projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: x86 emulator: add new ->wbinvd() callback
[deliverable/linux.git] / arch / x86 / kvm / x86.c
CommitLineData
043405e1
CO		 1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * derived from drivers/kvm/kvm_main.c
5 *
6 * Copyright (C) 2006 Qumranet, Inc.
4d5c5d0f
BAY		 7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9611c187 9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
043405e1
CO		 10 *
11 * Authors:
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
4d5c5d0f
BAY		 14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
043405e1
CO		 16 *
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
19 *
20 */
21
edf88417 22#include <linux/kvm_host.h>
313a3dc7 23#include "irq.h"
1d737c8a 24#include "mmu.h"
7837699f 25#include "i8254.h"
37817f29 26#include "tss.h"
5fdbf976 27#include "kvm_cache_regs.h"
26eef70c 28#include "x86.h"
313a3dc7 29
18068523 30#include <linux/clocksource.h>
4d5c5d0f 31#include <linux/interrupt.h>
313a3dc7
CO		 32#include <linux/kvm.h>
33#include <linux/fs.h>
34#include <linux/vmalloc.h>
5fb76f9b 35#include <linux/module.h>
0de10343 36#include <linux/mman.h>
2bacc55c 37#include <linux/highmem.h>
19de40a8 38#include <linux/iommu.h>
62c476c7 39#include <linux/intel-iommu.h>
c8076604 40#include <linux/cpufreq.h>
18863bdd 41#include <linux/user-return-notifier.h>
a983fb23 42#include <linux/srcu.h>
5a0e3ad6 43#include <linux/slab.h>
ff9d07a0 44#include <linux/perf_event.h>
7bee342a 45#include <linux/uaccess.h>
af585b92 46#include <linux/hash.h>
aec51dc4 47#include <trace/events/kvm.h>
2ed152af 48
229456fc
MT		 49#define CREATE_TRACE_POINTS
50#include "trace.h"
043405e1 51
24f1e32c 52#include <asm/debugreg.h>
d825ed0a 53#include <asm/msr.h>
a5f61300 54#include <asm/desc.h>
0bed3b56 55#include <asm/mtrr.h>
890ca9ae 56#include <asm/mce.h>
7cf30855 57#include <asm/i387.h>
98918833 58#include <asm/xcr.h>
1d5f066e 59#include <asm/pvclock.h>
217fc9cf 60#include <asm/div64.h>
043405e1 61
313a3dc7 62#define MAX_IO_MSRS 256
890ca9ae 63#define KVM_MAX_MCE_BANKS 32
5854dbca 64#define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
890ca9ae 65
0f65dd70
AK		 66#define emul_to_vcpu(ctxt) \
67 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
68
50a37eb4
JR		 69/* EFER defaults:
70 * - enable syscall per default because its emulated by KVM
71 * - enable LME and LMA per default on 64 bit KVM
72 */
73#ifdef CONFIG_X86_64
1260edbe
LJ		 74static
75u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
50a37eb4 76#else
1260edbe 77static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
50a37eb4 78#endif
313a3dc7 79
ba1389b7
AK		 80#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
81#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
417bc304 82
cb142eb7 83static void update_cr8_intercept(struct kvm_vcpu *vcpu);
674eea0f
AK		 84static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
85 struct kvm_cpuid_entry2 __user *entries);
86
97896d04 87struct kvm_x86_ops *kvm_x86_ops;
5fdbf976 88EXPORT_SYMBOL_GPL(kvm_x86_ops);
97896d04 89
ed85c068
AP		 90int ignore_msrs = 0;
91module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
92
92a1f12d
JR		 93bool kvm_has_tsc_control;
94EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
95u32 kvm_max_guest_tsc_khz;
96EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
97
18863bdd
AK		 98#define KVM_NR_SHARED_MSRS 16
99
100struct kvm_shared_msrs_global {
101 int nr;
2bf78fa7 102 u32 msrs[KVM_NR_SHARED_MSRS];
18863bdd
AK		 103};
104
105struct kvm_shared_msrs {
106 struct user_return_notifier urn;
107 bool registered;
2bf78fa7
SY		 108 struct kvm_shared_msr_values {
109 u64 host;
110 u64 curr;
111 } values[KVM_NR_SHARED_MSRS];
18863bdd
AK		 112};
113
114static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
115static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
116
417bc304 117struct kvm_stats_debugfs_item debugfs_entries[] = {
ba1389b7
AK		 118 { "pf_fixed", VCPU_STAT(pf_fixed) },
119 { "pf_guest", VCPU_STAT(pf_guest) },
120 { "tlb_flush", VCPU_STAT(tlb_flush) },
121 { "invlpg", VCPU_STAT(invlpg) },
122 { "exits", VCPU_STAT(exits) },
123 { "io_exits", VCPU_STAT(io_exits) },
124 { "mmio_exits", VCPU_STAT(mmio_exits) },
125 { "signal_exits", VCPU_STAT(signal_exits) },
126 { "irq_window", VCPU_STAT(irq_window_exits) },
f08864b4 127 { "nmi_window", VCPU_STAT(nmi_window_exits) },
ba1389b7
AK		 128 { "halt_exits", VCPU_STAT(halt_exits) },
129 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
f11c3a8d 130 { "hypercalls", VCPU_STAT(hypercalls) },
ba1389b7
AK		 131 { "request_irq", VCPU_STAT(request_irq_exits) },
132 { "irq_exits", VCPU_STAT(irq_exits) },
133 { "host_state_reload", VCPU_STAT(host_state_reload) },
134 { "efer_reload", VCPU_STAT(efer_reload) },
135 { "fpu_reload", VCPU_STAT(fpu_reload) },
136 { "insn_emulation", VCPU_STAT(insn_emulation) },
137 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
fa89a817 138 { "irq_injections", VCPU_STAT(irq_injections) },
c4abb7c9 139 { "nmi_injections", VCPU_STAT(nmi_injections) },
4cee5764
AK		 140 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
141 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
142 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
143 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
144 { "mmu_flooded", VM_STAT(mmu_flooded) },
145 { "mmu_recycled", VM_STAT(mmu_recycled) },
dfc5aa00 146 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
4731d4c7 147 { "mmu_unsync", VM_STAT(mmu_unsync) },
0f74a24c 148 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
05da4558 149 { "largepages", VM_STAT(lpages) },
417bc304
HB		 150 { NULL }
151};
152
2acf923e
DC		 153u64 __read_mostly host_xcr0;
154
d6aa1000
AK		 155int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
156
af585b92
GN		 157static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
158{
159 int i;
160 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
161 vcpu->arch.apf.gfns[i] = ~0;
162}
163
18863bdd
AK		 164static void kvm_on_user_return(struct user_return_notifier *urn)
165{
166 unsigned slot;
18863bdd
AK		 167 struct kvm_shared_msrs *locals
168 = container_of(urn, struct kvm_shared_msrs, urn);
2bf78fa7 169 struct kvm_shared_msr_values *values;
18863bdd
AK		 170
171 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
2bf78fa7
SY		 172 values = &locals->values[slot];
173 if (values->host != values->curr) {
174 wrmsrl(shared_msrs_global.msrs[slot], values->host);
175 values->curr = values->host;
18863bdd
AK		 176 }
177 }
178 locals->registered = false;
179 user_return_notifier_unregister(urn);
180}
181
2bf78fa7 182static void shared_msr_update(unsigned slot, u32 msr)
18863bdd 183{
2bf78fa7 184 struct kvm_shared_msrs *smsr;
18863bdd
AK		 185 u64 value;
186
2bf78fa7
SY		 187 smsr = &__get_cpu_var(shared_msrs);
188 /* only read, and nobody should modify it at this time,
189 * so don't need lock */
190 if (slot >= shared_msrs_global.nr) {
191 printk(KERN_ERR "kvm: invalid MSR slot!");
192 return;
193 }
194 rdmsrl_safe(msr, &value);
195 smsr->values[slot].host = value;
196 smsr->values[slot].curr = value;
197}
198
199void kvm_define_shared_msr(unsigned slot, u32 msr)
200{
18863bdd
AK		 201 if (slot >= shared_msrs_global.nr)
202 shared_msrs_global.nr = slot + 1;
2bf78fa7
SY		 203 shared_msrs_global.msrs[slot] = msr;
204 /* we need ensured the shared_msr_global have been updated */
205 smp_wmb();
18863bdd
AK		 206}
207EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
208
209static void kvm_shared_msr_cpu_online(void)
210{
211 unsigned i;
18863bdd
AK		 212
213 for (i = 0; i < shared_msrs_global.nr; ++i)
2bf78fa7 214 shared_msr_update(i, shared_msrs_global.msrs[i]);
18863bdd
AK		 215}
216
d5696725 217void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
18863bdd
AK		 218{
219 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
220
2bf78fa7 221 if (((value ^ smsr->values[slot].curr) & mask) == 0)
18863bdd 222 return;
2bf78fa7
SY		 223 smsr->values[slot].curr = value;
224 wrmsrl(shared_msrs_global.msrs[slot], value);
18863bdd
AK		 225 if (!smsr->registered) {
226 smsr->urn.on_user_return = kvm_on_user_return;
227 user_return_notifier_register(&smsr->urn);
228 smsr->registered = true;
229 }
230}
231EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
232
3548bab5
AK		 233static void drop_user_return_notifiers(void *ignore)
234{
235 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
236
237 if (smsr->registered)
238 kvm_on_user_return(&smsr->urn);
239}
240
6866b83e
CO		 241u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
242{
243 if (irqchip_in_kernel(vcpu->kvm))
ad312c7c 244 return vcpu->arch.apic_base;
6866b83e 245 else
ad312c7c 246 return vcpu->arch.apic_base;
6866b83e
CO		 247}
248EXPORT_SYMBOL_GPL(kvm_get_apic_base);
249
250void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
251{
252 /* TODO: reserve bits check */
253 if (irqchip_in_kernel(vcpu->kvm))
254 kvm_lapic_set_base(vcpu, data);
255 else
ad312c7c 256 vcpu->arch.apic_base = data;
6866b83e
CO		 257}
258EXPORT_SYMBOL_GPL(kvm_set_apic_base);
259
3fd28fce
ED		 260#define EXCPT_BENIGN 0
261#define EXCPT_CONTRIBUTORY 1
262#define EXCPT_PF 2
263
264static int exception_class(int vector)
265{
266 switch (vector) {
267 case PF_VECTOR:
268 return EXCPT_PF;
269 case DE_VECTOR:
270 case TS_VECTOR:
271 case NP_VECTOR:
272 case SS_VECTOR:
273 case GP_VECTOR:
274 return EXCPT_CONTRIBUTORY;
275 default:
276 break;
277 }
278 return EXCPT_BENIGN;
279}
280
281static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
ce7ddec4
JR		 282 unsigned nr, bool has_error, u32 error_code,
283 bool reinject)
3fd28fce
ED		 284{
285 u32 prev_nr;
286 int class1, class2;
287
3842d135
AK		 288 kvm_make_request(KVM_REQ_EVENT, vcpu);
289
3fd28fce
ED		 290 if (!vcpu->arch.exception.pending) {
291 queue:
292 vcpu->arch.exception.pending = true;
293 vcpu->arch.exception.has_error_code = has_error;
294 vcpu->arch.exception.nr = nr;
295 vcpu->arch.exception.error_code = error_code;
3f0fd292 296 vcpu->arch.exception.reinject = reinject;
3fd28fce
ED		 297 return;
298 }
299
300 /* to check exception */
301 prev_nr = vcpu->arch.exception.nr;
302 if (prev_nr == DF_VECTOR) {
303 /* triple fault -> shutdown */
a8eeb04a 304 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3fd28fce
ED		 305 return;
306 }
307 class1 = exception_class(prev_nr);
308 class2 = exception_class(nr);
309 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
310 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
311 /* generate double fault per SDM Table 5-5 */
312 vcpu->arch.exception.pending = true;
313 vcpu->arch.exception.has_error_code = true;
314 vcpu->arch.exception.nr = DF_VECTOR;
315 vcpu->arch.exception.error_code = 0;
316 } else
317 /* replace previous exception with a new one in a hope
318 that instruction re-execution will regenerate lost
319 exception */
320 goto queue;
321}
322
298101da
AK		 323void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
324{
ce7ddec4 325 kvm_multiple_exception(vcpu, nr, false, 0, false);
298101da
AK		 326}
327EXPORT_SYMBOL_GPL(kvm_queue_exception);
328
ce7ddec4
JR		 329void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
330{
331 kvm_multiple_exception(vcpu, nr, false, 0, true);
332}
333EXPORT_SYMBOL_GPL(kvm_requeue_exception);
334
db8fcefa 335void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
c3c91fee 336{
db8fcefa
AP		 337 if (err)
338 kvm_inject_gp(vcpu, 0);
339 else
340 kvm_x86_ops->skip_emulated_instruction(vcpu);
341}
342EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
8df25a32 343
6389ee94 344void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
c3c91fee
AK		 345{
346 ++vcpu->stat.pf_guest;
6389ee94
AK		 347 vcpu->arch.cr2 = fault->address;
348 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
c3c91fee
AK		 349}
350
6389ee94 351void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
d4f8cf66 352{
6389ee94
AK		 353 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
354 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
d4f8cf66 355 else
6389ee94 356 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
d4f8cf66
JR		 357}
358
3419ffc8
SY		 359void kvm_inject_nmi(struct kvm_vcpu *vcpu)
360{
3842d135 361 kvm_make_request(KVM_REQ_EVENT, vcpu);
c761e586 362 vcpu->arch.nmi_pending = 1;
3419ffc8
SY		 363}
364EXPORT_SYMBOL_GPL(kvm_inject_nmi);
365
298101da
AK		 366void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
367{
ce7ddec4 368 kvm_multiple_exception(vcpu, nr, true, error_code, false);
298101da
AK		 369}
370EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
371
ce7ddec4
JR		 372void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
373{
374 kvm_multiple_exception(vcpu, nr, true, error_code, true);
375}
376EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
377
0a79b009
AK		 378/*
379 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
380 * a #GP and return false.
381 */
382bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
298101da 383{
0a79b009
AK		 384 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
385 return true;
386 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
387 return false;
298101da 388}
0a79b009 389EXPORT_SYMBOL_GPL(kvm_require_cpl);
298101da 390
ec92fe44
JR		 391/*
392 * This function will be used to read from the physical memory of the currently
393 * running guest. The difference to kvm_read_guest_page is that this function
394 * can read from guest physical or from the guest's guest physical memory.
395 */
396int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
397 gfn_t ngfn, void *data, int offset, int len,
398 u32 access)
399{
400 gfn_t real_gfn;
401 gpa_t ngpa;
402
403 ngpa = gfn_to_gpa(ngfn);
404 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
405 if (real_gfn == UNMAPPED_GVA)
406 return -EFAULT;
407
408 real_gfn = gpa_to_gfn(real_gfn);
409
410 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
411}
412EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
413
3d06b8bf
JR		 414int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
415 void *data, int offset, int len, u32 access)
416{
417 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
418 data, offset, len, access);
419}
420
a03490ed
CO		 421/*
422 * Load the pae pdptrs. Return true is they are all valid.
423 */
ff03a073 424int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
a03490ed
CO		 425{
426 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
427 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
428 int i;
429 int ret;
ff03a073 430 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
a03490ed 431
ff03a073
JR		 432 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
433 offset * sizeof(u64), sizeof(pdpte),
434 PFERR_USER_MASK|PFERR_WRITE_MASK);
a03490ed
CO		 435 if (ret < 0) {
436 ret = 0;
437 goto out;
438 }
439 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
43a3795a 440 if (is_present_gpte(pdpte[i]) &&
20c466b5 441 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
a03490ed
CO		 442 ret = 0;
443 goto out;
444 }
445 }
446 ret = 1;
447
ff03a073 448 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
6de4f3ad
AK		 449 __set_bit(VCPU_EXREG_PDPTR,
450 (unsigned long *)&vcpu->arch.regs_avail);
451 __set_bit(VCPU_EXREG_PDPTR,
452 (unsigned long *)&vcpu->arch.regs_dirty);
a03490ed 453out:
a03490ed
CO		 454
455 return ret;
456}
cc4b6871 457EXPORT_SYMBOL_GPL(load_pdptrs);
a03490ed 458
d835dfec
AK		 459static bool pdptrs_changed(struct kvm_vcpu *vcpu)
460{
ff03a073 461 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
d835dfec 462 bool changed = true;
3d06b8bf
JR		 463 int offset;
464 gfn_t gfn;
d835dfec
AK		 465 int r;
466
467 if (is_long_mode(vcpu) || !is_pae(vcpu))
468 return false;
469
6de4f3ad
AK		 470 if (!test_bit(VCPU_EXREG_PDPTR,
471 (unsigned long *)&vcpu->arch.regs_avail))
472 return true;
473
9f8fe504
AK		 474 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
475 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
3d06b8bf
JR		 476 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
477 PFERR_USER_MASK | PFERR_WRITE_MASK);
d835dfec
AK		 478 if (r < 0)
479 goto out;
ff03a073 480 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
d835dfec 481out:
d835dfec
AK		 482
483 return changed;
484}
485
49a9b07e 486int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
a03490ed 487{
aad82703
SY		 488 unsigned long old_cr0 = kvm_read_cr0(vcpu);
489 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
490 X86_CR0_CD | X86_CR0_NW;
491
f9a48e6a
AK		 492 cr0 |= X86_CR0_ET;
493
ab344828 494#ifdef CONFIG_X86_64
0f12244f
GN		 495 if (cr0 & 0xffffffff00000000UL)
496 return 1;
ab344828
GN		 497#endif
498
499 cr0 &= ~CR0_RESERVED_BITS;
a03490ed 500
0f12244f
GN		 501 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
502 return 1;
a03490ed 503
0f12244f
GN		 504 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
505 return 1;
a03490ed
CO		 506
507 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
508#ifdef CONFIG_X86_64
f6801dff 509 if ((vcpu->arch.efer & EFER_LME)) {
a03490ed
CO		 510 int cs_db, cs_l;
511
0f12244f
GN		 512 if (!is_pae(vcpu))
513 return 1;
a03490ed 514 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
0f12244f
GN		 515 if (cs_l)
516 return 1;
a03490ed
CO		 517 } else
518#endif
ff03a073 519 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
9f8fe504 520 kvm_read_cr3(vcpu)))
0f12244f 521 return 1;
a03490ed
CO		 522 }
523
524 kvm_x86_ops->set_cr0(vcpu, cr0);
a03490ed 525
d170c419 526 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
e5f3f027 527 kvm_clear_async_pf_completion_queue(vcpu);
d170c419
LJ		 528 kvm_async_pf_hash_reset(vcpu);
529 }
e5f3f027 530
aad82703
SY		 531 if ((cr0 ^ old_cr0) & update_bits)
532 kvm_mmu_reset_context(vcpu);
0f12244f
GN		 533 return 0;
534}
2d3ad1f4 535EXPORT_SYMBOL_GPL(kvm_set_cr0);
a03490ed 536
2d3ad1f4 537void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
a03490ed 538{
49a9b07e 539 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
a03490ed 540}
2d3ad1f4 541EXPORT_SYMBOL_GPL(kvm_lmsw);
a03490ed 542
2acf923e
DC		 543int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
544{
545 u64 xcr0;
546
547 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
548 if (index != XCR_XFEATURE_ENABLED_MASK)
549 return 1;
550 xcr0 = xcr;
551 if (kvm_x86_ops->get_cpl(vcpu) != 0)
552 return 1;
553 if (!(xcr0 & XSTATE_FP))
554 return 1;
555 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
556 return 1;
557 if (xcr0 & ~host_xcr0)
558 return 1;
559 vcpu->arch.xcr0 = xcr0;
560 vcpu->guest_xcr0_loaded = 0;
561 return 0;
562}
563
564int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
565{
566 if (__kvm_set_xcr(vcpu, index, xcr)) {
567 kvm_inject_gp(vcpu, 0);
568 return 1;
569 }
570 return 0;
571}
572EXPORT_SYMBOL_GPL(kvm_set_xcr);
573
574static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
575{
576 struct kvm_cpuid_entry2 *best;
577
578 best = kvm_find_cpuid_entry(vcpu, 1, 0);
579 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
580}
581
582static void update_cpuid(struct kvm_vcpu *vcpu)
583{
584 struct kvm_cpuid_entry2 *best;
585
586 best = kvm_find_cpuid_entry(vcpu, 1, 0);
587 if (!best)
588 return;
589
590 /* Update OSXSAVE bit */
591 if (cpu_has_xsave && best->function == 0x1) {
592 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
593 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
594 best->ecx |= bit(X86_FEATURE_OSXSAVE);
595 }
596}
597
a83b29c6 598int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
a03490ed 599{
fc78f519 600 unsigned long old_cr4 = kvm_read_cr4(vcpu);
a2edf57f
AK		 601 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
602
0f12244f
GN		 603 if (cr4 & CR4_RESERVED_BITS)
604 return 1;
a03490ed 605
2acf923e
DC		 606 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
607 return 1;
608
a03490ed 609 if (is_long_mode(vcpu)) {
0f12244f
GN		 610 if (!(cr4 & X86_CR4_PAE))
611 return 1;
a2edf57f
AK		 612 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
613 && ((cr4 ^ old_cr4) & pdptr_bits)
9f8fe504
AK		 614 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
615 kvm_read_cr3(vcpu)))
0f12244f
GN		 616 return 1;
617
618 if (cr4 & X86_CR4_VMXE)
619 return 1;
a03490ed 620
a03490ed 621 kvm_x86_ops->set_cr4(vcpu, cr4);
62ad0755 622
aad82703
SY		 623 if ((cr4 ^ old_cr4) & pdptr_bits)
624 kvm_mmu_reset_context(vcpu);
0f12244f 625
2acf923e
DC		 626 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
627 update_cpuid(vcpu);
628
0f12244f
GN		 629 return 0;
630}
2d3ad1f4 631EXPORT_SYMBOL_GPL(kvm_set_cr4);
a03490ed 632
2390218b 633int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
a03490ed 634{
9f8fe504 635 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
0ba73cda 636 kvm_mmu_sync_roots(vcpu);
d835dfec 637 kvm_mmu_flush_tlb(vcpu);
0f12244f 638 return 0;
d835dfec
AK		 639 }
640
a03490ed 641 if (is_long_mode(vcpu)) {
0f12244f
GN		 642 if (cr3 & CR3_L_MODE_RESERVED_BITS)
643 return 1;
a03490ed
CO		 644 } else {
645 if (is_pae(vcpu)) {
0f12244f
GN		 646 if (cr3 & CR3_PAE_RESERVED_BITS)
647 return 1;
ff03a073
JR		 648 if (is_paging(vcpu) &&
649 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
0f12244f 650 return 1;
a03490ed
CO		 651 }
652 /*
653 * We don't check reserved bits in nonpae mode, because
654 * this isn't enforced, and VMware depends on this.
655 */
656 }
657
a03490ed
CO		 658 /*
659 * Does the new cr3 value map to physical memory? (Note, we
660 * catch an invalid cr3 even in real-mode, because it would
661 * cause trouble later on when we turn on paging anyway.)
662 *
663 * A real CPU would silently accept an invalid cr3 and would
664 * attempt to use it - with largely undefined (and often hard
665 * to debug) behavior on the guest side.
666 */
667 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
0f12244f
GN		 668 return 1;
669 vcpu->arch.cr3 = cr3;
aff48baa 670 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
0f12244f
GN		 671 vcpu->arch.mmu.new_cr3(vcpu);
672 return 0;
673}
2d3ad1f4 674EXPORT_SYMBOL_GPL(kvm_set_cr3);
a03490ed 675
eea1cff9 676int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
a03490ed 677{
0f12244f
GN		 678 if (cr8 & CR8_RESERVED_BITS)
679 return 1;
a03490ed
CO		 680 if (irqchip_in_kernel(vcpu->kvm))
681 kvm_lapic_set_tpr(vcpu, cr8);
682 else
ad312c7c 683 vcpu->arch.cr8 = cr8;
0f12244f
GN		 684 return 0;
685}
2d3ad1f4 686EXPORT_SYMBOL_GPL(kvm_set_cr8);
a03490ed 687
2d3ad1f4 688unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
a03490ed
CO		 689{
690 if (irqchip_in_kernel(vcpu->kvm))
691 return kvm_lapic_get_cr8(vcpu);
692 else
ad312c7c 693 return vcpu->arch.cr8;
a03490ed 694}
2d3ad1f4 695EXPORT_SYMBOL_GPL(kvm_get_cr8);
a03490ed 696
338dbc97 697static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
020df079
GN		 698{
699 switch (dr) {
700 case 0 ... 3:
701 vcpu->arch.db[dr] = val;
702 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
703 vcpu->arch.eff_db[dr] = val;
704 break;
705 case 4:
338dbc97
GN		 706 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
707 return 1; /* #UD */
020df079
GN		 708 /* fall through */
709 case 6:
338dbc97
GN		 710 if (val & 0xffffffff00000000ULL)
711 return -1; /* #GP */
020df079
GN		 712 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
713 break;
714 case 5:
338dbc97
GN		 715 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
716 return 1; /* #UD */
020df079
GN		 717 /* fall through */
718 default: /* 7 */
338dbc97
GN		 719 if (val & 0xffffffff00000000ULL)
720 return -1; /* #GP */
020df079
GN		 721 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
722 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
723 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
724 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
725 }
726 break;
727 }
728
729 return 0;
730}
338dbc97
GN		 731
732int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
733{
734 int res;
735
736 res = __kvm_set_dr(vcpu, dr, val);
737 if (res > 0)
738 kvm_queue_exception(vcpu, UD_VECTOR);
739 else if (res < 0)
740 kvm_inject_gp(vcpu, 0);
741
742 return res;
743}
020df079
GN		 744EXPORT_SYMBOL_GPL(kvm_set_dr);
745
338dbc97 746static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
020df079
GN		 747{
748 switch (dr) {
749 case 0 ... 3:
750 *val = vcpu->arch.db[dr];
751 break;
752 case 4:
338dbc97 753 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
020df079 754 return 1;
020df079
GN		 755 /* fall through */
756 case 6:
757 *val = vcpu->arch.dr6;
758 break;
759 case 5:
338dbc97 760 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
020df079 761 return 1;
020df079
GN		 762 /* fall through */
763 default: /* 7 */
764 *val = vcpu->arch.dr7;
765 break;
766 }
767
768 return 0;
769}
338dbc97
GN		 770
771int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
772{
773 if (_kvm_get_dr(vcpu, dr, val)) {
774 kvm_queue_exception(vcpu, UD_VECTOR);
775 return 1;
776 }
777 return 0;
778}
020df079
GN		 779EXPORT_SYMBOL_GPL(kvm_get_dr);
780
043405e1
CO		 781/*
782 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
783 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
784 *
785 * This list is modified at module load time to reflect the
e3267cbb
GC		 786 * capabilities of the host cpu. This capabilities test skips MSRs that are
787 * kvm-specific. Those are put in the beginning of the list.
