projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: MMU: Don't use RCU for lockless shadow walking
[GitHub/mt8127/android_kernel_alcatel_ttab.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"
313a3dc7 30
18068523 31#include <linux/clocksource.h>
4d5c5d0f 32#include <linux/interrupt.h>
313a3dc7
CO		 33#include <linux/kvm.h>
34#include <linux/fs.h>
35#include <linux/vmalloc.h>
5fb76f9b 36#include <linux/module.h>
0de10343 37#include <linux/mman.h>
2bacc55c 38#include <linux/highmem.h>
19de40a8 39#include <linux/iommu.h>
62c476c7 40#include <linux/intel-iommu.h>
c8076604 41#include <linux/cpufreq.h>
18863bdd 42#include <linux/user-return-notifier.h>
a983fb23 43#include <linux/srcu.h>
5a0e3ad6 44#include <linux/slab.h>
ff9d07a0 45#include <linux/perf_event.h>
7bee342a 46#include <linux/uaccess.h>
af585b92 47#include <linux/hash.h>
a1b60c1c 48#include <linux/pci.h>
aec51dc4 49#include <trace/events/kvm.h>
2ed152af 50
229456fc
MT		 51#define CREATE_TRACE_POINTS
52#include "trace.h"
043405e1 53
24f1e32c 54#include <asm/debugreg.h>
d825ed0a 55#include <asm/msr.h>
a5f61300 56#include <asm/desc.h>
0bed3b56 57#include <asm/mtrr.h>
890ca9ae 58#include <asm/mce.h>
7cf30855 59#include <asm/i387.h>
1361b83a 60#include <asm/fpu-internal.h> /* Ugh! */
98918833 61#include <asm/xcr.h>
1d5f066e 62#include <asm/pvclock.h>
217fc9cf 63#include <asm/div64.h>
043405e1 64
313a3dc7 65#define MAX_IO_MSRS 256
890ca9ae 66#define KVM_MAX_MCE_BANKS 32
5854dbca 67#define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
890ca9ae 68
0f65dd70
AK		 69#define emul_to_vcpu(ctxt) \
70 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
71
50a37eb4
JR		 72/* EFER defaults:
73 * - enable syscall per default because its emulated by KVM
74 * - enable LME and LMA per default on 64 bit KVM
75 */
76#ifdef CONFIG_X86_64
1260edbe
LJ		 77static
78u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
50a37eb4 79#else
1260edbe 80static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
50a37eb4 81#endif
313a3dc7 82
ba1389b7
AK		 83#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
84#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
417bc304 85
cb142eb7 86static void update_cr8_intercept(struct kvm_vcpu *vcpu);
7460fb4a 87static void process_nmi(struct kvm_vcpu *vcpu);
674eea0f 88
97896d04 89struct kvm_x86_ops *kvm_x86_ops;
5fdbf976 90EXPORT_SYMBOL_GPL(kvm_x86_ops);
97896d04 91
476bc001
RR		 92static bool ignore_msrs = 0;
93module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
ed85c068 94
92a1f12d
JR		 95bool kvm_has_tsc_control;
96EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
97u32 kvm_max_guest_tsc_khz;
98EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
99
cc578287
ZA		 100/* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
101static u32 tsc_tolerance_ppm = 250;
102module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
103
18863bdd
AK		 104#define KVM_NR_SHARED_MSRS 16
105
106struct kvm_shared_msrs_global {
107 int nr;
2bf78fa7 108 u32 msrs[KVM_NR_SHARED_MSRS];
18863bdd
AK		 109};
110
111struct kvm_shared_msrs {
112 struct user_return_notifier urn;
113 bool registered;
2bf78fa7
SY		 114 struct kvm_shared_msr_values {
115 u64 host;
116 u64 curr;
117 } values[KVM_NR_SHARED_MSRS];
18863bdd
AK		 118};
119
120static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
121static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
122
417bc304 123struct kvm_stats_debugfs_item debugfs_entries[] = {
ba1389b7
AK		 124 { "pf_fixed", VCPU_STAT(pf_fixed) },
125 { "pf_guest", VCPU_STAT(pf_guest) },
126 { "tlb_flush", VCPU_STAT(tlb_flush) },
127 { "invlpg", VCPU_STAT(invlpg) },
128 { "exits", VCPU_STAT(exits) },
129 { "io_exits", VCPU_STAT(io_exits) },
130 { "mmio_exits", VCPU_STAT(mmio_exits) },
131 { "signal_exits", VCPU_STAT(signal_exits) },
132 { "irq_window", VCPU_STAT(irq_window_exits) },
f08864b4 133 { "nmi_window", VCPU_STAT(nmi_window_exits) },
ba1389b7
AK		 134 { "halt_exits", VCPU_STAT(halt_exits) },
135 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
f11c3a8d 136 { "hypercalls", VCPU_STAT(hypercalls) },
ba1389b7
AK		 137 { "request_irq", VCPU_STAT(request_irq_exits) },
138 { "irq_exits", VCPU_STAT(irq_exits) },
139 { "host_state_reload", VCPU_STAT(host_state_reload) },
140 { "efer_reload", VCPU_STAT(efer_reload) },
141 { "fpu_reload", VCPU_STAT(fpu_reload) },
142 { "insn_emulation", VCPU_STAT(insn_emulation) },
143 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
fa89a817 144 { "irq_injections", VCPU_STAT(irq_injections) },
c4abb7c9 145 { "nmi_injections", VCPU_STAT(nmi_injections) },
4cee5764
AK		 146 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
147 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
148 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
149 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
150 { "mmu_flooded", VM_STAT(mmu_flooded) },
151 { "mmu_recycled", VM_STAT(mmu_recycled) },
dfc5aa00 152 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
4731d4c7 153 { "mmu_unsync", VM_STAT(mmu_unsync) },
0f74a24c 154 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
05da4558 155 { "largepages", VM_STAT(lpages) },
417bc304
HB		 156 { NULL }
157};
158
2acf923e
DC		 159u64 __read_mostly host_xcr0;
160
d6aa1000
AK		 161int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
162
af585b92
GN		 163static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
164{
165 int i;
166 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
167 vcpu->arch.apf.gfns[i] = ~0;
168}
169
18863bdd
AK		 170static void kvm_on_user_return(struct user_return_notifier *urn)
171{
172 unsigned slot;
18863bdd
AK		 173 struct kvm_shared_msrs *locals
174 = container_of(urn, struct kvm_shared_msrs, urn);
2bf78fa7 175 struct kvm_shared_msr_values *values;
18863bdd
AK		 176
177 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
2bf78fa7
SY		 178 values = &locals->values[slot];
179 if (values->host != values->curr) {
180 wrmsrl(shared_msrs_global.msrs[slot], values->host);
181 values->curr = values->host;
18863bdd
AK		 182 }
183 }
184 locals->registered = false;
185 user_return_notifier_unregister(urn);
186}
187
2bf78fa7 188static void shared_msr_update(unsigned slot, u32 msr)
18863bdd 189{
2bf78fa7 190 struct kvm_shared_msrs *smsr;
18863bdd
AK		 191 u64 value;
192
2bf78fa7
SY		 193 smsr = &__get_cpu_var(shared_msrs);
194 /* only read, and nobody should modify it at this time,
195 * so don't need lock */
196 if (slot >= shared_msrs_global.nr) {
197 printk(KERN_ERR "kvm: invalid MSR slot!");
198 return;
199 }
200 rdmsrl_safe(msr, &value);
201 smsr->values[slot].host = value;
202 smsr->values[slot].curr = value;
203}
204
205void kvm_define_shared_msr(unsigned slot, u32 msr)
206{
18863bdd
AK		 207 if (slot >= shared_msrs_global.nr)
208 shared_msrs_global.nr = slot + 1;
2bf78fa7
SY		 209 shared_msrs_global.msrs[slot] = msr;
210 /* we need ensured the shared_msr_global have been updated */
211 smp_wmb();
18863bdd
AK		 212}
213EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
214
215static void kvm_shared_msr_cpu_online(void)
216{
217 unsigned i;
18863bdd
AK		 218
219 for (i = 0; i < shared_msrs_global.nr; ++i)
2bf78fa7 220 shared_msr_update(i, shared_msrs_global.msrs[i]);
18863bdd
AK		 221}
222
d5696725 223void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
18863bdd
AK		 224{
225 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
226
2bf78fa7 227 if (((value ^ smsr->values[slot].curr) & mask) == 0)
18863bdd 228 return;
2bf78fa7
SY		 229 smsr->values[slot].curr = value;
230 wrmsrl(shared_msrs_global.msrs[slot], value);
18863bdd
AK		 231 if (!smsr->registered) {
232 smsr->urn.on_user_return = kvm_on_user_return;
233 user_return_notifier_register(&smsr->urn);
234 smsr->registered = true;
235 }
236}
237EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
238
3548bab5
AK		 239static void drop_user_return_notifiers(void *ignore)
240{
241 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
242
243 if (smsr->registered)
244 kvm_on_user_return(&smsr->urn);
245}
246
6866b83e
CO		 247u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
248{
249 if (irqchip_in_kernel(vcpu->kvm))
ad312c7c 250 return vcpu->arch.apic_base;
6866b83e 251 else
ad312c7c 252 return vcpu->arch.apic_base;
6866b83e
CO		 253}
254EXPORT_SYMBOL_GPL(kvm_get_apic_base);
255
256void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
257{
258 /* TODO: reserve bits check */
259 if (irqchip_in_kernel(vcpu->kvm))
260 kvm_lapic_set_base(vcpu, data);
261 else
ad312c7c 262 vcpu->arch.apic_base = data;
6866b83e
CO		 263}
264EXPORT_SYMBOL_GPL(kvm_set_apic_base);
265
3fd28fce
ED		 266#define EXCPT_BENIGN 0
267#define EXCPT_CONTRIBUTORY 1
268#define EXCPT_PF 2
269
270static int exception_class(int vector)
271{
272 switch (vector) {
273 case PF_VECTOR:
274 return EXCPT_PF;
275 case DE_VECTOR:
276 case TS_VECTOR:
277 case NP_VECTOR:
278 case SS_VECTOR:
279 case GP_VECTOR:
280 return EXCPT_CONTRIBUTORY;
281 default:
282 break;
283 }
284 return EXCPT_BENIGN;
285}
286
287static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
ce7ddec4
JR		 288 unsigned nr, bool has_error, u32 error_code,
289 bool reinject)
3fd28fce
ED		 290{
291 u32 prev_nr;
292 int class1, class2;
293
3842d135
AK		 294 kvm_make_request(KVM_REQ_EVENT, vcpu);
295
3fd28fce
ED		 296 if (!vcpu->arch.exception.pending) {
297 queue:
298 vcpu->arch.exception.pending = true;
299 vcpu->arch.exception.has_error_code = has_error;
300 vcpu->arch.exception.nr = nr;
301 vcpu->arch.exception.error_code = error_code;
3f0fd292 302 vcpu->arch.exception.reinject = reinject;
3fd28fce
ED		 303 return;
304 }
305
306 /* to check exception */
307 prev_nr = vcpu->arch.exception.nr;
308 if (prev_nr == DF_VECTOR) {
309 /* triple fault -> shutdown */
a8eeb04a 310 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3fd28fce
ED		 311 return;
312 }
313 class1 = exception_class(prev_nr);
314 class2 = exception_class(nr);
315 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
316 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
317 /* generate double fault per SDM Table 5-5 */
318 vcpu->arch.exception.pending = true;
319 vcpu->arch.exception.has_error_code = true;
320 vcpu->arch.exception.nr = DF_VECTOR;
321 vcpu->arch.exception.error_code = 0;
322 } else
323 /* replace previous exception with a new one in a hope
324 that instruction re-execution will regenerate lost
325 exception */
326 goto queue;
327}
328
298101da
AK		 329void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
330{
ce7ddec4 331 kvm_multiple_exception(vcpu, nr, false, 0, false);
298101da
AK		 332}
333EXPORT_SYMBOL_GPL(kvm_queue_exception);
334
ce7ddec4
JR		 335void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
336{
337 kvm_multiple_exception(vcpu, nr, false, 0, true);
338}
339EXPORT_SYMBOL_GPL(kvm_requeue_exception);
340
db8fcefa 341void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
c3c91fee 342{
db8fcefa
AP		 343 if (err)
344 kvm_inject_gp(vcpu, 0);
345 else
346 kvm_x86_ops->skip_emulated_instruction(vcpu);
347}
348EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
8df25a32 349
6389ee94 350void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
c3c91fee
AK		 351{
352 ++vcpu->stat.pf_guest;
6389ee94
AK		 353 vcpu->arch.cr2 = fault->address;
354 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
c3c91fee 355}
27d6c865 356EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
c3c91fee 357
6389ee94 358void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
d4f8cf66 359{
6389ee94
AK		 360 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
361 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
d4f8cf66 362 else
6389ee94 363 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
d4f8cf66
JR		 364}
365
3419ffc8
SY		 366void kvm_inject_nmi(struct kvm_vcpu *vcpu)
367{
7460fb4a
AK		 368 atomic_inc(&vcpu->arch.nmi_queued);
369 kvm_make_request(KVM_REQ_NMI, vcpu);
3419ffc8
SY		 370}
371EXPORT_SYMBOL_GPL(kvm_inject_nmi);
372
298101da
AK		 373void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
374{
ce7ddec4 375 kvm_multiple_exception(vcpu, nr, true, error_code, false);
298101da
AK		 376}
377EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
378
ce7ddec4
JR		 379void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
380{
381 kvm_multiple_exception(vcpu, nr, true, error_code, true);
382}
383EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
384
0a79b009
AK		 385/*
386 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
387 * a #GP and return false.
388 */
389bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
298101da 390{
0a79b009
AK		 391 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
392 return true;
393 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
394 return false;
298101da 395}
0a79b009 396EXPORT_SYMBOL_GPL(kvm_require_cpl);
298101da 397
ec92fe44
JR		 398/*
399 * This function will be used to read from the physical memory of the currently
400 * running guest. The difference to kvm_read_guest_page is that this function
401 * can read from guest physical or from the guest's guest physical memory.
402 */
403int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
404 gfn_t ngfn, void *data, int offset, int len,
405 u32 access)
406{
407 gfn_t real_gfn;
408 gpa_t ngpa;
409
410 ngpa = gfn_to_gpa(ngfn);
411 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
412 if (real_gfn == UNMAPPED_GVA)
413 return -EFAULT;
414
415 real_gfn = gpa_to_gfn(real_gfn);
416
417 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
418}
419EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
420
3d06b8bf
JR		 421int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
422 void *data, int offset, int len, u32 access)
423{
424 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
425 data, offset, len, access);
426}
427
a03490ed
CO		 428/*
429 * Load the pae pdptrs. Return true is they are all valid.
430 */
ff03a073 431int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
a03490ed
CO		 432{
433 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
434 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
435 int i;
436 int ret;
ff03a073 437 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
a03490ed 438
ff03a073
JR		 439 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
440 offset * sizeof(u64), sizeof(pdpte),
441 PFERR_USER_MASK|PFERR_WRITE_MASK);
a03490ed
CO		 442 if (ret < 0) {
443 ret = 0;
444 goto out;
445 }
446 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
43a3795a 447 if (is_present_gpte(pdpte[i]) &&
20c466b5 448 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
a03490ed
CO		 449 ret = 0;
450 goto out;
451 }
452 }
453 ret = 1;
454
ff03a073 455 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
6de4f3ad
AK		 456 __set_bit(VCPU_EXREG_PDPTR,
457 (unsigned long *)&vcpu->arch.regs_avail);
458 __set_bit(VCPU_EXREG_PDPTR,
459 (unsigned long *)&vcpu->arch.regs_dirty);
a03490ed 460out:
a03490ed
CO		 461
462 return ret;
463}
cc4b6871 464EXPORT_SYMBOL_GPL(load_pdptrs);
a03490ed 465
d835dfec
AK		 466static bool pdptrs_changed(struct kvm_vcpu *vcpu)
467{
ff03a073 468 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
d835dfec 469 bool changed = true;
3d06b8bf
JR		 470 int offset;
471 gfn_t gfn;
d835dfec
AK		 472 int r;
473
474 if (is_long_mode(vcpu) || !is_pae(vcpu))
475 return false;
476
6de4f3ad
AK		 477 if (!test_bit(VCPU_EXREG_PDPTR,
478 (unsigned long *)&vcpu->arch.regs_avail))
479 return true;
480
9f8fe504
AK		 481 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
482 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
3d06b8bf
JR		 483 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
484 PFERR_USER_MASK | PFERR_WRITE_MASK);
d835dfec
AK		 485 if (r < 0)
486 goto out;
ff03a073 487 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
d835dfec 488out:
d835dfec
AK		 489
490 return changed;
491}
492
49a9b07e 493int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
a03490ed 494{
aad82703
SY		 495 unsigned long old_cr0 = kvm_read_cr0(vcpu);
496 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
497 X86_CR0_CD | X86_CR0_NW;
498
f9a48e6a
AK		 499 cr0 |= X86_CR0_ET;
500
ab344828 501#ifdef CONFIG_X86_64
0f12244f
GN		 502 if (cr0 & 0xffffffff00000000UL)
503 return 1;
ab344828
GN		 504#endif
505
506 cr0 &= ~CR0_RESERVED_BITS;
a03490ed 507
0f12244f
GN		 508 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
509 return 1;
a03490ed 510
0f12244f
GN		 511 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
512 return 1;
a03490ed
CO		 513
514 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
515#ifdef CONFIG_X86_64
f6801dff 516 if ((vcpu->arch.efer & EFER_LME)) {
a03490ed
CO		 517 int cs_db, cs_l;
518
0f12244f
GN		 519 if (!is_pae(vcpu))
520 return 1;
a03490ed 521 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
0f12244f
GN		 522 if (cs_l)
523 return 1;
a03490ed
CO		 524 } else
525#endif
ff03a073 526 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
9f8fe504 527 kvm_read_cr3(vcpu)))
0f12244f 528 return 1;
a03490ed
CO		 529 }
530
531 kvm_x86_ops->set_cr0(vcpu, cr0);
a03490ed 532
d170c419 533 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
e5f3f027 534 kvm_clear_async_pf_completion_queue(vcpu);
d170c419
LJ		 535 kvm_async_pf_hash_reset(vcpu);
536 }
e5f3f027 537
aad82703
SY		 538 if ((cr0 ^ old_cr0) & update_bits)
539 kvm_mmu_reset_context(vcpu);
0f12244f
GN		 540 return 0;
541}
2d3ad1f4 542EXPORT_SYMBOL_GPL(kvm_set_cr0);
a03490ed 543
2d3ad1f4 544void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
a03490ed 545{
49a9b07e 546 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
a03490ed 547}
2d3ad1f4 548EXPORT_SYMBOL_GPL(kvm_lmsw);
a03490ed 549
2acf923e
DC		 550int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
551{
552 u64 xcr0;
553
554 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
555 if (index != XCR_XFEATURE_ENABLED_MASK)
556 return 1;
557 xcr0 = xcr;
558 if (kvm_x86_ops->get_cpl(vcpu) != 0)
559 return 1;
560 if (!(xcr0 & XSTATE_FP))
561 return 1;
562 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
563 return 1;
564 if (xcr0 & ~host_xcr0)
565 return 1;
566 vcpu->arch.xcr0 = xcr0;
567 vcpu->guest_xcr0_loaded = 0;
568 return 0;
569}
570
571int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
572{
573 if (__kvm_set_xcr(vcpu, index, xcr)) {
574 kvm_inject_gp(vcpu, 0);
575 return 1;
576 }
577 return 0;
578}
579EXPORT_SYMBOL_GPL(kvm_set_xcr);
580
a83b29c6 581int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
a03490ed 582{
fc78f519 583 unsigned long old_cr4 = kvm_read_cr4(vcpu);
c68b734f
YW		 584 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
585 X86_CR4_PAE | X86_CR4_SMEP;
0f12244f
GN		 586 if (cr4 & CR4_RESERVED_BITS)
587 return 1;
a03490ed 588
2acf923e
DC		 589 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
590 return 1;
591
c68b734f
YW		 592 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
593 return 1;
594
74dc2b4f
YW		 595 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
596 return 1;
597
a03490ed 598 if (is_long_mode(vcpu)) {
0f12244f
GN		 599 if (!(cr4 & X86_CR4_PAE))
600 return 1;
a2edf57f
AK		 601 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
602 && ((cr4 ^ old_cr4) & pdptr_bits)
9f8fe504
AK		 603 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
604 kvm_read_cr3(vcpu)))
0f12244f
GN		 605 return 1;
606
5e1746d6 607 if (kvm_x86_ops->set_cr4(vcpu, cr4))
0f12244f 608 return 1;
a03490ed 609
aad82703
SY		 610 if ((cr4 ^ old_cr4) & pdptr_bits)
611 kvm_mmu_reset_context(vcpu);
0f12244f 612
2acf923e 613 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
00b27a3e 614 kvm_update_cpuid(vcpu);
2acf923e 615
0f12244f
GN		 616 return 0;
617}
2d3ad1f4 618EXPORT_SYMBOL_GPL(kvm_set_cr4);
a03490ed 619
2390218b 620int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
a03490ed 621{
9f8fe504 622 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
0ba73cda 623 kvm_mmu_sync_roots(vcpu);
d835dfec 624 kvm_mmu_flush_tlb(vcpu);
0f12244f 625 return 0;
d835dfec
AK		 626 }
627
a03490ed 628 if (is_long_mode(vcpu)) {
0f12244f
GN		 629 if (cr3 & CR3_L_MODE_RESERVED_BITS)
630 return 1;
a03490ed
CO		 631 } else {
632 if (is_pae(vcpu)) {
0f12244f
GN		 633 if (cr3 & CR3_PAE_RESERVED_BITS)
634 return 1;
ff03a073
JR		 635 if (is_paging(vcpu) &&
636 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
0f12244f 637 return 1;
a03490ed
CO		 638 }
639 /*
640 * We don't check reserved bits in nonpae mode, because
641 * this isn't enforced, and VMware depends on this.
642 */
643 }
644
a03490ed
CO		 645 /*
646 * Does the new cr3 value map to physical memory? (Note, we
647 * catch an invalid cr3 even in real-mode, because it would
648 * cause trouble later on when we turn on paging anyway.)
649 *
650 * A real CPU would silently accept an invalid cr3 and would
651 * attempt to use it - with largely undefined (and often hard
652 * to debug) behavior on the guest side.
653 */
654 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
0f12244f
GN		 655 return 1;
656 vcpu->arch.cr3 = cr3;
aff48baa 657 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
0f12244f
GN		 658 vcpu->arch.mmu.new_cr3(vcpu);
659 return 0;
660}
2d3ad1f4 661EXPORT_SYMBOL_GPL(kvm_set_cr3);
a03490ed 662
eea1cff9 663int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
a03490ed 664{
0f12244f
GN		 665 if (cr8 & CR8_RESERVED_BITS)
666 return 1;
a03490ed
CO		 667 if (irqchip_in_kernel(vcpu->kvm))
668 kvm_lapic_set_tpr(vcpu, cr8);
669 else
ad312c7c 670 vcpu->arch.cr8 = cr8;
0f12244f
GN		 671 return 0;
672}
2d3ad1f4 673EXPORT_SYMBOL_GPL(kvm_set_cr8);
a03490ed 674
2d3ad1f4 675unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
a03490ed
CO		 676{
677 if (irqchip_in_kernel(vcpu->kvm))
678 return kvm_lapic_get_cr8(vcpu);
679 else
ad312c7c 680 return vcpu->arch.cr8;
a03490ed 681}
2d3ad1f4 682EXPORT_SYMBOL_GPL(kvm_get_cr8);
a03490ed 683
338dbc97 684static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
020df079
GN		 685{
686 switch (dr) {
687 case 0 ... 3:
688 vcpu->arch.db[dr] = val;
689 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
690 vcpu->arch.eff_db[dr] = val;
691 break;
692 case 4:
338dbc97
GN		 693 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
694 return 1; /* #UD */
020df079
GN		 695 /* fall through */
696 case 6:
338dbc97
GN		 697 if (val & 0xffffffff00000000ULL)
698 return -1; /* #GP */
020df079
GN		 699 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
700 break;
701 case 5:
338dbc97
GN		 702 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
703 return 1; /* #UD */
020df079
GN		 704 /* fall through */
705 default: /* 7 */
338dbc97
GN		 706 if (val & 0xffffffff00000000ULL)
707 return -1; /* #GP */
020df079
GN		 708 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
709 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
710 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
711 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
712 }
713 break;
714 }
715
716 return 0;
717}
338dbc97
GN		 718
719int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
720{
721 int res;
722
723 res = __kvm_set_dr(vcpu, dr, val);
724 if (res > 0)
725 kvm_queue_exception(vcpu, UD_VECTOR);
726 else if (res < 0)
727 kvm_inject_gp(vcpu, 0);
728
729 return res;
730}
020df079
GN		 731EXPORT_SYMBOL_GPL(kvm_set_dr);
732
338dbc97 733static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
020df079
GN		 734{
735 switch (dr) {
736 case 0 ... 3:
737 *val = vcpu->arch.db[dr];
738 break;
739 case 4:
338dbc97 740 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
020df079 741 return 1;
020df079
GN		 742 /* fall through */
743 case 6:
744 *val = vcpu->arch.dr6;
745 break;
746 case 5:
338dbc97 747 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
020df079 748 return 1;
020df079
GN		 749 /* fall through */
750 default: /* 7 */
751 *val = vcpu->arch.dr7;
752 break;
753 }
754
755 return 0;
756}
338dbc97
GN		 757
758int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
759{
760 if (_kvm_get_dr(vcpu, dr, val)) {
761 kvm_queue_exception(vcpu, UD_VECTOR);
762 return 1;
763 }
764 return 0;
765}
020df079
GN		 766EXPORT_SYMBOL_GPL(kvm_get_dr);
767
022cd0e8
AK		 768bool kvm_rdpmc(struct kvm_vcpu *vcpu)
769{
770 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
771 u64 data;
772 int err;
773
774 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
775 if (err)
776 return err;
777 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
778 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
779 return err;
780}
781EXPORT_SYMBOL_GPL(kvm_rdpmc);
782
043405e1
CO		 783/*
784 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
785 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
786 *
787 * This list is modified at module load time to reflect the
e3267cbb
GC		 788 * capabilities of the host cpu. This capabilities test skips MSRs that are
789 * kvm-specific. Those are put in the beginning of the list.
043405e1 790 */
e3267cbb 791
c9aaa895 792#define KVM_SAVE_MSRS_BEGIN 9
043405e1 793static u32 msrs_to_save[] = {
e3267cbb 794 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
11c6bffa 795 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
55cd8e5a 796 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
c9aaa895 797 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
043405e1 798 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
8c06585d 799 MSR_STAR,
043405e1
CO		 800#ifdef CONFIG_X86_64
801 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
802#endif
e90aa41e 803 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
043405e1
CO		 804};
805
806static unsigned num_msrs_to_save;
807
808static u32 emulated_msrs[] = {
a3e06bbe 809 MSR_IA32_TSCDEADLINE,
043405e1 810 MSR_IA32_MISC_ENABLE,
908e75f3
AK		 811 MSR_IA32_MCG_STATUS,
812 MSR_IA32_MCG_CTL,
043405e1
CO		 813};
814
b69e8cae 815static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
15c4a640 816{
aad82703
SY		 817 u64 old_efer = vcpu->arch.efer;
818
b69e8cae
RJ		 819 if (efer & efer_reserved_bits)
820 return 1;
15c4a640
CO		 821
822 if (is_paging(vcpu)
b69e8cae
RJ		 823 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
824 return 1;
15c4a640 825
1b2fd70c
AG		 826 if (efer & EFER_FFXSR) {
827 struct kvm_cpuid_entry2 *feat;
828
829 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
b69e8cae
RJ		 830 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
831 return 1;
1b2fd70c
AG		 832 }
833
d8017474
AG		 834 if (efer & EFER_SVME) {
835 struct kvm_cpuid_entry2 *feat;
836
837 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
b69e8cae
RJ		 838 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
839 return 1;
d8017474
AG		 840 }
841
15c4a640 842 efer &= ~EFER_LMA;
f6801dff 843 efer |= vcpu->arch.efer & EFER_LMA;
15c4a640 844
a3d204e2
SY		 845 kvm_x86_ops->set_efer(vcpu, efer);
846
9645bb56 847 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
b69e8cae 848
aad82703
SY		 849 /* Update reserved bits */
850 if ((efer ^ old_efer) & EFER_NX)
851 kvm_mmu_reset_context(vcpu);
852
b69e8cae 853 return 0;
15c4a640
CO		 854}
855
f2b4b7dd
JR		 856void kvm_enable_efer_bits(u64 mask)
857{
858 efer_reserved_bits &= ~mask;
859}
860EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
861
862
15c4a640
CO		 863/*
864 * Writes msr value into into the appropriate "register".
