MAX_VCPUS:
f725230a
AK		 2913 r = KVM_MAX_VCPUS;
2914 break;
a988b910 2915 case KVM_CAP_NR_MEMSLOTS:
bbacc0c1 2916 r = KVM_USER_MEM_SLOTS;
a988b910 2917 break;
a68a6a72
MT		 2918 case KVM_CAP_PV_MMU: /* obsolete */
2919 r = 0;
2f333bcb 2920 break;
4cee4b72 2921#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
62c476c7 2922 case KVM_CAP_IOMMU:
a1b60c1c 2923 r = iommu_present(&pci_bus_type);
62c476c7 2924 break;
4cee4b72 2925#endif
890ca9ae
HY		 2926 case KVM_CAP_MCE:
2927 r = KVM_MAX_MCE_BANKS;
2928 break;
2d5b5a66
SY		 2929 case KVM_CAP_XCRS:
2930 r = cpu_has_xsave;
2931 break;
92a1f12d
JR		 2932 case KVM_CAP_TSC_CONTROL:
2933 r = kvm_has_tsc_control;
2934 break;
018d00d2
ZX		 2935 default:
2936 r = 0;
2937 break;
2938 }
2939 return r;
2940
2941}
2942
043405e1
CO		 2943long kvm_arch_dev_ioctl(struct file *filp,
2944 unsigned int ioctl, unsigned long arg)
2945{
2946 void __user *argp = (void __user *)arg;
2947 long r;
2948
2949 switch (ioctl) {
2950 case KVM_GET_MSR_INDEX_LIST: {
2951 struct kvm_msr_list __user *user_msr_list = argp;
2952 struct kvm_msr_list msr_list;
2953 unsigned n;
2954
2955 r = -EFAULT;
2956 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2957 goto out;
2958 n = msr_list.nmsrs;
62ef68bb 2959 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
043405e1
CO		 2960 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2961 goto out;
2962 r = -E2BIG;
e125e7b6 2963 if (n < msr_list.nmsrs)
043405e1
CO		 2964 goto out;
2965 r = -EFAULT;
2966 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2967 num_msrs_to_save * sizeof(u32)))
2968 goto out;
e125e7b6 2969 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
043405e1 2970 &emulated_msrs,
62ef68bb 2971 num_emulated_msrs * sizeof(u32)))
043405e1
CO		 2972 goto out;
2973 r = 0;
2974 break;
2975 }
9c15bb1d
BP		 2976 case KVM_GET_SUPPORTED_CPUID:
2977 case KVM_GET_EMULATED_CPUID: {
674eea0f
AK		 2978 struct kvm_cpuid2 __user *cpuid_arg = argp;
2979 struct kvm_cpuid2 cpuid;
2980
2981 r = -EFAULT;
2982 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2983 goto out;
9c15bb1d
BP		 2984
2985 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2986 ioctl);
674eea0f
AK		 2987 if (r)
2988 goto out;
2989
2990 r = -EFAULT;
2991 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2992 goto out;
2993 r = 0;
2994 break;
2995 }
890ca9ae
HY		 2996 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2997 u64 mce_cap;
2998
2999 mce_cap = KVM_MCE_CAP_SUPPORTED;
3000 r = -EFAULT;
3001 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
3002 goto out;
3003 r = 0;
3004 break;
3005 }
043405e1
CO		 3006 default:
3007 r = -EINVAL;
3008 }
3009out:
3010 return r;
3011}
3012
f5f48ee1
SY		 3013static void wbinvd_ipi(void *garbage)
3014{
3015 wbinvd();
3016}
3017
3018static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3019{
e0f0bbc5 3020 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
f5f48ee1
SY		 3021}
3022
313a3dc7
CO		 3023void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3024{
f5f48ee1
SY		 3025 /* Address WBINVD may be executed by guest */
3026 if (need_emulate_wbinvd(vcpu)) {
3027 if (kvm_x86_ops->has_wbinvd_exit())
3028 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3029 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3030 smp_call_function_single(vcpu->cpu,
3031 wbinvd_ipi, NULL, 1);
3032 }
3033
313a3dc7 3034 kvm_x86_ops->vcpu_load(vcpu, cpu);
8f6055cb 3035
0dd6a6ed
ZA		 3036 /* Apply any externally detected TSC adjustments (due to suspend) */
3037 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3038 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3039 vcpu->arch.tsc_offset_adjustment = 0;
105b21bb 3040 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed 3041 }
8f6055cb 3042
48434c20 3043 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
6f526ec5
ZA		 3044 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3045 native_read_tsc() - vcpu->arch.last_host_tsc;
e48672fa
ZA		 3046 if (tsc_delta < 0)
3047 mark_tsc_unstable("KVM discovered backwards TSC");
c285545f 3048 if (check_tsc_unstable()) {
b183aa58
ZA		 3049 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
3050 vcpu->arch.last_guest_tsc);
3051 kvm_x86_ops->write_tsc_offset(vcpu, offset);
c285545f 3052 vcpu->arch.tsc_catchup = 1;
c285545f 3053 }
d98d07ca
MT		 3054 /*
3055 * On a host with synchronized TSC, there is no need to update
3056 * kvmclock on vcpu->cpu migration
3057 */
3058 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
0061d53d 3059 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
c285545f
ZA		 3060 if (vcpu->cpu != cpu)
3061 kvm_migrate_timers(vcpu);
e48672fa 3062 vcpu->cpu = cpu;
6b7d7e76 3063 }
c9aaa895
GC		 3064
3065 accumulate_steal_time(vcpu);
3066 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
313a3dc7
CO		 3067}
3068
3069void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3070{
02daab21 3071 kvm_x86_ops->vcpu_put(vcpu);
1c11e713 3072 kvm_put_guest_fpu(vcpu);
6f526ec5 3073 vcpu->arch.last_host_tsc = native_read_tsc();
313a3dc7
CO		 3074}
3075
313a3dc7
CO		 3076static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3077 struct kvm_lapic_state *s)
3078{
5a71785d 3079 kvm_x86_ops->sync_pir_to_irr(vcpu);
ad312c7c 3080 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
313a3dc7
CO		 3081
3082 return 0;
3083}
3084
3085static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3086 struct kvm_lapic_state *s)
3087{
64eb0620 3088 kvm_apic_post_state_restore(vcpu, s);
cb142eb7 3089 update_cr8_intercept(vcpu);
313a3dc7
CO		 3090
3091 return 0;
3092}
3093
f77bc6a4
ZX		 3094static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3095 struct kvm_interrupt *irq)
3096{
02cdb50f 3097 if (irq->irq >= KVM_NR_INTERRUPTS)
f77bc6a4
ZX		 3098 return -EINVAL;
3099 if (irqchip_in_kernel(vcpu->kvm))
3100 return -ENXIO;
f77bc6a4 3101
66fd3f7f 3102 kvm_queue_interrupt(vcpu, irq->irq, false);
3842d135 3103 kvm_make_request(KVM_REQ_EVENT, vcpu);
f77bc6a4 3104
f77bc6a4
ZX		 3105 return 0;
3106}
3107
c4abb7c9
JK		 3108static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3109{
c4abb7c9 3110 kvm_inject_nmi(vcpu);
c4abb7c9
JK		 3111
3112 return 0;
3113}
3114
f077825a
PB		 3115static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3116{
64d60670
PB		 3117 kvm_make_request(KVM_REQ_SMI, vcpu);
3118
f077825a
PB		 3119 return 0;
3120}
3121
b209749f
AK		 3122static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3123 struct kvm_tpr_access_ctl *tac)
3124{
3125 if (tac->flags)
3126 return -EINVAL;
3127 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3128 return 0;
3129}
3130
890ca9ae
HY		 3131static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3132 u64 mcg_cap)
3133{
3134 int r;
3135 unsigned bank_num = mcg_cap & 0xff, bank;
3136
3137 r = -EINVAL;
a9e38c3e 3138 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
890ca9ae
HY		 3139 goto out;
3140 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
3141 goto out;
3142 r = 0;
3143 vcpu->arch.mcg_cap = mcg_cap;
3144 /* Init IA32_MCG_CTL to all 1s */
3145 if (mcg_cap & MCG_CTL_P)
3146 vcpu->arch.mcg_ctl = ~(u64)0;
3147 /* Init IA32_MCi_CTL to all 1s */
3148 for (bank = 0; bank < bank_num; bank++)
3149 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3150out:
3151 return r;
3152}
3153
3154static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3155 struct kvm_x86_mce *mce)
3156{
3157 u64 mcg_cap = vcpu->arch.mcg_cap;
3158 unsigned bank_num = mcg_cap & 0xff;
3159 u64 *banks = vcpu->arch.mce_banks;
3160
3161 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3162 return -EINVAL;
3163 /*
3164 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3165 * reporting is disabled
3166 */
3167 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3168 vcpu->arch.mcg_ctl != ~(u64)0)
3169 return 0;
3170 banks += 4 * mce->bank;
3171 /*
3172 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3173 * reporting is disabled for the bank
3174 */
3175 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3176 return 0;
3177 if (mce->status & MCI_STATUS_UC) {
3178 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
fc78f519 3179 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
a8eeb04a 3180 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
890ca9ae
HY		 3181 return 0;
3182 }
3183 if (banks[1] & MCI_STATUS_VAL)
3184 mce->status |= MCI_STATUS_OVER;
3185 banks[2] = mce->addr;
3186 banks[3] = mce->misc;
3187 vcpu->arch.mcg_status = mce->mcg_status;
3188 banks[1] = mce->status;
3189 kvm_queue_exception(vcpu, MC_VECTOR);
3190 } else if (!(banks[1] & MCI_STATUS_VAL)
3191 || !(banks[1] & MCI_STATUS_UC)) {
3192 if (banks[1] & MCI_STATUS_VAL)
3193 mce->status |= MCI_STATUS_OVER;
3194 banks[2] = mce->addr;
3195 banks[3] = mce->misc;
3196 banks[1] = mce->status;
3197 } else
3198 banks[1] |= MCI_STATUS_OVER;
3199 return 0;
3200}
3201
3cfc3092
JK		 3202static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3203 struct kvm_vcpu_events *events)
3204{
7460fb4a 3205 process_nmi(vcpu);
03b82a30
JK		 3206 events->exception.injected =
3207 vcpu->arch.exception.pending &&
3208 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3cfc3092
JK		 3209 events->exception.nr = vcpu->arch.exception.nr;
3210 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
97e69aa6 3211 events->exception.pad = 0;
3cfc3092
JK		 3212 events->exception.error_code = vcpu->arch.exception.error_code;
3213
03b82a30
JK		 3214 events->interrupt.injected =
3215 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3cfc3092 3216 events->interrupt.nr = vcpu->arch.interrupt.nr;
03b82a30 3217 events->interrupt.soft = 0;
37ccdcbe 3218 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3cfc3092
JK		 3219
3220 events->nmi.injected = vcpu->arch.nmi_injected;
7460fb4a 3221 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3cfc3092 3222 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
97e69aa6 3223 events->nmi.pad = 0;
3cfc3092 3224
66450a21 3225 events->sipi_vector = 0; /* never valid when reporting to user space */
3cfc3092 3226
f077825a
PB		 3227 events->smi.smm = is_smm(vcpu);
3228 events->smi.pending = vcpu->arch.smi_pending;
3229 events->smi.smm_inside_nmi =
3230 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3231 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3232
dab4b911 3233 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
f077825a
PB		 3234 | KVM_VCPUEVENT_VALID_SHADOW
3235 | KVM_VCPUEVENT_VALID_SMM);
97e69aa6 3236 memset(&events->reserved, 0, sizeof(events->reserved));
3cfc3092
JK		 3237}
3238
3239static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3240 struct kvm_vcpu_events *events)
3241{
dab4b911 3242 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64 3243 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
f077825a
PB		 3244 | KVM_VCPUEVENT_VALID_SHADOW
3245 | KVM_VCPUEVENT_VALID_SMM))
3cfc3092
JK		 3246 return -EINVAL;
3247
7460fb4a 3248 process_nmi(vcpu);
3cfc3092
JK		 3249 vcpu->arch.exception.pending = events->exception.injected;
3250 vcpu->arch.exception.nr = events->exception.nr;
3251 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3252 vcpu->arch.exception.error_code = events->exception.error_code;
3253
3254 vcpu->arch.interrupt.pending = events->interrupt.injected;
3255 vcpu->arch.interrupt.nr = events->interrupt.nr;
3256 vcpu->arch.interrupt.soft = events->interrupt.soft;
48005f64
JK		 3257 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3258 kvm_x86_ops->set_interrupt_shadow(vcpu,
3259 events->interrupt.shadow);
3cfc3092
JK		 3260
3261 vcpu->arch.nmi_injected = events->nmi.injected;
dab4b911
JK		 3262 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3263 vcpu->arch.nmi_pending = events->nmi.pending;
3cfc3092
JK		 3264 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3265
66450a21
JK		 3266 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3267 kvm_vcpu_has_lapic(vcpu))
3268 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3cfc3092 3269
f077825a
PB		 3270 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3271 if (events->smi.smm)
3272 vcpu->arch.hflags |= HF_SMM_MASK;
3273 else
3274 vcpu->arch.hflags &= ~HF_SMM_MASK;
3275 vcpu->arch.smi_pending = events->smi.pending;
3276 if (events->smi.smm_inside_nmi)
3277 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3278 else
3279 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3280 if (kvm_vcpu_has_lapic(vcpu)) {
3281 if (events->smi.latched_init)
3282 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3283 else
3284 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3285 }
3286 }
3287
3842d135
AK		 3288 kvm_make_request(KVM_REQ_EVENT, vcpu);
3289
3cfc3092
JK		 3290 return 0;
3291}
3292
a1efbe77
JK		 3293static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3294 struct kvm_debugregs *dbgregs)
3295{
73aaf249
JK		 3296 unsigned long val;
3297
a1efbe77 3298 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
16f8a6f9 3299 kvm_get_dr(vcpu, 6, &val);
73aaf249 3300 dbgregs->dr6 = val;
a1efbe77
JK		 3301 dbgregs->dr7 = vcpu->arch.dr7;
3302 dbgregs->flags = 0;
97e69aa6 3303 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
a1efbe77
JK		 3304}
3305
3306static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3307 struct kvm_debugregs *dbgregs)
3308{
3309 if (dbgregs->flags)
3310 return -EINVAL;
3311
a1efbe77 3312 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
ae561ede 3313 kvm_update_dr0123(vcpu);
a1efbe77 3314 vcpu->arch.dr6 = dbgregs->dr6;
73aaf249 3315 kvm_update_dr6(vcpu);
a1efbe77 3316 vcpu->arch.dr7 = dbgregs->dr7;
9926c9fd 3317 kvm_update_dr7(vcpu);
a1efbe77 3318
a1efbe77
JK		 3319 return 0;
3320}
3321
df1daba7
PB		 3322#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3323
3324static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3325{
3326 struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3327 u64 xstate_bv = xsave->xsave_hdr.xstate_bv;
3328 u64 valid;
3329
3330 /*
3331 * Copy legacy XSAVE area, to avoid complications with CPUID
3332 * leaves 0 and 1 in the loop below.
3333 */
3334 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3335
3336 /* Set XSTATE_BV */
3337 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3338
3339 /*
3340 * Copy each region from the possibly compacted offset to the
3341 * non-compacted offset.
3342 */
3343 valid = xstate_bv & ~XSTATE_FPSSE;
3344 while (valid) {
3345 u64 feature = valid & -valid;
3346 int index = fls64(feature) - 1;
3347 void *src = get_xsave_addr(xsave, feature);
3348
3349 if (src) {
3350 u32 size, offset, ecx, edx;
3351 cpuid_count(XSTATE_CPUID, index,
3352 &size, &offset, &ecx, &edx);
3353 memcpy(dest + offset, src, size);
3354 }
3355
3356 valid -= feature;
3357 }
3358}
3359
3360static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3361{
3362 struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3363 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3364 u64 valid;
3365
3366 /*
3367 * Copy legacy XSAVE area, to avoid complications with CPUID
3368 * leaves 0 and 1 in the loop below.
3369 */
3370 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3371
3372 /* Set XSTATE_BV and possibly XCOMP_BV. */
3373 xsave->xsave_hdr.xstate_bv = xstate_bv;
3374 if (cpu_has_xsaves)
3375 xsave->xsave_hdr.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3376
3377 /*
3378 * Copy each region from the non-compacted offset to the
3379 * possibly compacted offset.
3380 */
3381 valid = xstate_bv & ~XSTATE_FPSSE;
3382 while (valid) {
3383 u64 feature = valid & -valid;
3384 int index = fls64(feature) - 1;
3385 void *dest = get_xsave_addr(xsave, feature);
3386
3387 if (dest) {
3388 u32 size, offset, ecx, edx;
3389 cpuid_count(XSTATE_CPUID, index,
3390 &size, &offset, &ecx, &edx);
3391 memcpy(dest, src + offset, size);
3392 } else
3393 WARN_ON_ONCE(1);
3394
3395 valid -= feature;
3396 }
3397}
3398
2d5b5a66
SY		 3399static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3400 struct kvm_xsave *guest_xsave)
3401{
4344ee98 3402 if (cpu_has_xsave) {
df1daba7
PB		 3403 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3404 fill_xsave((u8 *) guest_xsave->region, vcpu);
4344ee98 3405 } else {
2d5b5a66
SY		 3406 memcpy(guest_xsave->region,
3407 &vcpu->arch.guest_fpu.state->fxsave,
3408 sizeof(struct i387_fxsave_struct));
3409 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3410 XSTATE_FPSSE;
3411 }
3412}
3413
3414static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3415 struct kvm_xsave *guest_xsave)
3416{
3417 u64 xstate_bv =
3418 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3419
d7876f1b
PB		 3420 if (cpu_has_xsave) {
3421 /*
3422 * Here we allow setting states that are not present in
3423 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3424 * with old userspace.
3425 */
4ff41732 3426 if (xstate_bv & ~kvm_supported_xcr0())
d7876f1b 3427 return -EINVAL;
df1daba7 3428 load_xsave(vcpu, (u8 *)guest_xsave->region);
d7876f1b 3429 } else {
2d5b5a66
SY		 3430 if (xstate_bv & ~XSTATE_FPSSE)
3431 return -EINVAL;
3432 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3433 guest_xsave->region, sizeof(struct i387_fxsave_struct));
3434 }
3435 return 0;
3436}
3437
3438static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3439 struct kvm_xcrs *guest_xcrs)
3440{
3441 if (!cpu_has_xsave) {
3442 guest_xcrs->nr_xcrs = 0;
3443 return;
3444 }
3445
3446 guest_xcrs->nr_xcrs = 1;
3447 guest_xcrs->flags = 0;
3448 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3449 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3450}
3451
3452static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3453 struct kvm_xcrs *guest_xcrs)
3454{
3455 int i, r = 0;
3456
3457 if (!cpu_has_xsave)
3458 return -EINVAL;
3459
3460 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3461 return -EINVAL;
3462
3463 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3464 /* Only support XCR0 currently */
c67a04cb 3465 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
2d5b5a66 3466 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
c67a04cb 3467 guest_xcrs->xcrs[i].value);
2d5b5a66
SY		 3468 break;
3469 }
3470 if (r)
3471 r = -EINVAL;
3472 return r;
3473}
3474
1c0b28c2
EM		 3475/*
3476 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3477 * stopped by the hypervisor. This function will be called from the host only.
3478 * EINVAL is returned when the host attempts to set the flag for a guest that
3479 * does not support pv clocks.