043405e1 788 */
e3267cbb 789
344d9588 790#define KVM_SAVE_MSRS_BEGIN 8
043405e1 791static u32 msrs_to_save[] = {
e3267cbb 792 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
11c6bffa 793 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
55cd8e5a 794 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
344d9588 795 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
043405e1 796 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
8c06585d 797 MSR_STAR,
043405e1
CO		 798#ifdef CONFIG_X86_64
799 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
800#endif
e90aa41e 801 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
043405e1
CO		 802};
803
804static unsigned num_msrs_to_save;
805
806static u32 emulated_msrs[] = {
807 MSR_IA32_MISC_ENABLE,
908e75f3
AK		 808 MSR_IA32_MCG_STATUS,
809 MSR_IA32_MCG_CTL,
043405e1
CO		 810};
811
b69e8cae 812static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
15c4a640 813{
aad82703
SY		 814 u64 old_efer = vcpu->arch.efer;
815
b69e8cae
RJ		 816 if (efer & efer_reserved_bits)
817 return 1;
15c4a640
CO		 818
819 if (is_paging(vcpu)
b69e8cae
RJ		 820 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
821 return 1;
15c4a640 822
1b2fd70c
AG		 823 if (efer & EFER_FFXSR) {
824 struct kvm_cpuid_entry2 *feat;
825
826 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
b69e8cae
RJ		 827 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
828 return 1;
1b2fd70c
AG		 829 }
830
d8017474
AG		 831 if (efer & EFER_SVME) {
832 struct kvm_cpuid_entry2 *feat;
833
834 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
b69e8cae
RJ		 835 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
836 return 1;
d8017474
AG		 837 }
838
15c4a640 839 efer &= ~EFER_LMA;
f6801dff 840 efer |= vcpu->arch.efer & EFER_LMA;
15c4a640 841
a3d204e2
SY		 842 kvm_x86_ops->set_efer(vcpu, efer);
843
9645bb56 844 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
b69e8cae 845
aad82703
SY		 846 /* Update reserved bits */
847 if ((efer ^ old_efer) & EFER_NX)
848 kvm_mmu_reset_context(vcpu);
849
b69e8cae 850 return 0;
15c4a640
CO		 851}
852
f2b4b7dd
JR		 853void kvm_enable_efer_bits(u64 mask)
854{
855 efer_reserved_bits &= ~mask;
856}
857EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
858
859
15c4a640
CO		 860/*
861 * Writes msr value into into the appropriate "register".
862 * Returns 0 on success, non-0 otherwise.
863 * Assumes vcpu_load() was already called.
864 */
865int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
866{
867 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
868}
869
313a3dc7
CO		 870/*
871 * Adapt set_msr() to msr_io()'s calling convention
872 */
873static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
874{
875 return kvm_set_msr(vcpu, index, *data);
876}
877
18068523
GOC		 878static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
879{
9ed3c444
AK		 880 int version;
881 int r;
50d0a0f9 882 struct pvclock_wall_clock wc;
923de3cf 883 struct timespec boot;
18068523
GOC		 884
885 if (!wall_clock)
886 return;
887
9ed3c444
AK		 888 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
889 if (r)
890 return;
891
892 if (version & 1)
893 ++version; /* first time write, random junk */
894
895 ++version;
18068523 896
18068523
GOC		 897 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
898
50d0a0f9
GH		 899 /*
900 * The guest calculates current wall clock time by adding
34c238a1 901 * system time (updated by kvm_guest_time_update below) to the
50d0a0f9
GH		 902 * wall clock specified here. guest system time equals host
903 * system time for us, thus we must fill in host boot time here.
904 */
923de3cf 905 getboottime(&boot);
50d0a0f9
GH		 906
907 wc.sec = boot.tv_sec;
908 wc.nsec = boot.tv_nsec;
909 wc.version = version;
18068523
GOC		 910
911 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
912
913 version++;
914 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
18068523
GOC		 915}
916
50d0a0f9
GH		 917static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
918{
919 uint32_t quotient, remainder;
920
921 /* Don't try to replace with do_div(), this one calculates
922 * "(dividend << 32) / divisor" */
923 __asm__ ( "divl %4"
924 : "=a" (quotient), "=d" (remainder)
925 : "0" (0), "1" (dividend), "r" (divisor) );
926 return quotient;
927}
928
5f4e3f88
ZA		 929static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
930 s8 *pshift, u32 *pmultiplier)
50d0a0f9 931{
5f4e3f88 932 uint64_t scaled64;
50d0a0f9
GH		 933 int32_t shift = 0;
934 uint64_t tps64;
935 uint32_t tps32;
936
5f4e3f88
ZA		 937 tps64 = base_khz * 1000LL;
938 scaled64 = scaled_khz * 1000LL;
50933623 939 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
50d0a0f9
GH		 940 tps64 >>= 1;
941 shift--;
942 }
943
944 tps32 = (uint32_t)tps64;
50933623
JK		 945 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
946 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
5f4e3f88
ZA		 947 scaled64 >>= 1;
948 else
949 tps32 <<= 1;
50d0a0f9
GH		 950 shift++;
951 }
952
5f4e3f88
ZA		 953 *pshift = shift;
954 *pmultiplier = div_frac(scaled64, tps32);
50d0a0f9 955
5f4e3f88
ZA		 956 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
957 __func__, base_khz, scaled_khz, shift, *pmultiplier);
50d0a0f9
GH		 958}
959
759379dd
ZA		 960static inline u64 get_kernel_ns(void)
961{
962 struct timespec ts;
963
964 WARN_ON(preemptible());
965 ktime_get_ts(&ts);
966 monotonic_to_bootbased(&ts);
967 return timespec_to_ns(&ts);
50d0a0f9
GH		 968}
969
c8076604 970static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
c285545f 971unsigned long max_tsc_khz;
c8076604 972
8cfdc000
ZA		 973static inline int kvm_tsc_changes_freq(void)
974{
975 int cpu = get_cpu();
976 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
977 cpufreq_quick_get(cpu) != 0;
978 put_cpu();
979 return ret;
980}
981
1e993611
JR		 982static u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
983{
984 if (vcpu->arch.virtual_tsc_khz)
985 return vcpu->arch.virtual_tsc_khz;
986 else
987 return __this_cpu_read(cpu_tsc_khz);
988}
989
857e4099 990static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
759379dd 991{
217fc9cf
AK		 992 u64 ret;
993
759379dd
ZA		 994 WARN_ON(preemptible());
995 if (kvm_tsc_changes_freq())
996 printk_once(KERN_WARNING
997 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
857e4099 998 ret = nsec * vcpu_tsc_khz(vcpu);
217fc9cf
AK		 999 do_div(ret, USEC_PER_SEC);
1000 return ret;
759379dd
ZA		 1001}
1002
1e993611 1003static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
c285545f
ZA		 1004{
1005 /* Compute a scale to convert nanoseconds in TSC cycles */
1006 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1e993611
JR		 1007 &vcpu->arch.tsc_catchup_shift,
1008 &vcpu->arch.tsc_catchup_mult);
c285545f
ZA		 1009}
1010
1011static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1012{
1013 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1e993611
JR		 1014 vcpu->arch.tsc_catchup_mult,
1015 vcpu->arch.tsc_catchup_shift);
c285545f
ZA		 1016 tsc += vcpu->arch.last_tsc_write;
1017 return tsc;
1018}
1019
99e3e30a
ZA		 1020void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1021{
1022 struct kvm *kvm = vcpu->kvm;
f38e098f 1023 u64 offset, ns, elapsed;
99e3e30a 1024 unsigned long flags;
46543ba4 1025 s64 sdiff;
99e3e30a 1026
038f8c11 1027 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
857e4099 1028 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
759379dd 1029 ns = get_kernel_ns();
f38e098f 1030 elapsed = ns - kvm->arch.last_tsc_nsec;
46543ba4
ZA		 1031 sdiff = data - kvm->arch.last_tsc_write;
1032 if (sdiff < 0)
1033 sdiff = -sdiff;
f38e098f
ZA		 1034
1035 /*
46543ba4 1036 * Special case: close write to TSC within 5 seconds of
f38e098f 1037 * another CPU is interpreted as an attempt to synchronize
0d2eb44f 1038 * The 5 seconds is to accommodate host load / swapping as
46543ba4 1039 * well as any reset of TSC during the boot process.
f38e098f
ZA		 1040 *
1041 * In that case, for a reliable TSC, we can match TSC offsets,
46543ba4 1042 * or make a best guest using elapsed value.
f38e098f 1043 */
857e4099 1044 if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
46543ba4 1045 elapsed < 5ULL * NSEC_PER_SEC) {
f38e098f
ZA		 1046 if (!check_tsc_unstable()) {
1047 offset = kvm->arch.last_tsc_offset;
1048 pr_debug("kvm: matched tsc offset for %llu\n", data);
1049 } else {
857e4099 1050 u64 delta = nsec_to_cycles(vcpu, elapsed);
759379dd
ZA		 1051 offset += delta;
1052 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
f38e098f
ZA		 1053 }
1054 ns = kvm->arch.last_tsc_nsec;
1055 }
1056 kvm->arch.last_tsc_nsec = ns;
1057 kvm->arch.last_tsc_write = data;
1058 kvm->arch.last_tsc_offset = offset;
99e3e30a 1059 kvm_x86_ops->write_tsc_offset(vcpu, offset);
038f8c11 1060 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
99e3e30a
ZA		 1061
1062 /* Reset of TSC must disable overshoot protection below */
1063 vcpu->arch.hv_clock.tsc_timestamp = 0;
c285545f
ZA		 1064 vcpu->arch.last_tsc_write = data;
1065 vcpu->arch.last_tsc_nsec = ns;
99e3e30a
ZA		 1066}
1067EXPORT_SYMBOL_GPL(kvm_write_tsc);
1068
34c238a1 1069static int kvm_guest_time_update(struct kvm_vcpu *v)
18068523 1070{
18068523
GOC		 1071 unsigned long flags;
1072 struct kvm_vcpu_arch *vcpu = &v->arch;
1073 void *shared_kaddr;
463656c0 1074 unsigned long this_tsc_khz;
1d5f066e
ZA		 1075 s64 kernel_ns, max_kernel_ns;
1076 u64 tsc_timestamp;
18068523 1077
18068523
GOC		 1078 /* Keep irq disabled to prevent changes to the clock */
1079 local_irq_save(flags);
1d5f066e 1080 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
759379dd 1081 kernel_ns = get_kernel_ns();
1e993611 1082 this_tsc_khz = vcpu_tsc_khz(v);
8cfdc000 1083 if (unlikely(this_tsc_khz == 0)) {
c285545f 1084 local_irq_restore(flags);
34c238a1 1085 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
8cfdc000
ZA		 1086 return 1;
1087 }
18068523 1088
c285545f
ZA		 1089 /*
1090 * We may have to catch up the TSC to match elapsed wall clock
1091 * time for two reasons, even if kvmclock is used.
1092 * 1) CPU could have been running below the maximum TSC rate
1093 * 2) Broken TSC compensation resets the base at each VCPU
1094 * entry to avoid unknown leaps of TSC even when running
1095 * again on the same CPU. This may cause apparent elapsed
1096 * time to disappear, and the guest to stand still or run
1097 * very slowly.
1098 */
1099 if (vcpu->tsc_catchup) {
1100 u64 tsc = compute_guest_tsc(v, kernel_ns);
1101 if (tsc > tsc_timestamp) {
1102 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1103 tsc_timestamp = tsc;
1104 }
50d0a0f9
GH		 1105 }
1106
18068523
GOC		 1107 local_irq_restore(flags);
1108
c285545f
ZA		 1109 if (!vcpu->time_page)
1110 return 0;
18068523 1111
1d5f066e
ZA		 1112 /*
1113 * Time as measured by the TSC may go backwards when resetting the base
1114 * tsc_timestamp. The reason for this is that the TSC resolution is
1115 * higher than the resolution of the other clock scales. Thus, many
1116 * possible measurments of the TSC correspond to one measurement of any
1117 * other clock, and so a spread of values is possible. This is not a
1118 * problem for the computation of the nanosecond clock; with TSC rates
1119 * around 1GHZ, there can only be a few cycles which correspond to one
1120 * nanosecond value, and any path through this code will inevitably
1121 * take longer than that. However, with the kernel_ns value itself,
1122 * the precision may be much lower, down to HZ granularity. If the
1123 * first sampling of TSC against kernel_ns ends in the low part of the
1124 * range, and the second in the high end of the range, we can get:
1125 *
1126 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1127 *
1128 * As the sampling errors potentially range in the thousands of cycles,
1129 * it is possible such a time value has already been observed by the
1130 * guest. To protect against this, we must compute the system time as
1131 * observed by the guest and ensure the new system time is greater.
1132 */
1133 max_kernel_ns = 0;
1134 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1135 max_kernel_ns = vcpu->last_guest_tsc -
1136 vcpu->hv_clock.tsc_timestamp;
1137 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1138 vcpu->hv_clock.tsc_to_system_mul,
1139 vcpu->hv_clock.tsc_shift);
1140 max_kernel_ns += vcpu->last_kernel_ns;
1141 }
afbcf7ab 1142
e48672fa 1143 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
5f4e3f88
ZA		 1144 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1145 &vcpu->hv_clock.tsc_shift,
1146 &vcpu->hv_clock.tsc_to_system_mul);
e48672fa 1147 vcpu->hw_tsc_khz = this_tsc_khz;
8cfdc000
ZA		 1148 }
1149
1d5f066e
ZA		 1150 if (max_kernel_ns > kernel_ns)
1151 kernel_ns = max_kernel_ns;
1152
8cfdc000 1153 /* With all the info we got, fill in the values */
1d5f066e 1154 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
759379dd 1155 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1d5f066e 1156 vcpu->last_kernel_ns = kernel_ns;
28e4639a 1157 vcpu->last_guest_tsc = tsc_timestamp;
371bcf64
GC		 1158 vcpu->hv_clock.flags = 0;
1159
18068523
GOC		 1160 /*
1161 * The interface expects us to write an even number signaling that the
1162 * update is finished. Since the guest won't see the intermediate
50d0a0f9 1163 * state, we just increase by 2 at the end.
18068523 1164 */
50d0a0f9 1165 vcpu->hv_clock.version += 2;
18068523
GOC		 1166
1167 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1168
1169 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
50d0a0f9 1170 sizeof(vcpu->hv_clock));
18068523
GOC		 1171
1172 kunmap_atomic(shared_kaddr, KM_USER0);
1173
1174 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
8cfdc000 1175 return 0;
c8076604
GH		 1176}
1177
9ba075a6
AK		 1178static bool msr_mtrr_valid(unsigned msr)
1179{
1180 switch (msr) {
1181 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1182 case MSR_MTRRfix64K_00000:
1183 case MSR_MTRRfix16K_80000:
1184 case MSR_MTRRfix16K_A0000:
1185 case MSR_MTRRfix4K_C0000:
1186 case MSR_MTRRfix4K_C8000:
1187 case MSR_MTRRfix4K_D0000:
1188 case MSR_MTRRfix4K_D8000:
1189 case MSR_MTRRfix4K_E0000:
1190 case MSR_MTRRfix4K_E8000:
1191 case MSR_MTRRfix4K_F0000:
1192 case MSR_MTRRfix4K_F8000:
1193 case MSR_MTRRdefType:
1194 case MSR_IA32_CR_PAT:
1195 return true;
1196 case 0x2f8:
1197 return true;
1198 }
1199 return false;
1200}
1201
d6289b93
MT		 1202static bool valid_pat_type(unsigned t)
1203{
1204 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1205}
1206
1207static bool valid_mtrr_type(unsigned t)
1208{
1209 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1210}
1211
1212static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1213{
1214 int i;
1215
1216 if (!msr_mtrr_valid(msr))
1217 return false;
1218
1219 if (msr == MSR_IA32_CR_PAT) {
1220 for (i = 0; i < 8; i++)
1221 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1222 return false;
1223 return true;
1224 } else if (msr == MSR_MTRRdefType) {
1225 if (data & ~0xcff)
1226 return false;
1227 return valid_mtrr_type(data & 0xff);
1228 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1229 for (i = 0; i < 8 ; i++)
1230 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1231 return false;
1232 return true;
1233 }
1234
1235 /* variable MTRRs */
1236 return valid_mtrr_type(data & 0xff);
1237}
1238
9ba075a6
AK		 1239static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1240{
0bed3b56
SY		 1241 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1242
d6289b93 1243 if (!mtrr_valid(vcpu, msr, data))
9ba075a6
AK		 1244 return 1;
1245
0bed3b56
SY		 1246 if (msr == MSR_MTRRdefType) {
1247 vcpu->arch.mtrr_state.def_type = data;
1248 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1249 } else if (msr == MSR_MTRRfix64K_00000)
1250 p[0] = data;
1251 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1252 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1253 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1254 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1255 else if (msr == MSR_IA32_CR_PAT)
1256 vcpu->arch.pat = data;
1257 else { /* Variable MTRRs */
1258 int idx, is_mtrr_mask;
1259 u64 *pt;
1260
1261 idx = (msr - 0x200) / 2;
1262 is_mtrr_mask = msr - 0x200 - 2 * idx;
1263 if (!is_mtrr_mask)
1264 pt =
1265 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1266 else
1267 pt =
1268 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1269 *pt = data;
1270 }
1271
1272 kvm_mmu_reset_context(vcpu);
9ba075a6
AK		 1273 return 0;
1274}
15c4a640 1275
890ca9ae 1276static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
15c4a640 1277{
890ca9ae
HY		 1278 u64 mcg_cap = vcpu->arch.mcg_cap;
1279 unsigned bank_num = mcg_cap & 0xff;
1280
15c4a640 1281 switch (msr) {
15c4a640 1282 case MSR_IA32_MCG_STATUS:
890ca9ae 1283 vcpu->arch.mcg_status = data;
15c4a640 1284 break;
c7ac679c 1285 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 1286 if (!(mcg_cap & MCG_CTL_P))
1287 return 1;
1288 if (data != 0 && data != ~(u64)0)
1289 return -1;
1290 vcpu->arch.mcg_ctl = data;
1291 break;
1292 default:
1293 if (msr >= MSR_IA32_MC0_CTL &&
1294 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1295 u32 offset = msr - MSR_IA32_MC0_CTL;
114be429
AP		 1296 /* only 0 or all 1s can be written to IA32_MCi_CTL
1297 * some Linux kernels though clear bit 10 in bank 4 to
1298 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1299 * this to avoid an uncatched #GP in the guest
1300 */
890ca9ae 1301 if ((offset & 0x3) == 0 &&
114be429 1302 data != 0 && (data | (1 << 10)) != ~(u64)0)
890ca9ae
HY		 1303 return -1;
1304 vcpu->arch.mce_banks[offset] = data;
1305 break;
1306 }
1307 return 1;
1308 }
1309 return 0;
1310}
1311
ffde22ac
ES		 1312static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1313{
1314 struct kvm *kvm = vcpu->kvm;
1315 int lm = is_long_mode(vcpu);
1316 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1317 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1318 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1319 : kvm->arch.xen_hvm_config.blob_size_32;
1320 u32 page_num = data & ~PAGE_MASK;
1321 u64 page_addr = data & PAGE_MASK;
1322 u8 *page;
1323 int r;
1324
1325 r = -E2BIG;
1326 if (page_num >= blob_size)
1327 goto out;
1328 r = -ENOMEM;
1329 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1330 if (!page)
1331 goto out;
1332 r = -EFAULT;
1333 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1334 goto out_free;
1335 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1336 goto out_free;
1337 r = 0;
1338out_free:
1339 kfree(page);
1340out:
1341 return r;
1342}
1343
55cd8e5a
GN		 1344static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1345{
1346 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1347}
1348
1349static bool kvm_hv_msr_partition_wide(u32 msr)
1350{
1351 bool r = false;
1352 switch (msr) {
1353 case HV_X64_MSR_GUEST_OS_ID:
1354 case HV_X64_MSR_HYPERCALL:
1355 r = true;
1356 break;
1357 }
1358
1359 return r;
1360}
1361
1362static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1363{
1364 struct kvm *kvm = vcpu->kvm;
1365
1366 switch (msr) {
1367 case HV_X64_MSR_GUEST_OS_ID:
1368 kvm->arch.hv_guest_os_id = data;
1369 /* setting guest os id to zero disables hypercall page */
1370 if (!kvm->arch.hv_guest_os_id)
1371 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1372 break;
1373 case HV_X64_MSR_HYPERCALL: {
1374 u64 gfn;
1375 unsigned long addr;
1376 u8 instructions[4];
1377
1378 /* if guest os id is not set hypercall should remain disabled */
1379 if (!kvm->arch.hv_guest_os_id)
1380 break;
1381 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1382 kvm->arch.hv_hypercall = data;
1383 break;
1384 }
1385 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1386 addr = gfn_to_hva(kvm, gfn);
1387 if (kvm_is_error_hva(addr))
1388 return 1;
1389 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1390 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1391 if (copy_to_user((void __user *)addr, instructions, 4))
1392 return 1;
1393 kvm->arch.hv_hypercall = data;
1394 break;
1395 }
1396 default:
1397 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1398 "data 0x%llx\n", msr, data);
1399 return 1;
1400 }
1401 return 0;
1402}
1403
1404static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1405{
10388a07
GN		 1406 switch (msr) {
1407 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1408 unsigned long addr;
55cd8e5a 1409
10388a07
GN		 1410 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1411 vcpu->arch.hv_vapic = data;
1412 break;
1413 }
1414 addr = gfn_to_hva(vcpu->kvm, data >>
1415 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1416 if (kvm_is_error_hva(addr))
1417 return 1;
1418 if (clear_user((void __user *)addr, PAGE_SIZE))
1419 return 1;
1420 vcpu->arch.hv_vapic = data;
1421 break;
1422 }
1423 case HV_X64_MSR_EOI:
1424 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1425 case HV_X64_MSR_ICR:
1426 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1427 case HV_X64_MSR_TPR:
1428 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1429 default:
1430 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1431 "data 0x%llx\n", msr, data);
1432 return 1;
1433 }
1434
1435 return 0;
55cd8e5a
GN		 1436}
1437
344d9588
GN		 1438static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1439{
1440 gpa_t gpa = data & ~0x3f;
1441
6adba527
GN		 1442 /* Bits 2:5 are resrved, Should be zero */
1443 if (data & 0x3c)
344d9588
GN		 1444 return 1;
1445
1446 vcpu->arch.apf.msr_val = data;
1447
1448 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1449 kvm_clear_async_pf_completion_queue(vcpu);
1450 kvm_async_pf_hash_reset(vcpu);
1451 return 0;
1452 }
1453
1454 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1455 return 1;
1456
6adba527 1457 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
344d9588
GN		 1458 kvm_async_pf_wakeup_all(vcpu);
1459 return 0;
1460}
1461
12f9a48f
GC		 1462static void kvmclock_reset(struct kvm_vcpu *vcpu)
1463{
1464 if (vcpu->arch.time_page) {
1465 kvm_release_page_dirty(vcpu->arch.time_page);
1466 vcpu->arch.time_page = NULL;
1467 }
1468}
1469
15c4a640
CO		 1470int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1471{
1472 switch (msr) {
15c4a640 1473 case MSR_EFER:
b69e8cae 1474 return set_efer(vcpu, data);
8f1589d9
AP		 1475 case MSR_K7_HWCR:
1476 data &= ~(u64)0x40; /* ignore flush filter disable */
82494028 1477 data &= ~(u64)0x100; /* ignore ignne emulation enable */
8f1589d9
AP		 1478 if (data != 0) {
1479 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1480 data);
1481 return 1;
1482 }
15c4a640 1483 break;
f7c6d140
AP		 1484 case MSR_FAM10H_MMIO_CONF_BASE:
1485 if (data != 0) {
1486 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1487 "0x%llx\n", data);
1488 return 1;
1489 }
15c4a640 1490 break;
c323c0e5 1491 case MSR_AMD64_NB_CFG:
c7ac679c 1492 break;
b5e2fec0
AG		 1493 case MSR_IA32_DEBUGCTLMSR:
1494 if (!data) {
1495 /* We support the non-activated case already */
1496 break;
1497 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1498 /* Values other than LBR and BTF are vendor-specific,
1499 thus reserved and should throw a #GP */
1500 return 1;
1501 }
1502 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1503 __func__, data);
1504 break;
15c4a640
CO		 1505 case MSR_IA32_UCODE_REV:
1506 case MSR_IA32_UCODE_WRITE:
61a6bd67 1507 case MSR_VM_HSAVE_PA:
6098ca93 1508 case MSR_AMD64_PATCH_LOADER:
15c4a640 1509 break;
9ba075a6
AK		 1510 case 0x200 ... 0x2ff:
1511 return set_msr_mtrr(vcpu, msr, data);
15c4a640
CO		 1512 case MSR_IA32_APICBASE:
1513 kvm_set_apic_base(vcpu, data);
1514 break;
0105d1a5
GN		 1515 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1516 return kvm_x2apic_msr_write(vcpu, msr, data);
15c4a640 1517 case MSR_IA32_MISC_ENABLE:
ad312c7c 1518 vcpu->arch.ia32_misc_enable_msr = data;
15c4a640 1519 break;
11c6bffa 1520 case MSR_KVM_WALL_CLOCK_NEW:
18068523
GOC		 1521 case MSR_KVM_WALL_CLOCK:
1522 vcpu->kvm->arch.wall_clock = data;
1523 kvm_write_wall_clock(vcpu->kvm, data);
1524 break;
11c6bffa 1525 case MSR_KVM_SYSTEM_TIME_NEW:
18068523 1526 case MSR_KVM_SYSTEM_TIME: {
12f9a48f 1527 kvmclock_reset(vcpu);
18068523
GOC		 1528
1529 vcpu->arch.time = data;
c285545f 1530 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
18068523
GOC		 1531
1532 /* we verify if the enable bit is set... */
1533 if (!(data & 1))
1534 break;
1535
1536 /* ...but clean it before doing the actual write */
1537 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1538
18068523
GOC		 1539 vcpu->arch.time_page =
1540 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
18068523
GOC		 1541
1542 if (is_error_page(vcpu->arch.time_page)) {
1543 kvm_release_page_clean(vcpu->arch.time_page);
1544 vcpu->arch.time_page = NULL;
1545 }
18068523
GOC		 1546 break;
1547 }
344d9588
GN		 1548 case MSR_KVM_ASYNC_PF_EN:
1549 if (kvm_pv_enable_async_pf(vcpu, data))
1550 return 1;
1551 break;
890ca9ae
HY		 1552 case MSR_IA32_MCG_CTL:
1553 case MSR_IA32_MCG_STATUS:
1554 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1555 return set_msr_mce(vcpu, msr, data);
71db6023
AP		 1556
1557 /* Performance counters are not protected by a CPUID bit,
1558 * so we should check all of them in the generic path for the sake of
1559 * cross vendor migration.
1560 * Writing a zero into the event select MSRs disables them,
1561 * which we perfectly emulate ;-). Any other value should be at least
1562 * reported, some guests depend on them.
1563 */
1564 case MSR_P6_EVNTSEL0:
1565 case MSR_P6_EVNTSEL1:
1566 case MSR_K7_EVNTSEL0:
1567 case MSR_K7_EVNTSEL1:
1568 case MSR_K7_EVNTSEL2:
1569 case MSR_K7_EVNTSEL3:
1570 if (data != 0)
1571 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1572 "0x%x data 0x%llx\n", msr, data);
1573 break;
1574 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1575 * so we ignore writes to make it happy.
1576 */
1577 case MSR_P6_PERFCTR0:
1578 case MSR_P6_PERFCTR1:
1579 case MSR_K7_PERFCTR0:
1580 case MSR_K7_PERFCTR1:
1581 case MSR_K7_PERFCTR2:
1582 case MSR_K7_PERFCTR3:
1583 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1584 "0x%x data 0x%llx\n", msr, data);
1585 break;
84e0cefa
JS		 1586 case MSR_K7_CLK_CTL:
1587 /*
1588 * Ignore all writes to this no longer documented MSR.
1589 * Writes are only relevant for old K7 processors,
1590 * all pre-dating SVM, but a recommended workaround from
1591 * AMD for these chips. It is possible to speicify the
1592 * affected processor models on the command line, hence
1593 * the need to ignore the workaround.