865 * Returns 0 on success, non-0 otherwise.
866 * Assumes vcpu_load() was already called.
867 */
868int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
869{
870 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
871}
872
313a3dc7
CO		 873/*
874 * Adapt set_msr() to msr_io()'s calling convention
875 */
876static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
877{
878 return kvm_set_msr(vcpu, index, *data);
879}
880
18068523
GOC		 881static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
882{
9ed3c444
AK		 883 int version;
884 int r;
50d0a0f9 885 struct pvclock_wall_clock wc;
923de3cf 886 struct timespec boot;
18068523
GOC		 887
888 if (!wall_clock)
889 return;
890
9ed3c444
AK		 891 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
892 if (r)
893 return;
894
895 if (version & 1)
896 ++version; /* first time write, random junk */
897
898 ++version;
18068523 899
18068523
GOC		 900 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
901
50d0a0f9
GH		 902 /*
903 * The guest calculates current wall clock time by adding
34c238a1 904 * system time (updated by kvm_guest_time_update below) to the
50d0a0f9
GH		 905 * wall clock specified here. guest system time equals host
906 * system time for us, thus we must fill in host boot time here.
907 */
923de3cf 908 getboottime(&boot);
50d0a0f9
GH		 909
910 wc.sec = boot.tv_sec;
911 wc.nsec = boot.tv_nsec;
912 wc.version = version;
18068523
GOC		 913
914 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
915
916 version++;
917 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
18068523
GOC		 918}
919
50d0a0f9
GH		 920static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
921{
922 uint32_t quotient, remainder;
923
924 /* Don't try to replace with do_div(), this one calculates
925 * "(dividend << 32) / divisor" */
926 __asm__ ( "divl %4"
927 : "=a" (quotient), "=d" (remainder)
928 : "0" (0), "1" (dividend), "r" (divisor) );
929 return quotient;
930}
931
5f4e3f88
ZA		 932static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
933 s8 *pshift, u32 *pmultiplier)
50d0a0f9 934{
5f4e3f88 935 uint64_t scaled64;
50d0a0f9
GH		 936 int32_t shift = 0;
937 uint64_t tps64;
938 uint32_t tps32;
939
5f4e3f88
ZA		 940 tps64 = base_khz * 1000LL;
941 scaled64 = scaled_khz * 1000LL;
50933623 942 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
50d0a0f9
GH		 943 tps64 >>= 1;
944 shift--;
945 }
946
947 tps32 = (uint32_t)tps64;
50933623
JK		 948 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
949 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
5f4e3f88
ZA		 950 scaled64 >>= 1;
951 else
952 tps32 <<= 1;
50d0a0f9
GH		 953 shift++;
954 }
955
5f4e3f88
ZA		 956 *pshift = shift;
957 *pmultiplier = div_frac(scaled64, tps32);
50d0a0f9 958
5f4e3f88
ZA		 959 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
960 __func__, base_khz, scaled_khz, shift, *pmultiplier);
50d0a0f9
GH		 961}
962
759379dd
ZA		 963static inline u64 get_kernel_ns(void)
964{
965 struct timespec ts;
966
967 WARN_ON(preemptible());
968 ktime_get_ts(&ts);
969 monotonic_to_bootbased(&ts);
970 return timespec_to_ns(&ts);
50d0a0f9
GH		 971}
972
c8076604 973static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
c285545f 974unsigned long max_tsc_khz;
c8076604 975
cc578287 976static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
8cfdc000 977{
cc578287
ZA		 978 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
979 vcpu->arch.virtual_tsc_shift);
8cfdc000
ZA		 980}
981
cc578287 982static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1e993611 983{
cc578287
ZA		 984 u64 v = (u64)khz * (1000000 + ppm);
985 do_div(v, 1000000);
986 return v;
1e993611
JR		 987}
988
cc578287 989static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
759379dd 990{
cc578287
ZA		 991 u32 thresh_lo, thresh_hi;
992 int use_scaling = 0;
217fc9cf 993
c285545f
ZA		 994 /* Compute a scale to convert nanoseconds in TSC cycles */
995 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
cc578287
ZA		 996 &vcpu->arch.virtual_tsc_shift,
997 &vcpu->arch.virtual_tsc_mult);
998 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
999
1000 /*
1001 * Compute the variation in TSC rate which is acceptable
1002 * within the range of tolerance and decide if the
1003 * rate being applied is within that bounds of the hardware
1004 * rate. If so, no scaling or compensation need be done.
1005 */
1006 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1007 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1008 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1009 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1010 use_scaling = 1;
1011 }
1012 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
c285545f
ZA		 1013}
1014
1015static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1016{
e26101b1 1017 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
cc578287
ZA		 1018 vcpu->arch.virtual_tsc_mult,
1019 vcpu->arch.virtual_tsc_shift);
e26101b1 1020 tsc += vcpu->arch.this_tsc_write;
c285545f
ZA		 1021 return tsc;
1022}
1023
99e3e30a
ZA		 1024void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1025{
1026 struct kvm *kvm = vcpu->kvm;
f38e098f 1027 u64 offset, ns, elapsed;
99e3e30a 1028 unsigned long flags;
02626b6a 1029 s64 usdiff;
99e3e30a 1030
038f8c11 1031 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
857e4099 1032 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
759379dd 1033 ns = get_kernel_ns();
f38e098f 1034 elapsed = ns - kvm->arch.last_tsc_nsec;
5d3cb0f6
ZA		 1035
1036 /* n.b - signed multiplication and division required */
02626b6a 1037 usdiff = data - kvm->arch.last_tsc_write;
5d3cb0f6 1038#ifdef CONFIG_X86_64
02626b6a 1039 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
5d3cb0f6
ZA		 1040#else
1041 /* do_div() only does unsigned */
1042 asm("idivl %2; xor %%edx, %%edx"
02626b6a
MT		 1043 : "=A"(usdiff)
1044 : "A"(usdiff * 1000), "rm"(vcpu->arch.virtual_tsc_khz));
5d3cb0f6 1045#endif
02626b6a
MT		 1046 do_div(elapsed, 1000);
1047 usdiff -= elapsed;
1048 if (usdiff < 0)
1049 usdiff = -usdiff;
f38e098f
ZA		 1050
1051 /*
5d3cb0f6
ZA		 1052 * Special case: TSC write with a small delta (1 second) of virtual
1053 * cycle time against real time is interpreted as an attempt to
1054 * synchronize the CPU.
1055 *
1056 * For a reliable TSC, we can match TSC offsets, and for an unstable
1057 * TSC, we add elapsed time in this computation. We could let the
1058 * compensation code attempt to catch up if we fall behind, but
1059 * it's better to try to match offsets from the beginning.
1060 */
02626b6a 1061 if (usdiff < USEC_PER_SEC &&
5d3cb0f6 1062 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
f38e098f 1063 if (!check_tsc_unstable()) {
e26101b1 1064 offset = kvm->arch.cur_tsc_offset;
f38e098f
ZA		 1065 pr_debug("kvm: matched tsc offset for %llu\n", data);
1066 } else {
857e4099 1067 u64 delta = nsec_to_cycles(vcpu, elapsed);
5d3cb0f6
ZA		 1068 data += delta;
1069 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
759379dd 1070 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
f38e098f 1071 }
e26101b1
ZA		 1072 } else {
1073 /*
1074 * We split periods of matched TSC writes into generations.
1075 * For each generation, we track the original measured
1076 * nanosecond time, offset, and write, so if TSCs are in
1077 * sync, we can match exact offset, and if not, we can match
1078 * exact software computaion in compute_guest_tsc()
1079 *
1080 * These values are tracked in kvm->arch.cur_xxx variables.
1081 */
1082 kvm->arch.cur_tsc_generation++;
1083 kvm->arch.cur_tsc_nsec = ns;
1084 kvm->arch.cur_tsc_write = data;
1085 kvm->arch.cur_tsc_offset = offset;
1086 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1087 kvm->arch.cur_tsc_generation, data);
f38e098f 1088 }
e26101b1
ZA		 1089
1090 /*
1091 * We also track th most recent recorded KHZ, write and time to
1092 * allow the matching interval to be extended at each write.
1093 */
f38e098f
ZA		 1094 kvm->arch.last_tsc_nsec = ns;
1095 kvm->arch.last_tsc_write = data;
5d3cb0f6 1096 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
99e3e30a
ZA		 1097
1098 /* Reset of TSC must disable overshoot protection below */
1099 vcpu->arch.hv_clock.tsc_timestamp = 0;
b183aa58 1100 vcpu->arch.last_guest_tsc = data;
e26101b1
ZA		 1101
1102 /* Keep track of which generation this VCPU has synchronized to */
1103 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1104 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1105 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1106
1107 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1108 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
99e3e30a 1109}
e26101b1 1110
99e3e30a
ZA		 1111EXPORT_SYMBOL_GPL(kvm_write_tsc);
1112
34c238a1 1113static int kvm_guest_time_update(struct kvm_vcpu *v)
18068523 1114{
18068523
GOC		 1115 unsigned long flags;
1116 struct kvm_vcpu_arch *vcpu = &v->arch;
1117 void *shared_kaddr;
463656c0 1118 unsigned long this_tsc_khz;
1d5f066e
ZA		 1119 s64 kernel_ns, max_kernel_ns;
1120 u64 tsc_timestamp;
18068523 1121
18068523
GOC		 1122 /* Keep irq disabled to prevent changes to the clock */
1123 local_irq_save(flags);
d5c1785d 1124 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v);
759379dd 1125 kernel_ns = get_kernel_ns();
cc578287 1126 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
8cfdc000 1127 if (unlikely(this_tsc_khz == 0)) {
c285545f 1128 local_irq_restore(flags);
34c238a1 1129 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
8cfdc000
ZA		 1130 return 1;
1131 }
18068523 1132
c285545f
ZA		 1133 /*
1134 * We may have to catch up the TSC to match elapsed wall clock
1135 * time for two reasons, even if kvmclock is used.
1136 * 1) CPU could have been running below the maximum TSC rate
1137 * 2) Broken TSC compensation resets the base at each VCPU
1138 * entry to avoid unknown leaps of TSC even when running
1139 * again on the same CPU. This may cause apparent elapsed
1140 * time to disappear, and the guest to stand still or run
1141 * very slowly.
1142 */
1143 if (vcpu->tsc_catchup) {
1144 u64 tsc = compute_guest_tsc(v, kernel_ns);
1145 if (tsc > tsc_timestamp) {
f1e2b260 1146 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
c285545f
ZA		 1147 tsc_timestamp = tsc;
1148 }
50d0a0f9
GH		 1149 }
1150
18068523
GOC		 1151 local_irq_restore(flags);
1152
c285545f
ZA		 1153 if (!vcpu->time_page)
1154 return 0;
18068523 1155
1d5f066e
ZA		 1156 /*
1157 * Time as measured by the TSC may go backwards when resetting the base
1158 * tsc_timestamp. The reason for this is that the TSC resolution is
1159 * higher than the resolution of the other clock scales. Thus, many
1160 * possible measurments of the TSC correspond to one measurement of any
1161 * other clock, and so a spread of values is possible. This is not a
1162 * problem for the computation of the nanosecond clock; with TSC rates
1163 * around 1GHZ, there can only be a few cycles which correspond to one
1164 * nanosecond value, and any path through this code will inevitably
1165 * take longer than that. However, with the kernel_ns value itself,
1166 * the precision may be much lower, down to HZ granularity. If the
1167 * first sampling of TSC against kernel_ns ends in the low part of the
1168 * range, and the second in the high end of the range, we can get:
1169 *
1170 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1171 *
1172 * As the sampling errors potentially range in the thousands of cycles,
1173 * it is possible such a time value has already been observed by the
1174 * guest. To protect against this, we must compute the system time as
1175 * observed by the guest and ensure the new system time is greater.
1176 */
1177 max_kernel_ns = 0;
b183aa58 1178 if (vcpu->hv_clock.tsc_timestamp) {
1d5f066e
ZA		 1179 max_kernel_ns = vcpu->last_guest_tsc -
1180 vcpu->hv_clock.tsc_timestamp;
1181 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1182 vcpu->hv_clock.tsc_to_system_mul,
1183 vcpu->hv_clock.tsc_shift);
1184 max_kernel_ns += vcpu->last_kernel_ns;
1185 }
afbcf7ab 1186
e48672fa 1187 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
5f4e3f88
ZA		 1188 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1189 &vcpu->hv_clock.tsc_shift,
1190 &vcpu->hv_clock.tsc_to_system_mul);
e48672fa 1191 vcpu->hw_tsc_khz = this_tsc_khz;
8cfdc000
ZA		 1192 }
1193
1d5f066e
ZA		 1194 if (max_kernel_ns > kernel_ns)
1195 kernel_ns = max_kernel_ns;
1196
8cfdc000 1197 /* With all the info we got, fill in the values */
1d5f066e 1198 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
759379dd 1199 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1d5f066e 1200 vcpu->last_kernel_ns = kernel_ns;
28e4639a 1201 vcpu->last_guest_tsc = tsc_timestamp;
371bcf64
GC		 1202 vcpu->hv_clock.flags = 0;
1203
18068523
GOC		 1204 /*
1205 * The interface expects us to write an even number signaling that the
1206 * update is finished. Since the guest won't see the intermediate
50d0a0f9 1207 * state, we just increase by 2 at the end.
18068523 1208 */
50d0a0f9 1209 vcpu->hv_clock.version += 2;
18068523 1210
8fd75e12 1211 shared_kaddr = kmap_atomic(vcpu->time_page);
18068523
GOC		 1212
1213 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
50d0a0f9 1214 sizeof(vcpu->hv_clock));
18068523 1215
8fd75e12 1216 kunmap_atomic(shared_kaddr);
18068523
GOC		 1217
1218 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
8cfdc000 1219 return 0;
c8076604
GH		 1220}
1221
9ba075a6
AK		 1222static bool msr_mtrr_valid(unsigned msr)
1223{
1224 switch (msr) {
1225 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1226 case MSR_MTRRfix64K_00000:
1227 case MSR_MTRRfix16K_80000:
1228 case MSR_MTRRfix16K_A0000:
1229 case MSR_MTRRfix4K_C0000:
1230 case MSR_MTRRfix4K_C8000:
1231 case MSR_MTRRfix4K_D0000:
1232 case MSR_MTRRfix4K_D8000:
1233 case MSR_MTRRfix4K_E0000:
1234 case MSR_MTRRfix4K_E8000:
1235 case MSR_MTRRfix4K_F0000:
1236 case MSR_MTRRfix4K_F8000:
1237 case MSR_MTRRdefType:
1238 case MSR_IA32_CR_PAT:
1239 return true;
1240 case 0x2f8:
1241 return true;
1242 }
1243 return false;
1244}
1245
d6289b93
MT		 1246static bool valid_pat_type(unsigned t)
1247{
1248 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1249}
1250
1251static bool valid_mtrr_type(unsigned t)
1252{
1253 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1254}
1255
1256static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1257{
1258 int i;
1259
1260 if (!msr_mtrr_valid(msr))
1261 return false;
1262
1263 if (msr == MSR_IA32_CR_PAT) {
1264 for (i = 0; i < 8; i++)
1265 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1266 return false;
1267 return true;
1268 } else if (msr == MSR_MTRRdefType) {
1269 if (data & ~0xcff)
1270 return false;
1271 return valid_mtrr_type(data & 0xff);
1272 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1273 for (i = 0; i < 8 ; i++)
1274 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1275 return false;
1276 return true;
1277 }
1278
1279 /* variable MTRRs */
1280 return valid_mtrr_type(data & 0xff);
1281}
1282
9ba075a6
AK		 1283static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1284{
0bed3b56
SY		 1285 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1286
d6289b93 1287 if (!mtrr_valid(vcpu, msr, data))
9ba075a6
AK		 1288 return 1;
1289
0bed3b56
SY		 1290 if (msr == MSR_MTRRdefType) {
1291 vcpu->arch.mtrr_state.def_type = data;
1292 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1293 } else if (msr == MSR_MTRRfix64K_00000)
1294 p[0] = data;
1295 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1296 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1297 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1298 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1299 else if (msr == MSR_IA32_CR_PAT)
1300 vcpu->arch.pat = data;
1301 else { /* Variable MTRRs */
1302 int idx, is_mtrr_mask;
1303 u64 *pt;
1304
1305 idx = (msr - 0x200) / 2;
1306 is_mtrr_mask = msr - 0x200 - 2 * idx;
1307 if (!is_mtrr_mask)
1308 pt =
1309 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1310 else
1311 pt =
1312 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1313 *pt = data;
1314 }
1315
1316 kvm_mmu_reset_context(vcpu);
9ba075a6
AK		 1317 return 0;
1318}
15c4a640 1319
890ca9ae 1320static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
15c4a640 1321{
890ca9ae
HY		 1322 u64 mcg_cap = vcpu->arch.mcg_cap;
1323 unsigned bank_num = mcg_cap & 0xff;
1324
15c4a640 1325 switch (msr) {
15c4a640 1326 case MSR_IA32_MCG_STATUS:
890ca9ae 1327 vcpu->arch.mcg_status = data;
15c4a640 1328 break;
c7ac679c 1329 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 1330 if (!(mcg_cap & MCG_CTL_P))
1331 return 1;
1332 if (data != 0 && data != ~(u64)0)
1333 return -1;
1334 vcpu->arch.mcg_ctl = data;
1335 break;
1336 default:
1337 if (msr >= MSR_IA32_MC0_CTL &&
1338 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1339 u32 offset = msr - MSR_IA32_MC0_CTL;
114be429
AP		 1340 /* only 0 or all 1s can be written to IA32_MCi_CTL
1341 * some Linux kernels though clear bit 10 in bank 4 to
1342 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1343 * this to avoid an uncatched #GP in the guest
1344 */
890ca9ae 1345 if ((offset & 0x3) == 0 &&
114be429 1346 data != 0 && (data | (1 << 10)) != ~(u64)0)
890ca9ae
HY		 1347 return -1;
1348 vcpu->arch.mce_banks[offset] = data;
1349 break;
1350 }
1351 return 1;
1352 }
1353 return 0;
1354}
1355
ffde22ac
ES		 1356static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1357{
1358 struct kvm *kvm = vcpu->kvm;
1359 int lm = is_long_mode(vcpu);
1360 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1361 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1362 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1363 : kvm->arch.xen_hvm_config.blob_size_32;
1364 u32 page_num = data & ~PAGE_MASK;
1365 u64 page_addr = data & PAGE_MASK;
1366 u8 *page;
1367 int r;
1368
1369 r = -E2BIG;
1370 if (page_num >= blob_size)
1371 goto out;
1372 r = -ENOMEM;
ff5c2c03
SL		 1373 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1374 if (IS_ERR(page)) {
1375 r = PTR_ERR(page);
ffde22ac 1376 goto out;
ff5c2c03 1377 }
ffde22ac
ES		 1378 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1379 goto out_free;
1380 r = 0;
1381out_free:
1382 kfree(page);
1383out:
1384 return r;
1385}
1386
55cd8e5a
GN		 1387static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1388{
1389 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1390}
1391
1392static bool kvm_hv_msr_partition_wide(u32 msr)
1393{
1394 bool r = false;
1395 switch (msr) {
1396 case HV_X64_MSR_GUEST_OS_ID:
1397 case HV_X64_MSR_HYPERCALL:
1398 r = true;
1399 break;
1400 }
1401
1402 return r;
1403}
1404
1405static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1406{
1407 struct kvm *kvm = vcpu->kvm;
1408
1409 switch (msr) {
1410 case HV_X64_MSR_GUEST_OS_ID:
1411 kvm->arch.hv_guest_os_id = data;
1412 /* setting guest os id to zero disables hypercall page */
1413 if (!kvm->arch.hv_guest_os_id)
1414 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1415 break;
1416 case HV_X64_MSR_HYPERCALL: {
1417 u64 gfn;
1418 unsigned long addr;
1419 u8 instructions[4];
1420
1421 /* if guest os id is not set hypercall should remain disabled */
1422 if (!kvm->arch.hv_guest_os_id)
1423 break;
1424 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1425 kvm->arch.hv_hypercall = data;
1426 break;
1427 }
1428 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1429 addr = gfn_to_hva(kvm, gfn);
1430 if (kvm_is_error_hva(addr))
1431 return 1;
1432 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1433 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
8b0cedff 1434 if (__copy_to_user((void __user *)addr, instructions, 4))
55cd8e5a
GN		 1435 return 1;
1436 kvm->arch.hv_hypercall = data;
1437 break;
1438 }
1439 default:
1440 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1441 "data 0x%llx\n", msr, data);
1442 return 1;
1443 }
1444 return 0;
1445}
1446
1447static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1448{
10388a07
GN		 1449 switch (msr) {
1450 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1451 unsigned long addr;
55cd8e5a 1452
10388a07
GN		 1453 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1454 vcpu->arch.hv_vapic = data;
1455 break;
1456 }
1457 addr = gfn_to_hva(vcpu->kvm, data >>
1458 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1459 if (kvm_is_error_hva(addr))
1460 return 1;
8b0cedff 1461 if (__clear_user((void __user *)addr, PAGE_SIZE))
10388a07
GN		 1462 return 1;
1463 vcpu->arch.hv_vapic = data;
1464 break;
1465 }
1466 case HV_X64_MSR_EOI:
1467 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1468 case HV_X64_MSR_ICR:
1469 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1470 case HV_X64_MSR_TPR:
1471 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1472 default:
1473 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1474 "data 0x%llx\n", msr, data);
1475 return 1;
1476 }
1477
1478 return 0;
55cd8e5a
GN		 1479}
1480
344d9588
GN		 1481static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1482{
1483 gpa_t gpa = data & ~0x3f;
1484
6adba527
GN		 1485 /* Bits 2:5 are resrved, Should be zero */
1486 if (data & 0x3c)
344d9588
GN		 1487 return 1;
1488
1489 vcpu->arch.apf.msr_val = data;
1490
1491 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1492 kvm_clear_async_pf_completion_queue(vcpu);
1493 kvm_async_pf_hash_reset(vcpu);
1494 return 0;
1495 }
1496
1497 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1498 return 1;
1499
6adba527 1500 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
344d9588
GN		 1501 kvm_async_pf_wakeup_all(vcpu);
1502 return 0;
1503}
1504
12f9a48f
GC		 1505static void kvmclock_reset(struct kvm_vcpu *vcpu)
1506{
1507 if (vcpu->arch.time_page) {
1508 kvm_release_page_dirty(vcpu->arch.time_page);
1509 vcpu->arch.time_page = NULL;
1510 }
1511}
1512
c9aaa895
GC		 1513static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1514{
1515 u64 delta;
1516
1517 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1518 return;
1519
1520 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1521 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1522 vcpu->arch.st.accum_steal = delta;
1523}
1524
1525static void record_steal_time(struct kvm_vcpu *vcpu)
1526{
1527 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1528 return;
1529
1530 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1531 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1532 return;
1533
1534 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1535 vcpu->arch.st.steal.version += 2;
1536 vcpu->arch.st.accum_steal = 0;
1537
1538 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1539 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1540}
1541
15c4a640
CO		 1542int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1543{
5753785f
GN		 1544 bool pr = false;
1545
15c4a640 1546 switch (msr) {
15c4a640 1547 case MSR_EFER:
b69e8cae 1548 return set_efer(vcpu, data);
8f1589d9
AP		 1549 case MSR_K7_HWCR:
1550 data &= ~(u64)0x40; /* ignore flush filter disable */
82494028 1551 data &= ~(u64)0x100; /* ignore ignne emulation enable */
a223c313 1552 data &= ~(u64)0x8; /* ignore TLB cache disable */
8f1589d9
AP		 1553 if (data != 0) {
1554 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1555 data);
1556 return 1;
1557 }
15c4a640 1558 break;
f7c6d140
AP		 1559 case MSR_FAM10H_MMIO_CONF_BASE:
1560 if (data != 0) {
1561 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1562 "0x%llx\n", data);
1563 return 1;
1564 }
15c4a640 1565 break;
c323c0e5 1566 case MSR_AMD64_NB_CFG:
c7ac679c 1567 break;
b5e2fec0
AG		 1568 case MSR_IA32_DEBUGCTLMSR:
1569 if (!data) {
1570 /* We support the non-activated case already */
1571 break;
1572 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1573 /* Values other than LBR and BTF are vendor-specific,
1574 thus reserved and should throw a #GP */
1575 return 1;
1576 }
1577 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1578 __func__, data);
1579 break;
15c4a640
CO		 1580 case MSR_IA32_UCODE_REV:
1581 case MSR_IA32_UCODE_WRITE:
61a6bd67 1582 case MSR_VM_HSAVE_PA:
6098ca93 1583 case MSR_AMD64_PATCH_LOADER:
15c4a640 1584 break;
9ba075a6
AK		 1585 case 0x200 ... 0x2ff:
1586 return set_msr_mtrr(vcpu, msr, data);
15c4a640
CO		 1587 case MSR_IA32_APICBASE:
1588 kvm_set_apic_base(vcpu, data);
1589 break;
0105d1a5
GN		 1590 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1591 return kvm_x2apic_msr_write(vcpu, msr, data);
a3e06bbe
LJ		 1592 case MSR_IA32_TSCDEADLINE:
1593 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1594 break;
15c4a640 1595 case MSR_IA32_MISC_ENABLE:
ad312c7c 1596 vcpu->arch.ia32_misc_enable_msr = data;
15c4a640 1597 break;
11c6bffa 1598 case MSR_KVM_WALL_CLOCK_NEW:
18068523
GOC		 1599 case MSR_KVM_WALL_CLOCK:
1600 vcpu->kvm->arch.wall_clock = data;
1601 kvm_write_wall_clock(vcpu->kvm, data);
1602 break;
11c6bffa 1603 case MSR_KVM_SYSTEM_TIME_NEW:
18068523 1604 case MSR_KVM_SYSTEM_TIME: {
12f9a48f 1605 kvmclock_reset(vcpu);
18068523
GOC		 1606
1607 vcpu->arch.time = data;
c285545f 1608 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
18068523
GOC		 1609
1610 /* we verify if the enable bit is set... */
1611 if (!(data & 1))
1612 break;
1613
1614 /* ...but clean it before doing the actual write */
1615 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1616
18068523
GOC		 1617 vcpu->arch.time_page =
1618 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
18068523
GOC		 1619
1620 if (is_error_page(vcpu->arch.time_page)) {
1621 kvm_release_page_clean(vcpu->arch.time_page);
1622 vcpu->arch.time_page = NULL;
1623 }
18068523
GOC		 1624 break;
1625 }
344d9588
GN		 1626 case MSR_KVM_ASYNC_PF_EN:
1627 if (kvm_pv_enable_async_pf(vcpu, data))
1628 return 1;
1629 break;
c9aaa895
GC		 1630 case MSR_KVM_STEAL_TIME:
1631
1632 if (unlikely(!sched_info_on()))
1633 return 1;
1634
1635 if (data & KVM_STEAL_RESERVED_MASK)
1636 return 1;
1637
1638 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1639 data & KVM_STEAL_VALID_BITS))
1640 return 1;
1641
1642 vcpu->arch.st.msr_val = data;
1643
1644 if (!(data & KVM_MSR_ENABLED))
1645 break;
1646
1647 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1648
1649 preempt_disable();
1650 accumulate_steal_time(vcpu);
1651 preempt_enable();
1652
1653 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1654
1655 break;
1656
890ca9ae
HY		 1657 case MSR_IA32_MCG_CTL:
1658 case MSR_IA32_MCG_STATUS:
1659 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1660 return set_msr_mce(vcpu, msr, data);
71db6023
AP		 1661
1662 /* Performance counters are not protected by a CPUID bit,
1663 * so we should check all of them in the generic path for the sake of
1664 * cross vendor migration.
1665 * Writing a zero into the event select MSRs disables them,
1666 * which we perfectly emulate ;-). Any other value should be at least
1667 * reported, some guests depend on them.
1668 */
71db6023
AP		 1669 case MSR_K7_EVNTSEL0:
1670 case MSR_K7_EVNTSEL1:
1671 case MSR_K7_EVNTSEL2:
1672 case MSR_K7_EVNTSEL3:
1673 if (data != 0)
1674 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1675 "0x%x data 0x%llx\n", msr, data);
1676 break;
1677 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1678 * so we ignore writes to make it happy.