3480 */
3481static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3482{
0b79459b 3483 if (!vcpu->arch.pv_time_enabled)
1c0b28c2 3484 return -EINVAL;
51d59c6b 3485 vcpu->arch.pvclock_set_guest_stopped_request = true;
1c0b28c2
EM		 3486 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3487 return 0;
3488}
3489
313a3dc7
CO		 3490long kvm_arch_vcpu_ioctl(struct file *filp,
3491 unsigned int ioctl, unsigned long arg)
3492{
3493 struct kvm_vcpu *vcpu = filp->private_data;
3494 void __user *argp = (void __user *)arg;
3495 int r;
d1ac91d8
AK		 3496 union {
3497 struct kvm_lapic_state *lapic;
3498 struct kvm_xsave *xsave;
3499 struct kvm_xcrs *xcrs;
3500 void *buffer;
3501 } u;
3502
3503 u.buffer = NULL;
313a3dc7
CO		 3504 switch (ioctl) {
3505 case KVM_GET_LAPIC: {
2204ae3c
MT		 3506 r = -EINVAL;
3507 if (!vcpu->arch.apic)
3508 goto out;
d1ac91d8 3509 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
313a3dc7 3510
b772ff36 3511 r = -ENOMEM;
d1ac91d8 3512 if (!u.lapic)
b772ff36 3513 goto out;
d1ac91d8 3514 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
313a3dc7
CO		 3515 if (r)
3516 goto out;
3517 r = -EFAULT;
d1ac91d8 3518 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
313a3dc7
CO		 3519 goto out;
3520 r = 0;
3521 break;
3522 }
3523 case KVM_SET_LAPIC: {
2204ae3c
MT		 3524 r = -EINVAL;
3525 if (!vcpu->arch.apic)
3526 goto out;
ff5c2c03 3527 u.lapic = memdup_user(argp, sizeof(*u.lapic));
18595411
GC		 3528 if (IS_ERR(u.lapic))
3529 return PTR_ERR(u.lapic);
ff5c2c03 3530
d1ac91d8 3531 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
313a3dc7
CO		 3532 break;
3533 }
f77bc6a4
ZX		 3534 case KVM_INTERRUPT: {
3535 struct kvm_interrupt irq;
3536
3537 r = -EFAULT;
3538 if (copy_from_user(&irq, argp, sizeof irq))
3539 goto out;
3540 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
f77bc6a4
ZX		 3541 break;
3542 }
c4abb7c9
JK		 3543 case KVM_NMI: {
3544 r = kvm_vcpu_ioctl_nmi(vcpu);
c4abb7c9
JK		 3545 break;
3546 }
f077825a
PB		 3547 case KVM_SMI: {
3548 r = kvm_vcpu_ioctl_smi(vcpu);
3549 break;
3550 }
313a3dc7
CO		 3551 case KVM_SET_CPUID: {
3552 struct kvm_cpuid __user *cpuid_arg = argp;
3553 struct kvm_cpuid cpuid;
3554
3555 r = -EFAULT;
3556 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3557 goto out;
3558 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
313a3dc7
CO		 3559 break;
3560 }
07716717
DK		 3561 case KVM_SET_CPUID2: {
3562 struct kvm_cpuid2 __user *cpuid_arg = argp;
3563 struct kvm_cpuid2 cpuid;
3564
3565 r = -EFAULT;
3566 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3567 goto out;
3568 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
19355475 3569 cpuid_arg->entries);
07716717
DK		 3570 break;
3571 }
3572 case KVM_GET_CPUID2: {
3573 struct kvm_cpuid2 __user *cpuid_arg = argp;
3574 struct kvm_cpuid2 cpuid;
3575
3576 r = -EFAULT;
3577 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3578 goto out;
3579 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
19355475 3580 cpuid_arg->entries);
07716717
DK		 3581 if (r)
3582 goto out;
3583 r = -EFAULT;
3584 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3585 goto out;
3586 r = 0;
3587 break;
3588 }
313a3dc7 3589 case KVM_GET_MSRS:
609e36d3 3590 r = msr_io(vcpu, argp, do_get_msr, 1);
313a3dc7
CO		 3591 break;
3592 case KVM_SET_MSRS:
3593 r = msr_io(vcpu, argp, do_set_msr, 0);
3594 break;
b209749f
AK		 3595 case KVM_TPR_ACCESS_REPORTING: {
3596 struct kvm_tpr_access_ctl tac;
3597
3598 r = -EFAULT;
3599 if (copy_from_user(&tac, argp, sizeof tac))
3600 goto out;
3601 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3602 if (r)
3603 goto out;
3604 r = -EFAULT;
3605 if (copy_to_user(argp, &tac, sizeof tac))
3606 goto out;
3607 r = 0;
3608 break;
3609 };
b93463aa
AK		 3610 case KVM_SET_VAPIC_ADDR: {
3611 struct kvm_vapic_addr va;
3612
3613 r = -EINVAL;
3614 if (!irqchip_in_kernel(vcpu->kvm))
3615 goto out;
3616 r = -EFAULT;
3617 if (copy_from_user(&va, argp, sizeof va))
3618 goto out;
fda4e2e8 3619 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
b93463aa
AK		 3620 break;
3621 }
890ca9ae
HY		 3622 case KVM_X86_SETUP_MCE: {
3623 u64 mcg_cap;
3624
3625 r = -EFAULT;
3626 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3627 goto out;
3628 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3629 break;
3630 }
3631 case KVM_X86_SET_MCE: {
3632 struct kvm_x86_mce mce;
3633
3634 r = -EFAULT;
3635 if (copy_from_user(&mce, argp, sizeof mce))
3636 goto out;
3637 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3638 break;
3639 }
3cfc3092
JK		 3640 case KVM_GET_VCPU_EVENTS: {
3641 struct kvm_vcpu_events events;
3642
3643 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3644
3645 r = -EFAULT;
3646 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3647 break;
3648 r = 0;
3649 break;
3650 }
3651 case KVM_SET_VCPU_EVENTS: {
3652 struct kvm_vcpu_events events;
3653
3654 r = -EFAULT;
3655 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3656 break;
3657
3658 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3659 break;
3660 }
a1efbe77
JK		 3661 case KVM_GET_DEBUGREGS: {
3662 struct kvm_debugregs dbgregs;
3663
3664 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3665
3666 r = -EFAULT;
3667 if (copy_to_user(argp, &dbgregs,
3668 sizeof(struct kvm_debugregs)))
3669 break;
3670 r = 0;
3671 break;
3672 }
3673 case KVM_SET_DEBUGREGS: {
3674 struct kvm_debugregs dbgregs;
3675
3676 r = -EFAULT;
3677 if (copy_from_user(&dbgregs, argp,
3678 sizeof(struct kvm_debugregs)))
3679 break;
3680
3681 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3682 break;
3683 }
2d5b5a66 3684 case KVM_GET_XSAVE: {
d1ac91d8 3685 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2d5b5a66 3686 r = -ENOMEM;
d1ac91d8 3687 if (!u.xsave)
2d5b5a66
SY		 3688 break;
3689
d1ac91d8 3690 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3691
3692 r = -EFAULT;
d1ac91d8 3693 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2d5b5a66
SY		 3694 break;
3695 r = 0;
3696 break;
3697 }
3698 case KVM_SET_XSAVE: {
ff5c2c03 3699 u.xsave = memdup_user(argp, sizeof(*u.xsave));
18595411
GC		 3700 if (IS_ERR(u.xsave))
3701 return PTR_ERR(u.xsave);
2d5b5a66 3702
d1ac91d8 3703 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3704 break;
3705 }
3706 case KVM_GET_XCRS: {
d1ac91d8 3707 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2d5b5a66 3708 r = -ENOMEM;
d1ac91d8 3709 if (!u.xcrs)
2d5b5a66
SY		 3710 break;
3711
d1ac91d8 3712 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3713
3714 r = -EFAULT;
d1ac91d8 3715 if (copy_to_user(argp, u.xcrs,
2d5b5a66
SY		 3716 sizeof(struct kvm_xcrs)))
3717 break;
3718 r = 0;
3719 break;
3720 }
3721 case KVM_SET_XCRS: {
ff5c2c03 3722 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
18595411
GC		 3723 if (IS_ERR(u.xcrs))
3724 return PTR_ERR(u.xcrs);
2d5b5a66 3725
d1ac91d8 3726 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3727 break;
3728 }
92a1f12d
JR		 3729 case KVM_SET_TSC_KHZ: {
3730 u32 user_tsc_khz;
3731
3732 r = -EINVAL;
92a1f12d
JR		 3733 user_tsc_khz = (u32)arg;
3734
3735 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3736 goto out;
3737
cc578287
ZA		 3738 if (user_tsc_khz == 0)
3739 user_tsc_khz = tsc_khz;
3740
3741 kvm_set_tsc_khz(vcpu, user_tsc_khz);
92a1f12d
JR		 3742
3743 r = 0;
3744 goto out;
3745 }
3746 case KVM_GET_TSC_KHZ: {
cc578287 3747 r = vcpu->arch.virtual_tsc_khz;
92a1f12d
JR		 3748 goto out;
3749 }
1c0b28c2
EM		 3750 case KVM_KVMCLOCK_CTRL: {
3751 r = kvm_set_guest_paused(vcpu);
3752 goto out;
3753 }
313a3dc7
CO		 3754 default:
3755 r = -EINVAL;
3756 }
3757out:
d1ac91d8 3758 kfree(u.buffer);
313a3dc7
CO		 3759 return r;
3760}
3761
5b1c1493
CO		 3762int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3763{
3764 return VM_FAULT_SIGBUS;
3765}
3766
1fe779f8
CO		 3767static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3768{
3769 int ret;
3770
3771 if (addr > (unsigned int)(-3 * PAGE_SIZE))
951179ce 3772 return -EINVAL;
1fe779f8
CO		 3773 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3774 return ret;
3775}
3776
b927a3ce
SY		 3777static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3778 u64 ident_addr)
3779{
3780 kvm->arch.ept_identity_map_addr = ident_addr;
3781 return 0;
3782}
3783
1fe779f8
CO		 3784static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3785 u32 kvm_nr_mmu_pages)
3786{
3787 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3788 return -EINVAL;
3789
79fac95e 3790 mutex_lock(&kvm->slots_lock);
1fe779f8
CO		 3791
3792 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
f05e70ac 3793 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1fe779f8 3794
79fac95e 3795 mutex_unlock(&kvm->slots_lock);
1fe779f8
CO		 3796 return 0;
3797}
3798
3799static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3800{
39de71ec 3801 return kvm->arch.n_max_mmu_pages;
1fe779f8
CO		 3802}
3803
1fe779f8
CO		 3804static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3805{
3806 int r;
3807
3808 r = 0;
3809 switch (chip->chip_id) {
3810 case KVM_IRQCHIP_PIC_MASTER:
3811 memcpy(&chip->chip.pic,
3812 &pic_irqchip(kvm)->pics[0],
3813 sizeof(struct kvm_pic_state));
3814 break;
3815 case KVM_IRQCHIP_PIC_SLAVE:
3816 memcpy(&chip->chip.pic,
3817 &pic_irqchip(kvm)->pics[1],
3818 sizeof(struct kvm_pic_state));
3819 break;
3820 case KVM_IRQCHIP_IOAPIC:
eba0226b 3821 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3822 break;
3823 default:
3824 r = -EINVAL;
3825 break;
3826 }
3827 return r;
3828}
3829
3830static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3831{
3832 int r;
3833
3834 r = 0;
3835 switch (chip->chip_id) {
3836 case KVM_IRQCHIP_PIC_MASTER:
f4f51050 3837 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3838 memcpy(&pic_irqchip(kvm)->pics[0],
3839 &chip->chip.pic,
3840 sizeof(struct kvm_pic_state));
f4f51050 3841 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3842 break;
3843 case KVM_IRQCHIP_PIC_SLAVE:
f4f51050 3844 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3845 memcpy(&pic_irqchip(kvm)->pics[1],
3846 &chip->chip.pic,
3847 sizeof(struct kvm_pic_state));
f4f51050 3848 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3849 break;
3850 case KVM_IRQCHIP_IOAPIC:
eba0226b 3851 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3852 break;
3853 default:
3854 r = -EINVAL;
3855 break;
3856 }
3857 kvm_pic_update_irq(pic_irqchip(kvm));
3858 return r;
3859}
3860
e0f63cb9
SY		 3861static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3862{
3863 int r = 0;
3864
894a9c55 3865 mutex_lock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9 3866 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
894a9c55 3867 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9
SY		 3868 return r;
3869}
3870
3871static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3872{
3873 int r = 0;
3874
894a9c55 3875 mutex_lock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9 3876 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
e9f42757
BK		 3877 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3878 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3879 return r;
3880}
3881
3882static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3883{
3884 int r = 0;
3885
3886 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3887 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3888 sizeof(ps->channels));
3889 ps->flags = kvm->arch.vpit->pit_state.flags;
3890 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
97e69aa6 3891 memset(&ps->reserved, 0, sizeof(ps->reserved));
e9f42757
BK		 3892 return r;
3893}
3894
3895static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3896{
3897 int r = 0, start = 0;
3898 u32 prev_legacy, cur_legacy;
3899 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3900 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3901 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3902 if (!prev_legacy && cur_legacy)
3903 start = 1;
3904 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3905 sizeof(kvm->arch.vpit->pit_state.channels));
3906 kvm->arch.vpit->pit_state.flags = ps->flags;
3907 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
894a9c55 3908 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9
SY		 3909 return r;
3910}
3911
52d939a0
MT		 3912static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3913 struct kvm_reinject_control *control)
3914{
3915 if (!kvm->arch.vpit)
3916 return -ENXIO;
894a9c55 3917 mutex_lock(&kvm->arch.vpit->pit_state.lock);
26ef1924 3918 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
894a9c55 3919 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
52d939a0
MT		 3920 return 0;
3921}
3922
95d4c16c 3923/**
60c34612
TY		 3924 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3925 * @kvm: kvm instance
3926 * @log: slot id and address to which we copy the log
95d4c16c 3927 *
e108ff2f
PB		 3928 * Steps 1-4 below provide general overview of dirty page logging. See
3929 * kvm_get_dirty_log_protect() function description for additional details.
3930 *
3931 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3932 * always flush the TLB (step 4) even if previous step failed and the dirty
3933 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3934 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3935 * writes will be marked dirty for next log read.
95d4c16c 3936 *
60c34612
TY		 3937 * 1. Take a snapshot of the bit and clear it if needed.
3938 * 2. Write protect the corresponding page.
e108ff2f
PB		 3939 * 3. Copy the snapshot to the userspace.
3940 * 4. Flush TLB's if needed.
5bb064dc 3941 */
60c34612 3942int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
5bb064dc 3943{
60c34612 3944 bool is_dirty = false;
e108ff2f 3945 int r;
5bb064dc 3946
79fac95e 3947 mutex_lock(&kvm->slots_lock);
5bb064dc 3948
88178fd4
KH		 3949 /*
3950 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3951 */
3952 if (kvm_x86_ops->flush_log_dirty)
3953 kvm_x86_ops->flush_log_dirty(kvm);
3954
e108ff2f 3955 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
198c74f4
XG		 3956
3957 /*
3958 * All the TLBs can be flushed out of mmu lock, see the comments in
3959 * kvm_mmu_slot_remove_write_access().
3960 */
e108ff2f 3961 lockdep_assert_held(&kvm->slots_lock);
198c74f4
XG		 3962 if (is_dirty)
3963 kvm_flush_remote_tlbs(kvm);
3964
79fac95e 3965 mutex_unlock(&kvm->slots_lock);
5bb064dc
ZX		 3966 return r;
3967}
3968
aa2fbe6d
YZ		 3969int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3970 bool line_status)
23d43cf9
CD		 3971{
3972 if (!irqchip_in_kernel(kvm))
3973 return -ENXIO;
3974
3975 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
aa2fbe6d
YZ		 3976 irq_event->irq, irq_event->level,
3977 line_status);
23d43cf9
CD		 3978 return 0;
3979}
3980
90de4a18
NA		 3981static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3982 struct kvm_enable_cap *cap)
3983{
3984 int r;
3985
3986 if (cap->flags)
3987 return -EINVAL;
3988
3989 switch (cap->cap) {
3990 case KVM_CAP_DISABLE_QUIRKS:
3991 kvm->arch.disabled_quirks = cap->args[0];
3992 r = 0;
3993 break;
3994 default:
3995 r = -EINVAL;
3996 break;
3997 }
3998 return r;
3999}
4000
1fe779f8
CO		 4001long kvm_arch_vm_ioctl(struct file *filp,
4002 unsigned int ioctl, unsigned long arg)
4003{
4004 struct kvm *kvm = filp->private_data;
4005 void __user *argp = (void __user *)arg;
367e1319 4006 int r = -ENOTTY;
f0d66275
DH		 4007 /*
4008 * This union makes it completely explicit to gcc-3.x
4009 * that these two variables' stack usage should be
4010 * combined, not added together.
4011 */
4012 union {
4013 struct kvm_pit_state ps;
e9f42757 4014 struct kvm_pit_state2 ps2;
c5ff41ce 4015 struct kvm_pit_config pit_config;
f0d66275 4016 } u;
1fe779f8
CO		 4017
4018 switch (ioctl) {
4019 case KVM_SET_TSS_ADDR:
4020 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1fe779f8 4021 break;
b927a3ce
SY		 4022 case KVM_SET_IDENTITY_MAP_ADDR: {
4023 u64 ident_addr;
4024
4025 r = -EFAULT;
4026 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4027 goto out;
4028 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
b927a3ce
SY		 4029 break;
4030 }
1fe779f8
CO		 4031 case KVM_SET_NR_MMU_PAGES:
4032 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1fe779f8
CO		 4033 break;
4034 case KVM_GET_NR_MMU_PAGES:
4035 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4036 break;
3ddea128
MT		 4037 case KVM_CREATE_IRQCHIP: {
4038 struct kvm_pic *vpic;
4039
4040 mutex_lock(&kvm->lock);
4041 r = -EEXIST;
4042 if (kvm->arch.vpic)
4043 goto create_irqchip_unlock;
3e515705
AK		 4044 r = -EINVAL;
4045 if (atomic_read(&kvm->online_vcpus))
4046 goto create_irqchip_unlock;
1fe779f8 4047 r = -ENOMEM;
3ddea128
MT		 4048 vpic = kvm_create_pic(kvm);
4049 if (vpic) {
1fe779f8
CO		 4050 r = kvm_ioapic_init(kvm);
4051 if (r) {
175504cd 4052 mutex_lock(&kvm->slots_lock);
72bb2fcd 4053 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
743eeb0b
SL		 4054 &vpic->dev_master);
4055 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
4056 &vpic->dev_slave);
4057 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
4058 &vpic->dev_eclr);
175504cd 4059 mutex_unlock(&kvm->slots_lock);
3ddea128
MT		 4060 kfree(vpic);
4061 goto create_irqchip_unlock;
1fe779f8
CO		 4062 }
4063 } else
3ddea128
MT		 4064 goto create_irqchip_unlock;
4065 smp_wmb();
4066 kvm->arch.vpic = vpic;
4067 smp_wmb();
399ec807
AK		 4068 r = kvm_setup_default_irq_routing(kvm);
4069 if (r) {
175504cd 4070 mutex_lock(&kvm->slots_lock);
3ddea128 4071 mutex_lock(&kvm->irq_lock);
72bb2fcd
WY		 4072 kvm_ioapic_destroy(kvm);
4073 kvm_destroy_pic(kvm);
3ddea128 4074 mutex_unlock(&kvm->irq_lock);
175504cd 4075 mutex_unlock(&kvm->slots_lock);
399ec807 4076 }
3ddea128
MT		 4077 create_irqchip_unlock:
4078 mutex_unlock(&kvm->lock);
1fe779f8 4079 break;
3ddea128 4080 }
7837699f 4081 case KVM_CREATE_PIT:
c5ff41ce
JK		 4082 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4083 goto create_pit;
4084 case KVM_CREATE_PIT2:
4085 r = -EFAULT;
4086 if (copy_from_user(&u.pit_config, argp,
4087 sizeof(struct kvm_pit_config)))
4088 goto out;
4089 create_pit:
79fac95e 4090 mutex_lock(&kvm->slots_lock);
269e05e4
AK		 4091 r = -EEXIST;
4092 if (kvm->arch.vpit)
4093 goto create_pit_unlock;
7837699f 4094 r = -ENOMEM;
c5ff41ce 4095 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
7837699f
SY		 4096 if (kvm->arch.vpit)
4097 r = 0;
269e05e4 4098 create_pit_unlock:
79fac95e 4099 mutex_unlock(&kvm->slots_lock);
7837699f 4100 break;
1fe779f8
CO		 4101 case KVM_GET_IRQCHIP: {
4102 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 4103 struct kvm_irqchip *chip;
1fe779f8 4104
ff5c2c03
SL		 4105 chip = memdup_user(argp, sizeof(*chip));
4106 if (IS_ERR(chip)) {
4107 r = PTR_ERR(chip);
1fe779f8 4108 goto out;
ff5c2c03
SL		 4109 }
4110
1fe779f8
CO		 4111 r = -ENXIO;
4112 if (!irqchip_in_kernel(kvm))
f0d66275
DH		 4113 goto get_irqchip_out;
4114 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1fe779f8 4115 if (r)
f0d66275 4116 goto get_irqchip_out;
1fe779f8 4117 r = -EFAULT;
f0d66275
DH		 4118 if (copy_to_user(argp, chip, sizeof *chip))
4119 goto get_irqchip_out;
1fe779f8 4120 r = 0;
f0d66275
DH		 4121 get_irqchip_out:
4122 kfree(chip);
1fe779f8
CO		 4123 break;
4124 }
4125 case KVM_SET_IRQCHIP: {
4126 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 4127 struct kvm_irqchip *chip;
1fe779f8 4128
ff5c2c03
SL		 4129 chip = memdup_user(argp, sizeof(*chip));
4130 if (IS_ERR(chip)) {
4131 r = PTR_ERR(chip);
1fe779f8 4132 goto out;
ff5c2c03
SL		 4133 }
4134
1fe779f8
CO		 4135 r = -ENXIO;
4136 if (!irqchip_in_kernel(kvm))
f0d66275
DH		 4137 goto set_irqchip_out;
4138 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1fe779f8 4139 if (r)
f0d66275 4140 goto set_irqchip_out;
1fe779f8 4141 r = 0;
f0d66275
DH		 4142 set_irqchip_out:
4143 kfree(chip);
1fe779f8
CO		 4144 break;
4145 }
e0f63cb9 4146 case KVM_GET_PIT: {
e0f63cb9 4147 r = -EFAULT;
f0d66275 4148 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 4149 goto out;
4150 r = -ENXIO;
4151 if (!kvm->arch.vpit)
4152 goto out;
f0d66275 4153 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
e0f63cb9
SY		 4154 if (r)
4155 goto out;
4156 r = -EFAULT;
f0d66275 4157 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 4158 goto out;
4159 r = 0;
4160 break;
4161 }
4162 case KVM_SET_PIT: {
e0f63cb9 4163 r = -EFAULT;
f0d66275 4164 if (copy_from_user(&u.ps, argp, sizeof u.ps))
e0f63cb9
SY		 4165 goto out;
4166 r = -ENXIO;
4167 if (!kvm->arch.vpit)
4168 goto out;
f0d66275 4169 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
e0f63cb9
SY		 4170 break;
4171 }
e9f42757
BK		 4172 case KVM_GET_PIT2: {
4173 r = -ENXIO;
4174 if (!kvm->arch.vpit)
4175 goto out;
4176 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4177 if (r)
4178 goto out;
4179 r = -EFAULT;
4180 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4181 goto out;
4182 r = 0;
4183 break;
4184 }
4185 case KVM_SET_PIT2: {
4186 r = -EFAULT;
4187 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4188 goto out;
4189 r = -ENXIO;
4190 if (!kvm->arch.vpit)
4191 goto out;
4192 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
e9f42757
BK		 4193 break;
4194 }
52d939a0
MT		 4195 case KVM_REINJECT_CONTROL: {
4196 struct kvm_reinject_control control;
4197 r = -EFAULT;
4198 if (copy_from_user(&control, argp, sizeof(control)))
4199 goto out;
4200 r = kvm_vm_ioctl_reinject(kvm, &control);
52d939a0
MT		 4201 break;
4202 }
ffde22ac
ES		 4203 case KVM_XEN_HVM_CONFIG: {
4204 r = -EFAULT;
4205 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4206 sizeof(struct kvm_xen_hvm_config)))
4207 goto out;
4208 r = -EINVAL;
4209 if (kvm->arch.xen_hvm_config.flags)
4210 goto out;
4211 r = 0;
4212 break;
4213 }
afbcf7ab 4214 case KVM_SET_CLOCK: {
afbcf7ab
GC		 4215 struct kvm_clock_data user_ns;
4216 u64 now_ns;
4217 s64 delta;
4218
4219 r = -EFAULT;
4220 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4221 goto out;
4222
4223 r = -EINVAL;
4224 if (user_ns.flags)
4225 goto out;
4226
4227 r = 0;
395c6b0a 4228 local_irq_disable();
759379dd 4229 now_ns = get_kernel_ns();
afbcf7ab 4230 delta = user_ns.clock - now_ns;
395c6b0a 4231 local_irq_enable();
afbcf7ab 4232 kvm->arch.kvmclock_offset = delta;
2e762ff7 4233 kvm_gen_update_masterclock(kvm);
afbcf7ab
GC		 4234 break;
4235 }
4236 case KVM_GET_CLOCK: {
afbcf7ab
GC		 4237 struct kvm_clock_data user_ns;
4238 u64 now_ns;
4239
395c6b0a 4240 local_irq_disable();
759379dd 4241 now_ns = get_kernel_ns();
afbcf7ab 4242 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
395c6b0a 4243 local_irq_enable();
afbcf7ab 4244 user_ns.flags = 0;
97e69aa6 4245 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
afbcf7ab
GC		 4246
4247 r = -EFAULT;
4248 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4249 goto out;
4250 r = 0;
4251 break;
4252 }
90de4a18
NA		 4253 case KVM_ENABLE_CAP: {
4254 struct kvm_enable_cap cap;
afbcf7ab 4255
90de4a18
NA		 4256 r = -EFAULT;
4257 if (copy_from_user(&cap, argp, sizeof(cap)))
4258 goto out;
4259 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4260 break;
4261 }
1fe779f8 4262 default:
c274e03a 4263 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1fe779f8
CO		 4264 }
4265out:
4266 return r;
4267}
4268
a16b043c 4269static void kvm_init_msr_list(void)
043405e1
CO		 4270{
4271 u32 dummy[2];
4272 unsigned i, j;
4273
62ef68bb 4274 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
043405e1
CO		 4275 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4276 continue;
93c4adc7
PB		 4277
4278 /*
4279 * Even MSRs that are valid in the host may not be exposed
4280 * to the guests in some cases. We could work around this
4281 * in VMX with the generic MSR save/load machinery, but it
4282 * is not really worthwhile since it will really only
4283 * happen with nested virtualization.