1594 */
1595 break;
55cd8e5a
GN		 1596 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1597 if (kvm_hv_msr_partition_wide(msr)) {
1598 int r;
1599 mutex_lock(&vcpu->kvm->lock);
1600 r = set_msr_hyperv_pw(vcpu, msr, data);
1601 mutex_unlock(&vcpu->kvm->lock);
1602 return r;
1603 } else
1604 return set_msr_hyperv(vcpu, msr, data);
1605 break;
91c9c3ed 1606 case MSR_IA32_BBL_CR_CTL3:
1607 /* Drop writes to this legacy MSR -- see rdmsr
1608 * counterpart for further detail.
1609 */
1610 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1611 break;
15c4a640 1612 default:
ffde22ac
ES		 1613 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1614 return xen_hvm_config(vcpu, data);
ed85c068
AP		 1615 if (!ignore_msrs) {
1616 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1617 msr, data);
1618 return 1;
1619 } else {
1620 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1621 msr, data);
1622 break;
1623 }
15c4a640
CO		 1624 }
1625 return 0;
1626}
1627EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1628
1629
1630/*
1631 * Reads an msr value (of 'msr_index') into 'pdata'.
1632 * Returns 0 on success, non-0 otherwise.
1633 * Assumes vcpu_load() was already called.
1634 */
1635int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1636{
1637 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1638}
1639
9ba075a6
AK		 1640static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1641{
0bed3b56
SY		 1642 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1643
9ba075a6
AK		 1644 if (!msr_mtrr_valid(msr))
1645 return 1;
1646
0bed3b56
SY		 1647 if (msr == MSR_MTRRdefType)
1648 *pdata = vcpu->arch.mtrr_state.def_type +
1649 (vcpu->arch.mtrr_state.enabled << 10);
1650 else if (msr == MSR_MTRRfix64K_00000)
1651 *pdata = p[0];
1652 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1653 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1654 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1655 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1656 else if (msr == MSR_IA32_CR_PAT)
1657 *pdata = vcpu->arch.pat;
1658 else { /* Variable MTRRs */
1659 int idx, is_mtrr_mask;
1660 u64 *pt;
1661
1662 idx = (msr - 0x200) / 2;
1663 is_mtrr_mask = msr - 0x200 - 2 * idx;
1664 if (!is_mtrr_mask)
1665 pt =
1666 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1667 else
1668 pt =
1669 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1670 *pdata = *pt;
1671 }
1672
9ba075a6
AK		 1673 return 0;
1674}
1675
890ca9ae 1676static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
15c4a640
CO		 1677{
1678 u64 data;
890ca9ae
HY		 1679 u64 mcg_cap = vcpu->arch.mcg_cap;
1680 unsigned bank_num = mcg_cap & 0xff;
15c4a640
CO		 1681
1682 switch (msr) {
15c4a640
CO		 1683 case MSR_IA32_P5_MC_ADDR:
1684 case MSR_IA32_P5_MC_TYPE:
890ca9ae
HY		 1685 data = 0;
1686 break;
15c4a640 1687 case MSR_IA32_MCG_CAP:
890ca9ae
HY		 1688 data = vcpu->arch.mcg_cap;
1689 break;
c7ac679c 1690 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 1691 if (!(mcg_cap & MCG_CTL_P))
1692 return 1;
1693 data = vcpu->arch.mcg_ctl;
1694 break;
1695 case MSR_IA32_MCG_STATUS:
1696 data = vcpu->arch.mcg_status;
1697 break;
1698 default:
1699 if (msr >= MSR_IA32_MC0_CTL &&
1700 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1701 u32 offset = msr - MSR_IA32_MC0_CTL;
1702 data = vcpu->arch.mce_banks[offset];
1703 break;
1704 }
1705 return 1;
1706 }
1707 *pdata = data;
1708 return 0;
1709}
1710
55cd8e5a
GN		 1711static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1712{
1713 u64 data = 0;
1714 struct kvm *kvm = vcpu->kvm;
1715
1716 switch (msr) {
1717 case HV_X64_MSR_GUEST_OS_ID:
1718 data = kvm->arch.hv_guest_os_id;
1719 break;
1720 case HV_X64_MSR_HYPERCALL:
1721 data = kvm->arch.hv_hypercall;
1722 break;
1723 default:
1724 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1725 return 1;
1726 }
1727
1728 *pdata = data;
1729 return 0;
1730}
1731
1732static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1733{
1734 u64 data = 0;
1735
1736 switch (msr) {
1737 case HV_X64_MSR_VP_INDEX: {
1738 int r;
1739 struct kvm_vcpu *v;
1740 kvm_for_each_vcpu(r, v, vcpu->kvm)
1741 if (v == vcpu)
1742 data = r;
1743 break;
1744 }
10388a07
GN		 1745 case HV_X64_MSR_EOI:
1746 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1747 case HV_X64_MSR_ICR:
1748 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1749 case HV_X64_MSR_TPR:
1750 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
55cd8e5a
GN		 1751 default:
1752 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1753 return 1;
1754 }
1755 *pdata = data;
1756 return 0;
1757}
1758
890ca9ae
HY		 1759int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1760{
1761 u64 data;
1762
1763 switch (msr) {
890ca9ae 1764 case MSR_IA32_PLATFORM_ID:
15c4a640 1765 case MSR_IA32_UCODE_REV:
15c4a640 1766 case MSR_IA32_EBL_CR_POWERON:
b5e2fec0
AG		 1767 case MSR_IA32_DEBUGCTLMSR:
1768 case MSR_IA32_LASTBRANCHFROMIP:
1769 case MSR_IA32_LASTBRANCHTOIP:
1770 case MSR_IA32_LASTINTFROMIP:
1771 case MSR_IA32_LASTINTTOIP:
60af2ecd
JSR		 1772 case MSR_K8_SYSCFG:
1773 case MSR_K7_HWCR:
61a6bd67 1774 case MSR_VM_HSAVE_PA:
1f3ee616
AS		 1775 case MSR_P6_PERFCTR0:
1776 case MSR_P6_PERFCTR1:
7fe29e0f
AS		 1777 case MSR_P6_EVNTSEL0:
1778 case MSR_P6_EVNTSEL1:
9e699624 1779 case MSR_K7_EVNTSEL0:
1f3ee616 1780 case MSR_K7_PERFCTR0:
1fdbd48c 1781 case MSR_K8_INT_PENDING_MSG:
c323c0e5 1782 case MSR_AMD64_NB_CFG:
f7c6d140 1783 case MSR_FAM10H_MMIO_CONF_BASE:
15c4a640
CO		 1784 data = 0;
1785 break;
9ba075a6
AK		 1786 case MSR_MTRRcap:
1787 data = 0x500 | KVM_NR_VAR_MTRR;
1788 break;
1789 case 0x200 ... 0x2ff:
1790 return get_msr_mtrr(vcpu, msr, pdata);
15c4a640
CO		 1791 case 0xcd: /* fsb frequency */
1792 data = 3;
1793 break;
7b914098
JS		 1794 /*
1795 * MSR_EBC_FREQUENCY_ID
1796 * Conservative value valid for even the basic CPU models.
1797 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1798 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1799 * and 266MHz for model 3, or 4. Set Core Clock
1800 * Frequency to System Bus Frequency Ratio to 1 (bits
1801 * 31:24) even though these are only valid for CPU
1802 * models > 2, however guests may end up dividing or
1803 * multiplying by zero otherwise.
1804 */
1805 case MSR_EBC_FREQUENCY_ID:
1806 data = 1 << 24;
1807 break;
15c4a640
CO		 1808 case MSR_IA32_APICBASE:
1809 data = kvm_get_apic_base(vcpu);
1810 break;
0105d1a5
GN		 1811 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1812 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1813 break;
15c4a640 1814 case MSR_IA32_MISC_ENABLE:
ad312c7c 1815 data = vcpu->arch.ia32_misc_enable_msr;
15c4a640 1816 break;
847f0ad8
AG		 1817 case MSR_IA32_PERF_STATUS:
1818 /* TSC increment by tick */
1819 data = 1000ULL;
1820 /* CPU multiplier */
1821 data |= (((uint64_t)4ULL) << 40);
1822 break;
15c4a640 1823 case MSR_EFER:
f6801dff 1824 data = vcpu->arch.efer;
15c4a640 1825 break;
18068523 1826 case MSR_KVM_WALL_CLOCK:
11c6bffa 1827 case MSR_KVM_WALL_CLOCK_NEW:
18068523
GOC		 1828 data = vcpu->kvm->arch.wall_clock;
1829 break;
1830 case MSR_KVM_SYSTEM_TIME:
11c6bffa 1831 case MSR_KVM_SYSTEM_TIME_NEW:
18068523
GOC		 1832 data = vcpu->arch.time;
1833 break;
344d9588
GN		 1834 case MSR_KVM_ASYNC_PF_EN:
1835 data = vcpu->arch.apf.msr_val;
1836 break;
890ca9ae
HY		 1837 case MSR_IA32_P5_MC_ADDR:
1838 case MSR_IA32_P5_MC_TYPE:
1839 case MSR_IA32_MCG_CAP:
1840 case MSR_IA32_MCG_CTL:
1841 case MSR_IA32_MCG_STATUS:
1842 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1843 return get_msr_mce(vcpu, msr, pdata);
84e0cefa
JS		 1844 case MSR_K7_CLK_CTL:
1845 /*
1846 * Provide expected ramp-up count for K7. All other
1847 * are set to zero, indicating minimum divisors for
1848 * every field.
1849 *
1850 * This prevents guest kernels on AMD host with CPU
1851 * type 6, model 8 and higher from exploding due to
1852 * the rdmsr failing.
1853 */
1854 data = 0x20000000;
1855 break;
55cd8e5a
GN		 1856 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1857 if (kvm_hv_msr_partition_wide(msr)) {
1858 int r;
1859 mutex_lock(&vcpu->kvm->lock);
1860 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1861 mutex_unlock(&vcpu->kvm->lock);
1862 return r;
1863 } else
1864 return get_msr_hyperv(vcpu, msr, pdata);
1865 break;
91c9c3ed 1866 case MSR_IA32_BBL_CR_CTL3:
1867 /* This legacy MSR exists but isn't fully documented in current
1868 * silicon. It is however accessed by winxp in very narrow
1869 * scenarios where it sets bit #19, itself documented as
1870 * a "reserved" bit. Best effort attempt to source coherent
1871 * read data here should the balance of the register be
1872 * interpreted by the guest:
1873 *
1874 * L2 cache control register 3: 64GB range, 256KB size,
1875 * enabled, latency 0x1, configured
1876 */
1877 data = 0xbe702111;
1878 break;
15c4a640 1879 default:
ed85c068
AP		 1880 if (!ignore_msrs) {
1881 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1882 return 1;
1883 } else {
1884 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1885 data = 0;
1886 }
1887 break;
15c4a640
CO		 1888 }
1889 *pdata = data;
1890 return 0;
1891}
1892EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1893
313a3dc7
CO		 1894/*
1895 * Read or write a bunch of msrs. All parameters are kernel addresses.
1896 *
1897 * @return number of msrs set successfully.
1898 */
1899static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1900 struct kvm_msr_entry *entries,
1901 int (*do_msr)(struct kvm_vcpu *vcpu,
1902 unsigned index, u64 *data))
1903{
f656ce01 1904 int i, idx;
313a3dc7 1905
f656ce01 1906 idx = srcu_read_lock(&vcpu->kvm->srcu);
313a3dc7
CO		 1907 for (i = 0; i < msrs->nmsrs; ++i)
1908 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1909 break;
f656ce01 1910 srcu_read_unlock(&vcpu->kvm->srcu, idx);
313a3dc7 1911
313a3dc7
CO		 1912 return i;
1913}
1914
1915/*
1916 * Read or write a bunch of msrs. Parameters are user addresses.
1917 *
1918 * @return number of msrs set successfully.
1919 */
1920static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1921 int (*do_msr)(struct kvm_vcpu *vcpu,
1922 unsigned index, u64 *data),
1923 int writeback)
1924{
1925 struct kvm_msrs msrs;
1926 struct kvm_msr_entry *entries;
1927 int r, n;
1928 unsigned size;
1929
1930 r = -EFAULT;
1931 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1932 goto out;
1933
1934 r = -E2BIG;
1935 if (msrs.nmsrs >= MAX_IO_MSRS)
1936 goto out;
1937
1938 r = -ENOMEM;
1939 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
7a73c028 1940 entries = kmalloc(size, GFP_KERNEL);
313a3dc7
CO		 1941 if (!entries)
1942 goto out;
1943
1944 r = -EFAULT;
1945 if (copy_from_user(entries, user_msrs->entries, size))
1946 goto out_free;
1947
1948 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1949 if (r < 0)
1950 goto out_free;
1951
1952 r = -EFAULT;
1953 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1954 goto out_free;
1955
1956 r = n;
1957
1958out_free:
7a73c028 1959 kfree(entries);
313a3dc7
CO		 1960out:
1961 return r;
1962}
1963
018d00d2
ZX		 1964int kvm_dev_ioctl_check_extension(long ext)
1965{
1966 int r;
1967
1968 switch (ext) {
1969 case KVM_CAP_IRQCHIP:
1970 case KVM_CAP_HLT:
1971 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
018d00d2 1972 case KVM_CAP_SET_TSS_ADDR:
07716717 1973 case KVM_CAP_EXT_CPUID:
c8076604 1974 case KVM_CAP_CLOCKSOURCE:
7837699f 1975 case KVM_CAP_PIT:
a28e4f5a 1976 case KVM_CAP_NOP_IO_DELAY:
62d9f0db 1977 case KVM_CAP_MP_STATE:
ed848624 1978 case KVM_CAP_SYNC_MMU:
a355c85c 1979 case KVM_CAP_USER_NMI:
52d939a0 1980 case KVM_CAP_REINJECT_CONTROL:
4925663a 1981 case KVM_CAP_IRQ_INJECT_STATUS:
e56d532f 1982 case KVM_CAP_ASSIGN_DEV_IRQ:
721eecbf 1983 case KVM_CAP_IRQFD:
d34e6b17 1984 case KVM_CAP_IOEVENTFD:
c5ff41ce 1985 case KVM_CAP_PIT2:
e9f42757 1986 case KVM_CAP_PIT_STATE2:
b927a3ce 1987 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
ffde22ac 1988 case KVM_CAP_XEN_HVM:
afbcf7ab 1989 case KVM_CAP_ADJUST_CLOCK:
3cfc3092 1990 case KVM_CAP_VCPU_EVENTS:
55cd8e5a 1991 case KVM_CAP_HYPERV:
10388a07 1992 case KVM_CAP_HYPERV_VAPIC:
c25bc163 1993 case KVM_CAP_HYPERV_SPIN:
ab9f4ecb 1994 case KVM_CAP_PCI_SEGMENT:
a1efbe77 1995 case KVM_CAP_DEBUGREGS:
d2be1651 1996 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2d5b5a66 1997 case KVM_CAP_XSAVE:
344d9588 1998 case KVM_CAP_ASYNC_PF:
92a1f12d 1999 case KVM_CAP_GET_TSC_KHZ:
018d00d2
ZX		 2000 r = 1;
2001 break;
542472b5
LV		 2002 case KVM_CAP_COALESCED_MMIO:
2003 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2004 break;
774ead3a
AK		 2005 case KVM_CAP_VAPIC:
2006 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2007 break;
f725230a
AK		 2008 case KVM_CAP_NR_VCPUS:
2009 r = KVM_MAX_VCPUS;
2010 break;
a988b910
AK		 2011 case KVM_CAP_NR_MEMSLOTS:
2012 r = KVM_MEMORY_SLOTS;
2013 break;
a68a6a72
MT		 2014 case KVM_CAP_PV_MMU: /* obsolete */
2015 r = 0;
2f333bcb 2016 break;
62c476c7 2017 case KVM_CAP_IOMMU:
19de40a8 2018 r = iommu_found();
62c476c7 2019 break;
890ca9ae
HY		 2020 case KVM_CAP_MCE:
2021 r = KVM_MAX_MCE_BANKS;
2022 break;
2d5b5a66
SY		 2023 case KVM_CAP_XCRS:
2024 r = cpu_has_xsave;
2025 break;
92a1f12d
JR		 2026 case KVM_CAP_TSC_CONTROL:
2027 r = kvm_has_tsc_control;
2028 break;
018d00d2
ZX		 2029 default:
2030 r = 0;
2031 break;
2032 }
2033 return r;
2034
2035}
2036
043405e1
CO		 2037long kvm_arch_dev_ioctl(struct file *filp,
2038 unsigned int ioctl, unsigned long arg)
2039{
2040 void __user *argp = (void __user *)arg;
2041 long r;
2042
2043 switch (ioctl) {
2044 case KVM_GET_MSR_INDEX_LIST: {
2045 struct kvm_msr_list __user *user_msr_list = argp;
2046 struct kvm_msr_list msr_list;
2047 unsigned n;
2048
2049 r = -EFAULT;
2050 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2051 goto out;
2052 n = msr_list.nmsrs;
2053 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2054 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2055 goto out;
2056 r = -E2BIG;
e125e7b6 2057 if (n < msr_list.nmsrs)
043405e1
CO		 2058 goto out;
2059 r = -EFAULT;
2060 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2061 num_msrs_to_save * sizeof(u32)))
2062 goto out;
e125e7b6 2063 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
043405e1
CO		 2064 &emulated_msrs,
2065 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2066 goto out;
2067 r = 0;
2068 break;
2069 }
674eea0f
AK		 2070 case KVM_GET_SUPPORTED_CPUID: {
2071 struct kvm_cpuid2 __user *cpuid_arg = argp;
2072 struct kvm_cpuid2 cpuid;
2073
2074 r = -EFAULT;
2075 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2076 goto out;
2077 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
19355475 2078 cpuid_arg->entries);
674eea0f
AK		 2079 if (r)
2080 goto out;
2081
2082 r = -EFAULT;
2083 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2084 goto out;
2085 r = 0;
2086 break;
2087 }
890ca9ae
HY		 2088 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2089 u64 mce_cap;
2090
2091 mce_cap = KVM_MCE_CAP_SUPPORTED;
2092 r = -EFAULT;
2093 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2094 goto out;
2095 r = 0;
2096 break;
2097 }
043405e1
CO		 2098 default:
2099 r = -EINVAL;
2100 }
2101out:
2102 return r;
2103}
2104
f5f48ee1
SY		 2105static void wbinvd_ipi(void *garbage)
2106{
2107 wbinvd();
2108}
2109
2110static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2111{
2112 return vcpu->kvm->arch.iommu_domain &&
2113 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2114}
2115
313a3dc7
CO		 2116void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2117{
f5f48ee1
SY		 2118 /* Address WBINVD may be executed by guest */
2119 if (need_emulate_wbinvd(vcpu)) {
2120 if (kvm_x86_ops->has_wbinvd_exit())
2121 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2122 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2123 smp_call_function_single(vcpu->cpu,
2124 wbinvd_ipi, NULL, 1);
2125 }
2126
313a3dc7 2127 kvm_x86_ops->vcpu_load(vcpu, cpu);
48434c20 2128 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
e48672fa 2129 /* Make sure TSC doesn't go backwards */
8f6055cb
JR		 2130 s64 tsc_delta;
2131 u64 tsc;
2132
2133 kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
2134 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2135 tsc - vcpu->arch.last_guest_tsc;
2136
e48672fa
ZA		 2137 if (tsc_delta < 0)
2138 mark_tsc_unstable("KVM discovered backwards TSC");
c285545f 2139 if (check_tsc_unstable()) {
e48672fa 2140 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
c285545f 2141 vcpu->arch.tsc_catchup = 1;
c285545f 2142 }
1aa8ceef 2143 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
c285545f
ZA		 2144 if (vcpu->cpu != cpu)
2145 kvm_migrate_timers(vcpu);
e48672fa 2146 vcpu->cpu = cpu;
6b7d7e76 2147 }
313a3dc7
CO		 2148}
2149
2150void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2151{
02daab21 2152 kvm_x86_ops->vcpu_put(vcpu);
1c11e713 2153 kvm_put_guest_fpu(vcpu);
7c4c0f4f 2154 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
313a3dc7
CO		 2155}
2156
07716717 2157static int is_efer_nx(void)
313a3dc7 2158{
e286e86e 2159 unsigned long long efer = 0;
313a3dc7 2160
e286e86e 2161 rdmsrl_safe(MSR_EFER, &efer);
07716717
DK		 2162 return efer & EFER_NX;
2163}
2164
2165static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2166{
2167 int i;
2168 struct kvm_cpuid_entry2 *e, *entry;
2169
313a3dc7 2170 entry = NULL;
ad312c7c
ZX		 2171 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2172 e = &vcpu->arch.cpuid_entries[i];
313a3dc7
CO		 2173 if (e->function == 0x80000001) {
2174 entry = e;
2175 break;
2176 }
2177 }
07716717 2178 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
313a3dc7
CO		 2179 entry->edx &= ~(1 << 20);
2180 printk(KERN_INFO "kvm: guest NX capability removed\n");
2181 }
2182}
2183
07716717 2184/* when an old userspace process fills a new kernel module */
313a3dc7
CO		 2185static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2186 struct kvm_cpuid *cpuid,
2187 struct kvm_cpuid_entry __user *entries)
07716717
DK		 2188{
2189 int r, i;
2190 struct kvm_cpuid_entry *cpuid_entries;
2191
2192 r = -E2BIG;
2193 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2194 goto out;
2195 r = -ENOMEM;
2196 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2197 if (!cpuid_entries)
2198 goto out;
2199 r = -EFAULT;
2200 if (copy_from_user(cpuid_entries, entries,
2201 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2202 goto out_free;
2203 for (i = 0; i < cpuid->nent; i++) {
ad312c7c
ZX		 2204 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2205 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2206 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2207 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2208 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2209 vcpu->arch.cpuid_entries[i].index = 0;
2210 vcpu->arch.cpuid_entries[i].flags = 0;
2211 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2212 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2213 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2214 }
2215 vcpu->arch.cpuid_nent = cpuid->nent;
07716717
DK		 2216 cpuid_fix_nx_cap(vcpu);
2217 r = 0;
fc61b800 2218 kvm_apic_set_version(vcpu);
0e851880 2219 kvm_x86_ops->cpuid_update(vcpu);
2acf923e 2220 update_cpuid(vcpu);
07716717
DK		 2221
2222out_free:
2223 vfree(cpuid_entries);
2224out:
2225 return r;
2226}
2227
2228static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
19355475
AS		 2229 struct kvm_cpuid2 *cpuid,
2230 struct kvm_cpuid_entry2 __user *entries)
313a3dc7
CO		 2231{
2232 int r;
2233
2234 r = -E2BIG;
2235 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2236 goto out;
2237 r = -EFAULT;
ad312c7c 2238 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
07716717 2239 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
313a3dc7 2240 goto out;
ad312c7c 2241 vcpu->arch.cpuid_nent = cpuid->nent;
fc61b800 2242 kvm_apic_set_version(vcpu);
0e851880 2243 kvm_x86_ops->cpuid_update(vcpu);
2acf923e 2244 update_cpuid(vcpu);
313a3dc7
CO		 2245 return 0;
2246
2247out:
2248 return r;
2249}
2250
07716717 2251static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
19355475
AS		 2252 struct kvm_cpuid2 *cpuid,
2253 struct kvm_cpuid_entry2 __user *entries)
07716717
DK		 2254{
2255 int r;
2256
2257 r = -E2BIG;
ad312c7c 2258 if (cpuid->nent < vcpu->arch.cpuid_nent)
07716717
DK		 2259 goto out;
2260 r = -EFAULT;
ad312c7c 2261 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
19355475 2262 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
07716717
DK		 2263 goto out;
2264 return 0;
2265
2266out:
ad312c7c 2267 cpuid->nent = vcpu->arch.cpuid_nent;
07716717
DK		 2268 return r;
2269}
2270
945ee35e
AK		 2271static void cpuid_mask(u32 *word, int wordnum)
2272{
2273 *word &= boot_cpu_data.x86_capability[wordnum];
2274}
2275
07716717 2276static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
19355475 2277 u32 index)
07716717
DK		 2278{
2279 entry->function = function;
2280 entry->index = index;
2281 cpuid_count(entry->function, entry->index,
19355475 2282 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
07716717
DK		 2283 entry->flags = 0;
2284}
2285
7faa4ee1
AK		 2286#define F(x) bit(X86_FEATURE_##x)
2287
07716717
DK		 2288static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2289 u32 index, int *nent, int maxnent)
2290{
7faa4ee1 2291 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
07716717 2292#ifdef CONFIG_X86_64
17cc3935
SY		 2293 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2294 ? F(GBPAGES) : 0;
7faa4ee1
AK		 2295 unsigned f_lm = F(LM);
2296#else
17cc3935 2297 unsigned f_gbpages = 0;
7faa4ee1 2298 unsigned f_lm = 0;
07716717 2299#endif
4e47c7a6 2300 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
7faa4ee1
AK		 2301
2302 /* cpuid 1.edx */
2303 const u32 kvm_supported_word0_x86_features =
2304 F(FPU) | F(VME) | F(DE) | F(PSE) |
2305 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2306 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2307 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2308 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2309 0 /* Reserved, DS, ACPI */ | F(MMX) |
2310 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2311 0 /* HTT, TM, Reserved, PBE */;
2312 /* cpuid 0x80000001.edx */
2313 const u32 kvm_supported_word1_x86_features =
2314 F(FPU) | F(VME) | F(DE) | F(PSE) |
2315 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2316 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2317 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2318 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2319 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
4e47c7a6 2320 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
7faa4ee1
AK		 2321 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2322 /* cpuid 1.ecx */
2323 const u32 kvm_supported_word4_x86_features =
6c3f6041 2324 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
d149c731
AK		 2325 0 /* DS-CPL, VMX, SMX, EST */ |
2326 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2327 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2328 0 /* Reserved, DCA */ | F(XMM4_1) |
0105d1a5 2329 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
6d886fd0
AP		 2330 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2331 F(F16C);
7faa4ee1 2332 /* cpuid 0x80000001.ecx */
07716717 2333 const u32 kvm_supported_word6_x86_features =
4c62a2dc 2334 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
7faa4ee1 2335 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
7ef8aa72 2336 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
6d886fd0 2337 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
07716717 2338
19355475 2339 /* all calls to cpuid_count() should be made on the same cpu */
07716717
DK		 2340 get_cpu();
2341 do_cpuid_1_ent(entry, function, index);
2342 ++*nent;
2343
2344 switch (function) {
2345 case 0:
2acf923e 2346 entry->eax = min(entry->eax, (u32)0xd);
07716717
DK		 2347 break;
2348 case 1:
2349 entry->edx &= kvm_supported_word0_x86_features;
945ee35e 2350 cpuid_mask(&entry->edx, 0);
7faa4ee1 2351 entry->ecx &= kvm_supported_word4_x86_features;
945ee35e 2352 cpuid_mask(&entry->ecx, 4);
0d1de2d9
GN		 2353 /* we support x2apic emulation even if host does not support
2354 * it since we emulate x2apic in software */
2355 entry->ecx |= F(X2APIC);
07716717
DK		 2356 break;
2357 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2358 * may return different values. This forces us to get_cpu() before
2359 * issuing the first command, and also to emulate this annoying behavior
2360 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2361 case 2: {
2362 int t, times = entry->eax & 0xff;
2363
2364 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
0fdf8e59 2365 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
07716717
DK		 2366 for (t = 1; t < times && *nent < maxnent; ++t) {
2367 do_cpuid_1_ent(&entry[t], function, 0);
2368 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2369 ++*nent;
2370 }
2371 break;
2372 }
2373 /* function 4 and 0xb have additional index. */
2374 case 4: {
14af3f3c 2375 int i, cache_type;
07716717
DK		 2376
2377 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2378 /* read more entries until cache_type is zero */
14af3f3c
HH		 2379 for (i = 1; *nent < maxnent; ++i) {
2380 cache_type = entry[i - 1].eax & 0x1f;
07716717
DK		 2381 if (!cache_type)
2382 break;
14af3f3c
HH		 2383 do_cpuid_1_ent(&entry[i], function, i);
2384 entry[i].flags |=
07716717
DK		 2385 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2386 ++*nent;
2387 }
2388 break;
2389 }
2390 case 0xb: {
14af3f3c 2391 int i, level_type;
07716717
DK		 2392