1679 */
71db6023
AP		 1680 case MSR_K7_PERFCTR0:
1681 case MSR_K7_PERFCTR1:
1682 case MSR_K7_PERFCTR2:
1683 case MSR_K7_PERFCTR3:
1684 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1685 "0x%x data 0x%llx\n", msr, data);
1686 break;
5753785f
GN		 1687 case MSR_P6_PERFCTR0:
1688 case MSR_P6_PERFCTR1:
1689 pr = true;
1690 case MSR_P6_EVNTSEL0:
1691 case MSR_P6_EVNTSEL1:
1692 if (kvm_pmu_msr(vcpu, msr))
1693 return kvm_pmu_set_msr(vcpu, msr, data);
1694
1695 if (pr || data != 0)
1696 pr_unimpl(vcpu, "disabled perfctr wrmsr: "
1697 "0x%x data 0x%llx\n", msr, data);
1698 break;
84e0cefa
JS		 1699 case MSR_K7_CLK_CTL:
1700 /*
1701 * Ignore all writes to this no longer documented MSR.
1702 * Writes are only relevant for old K7 processors,
1703 * all pre-dating SVM, but a recommended workaround from
1704 * AMD for these chips. It is possible to speicify the
1705 * affected processor models on the command line, hence
1706 * the need to ignore the workaround.
1707 */
1708 break;
55cd8e5a
GN		 1709 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1710 if (kvm_hv_msr_partition_wide(msr)) {
1711 int r;
1712 mutex_lock(&vcpu->kvm->lock);
1713 r = set_msr_hyperv_pw(vcpu, msr, data);
1714 mutex_unlock(&vcpu->kvm->lock);
1715 return r;
1716 } else
1717 return set_msr_hyperv(vcpu, msr, data);
1718 break;
91c9c3ed 1719 case MSR_IA32_BBL_CR_CTL3:
1720 /* Drop writes to this legacy MSR -- see rdmsr
1721 * counterpart for further detail.
1722 */
1723 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1724 break;
2b036c6b
BO		 1725 case MSR_AMD64_OSVW_ID_LENGTH:
1726 if (!guest_cpuid_has_osvw(vcpu))
1727 return 1;
1728 vcpu->arch.osvw.length = data;
1729 break;
1730 case MSR_AMD64_OSVW_STATUS:
1731 if (!guest_cpuid_has_osvw(vcpu))
1732 return 1;
1733 vcpu->arch.osvw.status = data;
1734 break;
15c4a640 1735 default:
ffde22ac
ES		 1736 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1737 return xen_hvm_config(vcpu, data);
f5132b01
GN		 1738 if (kvm_pmu_msr(vcpu, msr))
1739 return kvm_pmu_set_msr(vcpu, msr, data);
ed85c068
AP		 1740 if (!ignore_msrs) {
1741 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1742 msr, data);
1743 return 1;
1744 } else {
1745 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1746 msr, data);
1747 break;
1748 }
15c4a640
CO		 1749 }
1750 return 0;
1751}
1752EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1753
1754
1755/*
1756 * Reads an msr value (of 'msr_index') into 'pdata'.
1757 * Returns 0 on success, non-0 otherwise.
1758 * Assumes vcpu_load() was already called.
1759 */
1760int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1761{
1762 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1763}
1764
9ba075a6
AK		 1765static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1766{
0bed3b56
SY		 1767 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1768
9ba075a6
AK		 1769 if (!msr_mtrr_valid(msr))
1770 return 1;
1771
0bed3b56
SY		 1772 if (msr == MSR_MTRRdefType)
1773 *pdata = vcpu->arch.mtrr_state.def_type +
1774 (vcpu->arch.mtrr_state.enabled << 10);
1775 else if (msr == MSR_MTRRfix64K_00000)
1776 *pdata = p[0];
1777 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1778 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1779 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1780 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1781 else if (msr == MSR_IA32_CR_PAT)
1782 *pdata = vcpu->arch.pat;
1783 else { /* Variable MTRRs */
1784 int idx, is_mtrr_mask;
1785 u64 *pt;
1786
1787 idx = (msr - 0x200) / 2;
1788 is_mtrr_mask = msr - 0x200 - 2 * idx;
1789 if (!is_mtrr_mask)
1790 pt =
1791 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1792 else
1793 pt =
1794 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1795 *pdata = *pt;
1796 }
1797
9ba075a6
AK		 1798 return 0;
1799}
1800
890ca9ae 1801static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
15c4a640
CO		 1802{
1803 u64 data;
890ca9ae
HY		 1804 u64 mcg_cap = vcpu->arch.mcg_cap;
1805 unsigned bank_num = mcg_cap & 0xff;
15c4a640
CO		 1806
1807 switch (msr) {
15c4a640
CO		 1808 case MSR_IA32_P5_MC_ADDR:
1809 case MSR_IA32_P5_MC_TYPE:
890ca9ae
HY		 1810 data = 0;
1811 break;
15c4a640 1812 case MSR_IA32_MCG_CAP:
890ca9ae
HY		 1813 data = vcpu->arch.mcg_cap;
1814 break;
c7ac679c 1815 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 1816 if (!(mcg_cap & MCG_CTL_P))
1817 return 1;
1818 data = vcpu->arch.mcg_ctl;
1819 break;
1820 case MSR_IA32_MCG_STATUS:
1821 data = vcpu->arch.mcg_status;
1822 break;
1823 default:
1824 if (msr >= MSR_IA32_MC0_CTL &&
1825 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1826 u32 offset = msr - MSR_IA32_MC0_CTL;
1827 data = vcpu->arch.mce_banks[offset];
1828 break;
1829 }
1830 return 1;
1831 }
1832 *pdata = data;
1833 return 0;
1834}
1835
55cd8e5a
GN		 1836static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1837{
1838 u64 data = 0;
1839 struct kvm *kvm = vcpu->kvm;
1840
1841 switch (msr) {
1842 case HV_X64_MSR_GUEST_OS_ID:
1843 data = kvm->arch.hv_guest_os_id;
1844 break;
1845 case HV_X64_MSR_HYPERCALL:
1846 data = kvm->arch.hv_hypercall;
1847 break;
1848 default:
1849 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1850 return 1;
1851 }
1852
1853 *pdata = data;
1854 return 0;
1855}
1856
1857static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1858{
1859 u64 data = 0;
1860
1861 switch (msr) {
1862 case HV_X64_MSR_VP_INDEX: {
1863 int r;
1864 struct kvm_vcpu *v;
1865 kvm_for_each_vcpu(r, v, vcpu->kvm)
1866 if (v == vcpu)
1867 data = r;
1868 break;
1869 }
10388a07
GN		 1870 case HV_X64_MSR_EOI:
1871 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1872 case HV_X64_MSR_ICR:
1873 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1874 case HV_X64_MSR_TPR:
1875 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
14fa67ee 1876 case HV_X64_MSR_APIC_ASSIST_PAGE:
d1613ad5
MW		 1877 data = vcpu->arch.hv_vapic;
1878 break;
55cd8e5a
GN		 1879 default:
1880 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1881 return 1;
1882 }
1883 *pdata = data;
1884 return 0;
1885}
1886
890ca9ae
HY		 1887int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1888{
1889 u64 data;
1890
1891 switch (msr) {
890ca9ae 1892 case MSR_IA32_PLATFORM_ID:
15c4a640 1893 case MSR_IA32_EBL_CR_POWERON:
b5e2fec0
AG		 1894 case MSR_IA32_DEBUGCTLMSR:
1895 case MSR_IA32_LASTBRANCHFROMIP:
1896 case MSR_IA32_LASTBRANCHTOIP:
1897 case MSR_IA32_LASTINTFROMIP:
1898 case MSR_IA32_LASTINTTOIP:
60af2ecd
JSR		 1899 case MSR_K8_SYSCFG:
1900 case MSR_K7_HWCR:
61a6bd67 1901 case MSR_VM_HSAVE_PA:
9e699624 1902 case MSR_K7_EVNTSEL0:
1f3ee616 1903 case MSR_K7_PERFCTR0:
1fdbd48c 1904 case MSR_K8_INT_PENDING_MSG:
c323c0e5 1905 case MSR_AMD64_NB_CFG:
f7c6d140 1906 case MSR_FAM10H_MMIO_CONF_BASE:
15c4a640
CO		 1907 data = 0;
1908 break;
5753785f
GN		 1909 case MSR_P6_PERFCTR0:
1910 case MSR_P6_PERFCTR1:
1911 case MSR_P6_EVNTSEL0:
1912 case MSR_P6_EVNTSEL1:
1913 if (kvm_pmu_msr(vcpu, msr))
1914 return kvm_pmu_get_msr(vcpu, msr, pdata);
1915 data = 0;
1916 break;
742bc670
MT		 1917 case MSR_IA32_UCODE_REV:
1918 data = 0x100000000ULL;
1919 break;
9ba075a6
AK		 1920 case MSR_MTRRcap:
1921 data = 0x500 | KVM_NR_VAR_MTRR;
1922 break;
1923 case 0x200 ... 0x2ff:
1924 return get_msr_mtrr(vcpu, msr, pdata);
15c4a640
CO		 1925 case 0xcd: /* fsb frequency */
1926 data = 3;
1927 break;
7b914098
JS		 1928 /*
1929 * MSR_EBC_FREQUENCY_ID
1930 * Conservative value valid for even the basic CPU models.
1931 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1932 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1933 * and 266MHz for model 3, or 4. Set Core Clock
1934 * Frequency to System Bus Frequency Ratio to 1 (bits
1935 * 31:24) even though these are only valid for CPU
1936 * models > 2, however guests may end up dividing or
1937 * multiplying by zero otherwise.
1938 */
1939 case MSR_EBC_FREQUENCY_ID:
1940 data = 1 << 24;
1941 break;
15c4a640
CO		 1942 case MSR_IA32_APICBASE:
1943 data = kvm_get_apic_base(vcpu);
1944 break;
0105d1a5
GN		 1945 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1946 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1947 break;
a3e06bbe
LJ		 1948 case MSR_IA32_TSCDEADLINE:
1949 data = kvm_get_lapic_tscdeadline_msr(vcpu);
1950 break;
15c4a640 1951 case MSR_IA32_MISC_ENABLE:
ad312c7c 1952 data = vcpu->arch.ia32_misc_enable_msr;
15c4a640 1953 break;
847f0ad8
AG		 1954 case MSR_IA32_PERF_STATUS:
1955 /* TSC increment by tick */
1956 data = 1000ULL;
1957 /* CPU multiplier */
1958 data |= (((uint64_t)4ULL) << 40);
1959 break;
15c4a640 1960 case MSR_EFER:
f6801dff 1961 data = vcpu->arch.efer;
15c4a640 1962 break;
18068523 1963 case MSR_KVM_WALL_CLOCK:
11c6bffa 1964 case MSR_KVM_WALL_CLOCK_NEW:
18068523
GOC		 1965 data = vcpu->kvm->arch.wall_clock;
1966 break;
1967 case MSR_KVM_SYSTEM_TIME:
11c6bffa 1968 case MSR_KVM_SYSTEM_TIME_NEW:
18068523
GOC		 1969 data = vcpu->arch.time;
1970 break;
344d9588
GN		 1971 case MSR_KVM_ASYNC_PF_EN:
1972 data = vcpu->arch.apf.msr_val;
1973 break;
c9aaa895
GC		 1974 case MSR_KVM_STEAL_TIME:
1975 data = vcpu->arch.st.msr_val;
1976 break;
890ca9ae
HY		 1977 case MSR_IA32_P5_MC_ADDR:
1978 case MSR_IA32_P5_MC_TYPE:
1979 case MSR_IA32_MCG_CAP:
1980 case MSR_IA32_MCG_CTL:
1981 case MSR_IA32_MCG_STATUS:
1982 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1983 return get_msr_mce(vcpu, msr, pdata);
84e0cefa
JS		 1984 case MSR_K7_CLK_CTL:
1985 /*
1986 * Provide expected ramp-up count for K7. All other
1987 * are set to zero, indicating minimum divisors for
1988 * every field.
1989 *
1990 * This prevents guest kernels on AMD host with CPU
1991 * type 6, model 8 and higher from exploding due to
1992 * the rdmsr failing.
1993 */
1994 data = 0x20000000;
1995 break;
55cd8e5a
GN		 1996 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1997 if (kvm_hv_msr_partition_wide(msr)) {
1998 int r;
1999 mutex_lock(&vcpu->kvm->lock);
2000 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2001 mutex_unlock(&vcpu->kvm->lock);
2002 return r;
2003 } else
2004 return get_msr_hyperv(vcpu, msr, pdata);
2005 break;
91c9c3ed 2006 case MSR_IA32_BBL_CR_CTL3:
2007 /* This legacy MSR exists but isn't fully documented in current
2008 * silicon. It is however accessed by winxp in very narrow
2009 * scenarios where it sets bit #19, itself documented as
2010 * a "reserved" bit. Best effort attempt to source coherent
2011 * read data here should the balance of the register be
2012 * interpreted by the guest:
2013 *
2014 * L2 cache control register 3: 64GB range, 256KB size,
2015 * enabled, latency 0x1, configured
2016 */
2017 data = 0xbe702111;
2018 break;
2b036c6b
BO		 2019 case MSR_AMD64_OSVW_ID_LENGTH:
2020 if (!guest_cpuid_has_osvw(vcpu))
2021 return 1;
2022 data = vcpu->arch.osvw.length;
2023 break;
2024 case MSR_AMD64_OSVW_STATUS:
2025 if (!guest_cpuid_has_osvw(vcpu))
2026 return 1;
2027 data = vcpu->arch.osvw.status;
2028 break;
15c4a640 2029 default:
f5132b01
GN		 2030 if (kvm_pmu_msr(vcpu, msr))
2031 return kvm_pmu_get_msr(vcpu, msr, pdata);
ed85c068
AP		 2032 if (!ignore_msrs) {
2033 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2034 return 1;
2035 } else {
2036 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2037 data = 0;
2038 }
2039 break;
15c4a640
CO		 2040 }
2041 *pdata = data;
2042 return 0;
2043}
2044EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2045
313a3dc7
CO		 2046/*
2047 * Read or write a bunch of msrs. All parameters are kernel addresses.
2048 *
2049 * @return number of msrs set successfully.
2050 */
2051static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2052 struct kvm_msr_entry *entries,
2053 int (*do_msr)(struct kvm_vcpu *vcpu,
2054 unsigned index, u64 *data))
2055{
f656ce01 2056 int i, idx;
313a3dc7 2057
f656ce01 2058 idx = srcu_read_lock(&vcpu->kvm->srcu);
313a3dc7
CO		 2059 for (i = 0; i < msrs->nmsrs; ++i)
2060 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2061 break;
f656ce01 2062 srcu_read_unlock(&vcpu->kvm->srcu, idx);
313a3dc7 2063
313a3dc7
CO		 2064 return i;
2065}
2066
2067/*
2068 * Read or write a bunch of msrs. Parameters are user addresses.
2069 *
2070 * @return number of msrs set successfully.
2071 */
2072static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2073 int (*do_msr)(struct kvm_vcpu *vcpu,
2074 unsigned index, u64 *data),
2075 int writeback)
2076{
2077 struct kvm_msrs msrs;
2078 struct kvm_msr_entry *entries;
2079 int r, n;
2080 unsigned size;
2081
2082 r = -EFAULT;
2083 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2084 goto out;
2085
2086 r = -E2BIG;
2087 if (msrs.nmsrs >= MAX_IO_MSRS)
2088 goto out;
2089
313a3dc7 2090 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
ff5c2c03
SL		 2091 entries = memdup_user(user_msrs->entries, size);
2092 if (IS_ERR(entries)) {
2093 r = PTR_ERR(entries);
313a3dc7 2094 goto out;
ff5c2c03 2095 }
313a3dc7
CO		 2096
2097 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2098 if (r < 0)
2099 goto out_free;
2100
2101 r = -EFAULT;
2102 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2103 goto out_free;
2104
2105 r = n;
2106
2107out_free:
7a73c028 2108 kfree(entries);
313a3dc7
CO		 2109out:
2110 return r;
2111}
2112
018d00d2
ZX		 2113int kvm_dev_ioctl_check_extension(long ext)
2114{
2115 int r;
2116
2117 switch (ext) {
2118 case KVM_CAP_IRQCHIP:
2119 case KVM_CAP_HLT:
2120 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
018d00d2 2121 case KVM_CAP_SET_TSS_ADDR:
07716717 2122 case KVM_CAP_EXT_CPUID:
c8076604 2123 case KVM_CAP_CLOCKSOURCE:
7837699f 2124 case KVM_CAP_PIT:
a28e4f5a 2125 case KVM_CAP_NOP_IO_DELAY:
62d9f0db 2126 case KVM_CAP_MP_STATE:
ed848624 2127 case KVM_CAP_SYNC_MMU:
a355c85c 2128 case KVM_CAP_USER_NMI:
52d939a0 2129 case KVM_CAP_REINJECT_CONTROL:
4925663a 2130 case KVM_CAP_IRQ_INJECT_STATUS:
e56d532f 2131 case KVM_CAP_ASSIGN_DEV_IRQ:
721eecbf 2132 case KVM_CAP_IRQFD:
d34e6b17 2133 case KVM_CAP_IOEVENTFD:
c5ff41ce 2134 case KVM_CAP_PIT2:
e9f42757 2135 case KVM_CAP_PIT_STATE2:
b927a3ce 2136 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
ffde22ac 2137 case KVM_CAP_XEN_HVM:
afbcf7ab 2138 case KVM_CAP_ADJUST_CLOCK:
3cfc3092 2139 case KVM_CAP_VCPU_EVENTS:
55cd8e5a 2140 case KVM_CAP_HYPERV:
10388a07 2141 case KVM_CAP_HYPERV_VAPIC:
c25bc163 2142 case KVM_CAP_HYPERV_SPIN:
ab9f4ecb 2143 case KVM_CAP_PCI_SEGMENT:
a1efbe77 2144 case KVM_CAP_DEBUGREGS:
d2be1651 2145 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2d5b5a66 2146 case KVM_CAP_XSAVE:
344d9588 2147 case KVM_CAP_ASYNC_PF:
92a1f12d 2148 case KVM_CAP_GET_TSC_KHZ:
07700a94 2149 case KVM_CAP_PCI_2_3:
1c0b28c2 2150 case KVM_CAP_KVMCLOCK_CTRL:
018d00d2
ZX		 2151 r = 1;
2152 break;
542472b5
LV		 2153 case KVM_CAP_COALESCED_MMIO:
2154 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2155 break;
774ead3a
AK		 2156 case KVM_CAP_VAPIC:
2157 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2158 break;
f725230a 2159 case KVM_CAP_NR_VCPUS:
8c3ba334
SL		 2160 r = KVM_SOFT_MAX_VCPUS;
2161 break;
2162 case KVM_CAP_MAX_VCPUS:
f725230a
AK		 2163 r = KVM_MAX_VCPUS;
2164 break;
a988b910
AK		 2165 case KVM_CAP_NR_MEMSLOTS:
2166 r = KVM_MEMORY_SLOTS;
2167 break;
a68a6a72
MT		 2168 case KVM_CAP_PV_MMU: /* obsolete */
2169 r = 0;
2f333bcb 2170 break;
62c476c7 2171 case KVM_CAP_IOMMU:
a1b60c1c 2172 r = iommu_present(&pci_bus_type);
62c476c7 2173 break;
890ca9ae
HY		 2174 case KVM_CAP_MCE:
2175 r = KVM_MAX_MCE_BANKS;
2176 break;
2d5b5a66
SY		 2177 case KVM_CAP_XCRS:
2178 r = cpu_has_xsave;
2179 break;
92a1f12d
JR		 2180 case KVM_CAP_TSC_CONTROL:
2181 r = kvm_has_tsc_control;
2182 break;
4d25a066
JK		 2183 case KVM_CAP_TSC_DEADLINE_TIMER:
2184 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2185 break;
018d00d2
ZX		 2186 default:
2187 r = 0;
2188 break;
2189 }
2190 return r;
2191
2192}
2193
043405e1
CO		 2194long kvm_arch_dev_ioctl(struct file *filp,
2195 unsigned int ioctl, unsigned long arg)
2196{
2197 void __user *argp = (void __user *)arg;
2198 long r;
2199
2200 switch (ioctl) {
2201 case KVM_GET_MSR_INDEX_LIST: {
2202 struct kvm_msr_list __user *user_msr_list = argp;
2203 struct kvm_msr_list msr_list;
2204 unsigned n;
2205
2206 r = -EFAULT;
2207 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2208 goto out;
2209 n = msr_list.nmsrs;
2210 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2211 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2212 goto out;
2213 r = -E2BIG;
e125e7b6 2214 if (n < msr_list.nmsrs)
043405e1
CO		 2215 goto out;
2216 r = -EFAULT;
2217 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2218 num_msrs_to_save * sizeof(u32)))
2219 goto out;
e125e7b6 2220 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
043405e1
CO		 2221 &emulated_msrs,
2222 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2223 goto out;
2224 r = 0;
2225 break;
2226 }
674eea0f
AK		 2227 case KVM_GET_SUPPORTED_CPUID: {
2228 struct kvm_cpuid2 __user *cpuid_arg = argp;
2229 struct kvm_cpuid2 cpuid;
2230
2231 r = -EFAULT;
2232 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2233 goto out;
2234 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
19355475 2235 cpuid_arg->entries);
674eea0f
AK		 2236 if (r)
2237 goto out;
2238
2239 r = -EFAULT;
2240 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2241 goto out;
2242 r = 0;
2243 break;
2244 }
890ca9ae
HY		 2245 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2246 u64 mce_cap;
2247
2248 mce_cap = KVM_MCE_CAP_SUPPORTED;
2249 r = -EFAULT;
2250 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2251 goto out;
2252 r = 0;
2253 break;
2254 }
043405e1
CO		 2255 default:
2256 r = -EINVAL;
2257 }
2258out:
2259 return r;
2260}
2261
f5f48ee1
SY		 2262static void wbinvd_ipi(void *garbage)
2263{
2264 wbinvd();
2265}
2266
2267static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2268{
2269 return vcpu->kvm->arch.iommu_domain &&
2270 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2271}
2272
313a3dc7
CO		 2273void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2274{
f5f48ee1
SY		 2275 /* Address WBINVD may be executed by guest */
2276 if (need_emulate_wbinvd(vcpu)) {
2277 if (kvm_x86_ops->has_wbinvd_exit())
2278 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2279 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2280 smp_call_function_single(vcpu->cpu,
2281 wbinvd_ipi, NULL, 1);
2282 }
2283
313a3dc7 2284 kvm_x86_ops->vcpu_load(vcpu, cpu);
8f6055cb 2285
0dd6a6ed
ZA		 2286 /* Apply any externally detected TSC adjustments (due to suspend) */
2287 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2288 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2289 vcpu->arch.tsc_offset_adjustment = 0;
2290 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2291 }
8f6055cb 2292
48434c20 2293 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
6f526ec5
ZA		 2294 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2295 native_read_tsc() - vcpu->arch.last_host_tsc;
e48672fa
ZA		 2296 if (tsc_delta < 0)
2297 mark_tsc_unstable("KVM discovered backwards TSC");
c285545f 2298 if (check_tsc_unstable()) {
b183aa58
ZA		 2299 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2300 vcpu->arch.last_guest_tsc);
2301 kvm_x86_ops->write_tsc_offset(vcpu, offset);
c285545f 2302 vcpu->arch.tsc_catchup = 1;
c285545f 2303 }
1aa8ceef 2304 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
c285545f
ZA		 2305 if (vcpu->cpu != cpu)
2306 kvm_migrate_timers(vcpu);
e48672fa 2307 vcpu->cpu = cpu;
6b7d7e76 2308 }
c9aaa895
GC		 2309
2310 accumulate_steal_time(vcpu);
2311 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
313a3dc7
CO		 2312}
2313
2314void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2315{
02daab21 2316 kvm_x86_ops->vcpu_put(vcpu);
1c11e713 2317 kvm_put_guest_fpu(vcpu);
6f526ec5 2318 vcpu->arch.last_host_tsc = native_read_tsc();
313a3dc7
CO		 2319}
2320
313a3dc7
CO		 2321static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2322 struct kvm_lapic_state *s)
2323{
ad312c7c 2324 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
313a3dc7
CO		 2325
2326 return 0;
2327}
2328
2329static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2330 struct kvm_lapic_state *s)
2331{
ad312c7c 2332 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
313a3dc7 2333 kvm_apic_post_state_restore(vcpu);
cb142eb7 2334 update_cr8_intercept(vcpu);
313a3dc7
CO		 2335
2336 return 0;
2337}
2338
f77bc6a4
ZX		 2339static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2340 struct kvm_interrupt *irq)
2341{
2342 if (irq->irq < 0 || irq->irq >= 256)
2343 return -EINVAL;
2344 if (irqchip_in_kernel(vcpu->kvm))
2345 return -ENXIO;
f77bc6a4 2346
66fd3f7f 2347 kvm_queue_interrupt(vcpu, irq->irq, false);
3842d135 2348 kvm_make_request(KVM_REQ_EVENT, vcpu);
f77bc6a4 2349
f77bc6a4
ZX		 2350 return 0;
2351}
2352
c4abb7c9
JK		 2353static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2354{
c4abb7c9 2355 kvm_inject_nmi(vcpu);
c4abb7c9
JK		 2356
2357 return 0;
2358}
2359
b209749f
AK		 2360static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2361 struct kvm_tpr_access_ctl *tac)
2362{
2363 if (tac->flags)
2364 return -EINVAL;
2365 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2366 return 0;
2367}
2368
890ca9ae
HY		 2369static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2370 u64 mcg_cap)
2371{
2372 int r;
2373 unsigned bank_num = mcg_cap & 0xff, bank;
2374
2375 r = -EINVAL;
a9e38c3e 2376 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
890ca9ae
HY		 2377 goto out;
2378 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2379 goto out;
2380 r = 0;
2381 vcpu->arch.mcg_cap = mcg_cap;
2382 /* Init IA32_MCG_CTL to all 1s */
2383 if (mcg_cap & MCG_CTL_P)
2384 vcpu->arch.mcg_ctl = ~(u64)0;
2385 /* Init IA32_MCi_CTL to all 1s */
2386 for (bank = 0; bank < bank_num; bank++)
2387 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2388out:
2389 return r;
2390}
2391
2392static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2393 struct kvm_x86_mce *mce)
2394{
2395 u64 mcg_cap = vcpu->arch.mcg_cap;
2396 unsigned bank_num = mcg_cap & 0xff;
2397 u64 *banks = vcpu->arch.mce_banks;
2398
2399 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2400 return -EINVAL;
2401 /*
2402 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2403 * reporting is disabled
2404 */
2405 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2406 vcpu->arch.mcg_ctl != ~(u64)0)
2407 return 0;
2408 banks += 4 * mce->bank;
2409 /*
2410 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2411 * reporting is disabled for the bank
2412 */
2413 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2414 return 0;
2415 if (mce->status & MCI_STATUS_UC) {
2416 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
fc78f519 2417 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
a8eeb04a 2418 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
890ca9ae
HY		 2419 return 0;
2420 }
2421 if (banks[1] & MCI_STATUS_VAL)
2422 mce->status |= MCI_STATUS_OVER;
2423 banks[2] = mce->addr;
2424 banks[3] = mce->misc;
2425 vcpu->arch.mcg_status = mce->mcg_status;
2426 banks[1] = mce->status;
2427 kvm_queue_exception(vcpu, MC_VECTOR);
2428 } else if (!(banks[1] & MCI_STATUS_VAL)
2429 || !(banks[1] & MCI_STATUS_UC)) {
2430 if (banks[1] & MCI_STATUS_VAL)
2431 mce->status |= MCI_STATUS_OVER;
2432 banks[2] = mce->addr;
2433 banks[3] = mce->misc;
2434 banks[1] = mce->status;
2435 } else
2436 banks[1] |= MCI_STATUS_OVER;
2437 return 0;
2438}
2439
3cfc3092
JK		 2440static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2441 struct kvm_vcpu_events *events)
2442{
7460fb4a 2443 process_nmi(vcpu);
03b82a30
JK		 2444 events->exception.injected =
2445 vcpu->arch.exception.pending &&
2446 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3cfc3092
JK		 2447 events->exception.nr = vcpu->arch.exception.nr;
2448 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
97e69aa6 2449 events->exception.pad = 0;
3cfc3092
JK		 2450 events->exception.error_code = vcpu->arch.exception.error_code;
2451
03b82a30
JK		 2452 events->interrupt.injected =
2453 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3cfc3092 2454 events->interrupt.nr = vcpu->arch.interrupt.nr;
03b82a30 2455 events->interrupt.soft = 0;
48005f64
JK		 2456 events->interrupt.shadow =
2457 kvm_x86_ops->get_interrupt_shadow(vcpu,
2458 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
3cfc3092
JK		 2459
2460 events->nmi.injected = vcpu->arch.nmi_injected;
7460fb4a 2461 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3cfc3092 2462 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
97e69aa6 2463 events->nmi.pad = 0;
3cfc3092
JK		 2464
2465 events->sipi_vector = vcpu->arch.sipi_vector;
2466
dab4b911 2467 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64
JK		 2468 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2469 | KVM_VCPUEVENT_VALID_SHADOW);
97e69aa6 2470 memset(&events->reserved, 0, sizeof(events->reserved));
3cfc3092
JK		 2471}
2472
2473static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2474 struct kvm_vcpu_events *events)
2475{
dab4b911 2476 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64
JK		 2477 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2478 | KVM_VCPUEVENT_VALID_SHADOW))
3cfc3092
JK		 2479 return -EINVAL;
2480
7460fb4a 2481 process_nmi(vcpu);
3cfc3092
JK		 2482 vcpu->arch.exception.pending = events->exception.injected;
2483 vcpu->arch.exception.nr = events->exception.nr;
2484 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2485 vcpu->arch.exception.error_code = events->exception.error_code;
2486
2487 vcpu->arch.interrupt.pending = events->interrupt.injected;
2488 vcpu->arch.interrupt.nr = events->interrupt.nr;
2489 vcpu->arch.interrupt.soft = events->interrupt.soft;
48005f64
JK		 2490 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2491 kvm_x86_ops->set_interrupt_shadow(vcpu,
2492 events->interrupt.shadow);
3cfc3092
JK		 2493
2494 vcpu->arch.nmi_injected = events->nmi.injected;
dab4b911
JK		 2495 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2496 vcpu->arch.nmi_pending = events->nmi.pending;
3cfc3092
JK		 2497 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2498
dab4b911
JK		 2499 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2500 vcpu->arch.sipi_vector = events->sipi_vector;
3cfc3092 2501
3842d135
AK		 2502 kvm_make_request(KVM_REQ_EVENT, vcpu);
2503
3cfc3092
JK		 2504 return 0;
2505}
2506
a1efbe77
JK		 2507static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2508 struct kvm_debugregs *dbgregs)
2509{
a1efbe77
JK		 2510 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2511 dbgregs->dr6 = vcpu->arch.dr6;
2512 dbgregs->dr7 = vcpu->arch.dr7;
2513 dbgregs->flags = 0;
97e69aa6 2514 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
a1efbe77
JK		 2515}
2516
2517static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2518 struct kvm_debugregs *dbgregs)
2519{
2520 if (dbgregs->flags)
2521 return -EINVAL;
2522
a1efbe77
JK		 2523 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2524 vcpu->arch.dr6 = dbgregs->dr6;
2525 vcpu->arch.dr7 = dbgregs->dr7;
2526
a1efbe77
JK		 2527 return 0;
2528}
2529
2d5b5a66
SY		 2530static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2531 struct kvm_xsave *guest_xsave)
2532{
2533 if (cpu_has_xsave)
2534 memcpy(guest_xsave->region,
2535 &vcpu->arch.guest_fpu.state->xsave,
f45755b8 2536 xstate_size);
2d5b5a66
SY		 2537 else {
2538 memcpy(guest_xsave->region,
2539 &vcpu->arch.guest_fpu.state->fxsave,
2540 sizeof(struct i387_fxsave_struct));
2541 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2542 XSTATE_FPSSE;
2543 }
2544}
2545
2546static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2547 struct kvm_xsave *guest_xsave)
2548{
2549 u64 xstate_bv =
2550 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2551
2552 if (cpu_has_xsave)
2553 memcpy(&vcpu->arch.guest_fpu.state->xsave,
f45755b8 2554 guest_xsave->region, xstate_size);
2d5b5a66
SY		 2555 else {
2556 if (xstate_bv & ~XSTATE_FPSSE)
2557 return -EINVAL;
2558 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2559 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2560 }
2561 return 0;
2562}
2563
2564static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2565 struct kvm_xcrs *guest_xcrs)
2566{
2567 if (!cpu_has_xsave) {
2568 guest_xcrs->nr_xcrs = 0;
2569 return;
2570 }
2571
2572 guest_xcrs->nr_xcrs = 1;
2573 guest_xcrs->flags = 0;
2574 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2575 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2576}
2577
2578static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2579 struct kvm_xcrs *guest_xcrs)
2580{
2581 int i, r = 0;
2582
2583 if (!cpu_has_xsave)
2584 return -EINVAL;
2585
2586 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2587 return -EINVAL;
2588
2589 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2590 /* Only support XCR0 currently */
2591 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2592 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2593 guest_xcrs->xcrs[0].value);
2594 break;
2595 }
2596 if (r)
2597 r = -EINVAL;
2598 return r;
2599}
2600
1c0b28c2
EM		 2601/*
2602 * kvm_set_guest_paused() indicates to the guest kernel that it has been
2603 * stopped by the hypervisor. This function will be called from the host only.