4284 */
4285 switch (msrs_to_save[i]) {
4286 case MSR_IA32_BNDCFGS:
4287 if (!kvm_x86_ops->mpx_supported())
4288 continue;
4289 break;
4290 default:
4291 break;
4292 }
4293
043405e1
CO		 4294 if (j < i)
4295 msrs_to_save[j] = msrs_to_save[i];
4296 j++;
4297 }
4298 num_msrs_to_save = j;
62ef68bb
PB		 4299
4300 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4301 switch (emulated_msrs[i]) {
4302 default:
4303 break;
4304 }
4305
4306 if (j < i)
4307 emulated_msrs[j] = emulated_msrs[i];
4308 j++;
4309 }
4310 num_emulated_msrs = j;
043405e1
CO		 4311}
4312
bda9020e
MT		 4313static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4314 const void *v)
bbd9b64e 4315{
70252a10
AK		 4316 int handled = 0;
4317 int n;
4318
4319 do {
4320 n = min(len, 8);
4321 if (!(vcpu->arch.apic &&
e32edf4f
NN		 4322 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4323 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10
AK		 4324 break;
4325 handled += n;
4326 addr += n;
4327 len -= n;
4328 v += n;
4329 } while (len);
bbd9b64e 4330
70252a10 4331 return handled;
bbd9b64e
CO		 4332}
4333
bda9020e 4334static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
bbd9b64e 4335{
70252a10
AK		 4336 int handled = 0;
4337 int n;
4338
4339 do {
4340 n = min(len, 8);
4341 if (!(vcpu->arch.apic &&
e32edf4f
NN		 4342 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4343 addr, n, v))
4344 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10
AK		 4345 break;
4346 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4347 handled += n;
4348 addr += n;
4349 len -= n;
4350 v += n;
4351 } while (len);
bbd9b64e 4352
70252a10 4353 return handled;
bbd9b64e
CO		 4354}
4355
2dafc6c2
GN		 4356static void kvm_set_segment(struct kvm_vcpu *vcpu,
4357 struct kvm_segment *var, int seg)
4358{
4359 kvm_x86_ops->set_segment(vcpu, var, seg);
4360}
4361
4362void kvm_get_segment(struct kvm_vcpu *vcpu,
4363 struct kvm_segment *var, int seg)
4364{
4365 kvm_x86_ops->get_segment(vcpu, var, seg);
4366}
4367
54987b7a
PB		 4368gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4369 struct x86_exception *exception)
02f59dc9
JR		 4370{
4371 gpa_t t_gpa;
02f59dc9
JR		 4372
4373 BUG_ON(!mmu_is_nested(vcpu));
4374
4375 /* NPT walks are always user-walks */
4376 access |= PFERR_USER_MASK;
54987b7a 4377 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
02f59dc9
JR		 4378
4379 return t_gpa;
4380}
4381
ab9ae313
AK		 4382gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4383 struct x86_exception *exception)
1871c602
GN		 4384{
4385 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
ab9ae313 4386 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4387}
4388
ab9ae313
AK		 4389 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4390 struct x86_exception *exception)
1871c602
GN		 4391{
4392 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4393 access |= PFERR_FETCH_MASK;
ab9ae313 4394 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4395}
4396
ab9ae313
AK		 4397gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4398 struct x86_exception *exception)
1871c602
GN		 4399{
4400 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4401 access |= PFERR_WRITE_MASK;
ab9ae313 4402 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4403}
4404
4405/* uses this to access any guest's mapped memory without checking CPL */
ab9ae313
AK		 4406gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4407 struct x86_exception *exception)
1871c602 4408{
ab9ae313 4409 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
1871c602
GN		 4410}
4411
4412static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4413 struct kvm_vcpu *vcpu, u32 access,
bcc55cba 4414 struct x86_exception *exception)
bbd9b64e
CO		 4415{
4416 void *data = val;
10589a46 4417 int r = X86EMUL_CONTINUE;
bbd9b64e
CO		 4418
4419 while (bytes) {
14dfe855 4420 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
ab9ae313 4421 exception);
bbd9b64e 4422 unsigned offset = addr & (PAGE_SIZE-1);
77c2002e 4423 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
bbd9b64e
CO		 4424 int ret;
4425
bcc55cba 4426 if (gpa == UNMAPPED_GVA)
ab9ae313 4427 return X86EMUL_PROPAGATE_FAULT;
44583cba
PB		 4428 ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, data,
4429 offset, toread);
10589a46 4430 if (ret < 0) {
c3cd7ffa 4431 r = X86EMUL_IO_NEEDED;
10589a46
MT		 4432 goto out;
4433 }
bbd9b64e 4434
77c2002e
IE		 4435 bytes -= toread;
4436 data += toread;
4437 addr += toread;
bbd9b64e 4438 }
10589a46 4439out:
10589a46 4440 return r;
bbd9b64e 4441}
77c2002e 4442
1871c602 4443/* used for instruction fetching */
0f65dd70
AK		 4444static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4445 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4446 struct x86_exception *exception)
1871c602 4447{
0f65dd70 4448 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4449 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
44583cba
PB		 4450 unsigned offset;
4451 int ret;
0f65dd70 4452
44583cba
PB		 4453 /* Inline kvm_read_guest_virt_helper for speed. */
4454 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4455 exception);
4456 if (unlikely(gpa == UNMAPPED_GVA))
4457 return X86EMUL_PROPAGATE_FAULT;
4458
4459 offset = addr & (PAGE_SIZE-1);
4460 if (WARN_ON(offset + bytes > PAGE_SIZE))
4461 bytes = (unsigned)PAGE_SIZE - offset;
4462 ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, val,
4463 offset, bytes);
4464 if (unlikely(ret < 0))
4465 return X86EMUL_IO_NEEDED;
4466
4467 return X86EMUL_CONTINUE;
1871c602
GN		 4468}
4469
064aea77 4470int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
0f65dd70 4471 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4472 struct x86_exception *exception)
1871c602 4473{
0f65dd70 4474 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4475 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
0f65dd70 4476
1871c602 4477 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
bcc55cba 4478 exception);
1871c602 4479}
064aea77 4480EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
1871c602 4481
0f65dd70
AK		 4482static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4483 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4484 struct x86_exception *exception)
1871c602 4485{
0f65dd70 4486 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
bcc55cba 4487 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
1871c602
GN		 4488}
4489
6a4d7550 4490int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
0f65dd70 4491 gva_t addr, void *val,
2dafc6c2 4492 unsigned int bytes,
bcc55cba 4493 struct x86_exception *exception)
77c2002e 4494{
0f65dd70 4495 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
77c2002e
IE		 4496 void *data = val;
4497 int r = X86EMUL_CONTINUE;
4498
4499 while (bytes) {
14dfe855
JR		 4500 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4501 PFERR_WRITE_MASK,
ab9ae313 4502 exception);
77c2002e
IE		 4503 unsigned offset = addr & (PAGE_SIZE-1);
4504 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4505 int ret;
4506
bcc55cba 4507 if (gpa == UNMAPPED_GVA)
ab9ae313 4508 return X86EMUL_PROPAGATE_FAULT;
77c2002e
IE		 4509 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4510 if (ret < 0) {
c3cd7ffa 4511 r = X86EMUL_IO_NEEDED;
77c2002e
IE		 4512 goto out;
4513 }
4514
4515 bytes -= towrite;
4516 data += towrite;
4517 addr += towrite;
4518 }
4519out:
4520 return r;
4521}
6a4d7550 4522EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
77c2002e 4523
af7cc7d1
XG		 4524static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4525 gpa_t *gpa, struct x86_exception *exception,
4526 bool write)
4527{
97d64b78
AK		 4528 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4529 | (write ? PFERR_WRITE_MASK : 0);
af7cc7d1 4530
97d64b78 4531 if (vcpu_match_mmio_gva(vcpu, gva)
97ec8c06
FW		 4532 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4533 vcpu->arch.access, access)) {
bebb106a
XG		 4534 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4535 (gva & (PAGE_SIZE - 1));
4f022648 4536 trace_vcpu_match_mmio(gva, *gpa, write, false);
bebb106a
XG		 4537 return 1;
4538 }
4539
af7cc7d1
XG		 4540 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4541
4542 if (*gpa == UNMAPPED_GVA)
4543 return -1;
4544
4545 /* For APIC access vmexit */
4546 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4547 return 1;
4548
4f022648
XG		 4549 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4550 trace_vcpu_match_mmio(gva, *gpa, write, true);
bebb106a 4551 return 1;
4f022648 4552 }
bebb106a 4553
af7cc7d1
XG		 4554 return 0;
4555}
4556
3200f405 4557int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
bcc55cba 4558 const void *val, int bytes)
bbd9b64e
CO		 4559{
4560 int ret;
4561
4562 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
9f811285 4563 if (ret < 0)
bbd9b64e 4564 return 0;
f57f2ef5 4565 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
bbd9b64e
CO		 4566 return 1;
4567}
4568
77d197b2
XG		 4569struct read_write_emulator_ops {
4570 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4571 int bytes);
4572 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4573 void *val, int bytes);
4574 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4575 int bytes, void *val);
4576 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4577 void *val, int bytes);
4578 bool write;
4579};
4580
4581static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4582{
4583 if (vcpu->mmio_read_completed) {
77d197b2 4584 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
f78146b0 4585 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
77d197b2
XG		 4586 vcpu->mmio_read_completed = 0;
4587 return 1;
4588 }
4589
4590 return 0;
4591}
4592
4593static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4594 void *val, int bytes)
4595{
4596 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4597}
4598
4599static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4600 void *val, int bytes)
4601{
4602 return emulator_write_phys(vcpu, gpa, val, bytes);
4603}
4604
4605static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4606{
4607 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4608 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4609}
4610
4611static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4612 void *val, int bytes)
4613{
4614 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4615 return X86EMUL_IO_NEEDED;
4616}
4617
4618static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4619 void *val, int bytes)
4620{
f78146b0
AK		 4621 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4622
87da7e66 4623 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
77d197b2
XG		 4624 return X86EMUL_CONTINUE;
4625}
4626
0fbe9b0b 4627static const struct read_write_emulator_ops read_emultor = {
77d197b2
XG		 4628 .read_write_prepare = read_prepare,
4629 .read_write_emulate = read_emulate,
4630 .read_write_mmio = vcpu_mmio_read,
4631 .read_write_exit_mmio = read_exit_mmio,
4632};
4633
0fbe9b0b 4634static const struct read_write_emulator_ops write_emultor = {
77d197b2
XG		 4635 .read_write_emulate = write_emulate,
4636 .read_write_mmio = write_mmio,
4637 .read_write_exit_mmio = write_exit_mmio,
4638 .write = true,
4639};
4640
22388a3c
XG		 4641static int emulator_read_write_onepage(unsigned long addr, void *val,
4642 unsigned int bytes,
4643 struct x86_exception *exception,
4644 struct kvm_vcpu *vcpu,
0fbe9b0b 4645 const struct read_write_emulator_ops *ops)
bbd9b64e 4646{
af7cc7d1
XG		 4647 gpa_t gpa;
4648 int handled, ret;
22388a3c 4649 bool write = ops->write;
f78146b0 4650 struct kvm_mmio_fragment *frag;
10589a46 4651
22388a3c 4652 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
bbd9b64e 4653
af7cc7d1 4654 if (ret < 0)
bbd9b64e 4655 return X86EMUL_PROPAGATE_FAULT;
bbd9b64e
CO		 4656
4657 /* For APIC access vmexit */
af7cc7d1 4658 if (ret)
bbd9b64e
CO		 4659 goto mmio;
4660
22388a3c 4661 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
bbd9b64e
CO		 4662 return X86EMUL_CONTINUE;
4663
4664mmio:
4665 /*
4666 * Is this MMIO handled locally?
4667 */
22388a3c 4668 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
70252a10 4669 if (handled == bytes)
bbd9b64e 4670 return X86EMUL_CONTINUE;
bbd9b64e 4671
70252a10
AK		 4672 gpa += handled;
4673 bytes -= handled;
4674 val += handled;
4675
87da7e66
XG		 4676 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4677 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4678 frag->gpa = gpa;
4679 frag->data = val;
4680 frag->len = bytes;
f78146b0 4681 return X86EMUL_CONTINUE;
bbd9b64e
CO		 4682}
4683
52eb5a6d
XL		 4684static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4685 unsigned long addr,
22388a3c
XG		 4686 void *val, unsigned int bytes,
4687 struct x86_exception *exception,
0fbe9b0b 4688 const struct read_write_emulator_ops *ops)
bbd9b64e 4689{
0f65dd70 4690 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
f78146b0
AK		 4691 gpa_t gpa;
4692 int rc;
4693
4694 if (ops->read_write_prepare &&
4695 ops->read_write_prepare(vcpu, val, bytes))
4696 return X86EMUL_CONTINUE;
4697
4698 vcpu->mmio_nr_fragments = 0;
0f65dd70 4699
bbd9b64e
CO		 4700 /* Crossing a page boundary? */
4701 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
f78146b0 4702 int now;
bbd9b64e
CO		 4703
4704 now = -addr & ~PAGE_MASK;
22388a3c
XG		 4705 rc = emulator_read_write_onepage(addr, val, now, exception,
4706 vcpu, ops);
4707
bbd9b64e
CO		 4708 if (rc != X86EMUL_CONTINUE)
4709 return rc;
4710 addr += now;
bac15531
NA		 4711 if (ctxt->mode != X86EMUL_MODE_PROT64)
4712 addr = (u32)addr;
bbd9b64e
CO		 4713 val += now;
4714 bytes -= now;
4715 }
22388a3c 4716
f78146b0
AK		 4717 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4718 vcpu, ops);
4719 if (rc != X86EMUL_CONTINUE)
4720 return rc;
4721
4722 if (!vcpu->mmio_nr_fragments)
4723 return rc;
4724
4725 gpa = vcpu->mmio_fragments[0].gpa;
4726
4727 vcpu->mmio_needed = 1;
4728 vcpu->mmio_cur_fragment = 0;
4729
87da7e66 4730 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
f78146b0
AK		 4731 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4732 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4733 vcpu->run->mmio.phys_addr = gpa;
4734
4735 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
22388a3c
XG		 4736}
4737
4738static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4739 unsigned long addr,
4740 void *val,
4741 unsigned int bytes,
4742 struct x86_exception *exception)
4743{
4744 return emulator_read_write(ctxt, addr, val, bytes,
4745 exception, &read_emultor);
4746}
4747
52eb5a6d 4748static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
22388a3c
XG		 4749 unsigned long addr,
4750 const void *val,
4751 unsigned int bytes,
4752 struct x86_exception *exception)
4753{
4754 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4755 exception, &write_emultor);
bbd9b64e 4756}
bbd9b64e 4757
daea3e73
AK		 4758#define CMPXCHG_TYPE(t, ptr, old, new) \
4759 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4760
4761#ifdef CONFIG_X86_64
4762# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4763#else
4764# define CMPXCHG64(ptr, old, new) \
9749a6c0 4765 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
daea3e73
AK		 4766#endif
4767
0f65dd70
AK		 4768static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4769 unsigned long addr,
bbd9b64e
CO		 4770 const void *old,
4771 const void *new,
4772 unsigned int bytes,
0f65dd70 4773 struct x86_exception *exception)
bbd9b64e 4774{
0f65dd70 4775 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
daea3e73
AK		 4776 gpa_t gpa;
4777 struct page *page;
4778 char *kaddr;
4779 bool exchanged;
2bacc55c 4780
daea3e73
AK		 4781 /* guests cmpxchg8b have to be emulated atomically */
4782 if (bytes > 8 || (bytes & (bytes - 1)))
4783 goto emul_write;
10589a46 4784
daea3e73 4785 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
2bacc55c 4786
daea3e73
AK		 4787 if (gpa == UNMAPPED_GVA ||
4788 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4789 goto emul_write;
2bacc55c 4790
daea3e73
AK		 4791 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4792 goto emul_write;
72dc67a6 4793
daea3e73 4794 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
32cad84f 4795 if (is_error_page(page))
c19b8bd6 4796 goto emul_write;
72dc67a6 4797
8fd75e12 4798 kaddr = kmap_atomic(page);
daea3e73
AK		 4799 kaddr += offset_in_page(gpa);
4800 switch (bytes) {
4801 case 1:
4802 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4803 break;
4804 case 2:
4805 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4806 break;
4807 case 4:
4808 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4809 break;
4810 case 8:
4811 exchanged = CMPXCHG64(kaddr, old, new);
4812 break;
4813 default:
4814 BUG();
2bacc55c 4815 }
8fd75e12 4816 kunmap_atomic(kaddr);
daea3e73
AK		 4817 kvm_release_page_dirty(page);
4818
4819 if (!exchanged)
4820 return X86EMUL_CMPXCHG_FAILED;
4821
d3714010 4822 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
f57f2ef5 4823 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
8f6abd06
GN		 4824
4825 return X86EMUL_CONTINUE;
4a5f48f6 4826
3200f405 4827emul_write:
daea3e73 4828 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2bacc55c 4829
0f65dd70 4830 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
bbd9b64e
CO		 4831}
4832
cf8f70bf
GN		 4833static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4834{
4835 /* TODO: String I/O for in kernel device */
4836 int r;
4837
4838 if (vcpu->arch.pio.in)
e32edf4f 4839 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
cf8f70bf
GN		 4840 vcpu->arch.pio.size, pd);
4841 else
e32edf4f 4842 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
cf8f70bf
GN		 4843 vcpu->arch.pio.port, vcpu->arch.pio.size,
4844 pd);
4845 return r;
4846}
4847
6f6fbe98
XG		 4848static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4849 unsigned short port, void *val,
4850 unsigned int count, bool in)
cf8f70bf 4851{
cf8f70bf 4852 vcpu->arch.pio.port = port;
6f6fbe98 4853 vcpu->arch.pio.in = in;
7972995b 4854 vcpu->arch.pio.count = count;
cf8f70bf
GN		 4855 vcpu->arch.pio.size = size;
4856
4857 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
7972995b 4858 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4859 return 1;
4860 }
4861
4862 vcpu->run->exit_reason = KVM_EXIT_IO;
6f6fbe98 4863 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
cf8f70bf
GN		 4864 vcpu->run->io.size = size;
4865 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4866 vcpu->run->io.count = count;
4867 vcpu->run->io.port = port;
4868
4869 return 0;
4870}
4871
6f6fbe98
XG		 4872static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4873 int size, unsigned short port, void *val,
4874 unsigned int count)
cf8f70bf 4875{
ca1d4a9e 4876 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6f6fbe98 4877 int ret;
ca1d4a9e 4878
6f6fbe98
XG		 4879 if (vcpu->arch.pio.count)
4880 goto data_avail;
cf8f70bf 4881
6f6fbe98
XG		 4882 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4883 if (ret) {
4884data_avail:
4885 memcpy(val, vcpu->arch.pio_data, size * count);
1171903d 4886 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
7972995b 4887 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4888 return 1;
4889 }
4890
cf8f70bf
GN		 4891 return 0;
4892}
4893
6f6fbe98
XG		 4894static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4895 int size, unsigned short port,
4896 const void *val, unsigned int count)
4897{
4898 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4899
4900 memcpy(vcpu->arch.pio_data, val, size * count);
1171903d 4901 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6f6fbe98
XG		 4902 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4903}
4904
bbd9b64e
CO		 4905static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4906{
4907 return kvm_x86_ops->get_segment_base(vcpu, seg);
4908}
4909
3cb16fe7 4910static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
bbd9b64e 4911{
3cb16fe7 4912 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
bbd9b64e
CO		 4913}
4914
5cb56059 4915int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
f5f48ee1
SY		 4916{
4917 if (!need_emulate_wbinvd(vcpu))
4918 return X86EMUL_CONTINUE;
4919
4920 if (kvm_x86_ops->has_wbinvd_exit()) {
2eec7343
JK		 4921 int cpu = get_cpu();
4922
4923 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
f5f48ee1
SY		 4924 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4925 wbinvd_ipi, NULL, 1);
2eec7343 4926 put_cpu();
f5f48ee1 4927 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
2eec7343
JK		 4928 } else
4929 wbinvd();
f5f48ee1
SY		 4930 return X86EMUL_CONTINUE;
4931}
5cb56059
JS		 4932
4933int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4934{
4935 kvm_x86_ops->skip_emulated_instruction(vcpu);
4936 return kvm_emulate_wbinvd_noskip(vcpu);
4937}
f5f48ee1
SY		 4938EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4939
5cb56059
JS		 4940
4941
bcaf5cc5
AK		 4942static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4943{
5cb56059 4944 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
bcaf5cc5
AK		 4945}
4946
52eb5a6d
XL		 4947static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4948 unsigned long *dest)
bbd9b64e 4949{
16f8a6f9 4950 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
bbd9b64e
CO		 4951}
4952
52eb5a6d
XL		 4953static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4954 unsigned long value)
bbd9b64e 4955{
338dbc97 4956
717746e3 4957 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
bbd9b64e
CO		 4958}
4959
52a46617 4960static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5fdbf976 4961{
52a46617 4962 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5fdbf976
MT		 4963}
4964
717746e3 4965static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
bbd9b64e 4966{
717746e3 4967 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
52a46617
GN		 4968 unsigned long value;
4969
4970 switch (cr) {
4971 case 0:
4972 value = kvm_read_cr0(vcpu);
4973 break;
4974 case 2:
4975 value = vcpu->arch.cr2;
4976 break;
4977 case 3:
9f8fe504 4978 value = kvm_read_cr3(vcpu);
52a46617
GN		 4979 break;
4980 case 4:
4981 value = kvm_read_cr4(vcpu);
4982 break;
4983 case 8:
4984 value = kvm_get_cr8(vcpu);
4985 break;
4986 default:
a737f256 4987 kvm_err("%s: unexpected cr %u\n", __func__, cr);
52a46617
GN		 4988 return 0;
4989 }
4990
4991 return value;
4992}
4993
717746e3 4994static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
52a46617 4995{
717746e3 4996 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
0f12244f
GN		 4997 int res = 0;
4998
52a46617
GN		 4999 switch (cr) {
5000 case 0:
49a9b07e 5001 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
52a46617
GN		 5002 break;
5003 case 2:
5004 vcpu->arch.cr2 = val;
5005 break;
5006 case 3:
2390218b 5007 res = kvm_set_cr3(vcpu, val);
52a46617
GN		 5008 break;
5009 case 4:
a83b29c6 5010 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
52a46617
GN		 5011 break;
5012 case 8:
eea1cff9 5013 res = kvm_set_cr8(vcpu, val);
52a46617
GN		 5014 break;
5015 default:
a737f256 5016 kvm_err("%s: unexpected cr %u\n", __func__, cr);
0f12244f 5017 res = -1;
52a46617 5018 }
0f12244f
GN		 5019
5020 return res;
52a46617
GN		 5021}
5022
717746e3 5023static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
9c537244 5024{
717746e3 5025 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
9c537244
GN		 5026}
5027
4bff1e86 5028static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
2dafc6c2 5029{
4bff1e86 5030 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
2dafc6c2
GN		 5031}
5032
4bff1e86 5033static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
160ce1f1 5034{
4bff1e86 5035 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
160ce1f1
MG		 5036}
5037
1ac9d0cf
AK		 5038static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5039{
5040 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5041}
5042
5043static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5044{
5045 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5046}
5047
4bff1e86
AK		 5048static unsigned long emulator_get_cached_segment_base(
5049 struct x86_emulate_ctxt *ctxt, int seg)
5951c442 5050{
4bff1e86 5051 return get_segment_base(emul_to_vcpu(ctxt), seg);
5951c442
GN		 5052}
5053
1aa36616
AK		 5054static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5055 struct desc_struct *desc, u32 *base3,
5056 int seg)
2dafc6c2
GN		 5057{
5058 struct kvm_segment var;
5059
4bff1e86 5060 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
1aa36616 5061 *selector = var.selector;
2dafc6c2 5062
378a8b09
GN		 5063 if (var.unusable) {
5064 memset(desc, 0, sizeof(*desc));
2dafc6c2 5065 return false;
378a8b09 5066 }
2dafc6c2
GN		 5067
5068 if (var.g)
5069 var.limit >>= 12;
5070 set_desc_limit(desc, var.limit);
5071 set_desc_base(desc, (unsigned long)var.base);
5601d05b
GN		 5072#ifdef CONFIG_X86_64
5073 if (base3)
5074 *base3 = var.base >> 32;
5075#endif
2dafc6c2
GN		 5076 desc->type = var.type;
5077 desc->s = var.s;
5078 desc->dpl = var.dpl;
5079 desc->p = var.present;
5080 desc->avl = var.avl;
5081 desc->l = var.l;
5082 desc->d = var.db;
5083 desc->g = var.g;
5084
5085 return true;
5086}
5087
1aa36616
AK		 5088static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5089 struct desc_struct *desc, u32 base3,
5090 int seg)
2dafc6c2 5091{
4bff1e86 5092 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
2dafc6c2
GN		 5093 struct kvm_segment var;
5094
1aa36616 5095 var.selector = selector;
2dafc6c2 5096 var.base = get_desc_base(desc);
5601d05b
GN		 5097#ifdef CONFIG_X86_64
5098 var.base |= ((u64)base3) << 32;
5099#endif
2dafc6c2
GN		 5100 var.limit = get_desc_limit(desc);
5101 if (desc->g)
5102 var.limit = (var.limit << 12) | 0xfff;
5103 var.type = desc->type;
2dafc6c2
GN		 5104 var.dpl = desc->dpl;
5105 var.db = desc->d;
5106 var.s = desc->s;
5107 var.l = desc->l;
5108 var.g = desc->g;
5109 var.avl = desc->avl;
5110 var.present = desc->p;
5111 var.unusable = !var.present;
5112 var.padding = 0;
5113
5114 kvm_set_segment(vcpu, &var, seg);
5115 return;
5116}
5117
717746e3
AK		 5118static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5119 u32 msr_index, u64 *pdata)
5120{
609e36d3
PB		 5121 struct msr_data msr;
5122 int r;
5123
5124 msr.index = msr_index;
5125 msr.host_initiated = false;
5126 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5127 if (r)
5128 return r;
5129
5130 *pdata = msr.data;
5131 return 0;
717746e3
AK		 5132}
5133
5134static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5135 u32 msr_index, u64 data)
5136{
8fe8ab46
WA		 5137 struct msr_data msr;
5138
5139 msr.data = data;
5140 msr.index = msr_index;
5141 msr.host_initiated = false;
5142 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
717746e3
AK		 5143}
5144
64d60670
PB		 5145static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5146{
5147 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5148
5149 return vcpu->arch.smbase;
5150}
5151
5152static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5153{
5154 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5155
5156 vcpu->arch.smbase = smbase;
5157}
5158
67f4d428
NA		 5159static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5160 u32 pmc)
5161{
5162 return kvm_pmu_check_pmc(emul_to_vcpu(ctxt), pmc);
5163}
5164
222d21aa
AK		 5165static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5166 u32 pmc, u64 *pdata)
5167{
5168 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
5169}
5170
6c3287f7
AK		 5171static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5172{
5173 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5174}
5175
5037f6f3
AK		 5176static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5177{
5178 preempt_disable();
5197b808 5179 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5037f6f3
AK		 5180 /*
5181 * CR0.TS may reference the host fpu state, not the guest fpu state,
5182 * so it may be clear at this point.