2393 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2394 /* read more entries until level_type is zero */
14af3f3c 2395 for (i = 1; *nent < maxnent; ++i) {
0853d2c1 2396 level_type = entry[i - 1].ecx & 0xff00;
07716717
DK		 2397 if (!level_type)
2398 break;
14af3f3c
HH		 2399 do_cpuid_1_ent(&entry[i], function, i);
2400 entry[i].flags |=
07716717
DK		 2401 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2402 ++*nent;
2403 }
2404 break;
2405 }
2acf923e
DC		 2406 case 0xd: {
2407 int i;
2408
2409 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
20800bc9
AP		 2410 for (i = 1; *nent < maxnent && i < 64; ++i) {
2411 if (entry[i].eax == 0)
2412 continue;
2acf923e
DC		 2413 do_cpuid_1_ent(&entry[i], function, i);
2414 entry[i].flags |=
2415 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2416 ++*nent;
2417 }
2418 break;
2419 }
84478c82
GC		 2420 case KVM_CPUID_SIGNATURE: {
2421 char signature[12] = "KVMKVMKVM\0\0";
2422 u32 *sigptr = (u32 *)signature;
2423 entry->eax = 0;
2424 entry->ebx = sigptr[0];
2425 entry->ecx = sigptr[1];
2426 entry->edx = sigptr[2];
2427 break;
2428 }
2429 case KVM_CPUID_FEATURES:
2430 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2431 (1 << KVM_FEATURE_NOP_IO_DELAY) |
371bcf64 2432 (1 << KVM_FEATURE_CLOCKSOURCE2) |
32918924 2433 (1 << KVM_FEATURE_ASYNC_PF) |
371bcf64 2434 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
84478c82
GC		 2435 entry->ebx = 0;
2436 entry->ecx = 0;
2437 entry->edx = 0;
2438 break;
07716717
DK		 2439 case 0x80000000:
2440 entry->eax = min(entry->eax, 0x8000001a);
2441 break;
2442 case 0x80000001:
2443 entry->edx &= kvm_supported_word1_x86_features;
945ee35e 2444 cpuid_mask(&entry->edx, 1);
07716717 2445 entry->ecx &= kvm_supported_word6_x86_features;
945ee35e 2446 cpuid_mask(&entry->ecx, 6);
07716717
DK		 2447 break;
2448 }
d4330ef2
JR		 2449
2450 kvm_x86_ops->set_supported_cpuid(function, entry);
2451
07716717
DK		 2452 put_cpu();
2453}
2454
7faa4ee1
AK		 2455#undef F
2456
674eea0f 2457static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
19355475 2458 struct kvm_cpuid_entry2 __user *entries)
07716717
DK		 2459{
2460 struct kvm_cpuid_entry2 *cpuid_entries;
2461 int limit, nent = 0, r = -E2BIG;
2462 u32 func;
2463
2464 if (cpuid->nent < 1)
2465 goto out;
6a544355
AK		 2466 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2467 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
07716717
DK		 2468 r = -ENOMEM;
2469 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2470 if (!cpuid_entries)
2471 goto out;
2472
2473 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2474 limit = cpuid_entries[0].eax;
2475 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2476 do_cpuid_ent(&cpuid_entries[nent], func, 0,
19355475 2477 &nent, cpuid->nent);
07716717
DK		 2478 r = -E2BIG;
2479 if (nent >= cpuid->nent)
2480 goto out_free;
2481
2482 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2483 limit = cpuid_entries[nent - 1].eax;
2484 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2485 do_cpuid_ent(&cpuid_entries[nent], func, 0,
19355475 2486 &nent, cpuid->nent);
84478c82
GC		 2487
2488
2489
2490 r = -E2BIG;
2491 if (nent >= cpuid->nent)
2492 goto out_free;
2493
2494 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2495 cpuid->nent);
2496
2497 r = -E2BIG;
2498 if (nent >= cpuid->nent)
2499 goto out_free;
2500
2501 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2502 cpuid->nent);
2503
cb007648
MM		 2504 r = -E2BIG;
2505 if (nent >= cpuid->nent)
2506 goto out_free;
2507
07716717
DK		 2508 r = -EFAULT;
2509 if (copy_to_user(entries, cpuid_entries,
19355475 2510 nent * sizeof(struct kvm_cpuid_entry2)))
07716717
DK		 2511 goto out_free;
2512 cpuid->nent = nent;
2513 r = 0;
2514
2515out_free:
2516 vfree(cpuid_entries);
2517out:
2518 return r;
2519}
2520
313a3dc7
CO		 2521static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2522 struct kvm_lapic_state *s)
2523{
ad312c7c 2524 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
313a3dc7
CO		 2525
2526 return 0;
2527}
2528
2529static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2530 struct kvm_lapic_state *s)
2531{
ad312c7c 2532 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
313a3dc7 2533 kvm_apic_post_state_restore(vcpu);
cb142eb7 2534 update_cr8_intercept(vcpu);
313a3dc7
CO		 2535
2536 return 0;
2537}
2538
f77bc6a4
ZX		 2539static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2540 struct kvm_interrupt *irq)
2541{
2542 if (irq->irq < 0 || irq->irq >= 256)
2543 return -EINVAL;
2544 if (irqchip_in_kernel(vcpu->kvm))
2545 return -ENXIO;
f77bc6a4 2546
66fd3f7f 2547 kvm_queue_interrupt(vcpu, irq->irq, false);
3842d135 2548 kvm_make_request(KVM_REQ_EVENT, vcpu);
f77bc6a4 2549
f77bc6a4
ZX		 2550 return 0;
2551}
2552
c4abb7c9
JK		 2553static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2554{
c4abb7c9 2555 kvm_inject_nmi(vcpu);
c4abb7c9
JK		 2556
2557 return 0;
2558}
2559
b209749f
AK		 2560static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2561 struct kvm_tpr_access_ctl *tac)
2562{
2563 if (tac->flags)
2564 return -EINVAL;
2565 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2566 return 0;
2567}
2568
890ca9ae
HY		 2569static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2570 u64 mcg_cap)
2571{
2572 int r;
2573 unsigned bank_num = mcg_cap & 0xff, bank;
2574
2575 r = -EINVAL;
a9e38c3e 2576 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
890ca9ae
HY		 2577 goto out;
2578 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2579 goto out;
2580 r = 0;
2581 vcpu->arch.mcg_cap = mcg_cap;
2582 /* Init IA32_MCG_CTL to all 1s */
2583 if (mcg_cap & MCG_CTL_P)
2584 vcpu->arch.mcg_ctl = ~(u64)0;
2585 /* Init IA32_MCi_CTL to all 1s */
2586 for (bank = 0; bank < bank_num; bank++)
2587 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2588out:
2589 return r;
2590}
2591
2592static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2593 struct kvm_x86_mce *mce)
2594{
2595 u64 mcg_cap = vcpu->arch.mcg_cap;
2596 unsigned bank_num = mcg_cap & 0xff;
2597 u64 *banks = vcpu->arch.mce_banks;
2598
2599 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2600 return -EINVAL;
2601 /*
2602 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2603 * reporting is disabled
2604 */
2605 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2606 vcpu->arch.mcg_ctl != ~(u64)0)
2607 return 0;
2608 banks += 4 * mce->bank;
2609 /*
2610 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2611 * reporting is disabled for the bank
2612 */
2613 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2614 return 0;
2615 if (mce->status & MCI_STATUS_UC) {
2616 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
fc78f519 2617 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
a8eeb04a 2618 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
890ca9ae
HY		 2619 return 0;
2620 }
2621 if (banks[1] & MCI_STATUS_VAL)
2622 mce->status |= MCI_STATUS_OVER;
2623 banks[2] = mce->addr;
2624 banks[3] = mce->misc;
2625 vcpu->arch.mcg_status = mce->mcg_status;
2626 banks[1] = mce->status;
2627 kvm_queue_exception(vcpu, MC_VECTOR);
2628 } else if (!(banks[1] & MCI_STATUS_VAL)
2629 || !(banks[1] & MCI_STATUS_UC)) {
2630 if (banks[1] & MCI_STATUS_VAL)
2631 mce->status |= MCI_STATUS_OVER;
2632 banks[2] = mce->addr;
2633 banks[3] = mce->misc;
2634 banks[1] = mce->status;
2635 } else
2636 banks[1] |= MCI_STATUS_OVER;
2637 return 0;
2638}
2639
3cfc3092
JK		 2640static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2641 struct kvm_vcpu_events *events)
2642{
03b82a30
JK		 2643 events->exception.injected =
2644 vcpu->arch.exception.pending &&
2645 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3cfc3092
JK		 2646 events->exception.nr = vcpu->arch.exception.nr;
2647 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
97e69aa6 2648 events->exception.pad = 0;
3cfc3092
JK		 2649 events->exception.error_code = vcpu->arch.exception.error_code;
2650
03b82a30
JK		 2651 events->interrupt.injected =
2652 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3cfc3092 2653 events->interrupt.nr = vcpu->arch.interrupt.nr;
03b82a30 2654 events->interrupt.soft = 0;
48005f64
JK		 2655 events->interrupt.shadow =
2656 kvm_x86_ops->get_interrupt_shadow(vcpu,
2657 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
3cfc3092
JK		 2658
2659 events->nmi.injected = vcpu->arch.nmi_injected;
2660 events->nmi.pending = vcpu->arch.nmi_pending;
2661 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
97e69aa6 2662 events->nmi.pad = 0;
3cfc3092
JK		 2663
2664 events->sipi_vector = vcpu->arch.sipi_vector;
2665
dab4b911 2666 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64
JK		 2667 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2668 | KVM_VCPUEVENT_VALID_SHADOW);
97e69aa6 2669 memset(&events->reserved, 0, sizeof(events->reserved));
3cfc3092
JK		 2670}
2671
2672static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2673 struct kvm_vcpu_events *events)
2674{
dab4b911 2675 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64
JK		 2676 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2677 | KVM_VCPUEVENT_VALID_SHADOW))
3cfc3092
JK		 2678 return -EINVAL;
2679
3cfc3092
JK		 2680 vcpu->arch.exception.pending = events->exception.injected;
2681 vcpu->arch.exception.nr = events->exception.nr;
2682 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2683 vcpu->arch.exception.error_code = events->exception.error_code;
2684
2685 vcpu->arch.interrupt.pending = events->interrupt.injected;
2686 vcpu->arch.interrupt.nr = events->interrupt.nr;
2687 vcpu->arch.interrupt.soft = events->interrupt.soft;
48005f64
JK		 2688 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2689 kvm_x86_ops->set_interrupt_shadow(vcpu,
2690 events->interrupt.shadow);
3cfc3092
JK		 2691
2692 vcpu->arch.nmi_injected = events->nmi.injected;
dab4b911
JK		 2693 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2694 vcpu->arch.nmi_pending = events->nmi.pending;
3cfc3092
JK		 2695 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2696
dab4b911
JK		 2697 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2698 vcpu->arch.sipi_vector = events->sipi_vector;
3cfc3092 2699
3842d135
AK		 2700 kvm_make_request(KVM_REQ_EVENT, vcpu);
2701
3cfc3092
JK		 2702 return 0;
2703}
2704
a1efbe77
JK		 2705static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2706 struct kvm_debugregs *dbgregs)
2707{
a1efbe77
JK		 2708 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2709 dbgregs->dr6 = vcpu->arch.dr6;
2710 dbgregs->dr7 = vcpu->arch.dr7;
2711 dbgregs->flags = 0;
97e69aa6 2712 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
a1efbe77
JK		 2713}
2714
2715static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2716 struct kvm_debugregs *dbgregs)
2717{
2718 if (dbgregs->flags)
2719 return -EINVAL;
2720
a1efbe77
JK		 2721 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2722 vcpu->arch.dr6 = dbgregs->dr6;
2723 vcpu->arch.dr7 = dbgregs->dr7;
2724
a1efbe77
JK		 2725 return 0;
2726}
2727
2d5b5a66
SY		 2728static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2729 struct kvm_xsave *guest_xsave)
2730{
2731 if (cpu_has_xsave)
2732 memcpy(guest_xsave->region,
2733 &vcpu->arch.guest_fpu.state->xsave,
f45755b8 2734 xstate_size);
2d5b5a66
SY		 2735 else {
2736 memcpy(guest_xsave->region,
2737 &vcpu->arch.guest_fpu.state->fxsave,
2738 sizeof(struct i387_fxsave_struct));
2739 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2740 XSTATE_FPSSE;
2741 }
2742}
2743
2744static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2745 struct kvm_xsave *guest_xsave)
2746{
2747 u64 xstate_bv =
2748 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2749
2750 if (cpu_has_xsave)
2751 memcpy(&vcpu->arch.guest_fpu.state->xsave,
f45755b8 2752 guest_xsave->region, xstate_size);
2d5b5a66
SY		 2753 else {
2754 if (xstate_bv & ~XSTATE_FPSSE)
2755 return -EINVAL;
2756 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2757 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2758 }
2759 return 0;
2760}
2761
2762static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2763 struct kvm_xcrs *guest_xcrs)
2764{
2765 if (!cpu_has_xsave) {
2766 guest_xcrs->nr_xcrs = 0;
2767 return;
2768 }
2769
2770 guest_xcrs->nr_xcrs = 1;
2771 guest_xcrs->flags = 0;
2772 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2773 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2774}
2775
2776static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2777 struct kvm_xcrs *guest_xcrs)
2778{
2779 int i, r = 0;
2780
2781 if (!cpu_has_xsave)
2782 return -EINVAL;
2783
2784 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2785 return -EINVAL;
2786
2787 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2788 /* Only support XCR0 currently */
2789 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2790 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2791 guest_xcrs->xcrs[0].value);
2792 break;
2793 }
2794 if (r)
2795 r = -EINVAL;
2796 return r;
2797}
2798
313a3dc7
CO		 2799long kvm_arch_vcpu_ioctl(struct file *filp,
2800 unsigned int ioctl, unsigned long arg)
2801{
2802 struct kvm_vcpu *vcpu = filp->private_data;
2803 void __user *argp = (void __user *)arg;
2804 int r;
d1ac91d8
AK		 2805 union {
2806 struct kvm_lapic_state *lapic;
2807 struct kvm_xsave *xsave;
2808 struct kvm_xcrs *xcrs;
2809 void *buffer;
2810 } u;
2811
2812 u.buffer = NULL;
313a3dc7
CO		 2813 switch (ioctl) {
2814 case KVM_GET_LAPIC: {
2204ae3c
MT		 2815 r = -EINVAL;
2816 if (!vcpu->arch.apic)
2817 goto out;
d1ac91d8 2818 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
313a3dc7 2819
b772ff36 2820 r = -ENOMEM;
d1ac91d8 2821 if (!u.lapic)
b772ff36 2822 goto out;
d1ac91d8 2823 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
313a3dc7
CO		 2824 if (r)
2825 goto out;
2826 r = -EFAULT;
d1ac91d8 2827 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
313a3dc7
CO		 2828 goto out;
2829 r = 0;
2830 break;
2831 }
2832 case KVM_SET_LAPIC: {
2204ae3c
MT		 2833 r = -EINVAL;
2834 if (!vcpu->arch.apic)
2835 goto out;
d1ac91d8 2836 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
b772ff36 2837 r = -ENOMEM;
d1ac91d8 2838 if (!u.lapic)
b772ff36 2839 goto out;
313a3dc7 2840 r = -EFAULT;
d1ac91d8 2841 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
313a3dc7 2842 goto out;
d1ac91d8 2843 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
313a3dc7
CO		 2844 if (r)
2845 goto out;
2846 r = 0;
2847 break;
2848 }
f77bc6a4
ZX		 2849 case KVM_INTERRUPT: {
2850 struct kvm_interrupt irq;
2851
2852 r = -EFAULT;
2853 if (copy_from_user(&irq, argp, sizeof irq))
2854 goto out;
2855 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2856 if (r)
2857 goto out;
2858 r = 0;
2859 break;
2860 }
c4abb7c9
JK		 2861 case KVM_NMI: {
2862 r = kvm_vcpu_ioctl_nmi(vcpu);
2863 if (r)
2864 goto out;
2865 r = 0;
2866 break;
2867 }
313a3dc7
CO		 2868 case KVM_SET_CPUID: {
2869 struct kvm_cpuid __user *cpuid_arg = argp;
2870 struct kvm_cpuid cpuid;
2871
2872 r = -EFAULT;
2873 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2874 goto out;
2875 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2876 if (r)
2877 goto out;
2878 break;
2879 }
07716717
DK		 2880 case KVM_SET_CPUID2: {
2881 struct kvm_cpuid2 __user *cpuid_arg = argp;
2882 struct kvm_cpuid2 cpuid;
2883
2884 r = -EFAULT;
2885 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2886 goto out;
2887 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
19355475 2888 cpuid_arg->entries);
07716717
DK		 2889 if (r)
2890 goto out;
2891 break;
2892 }
2893 case KVM_GET_CPUID2: {
2894 struct kvm_cpuid2 __user *cpuid_arg = argp;
2895 struct kvm_cpuid2 cpuid;
2896
2897 r = -EFAULT;
2898 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2899 goto out;
2900 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
19355475 2901 cpuid_arg->entries);
07716717
DK		 2902 if (r)
2903 goto out;
2904 r = -EFAULT;
2905 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2906 goto out;
2907 r = 0;
2908 break;
2909 }
313a3dc7
CO		 2910 case KVM_GET_MSRS:
2911 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2912 break;
2913 case KVM_SET_MSRS:
2914 r = msr_io(vcpu, argp, do_set_msr, 0);
2915 break;
b209749f
AK		 2916 case KVM_TPR_ACCESS_REPORTING: {
2917 struct kvm_tpr_access_ctl tac;
2918
2919 r = -EFAULT;
2920 if (copy_from_user(&tac, argp, sizeof tac))
2921 goto out;
2922 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2923 if (r)
2924 goto out;
2925 r = -EFAULT;
2926 if (copy_to_user(argp, &tac, sizeof tac))
2927 goto out;
2928 r = 0;
2929 break;
2930 };
b93463aa
AK		 2931 case KVM_SET_VAPIC_ADDR: {
2932 struct kvm_vapic_addr va;
2933
2934 r = -EINVAL;
2935 if (!irqchip_in_kernel(vcpu->kvm))
2936 goto out;
2937 r = -EFAULT;
2938 if (copy_from_user(&va, argp, sizeof va))
2939 goto out;
2940 r = 0;
2941 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2942 break;
2943 }
890ca9ae
HY		 2944 case KVM_X86_SETUP_MCE: {
2945 u64 mcg_cap;
2946
2947 r = -EFAULT;
2948 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2949 goto out;
2950 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2951 break;
2952 }
2953 case KVM_X86_SET_MCE: {
2954 struct kvm_x86_mce mce;
2955
2956 r = -EFAULT;
2957 if (copy_from_user(&mce, argp, sizeof mce))
2958 goto out;
2959 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2960 break;
2961 }
3cfc3092
JK		 2962 case KVM_GET_VCPU_EVENTS: {
2963 struct kvm_vcpu_events events;
2964
2965 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2966
2967 r = -EFAULT;
2968 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2969 break;
2970 r = 0;
2971 break;
2972 }
2973 case KVM_SET_VCPU_EVENTS: {
2974 struct kvm_vcpu_events events;
2975
2976 r = -EFAULT;
2977 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2978 break;
2979
2980 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2981 break;
2982 }
a1efbe77
JK		 2983 case KVM_GET_DEBUGREGS: {
2984 struct kvm_debugregs dbgregs;
2985
2986 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2987
2988 r = -EFAULT;
2989 if (copy_to_user(argp, &dbgregs,
2990 sizeof(struct kvm_debugregs)))
2991 break;
2992 r = 0;
2993 break;
2994 }
2995 case KVM_SET_DEBUGREGS: {
2996 struct kvm_debugregs dbgregs;
2997
2998 r = -EFAULT;
2999 if (copy_from_user(&dbgregs, argp,
3000 sizeof(struct kvm_debugregs)))
3001 break;
3002
3003 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3004 break;
3005 }
2d5b5a66 3006 case KVM_GET_XSAVE: {
d1ac91d8 3007 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2d5b5a66 3008 r = -ENOMEM;
d1ac91d8 3009 if (!u.xsave)
2d5b5a66
SY		 3010 break;
3011
d1ac91d8 3012 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3013
3014 r = -EFAULT;
d1ac91d8 3015 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2d5b5a66
SY		 3016 break;
3017 r = 0;
3018 break;
3019 }
3020 case KVM_SET_XSAVE: {
d1ac91d8 3021 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2d5b5a66 3022 r = -ENOMEM;
d1ac91d8 3023 if (!u.xsave)
2d5b5a66
SY		 3024 break;
3025
3026 r = -EFAULT;
d1ac91d8 3027 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
2d5b5a66
SY		 3028 break;
3029
d1ac91d8 3030 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3031 break;
3032 }
3033 case KVM_GET_XCRS: {
d1ac91d8 3034 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2d5b5a66 3035 r = -ENOMEM;
d1ac91d8 3036 if (!u.xcrs)
2d5b5a66
SY		 3037 break;
3038
d1ac91d8 3039 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3040
3041 r = -EFAULT;
d1ac91d8 3042 if (copy_to_user(argp, u.xcrs,
2d5b5a66
SY		 3043 sizeof(struct kvm_xcrs)))
3044 break;
3045 r = 0;
3046 break;
3047 }
3048 case KVM_SET_XCRS: {
d1ac91d8 3049 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2d5b5a66 3050 r = -ENOMEM;
d1ac91d8 3051 if (!u.xcrs)
2d5b5a66
SY		 3052 break;
3053
3054 r = -EFAULT;
d1ac91d8 3055 if (copy_from_user(u.xcrs, argp,
2d5b5a66
SY		 3056 sizeof(struct kvm_xcrs)))
3057 break;
3058
d1ac91d8 3059 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3060 break;
3061 }
92a1f12d
JR		 3062 case KVM_SET_TSC_KHZ: {
3063 u32 user_tsc_khz;
3064
3065 r = -EINVAL;
3066 if (!kvm_has_tsc_control)
3067 break;
3068
3069 user_tsc_khz = (u32)arg;
3070
3071 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3072 goto out;
3073
3074 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3075
3076 r = 0;
3077 goto out;
3078 }
3079 case KVM_GET_TSC_KHZ: {
3080 r = -EIO;
3081 if (check_tsc_unstable())
3082 goto out;
3083
3084 r = vcpu_tsc_khz(vcpu);
3085
3086 goto out;
3087 }
313a3dc7
CO		 3088 default:
3089 r = -EINVAL;
3090 }
3091out:
d1ac91d8 3092 kfree(u.buffer);
313a3dc7
CO		 3093 return r;
3094}
3095
1fe779f8
CO		 3096static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3097{
3098 int ret;
3099
3100 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3101 return -1;
3102 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3103 return ret;
3104}
3105
b927a3ce
SY		 3106static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3107 u64 ident_addr)
3108{
3109 kvm->arch.ept_identity_map_addr = ident_addr;
3110 return 0;
3111}
3112
1fe779f8
CO		 3113static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3114 u32 kvm_nr_mmu_pages)
3115{
3116 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3117 return -EINVAL;
3118
79fac95e 3119 mutex_lock(&kvm->slots_lock);
7c8a83b7 3120 spin_lock(&kvm->mmu_lock);
1fe779f8
CO		 3121
3122 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
f05e70ac 3123 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1fe779f8 3124
7c8a83b7 3125 spin_unlock(&kvm->mmu_lock);
79fac95e 3126 mutex_unlock(&kvm->slots_lock);
1fe779f8
CO		 3127 return 0;
3128}
3129
3130static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3131{
39de71ec 3132 return kvm->arch.n_max_mmu_pages;
1fe779f8
CO		 3133}
3134
1fe779f8
CO		 3135static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3136{
3137 int r;
3138
3139 r = 0;
3140 switch (chip->chip_id) {
3141 case KVM_IRQCHIP_PIC_MASTER:
3142 memcpy(&chip->chip.pic,
3143 &pic_irqchip(kvm)->pics[0],
3144 sizeof(struct kvm_pic_state));
3145 break;
3146 case KVM_IRQCHIP_PIC_SLAVE:
3147 memcpy(&chip->chip.pic,
3148 &pic_irqchip(kvm)->pics[1],
3149 sizeof(struct kvm_pic_state));
3150 break;
3151 case KVM_IRQCHIP_IOAPIC:
eba0226b 3152 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3153 break;
3154 default:
3155 r = -EINVAL;
3156 break;
3157 }
3158 return r;
3159}
3160
3161static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3162{
3163 int r;
3164
3165 r = 0;
3166 switch (chip->chip_id) {
3167 case KVM_IRQCHIP_PIC_MASTER:
f4f51050 3168 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3169 memcpy(&pic_irqchip(kvm)->pics[0],
3170 &chip->chip.pic,
3171 sizeof(struct kvm_pic_state));
f4f51050 3172 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3173 break;
3174 case KVM_IRQCHIP_PIC_SLAVE:
f4f51050 3175 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3176 memcpy(&pic_irqchip(kvm)->pics[1],
3177 &chip->chip.pic,
3178 sizeof(struct kvm_pic_state));
f4f51050 3179 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3180 break;
3181 case KVM_IRQCHIP_IOAPIC:
eba0226b 3182 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3183 break;
3184 default:
3185 r = -EINVAL;
3186 break;
3187 }
3188 kvm_pic_update_irq(pic_irqchip(kvm));
3189 return r;
3190}
3191
e0f63cb9
SY		 3192static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3193{
3194 int r = 0;
3195
894a9c55 3196 mutex_lock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9 3197 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
894a9c55 3198 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9
SY		 3199 return r;
3200}
3201
3202static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3203{
3204 int r = 0;
3205
894a9c55 3206 mutex_lock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9 3207 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
e9f42757
BK		 3208 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3209 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3210 return r;
3211}
3212
3213static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3214{
3215 int r = 0;
3216
3217 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3218 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3219 sizeof(ps->channels));
3220 ps->flags = kvm->arch.vpit->pit_state.flags;
3221 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
97e69aa6 3222 memset(&ps->reserved, 0, sizeof(ps->reserved));
e9f42757
BK		 3223 return r;
3224}
3225
3226static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3227{
3228 int r = 0, start = 0;
3229 u32 prev_legacy, cur_legacy;
3230 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3231 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3232 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3233 if (!prev_legacy && cur_legacy)
3234 start = 1;
3235 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3236 sizeof(kvm->arch.vpit->pit_state.channels));
3237 kvm->arch.vpit->pit_state.flags = ps->flags;
3238 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
894a9c55 3239 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9
SY		 3240 return r;
3241}
3242
52d939a0
MT		 3243static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3244 struct kvm_reinject_control *control)
3245{
3246 if (!kvm->arch.vpit)
3247 return -ENXIO;
894a9c55 3248 mutex_lock(&kvm->arch.vpit->pit_state.lock);
52d939a0 3249 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
894a9c55 3250 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
52d939a0
MT		 3251 return 0;
3252}
3253
5bb064dc
ZX		 3254/*
3255 * Get (and clear) the dirty memory log for a memory slot.