2604 * EINVAL is returned when the host attempts to set the flag for a guest that
2605 * does not support pv clocks.
2606 */
2607static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
2608{
2609 struct pvclock_vcpu_time_info *src = &vcpu->arch.hv_clock;
2610 if (!vcpu->arch.time_page)
2611 return -EINVAL;
2612 src->flags |= PVCLOCK_GUEST_STOPPED;
2613 mark_page_dirty(vcpu->kvm, vcpu->arch.time >> PAGE_SHIFT);
2614 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2615 return 0;
2616}
2617
313a3dc7
CO		 2618long kvm_arch_vcpu_ioctl(struct file *filp,
2619 unsigned int ioctl, unsigned long arg)
2620{
2621 struct kvm_vcpu *vcpu = filp->private_data;
2622 void __user *argp = (void __user *)arg;
2623 int r;
d1ac91d8
AK		 2624 union {
2625 struct kvm_lapic_state *lapic;
2626 struct kvm_xsave *xsave;
2627 struct kvm_xcrs *xcrs;
2628 void *buffer;
2629 } u;
2630
2631 u.buffer = NULL;
313a3dc7
CO		 2632 switch (ioctl) {
2633 case KVM_GET_LAPIC: {
2204ae3c
MT		 2634 r = -EINVAL;
2635 if (!vcpu->arch.apic)
2636 goto out;
d1ac91d8 2637 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
313a3dc7 2638
b772ff36 2639 r = -ENOMEM;
d1ac91d8 2640 if (!u.lapic)
b772ff36 2641 goto out;
d1ac91d8 2642 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
313a3dc7
CO		 2643 if (r)
2644 goto out;
2645 r = -EFAULT;
d1ac91d8 2646 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
313a3dc7
CO		 2647 goto out;
2648 r = 0;
2649 break;
2650 }
2651 case KVM_SET_LAPIC: {
2204ae3c
MT		 2652 r = -EINVAL;
2653 if (!vcpu->arch.apic)
2654 goto out;
ff5c2c03
SL		 2655 u.lapic = memdup_user(argp, sizeof(*u.lapic));
2656 if (IS_ERR(u.lapic)) {
2657 r = PTR_ERR(u.lapic);
313a3dc7 2658 goto out;
ff5c2c03
SL		 2659 }
2660
d1ac91d8 2661 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
313a3dc7
CO		 2662 if (r)
2663 goto out;
2664 r = 0;
2665 break;
2666 }
f77bc6a4
ZX		 2667 case KVM_INTERRUPT: {
2668 struct kvm_interrupt irq;
2669
2670 r = -EFAULT;
2671 if (copy_from_user(&irq, argp, sizeof irq))
2672 goto out;
2673 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2674 if (r)
2675 goto out;
2676 r = 0;
2677 break;
2678 }
c4abb7c9
JK		 2679 case KVM_NMI: {
2680 r = kvm_vcpu_ioctl_nmi(vcpu);
2681 if (r)
2682 goto out;
2683 r = 0;
2684 break;
2685 }
313a3dc7
CO		 2686 case KVM_SET_CPUID: {
2687 struct kvm_cpuid __user *cpuid_arg = argp;
2688 struct kvm_cpuid cpuid;
2689
2690 r = -EFAULT;
2691 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2692 goto out;
2693 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2694 if (r)
2695 goto out;
2696 break;
2697 }
07716717
DK		 2698 case KVM_SET_CPUID2: {
2699 struct kvm_cpuid2 __user *cpuid_arg = argp;
2700 struct kvm_cpuid2 cpuid;
2701
2702 r = -EFAULT;
2703 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2704 goto out;
2705 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
19355475 2706 cpuid_arg->entries);
07716717
DK		 2707 if (r)
2708 goto out;
2709 break;
2710 }
2711 case KVM_GET_CPUID2: {
2712 struct kvm_cpuid2 __user *cpuid_arg = argp;
2713 struct kvm_cpuid2 cpuid;
2714
2715 r = -EFAULT;
2716 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2717 goto out;
2718 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
19355475 2719 cpuid_arg->entries);
07716717
DK		 2720 if (r)
2721 goto out;
2722 r = -EFAULT;
2723 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2724 goto out;
2725 r = 0;
2726 break;
2727 }
313a3dc7
CO		 2728 case KVM_GET_MSRS:
2729 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2730 break;
2731 case KVM_SET_MSRS:
2732 r = msr_io(vcpu, argp, do_set_msr, 0);
2733 break;
b209749f
AK		 2734 case KVM_TPR_ACCESS_REPORTING: {
2735 struct kvm_tpr_access_ctl tac;
2736
2737 r = -EFAULT;
2738 if (copy_from_user(&tac, argp, sizeof tac))
2739 goto out;
2740 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2741 if (r)
2742 goto out;
2743 r = -EFAULT;
2744 if (copy_to_user(argp, &tac, sizeof tac))
2745 goto out;
2746 r = 0;
2747 break;
2748 };
b93463aa
AK		 2749 case KVM_SET_VAPIC_ADDR: {
2750 struct kvm_vapic_addr va;
2751
2752 r = -EINVAL;
2753 if (!irqchip_in_kernel(vcpu->kvm))
2754 goto out;
2755 r = -EFAULT;
2756 if (copy_from_user(&va, argp, sizeof va))
2757 goto out;
2758 r = 0;
2759 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2760 break;
2761 }
890ca9ae
HY		 2762 case KVM_X86_SETUP_MCE: {
2763 u64 mcg_cap;
2764
2765 r = -EFAULT;
2766 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2767 goto out;
2768 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2769 break;
2770 }
2771 case KVM_X86_SET_MCE: {
2772 struct kvm_x86_mce mce;
2773
2774 r = -EFAULT;
2775 if (copy_from_user(&mce, argp, sizeof mce))
2776 goto out;
2777 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2778 break;
2779 }
3cfc3092
JK		 2780 case KVM_GET_VCPU_EVENTS: {
2781 struct kvm_vcpu_events events;
2782
2783 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2784
2785 r = -EFAULT;
2786 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2787 break;
2788 r = 0;
2789 break;
2790 }
2791 case KVM_SET_VCPU_EVENTS: {
2792 struct kvm_vcpu_events events;
2793
2794 r = -EFAULT;
2795 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2796 break;
2797
2798 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2799 break;
2800 }
a1efbe77
JK		 2801 case KVM_GET_DEBUGREGS: {
2802 struct kvm_debugregs dbgregs;
2803
2804 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2805
2806 r = -EFAULT;
2807 if (copy_to_user(argp, &dbgregs,
2808 sizeof(struct kvm_debugregs)))
2809 break;
2810 r = 0;
2811 break;
2812 }
2813 case KVM_SET_DEBUGREGS: {
2814 struct kvm_debugregs dbgregs;
2815
2816 r = -EFAULT;
2817 if (copy_from_user(&dbgregs, argp,
2818 sizeof(struct kvm_debugregs)))
2819 break;
2820
2821 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2822 break;
2823 }
2d5b5a66 2824 case KVM_GET_XSAVE: {
d1ac91d8 2825 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2d5b5a66 2826 r = -ENOMEM;
d1ac91d8 2827 if (!u.xsave)
2d5b5a66
SY		 2828 break;
2829
d1ac91d8 2830 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2d5b5a66
SY		 2831
2832 r = -EFAULT;
d1ac91d8 2833 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2d5b5a66
SY		 2834 break;
2835 r = 0;
2836 break;
2837 }
2838 case KVM_SET_XSAVE: {
ff5c2c03
SL		 2839 u.xsave = memdup_user(argp, sizeof(*u.xsave));
2840 if (IS_ERR(u.xsave)) {
2841 r = PTR_ERR(u.xsave);
2842 goto out;
2843 }
2d5b5a66 2844
d1ac91d8 2845 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2d5b5a66
SY		 2846 break;
2847 }
2848 case KVM_GET_XCRS: {
d1ac91d8 2849 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2d5b5a66 2850 r = -ENOMEM;
d1ac91d8 2851 if (!u.xcrs)
2d5b5a66
SY		 2852 break;
2853
d1ac91d8 2854 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 2855
2856 r = -EFAULT;
d1ac91d8 2857 if (copy_to_user(argp, u.xcrs,
2d5b5a66
SY		 2858 sizeof(struct kvm_xcrs)))
2859 break;
2860 r = 0;
2861 break;
2862 }
2863 case KVM_SET_XCRS: {
ff5c2c03
SL		 2864 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
2865 if (IS_ERR(u.xcrs)) {
2866 r = PTR_ERR(u.xcrs);
2867 goto out;
2868 }
2d5b5a66 2869
d1ac91d8 2870 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 2871 break;
2872 }
92a1f12d
JR		 2873 case KVM_SET_TSC_KHZ: {
2874 u32 user_tsc_khz;
2875
2876 r = -EINVAL;
92a1f12d
JR		 2877 user_tsc_khz = (u32)arg;
2878
2879 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
2880 goto out;
2881
cc578287
ZA		 2882 if (user_tsc_khz == 0)
2883 user_tsc_khz = tsc_khz;
2884
2885 kvm_set_tsc_khz(vcpu, user_tsc_khz);
92a1f12d
JR		 2886
2887 r = 0;
2888 goto out;
2889 }
2890 case KVM_GET_TSC_KHZ: {
cc578287 2891 r = vcpu->arch.virtual_tsc_khz;
92a1f12d
JR		 2892 goto out;
2893 }
1c0b28c2
EM		 2894 case KVM_KVMCLOCK_CTRL: {
2895 r = kvm_set_guest_paused(vcpu);
2896 goto out;
2897 }
313a3dc7
CO		 2898 default:
2899 r = -EINVAL;
2900 }
2901out:
d1ac91d8 2902 kfree(u.buffer);
313a3dc7
CO		 2903 return r;
2904}
2905
5b1c1493
CO		 2906int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2907{
2908 return VM_FAULT_SIGBUS;
2909}
2910
1fe779f8
CO		 2911static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2912{
2913 int ret;
2914
2915 if (addr > (unsigned int)(-3 * PAGE_SIZE))
2916 return -1;
2917 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2918 return ret;
2919}
2920
b927a3ce
SY		 2921static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2922 u64 ident_addr)
2923{
2924 kvm->arch.ept_identity_map_addr = ident_addr;
2925 return 0;
2926}
2927
1fe779f8
CO		 2928static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2929 u32 kvm_nr_mmu_pages)
2930{
2931 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2932 return -EINVAL;
2933
79fac95e 2934 mutex_lock(&kvm->slots_lock);
7c8a83b7 2935 spin_lock(&kvm->mmu_lock);
1fe779f8
CO		 2936
2937 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
f05e70ac 2938 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1fe779f8 2939
7c8a83b7 2940 spin_unlock(&kvm->mmu_lock);
79fac95e 2941 mutex_unlock(&kvm->slots_lock);
1fe779f8
CO		 2942 return 0;
2943}
2944
2945static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2946{
39de71ec 2947 return kvm->arch.n_max_mmu_pages;
1fe779f8
CO		 2948}
2949
1fe779f8
CO		 2950static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2951{
2952 int r;
2953
2954 r = 0;
2955 switch (chip->chip_id) {
2956 case KVM_IRQCHIP_PIC_MASTER:
2957 memcpy(&chip->chip.pic,
2958 &pic_irqchip(kvm)->pics[0],
2959 sizeof(struct kvm_pic_state));
2960 break;
2961 case KVM_IRQCHIP_PIC_SLAVE:
2962 memcpy(&chip->chip.pic,
2963 &pic_irqchip(kvm)->pics[1],
2964 sizeof(struct kvm_pic_state));
2965 break;
2966 case KVM_IRQCHIP_IOAPIC:
eba0226b 2967 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 2968 break;
2969 default:
2970 r = -EINVAL;
2971 break;
2972 }
2973 return r;
2974}
2975
2976static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2977{
2978 int r;
2979
2980 r = 0;
2981 switch (chip->chip_id) {
2982 case KVM_IRQCHIP_PIC_MASTER:
f4f51050 2983 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 2984 memcpy(&pic_irqchip(kvm)->pics[0],
2985 &chip->chip.pic,
2986 sizeof(struct kvm_pic_state));
f4f51050 2987 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 2988 break;
2989 case KVM_IRQCHIP_PIC_SLAVE:
f4f51050 2990 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 2991 memcpy(&pic_irqchip(kvm)->pics[1],
2992 &chip->chip.pic,
2993 sizeof(struct kvm_pic_state));
f4f51050 2994 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 2995 break;
2996 case KVM_IRQCHIP_IOAPIC:
eba0226b 2997 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 2998 break;
2999 default:
3000 r = -EINVAL;
3001 break;
3002 }
3003 kvm_pic_update_irq(pic_irqchip(kvm));
3004 return r;
3005}
3006
e0f63cb9
SY		 3007static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3008{
3009 int r = 0;
3010
894a9c55 3011 mutex_lock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9 3012 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
894a9c55 3013 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9
SY		 3014 return r;
3015}
3016
3017static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3018{
3019 int r = 0;
3020
894a9c55 3021 mutex_lock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9 3022 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
e9f42757
BK		 3023 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3024 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3025 return r;
3026}
3027
3028static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3029{
3030 int r = 0;
3031
3032 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3033 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3034 sizeof(ps->channels));
3035 ps->flags = kvm->arch.vpit->pit_state.flags;
3036 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
97e69aa6 3037 memset(&ps->reserved, 0, sizeof(ps->reserved));
e9f42757
BK		 3038 return r;
3039}
3040
3041static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3042{
3043 int r = 0, start = 0;
3044 u32 prev_legacy, cur_legacy;
3045 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3046 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3047 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3048 if (!prev_legacy && cur_legacy)
3049 start = 1;
3050 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3051 sizeof(kvm->arch.vpit->pit_state.channels));
3052 kvm->arch.vpit->pit_state.flags = ps->flags;
3053 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
894a9c55 3054 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9
SY		 3055 return r;
3056}
3057
52d939a0
MT		 3058static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3059 struct kvm_reinject_control *control)
3060{
3061 if (!kvm->arch.vpit)
3062 return -ENXIO;
894a9c55 3063 mutex_lock(&kvm->arch.vpit->pit_state.lock);
52d939a0 3064 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
894a9c55 3065 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
52d939a0
MT		 3066 return 0;
3067}
3068
95d4c16c 3069/**
60c34612
TY		 3070 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3071 * @kvm: kvm instance
3072 * @log: slot id and address to which we copy the log
95d4c16c 3073 *
60c34612
TY		 3074 * We need to keep it in mind that VCPU threads can write to the bitmap
3075 * concurrently. So, to avoid losing data, we keep the following order for
3076 * each bit:
95d4c16c 3077 *
60c34612
TY		 3078 * 1. Take a snapshot of the bit and clear it if needed.
3079 * 2. Write protect the corresponding page.
3080 * 3. Flush TLB's if needed.
3081 * 4. Copy the snapshot to the userspace.
95d4c16c 3082 *
60c34612
TY		 3083 * Between 2 and 3, the guest may write to the page using the remaining TLB
3084 * entry. This is not a problem because the page will be reported dirty at
3085 * step 4 using the snapshot taken before and step 3 ensures that successive
3086 * writes will be logged for the next call.
95d4c16c 3087 */
60c34612 3088int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
5bb064dc 3089{
7850ac54 3090 int r;
5bb064dc 3091 struct kvm_memory_slot *memslot;
60c34612
TY		 3092 unsigned long n, i;
3093 unsigned long *dirty_bitmap;
3094 unsigned long *dirty_bitmap_buffer;
3095 bool is_dirty = false;
5bb064dc 3096
79fac95e 3097 mutex_lock(&kvm->slots_lock);
5bb064dc 3098
b050b015
MT		 3099 r = -EINVAL;
3100 if (log->slot >= KVM_MEMORY_SLOTS)
3101 goto out;
3102
28a37544 3103 memslot = id_to_memslot(kvm->memslots, log->slot);
60c34612
TY		 3104
3105 dirty_bitmap = memslot->dirty_bitmap;
b050b015 3106 r = -ENOENT;
60c34612 3107 if (!dirty_bitmap)
b050b015
MT		 3108 goto out;
3109
87bf6e7d 3110 n = kvm_dirty_bitmap_bytes(memslot);
b050b015 3111
60c34612
TY		 3112 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3113 memset(dirty_bitmap_buffer, 0, n);
b050b015 3114
60c34612 3115 spin_lock(&kvm->mmu_lock);
b050b015 3116
60c34612
TY		 3117 for (i = 0; i < n / sizeof(long); i++) {
3118 unsigned long mask;
3119 gfn_t offset;
cdfca7b3 3120
60c34612
TY		 3121 if (!dirty_bitmap[i])
3122 continue;
b050b015 3123
60c34612 3124 is_dirty = true;
914ebccd 3125
60c34612
TY		 3126 mask = xchg(&dirty_bitmap[i], 0);
3127 dirty_bitmap_buffer[i] = mask;
edde99ce 3128
60c34612
TY		 3129 offset = i * BITS_PER_LONG;
3130 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
5bb064dc 3131 }
60c34612
TY		 3132 if (is_dirty)
3133 kvm_flush_remote_tlbs(kvm);
3134
3135 spin_unlock(&kvm->mmu_lock);
3136
3137 r = -EFAULT;
3138 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3139 goto out;
b050b015 3140
5bb064dc
ZX		 3141 r = 0;
3142out:
79fac95e 3143 mutex_unlock(&kvm->slots_lock);
5bb064dc
ZX		 3144 return r;
3145}
3146
1fe779f8
CO		 3147long kvm_arch_vm_ioctl(struct file *filp,
3148 unsigned int ioctl, unsigned long arg)
3149{
3150 struct kvm *kvm = filp->private_data;
3151 void __user *argp = (void __user *)arg;
367e1319 3152 int r = -ENOTTY;
f0d66275
DH		 3153 /*
3154 * This union makes it completely explicit to gcc-3.x
3155 * that these two variables' stack usage should be
3156 * combined, not added together.