5183 */
5184 clts();
5185}
5186
5187static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5188{
5189 preempt_enable();
5190}
5191
2953538e 5192static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
8a76d7f2 5193 struct x86_instruction_info *info,
c4f035c6
AK		 5194 enum x86_intercept_stage stage)
5195{
2953538e 5196 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
c4f035c6
AK		 5197}
5198
0017f93a 5199static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
bdb42f5a
SB		 5200 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5201{
0017f93a 5202 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
bdb42f5a
SB		 5203}
5204
dd856efa
AK		 5205static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5206{
5207 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5208}
5209
5210static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5211{
5212 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5213}
5214
801806d9
NA		 5215static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5216{
5217 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5218}
5219
0225fb50 5220static const struct x86_emulate_ops emulate_ops = {
dd856efa
AK		 5221 .read_gpr = emulator_read_gpr,
5222 .write_gpr = emulator_write_gpr,
1871c602 5223 .read_std = kvm_read_guest_virt_system,
2dafc6c2 5224 .write_std = kvm_write_guest_virt_system,
1871c602 5225 .fetch = kvm_fetch_guest_virt,
bbd9b64e
CO		 5226 .read_emulated = emulator_read_emulated,
5227 .write_emulated = emulator_write_emulated,
5228 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3cb16fe7 5229 .invlpg = emulator_invlpg,
cf8f70bf
GN		 5230 .pio_in_emulated = emulator_pio_in_emulated,
5231 .pio_out_emulated = emulator_pio_out_emulated,
1aa36616
AK		 5232 .get_segment = emulator_get_segment,
5233 .set_segment = emulator_set_segment,
5951c442 5234 .get_cached_segment_base = emulator_get_cached_segment_base,
2dafc6c2 5235 .get_gdt = emulator_get_gdt,
160ce1f1 5236 .get_idt = emulator_get_idt,
1ac9d0cf
AK		 5237 .set_gdt = emulator_set_gdt,
5238 .set_idt = emulator_set_idt,
52a46617
GN		 5239 .get_cr = emulator_get_cr,
5240 .set_cr = emulator_set_cr,
9c537244 5241 .cpl = emulator_get_cpl,
35aa5375
GN		 5242 .get_dr = emulator_get_dr,
5243 .set_dr = emulator_set_dr,
64d60670
PB		 5244 .get_smbase = emulator_get_smbase,
5245 .set_smbase = emulator_set_smbase,
717746e3
AK		 5246 .set_msr = emulator_set_msr,
5247 .get_msr = emulator_get_msr,
67f4d428 5248 .check_pmc = emulator_check_pmc,
222d21aa 5249 .read_pmc = emulator_read_pmc,
6c3287f7 5250 .halt = emulator_halt,
bcaf5cc5 5251 .wbinvd = emulator_wbinvd,
d6aa1000 5252 .fix_hypercall = emulator_fix_hypercall,
5037f6f3
AK		 5253 .get_fpu = emulator_get_fpu,
5254 .put_fpu = emulator_put_fpu,
c4f035c6 5255 .intercept = emulator_intercept,
bdb42f5a 5256 .get_cpuid = emulator_get_cpuid,
801806d9 5257 .set_nmi_mask = emulator_set_nmi_mask,
bbd9b64e
CO		 5258};
5259
95cb2295
GN		 5260static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5261{
37ccdcbe 5262 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
95cb2295
GN		 5263 /*
5264 * an sti; sti; sequence only disable interrupts for the first
5265 * instruction. So, if the last instruction, be it emulated or
5266 * not, left the system with the INT_STI flag enabled, it
5267 * means that the last instruction is an sti. We should not
5268 * leave the flag on in this case. The same goes for mov ss
5269 */
37ccdcbe
PB		 5270 if (int_shadow & mask)
5271 mask = 0;
6addfc42 5272 if (unlikely(int_shadow || mask)) {
95cb2295 5273 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6addfc42
PB		 5274 if (!mask)
5275 kvm_make_request(KVM_REQ_EVENT, vcpu);
5276 }
95cb2295
GN		 5277}
5278
ef54bcfe 5279static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
54b8486f
GN		 5280{
5281 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
da9cb575 5282 if (ctxt->exception.vector == PF_VECTOR)
ef54bcfe
PB		 5283 return kvm_propagate_fault(vcpu, &ctxt->exception);
5284
5285 if (ctxt->exception.error_code_valid)
da9cb575
AK		 5286 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5287 ctxt->exception.error_code);
54b8486f 5288 else
da9cb575 5289 kvm_queue_exception(vcpu, ctxt->exception.vector);
ef54bcfe 5290 return false;
54b8486f
GN		 5291}
5292
8ec4722d
MG		 5293static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5294{
adf52235 5295 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d
MG		 5296 int cs_db, cs_l;
5297
8ec4722d
MG		 5298 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5299
adf52235
TY		 5300 ctxt->eflags = kvm_get_rflags(vcpu);
5301 ctxt->eip = kvm_rip_read(vcpu);
5302 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5303 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
42bf549f 5304 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
adf52235
TY		 5305 cs_db ? X86EMUL_MODE_PROT32 :
5306 X86EMUL_MODE_PROT16;
a584539b 5307 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
64d60670
PB		 5308 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5309 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
a584539b 5310 ctxt->emul_flags = vcpu->arch.hflags;
adf52235 5311
dd856efa 5312 init_decode_cache(ctxt);
7ae441ea 5313 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8ec4722d
MG		 5314}
5315
71f9833b 5316int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
63995653 5317{
9d74191a 5318 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
63995653
MG		 5319 int ret;
5320
5321 init_emulate_ctxt(vcpu);
5322
9dac77fa
AK		 5323 ctxt->op_bytes = 2;
5324 ctxt->ad_bytes = 2;
5325 ctxt->_eip = ctxt->eip + inc_eip;
9d74191a 5326 ret = emulate_int_real(ctxt, irq);
63995653
MG		 5327
5328 if (ret != X86EMUL_CONTINUE)
5329 return EMULATE_FAIL;
5330
9dac77fa 5331 ctxt->eip = ctxt->_eip;
9d74191a
TY		 5332 kvm_rip_write(vcpu, ctxt->eip);
5333 kvm_set_rflags(vcpu, ctxt->eflags);
63995653
MG		 5334
5335 if (irq == NMI_VECTOR)
7460fb4a 5336 vcpu->arch.nmi_pending = 0;
63995653
MG		 5337 else
5338 vcpu->arch.interrupt.pending = false;
5339
5340 return EMULATE_DONE;
5341}
5342EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5343
6d77dbfc
GN		 5344static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5345{
fc3a9157
JR		 5346 int r = EMULATE_DONE;
5347
6d77dbfc
GN		 5348 ++vcpu->stat.insn_emulation_fail;
5349 trace_kvm_emulate_insn_failed(vcpu);
a2b9e6c1 5350 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
fc3a9157
JR		 5351 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5352 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5353 vcpu->run->internal.ndata = 0;
5354 r = EMULATE_FAIL;
5355 }
6d77dbfc 5356 kvm_queue_exception(vcpu, UD_VECTOR);
fc3a9157
JR		 5357
5358 return r;
6d77dbfc
GN		 5359}
5360
93c05d3e 5361static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
991eebf9
GN		 5362 bool write_fault_to_shadow_pgtable,
5363 int emulation_type)
a6f177ef 5364{
95b3cf69 5365 gpa_t gpa = cr2;
8e3d9d06 5366 pfn_t pfn;
a6f177ef 5367
991eebf9
GN		 5368 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5369 return false;
5370
95b3cf69
XG		 5371 if (!vcpu->arch.mmu.direct_map) {
5372 /*
5373 * Write permission should be allowed since only
5374 * write access need to be emulated.
5375 */
5376 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
a6f177ef 5377
95b3cf69
XG		 5378 /*
5379 * If the mapping is invalid in guest, let cpu retry
5380 * it to generate fault.
5381 */
5382 if (gpa == UNMAPPED_GVA)
5383 return true;
5384 }
a6f177ef 5385
8e3d9d06
XG		 5386 /*
5387 * Do not retry the unhandleable instruction if it faults on the
5388 * readonly host memory, otherwise it will goto a infinite loop:
5389 * retry instruction -> write #PF -> emulation fail -> retry
5390 * instruction -> ...
5391 */
5392 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
95b3cf69
XG		 5393
5394 /*
5395 * If the instruction failed on the error pfn, it can not be fixed,
5396 * report the error to userspace.
5397 */
5398 if (is_error_noslot_pfn(pfn))
5399 return false;
5400
5401 kvm_release_pfn_clean(pfn);
5402
5403 /* The instructions are well-emulated on direct mmu. */
5404 if (vcpu->arch.mmu.direct_map) {
5405 unsigned int indirect_shadow_pages;
5406
5407 spin_lock(&vcpu->kvm->mmu_lock);
5408 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5409 spin_unlock(&vcpu->kvm->mmu_lock);
5410
5411 if (indirect_shadow_pages)
5412 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5413
a6f177ef 5414 return true;
8e3d9d06 5415 }
a6f177ef 5416
95b3cf69
XG		 5417 /*
5418 * if emulation was due to access to shadowed page table
5419 * and it failed try to unshadow page and re-enter the
5420 * guest to let CPU execute the instruction.
5421 */
5422 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
93c05d3e
XG		 5423
5424 /*
5425 * If the access faults on its page table, it can not
5426 * be fixed by unprotecting shadow page and it should
5427 * be reported to userspace.
5428 */
5429 return !write_fault_to_shadow_pgtable;
a6f177ef
GN		 5430}
5431
1cb3f3ae
XG		 5432static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5433 unsigned long cr2, int emulation_type)
5434{
5435 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5436 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5437
5438 last_retry_eip = vcpu->arch.last_retry_eip;
5439 last_retry_addr = vcpu->arch.last_retry_addr;
5440
5441 /*
5442 * If the emulation is caused by #PF and it is non-page_table
5443 * writing instruction, it means the VM-EXIT is caused by shadow
5444 * page protected, we can zap the shadow page and retry this
5445 * instruction directly.
5446 *
5447 * Note: if the guest uses a non-page-table modifying instruction
5448 * on the PDE that points to the instruction, then we will unmap
5449 * the instruction and go to an infinite loop. So, we cache the
5450 * last retried eip and the last fault address, if we meet the eip
5451 * and the address again, we can break out of the potential infinite
5452 * loop.
5453 */
5454 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5455
5456 if (!(emulation_type & EMULTYPE_RETRY))
5457 return false;
5458
5459 if (x86_page_table_writing_insn(ctxt))
5460 return false;
5461
5462 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5463 return false;
5464
5465 vcpu->arch.last_retry_eip = ctxt->eip;
5466 vcpu->arch.last_retry_addr = cr2;
5467
5468 if (!vcpu->arch.mmu.direct_map)
5469 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5470
22368028 5471 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
1cb3f3ae
XG		 5472
5473 return true;
5474}
5475
716d51ab
GN		 5476static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5477static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5478
64d60670 5479static void kvm_smm_changed(struct kvm_vcpu *vcpu)
a584539b 5480{
64d60670
PB		 5481 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5482 if (unlikely(vcpu->arch.smi_pending)) {
5483 kvm_make_request(KVM_REQ_SMI, vcpu);
5484 vcpu->arch.smi_pending = 0;
cd7764fe
PB		 5485 } else {
5486 /* Process a latched INIT, if any. */
5487 kvm_make_request(KVM_REQ_EVENT, vcpu);
64d60670
PB		 5488 }
5489 }
5490}
5491
5492static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5493{
5494 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5495
a584539b 5496 vcpu->arch.hflags = emul_flags;
64d60670
PB		 5497
5498 if (changed & HF_SMM_MASK)
5499 kvm_smm_changed(vcpu);
a584539b
PB		 5500}
5501
4a1e10d5
PB		 5502static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5503 unsigned long *db)
5504{
5505 u32 dr6 = 0;
5506 int i;
5507 u32 enable, rwlen;
5508
5509 enable = dr7;
5510 rwlen = dr7 >> 16;
5511 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5512 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5513 dr6 |= (1 << i);
5514 return dr6;
5515}
5516
6addfc42 5517static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
663f4c61
PB		 5518{
5519 struct kvm_run *kvm_run = vcpu->run;
5520
5521 /*
6addfc42
PB		 5522 * rflags is the old, "raw" value of the flags. The new value has
5523 * not been saved yet.
663f4c61
PB		 5524 *
5525 * This is correct even for TF set by the guest, because "the
5526 * processor will not generate this exception after the instruction
5527 * that sets the TF flag".
5528 */
663f4c61
PB		 5529 if (unlikely(rflags & X86_EFLAGS_TF)) {
5530 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6f43ed01
NA		 5531 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5532 DR6_RTM;
663f4c61
PB		 5533 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5534 kvm_run->debug.arch.exception = DB_VECTOR;
5535 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5536 *r = EMULATE_USER_EXIT;
5537 } else {
5538 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5539 /*
5540 * "Certain debug exceptions may clear bit 0-3. The
5541 * remaining contents of the DR6 register are never
5542 * cleared by the processor".
5543 */
5544 vcpu->arch.dr6 &= ~15;
6f43ed01 5545 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
663f4c61
PB		 5546 kvm_queue_exception(vcpu, DB_VECTOR);
5547 }
5548 }
5549}
5550
4a1e10d5
PB		 5551static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5552{
4a1e10d5
PB		 5553 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5554 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
82b32774
NA		 5555 struct kvm_run *kvm_run = vcpu->run;
5556 unsigned long eip = kvm_get_linear_rip(vcpu);
5557 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5558 vcpu->arch.guest_debug_dr7,
5559 vcpu->arch.eff_db);
5560
5561 if (dr6 != 0) {
6f43ed01 5562 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
82b32774 5563 kvm_run->debug.arch.pc = eip;
4a1e10d5
PB		 5564 kvm_run->debug.arch.exception = DB_VECTOR;
5565 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5566 *r = EMULATE_USER_EXIT;
5567 return true;
5568 }
5569 }
5570
4161a569
NA		 5571 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5572 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
82b32774
NA		 5573 unsigned long eip = kvm_get_linear_rip(vcpu);
5574 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5575 vcpu->arch.dr7,
5576 vcpu->arch.db);
5577
5578 if (dr6 != 0) {
5579 vcpu->arch.dr6 &= ~15;
6f43ed01 5580 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4a1e10d5
PB		 5581 kvm_queue_exception(vcpu, DB_VECTOR);
5582 *r = EMULATE_DONE;
5583 return true;
5584 }
5585 }
5586
5587 return false;
5588}
5589
51d8b661
AP		 5590int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5591 unsigned long cr2,
dc25e89e
AP		 5592 int emulation_type,
5593 void *insn,
5594 int insn_len)
bbd9b64e 5595{
95cb2295 5596 int r;
9d74191a 5597 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7ae441ea 5598 bool writeback = true;
93c05d3e 5599 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
bbd9b64e 5600
93c05d3e
XG		 5601 /*
5602 * Clear write_fault_to_shadow_pgtable here to ensure it is
5603 * never reused.
5604 */
5605 vcpu->arch.write_fault_to_shadow_pgtable = false;
26eef70c 5606 kvm_clear_exception_queue(vcpu);
8d7d8102 5607
571008da 5608 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8ec4722d 5609 init_emulate_ctxt(vcpu);
4a1e10d5
PB		 5610
5611 /*
5612 * We will reenter on the same instruction since
5613 * we do not set complete_userspace_io. This does not
5614 * handle watchpoints yet, those would be handled in
5615 * the emulate_ops.
5616 */
5617 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5618 return r;
5619
9d74191a
TY		 5620 ctxt->interruptibility = 0;
5621 ctxt->have_exception = false;
e0ad0b47 5622 ctxt->exception.vector = -1;
9d74191a 5623 ctxt->perm_ok = false;
bbd9b64e 5624
b51e974f 5625 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
4005996e 5626
9d74191a 5627 r = x86_decode_insn(ctxt, insn, insn_len);
bbd9b64e 5628
e46479f8 5629 trace_kvm_emulate_insn_start(vcpu);
f2b5756b 5630 ++vcpu->stat.insn_emulation;
1d2887e2 5631 if (r != EMULATION_OK) {
4005996e
AK		 5632 if (emulation_type & EMULTYPE_TRAP_UD)
5633 return EMULATE_FAIL;
991eebf9
GN		 5634 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5635 emulation_type))
bbd9b64e 5636 return EMULATE_DONE;
6d77dbfc
GN		 5637 if (emulation_type & EMULTYPE_SKIP)
5638 return EMULATE_FAIL;
5639 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 5640 }
5641 }
5642
ba8afb6b 5643 if (emulation_type & EMULTYPE_SKIP) {
9dac77fa 5644 kvm_rip_write(vcpu, ctxt->_eip);
bb663c7a
NA		 5645 if (ctxt->eflags & X86_EFLAGS_RF)
5646 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
ba8afb6b
GN		 5647 return EMULATE_DONE;
5648 }
5649
1cb3f3ae
XG		 5650 if (retry_instruction(ctxt, cr2, emulation_type))
5651 return EMULATE_DONE;
5652
7ae441ea 5653 /* this is needed for vmware backdoor interface to work since it
4d2179e1 5654 changes registers values during IO operation */
7ae441ea
GN		 5655 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5656 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
dd856efa 5657 emulator_invalidate_register_cache(ctxt);
7ae441ea 5658 }
4d2179e1 5659
5cd21917 5660restart:
9d74191a 5661 r = x86_emulate_insn(ctxt);
bbd9b64e 5662
775fde86
JR		 5663 if (r == EMULATION_INTERCEPTED)
5664 return EMULATE_DONE;
5665
d2ddd1c4 5666 if (r == EMULATION_FAILED) {
991eebf9
GN		 5667 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5668 emulation_type))
c3cd7ffa
GN		 5669 return EMULATE_DONE;
5670
6d77dbfc 5671 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 5672 }
5673
9d74191a 5674 if (ctxt->have_exception) {
d2ddd1c4 5675 r = EMULATE_DONE;
ef54bcfe
PB		 5676 if (inject_emulated_exception(vcpu))
5677 return r;
d2ddd1c4 5678 } else if (vcpu->arch.pio.count) {
0912c977
PB		 5679 if (!vcpu->arch.pio.in) {
5680 /* FIXME: return into emulator if single-stepping. */
3457e419 5681 vcpu->arch.pio.count = 0;
0912c977 5682 } else {
7ae441ea 5683 writeback = false;
716d51ab
GN		 5684 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5685 }
ac0a48c3 5686 r = EMULATE_USER_EXIT;
7ae441ea
GN		 5687 } else if (vcpu->mmio_needed) {
5688 if (!vcpu->mmio_is_write)
5689 writeback = false;
ac0a48c3 5690 r = EMULATE_USER_EXIT;
716d51ab 5691 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7ae441ea 5692 } else if (r == EMULATION_RESTART)
5cd21917 5693 goto restart;
d2ddd1c4
GN		 5694 else
5695 r = EMULATE_DONE;
f850e2e6 5696
7ae441ea 5697 if (writeback) {
6addfc42 5698 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
9d74191a 5699 toggle_interruptibility(vcpu, ctxt->interruptibility);
7ae441ea 5700 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
a584539b
PB		 5701 if (vcpu->arch.hflags != ctxt->emul_flags)
5702 kvm_set_hflags(vcpu, ctxt->emul_flags);
9d74191a 5703 kvm_rip_write(vcpu, ctxt->eip);
663f4c61 5704 if (r == EMULATE_DONE)
6addfc42 5705 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
38827dbd
NA		 5706 if (!ctxt->have_exception ||
5707 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5708 __kvm_set_rflags(vcpu, ctxt->eflags);
6addfc42
PB		 5709
5710 /*
5711 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5712 * do nothing, and it will be requested again as soon as
5713 * the shadow expires. But we still need to check here,
5714 * because POPF has no interrupt shadow.