3256 */
3257int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3258 struct kvm_dirty_log *log)
3259{
87bf6e7d 3260 int r, i;
5bb064dc 3261 struct kvm_memory_slot *memslot;
87bf6e7d 3262 unsigned long n;
b050b015 3263 unsigned long is_dirty = 0;
5bb064dc 3264
79fac95e 3265 mutex_lock(&kvm->slots_lock);
5bb064dc 3266
b050b015
MT		 3267 r = -EINVAL;
3268 if (log->slot >= KVM_MEMORY_SLOTS)
3269 goto out;
3270
3271 memslot = &kvm->memslots->memslots[log->slot];
3272 r = -ENOENT;
3273 if (!memslot->dirty_bitmap)
3274 goto out;
3275
87bf6e7d 3276 n = kvm_dirty_bitmap_bytes(memslot);
b050b015 3277
b050b015
MT		 3278 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3279 is_dirty = memslot->dirty_bitmap[i];
5bb064dc
ZX		 3280
3281 /* If nothing is dirty, don't bother messing with page tables. */
3282 if (is_dirty) {
b050b015 3283 struct kvm_memslots *slots, *old_slots;
914ebccd 3284 unsigned long *dirty_bitmap;
b050b015 3285
515a0127
TY		 3286 dirty_bitmap = memslot->dirty_bitmap_head;
3287 if (memslot->dirty_bitmap == dirty_bitmap)
3288 dirty_bitmap += n / sizeof(long);
914ebccd 3289 memset(dirty_bitmap, 0, n);
b050b015 3290
914ebccd
TY		 3291 r = -ENOMEM;
3292 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
515a0127 3293 if (!slots)
914ebccd 3294 goto out;
b050b015
MT		 3295 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3296 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
49c7754c 3297 slots->generation++;
b050b015
MT		 3298
3299 old_slots = kvm->memslots;
3300 rcu_assign_pointer(kvm->memslots, slots);
3301 synchronize_srcu_expedited(&kvm->srcu);
3302 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3303 kfree(old_slots);
914ebccd 3304
edde99ce
MT		 3305 spin_lock(&kvm->mmu_lock);
3306 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3307 spin_unlock(&kvm->mmu_lock);
3308
914ebccd 3309 r = -EFAULT;
515a0127 3310 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
914ebccd 3311 goto out;
914ebccd
TY		 3312 } else {
3313 r = -EFAULT;
3314 if (clear_user(log->dirty_bitmap, n))
3315 goto out;
5bb064dc 3316 }
b050b015 3317
5bb064dc
ZX		 3318 r = 0;
3319out:
79fac95e 3320 mutex_unlock(&kvm->slots_lock);
5bb064dc
ZX		 3321 return r;
3322}
3323
1fe779f8
CO		 3324long kvm_arch_vm_ioctl(struct file *filp,
3325 unsigned int ioctl, unsigned long arg)
3326{
3327 struct kvm *kvm = filp->private_data;
3328 void __user *argp = (void __user *)arg;
367e1319 3329 int r = -ENOTTY;
f0d66275
DH		 3330 /*
3331 * This union makes it completely explicit to gcc-3.x
3332 * that these two variables' stack usage should be
3333 * combined, not added together.
3334 */
3335 union {
3336 struct kvm_pit_state ps;
e9f42757 3337 struct kvm_pit_state2 ps2;
c5ff41ce 3338 struct kvm_pit_config pit_config;
f0d66275 3339 } u;
1fe779f8
CO		 3340
3341 switch (ioctl) {
3342 case KVM_SET_TSS_ADDR:
3343 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3344 if (r < 0)
3345 goto out;
3346 break;
b927a3ce
SY		 3347 case KVM_SET_IDENTITY_MAP_ADDR: {
3348 u64 ident_addr;
3349
3350 r = -EFAULT;
3351 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3352 goto out;
3353 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3354 if (r < 0)
3355 goto out;
3356 break;
3357 }
1fe779f8
CO		 3358 case KVM_SET_NR_MMU_PAGES:
3359 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3360 if (r)
3361 goto out;
3362 break;
3363 case KVM_GET_NR_MMU_PAGES:
3364 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3365 break;
3ddea128
MT		 3366 case KVM_CREATE_IRQCHIP: {
3367 struct kvm_pic *vpic;
3368
3369 mutex_lock(&kvm->lock);
3370 r = -EEXIST;
3371 if (kvm->arch.vpic)
3372 goto create_irqchip_unlock;
1fe779f8 3373 r = -ENOMEM;
3ddea128
MT		 3374 vpic = kvm_create_pic(kvm);
3375 if (vpic) {
1fe779f8
CO		 3376 r = kvm_ioapic_init(kvm);
3377 if (r) {
175504cd 3378 mutex_lock(&kvm->slots_lock);
72bb2fcd
WY		 3379 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3380 &vpic->dev);
175504cd 3381 mutex_unlock(&kvm->slots_lock);
3ddea128
MT		 3382 kfree(vpic);
3383 goto create_irqchip_unlock;
1fe779f8
CO		 3384 }
3385 } else
3ddea128
MT		 3386 goto create_irqchip_unlock;
3387 smp_wmb();
3388 kvm->arch.vpic = vpic;
3389 smp_wmb();
399ec807
AK		 3390 r = kvm_setup_default_irq_routing(kvm);
3391 if (r) {
175504cd 3392 mutex_lock(&kvm->slots_lock);
3ddea128 3393 mutex_lock(&kvm->irq_lock);
72bb2fcd
WY		 3394 kvm_ioapic_destroy(kvm);
3395 kvm_destroy_pic(kvm);
3ddea128 3396 mutex_unlock(&kvm->irq_lock);
175504cd 3397 mutex_unlock(&kvm->slots_lock);
399ec807 3398 }
3ddea128
MT		 3399 create_irqchip_unlock:
3400 mutex_unlock(&kvm->lock);
1fe779f8 3401 break;
3ddea128 3402 }
7837699f 3403 case KVM_CREATE_PIT:
c5ff41ce
JK		 3404 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3405 goto create_pit;
3406 case KVM_CREATE_PIT2:
3407 r = -EFAULT;
3408 if (copy_from_user(&u.pit_config, argp,
3409 sizeof(struct kvm_pit_config)))
3410 goto out;
3411 create_pit:
79fac95e 3412 mutex_lock(&kvm->slots_lock);
269e05e4
AK		 3413 r = -EEXIST;
3414 if (kvm->arch.vpit)
3415 goto create_pit_unlock;
7837699f 3416 r = -ENOMEM;
c5ff41ce 3417 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
7837699f
SY		 3418 if (kvm->arch.vpit)
3419 r = 0;
269e05e4 3420 create_pit_unlock:
79fac95e 3421 mutex_unlock(&kvm->slots_lock);
7837699f 3422 break;
4925663a 3423 case KVM_IRQ_LINE_STATUS:
1fe779f8
CO		 3424 case KVM_IRQ_LINE: {
3425 struct kvm_irq_level irq_event;
3426
3427 r = -EFAULT;
3428 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3429 goto out;
160d2f6c 3430 r = -ENXIO;
1fe779f8 3431 if (irqchip_in_kernel(kvm)) {
4925663a 3432 __s32 status;
4925663a
GN		 3433 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3434 irq_event.irq, irq_event.level);
4925663a 3435 if (ioctl == KVM_IRQ_LINE_STATUS) {
160d2f6c 3436 r = -EFAULT;
4925663a
GN		 3437 irq_event.status = status;
3438 if (copy_to_user(argp, &irq_event,
3439 sizeof irq_event))
3440 goto out;
3441 }
1fe779f8
CO		 3442 r = 0;
3443 }
3444 break;
3445 }
3446 case KVM_GET_IRQCHIP: {
3447 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
f0d66275 3448 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1fe779f8 3449
f0d66275
DH		 3450 r = -ENOMEM;
3451 if (!chip)
1fe779f8 3452 goto out;
f0d66275
DH		 3453 r = -EFAULT;
3454 if (copy_from_user(chip, argp, sizeof *chip))
3455 goto get_irqchip_out;
1fe779f8
CO		 3456 r = -ENXIO;
3457 if (!irqchip_in_kernel(kvm))
f0d66275
DH		 3458 goto get_irqchip_out;
3459 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1fe779f8 3460 if (r)
f0d66275 3461 goto get_irqchip_out;
1fe779f8 3462 r = -EFAULT;
f0d66275
DH		 3463 if (copy_to_user(argp, chip, sizeof *chip))
3464 goto get_irqchip_out;
1fe779f8 3465 r = 0;
f0d66275
DH		 3466 get_irqchip_out:
3467 kfree(chip);
3468 if (r)
3469 goto out;
1fe779f8
CO		 3470 break;
3471 }
3472 case KVM_SET_IRQCHIP: {
3473 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
f0d66275 3474 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1fe779f8 3475
f0d66275
DH		 3476 r = -ENOMEM;
3477 if (!chip)
1fe779f8 3478 goto out;
f0d66275
DH		 3479 r = -EFAULT;
3480 if (copy_from_user(chip, argp, sizeof *chip))
3481 goto set_irqchip_out;
1fe779f8
CO		 3482 r = -ENXIO;
3483 if (!irqchip_in_kernel(kvm))
f0d66275
DH		 3484 goto set_irqchip_out;
3485 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1fe779f8 3486 if (r)
f0d66275 3487 goto set_irqchip_out;
1fe779f8 3488 r = 0;
f0d66275
DH		 3489 set_irqchip_out:
3490 kfree(chip);
3491 if (r)
3492 goto out;
1fe779f8
CO		 3493 break;
3494 }
e0f63cb9 3495 case KVM_GET_PIT: {
e0f63cb9 3496 r = -EFAULT;
f0d66275 3497 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 3498 goto out;
3499 r = -ENXIO;
3500 if (!kvm->arch.vpit)
3501 goto out;
f0d66275 3502 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
e0f63cb9
SY		 3503 if (r)
3504 goto out;
3505 r = -EFAULT;
f0d66275 3506 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 3507 goto out;
3508 r = 0;
3509 break;
3510 }
3511 case KVM_SET_PIT: {
e0f63cb9 3512 r = -EFAULT;
f0d66275 3513 if (copy_from_user(&u.ps, argp, sizeof u.ps))
e0f63cb9
SY		 3514 goto out;
3515 r = -ENXIO;
3516 if (!kvm->arch.vpit)
3517 goto out;
f0d66275 3518 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
e0f63cb9
SY		 3519 if (r)
3520 goto out;
3521 r = 0;
3522 break;
3523 }
e9f42757
BK		 3524 case KVM_GET_PIT2: {
3525 r = -ENXIO;
3526 if (!kvm->arch.vpit)
3527 goto out;
3528 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3529 if (r)
3530 goto out;
3531 r = -EFAULT;
3532 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3533 goto out;
3534 r = 0;
3535 break;
3536 }
3537 case KVM_SET_PIT2: {
3538 r = -EFAULT;
3539 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3540 goto out;
3541 r = -ENXIO;
3542 if (!kvm->arch.vpit)
3543 goto out;
3544 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3545 if (r)
3546 goto out;
3547 r = 0;
3548 break;
3549 }
52d939a0
MT		 3550 case KVM_REINJECT_CONTROL: {
3551 struct kvm_reinject_control control;
3552 r = -EFAULT;
3553 if (copy_from_user(&control, argp, sizeof(control)))
3554 goto out;
3555 r = kvm_vm_ioctl_reinject(kvm, &control);
3556 if (r)
3557 goto out;
3558 r = 0;
3559 break;
3560 }
ffde22ac
ES		 3561 case KVM_XEN_HVM_CONFIG: {
3562 r = -EFAULT;
3563 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3564 sizeof(struct kvm_xen_hvm_config)))
3565 goto out;
3566 r = -EINVAL;
3567 if (kvm->arch.xen_hvm_config.flags)
3568 goto out;
3569 r = 0;
3570 break;
3571 }
afbcf7ab 3572 case KVM_SET_CLOCK: {
afbcf7ab
GC		 3573 struct kvm_clock_data user_ns;
3574 u64 now_ns;
3575 s64 delta;
3576
3577 r = -EFAULT;
3578 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3579 goto out;
3580
3581 r = -EINVAL;
3582 if (user_ns.flags)
3583 goto out;
3584
3585 r = 0;
395c6b0a 3586 local_irq_disable();
759379dd 3587 now_ns = get_kernel_ns();
afbcf7ab 3588 delta = user_ns.clock - now_ns;
395c6b0a 3589 local_irq_enable();
afbcf7ab
GC		 3590 kvm->arch.kvmclock_offset = delta;
3591 break;
3592 }
3593 case KVM_GET_CLOCK: {
afbcf7ab
GC		 3594 struct kvm_clock_data user_ns;
3595 u64 now_ns;
3596
395c6b0a 3597 local_irq_disable();
759379dd 3598 now_ns = get_kernel_ns();
afbcf7ab 3599 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
395c6b0a 3600 local_irq_enable();
afbcf7ab 3601 user_ns.flags = 0;
97e69aa6 3602 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
afbcf7ab
GC		 3603
3604 r = -EFAULT;
3605 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3606 goto out;
3607 r = 0;
3608 break;
3609 }
3610
1fe779f8
CO		 3611 default:
3612 ;
3613 }
3614out:
3615 return r;
3616}
3617
a16b043c 3618static void kvm_init_msr_list(void)
043405e1
CO		 3619{
3620 u32 dummy[2];
3621 unsigned i, j;
3622
e3267cbb
GC		 3623 /* skip the first msrs in the list. KVM-specific */
3624 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
043405e1
CO		 3625 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3626 continue;
3627 if (j < i)
3628 msrs_to_save[j] = msrs_to_save[i];
3629 j++;
3630 }
3631 num_msrs_to_save = j;
3632}
3633
bda9020e
MT		 3634static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3635 const void *v)
bbd9b64e 3636{
70252a10
AK		 3637 int handled = 0;
3638 int n;
3639
3640 do {
3641 n = min(len, 8);
3642 if (!(vcpu->arch.apic &&
3643 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3644 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3645 break;
3646 handled += n;
3647 addr += n;
3648 len -= n;
3649 v += n;
3650 } while (len);
bbd9b64e 3651
70252a10 3652 return handled;
bbd9b64e
CO		 3653}
3654
bda9020e 3655static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
bbd9b64e 3656{
70252a10
AK		 3657 int handled = 0;
3658 int n;
3659
3660 do {
3661 n = min(len, 8);
3662 if (!(vcpu->arch.apic &&
3663 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3664 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3665 break;
3666 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3667 handled += n;
3668 addr += n;
3669 len -= n;
3670 v += n;
3671 } while (len);
bbd9b64e 3672
70252a10 3673 return handled;
bbd9b64e
CO		 3674}
3675
2dafc6c2
GN		 3676static void kvm_set_segment(struct kvm_vcpu *vcpu,
3677 struct kvm_segment *var, int seg)
3678{
3679 kvm_x86_ops->set_segment(vcpu, var, seg);
3680}
3681
3682void kvm_get_segment(struct kvm_vcpu *vcpu,
3683 struct kvm_segment *var, int seg)
3684{
3685 kvm_x86_ops->get_segment(vcpu, var, seg);
3686}
3687
c30a358d
JR		 3688static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3689{
3690 return gpa;
3691}
3692
02f59dc9
JR		 3693static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3694{
3695 gpa_t t_gpa;
ab9ae313 3696 struct x86_exception exception;
02f59dc9
JR		 3697
3698 BUG_ON(!mmu_is_nested(vcpu));
3699
3700 /* NPT walks are always user-walks */
3701 access |= PFERR_USER_MASK;
ab9ae313 3702 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
02f59dc9
JR		 3703
3704 return t_gpa;
3705}
3706
ab9ae313
AK		 3707gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3708 struct x86_exception *exception)
1871c602
GN		 3709{
3710 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
ab9ae313 3711 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 3712}
3713
ab9ae313
AK		 3714 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3715 struct x86_exception *exception)
1871c602
GN		 3716{
3717 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3718 access |= PFERR_FETCH_MASK;
ab9ae313 3719 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 3720}
3721
ab9ae313
AK		 3722gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3723 struct x86_exception *exception)
1871c602
GN		 3724{
3725 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3726 access |= PFERR_WRITE_MASK;
ab9ae313 3727 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 3728}
3729
3730/* uses this to access any guest's mapped memory without checking CPL */
ab9ae313
AK		 3731gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3732 struct x86_exception *exception)
1871c602 3733{
ab9ae313 3734 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
1871c602
GN		 3735}
3736
3737static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3738 struct kvm_vcpu *vcpu, u32 access,
bcc55cba 3739 struct x86_exception *exception)
bbd9b64e
CO		 3740{
3741 void *data = val;
10589a46 3742 int r = X86EMUL_CONTINUE;
bbd9b64e
CO		 3743
3744 while (bytes) {
14dfe855 3745 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
ab9ae313 3746 exception);
bbd9b64e 3747 unsigned offset = addr & (PAGE_SIZE-1);
77c2002e 3748 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
bbd9b64e
CO		 3749 int ret;
3750
bcc55cba 3751 if (gpa == UNMAPPED_GVA)
ab9ae313 3752 return X86EMUL_PROPAGATE_FAULT;
77c2002e 3753 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
10589a46 3754 if (ret < 0) {
c3cd7ffa 3755 r = X86EMUL_IO_NEEDED;
10589a46
MT		 3756 goto out;
3757 }
bbd9b64e 3758
77c2002e
IE		 3759 bytes -= toread;
3760 data += toread;
3761 addr += toread;
bbd9b64e 3762 }
10589a46 3763out:
10589a46 3764 return r;
bbd9b64e 3765}
77c2002e 3766
1871c602 3767/* used for instruction fetching */
0f65dd70
AK		 3768static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3769 gva_t addr, void *val, unsigned int bytes,
bcc55cba 3770 struct x86_exception *exception)
1871c602 3771{
0f65dd70 3772 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 3773 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
0f65dd70 3774
1871c602 3775 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
bcc55cba
AK		 3776 access | PFERR_FETCH_MASK,
3777 exception);
1871c602
GN		 3778}
3779
0f65dd70
AK		 3780static int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3781 gva_t addr, void *val, unsigned int bytes,
bcc55cba 3782 struct x86_exception *exception)
1871c602 3783{
0f65dd70 3784 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 3785 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
0f65dd70 3786
1871c602 3787 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
bcc55cba 3788 exception);
1871c602
GN		 3789}
3790
0f65dd70
AK		 3791static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3792 gva_t addr, void *val, unsigned int bytes,
bcc55cba 3793 struct x86_exception *exception)
1871c602 3794{
0f65dd70 3795 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
bcc55cba 3796 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
1871c602
GN		 3797}
3798
0f65dd70
AK		 3799static int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3800 gva_t addr, void *val,
2dafc6c2 3801 unsigned int bytes,
bcc55cba 3802 struct x86_exception *exception)
77c2002e 3803{
0f65dd70 3804 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
77c2002e
IE		 3805 void *data = val;
3806 int r = X86EMUL_CONTINUE;
3807
3808 while (bytes) {
14dfe855
JR		 3809 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3810 PFERR_WRITE_MASK,
ab9ae313 3811 exception);
77c2002e
IE		 3812 unsigned offset = addr & (PAGE_SIZE-1);
3813 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3814 int ret;
3815
bcc55cba 3816 if (gpa == UNMAPPED_GVA)
ab9ae313 3817 return X86EMUL_PROPAGATE_FAULT;
77c2002e
IE		 3818 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3819 if (ret < 0) {
c3cd7ffa 3820 r = X86EMUL_IO_NEEDED;
77c2002e
IE		 3821 goto out;
3822 }
3823
3824 bytes -= towrite;
3825 data += towrite;
3826 addr += towrite;
3827 }
3828out:
3829 return r;
3830}
3831
0f65dd70
AK		 3832static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
3833 unsigned long addr,
bbd9b64e
CO		 3834 void *val,
3835 unsigned int bytes,
0f65dd70 3836 struct x86_exception *exception)
bbd9b64e 3837{
0f65dd70 3838 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
bbd9b64e 3839 gpa_t gpa;
70252a10 3840 int handled;
bbd9b64e
CO		 3841
3842 if (vcpu->mmio_read_completed) {
3843 memcpy(val, vcpu->mmio_data, bytes);
aec51dc4
AK		 3844 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3845 vcpu->mmio_phys_addr, *(u64 *)val);
bbd9b64e
CO		 3846 vcpu->mmio_read_completed = 0;
3847 return X86EMUL_CONTINUE;
3848 }
3849
ab9ae313 3850 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, exception);
1871c602 3851
8fe681e9 3852 if (gpa == UNMAPPED_GVA)
1871c602 3853 return X86EMUL_PROPAGATE_FAULT;
bbd9b64e
CO		 3854
3855 /* For APIC access vmexit */
3856 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3857 goto mmio;
3858
0f65dd70 3859 if (kvm_read_guest_virt(ctxt, addr, val, bytes, exception)
bcc55cba 3860 == X86EMUL_CONTINUE)
bbd9b64e 3861 return X86EMUL_CONTINUE;
bbd9b64e
CO		 3862
3863mmio:
3864 /*
3865 * Is this MMIO handled locally?
3866 */
70252a10
AK		 3867 handled = vcpu_mmio_read(vcpu, gpa, bytes, val);
3868
3869 if (handled == bytes)
bbd9b64e 3870 return X86EMUL_CONTINUE;
70252a10
AK		 3871
3872 gpa += handled;
3873 bytes -= handled;
3874 val += handled;
aec51dc4
AK		 3875
3876 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
bbd9b64e
CO		 3877
3878 vcpu->mmio_needed = 1;
411c35b7
GN		 3879 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3880 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
cef4dea0
AK		 3881 vcpu->mmio_size = bytes;
3882 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
411c35b7 3883 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
cef4dea0 3884 vcpu->mmio_index = 0;
bbd9b64e 3885
c3cd7ffa 3886 return X86EMUL_IO_NEEDED;
bbd9b64e
CO		 3887}
3888
3200f405 3889int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
bcc55cba 3890 const void *val, int bytes)
bbd9b64e
CO		 3891{
3892 int ret;
3893
3894 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
9f811285 3895 if (ret < 0)
bbd9b64e 3896 return 0;
ad218f85 3897 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
bbd9b64e
CO		 3898 return 1;
3899}
3900
3901static int emulator_write_emulated_onepage(unsigned long addr,
3902 const void *val,
3903 unsigned int bytes,
bcc55cba 3904 struct x86_exception *exception,
bbd9b64e
CO		 3905 struct kvm_vcpu *vcpu)
3906{
10589a46 3907 gpa_t gpa;
70252a10 3908 int handled;
10589a46 3909
ab9ae313 3910 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, exception);
bbd9b64e 3911
8fe681e9 3912 if (gpa == UNMAPPED_GVA)
bbd9b64e 3913 return X86EMUL_PROPAGATE_FAULT;
bbd9b64e
CO		 3914
3915 /* For APIC access vmexit */
3916 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3917 goto mmio;
3918
3919 if (emulator_write_phys(vcpu, gpa, val, bytes))
3920 return X86EMUL_CONTINUE;
3921
3922mmio:
aec51dc4 3923 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
bbd9b64e
CO		 3924 /*
3925 * Is this MMIO handled locally?