3157 */
3158 union {
3159 struct kvm_pit_state ps;
e9f42757 3160 struct kvm_pit_state2 ps2;
c5ff41ce 3161 struct kvm_pit_config pit_config;
f0d66275 3162 } u;
1fe779f8
CO		 3163
3164 switch (ioctl) {
3165 case KVM_SET_TSS_ADDR:
3166 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3167 if (r < 0)
3168 goto out;
3169 break;
b927a3ce
SY		 3170 case KVM_SET_IDENTITY_MAP_ADDR: {
3171 u64 ident_addr;
3172
3173 r = -EFAULT;
3174 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3175 goto out;
3176 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3177 if (r < 0)
3178 goto out;
3179 break;
3180 }
1fe779f8
CO		 3181 case KVM_SET_NR_MMU_PAGES:
3182 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3183 if (r)
3184 goto out;
3185 break;
3186 case KVM_GET_NR_MMU_PAGES:
3187 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3188 break;
3ddea128
MT		 3189 case KVM_CREATE_IRQCHIP: {
3190 struct kvm_pic *vpic;
3191
3192 mutex_lock(&kvm->lock);
3193 r = -EEXIST;
3194 if (kvm->arch.vpic)
3195 goto create_irqchip_unlock;
3e515705
AK		 3196 r = -EINVAL;
3197 if (atomic_read(&kvm->online_vcpus))
3198 goto create_irqchip_unlock;
1fe779f8 3199 r = -ENOMEM;
3ddea128
MT		 3200 vpic = kvm_create_pic(kvm);
3201 if (vpic) {
1fe779f8
CO		 3202 r = kvm_ioapic_init(kvm);
3203 if (r) {
175504cd 3204 mutex_lock(&kvm->slots_lock);
72bb2fcd 3205 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
743eeb0b
SL		 3206 &vpic->dev_master);
3207 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3208 &vpic->dev_slave);
3209 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3210 &vpic->dev_eclr);
175504cd 3211 mutex_unlock(&kvm->slots_lock);
3ddea128
MT		 3212 kfree(vpic);
3213 goto create_irqchip_unlock;
1fe779f8
CO		 3214 }
3215 } else
3ddea128
MT		 3216 goto create_irqchip_unlock;
3217 smp_wmb();
3218 kvm->arch.vpic = vpic;
3219 smp_wmb();
399ec807
AK		 3220 r = kvm_setup_default_irq_routing(kvm);
3221 if (r) {
175504cd 3222 mutex_lock(&kvm->slots_lock);
3ddea128 3223 mutex_lock(&kvm->irq_lock);
72bb2fcd
WY		 3224 kvm_ioapic_destroy(kvm);
3225 kvm_destroy_pic(kvm);
3ddea128 3226 mutex_unlock(&kvm->irq_lock);
175504cd 3227 mutex_unlock(&kvm->slots_lock);
399ec807 3228 }
3ddea128
MT		 3229 create_irqchip_unlock:
3230 mutex_unlock(&kvm->lock);
1fe779f8 3231 break;
3ddea128 3232 }
7837699f 3233 case KVM_CREATE_PIT:
c5ff41ce
JK		 3234 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3235 goto create_pit;
3236 case KVM_CREATE_PIT2:
3237 r = -EFAULT;
3238 if (copy_from_user(&u.pit_config, argp,
3239 sizeof(struct kvm_pit_config)))
3240 goto out;
3241 create_pit:
79fac95e 3242 mutex_lock(&kvm->slots_lock);
269e05e4
AK		 3243 r = -EEXIST;
3244 if (kvm->arch.vpit)
3245 goto create_pit_unlock;
7837699f 3246 r = -ENOMEM;
c5ff41ce 3247 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
7837699f
SY		 3248 if (kvm->arch.vpit)
3249 r = 0;
269e05e4 3250 create_pit_unlock:
79fac95e 3251 mutex_unlock(&kvm->slots_lock);
7837699f 3252 break;
4925663a 3253 case KVM_IRQ_LINE_STATUS:
1fe779f8
CO		 3254 case KVM_IRQ_LINE: {
3255 struct kvm_irq_level irq_event;
3256
3257 r = -EFAULT;
3258 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3259 goto out;
160d2f6c 3260 r = -ENXIO;
1fe779f8 3261 if (irqchip_in_kernel(kvm)) {
4925663a 3262 __s32 status;
4925663a
GN		 3263 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3264 irq_event.irq, irq_event.level);
4925663a 3265 if (ioctl == KVM_IRQ_LINE_STATUS) {
160d2f6c 3266 r = -EFAULT;
4925663a
GN		 3267 irq_event.status = status;
3268 if (copy_to_user(argp, &irq_event,
3269 sizeof irq_event))
3270 goto out;
3271 }
1fe779f8
CO		 3272 r = 0;
3273 }
3274 break;
3275 }
3276 case KVM_GET_IRQCHIP: {
3277 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 3278 struct kvm_irqchip *chip;
1fe779f8 3279
ff5c2c03
SL		 3280 chip = memdup_user(argp, sizeof(*chip));
3281 if (IS_ERR(chip)) {
3282 r = PTR_ERR(chip);
1fe779f8 3283 goto out;
ff5c2c03
SL		 3284 }
3285
1fe779f8
CO		 3286 r = -ENXIO;
3287 if (!irqchip_in_kernel(kvm))
f0d66275
DH		 3288 goto get_irqchip_out;
3289 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1fe779f8 3290 if (r)
f0d66275 3291 goto get_irqchip_out;
1fe779f8 3292 r = -EFAULT;
f0d66275
DH		 3293 if (copy_to_user(argp, chip, sizeof *chip))
3294 goto get_irqchip_out;
1fe779f8 3295 r = 0;
f0d66275
DH		 3296 get_irqchip_out:
3297 kfree(chip);
3298 if (r)
3299 goto out;
1fe779f8
CO		 3300 break;
3301 }
3302 case KVM_SET_IRQCHIP: {
3303 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 3304 struct kvm_irqchip *chip;
1fe779f8 3305
ff5c2c03
SL		 3306 chip = memdup_user(argp, sizeof(*chip));
3307 if (IS_ERR(chip)) {
3308 r = PTR_ERR(chip);
1fe779f8 3309 goto out;
ff5c2c03
SL		 3310 }
3311
1fe779f8
CO		 3312 r = -ENXIO;
3313 if (!irqchip_in_kernel(kvm))
f0d66275
DH		 3314 goto set_irqchip_out;
3315 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1fe779f8 3316 if (r)
f0d66275 3317 goto set_irqchip_out;
1fe779f8 3318 r = 0;
f0d66275
DH		 3319 set_irqchip_out:
3320 kfree(chip);
3321 if (r)
3322 goto out;
1fe779f8
CO		 3323 break;
3324 }
e0f63cb9 3325 case KVM_GET_PIT: {
e0f63cb9 3326 r = -EFAULT;
f0d66275 3327 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 3328 goto out;
3329 r = -ENXIO;
3330 if (!kvm->arch.vpit)
3331 goto out;
f0d66275 3332 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
e0f63cb9
SY		 3333 if (r)
3334 goto out;
3335 r = -EFAULT;
f0d66275 3336 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 3337 goto out;
3338 r = 0;
3339 break;
3340 }
3341 case KVM_SET_PIT: {
e0f63cb9 3342 r = -EFAULT;
f0d66275 3343 if (copy_from_user(&u.ps, argp, sizeof u.ps))
e0f63cb9
SY		 3344 goto out;
3345 r = -ENXIO;
3346 if (!kvm->arch.vpit)
3347 goto out;
f0d66275 3348 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
e0f63cb9
SY		 3349 if (r)
3350 goto out;
3351 r = 0;
3352 break;
3353 }
e9f42757
BK		 3354 case KVM_GET_PIT2: {
3355 r = -ENXIO;
3356 if (!kvm->arch.vpit)
3357 goto out;
3358 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3359 if (r)
3360 goto out;
3361 r = -EFAULT;
3362 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3363 goto out;
3364 r = 0;
3365 break;
3366 }
3367 case KVM_SET_PIT2: {
3368 r = -EFAULT;
3369 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3370 goto out;
3371 r = -ENXIO;
3372 if (!kvm->arch.vpit)
3373 goto out;
3374 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3375 if (r)
3376 goto out;
3377 r = 0;
3378 break;
3379 }
52d939a0
MT		 3380 case KVM_REINJECT_CONTROL: {
3381 struct kvm_reinject_control control;
3382 r = -EFAULT;
3383 if (copy_from_user(&control, argp, sizeof(control)))
3384 goto out;
3385 r = kvm_vm_ioctl_reinject(kvm, &control);
3386 if (r)
3387 goto out;
3388 r = 0;
3389 break;
3390 }
ffde22ac
ES		 3391 case KVM_XEN_HVM_CONFIG: {
3392 r = -EFAULT;
3393 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3394 sizeof(struct kvm_xen_hvm_config)))
3395 goto out;
3396 r = -EINVAL;
3397 if (kvm->arch.xen_hvm_config.flags)
3398 goto out;
3399 r = 0;
3400 break;
3401 }
afbcf7ab 3402 case KVM_SET_CLOCK: {
afbcf7ab
GC		 3403 struct kvm_clock_data user_ns;
3404 u64 now_ns;
3405 s64 delta;
3406
3407 r = -EFAULT;
3408 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3409 goto out;
3410
3411 r = -EINVAL;
3412 if (user_ns.flags)
3413 goto out;
3414
3415 r = 0;
395c6b0a 3416 local_irq_disable();
759379dd 3417 now_ns = get_kernel_ns();
afbcf7ab 3418 delta = user_ns.clock - now_ns;
395c6b0a 3419 local_irq_enable();
afbcf7ab
GC		 3420 kvm->arch.kvmclock_offset = delta;
3421 break;
3422 }
3423 case KVM_GET_CLOCK: {
afbcf7ab
GC		 3424 struct kvm_clock_data user_ns;
3425 u64 now_ns;
3426
395c6b0a 3427 local_irq_disable();
759379dd 3428 now_ns = get_kernel_ns();
afbcf7ab 3429 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
395c6b0a 3430 local_irq_enable();
afbcf7ab 3431 user_ns.flags = 0;
97e69aa6 3432 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
afbcf7ab
GC		 3433
3434 r = -EFAULT;
3435 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3436 goto out;
3437 r = 0;
3438 break;
3439 }
3440
1fe779f8
CO		 3441 default:
3442 ;
3443 }
3444out:
3445 return r;
3446}
3447
a16b043c 3448static void kvm_init_msr_list(void)
043405e1
CO		 3449{
3450 u32 dummy[2];
3451 unsigned i, j;
3452
e3267cbb
GC		 3453 /* skip the first msrs in the list. KVM-specific */
3454 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
043405e1
CO		 3455 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3456 continue;
3457 if (j < i)
3458 msrs_to_save[j] = msrs_to_save[i];
3459 j++;
3460 }
3461 num_msrs_to_save = j;
3462}
3463
bda9020e
MT		 3464static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3465 const void *v)
bbd9b64e 3466{
70252a10
AK		 3467 int handled = 0;
3468 int n;
3469
3470 do {
3471 n = min(len, 8);
3472 if (!(vcpu->arch.apic &&
3473 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3474 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3475 break;
3476 handled += n;
3477 addr += n;
3478 len -= n;
3479 v += n;
3480 } while (len);
bbd9b64e 3481
70252a10 3482 return handled;
bbd9b64e
CO		 3483}
3484
bda9020e 3485static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
bbd9b64e 3486{
70252a10
AK		 3487 int handled = 0;
3488 int n;
3489
3490 do {
3491 n = min(len, 8);
3492 if (!(vcpu->arch.apic &&
3493 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3494 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3495 break;
3496 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3497 handled += n;
3498 addr += n;
3499 len -= n;
3500 v += n;
3501 } while (len);
bbd9b64e 3502
70252a10 3503 return handled;
bbd9b64e
CO		 3504}
3505
2dafc6c2
GN		 3506static void kvm_set_segment(struct kvm_vcpu *vcpu,
3507 struct kvm_segment *var, int seg)
3508{
3509 kvm_x86_ops->set_segment(vcpu, var, seg);
3510}
3511
3512void kvm_get_segment(struct kvm_vcpu *vcpu,
3513 struct kvm_segment *var, int seg)
3514{
3515 kvm_x86_ops->get_segment(vcpu, var, seg);
3516}
3517
e459e322 3518gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
02f59dc9
JR		 3519{
3520 gpa_t t_gpa;
ab9ae313 3521 struct x86_exception exception;
02f59dc9
JR		 3522
3523 BUG_ON(!mmu_is_nested(vcpu));
3524
3525 /* NPT walks are always user-walks */
3526 access |= PFERR_USER_MASK;
ab9ae313 3527 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
02f59dc9
JR		 3528
3529 return t_gpa;
3530}
3531
ab9ae313
AK		 3532gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3533 struct x86_exception *exception)
1871c602
GN		 3534{
3535 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
ab9ae313 3536 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 3537}
3538
ab9ae313
AK		 3539 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3540 struct x86_exception *exception)
1871c602
GN		 3541{
3542 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3543 access |= PFERR_FETCH_MASK;
ab9ae313 3544 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 3545}
3546
ab9ae313
AK		 3547gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3548 struct x86_exception *exception)
1871c602
GN		 3549{
3550 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3551 access |= PFERR_WRITE_MASK;
ab9ae313 3552 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 3553}
3554
3555/* uses this to access any guest's mapped memory without checking CPL */
ab9ae313
AK		 3556gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3557 struct x86_exception *exception)
1871c602 3558{
ab9ae313 3559 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
1871c602
GN		 3560}
3561
3562static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3563 struct kvm_vcpu *vcpu, u32 access,
bcc55cba 3564 struct x86_exception *exception)
bbd9b64e
CO		 3565{
3566 void *data = val;
10589a46 3567 int r = X86EMUL_CONTINUE;
bbd9b64e
CO		 3568
3569 while (bytes) {
14dfe855 3570 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
ab9ae313 3571 exception);
bbd9b64e 3572 unsigned offset = addr & (PAGE_SIZE-1);
77c2002e 3573 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
bbd9b64e
CO		 3574 int ret;
3575
bcc55cba 3576 if (gpa == UNMAPPED_GVA)
ab9ae313 3577 return X86EMUL_PROPAGATE_FAULT;
77c2002e 3578 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
10589a46 3579 if (ret < 0) {
c3cd7ffa 3580 r = X86EMUL_IO_NEEDED;
10589a46
MT		 3581 goto out;
3582 }
bbd9b64e 3583
77c2002e
IE		 3584 bytes -= toread;
3585 data += toread;
3586 addr += toread;
bbd9b64e 3587 }
10589a46 3588out:
10589a46 3589 return r;
bbd9b64e 3590}
77c2002e 3591
1871c602 3592/* used for instruction fetching */
0f65dd70
AK		 3593static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3594 gva_t addr, void *val, unsigned int bytes,
bcc55cba 3595 struct x86_exception *exception)
1871c602 3596{
0f65dd70 3597 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 3598 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
0f65dd70 3599
1871c602 3600 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
bcc55cba
AK		 3601 access | PFERR_FETCH_MASK,
3602 exception);
1871c602
GN		 3603}
3604
064aea77 3605int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
0f65dd70 3606 gva_t addr, void *val, unsigned int bytes,
bcc55cba 3607 struct x86_exception *exception)
1871c602 3608{
0f65dd70 3609 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 3610 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
0f65dd70 3611
1871c602 3612 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
bcc55cba 3613 exception);
1871c602 3614}
064aea77 3615EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
1871c602 3616
0f65dd70
AK		 3617static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3618 gva_t addr, void *val, unsigned int bytes,
bcc55cba 3619 struct x86_exception *exception)
1871c602 3620{
0f65dd70 3621 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
bcc55cba 3622 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
1871c602
GN		 3623}
3624
6a4d7550 3625int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
0f65dd70 3626 gva_t addr, void *val,
2dafc6c2 3627 unsigned int bytes,
bcc55cba 3628 struct x86_exception *exception)
77c2002e 3629{
0f65dd70 3630 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
77c2002e
IE		 3631 void *data = val;
3632 int r = X86EMUL_CONTINUE;
3633
3634 while (bytes) {
14dfe855
JR		 3635 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3636 PFERR_WRITE_MASK,
ab9ae313 3637 exception);
77c2002e
IE		 3638 unsigned offset = addr & (PAGE_SIZE-1);
3639 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3640 int ret;
3641
bcc55cba 3642 if (gpa == UNMAPPED_GVA)
ab9ae313 3643 return X86EMUL_PROPAGATE_FAULT;
77c2002e
IE		 3644 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3645 if (ret < 0) {
c3cd7ffa 3646 r = X86EMUL_IO_NEEDED;
77c2002e
IE		 3647 goto out;
3648 }
3649
3650 bytes -= towrite;
3651 data += towrite;
3652 addr += towrite;
3653 }
3654out:
3655 return r;
3656}
6a4d7550 3657EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
77c2002e 3658
af7cc7d1
XG		 3659static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
3660 gpa_t *gpa, struct x86_exception *exception,
3661 bool write)
3662{
3663 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3664
bebb106a
XG		 3665 if (vcpu_match_mmio_gva(vcpu, gva) &&
3666 check_write_user_access(vcpu, write, access,
3667 vcpu->arch.access)) {
3668 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
3669 (gva & (PAGE_SIZE - 1));
4f022648 3670 trace_vcpu_match_mmio(gva, *gpa, write, false);
bebb106a
XG		 3671 return 1;
3672 }
3673
af7cc7d1
XG		 3674 if (write)
3675 access |= PFERR_WRITE_MASK;
3676
3677 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3678
3679 if (*gpa == UNMAPPED_GVA)
3680 return -1;
3681
3682 /* For APIC access vmexit */
3683 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3684 return 1;
3685
4f022648
XG		 3686 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
3687 trace_vcpu_match_mmio(gva, *gpa, write, true);
bebb106a 3688 return 1;
4f022648 3689 }
bebb106a 3690
af7cc7d1
XG		 3691 return 0;
3692}
3693
3200f405 3694int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
bcc55cba 3695 const void *val, int bytes)
bbd9b64e
CO		 3696{
3697 int ret;
3698
3699 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
9f811285 3700 if (ret < 0)
bbd9b64e 3701 return 0;
f57f2ef5 3702 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
bbd9b64e
CO		 3703 return 1;
3704}
3705
77d197b2
XG		 3706struct read_write_emulator_ops {
3707 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
3708 int bytes);
3709 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
3710 void *val, int bytes);
3711 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
3712 int bytes, void *val);
3713 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
3714 void *val, int bytes);
3715 bool write;
3716};
3717
3718static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
3719{
3720 if (vcpu->mmio_read_completed) {
77d197b2 3721 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
f78146b0 3722 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
77d197b2
XG		 3723 vcpu->mmio_read_completed = 0;
3724 return 1;
3725 }
3726
3727 return 0;
3728}
3729
3730static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
3731 void *val, int bytes)
3732{
3733 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
3734}
3735
3736static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
3737 void *val, int bytes)
3738{
3739 return emulator_write_phys(vcpu, gpa, val, bytes);
3740}
3741
3742static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
3743{
3744 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3745 return vcpu_mmio_write(vcpu, gpa, bytes, val);
3746}
3747
3748static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
3749 void *val, int bytes)
3750{
3751 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3752 return X86EMUL_IO_NEEDED;
3753}
3754
3755static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
3756 void *val, int bytes)
3757{
f78146b0
AK		 3758 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
3759
3760 memcpy(vcpu->run->mmio.data, frag->data, frag->len);
77d197b2
XG		 3761 return X86EMUL_CONTINUE;
3762}
3763
3764static struct read_write_emulator_ops read_emultor = {
3765 .read_write_prepare = read_prepare,
3766 .read_write_emulate = read_emulate,
3767 .read_write_mmio = vcpu_mmio_read,
3768 .read_write_exit_mmio = read_exit_mmio,
3769};
3770
3771static struct read_write_emulator_ops write_emultor = {
3772 .read_write_emulate = write_emulate,
3773 .read_write_mmio = write_mmio,
3774 .read_write_exit_mmio = write_exit_mmio,
3775 .write = true,
3776};
3777
22388a3c
XG		 3778static int emulator_read_write_onepage(unsigned long addr, void *val,
3779 unsigned int bytes,
3780 struct x86_exception *exception,
3781 struct kvm_vcpu *vcpu,
3782 struct read_write_emulator_ops *ops)
bbd9b64e 3783{
af7cc7d1
XG		 3784 gpa_t gpa;
3785 int handled, ret;
22388a3c 3786 bool write = ops->write;
f78146b0 3787 struct kvm_mmio_fragment *frag;
10589a46 3788
22388a3c 3789 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
bbd9b64e 3790
af7cc7d1 3791 if (ret < 0)
bbd9b64e 3792 return X86EMUL_PROPAGATE_FAULT;
bbd9b64e
CO		 3793
3794 /* For APIC access vmexit */
af7cc7d1 3795 if (ret)
bbd9b64e
CO		 3796 goto mmio;
3797
22388a3c 3798 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
bbd9b64e
CO		 3799 return X86EMUL_CONTINUE;
3800
3801mmio:
3802 /*
3803 * Is this MMIO handled locally?
3804 */
22388a3c 3805 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
70252a10 3806 if (handled == bytes)
bbd9b64e 3807 return X86EMUL_CONTINUE;
bbd9b64e 3808
70252a10
AK		 3809 gpa += handled;
3810 bytes -= handled;
3811 val += handled;
3812
f78146b0
AK		 3813 while (bytes) {
3814 unsigned now = min(bytes, 8U);
bbd9b64e 3815
f78146b0
AK		 3816 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
3817 frag->gpa = gpa;
3818 frag->data = val;
3819 frag->len = now;
3820
3821 gpa += now;
3822 val += now;
3823 bytes -= now;
3824 }
3825 return X86EMUL_CONTINUE;
bbd9b64e
CO		 3826}
3827
22388a3c
XG		 3828int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
3829 void *val, unsigned int bytes,
3830 struct x86_exception *exception,
3831 struct read_write_emulator_ops *ops)
bbd9b64e 3832{
0f65dd70 3833 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
f78146b0
AK		 3834 gpa_t gpa;
3835 int rc;
3836
3837 if (ops->read_write_prepare &&
3838 ops->read_write_prepare(vcpu, val, bytes))
3839 return X86EMUL_CONTINUE;
3840
3841 vcpu->mmio_nr_fragments = 0;
0f65dd70 3842
bbd9b64e
CO		 3843 /* Crossing a page boundary? */
3844 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
f78146b0 3845 int now;
bbd9b64e
CO		 3846
3847 now = -addr & ~PAGE_MASK;
22388a3c
XG		 3848 rc = emulator_read_write_onepage(addr, val, now, exception,
3849 vcpu, ops);
3850
bbd9b64e
CO		 3851 if (rc != X86EMUL_CONTINUE)
3852 return rc;
3853 addr += now;
3854 val += now;
3855 bytes -= now;
3856 }
22388a3c 3857
f78146b0
AK		 3858 rc = emulator_read_write_onepage(addr, val, bytes, exception,
3859 vcpu, ops);
3860 if (rc != X86EMUL_CONTINUE)
3861 return rc;
3862
3863 if (!vcpu->mmio_nr_fragments)
3864 return rc;
3865
3866 gpa = vcpu->mmio_fragments[0].gpa;
3867
3868 vcpu->mmio_needed = 1;
3869 vcpu->mmio_cur_fragment = 0;
3870
3871 vcpu->run->mmio.len = vcpu->mmio_fragments[0].len;
3872 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
3873 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3874 vcpu->run->mmio.phys_addr = gpa;
3875
3876 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
22388a3c
XG		 3877}
3878
3879static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
3880 unsigned long addr,
3881 void *val,
3882 unsigned int bytes,
3883 struct x86_exception *exception)
3884{
3885 return emulator_read_write(ctxt, addr, val, bytes,
3886 exception, &read_emultor);
3887}
3888
3889int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
3890 unsigned long addr,
3891 const void *val,
3892 unsigned int bytes,
3893 struct x86_exception *exception)
3894{
3895 return emulator_read_write(ctxt, addr, (void *)val, bytes,
3896 exception, &write_emultor);
bbd9b64e 3897}
bbd9b64e 3898
daea3e73
AK		 3899#define CMPXCHG_TYPE(t, ptr, old, new) \
3900 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3901
3902#ifdef CONFIG_X86_64
3903# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3904#else
3905# define CMPXCHG64(ptr, old, new) \
9749a6c0 3906 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
daea3e73
AK		 3907#endif
3908
0f65dd70
AK		 3909static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
3910 unsigned long addr,
bbd9b64e
CO		 3911 const void *old,
3912 const void *new,
3913 unsigned int bytes,
0f65dd70 3914 struct x86_exception *exception)
bbd9b64e 3915{
0f65dd70 3916 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
daea3e73
AK		 3917 gpa_t gpa;
3918 struct page *page;
3919 char *kaddr;
3920 bool exchanged;
2bacc55c 3921
daea3e73
AK		 3922 /* guests cmpxchg8b have to be emulated atomically */
3923 if (bytes > 8 || (bytes & (bytes - 1)))
3924 goto emul_write;
10589a46 3925
daea3e73 3926 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
2bacc55c 3927
daea3e73
AK		 3928 if (gpa == UNMAPPED_GVA ||
3929 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3930 goto emul_write;
2bacc55c 3931
daea3e73
AK		 3932 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3933 goto emul_write;
72dc67a6 3934
daea3e73 3935 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
c19b8bd6
WY		 3936 if (is_error_page(page)) {
3937 kvm_release_page_clean(page);
3938 goto emul_write;
3939 }
72dc67a6 3940
8fd75e12 3941 kaddr = kmap_atomic(page);
daea3e73
AK		 3942 kaddr += offset_in_page(gpa);
3943 switch (bytes) {
3944 case 1:
3945 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3946 break;
3947 case 2:
3948 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3949 break;
3950 case 4:
3951 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3952 break;
3953 case 8:
3954 exchanged = CMPXCHG64(kaddr, old, new);
3955 break;
3956 default:
3957 BUG();
2bacc55c 3958 }
8fd75e12 3959 kunmap_atomic(kaddr);
daea3e73
AK		 3960 kvm_release_page_dirty(page);
3961
3962 if (!exchanged)
3963 return X86EMUL_CMPXCHG_FAILED;
3964
f57f2ef5 3965 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
8f6abd06
GN		 3966
3967 return X86EMUL_CONTINUE;
4a5f48f6 3968
3200f405 3969emul_write:
daea3e73 3970 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2bacc55c 3971
0f65dd70 3972 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
bbd9b64e
CO		 3973}
3974
cf8f70bf
GN		 3975static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3976{
3977 /* TODO: String I/O for in kernel device */
3978 int r;
3979
3980 if (vcpu->arch.pio.in)
3981 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3982 vcpu->arch.pio.size, pd);
3983 else
3984 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3985 vcpu->arch.pio.port, vcpu->arch.pio.size,
3986 pd);
3987 return r;
3988}
3989
6f6fbe98
XG		 3990static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
3991 unsigned short port, void *val,
3992 unsigned int count, bool in)
cf8f70bf 3993{
6f6fbe98 3994 trace_kvm_pio(!in, port, size, count);
cf8f70bf
GN		 3995
3996 vcpu->arch.pio.port = port;
6f6fbe98 3997 vcpu->arch.pio.in = in;
7972995b 3998 vcpu->arch.pio.count = count;
cf8f70bf
GN		 3999 vcpu->arch.pio.size = size;
4000
4001 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
7972995b 4002 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4003 return 1;
4004 }
4005
4006 vcpu->run->exit_reason = KVM_EXIT_IO;
6f6fbe98 4007 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
cf8f70bf
GN		 4008 vcpu->run->io.size = size;
4009 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4010 vcpu->run->io.count = count;
4011 vcpu->run->io.port = port;
4012
4013 return 0;
4014}
4015
6f6fbe98
XG		 4016static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4017 int size, unsigned short port, void *val,
4018 unsigned int count)
cf8f70bf 4019{
ca1d4a9e 4020 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6f6fbe98 4021 int ret;
ca1d4a9e 4022
6f6fbe98
XG		 4023 if (vcpu->arch.pio.count)
4024 goto data_avail;
cf8f70bf 4025
6f6fbe98
XG		 4026 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4027 if (ret) {
4028data_avail:
4029 memcpy(val, vcpu->arch.pio_data, size * count);
7972995b 4030 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4031 return 1;
4032 }
4033
cf8f70bf
GN		 4034 return 0;
4035}
4036
6f6fbe98
XG		 4037static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4038 int size, unsigned short port,
4039 const void *val, unsigned int count)
4040{
4041 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4042
4043 memcpy(vcpu->arch.pio_data, val, size * count);
4044 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4045}
4046
bbd9b64e
CO		 4047static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4048{
4049 return kvm_x86_ops->get_segment_base(vcpu, seg);
4050}
4051
3cb16fe7 4052static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
bbd9b64e 4053{
3cb16fe7 4054 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
bbd9b64e
CO		 4055}
4056
f5f48ee1
SY		 4057int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4058{
4059 if (!need_emulate_wbinvd(vcpu))
4060 return X86EMUL_CONTINUE;
4061
4062 if (kvm_x86_ops->has_wbinvd_exit()) {
2eec7343
JK		 4063 int cpu = get_cpu();
4064
4065 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
f5f48ee1
SY		 4066 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4067 wbinvd_ipi, NULL, 1);
2eec7343 4068 put_cpu();
f5f48ee1 4069 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
2eec7343
JK		 4070 } else
4071 wbinvd();
f5f48ee1
SY		 4072 return X86EMUL_CONTINUE;
4073}
4074EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4075
bcaf5cc5
AK		 4076static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4077{
4078 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4079}
4080
717746e3 4081int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
bbd9b64e 4082{
717746e3 4083 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
bbd9b64e
CO		 4084}
4085
717746e3 4086int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
bbd9b64e 4087{
338dbc97 4088
717746e3 4089 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
bbd9b64e
CO		 4090}
4091
52a46617 4092static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5fdbf976 4093{
52a46617 4094 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5fdbf976
MT		 4095}
4096
717746e3 4097static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
bbd9b64e 4098{
717746e3 4099 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
52a46617
GN		 4100 unsigned long value;
4101
4102 switch (cr) {
4103 case 0:
4104 value = kvm_read_cr0(vcpu);
4105 break;
4106 case 2:
4107 value = vcpu->arch.cr2;
4108 break;
4109 case 3:
9f8fe504 4110 value = kvm_read_cr3(vcpu);
52a46617
GN		 4111 break;
4112 case 4:
4113 value = kvm_read_cr4(vcpu);
4114 break;
4115 case 8:
4116 value = kvm_get_cr8(vcpu);
4117 break;
4118 default:
4119 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4120 return 0;
4121 }
4122
4123 return value;
4124}
4125
717746e3 4126static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
52a46617 4127{
717746e3 4128 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
0f12244f
GN		 4129 int res = 0;
4130
52a46617
GN		 4131 switch (cr) {
4132 case 0:
49a9b07e 4133 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
52a46617
GN		 4134 break;
4135 case 2:
4136 vcpu->arch.cr2 = val;
4137 break;
4138 case 3:
2390218b 4139 res = kvm_set_cr3(vcpu, val);
52a46617
GN		 4140 break;
4141 case 4:
a83b29c6 4142 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
52a46617
GN		 4143 break;
4144 case 8:
eea1cff9 4145 res = kvm_set_cr8(vcpu, val);
52a46617
GN		 4146 break;
4147 default:
4148 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
0f12244f 4149 res = -1;
52a46617 4150 }
0f12244f
GN		 4151
4152 return res;
52a46617
GN		 4153}
4154
4cee4798
KW		 4155static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4156{
4157 kvm_set_rflags(emul_to_vcpu(ctxt), val);
4158}
4159
717746e3 4160static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
9c537244 4161{
717746e3 4162 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
9c537244
GN		 4163}
4164
4bff1e86 4165static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
2dafc6c2 4166{
4bff1e86 4167 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
2dafc6c2
GN		 4168}
4169
4bff1e86 4170static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
160ce1f1 4171{
4bff1e86 4172 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
160ce1f1
MG		 4173}
4174
1ac9d0cf
AK		 4175static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4176{
4177 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4178}
4179
4180static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4181{
4182 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4183}
4184
4bff1e86
AK		 4185static unsigned long emulator_get_cached_segment_base(
4186 struct x86_emulate_ctxt *ctxt, int seg)
5951c442 4187{
4bff1e86 4188 return get_segment_base(emul_to_vcpu(ctxt), seg);
5951c442
GN		 4189}
4190
1aa36616
AK		 4191static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4192 struct desc_struct *desc, u32 *base3,
4193 int seg)
2dafc6c2
GN		 4194{
4195 struct kvm_segment var;
4196
4bff1e86 4197 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
1aa36616 4198 *selector = var.selector;
2dafc6c2
GN		 4199
4200 if (var.unusable)
4201 return false;
4202
4203 if (var.g)
4204 var.limit >>= 12;
4205 set_desc_limit(desc, var.limit);
4206 set_desc_base(desc, (unsigned long)var.base);
5601d05b
GN		 4207#ifdef CONFIG_X86_64
4208 if (base3)
4209 *base3 = var.base >> 32;
4210#endif
2dafc6c2
GN		 4211 desc->type = var.type;
4212 desc->s = var.s;
4213 desc->dpl = var.dpl;
4214 desc->p = var.present;
4215 desc->avl = var.avl;
4216 desc->l = var.l;
4217 desc->d = var.db;
4218 desc->g = var.g;
4219
4220 return true;
4221}
4222
1aa36616
AK		 4223static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4224 struct desc_struct *desc, u32 base3,
4225 int seg)
2dafc6c2 4226{
4bff1e86 4227 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
2dafc6c2
GN		 4228 struct kvm_segment var;
4229
1aa36616 4230 var.selector = selector;
2dafc6c2 4231 var.base = get_desc_base(desc);
5601d05b
GN		 4232#ifdef CONFIG_X86_64
4233 var.base |= ((u64)base3) << 32;
4234#endif
2dafc6c2
GN		 4235 var.limit = get_desc_limit(desc);
4236 if (desc->g)
4237 var.limit = (var.limit << 12) | 0xfff;
4238 var.type = desc->type;
4239 var.present = desc->p;
4240 var.dpl = desc->dpl;
4241 var.db = desc->d;
4242 var.s = desc->s;
4243 var.l = desc->l;
4244 var.g = desc->g;
4245 var.avl = desc->avl;
4246 var.present = desc->p;
4247 var.unusable = !var.present;
4248 var.padding = 0;
4249
4250 kvm_set_segment(vcpu, &var, seg);
4251 return;
4252}
4253
717746e3
AK		 4254static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4255 u32 msr_index, u64 *pdata)
4256{
4257 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4258}
4259
4260static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4261 u32 msr_index, u64 data)
4262{
4263 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4264}
4265
222d21aa
AK		 4266static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4267 u32 pmc, u64 *pdata)
4268{
4269 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4270}
4271
6c3287f7
AK		 4272static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4273{
4274 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4275}
4276
5037f6f3
AK		 4277static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4278{
4279 preempt_disable();
5197b808 4280 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5037f6f3
AK		 4281 /*
4282 * CR0.TS may reference the host fpu state, not the guest fpu state,
4283 * so it may be clear at this point.