5715 */
5716 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5717 kvm_make_request(KVM_REQ_EVENT, vcpu);
7ae441ea
GN		 5718 } else
5719 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
e85d28f8
GN		 5720
5721 return r;
de7d789a 5722}
51d8b661 5723EXPORT_SYMBOL_GPL(x86_emulate_instruction);
de7d789a 5724
cf8f70bf 5725int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
de7d789a 5726{
cf8f70bf 5727 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
ca1d4a9e
AK		 5728 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5729 size, port, &val, 1);
cf8f70bf 5730 /* do not return to emulator after return from userspace */
7972995b 5731 vcpu->arch.pio.count = 0;
de7d789a
CO		 5732 return ret;
5733}
cf8f70bf 5734EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
de7d789a 5735
8cfdc000
ZA		 5736static void tsc_bad(void *info)
5737{
0a3aee0d 5738 __this_cpu_write(cpu_tsc_khz, 0);
8cfdc000
ZA		 5739}
5740
5741static void tsc_khz_changed(void *data)
c8076604 5742{
8cfdc000
ZA		 5743 struct cpufreq_freqs *freq = data;
5744 unsigned long khz = 0;
5745
5746 if (data)
5747 khz = freq->new;
5748 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5749 khz = cpufreq_quick_get(raw_smp_processor_id());
5750 if (!khz)
5751 khz = tsc_khz;
0a3aee0d 5752 __this_cpu_write(cpu_tsc_khz, khz);
c8076604
GH		 5753}
5754
c8076604
GH		 5755static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5756 void *data)
5757{
5758 struct cpufreq_freqs *freq = data;
5759 struct kvm *kvm;
5760 struct kvm_vcpu *vcpu;
5761 int i, send_ipi = 0;
5762
8cfdc000
ZA		 5763 /*
5764 * We allow guests to temporarily run on slowing clocks,
5765 * provided we notify them after, or to run on accelerating
5766 * clocks, provided we notify them before. Thus time never
5767 * goes backwards.
5768 *
5769 * However, we have a problem. We can't atomically update
5770 * the frequency of a given CPU from this function; it is
5771 * merely a notifier, which can be called from any CPU.
5772 * Changing the TSC frequency at arbitrary points in time
5773 * requires a recomputation of local variables related to
5774 * the TSC for each VCPU. We must flag these local variables
5775 * to be updated and be sure the update takes place with the
5776 * new frequency before any guests proceed.
5777 *
5778 * Unfortunately, the combination of hotplug CPU and frequency
5779 * change creates an intractable locking scenario; the order
5780 * of when these callouts happen is undefined with respect to
5781 * CPU hotplug, and they can race with each other. As such,
5782 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5783 * undefined; you can actually have a CPU frequency change take
5784 * place in between the computation of X and the setting of the
5785 * variable. To protect against this problem, all updates of
5786 * the per_cpu tsc_khz variable are done in an interrupt
5787 * protected IPI, and all callers wishing to update the value
5788 * must wait for a synchronous IPI to complete (which is trivial
5789 * if the caller is on the CPU already). This establishes the
5790 * necessary total order on variable updates.
5791 *
5792 * Note that because a guest time update may take place
5793 * anytime after the setting of the VCPU's request bit, the
5794 * correct TSC value must be set before the request. However,
5795 * to ensure the update actually makes it to any guest which
5796 * starts running in hardware virtualization between the set
5797 * and the acquisition of the spinlock, we must also ping the
5798 * CPU after setting the request bit.
5799 *
5800 */
5801
c8076604
GH		 5802 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5803 return 0;
5804 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5805 return 0;
8cfdc000
ZA		 5806
5807 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604 5808
2f303b74 5809 spin_lock(&kvm_lock);
c8076604 5810 list_for_each_entry(kvm, &vm_list, vm_list) {
988a2cae 5811 kvm_for_each_vcpu(i, vcpu, kvm) {
c8076604
GH		 5812 if (vcpu->cpu != freq->cpu)
5813 continue;
c285545f 5814 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
c8076604 5815 if (vcpu->cpu != smp_processor_id())
8cfdc000 5816 send_ipi = 1;
c8076604
GH		 5817 }
5818 }
2f303b74 5819 spin_unlock(&kvm_lock);
c8076604
GH		 5820
5821 if (freq->old < freq->new && send_ipi) {
5822 /*
5823 * We upscale the frequency. Must make the guest
5824 * doesn't see old kvmclock values while running with
5825 * the new frequency, otherwise we risk the guest sees
5826 * time go backwards.
5827 *
5828 * In case we update the frequency for another cpu
5829 * (which might be in guest context) send an interrupt
5830 * to kick the cpu out of guest context. Next time
5831 * guest context is entered kvmclock will be updated,
5832 * so the guest will not see stale values.
5833 */
8cfdc000 5834 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604
GH		 5835 }
5836 return 0;
5837}
5838
5839static struct notifier_block kvmclock_cpufreq_notifier_block = {
8cfdc000
ZA		 5840 .notifier_call = kvmclock_cpufreq_notifier
5841};
5842
5843static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5844 unsigned long action, void *hcpu)
5845{
5846 unsigned int cpu = (unsigned long)hcpu;
5847
5848 switch (action) {
5849 case CPU_ONLINE:
5850 case CPU_DOWN_FAILED:
5851 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5852 break;
5853 case CPU_DOWN_PREPARE:
5854 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5855 break;
5856 }
5857 return NOTIFY_OK;
5858}
5859
5860static struct notifier_block kvmclock_cpu_notifier_block = {
5861 .notifier_call = kvmclock_cpu_notifier,
5862 .priority = -INT_MAX
c8076604
GH		 5863};
5864
b820cc0c
ZA		 5865static void kvm_timer_init(void)
5866{
5867 int cpu;
5868
c285545f 5869 max_tsc_khz = tsc_khz;
460dd42e
SB		 5870
5871 cpu_notifier_register_begin();
b820cc0c 5872 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
c285545f
ZA		 5873#ifdef CONFIG_CPU_FREQ
5874 struct cpufreq_policy policy;
5875 memset(&policy, 0, sizeof(policy));
3e26f230
AK		 5876 cpu = get_cpu();
5877 cpufreq_get_policy(&policy, cpu);
c285545f
ZA		 5878 if (policy.cpuinfo.max_freq)
5879 max_tsc_khz = policy.cpuinfo.max_freq;
3e26f230 5880 put_cpu();
c285545f 5881#endif
b820cc0c
ZA		 5882 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5883 CPUFREQ_TRANSITION_NOTIFIER);
5884 }
c285545f 5885 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
8cfdc000
ZA		 5886 for_each_online_cpu(cpu)
5887 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
460dd42e
SB		 5888
5889 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5890 cpu_notifier_register_done();
5891
b820cc0c
ZA		 5892}
5893
ff9d07a0
ZY		 5894static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5895
f5132b01 5896int kvm_is_in_guest(void)
ff9d07a0 5897{
086c9855 5898 return __this_cpu_read(current_vcpu) != NULL;
ff9d07a0
ZY		 5899}
5900
5901static int kvm_is_user_mode(void)
5902{
5903 int user_mode = 3;
dcf46b94 5904
086c9855
AS		 5905 if (__this_cpu_read(current_vcpu))
5906 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
dcf46b94 5907
ff9d07a0
ZY		 5908 return user_mode != 0;
5909}
5910
5911static unsigned long kvm_get_guest_ip(void)
5912{
5913 unsigned long ip = 0;
dcf46b94 5914
086c9855
AS		 5915 if (__this_cpu_read(current_vcpu))
5916 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
dcf46b94 5917
ff9d07a0
ZY		 5918 return ip;
5919}
5920
5921static struct perf_guest_info_callbacks kvm_guest_cbs = {
5922 .is_in_guest = kvm_is_in_guest,
5923 .is_user_mode = kvm_is_user_mode,
5924 .get_guest_ip = kvm_get_guest_ip,
5925};
5926
5927void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5928{
086c9855 5929 __this_cpu_write(current_vcpu, vcpu);
ff9d07a0
ZY		 5930}
5931EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5932
5933void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5934{
086c9855 5935 __this_cpu_write(current_vcpu, NULL);
ff9d07a0
ZY		 5936}
5937EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5938
ce88decf
XG		 5939static void kvm_set_mmio_spte_mask(void)
5940{
5941 u64 mask;
5942 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5943
5944 /*
5945 * Set the reserved bits and the present bit of an paging-structure
5946 * entry to generate page fault with PFER.RSV = 1.
5947 */
885032b9 5948 /* Mask the reserved physical address bits. */
d1431483 5949 mask = rsvd_bits(maxphyaddr, 51);
885032b9
XG		 5950
5951 /* Bit 62 is always reserved for 32bit host. */
5952 mask |= 0x3ull << 62;
5953
5954 /* Set the present bit. */
ce88decf
XG		 5955 mask |= 1ull;
5956
5957#ifdef CONFIG_X86_64
5958 /*
5959 * If reserved bit is not supported, clear the present bit to disable
5960 * mmio page fault.
5961 */
5962 if (maxphyaddr == 52)
5963 mask &= ~1ull;
5964#endif
5965
5966 kvm_mmu_set_mmio_spte_mask(mask);
5967}
5968
16e8d74d
MT		 5969#ifdef CONFIG_X86_64
5970static void pvclock_gtod_update_fn(struct work_struct *work)
5971{
d828199e
MT		 5972 struct kvm *kvm;
5973
5974 struct kvm_vcpu *vcpu;
5975 int i;
5976
2f303b74 5977 spin_lock(&kvm_lock);
d828199e
MT		 5978 list_for_each_entry(kvm, &vm_list, vm_list)
5979 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 5980 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
d828199e 5981 atomic_set(&kvm_guest_has_master_clock, 0);
2f303b74 5982 spin_unlock(&kvm_lock);
16e8d74d
MT		 5983}
5984
5985static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5986
5987/*
5988 * Notification about pvclock gtod data update.
5989 */
5990static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5991 void *priv)
5992{
5993 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5994 struct timekeeper *tk = priv;
5995
5996 update_pvclock_gtod(tk);
5997
5998 /* disable master clock if host does not trust, or does not
5999 * use, TSC clocksource
6000 */
6001 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6002 atomic_read(&kvm_guest_has_master_clock) != 0)
6003 queue_work(system_long_wq, &pvclock_gtod_work);
6004
6005 return 0;
6006}
6007
6008static struct notifier_block pvclock_gtod_notifier = {
6009 .notifier_call = pvclock_gtod_notify,
6010};
6011#endif
6012
f8c16bba 6013int kvm_arch_init(void *opaque)
043405e1 6014{
b820cc0c 6015 int r;
6b61edf7 6016 struct kvm_x86_ops *ops = opaque;
f8c16bba 6017
f8c16bba
ZX		 6018 if (kvm_x86_ops) {
6019 printk(KERN_ERR "kvm: already loaded the other module\n");
56c6d28a
ZX		 6020 r = -EEXIST;
6021 goto out;
f8c16bba
ZX		 6022 }
6023
6024 if (!ops->cpu_has_kvm_support()) {
6025 printk(KERN_ERR "kvm: no hardware support\n");
56c6d28a
ZX		 6026 r = -EOPNOTSUPP;
6027 goto out;
f8c16bba
ZX		 6028 }
6029 if (ops->disabled_by_bios()) {
6030 printk(KERN_ERR "kvm: disabled by bios\n");
56c6d28a
ZX		 6031 r = -EOPNOTSUPP;
6032 goto out;
f8c16bba
ZX		 6033 }
6034
013f6a5d
MT		 6035 r = -ENOMEM;
6036 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6037 if (!shared_msrs) {
6038 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6039 goto out;
6040 }
6041
97db56ce
AK		 6042 r = kvm_mmu_module_init();
6043 if (r)
013f6a5d 6044 goto out_free_percpu;
97db56ce 6045
ce88decf 6046 kvm_set_mmio_spte_mask();
97db56ce 6047
f8c16bba 6048 kvm_x86_ops = ops;
920c8377 6049
7b52345e 6050 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4b12f0de 6051 PT_DIRTY_MASK, PT64_NX_MASK, 0);
c8076604 6052
b820cc0c 6053 kvm_timer_init();
c8076604 6054
ff9d07a0
ZY		 6055 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6056
2acf923e
DC		 6057 if (cpu_has_xsave)
6058 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6059
c5cc421b 6060 kvm_lapic_init();
16e8d74d
MT		 6061#ifdef CONFIG_X86_64
6062 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6063#endif
6064
f8c16bba 6065 return 0;
56c6d28a 6066
013f6a5d
MT		 6067out_free_percpu:
6068 free_percpu(shared_msrs);
56c6d28a 6069out:
56c6d28a 6070 return r;
043405e1 6071}
8776e519 6072
f8c16bba
ZX		 6073void kvm_arch_exit(void)
6074{
ff9d07a0
ZY		 6075 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6076
888d256e
JK		 6077 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6078 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6079 CPUFREQ_TRANSITION_NOTIFIER);
8cfdc000 6080 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
16e8d74d
MT		 6081#ifdef CONFIG_X86_64
6082 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6083#endif
f8c16bba 6084 kvm_x86_ops = NULL;
56c6d28a 6085 kvm_mmu_module_exit();
013f6a5d 6086 free_percpu(shared_msrs);
56c6d28a 6087}
f8c16bba 6088
5cb56059 6089int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8776e519
HB		 6090{
6091 ++vcpu->stat.halt_exits;
6092 if (irqchip_in_kernel(vcpu->kvm)) {
a4535290 6093 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
8776e519
HB		 6094 return 1;
6095 } else {
6096 vcpu->run->exit_reason = KVM_EXIT_HLT;
6097 return 0;
6098 }
6099}
5cb56059
JS		 6100EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6101
6102int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6103{
6104 kvm_x86_ops->skip_emulated_instruction(vcpu);
6105 return kvm_vcpu_halt(vcpu);
6106}
8776e519
HB		 6107EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6108
55cd8e5a
GN		 6109int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
6110{
6111 u64 param, ingpa, outgpa, ret;
6112 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
6113 bool fast, longmode;
55cd8e5a
GN		 6114
6115 /*
6116 * hypercall generates UD from non zero cpl and real mode
6117 * per HYPER-V spec
6118 */
3eeb3288 6119 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
55cd8e5a
GN		 6120 kvm_queue_exception(vcpu, UD_VECTOR);
6121 return 0;
6122 }
6123
a449c7aa 6124 longmode = is_64_bit_mode(vcpu);
55cd8e5a
GN		 6125
6126 if (!longmode) {
ccd46936
GN		 6127 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
6128 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
6129 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
6130 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
6131 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
6132 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
55cd8e5a
GN		 6133 }
6134#ifdef CONFIG_X86_64
6135 else {
6136 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
6137 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
6138 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
6139 }
6140#endif
6141
6142 code = param & 0xffff;
6143 fast = (param >> 16) & 0x1;
6144 rep_cnt = (param >> 32) & 0xfff;
6145 rep_idx = (param >> 48) & 0xfff;
6146
6147 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
6148
c25bc163
GN		 6149 switch (code) {
6150 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
6151 kvm_vcpu_on_spin(vcpu);
6152 break;
6153 default:
6154 res = HV_STATUS_INVALID_HYPERCALL_CODE;
6155 break;
6156 }
55cd8e5a
GN		 6157
6158 ret = res | (((u64)rep_done & 0xfff) << 32);
6159 if (longmode) {
6160 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6161 } else {
6162 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
6163 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
6164 }
6165
6166 return 1;
6167}
6168
6aef266c
SV		 6169/*
6170 * kvm_pv_kick_cpu_op: Kick a vcpu.
6171 *
6172 * @apicid - apicid of vcpu to be kicked.
6173 */
6174static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6175{
24d2166b 6176 struct kvm_lapic_irq lapic_irq;
6aef266c 6177
24d2166b
R		 6178 lapic_irq.shorthand = 0;
6179 lapic_irq.dest_mode = 0;
6180 lapic_irq.dest_id = apicid;
93bbf0b8 6181 lapic_irq.msi_redir_hint = false;
6aef266c 6182
24d2166b 6183 lapic_irq.delivery_mode = APIC_DM_REMRD;
795a149e 6184 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6aef266c
SV		 6185}
6186
8776e519
HB		 6187int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6188{
6189 unsigned long nr, a0, a1, a2, a3, ret;
a449c7aa 6190 int op_64_bit, r = 1;
8776e519 6191
5cb56059
JS		 6192 kvm_x86_ops->skip_emulated_instruction(vcpu);
6193
55cd8e5a
GN		 6194 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6195 return kvm_hv_hypercall(vcpu);
6196
5fdbf976
MT		 6197 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6198 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6199 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6200 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6201 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
8776e519 6202
229456fc 6203 trace_kvm_hypercall(nr, a0, a1, a2, a3);
2714d1d3 6204
a449c7aa
NA		 6205 op_64_bit = is_64_bit_mode(vcpu);
6206 if (!op_64_bit) {
8776e519
HB		 6207 nr &= 0xFFFFFFFF;
6208 a0 &= 0xFFFFFFFF;
6209 a1 &= 0xFFFFFFFF;
6210 a2 &= 0xFFFFFFFF;
6211 a3 &= 0xFFFFFFFF;
6212 }
6213
07708c4a
JK		 6214 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6215 ret = -KVM_EPERM;
6216 goto out;
6217 }
6218
8776e519 6219 switch (nr) {
b93463aa
AK		 6220 case KVM_HC_VAPIC_POLL_IRQ:
6221 ret = 0;
6222 break;
6aef266c
SV		 6223 case KVM_HC_KICK_CPU:
6224 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6225 ret = 0;
6226 break;
8776e519
HB		 6227 default:
6228 ret = -KVM_ENOSYS;
6229 break;
6230 }
07708c4a 6231out:
a449c7aa
NA		 6232 if (!op_64_bit)
6233 ret = (u32)ret;
5fdbf976 6234 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
f11c3a8d 6235 ++vcpu->stat.hypercalls;
2f333bcb 6236 return r;
8776e519
HB		 6237}
6238EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6239
b6785def 6240static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8776e519 6241{
d6aa1000 6242 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8776e519 6243 char instruction[3];
5fdbf976 6244 unsigned long rip = kvm_rip_read(vcpu);
8776e519 6245
8776e519 6246 kvm_x86_ops->patch_hypercall(vcpu, instruction);
8776e519 6247
9d74191a 6248 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
8776e519
HB		 6249}
6250
b6c7a5dc
HB		 6251/*
6252 * Check if userspace requested an interrupt window, and that the
6253 * interrupt window is open.
6254 *
6255 * No need to exit to userspace if we already have an interrupt queued.
6256 */
851ba692 6257static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
b6c7a5dc 6258{
8061823a 6259 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
851ba692 6260 vcpu->run->request_interrupt_window &&
5df56646 6261 kvm_arch_interrupt_allowed(vcpu));
b6c7a5dc
HB		 6262}
6263
851ba692 6264static void post_kvm_run_save(struct kvm_vcpu *vcpu)
b6c7a5dc 6265{
851ba692
AK		 6266 struct kvm_run *kvm_run = vcpu->run;
6267
91586a3b 6268 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
f077825a 6269 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
2d3ad1f4 6270 kvm_run->cr8 = kvm_get_cr8(vcpu);
b6c7a5dc 6271 kvm_run->apic_base = kvm_get_apic_base(vcpu);
4531220b 6272 if (irqchip_in_kernel(vcpu->kvm))
b6c7a5dc 6273 kvm_run->ready_for_interrupt_injection = 1;
4531220b 6274 else
b6c7a5dc 6275 kvm_run->ready_for_interrupt_injection =
fa9726b0
GN		 6276 kvm_arch_interrupt_allowed(vcpu) &&
6277 !kvm_cpu_has_interrupt(vcpu) &&
6278 !kvm_event_needs_reinjection(vcpu);
b6c7a5dc
HB		 6279}
6280
95ba8273
GN		 6281static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6282{
6283 int max_irr, tpr;
6284
6285 if (!kvm_x86_ops->update_cr8_intercept)
6286 return;
6287
88c808fd
AK		 6288 if (!vcpu->arch.apic)
6289 return;
6290
8db3baa2
GN		 6291 if (!vcpu->arch.apic->vapic_addr)
6292 max_irr = kvm_lapic_find_highest_irr(vcpu);
6293 else
6294 max_irr = -1;
95ba8273
GN		 6295
6296 if (max_irr != -1)
6297 max_irr >>= 4;
6298
6299 tpr = kvm_lapic_get_cr8(vcpu);
6300
6301 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6302}
6303
b6b8a145 6304static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
95ba8273 6305{
b6b8a145
JK		 6306 int r;
6307
95ba8273 6308 /* try to reinject previous events if any */
b59bb7bd 6309 if (vcpu->arch.exception.pending) {
5c1c85d0
AK		 6310 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6311 vcpu->arch.exception.has_error_code,
6312 vcpu->arch.exception.error_code);
d6e8c854
NA		 6313
6314 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6315 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6316 X86_EFLAGS_RF);
6317
6bdf0662
NA		 6318 if (vcpu->arch.exception.nr == DB_VECTOR &&
6319 (vcpu->arch.dr7 & DR7_GD)) {
6320 vcpu->arch.dr7 &= ~DR7_GD;
6321 kvm_update_dr7(vcpu);
6322 }
6323
b59bb7bd
GN		 6324 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6325 vcpu->arch.exception.has_error_code,
ce7ddec4
JR		 6326 vcpu->arch.exception.error_code,
6327 vcpu->arch.exception.reinject);
b6b8a145 6328 return 0;
b59bb7bd
GN		 6329 }
6330
95ba8273
GN		 6331 if (vcpu->arch.nmi_injected) {
6332 kvm_x86_ops->set_nmi(vcpu);
b6b8a145 6333 return 0;
95ba8273
GN		 6334 }
6335
6336 if (vcpu->arch.interrupt.pending) {
66fd3f7f 6337 kvm_x86_ops->set_irq(vcpu);
b6b8a145
JK		 6338 return 0;
6339 }
6340
6341 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6342 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6343 if (r != 0)
6344 return r;
95ba8273
GN		 6345 }
6346
6347 /* try to inject new event if pending */
6348 if (vcpu->arch.nmi_pending) {
6349 if (kvm_x86_ops->nmi_allowed(vcpu)) {
7460fb4a 6350 --vcpu->arch.nmi_pending;
95ba8273
GN		 6351 vcpu->arch.nmi_injected = true;
6352 kvm_x86_ops->set_nmi(vcpu);
6353 }
c7c9c56c 6354 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
9242b5b6
BD		 6355 /*
6356 * Because interrupts can be injected asynchronously, we are
6357 * calling check_nested_events again here to avoid a race condition.