3926 */
70252a10
AK		 3927 handled = vcpu_mmio_write(vcpu, gpa, bytes, val);
3928 if (handled == bytes)
bbd9b64e 3929 return X86EMUL_CONTINUE;
bbd9b64e 3930
70252a10
AK		 3931 gpa += handled;
3932 bytes -= handled;
3933 val += handled;
3934
bbd9b64e 3935 vcpu->mmio_needed = 1;
cef4dea0 3936 memcpy(vcpu->mmio_data, val, bytes);
411c35b7
GN		 3937 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3938 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
cef4dea0
AK		 3939 vcpu->mmio_size = bytes;
3940 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
411c35b7 3941 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
cef4dea0
AK		 3942 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3943 vcpu->mmio_index = 0;
bbd9b64e
CO		 3944
3945 return X86EMUL_CONTINUE;
3946}
3947
0f65dd70
AK		 3948int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
3949 unsigned long addr,
8f6abd06
GN		 3950 const void *val,
3951 unsigned int bytes,
0f65dd70 3952 struct x86_exception *exception)
bbd9b64e 3953{
0f65dd70
AK		 3954 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3955
bbd9b64e
CO		 3956 /* Crossing a page boundary? */
3957 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3958 int rc, now;
3959
3960 now = -addr & ~PAGE_MASK;
bcc55cba 3961 rc = emulator_write_emulated_onepage(addr, val, now, exception,
8fe681e9 3962 vcpu);
bbd9b64e
CO		 3963 if (rc != X86EMUL_CONTINUE)
3964 return rc;
3965 addr += now;
3966 val += now;
3967 bytes -= now;
3968 }
bcc55cba 3969 return emulator_write_emulated_onepage(addr, val, bytes, exception,
8fe681e9 3970 vcpu);
bbd9b64e 3971}
bbd9b64e 3972
daea3e73
AK		 3973#define CMPXCHG_TYPE(t, ptr, old, new) \
3974 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3975
3976#ifdef CONFIG_X86_64
3977# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3978#else
3979# define CMPXCHG64(ptr, old, new) \
9749a6c0 3980 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
daea3e73
AK		 3981#endif
3982
0f65dd70
AK		 3983static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
3984 unsigned long addr,
bbd9b64e
CO		 3985 const void *old,
3986 const void *new,
3987 unsigned int bytes,
0f65dd70 3988 struct x86_exception *exception)
bbd9b64e 3989{
0f65dd70 3990 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
daea3e73
AK		 3991 gpa_t gpa;
3992 struct page *page;
3993 char *kaddr;
3994 bool exchanged;
2bacc55c 3995
daea3e73
AK		 3996 /* guests cmpxchg8b have to be emulated atomically */
3997 if (bytes > 8 || (bytes & (bytes - 1)))
3998 goto emul_write;
10589a46 3999
daea3e73 4000 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
2bacc55c 4001
daea3e73
AK		 4002 if (gpa == UNMAPPED_GVA ||
4003 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4004 goto emul_write;
2bacc55c 4005
daea3e73
AK		 4006 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4007 goto emul_write;
72dc67a6 4008
daea3e73 4009 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
c19b8bd6
WY		 4010 if (is_error_page(page)) {
4011 kvm_release_page_clean(page);
4012 goto emul_write;
4013 }
72dc67a6 4014
daea3e73
AK		 4015 kaddr = kmap_atomic(page, KM_USER0);
4016 kaddr += offset_in_page(gpa);
4017 switch (bytes) {
4018 case 1:
4019 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4020 break;
4021 case 2:
4022 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4023 break;
4024 case 4:
4025 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4026 break;
4027 case 8:
4028 exchanged = CMPXCHG64(kaddr, old, new);
4029 break;
4030 default:
4031 BUG();
2bacc55c 4032 }
daea3e73
AK		 4033 kunmap_atomic(kaddr, KM_USER0);
4034 kvm_release_page_dirty(page);
4035
4036 if (!exchanged)
4037 return X86EMUL_CMPXCHG_FAILED;
4038
8f6abd06
GN		 4039 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4040
4041 return X86EMUL_CONTINUE;
4a5f48f6 4042
3200f405 4043emul_write:
daea3e73 4044 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2bacc55c 4045
0f65dd70 4046 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
bbd9b64e
CO		 4047}
4048
cf8f70bf
GN		 4049static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4050{
4051 /* TODO: String I/O for in kernel device */
4052 int r;
4053
4054 if (vcpu->arch.pio.in)
4055 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4056 vcpu->arch.pio.size, pd);
4057 else
4058 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4059 vcpu->arch.pio.port, vcpu->arch.pio.size,
4060 pd);
4061 return r;
4062}
4063
4064
ca1d4a9e
AK		 4065static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4066 int size, unsigned short port, void *val,
4067 unsigned int count)
cf8f70bf 4068{
ca1d4a9e
AK		 4069 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4070
7972995b 4071 if (vcpu->arch.pio.count)
cf8f70bf
GN		 4072 goto data_avail;
4073
61cfab2e 4074 trace_kvm_pio(0, port, size, count);
cf8f70bf
GN		 4075
4076 vcpu->arch.pio.port = port;
4077 vcpu->arch.pio.in = 1;
7972995b 4078 vcpu->arch.pio.count = count;
cf8f70bf
GN		 4079 vcpu->arch.pio.size = size;
4080
4081 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4082 data_avail:
4083 memcpy(val, vcpu->arch.pio_data, size * count);
7972995b 4084 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4085 return 1;
4086 }
4087
4088 vcpu->run->exit_reason = KVM_EXIT_IO;
4089 vcpu->run->io.direction = KVM_EXIT_IO_IN;
4090 vcpu->run->io.size = size;
4091 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4092 vcpu->run->io.count = count;
4093 vcpu->run->io.port = port;
4094
4095 return 0;
4096}
4097
ca1d4a9e
AK		 4098static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4099 int size, unsigned short port,
4100 const void *val, unsigned int count)
cf8f70bf 4101{
ca1d4a9e
AK		 4102 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4103
61cfab2e 4104 trace_kvm_pio(1, port, size, count);
cf8f70bf
GN		 4105
4106 vcpu->arch.pio.port = port;
4107 vcpu->arch.pio.in = 0;
7972995b 4108 vcpu->arch.pio.count = count;
cf8f70bf
GN		 4109 vcpu->arch.pio.size = size;
4110
4111 memcpy(vcpu->arch.pio_data, val, size * count);
4112
4113 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
7972995b 4114 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4115 return 1;
4116 }
4117
4118 vcpu->run->exit_reason = KVM_EXIT_IO;
4119 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4120 vcpu->run->io.size = size;
4121 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4122 vcpu->run->io.count = count;
4123 vcpu->run->io.port = port;
4124
4125 return 0;
4126}
4127
bbd9b64e
CO		 4128static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4129{
4130 return kvm_x86_ops->get_segment_base(vcpu, seg);
4131}
4132
3cb16fe7 4133static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
bbd9b64e 4134{
3cb16fe7 4135 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
bbd9b64e
CO		 4136}
4137
f5f48ee1
SY		 4138int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4139{
4140 if (!need_emulate_wbinvd(vcpu))
4141 return X86EMUL_CONTINUE;
4142
4143 if (kvm_x86_ops->has_wbinvd_exit()) {
2eec7343
JK		 4144 int cpu = get_cpu();
4145
4146 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
f5f48ee1
SY		 4147 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4148 wbinvd_ipi, NULL, 1);
2eec7343 4149 put_cpu();
f5f48ee1 4150 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
2eec7343
JK		 4151 } else
4152 wbinvd();
f5f48ee1
SY		 4153 return X86EMUL_CONTINUE;
4154}
4155EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4156
bcaf5cc5
AK		 4157static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4158{
4159 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4160}
4161
717746e3 4162int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
bbd9b64e 4163{
717746e3 4164 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
bbd9b64e
CO		 4165}
4166
717746e3 4167int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
bbd9b64e 4168{
338dbc97 4169
717746e3 4170 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
bbd9b64e
CO		 4171}
4172
52a46617 4173static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5fdbf976 4174{
52a46617 4175 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5fdbf976
MT		 4176}
4177
717746e3 4178static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
bbd9b64e 4179{
717746e3 4180 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
52a46617
GN		 4181 unsigned long value;
4182
4183 switch (cr) {
4184 case 0:
4185 value = kvm_read_cr0(vcpu);
4186 break;
4187 case 2:
4188 value = vcpu->arch.cr2;
4189 break;
4190 case 3:
9f8fe504 4191 value = kvm_read_cr3(vcpu);
52a46617
GN		 4192 break;
4193 case 4:
4194 value = kvm_read_cr4(vcpu);
4195 break;
4196 case 8:
4197 value = kvm_get_cr8(vcpu);
4198 break;
4199 default:
4200 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4201 return 0;
4202 }
4203
4204 return value;
4205}
4206
717746e3 4207static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
52a46617 4208{
717746e3 4209 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
0f12244f
GN		 4210 int res = 0;
4211
52a46617
GN		 4212 switch (cr) {
4213 case 0:
49a9b07e 4214 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
52a46617
GN		 4215 break;
4216 case 2:
4217 vcpu->arch.cr2 = val;
4218 break;
4219 case 3:
2390218b 4220 res = kvm_set_cr3(vcpu, val);
52a46617
GN		 4221 break;
4222 case 4:
a83b29c6 4223 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
52a46617
GN		 4224 break;
4225 case 8:
eea1cff9 4226 res = kvm_set_cr8(vcpu, val);
52a46617
GN		 4227 break;
4228 default:
4229 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
0f12244f 4230 res = -1;
52a46617 4231 }
0f12244f
GN		 4232
4233 return res;
52a46617
GN		 4234}
4235
717746e3 4236static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
9c537244 4237{
717746e3 4238 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
9c537244
GN		 4239}
4240
4bff1e86 4241static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
2dafc6c2 4242{
4bff1e86 4243 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
2dafc6c2
GN		 4244}
4245
4bff1e86 4246static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
160ce1f1 4247{
4bff1e86 4248 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
160ce1f1
MG		 4249}
4250
1ac9d0cf
AK		 4251static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4252{
4253 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4254}
4255
4256static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4257{
4258 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4259}
4260
4bff1e86
AK		 4261static unsigned long emulator_get_cached_segment_base(
4262 struct x86_emulate_ctxt *ctxt, int seg)
5951c442 4263{
4bff1e86 4264 return get_segment_base(emul_to_vcpu(ctxt), seg);
5951c442
GN		 4265}
4266
4bff1e86
AK		 4267static bool emulator_get_cached_descriptor(struct x86_emulate_ctxt *ctxt,
4268 struct desc_struct *desc, u32 *base3,
4269 int seg)
2dafc6c2
GN		 4270{
4271 struct kvm_segment var;
4272
4bff1e86 4273 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
2dafc6c2
GN		 4274
4275 if (var.unusable)
4276 return false;
4277
4278 if (var.g)
4279 var.limit >>= 12;
4280 set_desc_limit(desc, var.limit);
4281 set_desc_base(desc, (unsigned long)var.base);
5601d05b
GN		 4282#ifdef CONFIG_X86_64
4283 if (base3)
4284 *base3 = var.base >> 32;
4285#endif
2dafc6c2
GN		 4286 desc->type = var.type;
4287 desc->s = var.s;
4288 desc->dpl = var.dpl;
4289 desc->p = var.present;
4290 desc->avl = var.avl;
4291 desc->l = var.l;
4292 desc->d = var.db;
4293 desc->g = var.g;
4294
4295 return true;
4296}
4297
4bff1e86
AK		 4298static void emulator_set_cached_descriptor(struct x86_emulate_ctxt *ctxt,
4299 struct desc_struct *desc, u32 base3,
4300 int seg)
2dafc6c2 4301{
4bff1e86 4302 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
2dafc6c2
GN		 4303 struct kvm_segment var;
4304
4305 /* needed to preserve selector */
4306 kvm_get_segment(vcpu, &var, seg);
4307
4308 var.base = get_desc_base(desc);
5601d05b
GN		 4309#ifdef CONFIG_X86_64
4310 var.base |= ((u64)base3) << 32;
4311#endif
2dafc6c2
GN		 4312 var.limit = get_desc_limit(desc);
4313 if (desc->g)
4314 var.limit = (var.limit << 12) | 0xfff;
4315 var.type = desc->type;
4316 var.present = desc->p;
4317 var.dpl = desc->dpl;
4318 var.db = desc->d;
4319 var.s = desc->s;
4320 var.l = desc->l;
4321 var.g = desc->g;
4322 var.avl = desc->avl;
4323 var.present = desc->p;
4324 var.unusable = !var.present;
4325 var.padding = 0;
4326
4327 kvm_set_segment(vcpu, &var, seg);
4328 return;
4329}
4330
4bff1e86 4331static u16 emulator_get_segment_selector(struct x86_emulate_ctxt *ctxt, int seg)
2dafc6c2
GN		 4332{
4333 struct kvm_segment kvm_seg;
4334
4bff1e86 4335 kvm_get_segment(emul_to_vcpu(ctxt), &kvm_seg, seg);
2dafc6c2
GN		 4336 return kvm_seg.selector;
4337}
4338
4bff1e86
AK		 4339static void emulator_set_segment_selector(struct x86_emulate_ctxt *ctxt,
4340 u16 sel, int seg)
2dafc6c2
GN		 4341{
4342 struct kvm_segment kvm_seg;
4343
4bff1e86 4344 kvm_get_segment(emul_to_vcpu(ctxt), &kvm_seg, seg);
2dafc6c2 4345 kvm_seg.selector = sel;
4bff1e86 4346 kvm_set_segment(emul_to_vcpu(ctxt), &kvm_seg, seg);
2dafc6c2
GN		 4347}
4348
717746e3
AK		 4349static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4350 u32 msr_index, u64 *pdata)
4351{
4352 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4353}
4354
4355static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4356 u32 msr_index, u64 data)
4357{
4358 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4359}
4360
6c3287f7
AK		 4361static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4362{
4363 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4364}
4365
5037f6f3
AK		 4366static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4367{
4368 preempt_disable();
4369 kvm_load_guest_fpu(ctxt->vcpu);
4370 /*
4371 * CR0.TS may reference the host fpu state, not the guest fpu state,
4372 * so it may be clear at this point.
4373 */
4374 clts();
4375}
4376
4377static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4378{
4379 preempt_enable();
4380}
4381
2953538e 4382static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
8a76d7f2 4383 struct x86_instruction_info *info,
c4f035c6
AK		 4384 enum x86_intercept_stage stage)
4385{
2953538e 4386 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
c4f035c6
AK		 4387}
4388
14af3f3c 4389static struct x86_emulate_ops emulate_ops = {
1871c602 4390 .read_std = kvm_read_guest_virt_system,
2dafc6c2 4391 .write_std = kvm_write_guest_virt_system,
1871c602 4392 .fetch = kvm_fetch_guest_virt,
bbd9b64e
CO		 4393 .read_emulated = emulator_read_emulated,
4394 .write_emulated = emulator_write_emulated,
4395 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3cb16fe7 4396 .invlpg = emulator_invlpg,
cf8f70bf
GN		 4397 .pio_in_emulated = emulator_pio_in_emulated,
4398 .pio_out_emulated = emulator_pio_out_emulated,
2dafc6c2
GN		 4399 .get_cached_descriptor = emulator_get_cached_descriptor,
4400 .set_cached_descriptor = emulator_set_cached_descriptor,
4401 .get_segment_selector = emulator_get_segment_selector,
4402 .set_segment_selector = emulator_set_segment_selector,
5951c442 4403 .get_cached_segment_base = emulator_get_cached_segment_base,
2dafc6c2 4404 .get_gdt = emulator_get_gdt,
160ce1f1 4405 .get_idt = emulator_get_idt,
1ac9d0cf
AK		 4406 .set_gdt = emulator_set_gdt,
4407 .set_idt = emulator_set_idt,
52a46617
GN		 4408 .get_cr = emulator_get_cr,
4409 .set_cr = emulator_set_cr,
9c537244 4410 .cpl = emulator_get_cpl,
35aa5375
GN		 4411 .get_dr = emulator_get_dr,
4412 .set_dr = emulator_set_dr,
717746e3
AK		 4413 .set_msr = emulator_set_msr,
4414 .get_msr = emulator_get_msr,
6c3287f7 4415 .halt = emulator_halt,
bcaf5cc5 4416 .wbinvd = emulator_wbinvd,
d6aa1000 4417 .fix_hypercall = emulator_fix_hypercall,
5037f6f3
AK		 4418 .get_fpu = emulator_get_fpu,
4419 .put_fpu = emulator_put_fpu,
c4f035c6 4420 .intercept = emulator_intercept,
bbd9b64e
CO		 4421};
4422
5fdbf976
MT		 4423static void cache_all_regs(struct kvm_vcpu *vcpu)
4424{
4425 kvm_register_read(vcpu, VCPU_REGS_RAX);
4426 kvm_register_read(vcpu, VCPU_REGS_RSP);
4427 kvm_register_read(vcpu, VCPU_REGS_RIP);
4428 vcpu->arch.regs_dirty = ~0;
4429}
4430
95cb2295
GN		 4431static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4432{
4433 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4434 /*
4435 * an sti; sti; sequence only disable interrupts for the first
4436 * instruction. So, if the last instruction, be it emulated or
4437 * not, left the system with the INT_STI flag enabled, it
4438 * means that the last instruction is an sti. We should not
4439 * leave the flag on in this case. The same goes for mov ss
4440 */
4441 if (!(int_shadow & mask))
4442 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4443}
4444
54b8486f
GN		 4445static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4446{
4447 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
da9cb575 4448 if (ctxt->exception.vector == PF_VECTOR)
6389ee94 4449 kvm_propagate_fault(vcpu, &ctxt->exception);
da9cb575
AK		 4450 else if (ctxt->exception.error_code_valid)
4451 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4452 ctxt->exception.error_code);
54b8486f 4453 else
da9cb575 4454 kvm_queue_exception(vcpu, ctxt->exception.vector);
54b8486f
GN		 4455}
4456
8ec4722d
MG		 4457static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4458{
4459 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4460 int cs_db, cs_l;
4461
4462 cache_all_regs(vcpu);
4463
4464 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4465
4466 vcpu->arch.emulate_ctxt.vcpu = vcpu;
f6e78475 4467 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
8ec4722d
MG		 4468 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4469 vcpu->arch.emulate_ctxt.mode =
4470 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4471 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4472 ? X86EMUL_MODE_VM86 : cs_l
4473 ? X86EMUL_MODE_PROT64 : cs_db
4474 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
c4f035c6 4475 vcpu->arch.emulate_ctxt.guest_mode = is_guest_mode(vcpu);
8ec4722d
MG		 4476 memset(c, 0, sizeof(struct decode_cache));
4477 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
7ae441ea 4478 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8ec4722d
MG		 4479}
4480
71f9833b 4481int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
63995653
MG		 4482{
4483 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4484 int ret;
4485
4486 init_emulate_ctxt(vcpu);
4487
4488 vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4489 vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
71f9833b
SH		 4490 vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip +
4491 inc_eip;
63995653
MG		 4492 ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4493
4494 if (ret != X86EMUL_CONTINUE)
4495 return EMULATE_FAIL;
4496
4497 vcpu->arch.emulate_ctxt.eip = c->eip;
4498 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4499 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
f6e78475 4500 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
63995653
MG		 4501
4502 if (irq == NMI_VECTOR)
4503 vcpu->arch.nmi_pending = false;
4504 else
4505 vcpu->arch.interrupt.pending = false;
4506
4507 return EMULATE_DONE;
4508}
4509EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4510
6d77dbfc
GN		 4511static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4512{
fc3a9157
JR		 4513 int r = EMULATE_DONE;
4514
6d77dbfc
GN		 4515 ++vcpu->stat.insn_emulation_fail;
4516 trace_kvm_emulate_insn_failed(vcpu);
fc3a9157
JR		 4517 if (!is_guest_mode(vcpu)) {
4518 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4519 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4520 vcpu->run->internal.ndata = 0;
4521 r = EMULATE_FAIL;
4522 }
6d77dbfc 4523 kvm_queue_exception(vcpu, UD_VECTOR);
fc3a9157
JR		 4524
4525 return r;
6d77dbfc
GN		 4526}
4527
a6f177ef
GN		 4528static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4529{
4530 gpa_t gpa;
4531
68be0803
GN		 4532 if (tdp_enabled)
4533 return false;
4534
a6f177ef
GN		 4535 /*
4536 * if emulation was due to access to shadowed page table
4537 * and it failed try to unshadow page and re-entetr the
4538 * guest to let CPU execute the instruction.
4539 */
4540 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4541 return true;
4542
4543 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4544
4545 if (gpa == UNMAPPED_GVA)
4546 return true; /* let cpu generate fault */
4547
4548 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4549 return true;
4550
4551 return false;
4552}
4553
51d8b661
AP		 4554int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4555 unsigned long cr2,
dc25e89e
AP		 4556 int emulation_type,
4557 void *insn,
4558 int insn_len)
bbd9b64e 4559{
95cb2295 4560 int r;
4d2179e1 4561 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
7ae441ea 4562 bool writeback = true;
bbd9b64e 4563
26eef70c 4564 kvm_clear_exception_queue(vcpu);
ad312c7c 4565 vcpu->arch.mmio_fault_cr2 = cr2;
5fdbf976 4566 /*
56e82318 4567 * TODO: fix emulate.c to use guest_read/write_register
5fdbf976
MT		 4568 * instead of direct ->regs accesses, can save hundred cycles
4569 * on Intel for instructions that don't read/change RSP, for
4570 * for example.
4571 */
4572 cache_all_regs(vcpu);
bbd9b64e 4573
571008da 4574 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8ec4722d 4575 init_emulate_ctxt(vcpu);
95cb2295 4576 vcpu->arch.emulate_ctxt.interruptibility = 0;
da9cb575 4577 vcpu->arch.emulate_ctxt.have_exception = false;
4fc40f07 4578 vcpu->arch.emulate_ctxt.perm_ok = false;
bbd9b64e 4579
4005996e
AK		 4580 vcpu->arch.emulate_ctxt.only_vendor_specific_insn
4581 = emulation_type & EMULTYPE_TRAP_UD;
4582
dc25e89e 4583 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, insn, insn_len);
bbd9b64e 4584
e46479f8 4585 trace_kvm_emulate_insn_start(vcpu);
f2b5756b 4586 ++vcpu->stat.insn_emulation;
bbd9b64e 4587 if (r) {
4005996e
AK		 4588 if (emulation_type & EMULTYPE_TRAP_UD)
4589 return EMULATE_FAIL;
a6f177ef 4590 if (reexecute_instruction(vcpu, cr2))
bbd9b64e 4591 return EMULATE_DONE;
6d77dbfc
GN		 4592 if (emulation_type & EMULTYPE_SKIP)
4593 return EMULATE_FAIL;
4594 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 4595 }
4596 }
4597
ba8afb6b
GN		 4598 if (emulation_type & EMULTYPE_SKIP) {
4599 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4600 return EMULATE_DONE;
4601 }
4602
7ae441ea 4603 /* this is needed for vmware backdoor interface to work since it
4d2179e1 4604 changes registers values during IO operation */
7ae441ea
GN		 4605 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4606 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4607 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4608 }
4d2179e1 4609
5cd21917 4610restart:
9aabc88f 4611 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
bbd9b64e 4612
775fde86
JR		 4613 if (r == EMULATION_INTERCEPTED)
4614 return EMULATE_DONE;
4615
d2ddd1c4 4616 if (r == EMULATION_FAILED) {
a6f177ef 4617 if (reexecute_instruction(vcpu, cr2))
c3cd7ffa
GN		 4618 return EMULATE_DONE;
4619
6d77dbfc 4620 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 4621 }
4622
da9cb575 4623 if (vcpu->arch.emulate_ctxt.have_exception) {
54b8486f 4624 inject_emulated_exception(vcpu);
d2ddd1c4
GN		 4625 r = EMULATE_DONE;
4626 } else if (vcpu->arch.pio.count) {
3457e419
GN		 4627 if (!vcpu->arch.pio.in)
4628 vcpu->arch.pio.count = 0;
7ae441ea
GN		 4629 else
4630 writeback = false;
e85d28f8 4631 r = EMULATE_DO_MMIO;
7ae441ea
GN		 4632 } else if (vcpu->mmio_needed) {
4633 if (!vcpu->mmio_is_write)
4634 writeback = false;
e85d28f8 4635 r = EMULATE_DO_MMIO;
7ae441ea 4636 } else if (r == EMULATION_RESTART)
5cd21917 4637 goto restart;
d2ddd1c4
GN		 4638 else
4639 r = EMULATE_DONE;
f850e2e6 4640
7ae441ea
GN		 4641 if (writeback) {
4642 toggle_interruptibility(vcpu,
4643 vcpu->arch.emulate_ctxt.interruptibility);
4644 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4645 kvm_make_request(KVM_REQ_EVENT, vcpu);
4646 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4647 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4648 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4649 } else
4650 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
e85d28f8
GN		 4651
4652 return r;
de7d789a 4653}
51d8b661 4654EXPORT_SYMBOL_GPL(x86_emulate_instruction);
de7d789a 4655
cf8f70bf 4656int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
de7d789a 4657{
cf8f70bf 4658 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
ca1d4a9e
AK		 4659 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4660 size, port, &val, 1);
cf8f70bf 4661 /* do not return to emulator after return from userspace */
7972995b 4662 vcpu->arch.pio.count = 0;
de7d789a
CO		 4663 return ret;
4664}
cf8f70bf 4665EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
de7d789a 4666
8cfdc000
ZA		 4667static void tsc_bad(void *info)
4668{
0a3aee0d 4669 __this_cpu_write(cpu_tsc_khz, 0);
8cfdc000
ZA		 4670}
4671
4672static void tsc_khz_changed(void *data)
c8076604 4673{
8cfdc000
ZA		 4674 struct cpufreq_freqs *freq = data;
4675 unsigned long khz = 0;
4676
4677 if (data)
4678 khz = freq->new;
4679 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4680 khz = cpufreq_quick_get(raw_smp_processor_id());
4681 if (!khz)
4682 khz = tsc_khz;
0a3aee0d 4683 __this_cpu_write(cpu_tsc_khz, khz);
c8076604
GH		 4684}
4685
c8076604
GH		 4686static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4687 void *data)
4688{
4689 struct cpufreq_freqs *freq = data;
4690 struct kvm *kvm;
4691 struct kvm_vcpu *vcpu;
4692 int i, send_ipi = 0;
4693
8cfdc000
ZA		 4694 /*
4695 * We allow guests to temporarily run on slowing clocks,
4696 * provided we notify them after, or to run on accelerating
4697 * clocks, provided we notify them before. Thus time never
4698 * goes backwards.
4699 *
4700 * However, we have a problem. We can't atomically update
4701 * the frequency of a given CPU from this function; it is
4702 * merely a notifier, which can be called from any CPU.
4703 * Changing the TSC frequency at arbitrary points in time
4704 * requires a recomputation of local variables related to
4705 * the TSC for each VCPU. We must flag these local variables
4706 * to be updated and be sure the update takes place with the
4707 * new frequency before any guests proceed.
4708 *
4709 * Unfortunately, the combination of hotplug CPU and frequency
4710 * change creates an intractable locking scenario; the order
4711 * of when these callouts happen is undefined with respect to
4712 * CPU hotplug, and they can race with each other. As such,
4713 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4714 * undefined; you can actually have a CPU frequency change take
4715 * place in between the computation of X and the setting of the
4716 * variable. To protect against this problem, all updates of
4717 * the per_cpu tsc_khz variable are done in an interrupt
4718 * protected IPI, and all callers wishing to update the value
4719 * must wait for a synchronous IPI to complete (which is trivial
4720 * if the caller is on the CPU already). This establishes the
4721 * necessary total order on variable updates.
4722 *
4723 * Note that because a guest time update may take place
4724 * anytime after the setting of the VCPU's request bit, the
4725 * correct TSC value must be set before the request. However,
4726 * to ensure the update actually makes it to any guest which
4727 * starts running in hardware virtualization between the set
4728 * and the acquisition of the spinlock, we must also ping the
4729 * CPU after setting the request bit.
4730 *
4731 */
4732
c8076604
GH		 4733 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4734 return 0;
4735 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4736 return 0;
8cfdc000
ZA		 4737
4738 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604 4739
e935b837 4740 raw_spin_lock(&kvm_lock);
c8076604 4741 list_for_each_entry(kvm, &vm_list, vm_list) {
988a2cae 4742 kvm_for_each_vcpu(i, vcpu, kvm) {
c8076604
GH		 4743 if (vcpu->cpu != freq->cpu)
4744 continue;
c285545f 4745 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
c8076604 4746 if (vcpu->cpu != smp_processor_id())
8cfdc000 4747 send_ipi = 1;
c8076604
GH		 4748 }
4749 }
e935b837 4750 raw_spin_unlock(&kvm_lock);
c8076604
GH		 4751
4752 if (freq->old < freq->new && send_ipi) {
4753 /*
4754 * We upscale the frequency. Must make the guest
4755 * doesn't see old kvmclock values while running with
4756 * the new frequency, otherwise we risk the guest sees
4757 * time go backwards.
4758 *
4759 * In case we update the frequency for another cpu
4760 * (which might be in guest context) send an interrupt
4761 * to kick the cpu out of guest context. Next time
4762 * guest context is entered kvmclock will be updated,
4763 * so the guest will not see stale values.