4284 */
4285 clts();
4286}
4287
4288static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4289{
4290 preempt_enable();
4291}
4292
2953538e 4293static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
8a76d7f2 4294 struct x86_instruction_info *info,
c4f035c6
AK		 4295 enum x86_intercept_stage stage)
4296{
2953538e 4297 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
c4f035c6
AK		 4298}
4299
bdb42f5a
SB		 4300static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4301 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4302{
4303 struct kvm_cpuid_entry2 *cpuid = NULL;
4304
4305 if (eax && ecx)
4306 cpuid = kvm_find_cpuid_entry(emul_to_vcpu(ctxt),
4307 *eax, *ecx);
4308
4309 if (cpuid) {
4310 *eax = cpuid->eax;
4311 *ecx = cpuid->ecx;
4312 if (ebx)
4313 *ebx = cpuid->ebx;
4314 if (edx)
4315 *edx = cpuid->edx;
4316 return true;
4317 }
4318
4319 return false;
4320}
4321
14af3f3c 4322static struct x86_emulate_ops emulate_ops = {
1871c602 4323 .read_std = kvm_read_guest_virt_system,
2dafc6c2 4324 .write_std = kvm_write_guest_virt_system,
1871c602 4325 .fetch = kvm_fetch_guest_virt,
bbd9b64e
CO		 4326 .read_emulated = emulator_read_emulated,
4327 .write_emulated = emulator_write_emulated,
4328 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3cb16fe7 4329 .invlpg = emulator_invlpg,
cf8f70bf
GN		 4330 .pio_in_emulated = emulator_pio_in_emulated,
4331 .pio_out_emulated = emulator_pio_out_emulated,
1aa36616
AK		 4332 .get_segment = emulator_get_segment,
4333 .set_segment = emulator_set_segment,
5951c442 4334 .get_cached_segment_base = emulator_get_cached_segment_base,
2dafc6c2 4335 .get_gdt = emulator_get_gdt,
160ce1f1 4336 .get_idt = emulator_get_idt,
1ac9d0cf
AK		 4337 .set_gdt = emulator_set_gdt,
4338 .set_idt = emulator_set_idt,
52a46617
GN		 4339 .get_cr = emulator_get_cr,
4340 .set_cr = emulator_set_cr,
4cee4798 4341 .set_rflags = emulator_set_rflags,
9c537244 4342 .cpl = emulator_get_cpl,
35aa5375
GN		 4343 .get_dr = emulator_get_dr,
4344 .set_dr = emulator_set_dr,
717746e3
AK		 4345 .set_msr = emulator_set_msr,
4346 .get_msr = emulator_get_msr,
222d21aa 4347 .read_pmc = emulator_read_pmc,
6c3287f7 4348 .halt = emulator_halt,
bcaf5cc5 4349 .wbinvd = emulator_wbinvd,
d6aa1000 4350 .fix_hypercall = emulator_fix_hypercall,
5037f6f3
AK		 4351 .get_fpu = emulator_get_fpu,
4352 .put_fpu = emulator_put_fpu,
c4f035c6 4353 .intercept = emulator_intercept,
bdb42f5a 4354 .get_cpuid = emulator_get_cpuid,
bbd9b64e
CO		 4355};
4356
5fdbf976
MT		 4357static void cache_all_regs(struct kvm_vcpu *vcpu)
4358{
4359 kvm_register_read(vcpu, VCPU_REGS_RAX);
4360 kvm_register_read(vcpu, VCPU_REGS_RSP);
4361 kvm_register_read(vcpu, VCPU_REGS_RIP);
4362 vcpu->arch.regs_dirty = ~0;
4363}
4364
95cb2295
GN		 4365static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4366{
4367 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4368 /*
4369 * an sti; sti; sequence only disable interrupts for the first
4370 * instruction. So, if the last instruction, be it emulated or
4371 * not, left the system with the INT_STI flag enabled, it
4372 * means that the last instruction is an sti. We should not
4373 * leave the flag on in this case. The same goes for mov ss
4374 */
4375 if (!(int_shadow & mask))
4376 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4377}
4378
54b8486f
GN		 4379static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4380{
4381 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
da9cb575 4382 if (ctxt->exception.vector == PF_VECTOR)
6389ee94 4383 kvm_propagate_fault(vcpu, &ctxt->exception);
da9cb575
AK		 4384 else if (ctxt->exception.error_code_valid)
4385 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4386 ctxt->exception.error_code);
54b8486f 4387 else
da9cb575 4388 kvm_queue_exception(vcpu, ctxt->exception.vector);
54b8486f
GN		 4389}
4390
9dac77fa 4391static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
b5c9ff73
TY		 4392 const unsigned long *regs)
4393{
9dac77fa
AK		 4394 memset(&ctxt->twobyte, 0,
4395 (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4396 memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
b5c9ff73 4397
9dac77fa
AK		 4398 ctxt->fetch.start = 0;
4399 ctxt->fetch.end = 0;
4400 ctxt->io_read.pos = 0;
4401 ctxt->io_read.end = 0;
4402 ctxt->mem_read.pos = 0;
4403 ctxt->mem_read.end = 0;
b5c9ff73
TY		 4404}
4405
8ec4722d
MG		 4406static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4407{
adf52235 4408 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d
MG		 4409 int cs_db, cs_l;
4410
2aab2c5b
GN		 4411 /*
4412 * TODO: fix emulate.c to use guest_read/write_register
4413 * instead of direct ->regs accesses, can save hundred cycles
4414 * on Intel for instructions that don't read/change RSP, for
4415 * for example.
4416 */
8ec4722d
MG		 4417 cache_all_regs(vcpu);
4418
4419 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4420
adf52235
TY		 4421 ctxt->eflags = kvm_get_rflags(vcpu);
4422 ctxt->eip = kvm_rip_read(vcpu);
4423 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4424 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4425 cs_l ? X86EMUL_MODE_PROT64 :
4426 cs_db ? X86EMUL_MODE_PROT32 :
4427 X86EMUL_MODE_PROT16;
4428 ctxt->guest_mode = is_guest_mode(vcpu);
4429
9dac77fa 4430 init_decode_cache(ctxt, vcpu->arch.regs);
7ae441ea 4431 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8ec4722d
MG		 4432}
4433
71f9833b 4434int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
63995653 4435{
9d74191a 4436 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
63995653
MG		 4437 int ret;
4438
4439 init_emulate_ctxt(vcpu);
4440
9dac77fa
AK		 4441 ctxt->op_bytes = 2;
4442 ctxt->ad_bytes = 2;
4443 ctxt->_eip = ctxt->eip + inc_eip;
9d74191a 4444 ret = emulate_int_real(ctxt, irq);
63995653
MG		 4445
4446 if (ret != X86EMUL_CONTINUE)
4447 return EMULATE_FAIL;
4448
9dac77fa
AK		 4449 ctxt->eip = ctxt->_eip;
4450 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
9d74191a
TY		 4451 kvm_rip_write(vcpu, ctxt->eip);
4452 kvm_set_rflags(vcpu, ctxt->eflags);
63995653
MG		 4453
4454 if (irq == NMI_VECTOR)
7460fb4a 4455 vcpu->arch.nmi_pending = 0;
63995653
MG		 4456 else
4457 vcpu->arch.interrupt.pending = false;
4458
4459 return EMULATE_DONE;
4460}
4461EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4462
6d77dbfc
GN		 4463static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4464{
fc3a9157
JR		 4465 int r = EMULATE_DONE;
4466
6d77dbfc
GN		 4467 ++vcpu->stat.insn_emulation_fail;
4468 trace_kvm_emulate_insn_failed(vcpu);
fc3a9157
JR		 4469 if (!is_guest_mode(vcpu)) {
4470 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4471 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4472 vcpu->run->internal.ndata = 0;
4473 r = EMULATE_FAIL;
4474 }
6d77dbfc 4475 kvm_queue_exception(vcpu, UD_VECTOR);
fc3a9157
JR		 4476
4477 return r;
6d77dbfc
GN		 4478}
4479
a6f177ef
GN		 4480static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4481{
4482 gpa_t gpa;
4483
68be0803
GN		 4484 if (tdp_enabled)
4485 return false;
4486
a6f177ef
GN		 4487 /*
4488 * if emulation was due to access to shadowed page table
4489 * and it failed try to unshadow page and re-entetr the
4490 * guest to let CPU execute the instruction.
4491 */
4492 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4493 return true;
4494
4495 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4496
4497 if (gpa == UNMAPPED_GVA)
4498 return true; /* let cpu generate fault */
4499
4500 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4501 return true;
4502
4503 return false;
4504}
4505
1cb3f3ae
XG		 4506static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4507 unsigned long cr2, int emulation_type)
4508{
4509 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4510 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4511
4512 last_retry_eip = vcpu->arch.last_retry_eip;
4513 last_retry_addr = vcpu->arch.last_retry_addr;
4514
4515 /*
4516 * If the emulation is caused by #PF and it is non-page_table
4517 * writing instruction, it means the VM-EXIT is caused by shadow
4518 * page protected, we can zap the shadow page and retry this
4519 * instruction directly.
4520 *
4521 * Note: if the guest uses a non-page-table modifying instruction
4522 * on the PDE that points to the instruction, then we will unmap
4523 * the instruction and go to an infinite loop. So, we cache the
4524 * last retried eip and the last fault address, if we meet the eip
4525 * and the address again, we can break out of the potential infinite
4526 * loop.
4527 */
4528 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4529
4530 if (!(emulation_type & EMULTYPE_RETRY))
4531 return false;
4532
4533 if (x86_page_table_writing_insn(ctxt))
4534 return false;
4535
4536 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4537 return false;
4538
4539 vcpu->arch.last_retry_eip = ctxt->eip;
4540 vcpu->arch.last_retry_addr = cr2;
4541
4542 if (!vcpu->arch.mmu.direct_map)
4543 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4544
4545 kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4546
4547 return true;
4548}
4549
51d8b661
AP		 4550int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4551 unsigned long cr2,
dc25e89e
AP		 4552 int emulation_type,
4553 void *insn,
4554 int insn_len)
bbd9b64e 4555{
95cb2295 4556 int r;
9d74191a 4557 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7ae441ea 4558 bool writeback = true;
bbd9b64e 4559
26eef70c 4560 kvm_clear_exception_queue(vcpu);
8d7d8102 4561
571008da 4562 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8ec4722d 4563 init_emulate_ctxt(vcpu);
9d74191a
TY		 4564 ctxt->interruptibility = 0;
4565 ctxt->have_exception = false;
4566 ctxt->perm_ok = false;
bbd9b64e 4567
9d74191a 4568 ctxt->only_vendor_specific_insn
4005996e
AK		 4569 = emulation_type & EMULTYPE_TRAP_UD;
4570
9d74191a 4571 r = x86_decode_insn(ctxt, insn, insn_len);
bbd9b64e 4572
e46479f8 4573 trace_kvm_emulate_insn_start(vcpu);
f2b5756b 4574 ++vcpu->stat.insn_emulation;
1d2887e2 4575 if (r != EMULATION_OK) {
4005996e
AK		 4576 if (emulation_type & EMULTYPE_TRAP_UD)
4577 return EMULATE_FAIL;
a6f177ef 4578 if (reexecute_instruction(vcpu, cr2))
bbd9b64e 4579 return EMULATE_DONE;
6d77dbfc
GN		 4580 if (emulation_type & EMULTYPE_SKIP)
4581 return EMULATE_FAIL;
4582 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 4583 }
4584 }
4585
ba8afb6b 4586 if (emulation_type & EMULTYPE_SKIP) {
9dac77fa 4587 kvm_rip_write(vcpu, ctxt->_eip);
ba8afb6b
GN		 4588 return EMULATE_DONE;
4589 }
4590
1cb3f3ae
XG		 4591 if (retry_instruction(ctxt, cr2, emulation_type))
4592 return EMULATE_DONE;
4593
7ae441ea 4594 /* this is needed for vmware backdoor interface to work since it
4d2179e1 4595 changes registers values during IO operation */
7ae441ea
GN		 4596 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4597 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
9dac77fa 4598 memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
7ae441ea 4599 }
4d2179e1 4600
5cd21917 4601restart:
9d74191a 4602 r = x86_emulate_insn(ctxt);
bbd9b64e 4603
775fde86
JR		 4604 if (r == EMULATION_INTERCEPTED)
4605 return EMULATE_DONE;
4606
d2ddd1c4 4607 if (r == EMULATION_FAILED) {
a6f177ef 4608 if (reexecute_instruction(vcpu, cr2))
c3cd7ffa
GN		 4609 return EMULATE_DONE;
4610
6d77dbfc 4611 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 4612 }
4613
9d74191a 4614 if (ctxt->have_exception) {
54b8486f 4615 inject_emulated_exception(vcpu);
d2ddd1c4
GN		 4616 r = EMULATE_DONE;
4617 } else if (vcpu->arch.pio.count) {
3457e419
GN		 4618 if (!vcpu->arch.pio.in)
4619 vcpu->arch.pio.count = 0;
7ae441ea
GN		 4620 else
4621 writeback = false;
e85d28f8 4622 r = EMULATE_DO_MMIO;
7ae441ea
GN		 4623 } else if (vcpu->mmio_needed) {
4624 if (!vcpu->mmio_is_write)
4625 writeback = false;
e85d28f8 4626 r = EMULATE_DO_MMIO;
7ae441ea 4627 } else if (r == EMULATION_RESTART)
5cd21917 4628 goto restart;
d2ddd1c4
GN		 4629 else
4630 r = EMULATE_DONE;
f850e2e6 4631
7ae441ea 4632 if (writeback) {
9d74191a
TY		 4633 toggle_interruptibility(vcpu, ctxt->interruptibility);
4634 kvm_set_rflags(vcpu, ctxt->eflags);
7ae441ea 4635 kvm_make_request(KVM_REQ_EVENT, vcpu);
9dac77fa 4636 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
7ae441ea 4637 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9d74191a 4638 kvm_rip_write(vcpu, ctxt->eip);
7ae441ea
GN		 4639 } else
4640 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
e85d28f8
GN		 4641
4642 return r;
de7d789a 4643}
51d8b661 4644EXPORT_SYMBOL_GPL(x86_emulate_instruction);
de7d789a 4645
cf8f70bf 4646int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
de7d789a 4647{
cf8f70bf 4648 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
ca1d4a9e
AK		 4649 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4650 size, port, &val, 1);
cf8f70bf 4651 /* do not return to emulator after return from userspace */
7972995b 4652 vcpu->arch.pio.count = 0;
de7d789a
CO		 4653 return ret;
4654}
cf8f70bf 4655EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
de7d789a 4656
8cfdc000
ZA		 4657static void tsc_bad(void *info)
4658{
0a3aee0d 4659 __this_cpu_write(cpu_tsc_khz, 0);
8cfdc000
ZA		 4660}
4661
4662static void tsc_khz_changed(void *data)
c8076604 4663{
8cfdc000
ZA		 4664 struct cpufreq_freqs *freq = data;
4665 unsigned long khz = 0;
4666
4667 if (data)
4668 khz = freq->new;
4669 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4670 khz = cpufreq_quick_get(raw_smp_processor_id());
4671 if (!khz)
4672 khz = tsc_khz;
0a3aee0d 4673 __this_cpu_write(cpu_tsc_khz, khz);
c8076604
GH		 4674}
4675
c8076604
GH		 4676static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4677 void *data)
4678{
4679 struct cpufreq_freqs *freq = data;
4680 struct kvm *kvm;
4681 struct kvm_vcpu *vcpu;
4682 int i, send_ipi = 0;
4683
8cfdc000
ZA		 4684 /*
4685 * We allow guests to temporarily run on slowing clocks,
4686 * provided we notify them after, or to run on accelerating
4687 * clocks, provided we notify them before. Thus time never
4688 * goes backwards.
4689 *
4690 * However, we have a problem. We can't atomically update
4691 * the frequency of a given CPU from this function; it is
4692 * merely a notifier, which can be called from any CPU.
4693 * Changing the TSC frequency at arbitrary points in time
4694 * requires a recomputation of local variables related to
4695 * the TSC for each VCPU. We must flag these local variables
4696 * to be updated and be sure the update takes place with the
4697 * new frequency before any guests proceed.
4698 *
4699 * Unfortunately, the combination of hotplug CPU and frequency
4700 * change creates an intractable locking scenario; the order
4701 * of when these callouts happen is undefined with respect to
4702 * CPU hotplug, and they can race with each other. As such,
4703 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4704 * undefined; you can actually have a CPU frequency change take
4705 * place in between the computation of X and the setting of the
4706 * variable. To protect against this problem, all updates of
4707 * the per_cpu tsc_khz variable are done in an interrupt
4708 * protected IPI, and all callers wishing to update the value
4709 * must wait for a synchronous IPI to complete (which is trivial
4710 * if the caller is on the CPU already). This establishes the
4711 * necessary total order on variable updates.
4712 *
4713 * Note that because a guest time update may take place
4714 * anytime after the setting of the VCPU's request bit, the
4715 * correct TSC value must be set before the request. However,
4716 * to ensure the update actually makes it to any guest which
4717 * starts running in hardware virtualization between the set
4718 * and the acquisition of the spinlock, we must also ping the
4719 * CPU after setting the request bit.
4720 *
4721 */
4722
c8076604
GH		 4723 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4724 return 0;
4725 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4726 return 0;
8cfdc000
ZA		 4727
4728 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604 4729
e935b837 4730 raw_spin_lock(&kvm_lock);
c8076604 4731 list_for_each_entry(kvm, &vm_list, vm_list) {
988a2cae 4732 kvm_for_each_vcpu(i, vcpu, kvm) {
c8076604
GH		 4733 if (vcpu->cpu != freq->cpu)
4734 continue;
c285545f 4735 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
c8076604 4736 if (vcpu->cpu != smp_processor_id())
8cfdc000 4737 send_ipi = 1;
c8076604
GH		 4738 }
4739 }
e935b837 4740 raw_spin_unlock(&kvm_lock);
c8076604
GH		 4741
4742 if (freq->old < freq->new && send_ipi) {
4743 /*
4744 * We upscale the frequency. Must make the guest
4745 * doesn't see old kvmclock values while running with
4746 * the new frequency, otherwise we risk the guest sees
4747 * time go backwards.
4748 *
4749 * In case we update the frequency for another cpu
4750 * (which might be in guest context) send an interrupt
4751 * to kick the cpu out of guest context. Next time
4752 * guest context is entered kvmclock will be updated,
4753 * so the guest will not see stale values.
4754 */
8cfdc000 4755 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604
GH		 4756 }
4757 return 0;
4758}
4759
4760static struct notifier_block kvmclock_cpufreq_notifier_block = {
8cfdc000
ZA		 4761 .notifier_call = kvmclock_cpufreq_notifier
4762};
4763
4764static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4765 unsigned long action, void *hcpu)
4766{
4767 unsigned int cpu = (unsigned long)hcpu;
4768
4769 switch (action) {
4770 case CPU_ONLINE:
4771 case CPU_DOWN_FAILED:
4772 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4773 break;
4774 case CPU_DOWN_PREPARE:
4775 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4776 break;
4777 }
4778 return NOTIFY_OK;
4779}
4780
4781static struct notifier_block kvmclock_cpu_notifier_block = {
4782 .notifier_call = kvmclock_cpu_notifier,
4783 .priority = -INT_MAX
c8076604
GH		 4784};
4785
b820cc0c
ZA		 4786static void kvm_timer_init(void)
4787{
4788 int cpu;
4789
c285545f 4790 max_tsc_khz = tsc_khz;
8cfdc000 4791 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
b820cc0c 4792 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
c285545f
ZA		 4793#ifdef CONFIG_CPU_FREQ
4794 struct cpufreq_policy policy;
4795 memset(&policy, 0, sizeof(policy));
3e26f230
AK		 4796 cpu = get_cpu();
4797 cpufreq_get_policy(&policy, cpu);
c285545f
ZA		 4798 if (policy.cpuinfo.max_freq)
4799 max_tsc_khz = policy.cpuinfo.max_freq;
3e26f230 4800 put_cpu();
c285545f 4801#endif
b820cc0c
ZA		 4802 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4803 CPUFREQ_TRANSITION_NOTIFIER);
4804 }
c285545f 4805 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
8cfdc000
ZA		 4806 for_each_online_cpu(cpu)
4807 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
b820cc0c
ZA		 4808}
4809
ff9d07a0
ZY		 4810static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4811
f5132b01 4812int kvm_is_in_guest(void)
ff9d07a0 4813{
086c9855 4814 return __this_cpu_read(current_vcpu) != NULL;
ff9d07a0
ZY		 4815}
4816
4817static int kvm_is_user_mode(void)
4818{
4819 int user_mode = 3;
dcf46b94 4820
086c9855
AS		 4821 if (__this_cpu_read(current_vcpu))
4822 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
dcf46b94 4823
ff9d07a0
ZY		 4824 return user_mode != 0;
4825}
4826
4827static unsigned long kvm_get_guest_ip(void)
4828{
4829 unsigned long ip = 0;
dcf46b94 4830
086c9855
AS		 4831 if (__this_cpu_read(current_vcpu))
4832 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
dcf46b94 4833
ff9d07a0
ZY		 4834 return ip;
4835}
4836
4837static struct perf_guest_info_callbacks kvm_guest_cbs = {
4838 .is_in_guest = kvm_is_in_guest,
4839 .is_user_mode = kvm_is_user_mode,
4840 .get_guest_ip = kvm_get_guest_ip,
4841};
4842
4843void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4844{
086c9855 4845 __this_cpu_write(current_vcpu, vcpu);
ff9d07a0
ZY		 4846}
4847EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4848
4849void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4850{
086c9855 4851 __this_cpu_write(current_vcpu, NULL);
ff9d07a0
ZY		 4852}
4853EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4854
ce88decf
XG		 4855static void kvm_set_mmio_spte_mask(void)
4856{
4857 u64 mask;
4858 int maxphyaddr = boot_cpu_data.x86_phys_bits;
4859
4860 /*
4861 * Set the reserved bits and the present bit of an paging-structure
4862 * entry to generate page fault with PFER.RSV = 1.
4863 */
4864 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
4865 mask |= 1ull;
4866
4867#ifdef CONFIG_X86_64
4868 /*
4869 * If reserved bit is not supported, clear the present bit to disable
4870 * mmio page fault.
4871 */
4872 if (maxphyaddr == 52)
4873 mask &= ~1ull;
4874#endif
4875
4876 kvm_mmu_set_mmio_spte_mask(mask);
4877}
4878
f8c16bba 4879int kvm_arch_init(void *opaque)
043405e1 4880{
b820cc0c 4881 int r;
f8c16bba
ZX		 4882 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4883
f8c16bba
ZX		 4884 if (kvm_x86_ops) {
4885 printk(KERN_ERR "kvm: already loaded the other module\n");
56c6d28a
ZX		 4886 r = -EEXIST;
4887 goto out;
f8c16bba
ZX		 4888 }
4889
4890 if (!ops->cpu_has_kvm_support()) {
4891 printk(KERN_ERR "kvm: no hardware support\n");
56c6d28a
ZX		 4892 r = -EOPNOTSUPP;
4893 goto out;
f8c16bba
ZX		 4894 }
4895 if (ops->disabled_by_bios()) {
4896 printk(KERN_ERR "kvm: disabled by bios\n");
56c6d28a
ZX		 4897 r = -EOPNOTSUPP;
4898 goto out;
f8c16bba
ZX		 4899 }
4900
97db56ce
AK		 4901 r = kvm_mmu_module_init();
4902 if (r)
4903 goto out;
4904
ce88decf 4905 kvm_set_mmio_spte_mask();
97db56ce
AK		 4906 kvm_init_msr_list();
4907
f8c16bba 4908 kvm_x86_ops = ops;
7b52345e 4909 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4b12f0de 4910 PT_DIRTY_MASK, PT64_NX_MASK, 0);
c8076604 4911
b820cc0c 4912 kvm_timer_init();
c8076604 4913
ff9d07a0
ZY		 4914 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4915
2acf923e
DC		 4916 if (cpu_has_xsave)
4917 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4918
f8c16bba 4919 return 0;
56c6d28a
ZX		 4920
4921out:
56c6d28a 4922 return r;
043405e1 4923}
8776e519 4924
f8c16bba
ZX		 4925void kvm_arch_exit(void)
4926{
ff9d07a0
ZY		 4927 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4928
888d256e
JK		 4929 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4930 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4931 CPUFREQ_TRANSITION_NOTIFIER);
8cfdc000 4932 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
f8c16bba 4933 kvm_x86_ops = NULL;
56c6d28a
ZX		 4934 kvm_mmu_module_exit();
4935}
f8c16bba 4936
8776e519
HB		 4937int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4938{
4939 ++vcpu->stat.halt_exits;
4940 if (irqchip_in_kernel(vcpu->kvm)) {
a4535290 4941 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
8776e519
HB		 4942 return 1;
4943 } else {
4944 vcpu->run->exit_reason = KVM_EXIT_HLT;
4945 return 0;
4946 }
4947}
4948EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4949
55cd8e5a
GN		 4950int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4951{
4952 u64 param, ingpa, outgpa, ret;
4953 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4954 bool fast, longmode;
4955 int cs_db, cs_l;
4956
4957 /*
4958 * hypercall generates UD from non zero cpl and real mode
4959 * per HYPER-V spec
4960 */
3eeb3288 4961 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
55cd8e5a
GN		 4962 kvm_queue_exception(vcpu, UD_VECTOR);
4963 return 0;
4964 }
4965
4966 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4967 longmode = is_long_mode(vcpu) && cs_l == 1;
4968
4969 if (!longmode) {
ccd46936
GN		 4970 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4971 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4972 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4973 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4974 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4975 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
55cd8e5a
GN		 4976 }
4977#ifdef CONFIG_X86_64
4978 else {
4979 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4980 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4981 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4982 }
4983#endif
4984
4985 code = param & 0xffff;
4986 fast = (param >> 16) & 0x1;
4987 rep_cnt = (param >> 32) & 0xfff;
4988 rep_idx = (param >> 48) & 0xfff;
4989
4990 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4991
c25bc163
GN		 4992 switch (code) {
4993 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4994 kvm_vcpu_on_spin(vcpu);
4995 break;
4996 default:
4997 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4998 break;
4999 }
55cd8e5a
GN		 5000
5001 ret = res | (((u64)rep_done & 0xfff) << 32);
5002 if (longmode) {
5003 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5004 } else {
5005 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5006 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5007 }
5008
5009 return 1;
5010}
5011
8776e519
HB		 5012int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5013{
5014 unsigned long nr, a0, a1, a2, a3, ret;
2f333bcb 5015 int r = 1;
8776e519 5016
55cd8e5a
GN		 5017 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5018 return kvm_hv_hypercall(vcpu);
5019
5fdbf976
MT		 5020 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5021 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5022 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5023 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5024 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
8776e519 5025
229456fc 5026 trace_kvm_hypercall(nr, a0, a1, a2, a3);
2714d1d3 5027
8776e519
HB		 5028 if (!is_long_mode(vcpu)) {
5029 nr &= 0xFFFFFFFF;
5030 a0 &= 0xFFFFFFFF;
5031 a1 &= 0xFFFFFFFF;
5032 a2 &= 0xFFFFFFFF;
5033 a3 &= 0xFFFFFFFF;
5034 }
5035
07708c4a
JK		 5036 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5037 ret = -KVM_EPERM;
5038 goto out;
5039 }
5040
8776e519 5041 switch (nr) {
b93463aa
AK		 5042 case KVM_HC_VAPIC_POLL_IRQ:
5043 ret = 0;
5044 break;
8776e519
HB		 5045 default:
5046 ret = -KVM_ENOSYS;
5047 break;
5048 }
07708c4a 5049out:
5fdbf976 5050 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
f11c3a8d 5051 ++vcpu->stat.hypercalls;
2f333bcb 5052 return r;
8776e519
HB		 5053}
5054EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5055
d6aa1000 5056int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8776e519 5057{
d6aa1000 5058 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8776e519 5059 char instruction[3];
5fdbf976 5060 unsigned long rip = kvm_rip_read(vcpu);
8776e519 5061
8776e519
HB		 5062 /*
5063 * Blow out the MMU to ensure that no other VCPU has an active mapping
5064 * to ensure that the updated hypercall appears atomically across all
5065 * VCPUs.