6358 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6359 * proposal and current concerns. Perhaps we should be setting
6360 * KVM_REQ_EVENT only on certain events and not unconditionally?
6361 */
6362 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6363 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6364 if (r != 0)
6365 return r;
6366 }
95ba8273 6367 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
66fd3f7f
GN		 6368 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6369 false);
6370 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 6371 }
6372 }
b6b8a145 6373 return 0;
95ba8273
GN		 6374}
6375
7460fb4a
AK		 6376static void process_nmi(struct kvm_vcpu *vcpu)
6377{
6378 unsigned limit = 2;
6379
6380 /*
6381 * x86 is limited to one NMI running, and one NMI pending after it.
6382 * If an NMI is already in progress, limit further NMIs to just one.
6383 * Otherwise, allow two (and we'll inject the first one immediately).
6384 */
6385 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6386 limit = 1;
6387
6388 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6389 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6390 kvm_make_request(KVM_REQ_EVENT, vcpu);
6391}
6392
64d60670
PB		 6393static void process_smi(struct kvm_vcpu *vcpu)
6394{
6395 if (is_smm(vcpu)) {
6396 vcpu->arch.smi_pending = true;
6397 return;
6398 }
6399
6400 printk_once(KERN_DEBUG "Ignoring guest SMI\n");
6401}
6402
3d81bc7e 6403static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
c7c9c56c
YZ		 6404{
6405 u64 eoi_exit_bitmap[4];
cf9e65b7 6406 u32 tmr[8];
c7c9c56c 6407
3d81bc7e
YZ		 6408 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6409 return;
c7c9c56c
YZ		 6410
6411 memset(eoi_exit_bitmap, 0, 32);
cf9e65b7 6412 memset(tmr, 0, 32);
c7c9c56c 6413
cf9e65b7 6414 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
c7c9c56c 6415 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
cf9e65b7 6416 kvm_apic_update_tmr(vcpu, tmr);
c7c9c56c
YZ		 6417}
6418
a70656b6
RK		 6419static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6420{
6421 ++vcpu->stat.tlb_flush;
6422 kvm_x86_ops->tlb_flush(vcpu);
6423}
6424
4256f43f
TC		 6425void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6426{
c24ae0dc
TC		 6427 struct page *page = NULL;
6428
f439ed27
PB		 6429 if (!irqchip_in_kernel(vcpu->kvm))
6430 return;
6431
4256f43f
TC		 6432 if (!kvm_x86_ops->set_apic_access_page_addr)
6433 return;
6434
c24ae0dc 6435 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
e8fd5e9e
AA		 6436 if (is_error_page(page))
6437 return;
c24ae0dc
TC		 6438 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6439
6440 /*
6441 * Do not pin apic access page in memory, the MMU notifier
6442 * will call us again if it is migrated or swapped out.
6443 */
6444 put_page(page);
4256f43f
TC		 6445}
6446EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6447
fe71557a
TC		 6448void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6449 unsigned long address)
6450{
c24ae0dc
TC		 6451 /*
6452 * The physical address of apic access page is stored in the VMCS.
6453 * Update it when it becomes invalid.
6454 */
6455 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6456 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
fe71557a
TC		 6457}
6458
9357d939 6459/*
362c698f 6460 * Returns 1 to let vcpu_run() continue the guest execution loop without
9357d939
TY		 6461 * exiting to the userspace. Otherwise, the value will be returned to the
6462 * userspace.
6463 */
851ba692 6464static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
b6c7a5dc
HB		 6465{
6466 int r;
6a8b1d13 6467 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
851ba692 6468 vcpu->run->request_interrupt_window;
730dca42 6469 bool req_immediate_exit = false;
b6c7a5dc 6470
3e007509 6471 if (vcpu->requests) {
a8eeb04a 6472 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
2e53d63a 6473 kvm_mmu_unload(vcpu);
a8eeb04a 6474 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
2f599714 6475 __kvm_migrate_timers(vcpu);
d828199e
MT		 6476 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6477 kvm_gen_update_masterclock(vcpu->kvm);
0061d53d
MT		 6478 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6479 kvm_gen_kvmclock_update(vcpu);
34c238a1
ZA		 6480 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6481 r = kvm_guest_time_update(vcpu);
8cfdc000
ZA		 6482 if (unlikely(r))
6483 goto out;
6484 }
a8eeb04a 6485 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
4731d4c7 6486 kvm_mmu_sync_roots(vcpu);
a8eeb04a 6487 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
a70656b6 6488 kvm_vcpu_flush_tlb(vcpu);
a8eeb04a 6489 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
851ba692 6490 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
b93463aa
AK		 6491 r = 0;
6492 goto out;
6493 }
a8eeb04a 6494 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
851ba692 6495 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
71c4dfaf
JR		 6496 r = 0;
6497 goto out;
6498 }
a8eeb04a 6499 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
02daab21
AK		 6500 vcpu->fpu_active = 0;
6501 kvm_x86_ops->fpu_deactivate(vcpu);
6502 }
af585b92
GN		 6503 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6504 /* Page is swapped out. Do synthetic halt */
6505 vcpu->arch.apf.halted = true;
6506 r = 1;
6507 goto out;
6508 }
c9aaa895
GC		 6509 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6510 record_steal_time(vcpu);
64d60670
PB		 6511 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6512 process_smi(vcpu);
7460fb4a
AK		 6513 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6514 process_nmi(vcpu);
f5132b01
GN		 6515 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6516 kvm_handle_pmu_event(vcpu);
6517 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6518 kvm_deliver_pmi(vcpu);
3d81bc7e
YZ		 6519 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6520 vcpu_scan_ioapic(vcpu);
4256f43f
TC		 6521 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6522 kvm_vcpu_reload_apic_access_page(vcpu);
2f52d58c 6523 }
b93463aa 6524
b463a6f7 6525 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
66450a21
JK		 6526 kvm_apic_accept_events(vcpu);
6527 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6528 r = 1;
6529 goto out;
6530 }
6531
b6b8a145
JK		 6532 if (inject_pending_event(vcpu, req_int_win) != 0)
6533 req_immediate_exit = true;
b463a6f7 6534 /* enable NMI/IRQ window open exits if needed */
b6b8a145 6535 else if (vcpu->arch.nmi_pending)
c9a7953f 6536 kvm_x86_ops->enable_nmi_window(vcpu);
c7c9c56c 6537 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
c9a7953f 6538 kvm_x86_ops->enable_irq_window(vcpu);
b463a6f7
AK		 6539
6540 if (kvm_lapic_enabled(vcpu)) {
c7c9c56c
YZ		 6541 /*
6542 * Update architecture specific hints for APIC
6543 * virtual interrupt delivery.
6544 */
6545 if (kvm_x86_ops->hwapic_irr_update)
6546 kvm_x86_ops->hwapic_irr_update(vcpu,
6547 kvm_lapic_find_highest_irr(vcpu));
b463a6f7
AK		 6548 update_cr8_intercept(vcpu);
6549 kvm_lapic_sync_to_vapic(vcpu);
6550 }
6551 }
6552
d8368af8
AK		 6553 r = kvm_mmu_reload(vcpu);
6554 if (unlikely(r)) {
d905c069 6555 goto cancel_injection;
d8368af8
AK		 6556 }
6557
b6c7a5dc
HB		 6558 preempt_disable();
6559
6560 kvm_x86_ops->prepare_guest_switch(vcpu);
2608d7a1
AK		 6561 if (vcpu->fpu_active)
6562 kvm_load_guest_fpu(vcpu);
2acf923e 6563 kvm_load_guest_xcr0(vcpu);
b6c7a5dc 6564
6b7e2d09
XG		 6565 vcpu->mode = IN_GUEST_MODE;
6566
01b71917
MT		 6567 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6568
6b7e2d09
XG		 6569 /* We should set ->mode before check ->requests,
6570 * see the comment in make_all_cpus_request.
6571 */
01b71917 6572 smp_mb__after_srcu_read_unlock();
b6c7a5dc 6573
d94e1dc9 6574 local_irq_disable();
32f88400 6575
6b7e2d09 6576 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
d94e1dc9 6577 || need_resched() || signal_pending(current)) {
6b7e2d09 6578 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 6579 smp_wmb();
6c142801
AK		 6580 local_irq_enable();
6581 preempt_enable();
01b71917 6582 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6c142801 6583 r = 1;
d905c069 6584 goto cancel_injection;
6c142801
AK		 6585 }
6586
d6185f20
NHE		 6587 if (req_immediate_exit)
6588 smp_send_reschedule(vcpu->cpu);
6589
ccf73aaf 6590 __kvm_guest_enter();
b6c7a5dc 6591
42dbaa5a 6592 if (unlikely(vcpu->arch.switch_db_regs)) {
42dbaa5a
JK		 6593 set_debugreg(0, 7);
6594 set_debugreg(vcpu->arch.eff_db[0], 0);
6595 set_debugreg(vcpu->arch.eff_db[1], 1);
6596 set_debugreg(vcpu->arch.eff_db[2], 2);
6597 set_debugreg(vcpu->arch.eff_db[3], 3);
c77fb5fe 6598 set_debugreg(vcpu->arch.dr6, 6);
ae561ede 6599 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
42dbaa5a 6600 }
b6c7a5dc 6601
229456fc 6602 trace_kvm_entry(vcpu->vcpu_id);
d0659d94 6603 wait_lapic_expire(vcpu);
851ba692 6604 kvm_x86_ops->run(vcpu);
b6c7a5dc 6605
c77fb5fe
PB		 6606 /*
6607 * Do this here before restoring debug registers on the host. And
6608 * since we do this before handling the vmexit, a DR access vmexit
6609 * can (a) read the correct value of the debug registers, (b) set
6610 * KVM_DEBUGREG_WONT_EXIT again.
6611 */
6612 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6613 int i;
6614
6615 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6616 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6617 for (i = 0; i < KVM_NR_DB_REGS; i++)
6618 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6619 }
6620
24f1e32c
FW		 6621 /*
6622 * If the guest has used debug registers, at least dr7
6623 * will be disabled while returning to the host.
6624 * If we don't have active breakpoints in the host, we don't
6625 * care about the messed up debug address registers. But if
6626 * we have some of them active, restore the old state.
6627 */
59d8eb53 6628 if (hw_breakpoint_active())
24f1e32c 6629 hw_breakpoint_restore();
42dbaa5a 6630
886b470c
MT		 6631 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6632 native_read_tsc());
1d5f066e 6633
6b7e2d09 6634 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 6635 smp_wmb();
a547c6db
YZ		 6636
6637 /* Interrupt is enabled by handle_external_intr() */
6638 kvm_x86_ops->handle_external_intr(vcpu);
b6c7a5dc
HB		 6639
6640 ++vcpu->stat.exits;
6641
6642 /*
6643 * We must have an instruction between local_irq_enable() and
6644 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6645 * the interrupt shadow. The stat.exits increment will do nicely.
6646 * But we need to prevent reordering, hence this barrier():
6647 */
6648 barrier();
6649
6650 kvm_guest_exit();
6651
6652 preempt_enable();
6653
f656ce01 6654 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3200f405 6655
b6c7a5dc
HB		 6656 /*
6657 * Profile KVM exit RIPs:
6658 */
6659 if (unlikely(prof_on == KVM_PROFILING)) {
5fdbf976
MT		 6660 unsigned long rip = kvm_rip_read(vcpu);
6661 profile_hit(KVM_PROFILING, (void *)rip);
b6c7a5dc
HB		 6662 }
6663
cc578287
ZA		 6664 if (unlikely(vcpu->arch.tsc_always_catchup))
6665 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
298101da 6666
5cfb1d5a
MT		 6667 if (vcpu->arch.apic_attention)
6668 kvm_lapic_sync_from_vapic(vcpu);
b93463aa 6669
851ba692 6670 r = kvm_x86_ops->handle_exit(vcpu);
d905c069
MT		 6671 return r;
6672
6673cancel_injection:
6674 kvm_x86_ops->cancel_injection(vcpu);
ae7a2a3f
MT		 6675 if (unlikely(vcpu->arch.apic_attention))
6676 kvm_lapic_sync_from_vapic(vcpu);
d7690175
MT		 6677out:
6678 return r;
6679}
b6c7a5dc 6680
362c698f
PB		 6681static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6682{
9c8fd1ba
PB		 6683 if (!kvm_arch_vcpu_runnable(vcpu)) {
6684 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6685 kvm_vcpu_block(vcpu);
6686 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6687 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6688 return 1;
6689 }
362c698f
PB		 6690
6691 kvm_apic_accept_events(vcpu);
6692 switch(vcpu->arch.mp_state) {
6693 case KVM_MP_STATE_HALTED:
6694 vcpu->arch.pv.pv_unhalted = false;
6695 vcpu->arch.mp_state =
6696 KVM_MP_STATE_RUNNABLE;
6697 case KVM_MP_STATE_RUNNABLE:
6698 vcpu->arch.apf.halted = false;
6699 break;
6700 case KVM_MP_STATE_INIT_RECEIVED:
6701 break;
6702 default:
6703 return -EINTR;
6704 break;
6705 }
6706 return 1;
6707}
09cec754 6708
362c698f 6709static int vcpu_run(struct kvm_vcpu *vcpu)
d7690175
MT		 6710{
6711 int r;
f656ce01 6712 struct kvm *kvm = vcpu->kvm;
d7690175 6713
f656ce01 6714 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 6715
362c698f 6716 for (;;) {
af585b92
GN		 6717 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6718 !vcpu->arch.apf.halted)
851ba692 6719 r = vcpu_enter_guest(vcpu);
362c698f
PB		 6720 else
6721 r = vcpu_block(kvm, vcpu);
09cec754
GN		 6722 if (r <= 0)
6723 break;
6724
6725 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6726 if (kvm_cpu_has_pending_timer(vcpu))
6727 kvm_inject_pending_timer_irqs(vcpu);
6728
851ba692 6729 if (dm_request_for_irq_injection(vcpu)) {
09cec754 6730 r = -EINTR;
851ba692 6731 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754 6732 ++vcpu->stat.request_irq_exits;
362c698f 6733 break;
09cec754 6734 }
af585b92
GN		 6735
6736 kvm_check_async_pf_completion(vcpu);
6737
09cec754
GN		 6738 if (signal_pending(current)) {
6739 r = -EINTR;
851ba692 6740 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754 6741 ++vcpu->stat.signal_exits;
362c698f 6742 break;
09cec754
GN		 6743 }
6744 if (need_resched()) {
f656ce01 6745 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
c08ac06a 6746 cond_resched();
f656ce01 6747 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 6748 }
b6c7a5dc
HB		 6749 }
6750
f656ce01 6751 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
b6c7a5dc
HB		 6752
6753 return r;
6754}
6755
716d51ab
GN		 6756static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6757{
6758 int r;
6759 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6760 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6761 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6762 if (r != EMULATE_DONE)
6763 return 0;
6764 return 1;
6765}
6766
6767static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6768{
6769 BUG_ON(!vcpu->arch.pio.count);
6770
6771 return complete_emulated_io(vcpu);
6772}
6773
f78146b0
AK		 6774/*
6775 * Implements the following, as a state machine:
6776 *
6777 * read:
6778 * for each fragment
87da7e66
XG		 6779 * for each mmio piece in the fragment
6780 * write gpa, len
6781 * exit
6782 * copy data
f78146b0
AK		 6783 * execute insn
6784 *
6785 * write:
6786 * for each fragment
87da7e66
XG		 6787 * for each mmio piece in the fragment
6788 * write gpa, len
6789 * copy data
6790 * exit
f78146b0 6791 */
716d51ab 6792static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
5287f194
AK		 6793{
6794 struct kvm_run *run = vcpu->run;
f78146b0 6795 struct kvm_mmio_fragment *frag;
87da7e66 6796 unsigned len;
5287f194 6797
716d51ab 6798 BUG_ON(!vcpu->mmio_needed);
5287f194 6799
716d51ab 6800 /* Complete previous fragment */
87da7e66
XG		 6801 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6802 len = min(8u, frag->len);
716d51ab 6803 if (!vcpu->mmio_is_write)
87da7e66
XG		 6804 memcpy(frag->data, run->mmio.data, len);
6805
6806 if (frag->len <= 8) {
6807 /* Switch to the next fragment. */
6808 frag++;
6809 vcpu->mmio_cur_fragment++;
6810 } else {
6811 /* Go forward to the next mmio piece. */
6812 frag->data += len;
6813 frag->gpa += len;
6814 frag->len -= len;
6815 }
6816
a08d3b3b 6817 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
716d51ab 6818 vcpu->mmio_needed = 0;
0912c977
PB		 6819
6820 /* FIXME: return into emulator if single-stepping. */
cef4dea0 6821 if (vcpu->mmio_is_write)
716d51ab
GN		 6822 return 1;
6823 vcpu->mmio_read_completed = 1;
6824 return complete_emulated_io(vcpu);
6825 }
87da7e66 6826
716d51ab
GN		 6827 run->exit_reason = KVM_EXIT_MMIO;
6828 run->mmio.phys_addr = frag->gpa;
6829 if (vcpu->mmio_is_write)
87da7e66
XG		 6830 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6831 run->mmio.len = min(8u, frag->len);
716d51ab
GN		 6832 run->mmio.is_write = vcpu->mmio_is_write;
6833 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6834 return 0;
5287f194
AK		 6835}
6836
716d51ab 6837
b6c7a5dc
HB		 6838int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6839{
6840 int r;
6841 sigset_t sigsaved;
6842
e5c30142
AK		 6843 if (!tsk_used_math(current) && init_fpu(current))
6844 return -ENOMEM;
6845
ac9f6dc0
AK		 6846 if (vcpu->sigset_active)
6847 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6848
a4535290 6849 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
b6c7a5dc 6850 kvm_vcpu_block(vcpu);
66450a21 6851 kvm_apic_accept_events(vcpu);
d7690175 6852 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
ac9f6dc0
AK		 6853 r = -EAGAIN;
6854 goto out;
b6c7a5dc
HB		 6855 }
6856
b6c7a5dc 6857 /* re-sync apic's tpr */
eea1cff9
AP		 6858 if (!irqchip_in_kernel(vcpu->kvm)) {
6859 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6860 r = -EINVAL;
6861 goto out;
6862 }
6863 }
b6c7a5dc 6864
716d51ab
GN		 6865 if (unlikely(vcpu->arch.complete_userspace_io)) {
6866 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6867 vcpu->arch.complete_userspace_io = NULL;
6868 r = cui(vcpu);
6869 if (r <= 0)
6870 goto out;
6871 } else
6872 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
5287f194 6873
362c698f 6874 r = vcpu_run(vcpu);
b6c7a5dc
HB		 6875
6876out:
f1d86e46 6877 post_kvm_run_save(vcpu);
b6c7a5dc
HB		 6878 if (vcpu->sigset_active)
6879 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6880
b6c7a5dc
HB		 6881 return r;
6882}
6883
6884int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6885{
7ae441ea
GN		 6886 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6887 /*
6888 * We are here if userspace calls get_regs() in the middle of
6889 * instruction emulation. Registers state needs to be copied
4a969980 6890 * back from emulation context to vcpu. Userspace shouldn't do
7ae441ea
GN		 6891 * that usually, but some bad designed PV devices (vmware
6892 * backdoor interface) need this to work
6893 */
dd856efa 6894 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7ae441ea
GN		 6895 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6896 }
5fdbf976
MT		 6897 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6898 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6899 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6900 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6901 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6902 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6903 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6904 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
b6c7a5dc 6905#ifdef CONFIG_X86_64
5fdbf976
MT		 6906 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6907 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6908 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6909 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6910 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6911 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6912 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6913 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
b6c7a5dc
HB		 6914#endif
6915
5fdbf976 6916 regs->rip = kvm_rip_read(vcpu);
91586a3b 6917 regs->rflags = kvm_get_rflags(vcpu);
b6c7a5dc 6918
b6c7a5dc
HB		 6919 return 0;
6920}
6921
6922int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6923{
7ae441ea
GN		 6924 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6925 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6926
5fdbf976
MT		 6927 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6928 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6929 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6930 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6931 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6932 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6933 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6934 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
b6c7a5dc 6935#ifdef CONFIG_X86_64
5fdbf976
MT		 6936 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6937 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6938 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6939 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6940 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6941 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6942 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6943 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
b6c7a5dc
HB		 6944#endif
6945
5fdbf976 6946 kvm_rip_write(vcpu, regs->rip);
91586a3b 6947 kvm_set_rflags(vcpu, regs->rflags);
b6c7a5dc 6948
b4f14abd
JK		 6949 vcpu->arch.exception.pending = false;
6950
3842d135
AK		 6951 kvm_make_request(KVM_REQ_EVENT, vcpu);
6952
b6c7a5dc
HB		 6953 return 0;
6954}
6955
b6c7a5dc
HB		 6956void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6957{
6958 struct kvm_segment cs;
6959
3e6e0aab 6960 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
b6c7a5dc
HB		 6961 *db = cs.db;
6962 *l = cs.l;
6963}
6964EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6965
6966int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6967 struct kvm_sregs *sregs)
6968{
89a27f4d 6969 struct desc_ptr dt;
b6c7a5dc 6970
3e6e0aab
GT		 6971 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6972 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6973 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6974 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6975 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6976 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 6977
3e6e0aab
GT		 6978 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6979 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc
HB		 6980
6981 kvm_x86_ops->get_idt(vcpu, &dt);
89a27f4d
GN		 6982 sregs->idt.limit = dt.size;
6983 sregs->idt.base = dt.address;
b6c7a5dc 6984 kvm_x86_ops->get_gdt(vcpu, &dt);
89a27f4d
GN		 6985 sregs->gdt.limit = dt.size;
6986 sregs->gdt.base = dt.address;
b6c7a5dc 6987
4d4ec087 6988 sregs->cr0 = kvm_read_cr0(vcpu);
ad312c7c 6989 sregs->cr2 = vcpu->arch.cr2;
9f8fe504 6990 sregs->cr3 = kvm_read_cr3(vcpu);
fc78f519 6991 sregs->cr4 = kvm_read_cr4(vcpu);
2d3ad1f4 6992 sregs->cr8 = kvm_get_cr8(vcpu);
f6801dff 6993 sregs->efer = vcpu->arch.efer;
b6c7a5dc
HB		 6994 sregs->apic_base = kvm_get_apic_base(vcpu);
6995
923c61bb 6996 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
b6c7a5dc 6997
36752c9b 6998 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
14d0bc1f
GN		 6999 set_bit(vcpu->arch.interrupt.nr,
7000 (unsigned long *)sregs->interrupt_bitmap);
16d7a191 7001
b6c7a5dc
HB		 7002 return 0;
7003}
7004
62d9f0db
MT		 7005int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7006 struct kvm_mp_state *mp_state)
7007{
66450a21 7008 kvm_apic_accept_events(vcpu);
6aef266c
SV		 7009 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7010 vcpu->arch.pv.pv_unhalted)
7011 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7012 else
7013 mp_state->mp_state = vcpu->arch.mp_state;
7014
62d9f0db
MT		 7015 return 0;
7016}
7017
7018int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7019 struct kvm_mp_state *mp_state)
7020{
66450a21
JK		 7021 if (!kvm_vcpu_has_lapic(vcpu) &&
7022 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7023 return -EINVAL;
7024
7025 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7026 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7027 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7028 } else
7029 vcpu->arch.mp_state = mp_state->mp_state;
3842d135 7030 kvm_make_request(KVM_REQ_EVENT, vcpu);
62d9f0db
MT		 7031 return 0;
7032}
7033
7f3d35fd
KW		 7034int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7035 int reason, bool has_error_code, u32 error_code)
b6c7a5dc 7036{
9d74191a 7037 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d 7038 int ret;
e01c2426 7039
8ec4722d 7040 init_emulate_ctxt(vcpu);
c697518a 7041
7f3d35fd 7042 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9d74191a 7043 has_error_code, error_code);
c697518a 7044
c697518a 7045 if (ret)
19d04437 7046 return EMULATE_FAIL;
37817f29 7047
9d74191a
TY		 7048 kvm_rip_write(vcpu, ctxt->eip);
7049 kvm_set_rflags(vcpu, ctxt->eflags);
3842d135 7050 kvm_make_request(KVM_REQ_EVENT, vcpu);
19d04437 7051 return EMULATE_DONE;
37817f29
IE		 7052}
7053EXPORT_SYMBOL_GPL(kvm_task_switch);
7054
b6c7a5dc
HB		 7055int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7056 struct kvm_sregs *sregs)
7057{
58cb628d 7058 struct msr_data apic_base_msr;
b6c7a5dc 7059 int mmu_reset_needed = 0;
63f42e02 7060 int pending_vec, max_bits, idx;
89a27f4d 7061 struct desc_ptr dt;
b6c7a5dc 7062
6d1068b3
PM		 7063 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7064 return -EINVAL;
7065
89a27f4d
GN		 7066 dt.size = sregs->idt.limit;
7067 dt.address = sregs->idt.base;
b6c7a5dc 7068 kvm_x86_ops->set_idt(vcpu, &dt);
89a27f4d
GN		 7069 dt.size = sregs->gdt.limit;
7070 dt.address = sregs->gdt.base;
b6c7a5dc
HB		 7071 kvm_x86_ops->set_gdt(vcpu, &dt);
7072
ad312c7c 7073 vcpu->arch.cr2 = sregs->cr2;
9f8fe504 7074 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
dc7e795e 7075 vcpu->arch.cr3 = sregs->cr3;
aff48baa 7076 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
b6c7a5dc 7077
2d3ad1f4 7078 kvm_set_cr8(vcpu, sregs->cr8);
b6c7a5dc 7079
f6801dff 7080 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
b6c7a5dc 7081 kvm_x86_ops->set_efer(vcpu, sregs->efer);
58cb628d
JK		 7082 apic_base_msr.data = sregs->apic_base;
7083 apic_base_msr.host_initiated = true;
7084 kvm_set_apic_base(vcpu, &apic_base_msr);
b6c7a5dc 7085
4d4ec087 7086 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
b6c7a5dc 7087 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
d7306163 7088 vcpu->arch.cr0 = sregs->cr0;
b6c7a5dc 7089
fc78f519 7090 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
b6c7a5dc 7091 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3ea3aa8c 7092 if (sregs->cr4 & X86_CR4_OSXSAVE)
00b27a3e 7093 kvm_update_cpuid(vcpu);
63f42e02
XG		 7094
7095 idx = srcu_read_lock(&vcpu->kvm->srcu);
7c93be44 7096 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
9f8fe504 7097 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7c93be44
MT		 7098 mmu_reset_needed = 1;
7099 }
63f42e02 7100 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b6c7a5dc
HB		 7101
7102 if (mmu_reset_needed)
7103 kvm_mmu_reset_context(vcpu);
7104
a50abc3b 7105 max_bits = KVM_NR_INTERRUPTS;
923c61bb
GN		 7106 pending_vec = find_first_bit(
7107 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7108 if (pending_vec < max_bits) {
66fd3f7f 7109 kvm_queue_interrupt(vcpu, pending_vec, false);
923c61bb 7110 pr_debug("Set back pending irq %d\n", pending_vec);
b6c7a5dc
HB		 7111 }
7112
3e6e0aab
GT		 7113 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7114 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7115 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7116 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7117 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7118 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 7119
3e6e0aab
GT		 7120 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7121 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc 7122
5f0269f5
ME		 7123 update_cr8_intercept(vcpu);
7124
9c3e4aab 7125 /* Older userspace won't unhalt the vcpu on reset. */
c5af89b6 7126 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9c3e4aab 7127 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3eeb3288 7128 !is_protmode(vcpu))
9c3e4aab
MT		 7129 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7130
3842d135
AK		 7131 kvm_make_request(KVM_REQ_EVENT, vcpu);
7132
b6c7a5dc
HB		 7133 return 0;
7134}
7135
d0bfb940
JK		 7136int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7137 struct kvm_guest_debug *dbg)
b6c7a5dc 7138{
355be0b9 7139 unsigned long rflags;
ae675ef0 7140 int i, r;
b6c7a5dc 7141
4f926bf2
JK		 7142 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7143 r = -EBUSY;
7144 if (vcpu->arch.exception.pending)
2122ff5e 7145 goto out;
4f926bf2
JK		 7146 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7147 kvm_queue_exception(vcpu, DB_VECTOR);
7148 else
7149 kvm_queue_exception(vcpu, BP_VECTOR);
7150 }
7151
91586a3b
JK		 7152 /*
7153 * Read rflags as long as potentially injected trace flags are still
7154 * filtered out.