4764 */
8cfdc000 4765 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604
GH		 4766 }
4767 return 0;
4768}
4769
4770static struct notifier_block kvmclock_cpufreq_notifier_block = {
8cfdc000
ZA		 4771 .notifier_call = kvmclock_cpufreq_notifier
4772};
4773
4774static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4775 unsigned long action, void *hcpu)
4776{
4777 unsigned int cpu = (unsigned long)hcpu;
4778
4779 switch (action) {
4780 case CPU_ONLINE:
4781 case CPU_DOWN_FAILED:
4782 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4783 break;
4784 case CPU_DOWN_PREPARE:
4785 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4786 break;
4787 }
4788 return NOTIFY_OK;
4789}
4790
4791static struct notifier_block kvmclock_cpu_notifier_block = {
4792 .notifier_call = kvmclock_cpu_notifier,
4793 .priority = -INT_MAX
c8076604
GH		 4794};
4795
b820cc0c
ZA		 4796static void kvm_timer_init(void)
4797{
4798 int cpu;
4799
c285545f 4800 max_tsc_khz = tsc_khz;
8cfdc000 4801 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
b820cc0c 4802 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
c285545f
ZA		 4803#ifdef CONFIG_CPU_FREQ
4804 struct cpufreq_policy policy;
4805 memset(&policy, 0, sizeof(policy));
3e26f230
AK		 4806 cpu = get_cpu();
4807 cpufreq_get_policy(&policy, cpu);
c285545f
ZA		 4808 if (policy.cpuinfo.max_freq)
4809 max_tsc_khz = policy.cpuinfo.max_freq;
3e26f230 4810 put_cpu();
c285545f 4811#endif
b820cc0c
ZA		 4812 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4813 CPUFREQ_TRANSITION_NOTIFIER);
4814 }
c285545f 4815 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
8cfdc000
ZA		 4816 for_each_online_cpu(cpu)
4817 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
b820cc0c
ZA		 4818}
4819
ff9d07a0
ZY		 4820static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4821
4822static int kvm_is_in_guest(void)
4823{
4824 return percpu_read(current_vcpu) != NULL;
4825}
4826
4827static int kvm_is_user_mode(void)
4828{
4829 int user_mode = 3;
dcf46b94 4830
ff9d07a0
ZY		 4831 if (percpu_read(current_vcpu))
4832 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
dcf46b94 4833
ff9d07a0
ZY		 4834 return user_mode != 0;
4835}
4836
4837static unsigned long kvm_get_guest_ip(void)
4838{
4839 unsigned long ip = 0;
dcf46b94 4840
ff9d07a0
ZY		 4841 if (percpu_read(current_vcpu))
4842 ip = kvm_rip_read(percpu_read(current_vcpu));
dcf46b94 4843
ff9d07a0
ZY		 4844 return ip;
4845}
4846
4847static struct perf_guest_info_callbacks kvm_guest_cbs = {
4848 .is_in_guest = kvm_is_in_guest,
4849 .is_user_mode = kvm_is_user_mode,
4850 .get_guest_ip = kvm_get_guest_ip,
4851};
4852
4853void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4854{
4855 percpu_write(current_vcpu, vcpu);
4856}
4857EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4858
4859void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4860{
4861 percpu_write(current_vcpu, NULL);
4862}
4863EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4864
f8c16bba 4865int kvm_arch_init(void *opaque)
043405e1 4866{
b820cc0c 4867 int r;
f8c16bba
ZX		 4868 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4869
f8c16bba
ZX		 4870 if (kvm_x86_ops) {
4871 printk(KERN_ERR "kvm: already loaded the other module\n");
56c6d28a
ZX		 4872 r = -EEXIST;
4873 goto out;
f8c16bba
ZX		 4874 }
4875
4876 if (!ops->cpu_has_kvm_support()) {
4877 printk(KERN_ERR "kvm: no hardware support\n");
56c6d28a
ZX		 4878 r = -EOPNOTSUPP;
4879 goto out;
f8c16bba
ZX		 4880 }
4881 if (ops->disabled_by_bios()) {
4882 printk(KERN_ERR "kvm: disabled by bios\n");
56c6d28a
ZX		 4883 r = -EOPNOTSUPP;
4884 goto out;
f8c16bba
ZX		 4885 }
4886
97db56ce
AK		 4887 r = kvm_mmu_module_init();
4888 if (r)
4889 goto out;
4890
4891 kvm_init_msr_list();
4892
f8c16bba 4893 kvm_x86_ops = ops;
56c6d28a 4894 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
7b52345e 4895 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4b12f0de 4896 PT_DIRTY_MASK, PT64_NX_MASK, 0);
c8076604 4897
b820cc0c 4898 kvm_timer_init();
c8076604 4899
ff9d07a0
ZY		 4900 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4901
2acf923e
DC		 4902 if (cpu_has_xsave)
4903 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4904
f8c16bba 4905 return 0;
56c6d28a
ZX		 4906
4907out:
56c6d28a 4908 return r;
043405e1 4909}
8776e519 4910
f8c16bba
ZX		 4911void kvm_arch_exit(void)
4912{
ff9d07a0
ZY		 4913 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4914
888d256e
JK		 4915 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4916 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4917 CPUFREQ_TRANSITION_NOTIFIER);
8cfdc000 4918 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
f8c16bba 4919 kvm_x86_ops = NULL;
56c6d28a
ZX		 4920 kvm_mmu_module_exit();
4921}
f8c16bba 4922
8776e519
HB		 4923int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4924{
4925 ++vcpu->stat.halt_exits;
4926 if (irqchip_in_kernel(vcpu->kvm)) {
a4535290 4927 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
8776e519
HB		 4928 return 1;
4929 } else {
4930 vcpu->run->exit_reason = KVM_EXIT_HLT;
4931 return 0;
4932 }
4933}
4934EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4935
2f333bcb
MT		 4936static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4937 unsigned long a1)
4938{
4939 if (is_long_mode(vcpu))
4940 return a0;
4941 else
4942 return a0 | ((gpa_t)a1 << 32);
4943}
4944
55cd8e5a
GN		 4945int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4946{
4947 u64 param, ingpa, outgpa, ret;
4948 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4949 bool fast, longmode;
4950 int cs_db, cs_l;
4951
4952 /*
4953 * hypercall generates UD from non zero cpl and real mode
4954 * per HYPER-V spec
4955 */
3eeb3288 4956 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
55cd8e5a
GN		 4957 kvm_queue_exception(vcpu, UD_VECTOR);
4958 return 0;
4959 }
4960
4961 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4962 longmode = is_long_mode(vcpu) && cs_l == 1;
4963
4964 if (!longmode) {
ccd46936
GN		 4965 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4966 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4967 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4968 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4969 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4970 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
55cd8e5a
GN		 4971 }
4972#ifdef CONFIG_X86_64
4973 else {
4974 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4975 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4976 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4977 }
4978#endif
4979
4980 code = param & 0xffff;
4981 fast = (param >> 16) & 0x1;
4982 rep_cnt = (param >> 32) & 0xfff;
4983 rep_idx = (param >> 48) & 0xfff;
4984
4985 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4986
c25bc163
GN		 4987 switch (code) {
4988 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4989 kvm_vcpu_on_spin(vcpu);
4990 break;
4991 default:
4992 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4993 break;
4994 }
55cd8e5a
GN		 4995
4996 ret = res | (((u64)rep_done & 0xfff) << 32);
4997 if (longmode) {
4998 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4999 } else {
5000 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5001 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5002 }
5003
5004 return 1;
5005}
5006
8776e519
HB		 5007int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5008{
5009 unsigned long nr, a0, a1, a2, a3, ret;
2f333bcb 5010 int r = 1;
8776e519 5011
55cd8e5a
GN		 5012 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5013 return kvm_hv_hypercall(vcpu);
5014
5fdbf976
MT		 5015 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5016 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5017 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5018 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5019 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
8776e519 5020
229456fc 5021 trace_kvm_hypercall(nr, a0, a1, a2, a3);
2714d1d3 5022
8776e519
HB		 5023 if (!is_long_mode(vcpu)) {
5024 nr &= 0xFFFFFFFF;
5025 a0 &= 0xFFFFFFFF;
5026 a1 &= 0xFFFFFFFF;
5027 a2 &= 0xFFFFFFFF;
5028 a3 &= 0xFFFFFFFF;
5029 }
5030
07708c4a
JK		 5031 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5032 ret = -KVM_EPERM;
5033 goto out;
5034 }
5035
8776e519 5036 switch (nr) {
b93463aa
AK		 5037 case KVM_HC_VAPIC_POLL_IRQ:
5038 ret = 0;
5039 break;
2f333bcb
MT		 5040 case KVM_HC_MMU_OP:
5041 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5042 break;
8776e519
HB		 5043 default:
5044 ret = -KVM_ENOSYS;
5045 break;
5046 }
07708c4a 5047out:
5fdbf976 5048 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
f11c3a8d 5049 ++vcpu->stat.hypercalls;
2f333bcb 5050 return r;
8776e519
HB		 5051}
5052EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5053
d6aa1000 5054int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8776e519 5055{
d6aa1000 5056 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8776e519 5057 char instruction[3];
5fdbf976 5058 unsigned long rip = kvm_rip_read(vcpu);
8776e519 5059
8776e519
HB		 5060 /*
5061 * Blow out the MMU to ensure that no other VCPU has an active mapping
5062 * to ensure that the updated hypercall appears atomically across all
5063 * VCPUs.
5064 */
5065 kvm_mmu_zap_all(vcpu->kvm);
5066
8776e519 5067 kvm_x86_ops->patch_hypercall(vcpu, instruction);
8776e519 5068
0f65dd70
AK		 5069 return emulator_write_emulated(&vcpu->arch.emulate_ctxt,
5070 rip, instruction, 3, NULL);
8776e519
HB		 5071}
5072
07716717
DK		 5073static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5074{
ad312c7c
ZX		 5075 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5076 int j, nent = vcpu->arch.cpuid_nent;
07716717
DK		 5077
5078 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5079 /* when no next entry is found, the current entry[i] is reselected */
0fdf8e59 5080 for (j = i + 1; ; j = (j + 1) % nent) {
ad312c7c 5081 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
07716717
DK		 5082 if (ej->function == e->function) {
5083 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5084 return j;
5085 }
5086 }
5087 return 0; /* silence gcc, even though control never reaches here */
5088}
5089
5090/* find an entry with matching function, matching index (if needed), and that
5091 * should be read next (if it's stateful) */
5092static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5093 u32 function, u32 index)
5094{
5095 if (e->function != function)
5096 return 0;
5097 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5098 return 0;
5099 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
19355475 5100 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
07716717
DK		 5101 return 0;
5102 return 1;
5103}
5104
d8017474
AG		 5105struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5106 u32 function, u32 index)
8776e519
HB		 5107{
5108 int i;
d8017474 5109 struct kvm_cpuid_entry2 *best = NULL;
8776e519 5110
ad312c7c 5111 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
d8017474
AG		 5112 struct kvm_cpuid_entry2 *e;
5113
ad312c7c 5114 e = &vcpu->arch.cpuid_entries[i];
07716717
DK		 5115 if (is_matching_cpuid_entry(e, function, index)) {
5116 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5117 move_to_next_stateful_cpuid_entry(vcpu, i);
8776e519
HB		 5118 best = e;
5119 break;
5120 }
8776e519 5121 }
d8017474
AG		 5122 return best;
5123}
0e851880 5124EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
d8017474 5125
82725b20
DE		 5126int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5127{
5128 struct kvm_cpuid_entry2 *best;
5129
f7a71197
AK		 5130 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5131 if (!best || best->eax < 0x80000008)
5132 goto not_found;
82725b20
DE		 5133 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5134 if (best)
5135 return best->eax & 0xff;
f7a71197 5136not_found:
82725b20
DE		 5137 return 36;
5138}
5139
bd22f5cf
AP		 5140/*
5141 * If no match is found, check whether we exceed the vCPU's limit
5142 * and return the content of the highest valid _standard_ leaf instead.
5143 * This is to satisfy the CPUID specification.
5144 */
5145static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5146 u32 function, u32 index)
5147{
5148 struct kvm_cpuid_entry2 *maxlevel;
5149
5150 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5151 if (!maxlevel || maxlevel->eax >= function)
5152 return NULL;
5153 if (function & 0x80000000) {
5154 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5155 if (!maxlevel)
5156 return NULL;
5157 }
5158 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5159}
5160
d8017474
AG		 5161void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5162{
5163 u32 function, index;
5164 struct kvm_cpuid_entry2 *best;
5165
5166 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5167 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5168 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5169 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5170 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5171 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5172 best = kvm_find_cpuid_entry(vcpu, function, index);
bd22f5cf
AP		 5173
5174 if (!best)
5175 best = check_cpuid_limit(vcpu, function, index);
5176
8776e519 5177 if (best) {
5fdbf976
MT		 5178 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5179 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5180 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5181 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
8776e519 5182 }
8776e519 5183 kvm_x86_ops->skip_emulated_instruction(vcpu);
229456fc
MT		 5184 trace_kvm_cpuid(function,
5185 kvm_register_read(vcpu, VCPU_REGS_RAX),
5186 kvm_register_read(vcpu, VCPU_REGS_RBX),
5187 kvm_register_read(vcpu, VCPU_REGS_RCX),
5188 kvm_register_read(vcpu, VCPU_REGS_RDX));
8776e519
HB		 5189}
5190EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
d0752060 5191
b6c7a5dc
HB		 5192/*
5193 * Check if userspace requested an interrupt window, and that the
5194 * interrupt window is open.
5195 *
5196 * No need to exit to userspace if we already have an interrupt queued.
5197 */
851ba692 5198static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
b6c7a5dc 5199{
8061823a 5200 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
851ba692 5201 vcpu->run->request_interrupt_window &&
5df56646 5202 kvm_arch_interrupt_allowed(vcpu));
b6c7a5dc
HB		 5203}
5204
851ba692 5205static void post_kvm_run_save(struct kvm_vcpu *vcpu)
b6c7a5dc 5206{
851ba692
AK		 5207 struct kvm_run *kvm_run = vcpu->run;
5208
91586a3b 5209 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2d3ad1f4 5210 kvm_run->cr8 = kvm_get_cr8(vcpu);
b6c7a5dc 5211 kvm_run->apic_base = kvm_get_apic_base(vcpu);
4531220b 5212 if (irqchip_in_kernel(vcpu->kvm))
b6c7a5dc 5213 kvm_run->ready_for_interrupt_injection = 1;
4531220b 5214 else
b6c7a5dc 5215 kvm_run->ready_for_interrupt_injection =
fa9726b0
GN		 5216 kvm_arch_interrupt_allowed(vcpu) &&
5217 !kvm_cpu_has_interrupt(vcpu) &&
5218 !kvm_event_needs_reinjection(vcpu);
b6c7a5dc
HB		 5219}
5220
b93463aa
AK		 5221static void vapic_enter(struct kvm_vcpu *vcpu)
5222{
5223 struct kvm_lapic *apic = vcpu->arch.apic;
5224 struct page *page;
5225
5226 if (!apic || !apic->vapic_addr)
5227 return;
5228
5229 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
72dc67a6
IE		 5230
5231 vcpu->arch.apic->vapic_page = page;
b93463aa
AK		 5232}
5233
5234static void vapic_exit(struct kvm_vcpu *vcpu)
5235{
5236 struct kvm_lapic *apic = vcpu->arch.apic;
f656ce01 5237 int idx;
b93463aa
AK		 5238
5239 if (!apic || !apic->vapic_addr)
5240 return;
5241
f656ce01 5242 idx = srcu_read_lock(&vcpu->kvm->srcu);
b93463aa
AK		 5243 kvm_release_page_dirty(apic->vapic_page);
5244 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
f656ce01 5245 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b93463aa
AK		 5246}
5247
95ba8273
GN		 5248static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5249{
5250 int max_irr, tpr;
5251
5252 if (!kvm_x86_ops->update_cr8_intercept)
5253 return;
5254
88c808fd
AK		 5255 if (!vcpu->arch.apic)
5256 return;
5257
8db3baa2
GN		 5258 if (!vcpu->arch.apic->vapic_addr)
5259 max_irr = kvm_lapic_find_highest_irr(vcpu);
5260 else
5261 max_irr = -1;
95ba8273
GN		 5262
5263 if (max_irr != -1)
5264 max_irr >>= 4;
5265
5266 tpr = kvm_lapic_get_cr8(vcpu);
5267
5268 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5269}
5270
851ba692 5271static void inject_pending_event(struct kvm_vcpu *vcpu)
95ba8273
GN		 5272{
5273 /* try to reinject previous events if any */
b59bb7bd 5274 if (vcpu->arch.exception.pending) {
5c1c85d0
AK		 5275 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5276 vcpu->arch.exception.has_error_code,
5277 vcpu->arch.exception.error_code);
b59bb7bd
GN		 5278 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5279 vcpu->arch.exception.has_error_code,
ce7ddec4
JR		 5280 vcpu->arch.exception.error_code,
5281 vcpu->arch.exception.reinject);
b59bb7bd
GN		 5282 return;
5283 }
5284
95ba8273
GN		 5285 if (vcpu->arch.nmi_injected) {
5286 kvm_x86_ops->set_nmi(vcpu);
5287 return;
5288 }
5289
5290 if (vcpu->arch.interrupt.pending) {
66fd3f7f 5291 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 5292 return;
5293 }
5294
5295 /* try to inject new event if pending */
5296 if (vcpu->arch.nmi_pending) {
5297 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5298 vcpu->arch.nmi_pending = false;
5299 vcpu->arch.nmi_injected = true;
5300 kvm_x86_ops->set_nmi(vcpu);
5301 }
5302 } else if (kvm_cpu_has_interrupt(vcpu)) {
5303 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
66fd3f7f
GN		 5304 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5305 false);
5306 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 5307 }
5308 }
5309}
5310
2acf923e
DC		 5311static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5312{
5313 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5314 !vcpu->guest_xcr0_loaded) {
5315 /* kvm_set_xcr() also depends on this */
5316 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5317 vcpu->guest_xcr0_loaded = 1;
5318 }
5319}
5320
5321static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5322{
5323 if (vcpu->guest_xcr0_loaded) {
5324 if (vcpu->arch.xcr0 != host_xcr0)
5325 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5326 vcpu->guest_xcr0_loaded = 0;
5327 }
5328}
5329
851ba692 5330static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
b6c7a5dc
HB		 5331{
5332 int r;
1499e54a 5333 bool nmi_pending;
6a8b1d13 5334 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
851ba692 5335 vcpu->run->request_interrupt_window;
b6c7a5dc 5336
3e007509 5337 if (vcpu->requests) {
a8eeb04a 5338 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
2e53d63a 5339 kvm_mmu_unload(vcpu);
a8eeb04a 5340 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
2f599714 5341 __kvm_migrate_timers(vcpu);
34c238a1
ZA		 5342 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5343 r = kvm_guest_time_update(vcpu);
8cfdc000
ZA		 5344 if (unlikely(r))
5345 goto out;
5346 }
a8eeb04a 5347 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
4731d4c7 5348 kvm_mmu_sync_roots(vcpu);
a8eeb04a 5349 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
d4acf7e7 5350 kvm_x86_ops->tlb_flush(vcpu);
a8eeb04a 5351 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
851ba692 5352 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
b93463aa
AK		 5353 r = 0;
5354 goto out;
5355 }
a8eeb04a 5356 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
851ba692 5357 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
71c4dfaf
JR		 5358 r = 0;
5359 goto out;
5360 }
a8eeb04a 5361 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
02daab21
AK		 5362 vcpu->fpu_active = 0;
5363 kvm_x86_ops->fpu_deactivate(vcpu);
5364 }
af585b92
GN		 5365 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5366 /* Page is swapped out. Do synthetic halt */
5367 vcpu->arch.apf.halted = true;
5368 r = 1;
5369 goto out;
5370 }
2f52d58c 5371 }
b93463aa 5372
3e007509
AK		 5373 r = kvm_mmu_reload(vcpu);
5374 if (unlikely(r))
5375 goto out;
5376
1499e54a
GN		 5377 /*
5378 * An NMI can be injected between local nmi_pending read and
5379 * vcpu->arch.nmi_pending read inside inject_pending_event().
5380 * But in that case, KVM_REQ_EVENT will be set, which makes
5381 * the race described above benign.
5382 */
5383 nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5384
b463a6f7
AK		 5385 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5386 inject_pending_event(vcpu);
5387
5388 /* enable NMI/IRQ window open exits if needed */
1499e54a 5389 if (nmi_pending)
b463a6f7
AK		 5390 kvm_x86_ops->enable_nmi_window(vcpu);
5391 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5392 kvm_x86_ops->enable_irq_window(vcpu);
5393
5394 if (kvm_lapic_enabled(vcpu)) {
5395 update_cr8_intercept(vcpu);
5396 kvm_lapic_sync_to_vapic(vcpu);
5397 }
5398 }
5399
b6c7a5dc
HB		 5400 preempt_disable();
5401
5402 kvm_x86_ops->prepare_guest_switch(vcpu);
2608d7a1
AK		 5403 if (vcpu->fpu_active)
5404 kvm_load_guest_fpu(vcpu);
2acf923e 5405 kvm_load_guest_xcr0(vcpu);
b6c7a5dc 5406
6b7e2d09
XG		 5407 vcpu->mode = IN_GUEST_MODE;
5408
5409 /* We should set ->mode before check ->requests,
5410 * see the comment in make_all_cpus_request.
5411 */
5412 smp_mb();
b6c7a5dc 5413
d94e1dc9 5414 local_irq_disable();
32f88400 5415
6b7e2d09 5416 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
d94e1dc9 5417 || need_resched() || signal_pending(current)) {
6b7e2d09 5418 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 5419 smp_wmb();
6c142801
AK		 5420 local_irq_enable();
5421 preempt_enable();
b463a6f7 5422 kvm_x86_ops->cancel_injection(vcpu);
6c142801
AK		 5423 r = 1;
5424 goto out;
5425 }
5426
f656ce01 5427 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
3200f405 5428
b6c7a5dc
HB		 5429 kvm_guest_enter();
5430
42dbaa5a 5431 if (unlikely(vcpu->arch.switch_db_regs)) {
42dbaa5a
JK		 5432 set_debugreg(0, 7);
5433 set_debugreg(vcpu->arch.eff_db[0], 0);
5434 set_debugreg(vcpu->arch.eff_db[1], 1);
5435 set_debugreg(vcpu->arch.eff_db[2], 2);
5436 set_debugreg(vcpu->arch.eff_db[3], 3);
5437 }
b6c7a5dc 5438
229456fc 5439 trace_kvm_entry(vcpu->vcpu_id);
851ba692 5440 kvm_x86_ops->run(vcpu);
b6c7a5dc 5441
24f1e32c
FW		 5442 /*
5443 * If the guest has used debug registers, at least dr7
5444 * will be disabled while returning to the host.
5445 * If we don't have active breakpoints in the host, we don't
5446 * care about the messed up debug address registers. But if
5447 * we have some of them active, restore the old state.
5448 */
59d8eb53 5449 if (hw_breakpoint_active())
24f1e32c 5450 hw_breakpoint_restore();
42dbaa5a 5451
1d5f066e
ZA		 5452 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5453
6b7e2d09 5454 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 5455 smp_wmb();
b6c7a5dc
HB		 5456 local_irq_enable();
5457
5458 ++vcpu->stat.exits;
5459
5460 /*
5461 * We must have an instruction between local_irq_enable() and
5462 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5463 * the interrupt shadow. The stat.exits increment will do nicely.