5066 */
5067 kvm_mmu_zap_all(vcpu->kvm);
5068
8776e519 5069 kvm_x86_ops->patch_hypercall(vcpu, instruction);
8776e519 5070
9d74191a 5071 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
8776e519
HB		 5072}
5073
b6c7a5dc
HB		 5074/*
5075 * Check if userspace requested an interrupt window, and that the
5076 * interrupt window is open.
5077 *
5078 * No need to exit to userspace if we already have an interrupt queued.
5079 */
851ba692 5080static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
b6c7a5dc 5081{
8061823a 5082 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
851ba692 5083 vcpu->run->request_interrupt_window &&
5df56646 5084 kvm_arch_interrupt_allowed(vcpu));
b6c7a5dc
HB		 5085}
5086
851ba692 5087static void post_kvm_run_save(struct kvm_vcpu *vcpu)
b6c7a5dc 5088{
851ba692
AK		 5089 struct kvm_run *kvm_run = vcpu->run;
5090
91586a3b 5091 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2d3ad1f4 5092 kvm_run->cr8 = kvm_get_cr8(vcpu);
b6c7a5dc 5093 kvm_run->apic_base = kvm_get_apic_base(vcpu);
4531220b 5094 if (irqchip_in_kernel(vcpu->kvm))
b6c7a5dc 5095 kvm_run->ready_for_interrupt_injection = 1;
4531220b 5096 else
b6c7a5dc 5097 kvm_run->ready_for_interrupt_injection =
fa9726b0
GN		 5098 kvm_arch_interrupt_allowed(vcpu) &&
5099 !kvm_cpu_has_interrupt(vcpu) &&
5100 !kvm_event_needs_reinjection(vcpu);
b6c7a5dc
HB		 5101}
5102
b93463aa
AK		 5103static void vapic_enter(struct kvm_vcpu *vcpu)
5104{
5105 struct kvm_lapic *apic = vcpu->arch.apic;
5106 struct page *page;
5107
5108 if (!apic || !apic->vapic_addr)
5109 return;
5110
5111 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
72dc67a6
IE		 5112
5113 vcpu->arch.apic->vapic_page = page;
b93463aa
AK		 5114}
5115
5116static void vapic_exit(struct kvm_vcpu *vcpu)
5117{
5118 struct kvm_lapic *apic = vcpu->arch.apic;
f656ce01 5119 int idx;
b93463aa
AK		 5120
5121 if (!apic || !apic->vapic_addr)
5122 return;
5123
f656ce01 5124 idx = srcu_read_lock(&vcpu->kvm->srcu);
b93463aa
AK		 5125 kvm_release_page_dirty(apic->vapic_page);
5126 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
f656ce01 5127 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b93463aa
AK		 5128}
5129
95ba8273
GN		 5130static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5131{
5132 int max_irr, tpr;
5133
5134 if (!kvm_x86_ops->update_cr8_intercept)
5135 return;
5136
88c808fd
AK		 5137 if (!vcpu->arch.apic)
5138 return;
5139
8db3baa2
GN		 5140 if (!vcpu->arch.apic->vapic_addr)
5141 max_irr = kvm_lapic_find_highest_irr(vcpu);
5142 else
5143 max_irr = -1;
95ba8273
GN		 5144
5145 if (max_irr != -1)
5146 max_irr >>= 4;
5147
5148 tpr = kvm_lapic_get_cr8(vcpu);
5149
5150 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5151}
5152
851ba692 5153static void inject_pending_event(struct kvm_vcpu *vcpu)
95ba8273
GN		 5154{
5155 /* try to reinject previous events if any */
b59bb7bd 5156 if (vcpu->arch.exception.pending) {
5c1c85d0
AK		 5157 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5158 vcpu->arch.exception.has_error_code,
5159 vcpu->arch.exception.error_code);
b59bb7bd
GN		 5160 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5161 vcpu->arch.exception.has_error_code,
ce7ddec4
JR		 5162 vcpu->arch.exception.error_code,
5163 vcpu->arch.exception.reinject);
b59bb7bd
GN		 5164 return;
5165 }
5166
95ba8273
GN		 5167 if (vcpu->arch.nmi_injected) {
5168 kvm_x86_ops->set_nmi(vcpu);
5169 return;
5170 }
5171
5172 if (vcpu->arch.interrupt.pending) {
66fd3f7f 5173 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 5174 return;
5175 }
5176
5177 /* try to inject new event if pending */
5178 if (vcpu->arch.nmi_pending) {
5179 if (kvm_x86_ops->nmi_allowed(vcpu)) {
7460fb4a 5180 --vcpu->arch.nmi_pending;
95ba8273
GN		 5181 vcpu->arch.nmi_injected = true;
5182 kvm_x86_ops->set_nmi(vcpu);
5183 }
5184 } else if (kvm_cpu_has_interrupt(vcpu)) {
5185 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
66fd3f7f
GN		 5186 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5187 false);
5188 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 5189 }
5190 }
5191}
5192
2acf923e
DC		 5193static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5194{
5195 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5196 !vcpu->guest_xcr0_loaded) {
5197 /* kvm_set_xcr() also depends on this */
5198 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5199 vcpu->guest_xcr0_loaded = 1;
5200 }
5201}
5202
5203static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5204{
5205 if (vcpu->guest_xcr0_loaded) {
5206 if (vcpu->arch.xcr0 != host_xcr0)
5207 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5208 vcpu->guest_xcr0_loaded = 0;
5209 }
5210}
5211
7460fb4a
AK		 5212static void process_nmi(struct kvm_vcpu *vcpu)
5213{
5214 unsigned limit = 2;
5215
5216 /*
5217 * x86 is limited to one NMI running, and one NMI pending after it.
5218 * If an NMI is already in progress, limit further NMIs to just one.
5219 * Otherwise, allow two (and we'll inject the first one immediately).
5220 */
5221 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5222 limit = 1;
5223
5224 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5225 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5226 kvm_make_request(KVM_REQ_EVENT, vcpu);
5227}
5228
851ba692 5229static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
b6c7a5dc
HB		 5230{
5231 int r;
6a8b1d13 5232 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
851ba692 5233 vcpu->run->request_interrupt_window;
d6185f20 5234 bool req_immediate_exit = 0;
b6c7a5dc 5235
3e007509 5236 if (vcpu->requests) {
a8eeb04a 5237 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
2e53d63a 5238 kvm_mmu_unload(vcpu);
a8eeb04a 5239 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
2f599714 5240 __kvm_migrate_timers(vcpu);
34c238a1
ZA		 5241 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5242 r = kvm_guest_time_update(vcpu);
8cfdc000
ZA		 5243 if (unlikely(r))
5244 goto out;
5245 }
a8eeb04a 5246 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
4731d4c7 5247 kvm_mmu_sync_roots(vcpu);
a8eeb04a 5248 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
d4acf7e7 5249 kvm_x86_ops->tlb_flush(vcpu);
a8eeb04a 5250 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
851ba692 5251 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
b93463aa
AK		 5252 r = 0;
5253 goto out;
5254 }
a8eeb04a 5255 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
851ba692 5256 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
71c4dfaf
JR		 5257 r = 0;
5258 goto out;
5259 }
a8eeb04a 5260 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
02daab21
AK		 5261 vcpu->fpu_active = 0;
5262 kvm_x86_ops->fpu_deactivate(vcpu);
5263 }
af585b92
GN		 5264 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5265 /* Page is swapped out. Do synthetic halt */
5266 vcpu->arch.apf.halted = true;
5267 r = 1;
5268 goto out;
5269 }
c9aaa895
GC		 5270 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5271 record_steal_time(vcpu);
7460fb4a
AK		 5272 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5273 process_nmi(vcpu);
d6185f20
NHE		 5274 req_immediate_exit =
5275 kvm_check_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
f5132b01
GN		 5276 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5277 kvm_handle_pmu_event(vcpu);
5278 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5279 kvm_deliver_pmi(vcpu);
2f52d58c 5280 }
b93463aa 5281
3e007509
AK		 5282 r = kvm_mmu_reload(vcpu);
5283 if (unlikely(r))
5284 goto out;
5285
b463a6f7
AK		 5286 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5287 inject_pending_event(vcpu);
5288
5289 /* enable NMI/IRQ window open exits if needed */
7460fb4a 5290 if (vcpu->arch.nmi_pending)
b463a6f7
AK		 5291 kvm_x86_ops->enable_nmi_window(vcpu);
5292 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5293 kvm_x86_ops->enable_irq_window(vcpu);
5294
5295 if (kvm_lapic_enabled(vcpu)) {
5296 update_cr8_intercept(vcpu);
5297 kvm_lapic_sync_to_vapic(vcpu);
5298 }
5299 }
5300
b6c7a5dc
HB		 5301 preempt_disable();
5302
5303 kvm_x86_ops->prepare_guest_switch(vcpu);
2608d7a1
AK		 5304 if (vcpu->fpu_active)
5305 kvm_load_guest_fpu(vcpu);
2acf923e 5306 kvm_load_guest_xcr0(vcpu);
b6c7a5dc 5307
6b7e2d09
XG		 5308 vcpu->mode = IN_GUEST_MODE;
5309
5310 /* We should set ->mode before check ->requests,
5311 * see the comment in make_all_cpus_request.
5312 */
5313 smp_mb();
b6c7a5dc 5314
d94e1dc9 5315 local_irq_disable();
32f88400 5316
6b7e2d09 5317 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
d94e1dc9 5318 || need_resched() || signal_pending(current)) {
6b7e2d09 5319 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 5320 smp_wmb();
6c142801
AK		 5321 local_irq_enable();
5322 preempt_enable();
b463a6f7 5323 kvm_x86_ops->cancel_injection(vcpu);
6c142801
AK		 5324 r = 1;
5325 goto out;
5326 }
5327
f656ce01 5328 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
3200f405 5329
d6185f20
NHE		 5330 if (req_immediate_exit)
5331 smp_send_reschedule(vcpu->cpu);
5332
b6c7a5dc
HB		 5333 kvm_guest_enter();
5334
42dbaa5a 5335 if (unlikely(vcpu->arch.switch_db_regs)) {
42dbaa5a
JK		 5336 set_debugreg(0, 7);
5337 set_debugreg(vcpu->arch.eff_db[0], 0);
5338 set_debugreg(vcpu->arch.eff_db[1], 1);
5339 set_debugreg(vcpu->arch.eff_db[2], 2);
5340 set_debugreg(vcpu->arch.eff_db[3], 3);
5341 }
b6c7a5dc 5342
229456fc 5343 trace_kvm_entry(vcpu->vcpu_id);
851ba692 5344 kvm_x86_ops->run(vcpu);
b6c7a5dc 5345
24f1e32c
FW		 5346 /*
5347 * If the guest has used debug registers, at least dr7
5348 * will be disabled while returning to the host.
5349 * If we don't have active breakpoints in the host, we don't
5350 * care about the messed up debug address registers. But if
5351 * we have some of them active, restore the old state.
5352 */
59d8eb53 5353 if (hw_breakpoint_active())
24f1e32c 5354 hw_breakpoint_restore();
42dbaa5a 5355
d5c1785d 5356 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
1d5f066e 5357
6b7e2d09 5358 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 5359 smp_wmb();
b6c7a5dc
HB		 5360 local_irq_enable();
5361
5362 ++vcpu->stat.exits;
5363
5364 /*
5365 * We must have an instruction between local_irq_enable() and
5366 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5367 * the interrupt shadow. The stat.exits increment will do nicely.
5368 * But we need to prevent reordering, hence this barrier():
5369 */
5370 barrier();
5371
5372 kvm_guest_exit();
5373
5374 preempt_enable();
5375
f656ce01 5376 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3200f405 5377
b6c7a5dc
HB		 5378 /*
5379 * Profile KVM exit RIPs:
5380 */
5381 if (unlikely(prof_on == KVM_PROFILING)) {
5fdbf976
MT		 5382 unsigned long rip = kvm_rip_read(vcpu);
5383 profile_hit(KVM_PROFILING, (void *)rip);
b6c7a5dc
HB		 5384 }
5385
cc578287
ZA		 5386 if (unlikely(vcpu->arch.tsc_always_catchup))
5387 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
298101da 5388
b93463aa
AK		 5389 kvm_lapic_sync_from_vapic(vcpu);
5390
851ba692 5391 r = kvm_x86_ops->handle_exit(vcpu);
d7690175
MT		 5392out:
5393 return r;
5394}
b6c7a5dc 5395
09cec754 5396
851ba692 5397static int __vcpu_run(struct kvm_vcpu *vcpu)
d7690175
MT		 5398{
5399 int r;
f656ce01 5400 struct kvm *kvm = vcpu->kvm;
d7690175
MT		 5401
5402 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
1b10bf31
JK		 5403 pr_debug("vcpu %d received sipi with vector # %x\n",
5404 vcpu->vcpu_id, vcpu->arch.sipi_vector);
d7690175 5405 kvm_lapic_reset(vcpu);
5f179287 5406 r = kvm_arch_vcpu_reset(vcpu);
d7690175
MT		 5407 if (r)
5408 return r;
5409 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
b6c7a5dc
HB		 5410 }
5411
f656ce01 5412 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175
MT		 5413 vapic_enter(vcpu);
5414
5415 r = 1;
5416 while (r > 0) {
af585b92
GN		 5417 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5418 !vcpu->arch.apf.halted)
851ba692 5419 r = vcpu_enter_guest(vcpu);
d7690175 5420 else {
f656ce01 5421 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
d7690175 5422 kvm_vcpu_block(vcpu);
f656ce01 5423 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
a8eeb04a 5424 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
09cec754
GN		 5425 {
5426 switch(vcpu->arch.mp_state) {
5427 case KVM_MP_STATE_HALTED:
d7690175 5428 vcpu->arch.mp_state =
09cec754
GN		 5429 KVM_MP_STATE_RUNNABLE;
5430 case KVM_MP_STATE_RUNNABLE:
af585b92 5431 vcpu->arch.apf.halted = false;
09cec754
GN		 5432 break;
5433 case KVM_MP_STATE_SIPI_RECEIVED:
5434 default:
5435 r = -EINTR;
5436 break;
5437 }
5438 }
d7690175
MT		 5439 }
5440
09cec754
GN		 5441 if (r <= 0)
5442 break;
5443
5444 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5445 if (kvm_cpu_has_pending_timer(vcpu))
5446 kvm_inject_pending_timer_irqs(vcpu);
5447
851ba692 5448 if (dm_request_for_irq_injection(vcpu)) {
09cec754 5449 r = -EINTR;
851ba692 5450 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754
GN		 5451 ++vcpu->stat.request_irq_exits;
5452 }
af585b92
GN		 5453
5454 kvm_check_async_pf_completion(vcpu);
5455
09cec754
GN		 5456 if (signal_pending(current)) {
5457 r = -EINTR;
851ba692 5458 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754
GN		 5459 ++vcpu->stat.signal_exits;
5460 }
5461 if (need_resched()) {
f656ce01 5462 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
09cec754 5463 kvm_resched(vcpu);
f656ce01 5464 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 5465 }
b6c7a5dc
HB		 5466 }
5467
f656ce01 5468 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
b6c7a5dc 5469
b93463aa
AK		 5470 vapic_exit(vcpu);
5471
b6c7a5dc
HB		 5472 return r;
5473}
5474
f78146b0
AK		 5475/*
5476 * Implements the following, as a state machine:
5477 *
5478 * read:
5479 * for each fragment
5480 * write gpa, len
5481 * exit
5482 * copy data
5483 * execute insn
5484 *
5485 * write:
5486 * for each fragment
5487 * write gpa, len
5488 * copy data
5489 * exit
5490 */
5287f194
AK		 5491static int complete_mmio(struct kvm_vcpu *vcpu)
5492{
5493 struct kvm_run *run = vcpu->run;
f78146b0 5494 struct kvm_mmio_fragment *frag;
5287f194
AK		 5495 int r;
5496
5497 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5498 return 1;
5499
5500 if (vcpu->mmio_needed) {
f78146b0
AK		 5501 /* Complete previous fragment */
5502 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment++];
cef4dea0 5503 if (!vcpu->mmio_is_write)
f78146b0
AK		 5504 memcpy(frag->data, run->mmio.data, frag->len);
5505 if (vcpu->mmio_cur_fragment == vcpu->mmio_nr_fragments) {
5506 vcpu->mmio_needed = 0;
5507 if (vcpu->mmio_is_write)
5508 return 1;
5509 vcpu->mmio_read_completed = 1;
5510 goto done;
cef4dea0 5511 }
f78146b0
AK		 5512 /* Initiate next fragment */
5513 ++frag;
5514 run->exit_reason = KVM_EXIT_MMIO;
5515 run->mmio.phys_addr = frag->gpa;
cef4dea0 5516 if (vcpu->mmio_is_write)
f78146b0
AK		 5517 memcpy(run->mmio.data, frag->data, frag->len);
5518 run->mmio.len = frag->len;
5519 run->mmio.is_write = vcpu->mmio_is_write;
5520 return 0;
5521
5287f194 5522 }
f78146b0 5523done:
5287f194
AK		 5524 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5525 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5526 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5527 if (r != EMULATE_DONE)
5528 return 0;
5529 return 1;
5530}
5531
b6c7a5dc
HB		 5532int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5533{
5534 int r;
5535 sigset_t sigsaved;
5536
e5c30142
AK		 5537 if (!tsk_used_math(current) && init_fpu(current))
5538 return -ENOMEM;
5539
ac9f6dc0
AK		 5540 if (vcpu->sigset_active)
5541 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5542
a4535290 5543 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
b6c7a5dc 5544 kvm_vcpu_block(vcpu);
d7690175 5545 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
ac9f6dc0
AK		 5546 r = -EAGAIN;
5547 goto out;
b6c7a5dc
HB		 5548 }
5549
b6c7a5dc 5550 /* re-sync apic's tpr */
eea1cff9
AP		 5551 if (!irqchip_in_kernel(vcpu->kvm)) {
5552 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5553 r = -EINVAL;
5554 goto out;
5555 }
5556 }
b6c7a5dc 5557
5287f194
AK		 5558 r = complete_mmio(vcpu);
5559 if (r <= 0)
5560 goto out;
5561
851ba692 5562 r = __vcpu_run(vcpu);
b6c7a5dc
HB		 5563
5564out:
f1d86e46 5565 post_kvm_run_save(vcpu);
b6c7a5dc
HB		 5566 if (vcpu->sigset_active)
5567 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5568
b6c7a5dc
HB		 5569 return r;
5570}
5571
5572int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5573{
7ae441ea
GN		 5574 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5575 /*
5576 * We are here if userspace calls get_regs() in the middle of
5577 * instruction emulation. Registers state needs to be copied
5578 * back from emulation context to vcpu. Usrapace shouldn't do
5579 * that usually, but some bad designed PV devices (vmware
5580 * backdoor interface) need this to work
5581 */
9dac77fa
AK		 5582 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5583 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
7ae441ea
GN		 5584 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5585 }
5fdbf976
MT		 5586 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5587 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5588 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5589 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5590 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5591 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5592 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5593 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
b6c7a5dc 5594#ifdef CONFIG_X86_64
5fdbf976
MT		 5595 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5596 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5597 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5598 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5599 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5600 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5601 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5602 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
b6c7a5dc
HB		 5603#endif
5604
5fdbf976 5605 regs->rip = kvm_rip_read(vcpu);
91586a3b 5606 regs->rflags = kvm_get_rflags(vcpu);
b6c7a5dc 5607
b6c7a5dc
HB		 5608 return 0;
5609}
5610
5611int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5612{
7ae441ea
GN		 5613 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5614 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5615
5fdbf976
MT		 5616 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5617 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5618 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5619 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5620 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5621 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5622 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5623 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
b6c7a5dc 5624#ifdef CONFIG_X86_64
5fdbf976
MT		 5625 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5626 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5627 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5628 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5629 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5630 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5631 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5632 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
b6c7a5dc
HB		 5633#endif
5634
5fdbf976 5635 kvm_rip_write(vcpu, regs->rip);
91586a3b 5636 kvm_set_rflags(vcpu, regs->rflags);
b6c7a5dc 5637
b4f14abd
JK		 5638 vcpu->arch.exception.pending = false;
5639
3842d135
AK		 5640 kvm_make_request(KVM_REQ_EVENT, vcpu);
5641
b6c7a5dc
HB		 5642 return 0;
5643}
5644
b6c7a5dc
HB		 5645void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5646{
5647 struct kvm_segment cs;
5648
3e6e0aab 5649 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
b6c7a5dc
HB		 5650 *db = cs.db;
5651 *l = cs.l;
5652}
5653EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5654
5655int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5656 struct kvm_sregs *sregs)
5657{
89a27f4d 5658 struct desc_ptr dt;
b6c7a5dc 5659
3e6e0aab
GT		 5660 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5661 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5662 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5663 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5664 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5665 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 5666
3e6e0aab
GT		 5667 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5668 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc
HB		 5669
5670 kvm_x86_ops->get_idt(vcpu, &dt);
89a27f4d
GN		 5671 sregs->idt.limit = dt.size;
5672 sregs->idt.base = dt.address;
b6c7a5dc 5673 kvm_x86_ops->get_gdt(vcpu, &dt);
89a27f4d
GN		 5674 sregs->gdt.limit = dt.size;
5675 sregs->gdt.base = dt.address;
b6c7a5dc 5676
4d4ec087 5677 sregs->cr0 = kvm_read_cr0(vcpu);
ad312c7c 5678 sregs->cr2 = vcpu->arch.cr2;
9f8fe504 5679 sregs->cr3 = kvm_read_cr3(vcpu);
fc78f519 5680 sregs->cr4 = kvm_read_cr4(vcpu);
2d3ad1f4 5681 sregs->cr8 = kvm_get_cr8(vcpu);
f6801dff 5682 sregs->efer = vcpu->arch.efer;
b6c7a5dc
HB		 5683 sregs->apic_base = kvm_get_apic_base(vcpu);
5684
923c61bb 5685 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
b6c7a5dc 5686
36752c9b 5687 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
14d0bc1f
GN		 5688 set_bit(vcpu->arch.interrupt.nr,
5689 (unsigned long *)sregs->interrupt_bitmap);
16d7a191 5690
b6c7a5dc
HB		 5691 return 0;
5692}
5693
62d9f0db
MT		 5694int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5695 struct kvm_mp_state *mp_state)
5696{
62d9f0db 5697 mp_state->mp_state = vcpu->arch.mp_state;
62d9f0db
MT		 5698 return 0;
5699}
5700
5701int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5702 struct kvm_mp_state *mp_state)
5703{
62d9f0db 5704 vcpu->arch.mp_state = mp_state->mp_state;
3842d135 5705 kvm_make_request(KVM_REQ_EVENT, vcpu);
62d9f0db
MT		 5706 return 0;
5707}
5708
7f3d35fd
KW		 5709int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
5710 int reason, bool has_error_code, u32 error_code)
b6c7a5dc 5711{
9d74191a 5712 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d 5713 int ret;
e01c2426 5714
8ec4722d 5715 init_emulate_ctxt(vcpu);
c697518a 5716
7f3d35fd 5717 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9d74191a 5718 has_error_code, error_code);
c697518a 5719
c697518a 5720 if (ret)
19d04437 5721 return EMULATE_FAIL;
37817f29 5722
9dac77fa 5723 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
9d74191a
TY		 5724 kvm_rip_write(vcpu, ctxt->eip);
5725 kvm_set_rflags(vcpu, ctxt->eflags);
3842d135 5726 kvm_make_request(KVM_REQ_EVENT, vcpu);
19d04437 5727 return EMULATE_DONE;
37817f29
IE		 5728}
5729EXPORT_SYMBOL_GPL(kvm_task_switch);
5730
b6c7a5dc
HB		 5731int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5732 struct kvm_sregs *sregs)
5733{
5734 int mmu_reset_needed = 0;
63f42e02 5735 int pending_vec, max_bits, idx;
89a27f4d 5736 struct desc_ptr dt;
b6c7a5dc 5737
89a27f4d
GN		 5738 dt.size = sregs->idt.limit;
5739 dt.address = sregs->idt.base;
b6c7a5dc 5740 kvm_x86_ops->set_idt(vcpu, &dt);
89a27f4d
GN		 5741 dt.size = sregs->gdt.limit;
5742 dt.address = sregs->gdt.base;
b6c7a5dc
HB		 5743 kvm_x86_ops->set_gdt(vcpu, &dt);
5744
ad312c7c 5745 vcpu->arch.cr2 = sregs->cr2;
9f8fe504 5746 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
dc7e795e 5747 vcpu->arch.cr3 = sregs->cr3;
aff48baa 5748 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
b6c7a5dc 5749
2d3ad1f4 5750 kvm_set_cr8(vcpu, sregs->cr8);
b6c7a5dc 5751
f6801dff 5752 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
b6c7a5dc 5753 kvm_x86_ops->set_efer(vcpu, sregs->efer);
b6c7a5dc
HB		 5754 kvm_set_apic_base(vcpu, sregs->apic_base);
5755
4d4ec087 5756 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
b6c7a5dc 5757 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
d7306163 5758 vcpu->arch.cr0 = sregs->cr0;
b6c7a5dc 5759
fc78f519 5760 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
b6c7a5dc 5761 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3ea3aa8c 5762 if (sregs->cr4 & X86_CR4_OSXSAVE)
00b27a3e 5763 kvm_update_cpuid(vcpu);
63f42e02
XG		 5764
5765 idx = srcu_read_lock(&vcpu->kvm->srcu);
7c93be44 5766 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
9f8fe504 5767 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7c93be44
MT		 5768 mmu_reset_needed = 1;
5769 }
63f42e02 5770 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b6c7a5dc
HB		 5771
5772 if (mmu_reset_needed)
5773 kvm_mmu_reset_context(vcpu);
5774
923c61bb
GN		 5775 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5776 pending_vec = find_first_bit(
5777 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5778 if (pending_vec < max_bits) {
66fd3f7f 5779 kvm_queue_interrupt(vcpu, pending_vec, false);
923c61bb 5780 pr_debug("Set back pending irq %d\n", pending_vec);
b6c7a5dc
HB		 5781 }
5782
3e6e0aab
GT		 5783 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5784 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5785 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5786 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5787 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5788 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 5789
3e6e0aab
GT		 5790 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5791 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc 5792
5f0269f5
ME		 5793 update_cr8_intercept(vcpu);
5794
9c3e4aab 5795 /* Older userspace won't unhalt the vcpu on reset. */
c5af89b6 5796 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9c3e4aab 5797 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3eeb3288 5798 !is_protmode(vcpu))
9c3e4aab
MT		 5799 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5800
3842d135
AK		 5801 kvm_make_request(KVM_REQ_EVENT, vcpu);
5802
b6c7a5dc
HB		 5803 return 0;
5804}
5805
d0bfb940
JK		 5806int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5807 struct kvm_guest_debug *dbg)
b6c7a5dc 5808{
355be0b9 5809 unsigned long rflags;
ae675ef0 5810 int i, r;
b6c7a5dc 5811
4f926bf2
JK		 5812 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5813 r = -EBUSY;
5814 if (vcpu->arch.exception.pending)
2122ff5e 5815 goto out;
4f926bf2
JK		 5816 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5817 kvm_queue_exception(vcpu, DB_VECTOR);
5818 else
5819 kvm_queue_exception(vcpu, BP_VECTOR);
5820 }
5821
91586a3b
JK		 5822 /*
5823 * Read rflags as long as potentially injected trace flags are still
5824 * filtered out.