7155 */
7156 rflags = kvm_get_rflags(vcpu);
355be0b9
JK		 7157
7158 vcpu->guest_debug = dbg->control;
7159 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7160 vcpu->guest_debug = 0;
7161
7162 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
ae675ef0
JK		 7163 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7164 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
c8639010 7165 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
ae675ef0
JK		 7166 } else {
7167 for (i = 0; i < KVM_NR_DB_REGS; i++)
7168 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
ae675ef0 7169 }
c8639010 7170 kvm_update_dr7(vcpu);
ae675ef0 7171
f92653ee
JK		 7172 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7173 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7174 get_segment_base(vcpu, VCPU_SREG_CS);
94fe45da 7175
91586a3b
JK		 7176 /*
7177 * Trigger an rflags update that will inject or remove the trace
7178 * flags.
7179 */
7180 kvm_set_rflags(vcpu, rflags);
b6c7a5dc 7181
c8639010 7182 kvm_x86_ops->update_db_bp_intercept(vcpu);
b6c7a5dc 7183
4f926bf2 7184 r = 0;
d0bfb940 7185
2122ff5e 7186out:
b6c7a5dc
HB		 7187
7188 return r;
7189}
7190
8b006791
ZX		 7191/*
7192 * Translate a guest virtual address to a guest physical address.
7193 */
7194int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7195 struct kvm_translation *tr)
7196{
7197 unsigned long vaddr = tr->linear_address;
7198 gpa_t gpa;
f656ce01 7199 int idx;
8b006791 7200
f656ce01 7201 idx = srcu_read_lock(&vcpu->kvm->srcu);
1871c602 7202 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
f656ce01 7203 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8b006791
ZX		 7204 tr->physical_address = gpa;
7205 tr->valid = gpa != UNMAPPED_GVA;
7206 tr->writeable = 1;
7207 tr->usermode = 0;
8b006791
ZX		 7208
7209 return 0;
7210}
7211
d0752060
HB		 7212int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7213{
98918833
SY		 7214 struct i387_fxsave_struct *fxsave =
7215 &vcpu->arch.guest_fpu.state->fxsave;
d0752060 7216
d0752060
HB		 7217 memcpy(fpu->fpr, fxsave->st_space, 128);
7218 fpu->fcw = fxsave->cwd;
7219 fpu->fsw = fxsave->swd;
7220 fpu->ftwx = fxsave->twd;
7221 fpu->last_opcode = fxsave->fop;
7222 fpu->last_ip = fxsave->rip;
7223 fpu->last_dp = fxsave->rdp;
7224 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7225
d0752060
HB		 7226 return 0;
7227}
7228
7229int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7230{
98918833
SY		 7231 struct i387_fxsave_struct *fxsave =
7232 &vcpu->arch.guest_fpu.state->fxsave;
d0752060 7233
d0752060
HB		 7234 memcpy(fxsave->st_space, fpu->fpr, 128);
7235 fxsave->cwd = fpu->fcw;
7236 fxsave->swd = fpu->fsw;
7237 fxsave->twd = fpu->ftwx;
7238 fxsave->fop = fpu->last_opcode;
7239 fxsave->rip = fpu->last_ip;
7240 fxsave->rdp = fpu->last_dp;
7241 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7242
d0752060
HB		 7243 return 0;
7244}
7245
d28bc9dd 7246int fx_init(struct kvm_vcpu *vcpu, bool init_event)
d0752060 7247{
10ab25cd
JK		 7248 int err;
7249
7250 err = fpu_alloc(&vcpu->arch.guest_fpu);
7251 if (err)
7252 return err;
7253
d28bc9dd
NA		 7254 if (!init_event)
7255 fpu_finit(&vcpu->arch.guest_fpu);
7256
df1daba7
PB		 7257 if (cpu_has_xsaves)
7258 vcpu->arch.guest_fpu.state->xsave.xsave_hdr.xcomp_bv =
7259 host_xcr0 | XSTATE_COMPACTION_ENABLED;
d0752060 7260
2acf923e
DC		 7261 /*
7262 * Ensure guest xcr0 is valid for loading
7263 */
7264 vcpu->arch.xcr0 = XSTATE_FP;
7265
ad312c7c 7266 vcpu->arch.cr0 |= X86_CR0_ET;
10ab25cd
JK		 7267
7268 return 0;
d0752060
HB		 7269}
7270EXPORT_SYMBOL_GPL(fx_init);
7271
98918833
SY		 7272static void fx_free(struct kvm_vcpu *vcpu)
7273{
7274 fpu_free(&vcpu->arch.guest_fpu);
7275}
7276
d0752060
HB		 7277void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7278{
2608d7a1 7279 if (vcpu->guest_fpu_loaded)
d0752060
HB		 7280 return;
7281
2acf923e
DC		 7282 /*
7283 * Restore all possible states in the guest,
7284 * and assume host would use all available bits.
7285 * Guest xcr0 would be loaded later.
7286 */
7287 kvm_put_guest_xcr0(vcpu);
d0752060 7288 vcpu->guest_fpu_loaded = 1;
b1a74bf8 7289 __kernel_fpu_begin();
98918833 7290 fpu_restore_checking(&vcpu->arch.guest_fpu);
0c04851c 7291 trace_kvm_fpu(1);
d0752060 7292}
d0752060
HB		 7293
7294void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7295{
2acf923e
DC		 7296 kvm_put_guest_xcr0(vcpu);
7297
653f52c3
RR		 7298 if (!vcpu->guest_fpu_loaded) {
7299 vcpu->fpu_counter = 0;
d0752060 7300 return;
653f52c3 7301 }
d0752060
HB		 7302
7303 vcpu->guest_fpu_loaded = 0;
98918833 7304 fpu_save_init(&vcpu->arch.guest_fpu);
b1a74bf8 7305 __kernel_fpu_end();
f096ed85 7306 ++vcpu->stat.fpu_reload;
653f52c3
RR		 7307 /*
7308 * If using eager FPU mode, or if the guest is a frequent user
7309 * of the FPU, just leave the FPU active for next time.
7310 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7311 * the FPU in bursts will revert to loading it on demand.
7312 */
a9b4fb7e 7313 if (!vcpu->arch.eager_fpu) {
653f52c3
RR		 7314 if (++vcpu->fpu_counter < 5)
7315 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7316 }
0c04851c 7317 trace_kvm_fpu(0);
d0752060 7318}
e9b11c17
ZX		 7319
7320void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7321{
12f9a48f 7322 kvmclock_reset(vcpu);
7f1ea208 7323
f5f48ee1 7324 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
98918833 7325 fx_free(vcpu);
e9b11c17
ZX		 7326 kvm_x86_ops->vcpu_free(vcpu);
7327}
7328
7329struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7330 unsigned int id)
7331{
c447e76b
LL		 7332 struct kvm_vcpu *vcpu;
7333
6755bae8
ZA		 7334 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7335 printk_once(KERN_WARNING
7336 "kvm: SMP vm created on host with unstable TSC; "
7337 "guest TSC will not be reliable\n");
c447e76b
LL		 7338
7339 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7340
7341 /*
7342 * Activate fpu unconditionally in case the guest needs eager FPU. It will be
7343 * deactivated soon if it doesn't.
7344 */
7345 kvm_x86_ops->fpu_activate(vcpu);
7346 return vcpu;
26e5215f 7347}
e9b11c17 7348
26e5215f
AK		 7349int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7350{
7351 int r;
e9b11c17 7352
0bed3b56 7353 vcpu->arch.mtrr_state.have_fixed = 1;
9fc77441
MT		 7354 r = vcpu_load(vcpu);
7355 if (r)
7356 return r;
d28bc9dd 7357 kvm_vcpu_reset(vcpu, false);
8a3c1a33 7358 kvm_mmu_setup(vcpu);
e9b11c17 7359 vcpu_put(vcpu);
e9b11c17 7360
26e5215f 7361 return r;
e9b11c17
ZX		 7362}
7363
31928aa5 7364void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
42897d86 7365{
8fe8ab46 7366 struct msr_data msr;
332967a3 7367 struct kvm *kvm = vcpu->kvm;
42897d86 7368
31928aa5
DD		 7369 if (vcpu_load(vcpu))
7370 return;
8fe8ab46
WA		 7371 msr.data = 0x0;
7372 msr.index = MSR_IA32_TSC;
7373 msr.host_initiated = true;
7374 kvm_write_tsc(vcpu, &msr);
42897d86
MT		 7375 vcpu_put(vcpu);
7376
630994b3
MT		 7377 if (!kvmclock_periodic_sync)
7378 return;
7379
332967a3
AJ		 7380 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7381 KVMCLOCK_SYNC_PERIOD);
42897d86
MT		 7382}
7383
d40ccc62 7384void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
e9b11c17 7385{
9fc77441 7386 int r;
344d9588
GN		 7387 vcpu->arch.apf.msr_val = 0;
7388
9fc77441
MT		 7389 r = vcpu_load(vcpu);
7390 BUG_ON(r);
e9b11c17
ZX		 7391 kvm_mmu_unload(vcpu);
7392 vcpu_put(vcpu);
7393
98918833 7394 fx_free(vcpu);
e9b11c17
ZX		 7395 kvm_x86_ops->vcpu_free(vcpu);
7396}
7397
d28bc9dd 7398void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
e9b11c17 7399{
e69fab5d
PB		 7400 vcpu->arch.hflags = 0;
7401
7460fb4a
AK		 7402 atomic_set(&vcpu->arch.nmi_queued, 0);
7403 vcpu->arch.nmi_pending = 0;
448fa4a9 7404 vcpu->arch.nmi_injected = false;
5f7552d4
NA		 7405 kvm_clear_interrupt_queue(vcpu);
7406 kvm_clear_exception_queue(vcpu);
448fa4a9 7407
42dbaa5a 7408 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
ae561ede 7409 kvm_update_dr0123(vcpu);
6f43ed01 7410 vcpu->arch.dr6 = DR6_INIT;
73aaf249 7411 kvm_update_dr6(vcpu);
42dbaa5a 7412 vcpu->arch.dr7 = DR7_FIXED_1;
c8639010 7413 kvm_update_dr7(vcpu);
42dbaa5a 7414
1119022c
NA		 7415 vcpu->arch.cr2 = 0;
7416
3842d135 7417 kvm_make_request(KVM_REQ_EVENT, vcpu);
344d9588 7418 vcpu->arch.apf.msr_val = 0;
c9aaa895 7419 vcpu->arch.st.msr_val = 0;
3842d135 7420
12f9a48f
GC		 7421 kvmclock_reset(vcpu);
7422
af585b92
GN		 7423 kvm_clear_async_pf_completion_queue(vcpu);
7424 kvm_async_pf_hash_reset(vcpu);
7425 vcpu->arch.apf.halted = false;
3842d135 7426
64d60670 7427 if (!init_event) {
d28bc9dd 7428 kvm_pmu_reset(vcpu);
64d60670
PB		 7429 vcpu->arch.smbase = 0x30000;
7430 }
f5132b01 7431
66f7b72e
JS		 7432 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7433 vcpu->arch.regs_avail = ~0;
7434 vcpu->arch.regs_dirty = ~0;
7435
d28bc9dd 7436 kvm_x86_ops->vcpu_reset(vcpu, init_event);
e9b11c17
ZX		 7437}
7438
2b4a273b 7439void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
66450a21
JK		 7440{
7441 struct kvm_segment cs;
7442
7443 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7444 cs.selector = vector << 8;
7445 cs.base = vector << 12;
7446 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7447 kvm_rip_write(vcpu, 0);
e9b11c17
ZX		 7448}
7449
13a34e06 7450int kvm_arch_hardware_enable(void)
e9b11c17 7451{
ca84d1a2
ZA		 7452 struct kvm *kvm;
7453 struct kvm_vcpu *vcpu;
7454 int i;
0dd6a6ed
ZA		 7455 int ret;
7456 u64 local_tsc;
7457 u64 max_tsc = 0;
7458 bool stable, backwards_tsc = false;
18863bdd
AK		 7459
7460 kvm_shared_msr_cpu_online();
13a34e06 7461 ret = kvm_x86_ops->hardware_enable();
0dd6a6ed
ZA		 7462 if (ret != 0)
7463 return ret;
7464
7465 local_tsc = native_read_tsc();
7466 stable = !check_tsc_unstable();
7467 list_for_each_entry(kvm, &vm_list, vm_list) {
7468 kvm_for_each_vcpu(i, vcpu, kvm) {
7469 if (!stable && vcpu->cpu == smp_processor_id())
105b21bb 7470 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 7471 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7472 backwards_tsc = true;
7473 if (vcpu->arch.last_host_tsc > max_tsc)
7474 max_tsc = vcpu->arch.last_host_tsc;
7475 }
7476 }
7477 }
7478
7479 /*
7480 * Sometimes, even reliable TSCs go backwards. This happens on
7481 * platforms that reset TSC during suspend or hibernate actions, but
7482 * maintain synchronization. We must compensate. Fortunately, we can
7483 * detect that condition here, which happens early in CPU bringup,
7484 * before any KVM threads can be running. Unfortunately, we can't
7485 * bring the TSCs fully up to date with real time, as we aren't yet far
7486 * enough into CPU bringup that we know how much real time has actually
7487 * elapsed; our helper function, get_kernel_ns() will be using boot
7488 * variables that haven't been updated yet.
7489 *
7490 * So we simply find the maximum observed TSC above, then record the
7491 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7492 * the adjustment will be applied. Note that we accumulate
7493 * adjustments, in case multiple suspend cycles happen before some VCPU
7494 * gets a chance to run again. In the event that no KVM threads get a
7495 * chance to run, we will miss the entire elapsed period, as we'll have
7496 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7497 * loose cycle time. This isn't too big a deal, since the loss will be
7498 * uniform across all VCPUs (not to mention the scenario is extremely
7499 * unlikely). It is possible that a second hibernate recovery happens
7500 * much faster than a first, causing the observed TSC here to be
7501 * smaller; this would require additional padding adjustment, which is
7502 * why we set last_host_tsc to the local tsc observed here.
7503 *
7504 * N.B. - this code below runs only on platforms with reliable TSC,
7505 * as that is the only way backwards_tsc is set above. Also note
7506 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7507 * have the same delta_cyc adjustment applied if backwards_tsc
7508 * is detected. Note further, this adjustment is only done once,
7509 * as we reset last_host_tsc on all VCPUs to stop this from being
7510 * called multiple times (one for each physical CPU bringup).
7511 *
4a969980 7512 * Platforms with unreliable TSCs don't have to deal with this, they
0dd6a6ed
ZA		 7513 * will be compensated by the logic in vcpu_load, which sets the TSC to
7514 * catchup mode. This will catchup all VCPUs to real time, but cannot
7515 * guarantee that they stay in perfect synchronization.
7516 */
7517 if (backwards_tsc) {
7518 u64 delta_cyc = max_tsc - local_tsc;
16a96021 7519 backwards_tsc_observed = true;
0dd6a6ed
ZA		 7520 list_for_each_entry(kvm, &vm_list, vm_list) {
7521 kvm_for_each_vcpu(i, vcpu, kvm) {
7522 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7523 vcpu->arch.last_host_tsc = local_tsc;
105b21bb 7524 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 7525 }
7526
7527 /*
7528 * We have to disable TSC offset matching.. if you were
7529 * booting a VM while issuing an S4 host suspend....
7530 * you may have some problem. Solving this issue is
7531 * left as an exercise to the reader.