5825 */
5826 rflags = kvm_get_rflags(vcpu);
355be0b9
JK		 5827
5828 vcpu->guest_debug = dbg->control;
5829 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5830 vcpu->guest_debug = 0;
5831
5832 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
ae675ef0
JK		 5833 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5834 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5835 vcpu->arch.switch_db_regs =
5836 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5837 } else {
5838 for (i = 0; i < KVM_NR_DB_REGS; i++)
5839 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5840 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5841 }
5842
f92653ee
JK		 5843 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5844 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5845 get_segment_base(vcpu, VCPU_SREG_CS);
94fe45da 5846
91586a3b
JK		 5847 /*
5848 * Trigger an rflags update that will inject or remove the trace
5849 * flags.
5850 */
5851 kvm_set_rflags(vcpu, rflags);
b6c7a5dc 5852
355be0b9 5853 kvm_x86_ops->set_guest_debug(vcpu, dbg);
b6c7a5dc 5854
4f926bf2 5855 r = 0;
d0bfb940 5856
2122ff5e 5857out:
b6c7a5dc
HB		 5858
5859 return r;
5860}
5861
8b006791
ZX		 5862/*
5863 * Translate a guest virtual address to a guest physical address.
5864 */
5865int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5866 struct kvm_translation *tr)
5867{
5868 unsigned long vaddr = tr->linear_address;
5869 gpa_t gpa;
f656ce01 5870 int idx;
8b006791 5871
f656ce01 5872 idx = srcu_read_lock(&vcpu->kvm->srcu);
1871c602 5873 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
f656ce01 5874 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8b006791
ZX		 5875 tr->physical_address = gpa;
5876 tr->valid = gpa != UNMAPPED_GVA;
5877 tr->writeable = 1;
5878 tr->usermode = 0;
8b006791
ZX		 5879
5880 return 0;
5881}
5882
d0752060
HB		 5883int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5884{
98918833
SY		 5885 struct i387_fxsave_struct *fxsave =
5886 &vcpu->arch.guest_fpu.state->fxsave;
d0752060 5887
d0752060
HB		 5888 memcpy(fpu->fpr, fxsave->st_space, 128);
5889 fpu->fcw = fxsave->cwd;
5890 fpu->fsw = fxsave->swd;
5891 fpu->ftwx = fxsave->twd;
5892 fpu->last_opcode = fxsave->fop;
5893 fpu->last_ip = fxsave->rip;
5894 fpu->last_dp = fxsave->rdp;
5895 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5896
d0752060
HB		 5897 return 0;
5898}
5899
5900int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5901{
98918833
SY		 5902 struct i387_fxsave_struct *fxsave =
5903 &vcpu->arch.guest_fpu.state->fxsave;
d0752060 5904
d0752060
HB		 5905 memcpy(fxsave->st_space, fpu->fpr, 128);
5906 fxsave->cwd = fpu->fcw;
5907 fxsave->swd = fpu->fsw;
5908 fxsave->twd = fpu->ftwx;
5909 fxsave->fop = fpu->last_opcode;
5910 fxsave->rip = fpu->last_ip;
5911 fxsave->rdp = fpu->last_dp;
5912 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5913
d0752060
HB		 5914 return 0;
5915}
5916
10ab25cd 5917int fx_init(struct kvm_vcpu *vcpu)
d0752060 5918{
10ab25cd
JK		 5919 int err;
5920
5921 err = fpu_alloc(&vcpu->arch.guest_fpu);
5922 if (err)
5923 return err;
5924
98918833 5925 fpu_finit(&vcpu->arch.guest_fpu);
d0752060 5926
2acf923e
DC		 5927 /*
5928 * Ensure guest xcr0 is valid for loading
5929 */
5930 vcpu->arch.xcr0 = XSTATE_FP;
5931
ad312c7c 5932 vcpu->arch.cr0 |= X86_CR0_ET;
10ab25cd
JK		 5933
5934 return 0;
d0752060
HB		 5935}
5936EXPORT_SYMBOL_GPL(fx_init);
5937
98918833
SY		 5938static void fx_free(struct kvm_vcpu *vcpu)
5939{
5940 fpu_free(&vcpu->arch.guest_fpu);
5941}
5942
d0752060
HB		 5943void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5944{
2608d7a1 5945 if (vcpu->guest_fpu_loaded)
d0752060
HB		 5946 return;
5947
2acf923e
DC		 5948 /*
5949 * Restore all possible states in the guest,
5950 * and assume host would use all available bits.
5951 * Guest xcr0 would be loaded later.
5952 */
5953 kvm_put_guest_xcr0(vcpu);
d0752060 5954 vcpu->guest_fpu_loaded = 1;
7cf30855 5955 unlazy_fpu(current);
98918833 5956 fpu_restore_checking(&vcpu->arch.guest_fpu);
0c04851c 5957 trace_kvm_fpu(1);
d0752060 5958}
d0752060
HB		 5959
5960void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5961{
2acf923e
DC		 5962 kvm_put_guest_xcr0(vcpu);
5963
d0752060
HB		 5964 if (!vcpu->guest_fpu_loaded)
5965 return;
5966
5967 vcpu->guest_fpu_loaded = 0;
98918833 5968 fpu_save_init(&vcpu->arch.guest_fpu);
f096ed85 5969 ++vcpu->stat.fpu_reload;
a8eeb04a 5970 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
0c04851c 5971 trace_kvm_fpu(0);
d0752060 5972}
e9b11c17
ZX		 5973
5974void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5975{
12f9a48f 5976 kvmclock_reset(vcpu);
7f1ea208 5977
f5f48ee1 5978 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
98918833 5979 fx_free(vcpu);
e9b11c17
ZX		 5980 kvm_x86_ops->vcpu_free(vcpu);
5981}
5982
5983struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5984 unsigned int id)
5985{
6755bae8
ZA		 5986 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5987 printk_once(KERN_WARNING
5988 "kvm: SMP vm created on host with unstable TSC; "
5989 "guest TSC will not be reliable\n");
26e5215f
AK		 5990 return kvm_x86_ops->vcpu_create(kvm, id);
5991}
e9b11c17 5992
26e5215f
AK		 5993int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5994{
5995 int r;
e9b11c17 5996
0bed3b56 5997 vcpu->arch.mtrr_state.have_fixed = 1;
e9b11c17
ZX		 5998 vcpu_load(vcpu);
5999 r = kvm_arch_vcpu_reset(vcpu);
6000 if (r == 0)
6001 r = kvm_mmu_setup(vcpu);
6002 vcpu_put(vcpu);
e9b11c17 6003
26e5215f 6004 return r;
e9b11c17
ZX		 6005}
6006
d40ccc62 6007void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
e9b11c17 6008{
344d9588
GN		 6009 vcpu->arch.apf.msr_val = 0;
6010
e9b11c17
ZX		 6011 vcpu_load(vcpu);
6012 kvm_mmu_unload(vcpu);
6013 vcpu_put(vcpu);
6014
98918833 6015 fx_free(vcpu);
e9b11c17
ZX		 6016 kvm_x86_ops->vcpu_free(vcpu);
6017}
6018
6019int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6020{
7460fb4a
AK		 6021 atomic_set(&vcpu->arch.nmi_queued, 0);
6022 vcpu->arch.nmi_pending = 0;
448fa4a9
JK		 6023 vcpu->arch.nmi_injected = false;
6024
42dbaa5a
JK		 6025 vcpu->arch.switch_db_regs = 0;
6026 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6027 vcpu->arch.dr6 = DR6_FIXED_1;
6028 vcpu->arch.dr7 = DR7_FIXED_1;
6029
3842d135 6030 kvm_make_request(KVM_REQ_EVENT, vcpu);
344d9588 6031 vcpu->arch.apf.msr_val = 0;
c9aaa895 6032 vcpu->arch.st.msr_val = 0;
3842d135 6033
12f9a48f
GC		 6034 kvmclock_reset(vcpu);
6035
af585b92
GN		 6036 kvm_clear_async_pf_completion_queue(vcpu);
6037 kvm_async_pf_hash_reset(vcpu);
6038 vcpu->arch.apf.halted = false;
3842d135 6039
f5132b01
GN		 6040 kvm_pmu_reset(vcpu);
6041
e9b11c17
ZX		 6042 return kvm_x86_ops->vcpu_reset(vcpu);
6043}
6044
10474ae8 6045int kvm_arch_hardware_enable(void *garbage)
e9b11c17 6046{
ca84d1a2
ZA		 6047 struct kvm *kvm;
6048 struct kvm_vcpu *vcpu;
6049 int i;
0dd6a6ed
ZA		 6050 int ret;
6051 u64 local_tsc;
6052 u64 max_tsc = 0;
6053 bool stable, backwards_tsc = false;
18863bdd
AK		 6054
6055 kvm_shared_msr_cpu_online();
0dd6a6ed
ZA		 6056 ret = kvm_x86_ops->hardware_enable(garbage);
6057 if (ret != 0)
6058 return ret;
6059
6060 local_tsc = native_read_tsc();
6061 stable = !check_tsc_unstable();
6062 list_for_each_entry(kvm, &vm_list, vm_list) {
6063 kvm_for_each_vcpu(i, vcpu, kvm) {
6064 if (!stable && vcpu->cpu == smp_processor_id())
6065 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6066 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6067 backwards_tsc = true;
6068 if (vcpu->arch.last_host_tsc > max_tsc)
6069 max_tsc = vcpu->arch.last_host_tsc;
6070 }
6071 }
6072 }
6073
6074 /*
6075 * Sometimes, even reliable TSCs go backwards. This happens on
6076 * platforms that reset TSC during suspend or hibernate actions, but
6077 * maintain synchronization. We must compensate. Fortunately, we can
6078 * detect that condition here, which happens early in CPU bringup,
6079 * before any KVM threads can be running. Unfortunately, we can't
6080 * bring the TSCs fully up to date with real time, as we aren't yet far
6081 * enough into CPU bringup that we know how much real time has actually
6082 * elapsed; our helper function, get_kernel_ns() will be using boot
6083 * variables that haven't been updated yet.
6084 *
6085 * So we simply find the maximum observed TSC above, then record the
6086 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6087 * the adjustment will be applied. Note that we accumulate
6088 * adjustments, in case multiple suspend cycles happen before some VCPU
6089 * gets a chance to run again. In the event that no KVM threads get a
6090 * chance to run, we will miss the entire elapsed period, as we'll have
6091 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6092 * loose cycle time. This isn't too big a deal, since the loss will be
6093 * uniform across all VCPUs (not to mention the scenario is extremely
6094 * unlikely). It is possible that a second hibernate recovery happens
6095 * much faster than a first, causing the observed TSC here to be
6096 * smaller; this would require additional padding adjustment, which is
6097 * why we set last_host_tsc to the local tsc observed here.
6098 *
6099 * N.B. - this code below runs only on platforms with reliable TSC,
6100 * as that is the only way backwards_tsc is set above. Also note
6101 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6102 * have the same delta_cyc adjustment applied if backwards_tsc
6103 * is detected. Note further, this adjustment is only done once,
6104 * as we reset last_host_tsc on all VCPUs to stop this from being
6105 * called multiple times (one for each physical CPU bringup).
6106 *
6107 * Platforms with unnreliable TSCs don't have to deal with this, they
6108 * will be compensated by the logic in vcpu_load, which sets the TSC to
6109 * catchup mode. This will catchup all VCPUs to real time, but cannot
6110 * guarantee that they stay in perfect synchronization.
6111 */
6112 if (backwards_tsc) {
6113 u64 delta_cyc = max_tsc - local_tsc;
6114 list_for_each_entry(kvm, &vm_list, vm_list) {
6115 kvm_for_each_vcpu(i, vcpu, kvm) {
6116 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6117 vcpu->arch.last_host_tsc = local_tsc;
6118 }
6119
6120 /*
6121 * We have to disable TSC offset matching.. if you were
6122 * booting a VM while issuing an S4 host suspend....
6123 * you may have some problem. Solving this issue is
6124 * left as an exercise to the reader.
6125 */
6126 kvm->arch.last_tsc_nsec = 0;
6127 kvm->arch.last_tsc_write = 0;
6128 }
6129
6130 }
6131 return 0;
e9b11c17
ZX		 6132}
6133
6134void kvm_arch_hardware_disable(void *garbage)
6135{
6136 kvm_x86_ops->hardware_disable(garbage);
3548bab5 6137 drop_user_return_notifiers(garbage);
e9b11c17
ZX		 6138}
6139
6140int kvm_arch_hardware_setup(void)
6141{
6142 return kvm_x86_ops->hardware_setup();
6143}
6144
6145void kvm_arch_hardware_unsetup(void)
6146{
6147 kvm_x86_ops->hardware_unsetup();
6148}
6149
6150void kvm_arch_check_processor_compat(void *rtn)
6151{
6152 kvm_x86_ops->check_processor_compatibility(rtn);
6153}
6154
3e515705
AK		 6155bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6156{
6157 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6158}
6159
e9b11c17
ZX		 6160int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6161{
6162 struct page *page;
6163 struct kvm *kvm;
6164 int r;
6165
6166 BUG_ON(vcpu->kvm == NULL);
6167 kvm = vcpu->kvm;
6168
9aabc88f 6169 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
c5af89b6 6170 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
a4535290 6171 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
e9b11c17 6172 else
a4535290 6173 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
e9b11c17
ZX		 6174
6175 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6176 if (!page) {
6177 r = -ENOMEM;
6178 goto fail;
6179 }
ad312c7c 6180 vcpu->arch.pio_data = page_address(page);
e9b11c17 6181
cc578287 6182 kvm_set_tsc_khz(vcpu, max_tsc_khz);
c285545f 6183
e9b11c17
ZX		 6184 r = kvm_mmu_create(vcpu);
6185 if (r < 0)
6186 goto fail_free_pio_data;
6187
6188 if (irqchip_in_kernel(kvm)) {
6189 r = kvm_create_lapic(vcpu);
6190 if (r < 0)
6191 goto fail_mmu_destroy;
6192 }
6193
890ca9ae
HY		 6194 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6195 GFP_KERNEL);
6196 if (!vcpu->arch.mce_banks) {
6197 r = -ENOMEM;
443c39bc 6198 goto fail_free_lapic;
890ca9ae
HY		 6199 }
6200 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6201
f5f48ee1
SY		 6202 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6203 goto fail_free_mce_banks;
6204
af585b92 6205 kvm_async_pf_hash_reset(vcpu);
f5132b01 6206 kvm_pmu_init(vcpu);
af585b92 6207
e9b11c17 6208 return 0;
f5f48ee1
SY		 6209fail_free_mce_banks:
6210 kfree(vcpu->arch.mce_banks);
443c39bc
WY		 6211fail_free_lapic:
6212 kvm_free_lapic(vcpu);
e9b11c17
ZX		 6213fail_mmu_destroy:
6214 kvm_mmu_destroy(vcpu);
6215fail_free_pio_data:
ad312c7c 6216 free_page((unsigned long)vcpu->arch.pio_data);
e9b11c17
ZX		 6217fail:
6218 return r;
6219}
6220
6221void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6222{
f656ce01
MT		 6223 int idx;
6224
f5132b01 6225 kvm_pmu_destroy(vcpu);
36cb93fd 6226 kfree(vcpu->arch.mce_banks);
e9b11c17 6227 kvm_free_lapic(vcpu);
f656ce01 6228 idx = srcu_read_lock(&vcpu->kvm->srcu);
e9b11c17 6229 kvm_mmu_destroy(vcpu);
f656ce01 6230 srcu_read_unlock(&vcpu->kvm->srcu, idx);
ad312c7c 6231 free_page((unsigned long)vcpu->arch.pio_data);
e9b11c17 6232}
d19a9cd2 6233
e08b9637 6234int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
d19a9cd2 6235{
e08b9637
CO		 6236 if (type)
6237 return -EINVAL;
6238
f05e70ac 6239 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4d5c5d0f 6240 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
d19a9cd2 6241
5550af4d
SY		 6242 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6243 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6244
038f8c11 6245 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
53f658b3 6246
d89f5eff 6247 return 0;
d19a9cd2
ZX		 6248}
6249
6250static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6251{
6252 vcpu_load(vcpu);
6253 kvm_mmu_unload(vcpu);
6254 vcpu_put(vcpu);
6255}
6256
6257static void kvm_free_vcpus(struct kvm *kvm)
6258{
6259 unsigned int i;
988a2cae 6260 struct kvm_vcpu *vcpu;
d19a9cd2
ZX		 6261
6262 /*
6263 * Unpin any mmu pages first.
6264 */
af585b92
GN		 6265 kvm_for_each_vcpu(i, vcpu, kvm) {
6266 kvm_clear_async_pf_completion_queue(vcpu);
988a2cae 6267 kvm_unload_vcpu_mmu(vcpu);
af585b92 6268 }
988a2cae
GN		 6269 kvm_for_each_vcpu(i, vcpu, kvm)
6270 kvm_arch_vcpu_free(vcpu);
6271
6272 mutex_lock(&kvm->lock);
6273 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6274 kvm->vcpus[i] = NULL;
d19a9cd2 6275
988a2cae
GN		 6276 atomic_set(&kvm->online_vcpus, 0);
6277 mutex_unlock(&kvm->lock);
d19a9cd2
ZX		 6278}
6279
ad8ba2cd
SY		 6280void kvm_arch_sync_events(struct kvm *kvm)
6281{
ba4cef31 6282 kvm_free_all_assigned_devices(kvm);
aea924f6 6283 kvm_free_pit(kvm);
ad8ba2cd
SY		 6284}
6285
d19a9cd2
ZX		 6286void kvm_arch_destroy_vm(struct kvm *kvm)
6287{
6eb55818 6288 kvm_iommu_unmap_guest(kvm);
d7deeeb0
ZX		 6289 kfree(kvm->arch.vpic);
6290 kfree(kvm->arch.vioapic);
d19a9cd2 6291 kvm_free_vcpus(kvm);
3d45830c
AK		 6292 if (kvm->arch.apic_access_page)
6293 put_page(kvm->arch.apic_access_page);
b7ebfb05
SY		 6294 if (kvm->arch.ept_identity_pagetable)
6295 put_page(kvm->arch.ept_identity_pagetable);
d19a9cd2 6296}
0de10343 6297
db3fe4eb
TY		 6298void kvm_arch_free_memslot(struct kvm_memory_slot *free,
6299 struct kvm_memory_slot *dont)
6300{
6301 int i;
6302
6303 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
6304 if (!dont || free->arch.lpage_info[i] != dont->arch.lpage_info[i]) {
6305 vfree(free->arch.lpage_info[i]);
6306 free->arch.lpage_info[i] = NULL;
6307 }
6308 }
6309}
6310
6311int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
6312{
6313 int i;
6314
6315 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
6316 unsigned long ugfn;
6317 int lpages;
6318 int level = i + 2;
6319
6320 lpages = gfn_to_index(slot->base_gfn + npages - 1,
6321 slot->base_gfn, level) + 1;
6322
6323 slot->arch.lpage_info[i] =
6324 vzalloc(lpages * sizeof(*slot->arch.lpage_info[i]));
6325 if (!slot->arch.lpage_info[i])
6326 goto out_free;
6327
6328 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
6329 slot->arch.lpage_info[i][0].write_count = 1;
6330 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
6331 slot->arch.lpage_info[i][lpages - 1].write_count = 1;
6332 ugfn = slot->userspace_addr >> PAGE_SHIFT;
6333 /*
6334 * If the gfn and userspace address are not aligned wrt each
6335 * other, or if explicitly asked to, disable large page
6336 * support for this slot
6337 */
6338 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
6339 !kvm_largepages_enabled()) {
6340 unsigned long j;
6341
6342 for (j = 0; j < lpages; ++j)
6343 slot->arch.lpage_info[i][j].write_count = 1;
6344 }
6345 }
6346
6347 return 0;
6348
6349out_free:
6350 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
6351 vfree(slot->arch.lpage_info[i]);
6352 slot->arch.lpage_info[i] = NULL;
6353 }
6354 return -ENOMEM;
6355}
6356
f7784b8e
MT		 6357int kvm_arch_prepare_memory_region(struct kvm *kvm,
6358 struct kvm_memory_slot *memslot,
0de10343 6359 struct kvm_memory_slot old,
f7784b8e 6360 struct kvm_userspace_memory_region *mem,
0de10343
ZX		 6361 int user_alloc)
6362{
f7784b8e 6363 int npages = memslot->npages;
7ac77099
AK		 6364 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6365
6366 /* Prevent internal slot pages from being moved by fork()/COW. */
6367 if (memslot->id >= KVM_MEMORY_SLOTS)
6368 map_flags = MAP_SHARED | MAP_ANONYMOUS;
0de10343
ZX		 6369
6370 /*To keep backward compatibility with older userspace,
6371 *x86 needs to hanlde !user_alloc case.
6372 */
6373 if (!user_alloc) {
6374 if (npages && !old.rmap) {
604b38ac
AA		 6375 unsigned long userspace_addr;
6376
72dc67a6 6377 down_write(&current->mm->mmap_sem);
604b38ac
AA		 6378 userspace_addr = do_mmap(NULL, 0,
6379 npages * PAGE_SIZE,
6380 PROT_READ | PROT_WRITE,
7ac77099 6381 map_flags,
604b38ac 6382 0);
72dc67a6 6383 up_write(&current->mm->mmap_sem);
0de10343 6384
604b38ac
AA		 6385 if (IS_ERR((void *)userspace_addr))
6386 return PTR_ERR((void *)userspace_addr);
6387
604b38ac 6388 memslot->userspace_addr = userspace_addr;
0de10343
ZX		 6389 }
6390 }
6391
f7784b8e
MT		 6392
6393 return 0;
6394}
6395
6396void kvm_arch_commit_memory_region(struct kvm *kvm,
6397 struct kvm_userspace_memory_region *mem,
6398 struct kvm_memory_slot old,
6399 int user_alloc)
6400{
6401
48c0e4e9 6402 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
f7784b8e
MT		 6403
6404 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6405 int ret;
6406
6407 down_write(&current->mm->mmap_sem);
6408 ret = do_munmap(current->mm, old.userspace_addr,
6409 old.npages * PAGE_SIZE);
6410 up_write(&current->mm->mmap_sem);
6411 if (ret < 0)
6412 printk(KERN_WARNING
6413 "kvm_vm_ioctl_set_memory_region: "
6414 "failed to munmap memory\n");
6415 }
6416
48c0e4e9
XG		 6417 if (!kvm->arch.n_requested_mmu_pages)
6418 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6419
7c8a83b7 6420 spin_lock(&kvm->mmu_lock);
48c0e4e9 6421 if (nr_mmu_pages)
0de10343 6422 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
0de10343 6423 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7c8a83b7 6424 spin_unlock(&kvm->mmu_lock);
0de10343 6425}
1d737c8a 6426
34d4cb8f
MT		 6427void kvm_arch_flush_shadow(struct kvm *kvm)
6428{
6429 kvm_mmu_zap_all(kvm);
8986ecc0 6430 kvm_reload_remote_mmus(kvm);
34d4cb8f
MT		 6431}
6432
1d737c8a
ZX		 6433int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6434{
af585b92
GN		 6435 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6436 !vcpu->arch.apf.halted)
6437 || !list_empty_careful(&vcpu->async_pf.done)
a1b37100 6438 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
7460fb4a 6439 || atomic_read(&vcpu->arch.nmi_queued) ||
a1b37100
GN		 6440 (kvm_arch_interrupt_allowed(vcpu) &&
6441 kvm_cpu_has_interrupt(vcpu));
1d737c8a 6442}
5736199a 6443
b6d33834 6444int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
5736199a 6445{
b6d33834 6446 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
5736199a 6447}
78646121
GN		 6448
6449int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6450{
6451 return kvm_x86_ops->interrupt_allowed(vcpu);
6452}
229456fc 6453
f92653ee
JK		 6454bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6455{
6456 unsigned long current_rip = kvm_rip_read(vcpu) +
6457 get_segment_base(vcpu, VCPU_SREG_CS);
6458
6459 return current_rip == linear_rip;
6460}
6461EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6462
94fe45da
JK		 6463unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6464{
6465 unsigned long rflags;
6466
6467 rflags = kvm_x86_ops->get_rflags(vcpu);
6468 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
c310bac5 6469 rflags &= ~X86_EFLAGS_TF;
94fe45da
JK		 6470 return rflags;
6471}
6472EXPORT_SYMBOL_GPL(kvm_get_rflags);
6473
6474void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6475{
6476 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
f92653ee 6477 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
c310bac5 6478 rflags |= X86_EFLAGS_TF;
94fe45da 6479 kvm_x86_ops->set_rflags(vcpu, rflags);
3842d135 6480 kvm_make_request(KVM_REQ_EVENT, vcpu);
94fe45da
JK		 6481}
6482EXPORT_SYMBOL_GPL(kvm_set_rflags);
6483
56028d08
GN		 6484void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6485{
6486 int r;
6487
fb67e14f 6488 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
c4806acd 6489 is_error_page(work->page))
56028d08
GN		 6490 return;
6491
6492 r = kvm_mmu_reload(vcpu);
6493 if (unlikely(r))
6494 return;
6495
fb67e14f
XG		 6496 if (!vcpu->arch.mmu.direct_map &&
6497 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6498 return;
6499
56028d08
GN		 6500 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6501}
6502
af585b92
GN		 6503static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6504{
6505 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6506}
6507
6508static inline u32 kvm_async_pf_next_probe(u32 key)
6509{
6510 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6511}
6512
6513static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6514{
6515 u32 key = kvm_async_pf_hash_fn(gfn);
6516
6517 while (vcpu->arch.apf.gfns[key] != ~0)
6518 key = kvm_async_pf_next_probe(key);
6519
6520 vcpu->arch.apf.gfns[key] = gfn;
6521}
6522
6523static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6524{
6525 int i;
6526 u32 key = kvm_async_pf_hash_fn(gfn);
6527
6528 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
c7d28c24
XG		 6529 (vcpu->arch.apf.gfns[key] != gfn &&
6530 vcpu->arch.apf.gfns[key] != ~0); i++)
af585b92
GN		 6531 key = kvm_async_pf_next_probe(key);
6532
6533 return key;
6534}
6535
6536bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6537{
6538 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6539}
6540
6541static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6542{
6543 u32 i, j, k;
6544
6545 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6546 while (true) {
6547 vcpu->arch.apf.gfns[i] = ~0;
6548 do {
6549 j = kvm_async_pf_next_probe(j);
6550 if (vcpu->arch.apf.gfns[j] == ~0)
6551 return;
6552 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6553 /*
6554 * k lies cyclically in ]i,j]
6555 * | i.k.j |
6556 * |....j i.k.| or |.k..j i...|
6557 */
6558 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6559 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6560 i = j;
6561 }
6562}
6563
7c90705b
GN		 6564static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6565{
6566
6567 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6568 sizeof(val));
6569}
6570
af585b92
GN		 6571void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6572 struct kvm_async_pf *work)
6573{
6389ee94
AK		 6574 struct x86_exception fault;
6575
7c90705b 6576 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
af585b92 6577 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7c90705b
GN		 6578
6579 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
fc5f06fa
GN		 6580 (vcpu->arch.apf.send_user_only &&
6581 kvm_x86_ops->get_cpl(vcpu) == 0))
7c90705b
GN		 6582 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6583 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6389ee94
AK		 6584 fault.vector = PF_VECTOR;
6585 fault.error_code_valid = true;
6586 fault.error_code = 0;
6587 fault.nested_page_fault = false;
6588 fault.address = work->arch.token;
6589 kvm_inject_page_fault(vcpu, &fault);
7c90705b 6590 }
af585b92
GN		 6591}
6592
6593void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6594 struct kvm_async_pf *work)
6595{
6389ee94
AK		 6596 struct x86_exception fault;
6597
7c90705b
GN		 6598 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6599 if (is_error_page(work->page))
6600 work->arch.token = ~0; /* broadcast wakeup */
6601 else
6602 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6603
6604 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6605 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6389ee94
AK		 6606 fault.vector = PF_VECTOR;
6607 fault.error_code_valid = true;
6608 fault.error_code = 0;
6609 fault.nested_page_fault = false;
6610 fault.address = work->arch.token;
6611 kvm_inject_page_fault(vcpu, &fault);
7c90705b 6612 }
e6d53e3b 6613 vcpu->arch.apf.halted = false;
7c90705b
GN		 6614}
6615
6616bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6617{
6618 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6619 return true;
6620 else
6621 return !kvm_event_needs_reinjection(vcpu) &&
6622 kvm_x86_ops->interrupt_allowed(vcpu);
af585b92
GN		 6623}
6624
229456fc
MT		 6625EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6626EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6627EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6628EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6629EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
0ac406de 6630EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
d8cabddf 6631EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
17897f36 6632EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
236649de 6633EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
ec1ff790 6634EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
532a46b9 6635EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
2e554e8d 6636EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
kernel source for telekom puls (alcatel/mediatek)