7532 */
7533 kvm->arch.last_tsc_nsec = 0;
7534 kvm->arch.last_tsc_write = 0;
7535 }
7536
7537 }
7538 return 0;
e9b11c17
ZX		 7539}
7540
13a34e06 7541void kvm_arch_hardware_disable(void)
e9b11c17 7542{
13a34e06
RK		 7543 kvm_x86_ops->hardware_disable();
7544 drop_user_return_notifiers();
e9b11c17
ZX		 7545}
7546
7547int kvm_arch_hardware_setup(void)
7548{
9e9c3fe4
NA		 7549 int r;
7550
7551 r = kvm_x86_ops->hardware_setup();
7552 if (r != 0)
7553 return r;
7554
7555 kvm_init_msr_list();
7556 return 0;
e9b11c17
ZX		 7557}
7558
7559void kvm_arch_hardware_unsetup(void)
7560{
7561 kvm_x86_ops->hardware_unsetup();
7562}
7563
7564void kvm_arch_check_processor_compat(void *rtn)
7565{
7566 kvm_x86_ops->check_processor_compatibility(rtn);
7567}
7568
3e515705
AK		 7569bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7570{
7571 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7572}
7573
54e9818f
GN		 7574struct static_key kvm_no_apic_vcpu __read_mostly;
7575
e9b11c17
ZX		 7576int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7577{
7578 struct page *page;
7579 struct kvm *kvm;
7580 int r;
7581
7582 BUG_ON(vcpu->kvm == NULL);
7583 kvm = vcpu->kvm;
7584
6aef266c 7585 vcpu->arch.pv.pv_unhalted = false;
9aabc88f 7586 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
58d269d8 7587 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
a4535290 7588 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
e9b11c17 7589 else
a4535290 7590 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
e9b11c17
ZX		 7591
7592 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7593 if (!page) {
7594 r = -ENOMEM;
7595 goto fail;
7596 }
ad312c7c 7597 vcpu->arch.pio_data = page_address(page);
e9b11c17 7598
cc578287 7599 kvm_set_tsc_khz(vcpu, max_tsc_khz);
c285545f 7600
e9b11c17
ZX		 7601 r = kvm_mmu_create(vcpu);
7602 if (r < 0)
7603 goto fail_free_pio_data;
7604
7605 if (irqchip_in_kernel(kvm)) {
7606 r = kvm_create_lapic(vcpu);
7607 if (r < 0)
7608 goto fail_mmu_destroy;
54e9818f
GN		 7609 } else
7610 static_key_slow_inc(&kvm_no_apic_vcpu);
e9b11c17 7611
890ca9ae
HY		 7612 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7613 GFP_KERNEL);
7614 if (!vcpu->arch.mce_banks) {
7615 r = -ENOMEM;
443c39bc 7616 goto fail_free_lapic;
890ca9ae
HY		 7617 }
7618 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7619
f1797359
WY		 7620 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7621 r = -ENOMEM;
f5f48ee1 7622 goto fail_free_mce_banks;
f1797359 7623 }
f5f48ee1 7624
d28bc9dd 7625 r = fx_init(vcpu, false);
66f7b72e
JS		 7626 if (r)
7627 goto fail_free_wbinvd_dirty_mask;
7628
ba904635 7629 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
0b79459b 7630 vcpu->arch.pv_time_enabled = false;
d7876f1b
PB		 7631
7632 vcpu->arch.guest_supported_xcr0 = 0;
4344ee98 7633 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
d7876f1b 7634
5a4f55cd
EK		 7635 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7636
74545705
RK		 7637 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7638
af585b92 7639 kvm_async_pf_hash_reset(vcpu);
f5132b01 7640 kvm_pmu_init(vcpu);
af585b92 7641
e9b11c17 7642 return 0;
66f7b72e
JS		 7643fail_free_wbinvd_dirty_mask:
7644 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
f5f48ee1
SY		 7645fail_free_mce_banks:
7646 kfree(vcpu->arch.mce_banks);
443c39bc
WY		 7647fail_free_lapic:
7648 kvm_free_lapic(vcpu);
e9b11c17
ZX		 7649fail_mmu_destroy:
7650 kvm_mmu_destroy(vcpu);
7651fail_free_pio_data:
ad312c7c 7652 free_page((unsigned long)vcpu->arch.pio_data);
e9b11c17
ZX		 7653fail:
7654 return r;
7655}
7656
7657void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7658{
f656ce01
MT		 7659 int idx;
7660
f5132b01 7661 kvm_pmu_destroy(vcpu);
36cb93fd 7662 kfree(vcpu->arch.mce_banks);
e9b11c17 7663 kvm_free_lapic(vcpu);
f656ce01 7664 idx = srcu_read_lock(&vcpu->kvm->srcu);
e9b11c17 7665 kvm_mmu_destroy(vcpu);
f656ce01 7666 srcu_read_unlock(&vcpu->kvm->srcu, idx);
ad312c7c 7667 free_page((unsigned long)vcpu->arch.pio_data);
54e9818f
GN		 7668 if (!irqchip_in_kernel(vcpu->kvm))
7669 static_key_slow_dec(&kvm_no_apic_vcpu);
e9b11c17 7670}
d19a9cd2 7671
e790d9ef
RK		 7672void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7673{
ae97a3b8 7674 kvm_x86_ops->sched_in(vcpu, cpu);
e790d9ef
RK		 7675}
7676
e08b9637 7677int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
d19a9cd2 7678{
e08b9637
CO		 7679 if (type)
7680 return -EINVAL;
7681
6ef768fa 7682 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
f05e70ac 7683 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
365c8868 7684 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
4d5c5d0f 7685 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
e0f0bbc5 7686 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
d19a9cd2 7687
5550af4d
SY		 7688 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7689 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7a84428a
AW		 7690 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7691 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7692 &kvm->arch.irq_sources_bitmap);
5550af4d 7693
038f8c11 7694 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
1e08ec4a 7695 mutex_init(&kvm->arch.apic_map_lock);
d828199e
MT		 7696 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7697
7698 pvclock_update_vm_gtod_copy(kvm);
53f658b3 7699
7e44e449 7700 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
332967a3 7701 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7e44e449 7702
d89f5eff 7703 return 0;
d19a9cd2
ZX		 7704}
7705
7706static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7707{
9fc77441
MT		 7708 int r;
7709 r = vcpu_load(vcpu);
7710 BUG_ON(r);
d19a9cd2
ZX		 7711 kvm_mmu_unload(vcpu);
7712 vcpu_put(vcpu);
7713}
7714
7715static void kvm_free_vcpus(struct kvm *kvm)
7716{
7717 unsigned int i;
988a2cae 7718 struct kvm_vcpu *vcpu;
d19a9cd2
ZX		 7719
7720 /*
7721 * Unpin any mmu pages first.
7722 */
af585b92
GN		 7723 kvm_for_each_vcpu(i, vcpu, kvm) {
7724 kvm_clear_async_pf_completion_queue(vcpu);
988a2cae 7725 kvm_unload_vcpu_mmu(vcpu);
af585b92 7726 }
988a2cae
GN		 7727 kvm_for_each_vcpu(i, vcpu, kvm)
7728 kvm_arch_vcpu_free(vcpu);
7729
7730 mutex_lock(&kvm->lock);
7731 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7732 kvm->vcpus[i] = NULL;
d19a9cd2 7733
988a2cae
GN		 7734 atomic_set(&kvm->online_vcpus, 0);
7735 mutex_unlock(&kvm->lock);
d19a9cd2
ZX		 7736}
7737
ad8ba2cd
SY		 7738void kvm_arch_sync_events(struct kvm *kvm)
7739{
332967a3 7740 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7e44e449 7741 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
ba4cef31 7742 kvm_free_all_assigned_devices(kvm);
aea924f6 7743 kvm_free_pit(kvm);
ad8ba2cd
SY		 7744}
7745
d19a9cd2
ZX		 7746void kvm_arch_destroy_vm(struct kvm *kvm)
7747{
27469d29
AH		 7748 if (current->mm == kvm->mm) {
7749 /*
7750 * Free memory regions allocated on behalf of userspace,
7751 * unless the the memory map has changed due to process exit
7752 * or fd copying.
7753 */
7754 struct kvm_userspace_memory_region mem;
7755 memset(&mem, 0, sizeof(mem));
7756 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7757 kvm_set_memory_region(kvm, &mem);
7758
7759 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7760 kvm_set_memory_region(kvm, &mem);
7761
7762 mem.slot = TSS_PRIVATE_MEMSLOT;
7763 kvm_set_memory_region(kvm, &mem);
7764 }
6eb55818 7765 kvm_iommu_unmap_guest(kvm);
d7deeeb0
ZX		 7766 kfree(kvm->arch.vpic);
7767 kfree(kvm->arch.vioapic);
d19a9cd2 7768 kvm_free_vcpus(kvm);
1e08ec4a 7769 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
d19a9cd2 7770}
0de10343 7771
5587027c 7772void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
db3fe4eb
TY		 7773 struct kvm_memory_slot *dont)
7774{
7775 int i;
7776
d89cc617
TY		 7777 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7778 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
548ef284 7779 kvfree(free->arch.rmap[i]);
d89cc617 7780 free->arch.rmap[i] = NULL;
77d11309 7781 }
d89cc617
TY		 7782 if (i == 0)
7783 continue;
7784
7785 if (!dont || free->arch.lpage_info[i - 1] !=
7786 dont->arch.lpage_info[i - 1]) {
548ef284 7787 kvfree(free->arch.lpage_info[i - 1]);
d89cc617 7788 free->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 7789 }
7790 }
7791}
7792
5587027c
AK		 7793int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7794 unsigned long npages)
db3fe4eb
TY		 7795{
7796 int i;
7797
d89cc617 7798 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
db3fe4eb
TY		 7799 unsigned long ugfn;
7800 int lpages;
d89cc617 7801 int level = i + 1;
db3fe4eb
TY		 7802
7803 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7804 slot->base_gfn, level) + 1;
7805
d89cc617
TY		 7806 slot->arch.rmap[i] =
7807 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7808 if (!slot->arch.rmap[i])
77d11309 7809 goto out_free;
d89cc617
TY		 7810 if (i == 0)
7811 continue;
77d11309 7812
d89cc617
TY		 7813 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7814 sizeof(*slot->arch.lpage_info[i - 1]));
7815 if (!slot->arch.lpage_info[i - 1])
db3fe4eb
TY		 7816 goto out_free;
7817
7818 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
d89cc617 7819 slot->arch.lpage_info[i - 1][0].write_count = 1;
db3fe4eb 7820 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
d89cc617 7821 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
db3fe4eb
TY		 7822 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7823 /*
7824 * If the gfn and userspace address are not aligned wrt each
7825 * other, or if explicitly asked to, disable large page
7826 * support for this slot
7827 */
7828 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7829 !kvm_largepages_enabled()) {
7830 unsigned long j;
7831
7832 for (j = 0; j < lpages; ++j)
d89cc617 7833 slot->arch.lpage_info[i - 1][j].write_count = 1;
db3fe4eb
TY		 7834 }
7835 }
7836
7837 return 0;
7838
7839out_free:
d89cc617 7840 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
548ef284 7841 kvfree(slot->arch.rmap[i]);
d89cc617
TY		 7842 slot->arch.rmap[i] = NULL;
7843 if (i == 0)
7844 continue;
7845
548ef284 7846 kvfree(slot->arch.lpage_info[i - 1]);
d89cc617 7847 slot->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 7848 }
7849 return -ENOMEM;
7850}
7851
15f46015 7852void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
e59dbe09 7853{
e6dff7d1
TY		 7854 /*
7855 * memslots->generation has been incremented.
7856 * mmio generation may have reached its maximum value.
7857 */
7858 kvm_mmu_invalidate_mmio_sptes(kvm);
e59dbe09
TY		 7859}
7860
f7784b8e
MT		 7861int kvm_arch_prepare_memory_region(struct kvm *kvm,
7862 struct kvm_memory_slot *memslot,
09170a49 7863 const struct kvm_userspace_memory_region *mem,
7b6195a9 7864 enum kvm_mr_change change)
0de10343 7865{
7a905b14
TY		 7866 /*
7867 * Only private memory slots need to be mapped here since
7868 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
0de10343 7869 */
7b6195a9 7870 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7a905b14 7871 unsigned long userspace_addr;
604b38ac 7872
7a905b14
TY		 7873 /*
7874 * MAP_SHARED to prevent internal slot pages from being moved
7875 * by fork()/COW.
7876 */
7b6195a9 7877 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7a905b14
TY		 7878 PROT_READ | PROT_WRITE,
7879 MAP_SHARED | MAP_ANONYMOUS, 0);
0de10343 7880
7a905b14
TY		 7881 if (IS_ERR((void *)userspace_addr))
7882 return PTR_ERR((void *)userspace_addr);
604b38ac 7883
7a905b14 7884 memslot->userspace_addr = userspace_addr;
0de10343
ZX		 7885 }
7886
f7784b8e
MT		 7887 return 0;
7888}
7889
88178fd4
KH		 7890static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7891 struct kvm_memory_slot *new)
7892{
7893 /* Still write protect RO slot */
7894 if (new->flags & KVM_MEM_READONLY) {
7895 kvm_mmu_slot_remove_write_access(kvm, new);
7896 return;
7897 }
7898
7899 /*
7900 * Call kvm_x86_ops dirty logging hooks when they are valid.
7901 *
7902 * kvm_x86_ops->slot_disable_log_dirty is called when:
7903 *
7904 * - KVM_MR_CREATE with dirty logging is disabled
7905 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7906 *
7907 * The reason is, in case of PML, we need to set D-bit for any slots
7908 * with dirty logging disabled in order to eliminate unnecessary GPA
7909 * logging in PML buffer (and potential PML buffer full VMEXT). This
7910 * guarantees leaving PML enabled during guest's lifetime won't have
7911 * any additonal overhead from PML when guest is running with dirty
7912 * logging disabled for memory slots.
7913 *
7914 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7915 * to dirty logging mode.
7916 *
7917 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7918 *
7919 * In case of write protect:
7920 *
7921 * Write protect all pages for dirty logging.
7922 *
7923 * All the sptes including the large sptes which point to this
7924 * slot are set to readonly. We can not create any new large
7925 * spte on this slot until the end of the logging.
7926 *
7927 * See the comments in fast_page_fault().
7928 */
7929 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7930 if (kvm_x86_ops->slot_enable_log_dirty)
7931 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7932 else
7933 kvm_mmu_slot_remove_write_access(kvm, new);
7934 } else {
7935 if (kvm_x86_ops->slot_disable_log_dirty)
7936 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7937 }
7938}
7939
f7784b8e 7940void kvm_arch_commit_memory_region(struct kvm *kvm,
09170a49 7941 const struct kvm_userspace_memory_region *mem,
8482644a 7942 const struct kvm_memory_slot *old,
f36f3f28 7943 const struct kvm_memory_slot *new,
8482644a 7944 enum kvm_mr_change change)
f7784b8e 7945{
8482644a 7946 int nr_mmu_pages = 0;
f7784b8e 7947
f36f3f28 7948 if (change == KVM_MR_DELETE && old->id >= KVM_USER_MEM_SLOTS) {
f7784b8e
MT		 7949 int ret;
7950
8482644a
TY		 7951 ret = vm_munmap(old->userspace_addr,
7952 old->npages * PAGE_SIZE);
f7784b8e
MT		 7953 if (ret < 0)
7954 printk(KERN_WARNING
7955 "kvm_vm_ioctl_set_memory_region: "
7956 "failed to munmap memory\n");
7957 }
7958
48c0e4e9
XG		 7959 if (!kvm->arch.n_requested_mmu_pages)
7960 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7961
48c0e4e9 7962 if (nr_mmu_pages)
0de10343 7963 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
1c91cad4 7964
3ea3b7fa
WL		 7965 /*
7966 * Dirty logging tracks sptes in 4k granularity, meaning that large
7967 * sptes have to be split. If live migration is successful, the guest
7968 * in the source machine will be destroyed and large sptes will be
7969 * created in the destination. However, if the guest continues to run
7970 * in the source machine (for example if live migration fails), small
7971 * sptes will remain around and cause bad performance.
7972 *
7973 * Scan sptes if dirty logging has been stopped, dropping those
7974 * which can be collapsed into a single large-page spte. Later
7975 * page faults will create the large-page sptes.
7976 */
7977 if ((change != KVM_MR_DELETE) &&
7978 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7979 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7980 kvm_mmu_zap_collapsible_sptes(kvm, new);
7981
c972f3b1 7982 /*
88178fd4 7983 * Set up write protection and/or dirty logging for the new slot.
c126d94f 7984 *
88178fd4
KH		 7985 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7986 * been zapped so no dirty logging staff is needed for old slot. For
7987 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7988 * new and it's also covered when dealing with the new slot.
f36f3f28
PB		 7989 *
7990 * FIXME: const-ify all uses of struct kvm_memory_slot.
c972f3b1 7991 */
88178fd4 7992 if (change != KVM_MR_DELETE)
f36f3f28 7993 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
0de10343 7994}
1d737c8a 7995
2df72e9b 7996void kvm_arch_flush_shadow_all(struct kvm *kvm)
34d4cb8f 7997{
6ca18b69 7998 kvm_mmu_invalidate_zap_all_pages(kvm);
34d4cb8f
MT		 7999}
8000
2df72e9b
MT		 8001void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8002 struct kvm_memory_slot *slot)
8003{
6ca18b69 8004 kvm_mmu_invalidate_zap_all_pages(kvm);
2df72e9b
MT		 8005}
8006
1d737c8a
ZX		 8007int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8008{
b6b8a145
JK		 8009 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8010 kvm_x86_ops->check_nested_events(vcpu, false);
8011
af585b92
GN		 8012 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8013 !vcpu->arch.apf.halted)
8014 || !list_empty_careful(&vcpu->async_pf.done)
66450a21 8015 || kvm_apic_has_events(vcpu)
6aef266c 8016 || vcpu->arch.pv.pv_unhalted
7460fb4a 8017 || atomic_read(&vcpu->arch.nmi_queued) ||
a1b37100
GN		 8018 (kvm_arch_interrupt_allowed(vcpu) &&
8019 kvm_cpu_has_interrupt(vcpu));
1d737c8a 8020}
5736199a 8021
b6d33834 8022int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
5736199a 8023{
b6d33834 8024 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
5736199a 8025}
78646121
GN		 8026
8027int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8028{
8029 return kvm_x86_ops->interrupt_allowed(vcpu);
8030}
229456fc 8031
82b32774 8032unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
f92653ee 8033{
82b32774
NA		 8034 if (is_64_bit_mode(vcpu))
8035 return kvm_rip_read(vcpu);
8036 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8037 kvm_rip_read(vcpu));
8038}
8039EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
f92653ee 8040
82b32774
NA		 8041bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8042{
8043 return kvm_get_linear_rip(vcpu) == linear_rip;
f92653ee
JK		 8044}
8045EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8046
94fe45da
JK		 8047unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8048{
8049 unsigned long rflags;
8050
8051 rflags = kvm_x86_ops->get_rflags(vcpu);
8052 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
c310bac5 8053 rflags &= ~X86_EFLAGS_TF;
94fe45da
JK		 8054 return rflags;
8055}
8056EXPORT_SYMBOL_GPL(kvm_get_rflags);
8057
6addfc42 8058static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
94fe45da
JK		 8059{
8060 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
f92653ee 8061 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
c310bac5 8062 rflags |= X86_EFLAGS_TF;
94fe45da 8063 kvm_x86_ops->set_rflags(vcpu, rflags);
6addfc42
PB		 8064}
8065
8066void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8067{
8068 __kvm_set_rflags(vcpu, rflags);
3842d135 8069 kvm_make_request(KVM_REQ_EVENT, vcpu);
94fe45da
JK		 8070}
8071EXPORT_SYMBOL_GPL(kvm_set_rflags);
8072
56028d08
GN		 8073void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8074{
8075 int r;
8076
fb67e14f 8077 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
f2e10669 8078 work->wakeup_all)
56028d08
GN		 8079 return;
8080
8081 r = kvm_mmu_reload(vcpu);
8082 if (unlikely(r))
8083 return;
8084
fb67e14f
XG		 8085 if (!vcpu->arch.mmu.direct_map &&
8086 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8087 return;
8088
56028d08
GN		 8089 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8090}
8091
af585b92
GN		 8092static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8093{
8094 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8095}
8096
8097static inline u32 kvm_async_pf_next_probe(u32 key)
8098{
8099 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8100}
8101
8102static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8103{
8104 u32 key = kvm_async_pf_hash_fn(gfn);
8105
8106 while (vcpu->arch.apf.gfns[key] != ~0)
8107 key = kvm_async_pf_next_probe(key);
8108
8109 vcpu->arch.apf.gfns[key] = gfn;
8110}
8111
8112static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8113{
8114 int i;
8115 u32 key = kvm_async_pf_hash_fn(gfn);
8116
8117 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
c7d28c24
XG		 8118 (vcpu->arch.apf.gfns[key] != gfn &&
8119 vcpu->arch.apf.gfns[key] != ~0); i++)
af585b92
GN		 8120 key = kvm_async_pf_next_probe(key);
8121
8122 return key;
8123}
8124
8125bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8126{
8127 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8128}
8129
8130static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8131{
8132 u32 i, j, k;
8133
8134 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8135 while (true) {
8136 vcpu->arch.apf.gfns[i] = ~0;
8137 do {
8138 j = kvm_async_pf_next_probe(j);
8139 if (vcpu->arch.apf.gfns[j] == ~0)
8140 return;
8141 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8142 /*
8143 * k lies cyclically in ]i,j]
8144 * | i.k.j |
8145 * |....j i.k.| or |.k..j i...|
8146 */
8147 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8148 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8149 i = j;
8150 }
8151}
8152
7c90705b
GN		 8153static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8154{
8155
8156 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8157 sizeof(val));
8158}
8159
af585b92
GN		 8160void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8161 struct kvm_async_pf *work)
8162{
6389ee94
AK		 8163 struct x86_exception fault;
8164
7c90705b 8165 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
af585b92 8166 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7c90705b
GN		 8167
8168 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
fc5f06fa
GN		 8169 (vcpu->arch.apf.send_user_only &&
8170 kvm_x86_ops->get_cpl(vcpu) == 0))
7c90705b
GN		 8171 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8172 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6389ee94
AK		 8173 fault.vector = PF_VECTOR;
8174 fault.error_code_valid = true;
8175 fault.error_code = 0;
8176 fault.nested_page_fault = false;
8177 fault.address = work->arch.token;
8178 kvm_inject_page_fault(vcpu, &fault);
7c90705b 8179 }
af585b92
GN		 8180}
8181
8182void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8183 struct kvm_async_pf *work)
8184{
6389ee94
AK		 8185 struct x86_exception fault;
8186
7c90705b 8187 trace_kvm_async_pf_ready(work->arch.token, work->gva);
f2e10669 8188 if (work->wakeup_all)
7c90705b
GN		 8189 work->arch.token = ~0; /* broadcast wakeup */
8190 else
8191 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8192
8193 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8194 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6389ee94
AK		 8195 fault.vector = PF_VECTOR;
8196 fault.error_code_valid = true;
8197 fault.error_code = 0;
8198 fault.nested_page_fault = false;
8199 fault.address = work->arch.token;
8200 kvm_inject_page_fault(vcpu, &fault);
7c90705b 8201 }
e6d53e3b 8202 vcpu->arch.apf.halted = false;
a4fa1635 8203 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7c90705b
GN		 8204}
8205
8206bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8207{
8208 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8209 return true;
8210 else
8211 return !kvm_event_needs_reinjection(vcpu) &&
8212 kvm_x86_ops->interrupt_allowed(vcpu);
af585b92
GN		 8213}
8214
e0f0bbc5
AW		 8215void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8216{
8217 atomic_inc(&kvm->arch.noncoherent_dma_count);
8218}
8219EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8220
8221void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8222{
8223 atomic_dec(&kvm->arch.noncoherent_dma_count);
8224}
8225EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8226
8227bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8228{
8229 return atomic_read(&kvm->arch.noncoherent_dma_count);
8230}
8231EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8232
229456fc
MT		 8233EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8234EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8235EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8236EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8237EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
0ac406de 8238EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
d8cabddf 8239EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
17897f36 8240EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
236649de 8241EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
ec1ff790 8242EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
532a46b9 8243EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
2e554e8d 8244EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
489223ed 8245EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7b46268d 8246EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
843e4330 8247EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
ubuntu-bionic-kernel mirror