5464 * But we need to prevent reordering, hence this barrier():
5465 */
5466 barrier();
5467
5468 kvm_guest_exit();
5469
5470 preempt_enable();
5471
f656ce01 5472 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3200f405 5473
b6c7a5dc
HB		 5474 /*
5475 * Profile KVM exit RIPs:
5476 */
5477 if (unlikely(prof_on == KVM_PROFILING)) {
5fdbf976
MT		 5478 unsigned long rip = kvm_rip_read(vcpu);
5479 profile_hit(KVM_PROFILING, (void *)rip);
b6c7a5dc
HB		 5480 }
5481
298101da 5482
b93463aa
AK		 5483 kvm_lapic_sync_from_vapic(vcpu);
5484
851ba692 5485 r = kvm_x86_ops->handle_exit(vcpu);
d7690175
MT		 5486out:
5487 return r;
5488}
b6c7a5dc 5489
09cec754 5490
851ba692 5491static int __vcpu_run(struct kvm_vcpu *vcpu)
d7690175
MT		 5492{
5493 int r;
f656ce01 5494 struct kvm *kvm = vcpu->kvm;
d7690175
MT		 5495
5496 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
1b10bf31
JK		 5497 pr_debug("vcpu %d received sipi with vector # %x\n",
5498 vcpu->vcpu_id, vcpu->arch.sipi_vector);
d7690175 5499 kvm_lapic_reset(vcpu);
5f179287 5500 r = kvm_arch_vcpu_reset(vcpu);
d7690175
MT		 5501 if (r)
5502 return r;
5503 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
b6c7a5dc
HB		 5504 }
5505
f656ce01 5506 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175
MT		 5507 vapic_enter(vcpu);
5508
5509 r = 1;
5510 while (r > 0) {
af585b92
GN		 5511 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5512 !vcpu->arch.apf.halted)
851ba692 5513 r = vcpu_enter_guest(vcpu);
d7690175 5514 else {
f656ce01 5515 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
d7690175 5516 kvm_vcpu_block(vcpu);
f656ce01 5517 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
a8eeb04a 5518 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
09cec754
GN		 5519 {
5520 switch(vcpu->arch.mp_state) {
5521 case KVM_MP_STATE_HALTED:
d7690175 5522 vcpu->arch.mp_state =
09cec754
GN		 5523 KVM_MP_STATE_RUNNABLE;
5524 case KVM_MP_STATE_RUNNABLE:
af585b92 5525 vcpu->arch.apf.halted = false;
09cec754
GN		 5526 break;
5527 case KVM_MP_STATE_SIPI_RECEIVED:
5528 default:
5529 r = -EINTR;
5530 break;
5531 }
5532 }
d7690175
MT		 5533 }
5534
09cec754
GN		 5535 if (r <= 0)
5536 break;
5537
5538 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5539 if (kvm_cpu_has_pending_timer(vcpu))
5540 kvm_inject_pending_timer_irqs(vcpu);
5541
851ba692 5542 if (dm_request_for_irq_injection(vcpu)) {
09cec754 5543 r = -EINTR;
851ba692 5544 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754
GN		 5545 ++vcpu->stat.request_irq_exits;
5546 }
af585b92
GN		 5547
5548 kvm_check_async_pf_completion(vcpu);
5549
09cec754
GN		 5550 if (signal_pending(current)) {
5551 r = -EINTR;
851ba692 5552 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754
GN		 5553 ++vcpu->stat.signal_exits;
5554 }
5555 if (need_resched()) {
f656ce01 5556 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
09cec754 5557 kvm_resched(vcpu);
f656ce01 5558 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 5559 }
b6c7a5dc
HB		 5560 }
5561
f656ce01 5562 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
b6c7a5dc 5563
b93463aa
AK		 5564 vapic_exit(vcpu);
5565
b6c7a5dc
HB		 5566 return r;
5567}
5568
5287f194
AK		 5569static int complete_mmio(struct kvm_vcpu *vcpu)
5570{
5571 struct kvm_run *run = vcpu->run;
5572 int r;
5573
5574 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5575 return 1;
5576
5577 if (vcpu->mmio_needed) {
5287f194 5578 vcpu->mmio_needed = 0;
cef4dea0
AK		 5579 if (!vcpu->mmio_is_write)
5580 memcpy(vcpu->mmio_data, run->mmio.data, 8);
5581 vcpu->mmio_index += 8;
5582 if (vcpu->mmio_index < vcpu->mmio_size) {
5583 run->exit_reason = KVM_EXIT_MMIO;
5584 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5585 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5586 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5587 run->mmio.is_write = vcpu->mmio_is_write;
5588 vcpu->mmio_needed = 1;
5589 return 0;
5590 }
5591 if (vcpu->mmio_is_write)
5592 return 1;
5593 vcpu->mmio_read_completed = 1;
5287f194
AK		 5594 }
5595 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5596 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5597 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5598 if (r != EMULATE_DONE)
5599 return 0;
5600 return 1;
5601}
5602
b6c7a5dc
HB		 5603int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5604{
5605 int r;
5606 sigset_t sigsaved;
5607
e5c30142
AK		 5608 if (!tsk_used_math(current) && init_fpu(current))
5609 return -ENOMEM;
5610
ac9f6dc0
AK		 5611 if (vcpu->sigset_active)
5612 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5613
a4535290 5614 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
b6c7a5dc 5615 kvm_vcpu_block(vcpu);
d7690175 5616 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
ac9f6dc0
AK		 5617 r = -EAGAIN;
5618 goto out;
b6c7a5dc
HB		 5619 }
5620
b6c7a5dc 5621 /* re-sync apic's tpr */
eea1cff9
AP		 5622 if (!irqchip_in_kernel(vcpu->kvm)) {
5623 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5624 r = -EINVAL;
5625 goto out;
5626 }
5627 }
b6c7a5dc 5628
5287f194
AK		 5629 r = complete_mmio(vcpu);
5630 if (r <= 0)
5631 goto out;
5632
5fdbf976
MT		 5633 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5634 kvm_register_write(vcpu, VCPU_REGS_RAX,
5635 kvm_run->hypercall.ret);
b6c7a5dc 5636
851ba692 5637 r = __vcpu_run(vcpu);
b6c7a5dc
HB		 5638
5639out:
f1d86e46 5640 post_kvm_run_save(vcpu);
b6c7a5dc
HB		 5641 if (vcpu->sigset_active)
5642 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5643
b6c7a5dc
HB		 5644 return r;
5645}
5646
5647int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5648{
7ae441ea
GN		 5649 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5650 /*
5651 * We are here if userspace calls get_regs() in the middle of
5652 * instruction emulation. Registers state needs to be copied
5653 * back from emulation context to vcpu. Usrapace shouldn't do
5654 * that usually, but some bad designed PV devices (vmware
5655 * backdoor interface) need this to work
5656 */
5657 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5658 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5659 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5660 }
5fdbf976
MT		 5661 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5662 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5663 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5664 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5665 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5666 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5667 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5668 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
b6c7a5dc 5669#ifdef CONFIG_X86_64
5fdbf976
MT		 5670 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5671 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5672 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5673 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5674 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5675 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5676 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5677 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
b6c7a5dc
HB		 5678#endif
5679
5fdbf976 5680 regs->rip = kvm_rip_read(vcpu);
91586a3b 5681 regs->rflags = kvm_get_rflags(vcpu);
b6c7a5dc 5682
b6c7a5dc
HB		 5683 return 0;
5684}
5685
5686int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5687{
7ae441ea
GN		 5688 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5689 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5690
5fdbf976
MT		 5691 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5692 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5693 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5694 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5695 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5696 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5697 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5698 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
b6c7a5dc 5699#ifdef CONFIG_X86_64
5fdbf976
MT		 5700 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5701 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5702 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5703 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5704 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5705 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5706 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5707 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
b6c7a5dc
HB		 5708#endif
5709
5fdbf976 5710 kvm_rip_write(vcpu, regs->rip);
91586a3b 5711 kvm_set_rflags(vcpu, regs->rflags);
b6c7a5dc 5712
b4f14abd
JK		 5713 vcpu->arch.exception.pending = false;
5714
3842d135
AK		 5715 kvm_make_request(KVM_REQ_EVENT, vcpu);
5716
b6c7a5dc
HB		 5717 return 0;
5718}
5719
b6c7a5dc
HB		 5720void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5721{
5722 struct kvm_segment cs;
5723
3e6e0aab 5724 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
b6c7a5dc
HB		 5725 *db = cs.db;
5726 *l = cs.l;
5727}
5728EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5729
5730int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5731 struct kvm_sregs *sregs)
5732{
89a27f4d 5733 struct desc_ptr dt;
b6c7a5dc 5734
3e6e0aab
GT		 5735 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5736 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5737 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5738 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5739 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5740 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 5741
3e6e0aab
GT		 5742 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5743 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc
HB		 5744
5745 kvm_x86_ops->get_idt(vcpu, &dt);
89a27f4d
GN		 5746 sregs->idt.limit = dt.size;
5747 sregs->idt.base = dt.address;
b6c7a5dc 5748 kvm_x86_ops->get_gdt(vcpu, &dt);
89a27f4d
GN		 5749 sregs->gdt.limit = dt.size;
5750 sregs->gdt.base = dt.address;
b6c7a5dc 5751
4d4ec087 5752 sregs->cr0 = kvm_read_cr0(vcpu);
ad312c7c 5753 sregs->cr2 = vcpu->arch.cr2;
9f8fe504 5754 sregs->cr3 = kvm_read_cr3(vcpu);
fc78f519 5755 sregs->cr4 = kvm_read_cr4(vcpu);
2d3ad1f4 5756 sregs->cr8 = kvm_get_cr8(vcpu);
f6801dff 5757 sregs->efer = vcpu->arch.efer;
b6c7a5dc
HB		 5758 sregs->apic_base = kvm_get_apic_base(vcpu);
5759
923c61bb 5760 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
b6c7a5dc 5761
36752c9b 5762 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
14d0bc1f
GN		 5763 set_bit(vcpu->arch.interrupt.nr,
5764 (unsigned long *)sregs->interrupt_bitmap);
16d7a191 5765
b6c7a5dc
HB		 5766 return 0;
5767}
5768
62d9f0db
MT		 5769int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5770 struct kvm_mp_state *mp_state)
5771{
62d9f0db 5772 mp_state->mp_state = vcpu->arch.mp_state;
62d9f0db
MT		 5773 return 0;
5774}
5775
5776int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5777 struct kvm_mp_state *mp_state)
5778{
62d9f0db 5779 vcpu->arch.mp_state = mp_state->mp_state;
3842d135 5780 kvm_make_request(KVM_REQ_EVENT, vcpu);
62d9f0db
MT		 5781 return 0;
5782}
5783
e269fb21
JK		 5784int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5785 bool has_error_code, u32 error_code)
b6c7a5dc 5786{
4d2179e1 5787 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
8ec4722d 5788 int ret;
e01c2426 5789
8ec4722d 5790 init_emulate_ctxt(vcpu);
c697518a 5791
9aabc88f 5792 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
e269fb21
JK		 5793 tss_selector, reason, has_error_code,
5794 error_code);
c697518a 5795
c697518a 5796 if (ret)
19d04437 5797 return EMULATE_FAIL;
37817f29 5798
4d2179e1 5799 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
95c55886 5800 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
f6e78475 5801 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3842d135 5802 kvm_make_request(KVM_REQ_EVENT, vcpu);
19d04437 5803 return EMULATE_DONE;
37817f29
IE		 5804}
5805EXPORT_SYMBOL_GPL(kvm_task_switch);
5806
b6c7a5dc
HB		 5807int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5808 struct kvm_sregs *sregs)
5809{
5810 int mmu_reset_needed = 0;
63f42e02 5811 int pending_vec, max_bits, idx;
89a27f4d 5812 struct desc_ptr dt;
b6c7a5dc 5813
89a27f4d
GN		 5814 dt.size = sregs->idt.limit;
5815 dt.address = sregs->idt.base;
b6c7a5dc 5816 kvm_x86_ops->set_idt(vcpu, &dt);
89a27f4d
GN		 5817 dt.size = sregs->gdt.limit;
5818 dt.address = sregs->gdt.base;
b6c7a5dc
HB		 5819 kvm_x86_ops->set_gdt(vcpu, &dt);
5820
ad312c7c 5821 vcpu->arch.cr2 = sregs->cr2;
9f8fe504 5822 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
dc7e795e 5823 vcpu->arch.cr3 = sregs->cr3;
aff48baa 5824 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
b6c7a5dc 5825
2d3ad1f4 5826 kvm_set_cr8(vcpu, sregs->cr8);
b6c7a5dc 5827
f6801dff 5828 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
b6c7a5dc 5829 kvm_x86_ops->set_efer(vcpu, sregs->efer);
b6c7a5dc
HB		 5830 kvm_set_apic_base(vcpu, sregs->apic_base);
5831
4d4ec087 5832 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
b6c7a5dc 5833 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
d7306163 5834 vcpu->arch.cr0 = sregs->cr0;
b6c7a5dc 5835
fc78f519 5836 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
b6c7a5dc 5837 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
3ea3aa8c
SY		 5838 if (sregs->cr4 & X86_CR4_OSXSAVE)
5839 update_cpuid(vcpu);
63f42e02
XG		 5840
5841 idx = srcu_read_lock(&vcpu->kvm->srcu);
7c93be44 5842 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
9f8fe504 5843 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7c93be44
MT		 5844 mmu_reset_needed = 1;
5845 }
63f42e02 5846 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b6c7a5dc
HB		 5847
5848 if (mmu_reset_needed)
5849 kvm_mmu_reset_context(vcpu);
5850
923c61bb
GN		 5851 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5852 pending_vec = find_first_bit(
5853 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5854 if (pending_vec < max_bits) {
66fd3f7f 5855 kvm_queue_interrupt(vcpu, pending_vec, false);
923c61bb 5856 pr_debug("Set back pending irq %d\n", pending_vec);
b6c7a5dc
HB		 5857 }
5858
3e6e0aab
GT		 5859 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5860 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5861 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5862 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5863 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5864 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 5865
3e6e0aab
GT		 5866 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5867 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc 5868
5f0269f5
ME		 5869 update_cr8_intercept(vcpu);
5870
9c3e4aab 5871 /* Older userspace won't unhalt the vcpu on reset. */
c5af89b6 5872 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9c3e4aab 5873 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3eeb3288 5874 !is_protmode(vcpu))
9c3e4aab
MT		 5875 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5876
3842d135
AK		 5877 kvm_make_request(KVM_REQ_EVENT, vcpu);
5878
b6c7a5dc
HB		 5879 return 0;
5880}
5881
d0bfb940
JK		 5882int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5883 struct kvm_guest_debug *dbg)
b6c7a5dc 5884{
355be0b9 5885 unsigned long rflags;
ae675ef0 5886 int i, r;
b6c7a5dc 5887
4f926bf2
JK		 5888 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5889 r = -EBUSY;
5890 if (vcpu->arch.exception.pending)
2122ff5e 5891 goto out;
4f926bf2
JK		 5892 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5893 kvm_queue_exception(vcpu, DB_VECTOR);
5894 else
5895 kvm_queue_exception(vcpu, BP_VECTOR);
5896 }
5897
91586a3b
JK		 5898 /*
5899 * Read rflags as long as potentially injected trace flags are still
5900 * filtered out.
5901 */
5902 rflags = kvm_get_rflags(vcpu);
355be0b9
JK		 5903
5904 vcpu->guest_debug = dbg->control;
5905 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5906 vcpu->guest_debug = 0;
5907
5908 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
ae675ef0
JK		 5909 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5910 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5911 vcpu->arch.switch_db_regs =
5912 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5913 } else {
5914 for (i = 0; i < KVM_NR_DB_REGS; i++)
5915 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5916 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5917 }
5918
f92653ee
JK		 5919 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5920 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5921 get_segment_base(vcpu, VCPU_SREG_CS);
94fe45da 5922
91586a3b
JK		 5923 /*
5924 * Trigger an rflags update that will inject or remove the trace
5925 * flags.
5926 */
5927 kvm_set_rflags(vcpu, rflags);
b6c7a5dc 5928
355be0b9 5929 kvm_x86_ops->set_guest_debug(vcpu, dbg);
b6c7a5dc 5930
4f926bf2 5931 r = 0;
d0bfb940 5932
2122ff5e 5933out:
b6c7a5dc
HB		 5934
5935 return r;
5936}
5937
8b006791
ZX		 5938/*
5939 * Translate a guest virtual address to a guest physical address.
5940 */
5941int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5942 struct kvm_translation *tr)
5943{
5944 unsigned long vaddr = tr->linear_address;
5945 gpa_t gpa;
f656ce01 5946 int idx;
8b006791 5947
f656ce01 5948 idx = srcu_read_lock(&vcpu->kvm->srcu);
1871c602 5949 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
f656ce01 5950 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8b006791
ZX		 5951 tr->physical_address = gpa;
5952 tr->valid = gpa != UNMAPPED_GVA;
5953 tr->writeable = 1;
5954 tr->usermode = 0;
8b006791
ZX		 5955
5956 return 0;
5957}
5958
d0752060
HB		 5959int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5960{
98918833
SY		 5961 struct i387_fxsave_struct *fxsave =
5962 &vcpu->arch.guest_fpu.state->fxsave;
d0752060 5963
d0752060
HB		 5964 memcpy(fpu->fpr, fxsave->st_space, 128);
5965 fpu->fcw = fxsave->cwd;
5966 fpu->fsw = fxsave->swd;
5967 fpu->ftwx = fxsave->twd;
5968 fpu->last_opcode = fxsave->fop;
5969 fpu->last_ip = fxsave->rip;
5970 fpu->last_dp = fxsave->rdp;
5971 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5972
d0752060
HB		 5973 return 0;
5974}
5975
5976int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5977{
98918833
SY		 5978 struct i387_fxsave_struct *fxsave =
5979 &vcpu->arch.guest_fpu.state->fxsave;
d0752060 5980
d0752060
HB		 5981 memcpy(fxsave->st_space, fpu->fpr, 128);
5982 fxsave->cwd = fpu->fcw;
5983 fxsave->swd = fpu->fsw;
5984 fxsave->twd = fpu->ftwx;
5985 fxsave->fop = fpu->last_opcode;
5986 fxsave->rip = fpu->last_ip;
5987 fxsave->rdp = fpu->last_dp;
5988 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5989
d0752060
HB		 5990 return 0;
5991}
5992
10ab25cd 5993int fx_init(struct kvm_vcpu *vcpu)
d0752060 5994{
10ab25cd
JK		 5995 int err;
5996
5997 err = fpu_alloc(&vcpu->arch.guest_fpu);
5998 if (err)
5999 return err;
6000
98918833 6001 fpu_finit(&vcpu->arch.guest_fpu);
d0752060 6002
2acf923e
DC		 6003 /*
6004 * Ensure guest xcr0 is valid for loading
6005 */
6006 vcpu->arch.xcr0 = XSTATE_FP;
6007
ad312c7c 6008 vcpu->arch.cr0 |= X86_CR0_ET;
10ab25cd
JK		 6009
6010 return 0;
d0752060
HB		 6011}
6012EXPORT_SYMBOL_GPL(fx_init);
6013
98918833
SY		 6014static void fx_free(struct kvm_vcpu *vcpu)
6015{
6016 fpu_free(&vcpu->arch.guest_fpu);
6017}
6018
d0752060
HB		 6019void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6020{
2608d7a1 6021 if (vcpu->guest_fpu_loaded)
d0752060
HB		 6022 return;
6023
2acf923e
DC		 6024 /*
6025 * Restore all possible states in the guest,
6026 * and assume host would use all available bits.
6027 * Guest xcr0 would be loaded later.
6028 */
6029 kvm_put_guest_xcr0(vcpu);
d0752060 6030 vcpu->guest_fpu_loaded = 1;
7cf30855 6031 unlazy_fpu(current);
98918833 6032 fpu_restore_checking(&vcpu->arch.guest_fpu);
0c04851c 6033 trace_kvm_fpu(1);
d0752060 6034}
d0752060
HB		 6035
6036void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6037{
2acf923e
DC		 6038 kvm_put_guest_xcr0(vcpu);
6039
d0752060
HB		 6040 if (!vcpu->guest_fpu_loaded)
6041 return;
6042
6043 vcpu->guest_fpu_loaded = 0;
98918833 6044 fpu_save_init(&vcpu->arch.guest_fpu);
f096ed85 6045 ++vcpu->stat.fpu_reload;
a8eeb04a 6046 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
0c04851c 6047 trace_kvm_fpu(0);
d0752060 6048}
e9b11c17
ZX		 6049
6050void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6051{
12f9a48f 6052 kvmclock_reset(vcpu);
7f1ea208 6053
f5f48ee1 6054 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
98918833 6055 fx_free(vcpu);
e9b11c17
ZX		 6056 kvm_x86_ops->vcpu_free(vcpu);
6057}
6058
6059struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6060 unsigned int id)
6061{
6755bae8
ZA		 6062 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6063 printk_once(KERN_WARNING
6064 "kvm: SMP vm created on host with unstable TSC; "
6065 "guest TSC will not be reliable\n");
26e5215f
AK		 6066 return kvm_x86_ops->vcpu_create(kvm, id);
6067}
e9b11c17 6068
26e5215f
AK		 6069int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6070{
6071 int r;
e9b11c17 6072
0bed3b56 6073 vcpu->arch.mtrr_state.have_fixed = 1;
e9b11c17
ZX		 6074 vcpu_load(vcpu);
6075 r = kvm_arch_vcpu_reset(vcpu);
6076 if (r == 0)
6077 r = kvm_mmu_setup(vcpu);
6078 vcpu_put(vcpu);
6079 if (r < 0)
6080 goto free_vcpu;
6081
26e5215f 6082 return 0;
e9b11c17
ZX		 6083free_vcpu:
6084 kvm_x86_ops->vcpu_free(vcpu);
26e5215f 6085 return r;
e9b11c17
ZX		 6086}
6087
d40ccc62 6088void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
e9b11c17 6089{
344d9588
GN		 6090 vcpu->arch.apf.msr_val = 0;
6091
e9b11c17
ZX		 6092 vcpu_load(vcpu);
6093 kvm_mmu_unload(vcpu);
6094 vcpu_put(vcpu);
6095
98918833 6096 fx_free(vcpu);
e9b11c17
ZX		 6097 kvm_x86_ops->vcpu_free(vcpu);
6098}
6099
6100int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6101{
448fa4a9
JK		 6102 vcpu->arch.nmi_pending = false;
6103 vcpu->arch.nmi_injected = false;
6104
42dbaa5a
JK		 6105 vcpu->arch.switch_db_regs = 0;
6106 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6107 vcpu->arch.dr6 = DR6_FIXED_1;
6108 vcpu->arch.dr7 = DR7_FIXED_1;
6109
3842d135 6110 kvm_make_request(KVM_REQ_EVENT, vcpu);
344d9588 6111 vcpu->arch.apf.msr_val = 0;
3842d135 6112
12f9a48f
GC		 6113 kvmclock_reset(vcpu);
6114
af585b92
GN		 6115 kvm_clear_async_pf_completion_queue(vcpu);
6116 kvm_async_pf_hash_reset(vcpu);
6117 vcpu->arch.apf.halted = false;
3842d135 6118
e9b11c17
ZX		 6119 return kvm_x86_ops->vcpu_reset(vcpu);
6120}
6121
10474ae8 6122int kvm_arch_hardware_enable(void *garbage)
e9b11c17 6123{
ca84d1a2
ZA		 6124 struct kvm *kvm;
6125 struct kvm_vcpu *vcpu;
6126 int i;
18863bdd
AK		 6127
6128 kvm_shared_msr_cpu_online();
ca84d1a2
ZA		 6129 list_for_each_entry(kvm, &vm_list, vm_list)
6130 kvm_for_each_vcpu(i, vcpu, kvm)
6131 if (vcpu->cpu == smp_processor_id())
c285545f 6132 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10474ae8 6133 return kvm_x86_ops->hardware_enable(garbage);
e9b11c17
ZX		 6134}
6135
6136void kvm_arch_hardware_disable(void *garbage)
6137{
6138 kvm_x86_ops->hardware_disable(garbage);
3548bab5 6139 drop_user_return_notifiers(garbage);
e9b11c17
ZX		 6140}
6141
6142int kvm_arch_hardware_setup(void)
6143{
6144 return kvm_x86_ops->hardware_setup();
6145}
6146
6147void kvm_arch_hardware_unsetup(void)
6148{
6149 kvm_x86_ops->hardware_unsetup();
6150}
6151
6152void kvm_arch_check_processor_compat(void *rtn)
6153{
6154 kvm_x86_ops->check_processor_compatibility(rtn);
6155}
6156
6157int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6158{
6159 struct page *page;
6160 struct kvm *kvm;
6161 int r;
6162
6163 BUG_ON(vcpu->kvm == NULL);
6164 kvm = vcpu->kvm;
6165
9aabc88f 6166 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
14dfe855 6167 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
ad312c7c 6168 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
c30a358d 6169 vcpu->arch.mmu.translate_gpa = translate_gpa;
02f59dc9 6170 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
c5af89b6 6171 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
a4535290 6172 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
e9b11c17 6173 else
a4535290 6174 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
e9b11c17
ZX		 6175
6176 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6177 if (!page) {
6178 r = -ENOMEM;
6179 goto fail;
6180 }
ad312c7c 6181 vcpu->arch.pio_data = page_address(page);
e9b11c17 6182
1e993611 6183 kvm_init_tsc_catchup(vcpu, max_tsc_khz);
c285545f 6184
e9b11c17
ZX		 6185 r = kvm_mmu_create(vcpu);
6186 if (r < 0)
6187 goto fail_free_pio_data;
6188
6189 if (irqchip_in_kernel(kvm)) {
6190 r = kvm_create_lapic(vcpu);
6191 if (r < 0)
6192 goto fail_mmu_destroy;
6193 }
6194
890ca9ae
HY		 6195 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6196 GFP_KERNEL);
6197 if (!vcpu->arch.mce_banks) {
6198 r = -ENOMEM;
443c39bc 6199 goto fail_free_lapic;
890ca9ae
HY		 6200 }
6201 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6202
f5f48ee1
SY		 6203 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6204 goto fail_free_mce_banks;
6205
af585b92
GN		 6206 kvm_async_pf_hash_reset(vcpu);
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
36cb93fd 6225 kfree(vcpu->arch.mce_banks);
e9b11c17 6226 kvm_free_lapic(vcpu);
f656ce01 6227 idx = srcu_read_lock(&vcpu->kvm->srcu);
e9b11c17 6228 kvm_mmu_destroy(vcpu);
f656ce01 6229 srcu_read_unlock(&vcpu->kvm->srcu, idx);
ad312c7c 6230 free_page((unsigned long)vcpu->arch.pio_data);
e9b11c17 6231}
d19a9cd2 6232
d89f5eff 6233int kvm_arch_init_vm(struct kvm *kvm)
d19a9cd2 6234{
f05e70ac 6235 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4d5c5d0f 6236 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
d19a9cd2 6237
5550af4d
SY		 6238 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6239 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6240
038f8c11 6241 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
53f658b3 6242
d89f5eff 6243 return 0;
d19a9cd2
ZX		 6244}
6245
6246static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6247{
6248 vcpu_load(vcpu);
6249 kvm_mmu_unload(vcpu);
6250 vcpu_put(vcpu);
6251}
6252
6253static void kvm_free_vcpus(struct kvm *kvm)
6254{
6255 unsigned int i;
988a2cae 6256 struct kvm_vcpu *vcpu;
d19a9cd2
ZX		 6257
6258 /*
6259 * Unpin any mmu pages first.
6260 */
af585b92
GN		 6261 kvm_for_each_vcpu(i, vcpu, kvm) {
6262 kvm_clear_async_pf_completion_queue(vcpu);
988a2cae 6263 kvm_unload_vcpu_mmu(vcpu);
af585b92 6264 }
988a2cae
GN		 6265 kvm_for_each_vcpu(i, vcpu, kvm)
6266 kvm_arch_vcpu_free(vcpu);
6267
6268 mutex_lock(&kvm->lock);
6269 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6270 kvm->vcpus[i] = NULL;
d19a9cd2 6271
988a2cae
GN		 6272 atomic_set(&kvm->online_vcpus, 0);
6273 mutex_unlock(&kvm->lock);
d19a9cd2
ZX		 6274}
6275
ad8ba2cd
SY		 6276void kvm_arch_sync_events(struct kvm *kvm)
6277{
ba4cef31 6278 kvm_free_all_assigned_devices(kvm);
aea924f6 6279 kvm_free_pit(kvm);
ad8ba2cd
SY		 6280}
6281
d19a9cd2
ZX		 6282void kvm_arch_destroy_vm(struct kvm *kvm)
6283{
6eb55818 6284 kvm_iommu_unmap_guest(kvm);
d7deeeb0
ZX		 6285 kfree(kvm->arch.vpic);
6286 kfree(kvm->arch.vioapic);
d19a9cd2 6287 kvm_free_vcpus(kvm);
3d45830c
AK		 6288 if (kvm->arch.apic_access_page)
6289 put_page(kvm->arch.apic_access_page);
b7ebfb05
SY		 6290 if (kvm->arch.ept_identity_pagetable)
6291 put_page(kvm->arch.ept_identity_pagetable);
d19a9cd2 6292}
0de10343 6293
f7784b8e
MT		 6294int kvm_arch_prepare_memory_region(struct kvm *kvm,
6295 struct kvm_memory_slot *memslot,
0de10343 6296 struct kvm_memory_slot old,
f7784b8e 6297 struct kvm_userspace_memory_region *mem,
0de10343
ZX		 6298 int user_alloc)
6299{
f7784b8e 6300 int npages = memslot->npages;
7ac77099
AK		 6301 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6302
6303 /* Prevent internal slot pages from being moved by fork()/COW. */
6304 if (memslot->id >= KVM_MEMORY_SLOTS)
6305 map_flags = MAP_SHARED | MAP_ANONYMOUS;
0de10343
ZX		 6306
6307 /*To keep backward compatibility with older userspace,
6308 *x86 needs to hanlde !user_alloc case.
6309 */
6310 if (!user_alloc) {
6311 if (npages && !old.rmap) {
604b38ac
AA		 6312 unsigned long userspace_addr;
6313
72dc67a6 6314 down_write(&current->mm->mmap_sem);
604b38ac
AA		 6315 userspace_addr = do_mmap(NULL, 0,
6316 npages * PAGE_SIZE,
6317 PROT_READ | PROT_WRITE,
7ac77099 6318 map_flags,
604b38ac 6319 0);
72dc67a6 6320 up_write(&current->mm->mmap_sem);
0de10343 6321
604b38ac
AA		 6322 if (IS_ERR((void *)userspace_addr))
6323 return PTR_ERR((void *)userspace_addr);
6324
604b38ac 6325 memslot->userspace_addr = userspace_addr;
0de10343
ZX		 6326 }
6327 }
6328
f7784b8e
MT		 6329
6330 return 0;
6331}
6332
6333void kvm_arch_commit_memory_region(struct kvm *kvm,
6334 struct kvm_userspace_memory_region *mem,
6335 struct kvm_memory_slot old,
6336 int user_alloc)
6337{
6338
48c0e4e9 6339 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
f7784b8e
MT		 6340
6341 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6342 int ret;
6343
6344 down_write(&current->mm->mmap_sem);
6345 ret = do_munmap(current->mm, old.userspace_addr,
6346 old.npages * PAGE_SIZE);
6347 up_write(&current->mm->mmap_sem);
6348 if (ret < 0)
6349 printk(KERN_WARNING
6350 "kvm_vm_ioctl_set_memory_region: "
6351 "failed to munmap memory\n");
6352 }
6353
48c0e4e9
XG		 6354 if (!kvm->arch.n_requested_mmu_pages)
6355 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6356
7c8a83b7 6357 spin_lock(&kvm->mmu_lock);
48c0e4e9 6358 if (nr_mmu_pages)
0de10343 6359 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
0de10343 6360 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7c8a83b7 6361 spin_unlock(&kvm->mmu_lock);
0de10343 6362}
1d737c8a 6363
34d4cb8f
MT		 6364void kvm_arch_flush_shadow(struct kvm *kvm)
6365{
6366 kvm_mmu_zap_all(kvm);
8986ecc0 6367 kvm_reload_remote_mmus(kvm);
34d4cb8f
MT		 6368}
6369
1d737c8a
ZX		 6370int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6371{
af585b92
GN		 6372 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6373 !vcpu->arch.apf.halted)
6374 || !list_empty_careful(&vcpu->async_pf.done)
a1b37100
GN		 6375 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6376 || vcpu->arch.nmi_pending ||
6377 (kvm_arch_interrupt_allowed(vcpu) &&
6378 kvm_cpu_has_interrupt(vcpu));
1d737c8a 6379}
5736199a 6380
5736199a
ZX		 6381void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6382{
32f88400
MT		 6383 int me;
6384 int cpu = vcpu->cpu;
5736199a
ZX		 6385
6386 if (waitqueue_active(&vcpu->wq)) {
6387 wake_up_interruptible(&vcpu->wq);
6388 ++vcpu->stat.halt_wakeup;
6389 }
32f88400
MT		 6390
6391 me = get_cpu();
6392 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6b7e2d09 6393 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
32f88400 6394 smp_send_reschedule(cpu);
e9571ed5 6395 put_cpu();
5736199a 6396}
78646121
GN		 6397
6398int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6399{
6400 return kvm_x86_ops->interrupt_allowed(vcpu);
6401}
229456fc 6402
f92653ee
JK		 6403bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6404{
6405 unsigned long current_rip = kvm_rip_read(vcpu) +
6406 get_segment_base(vcpu, VCPU_SREG_CS);
6407
6408 return current_rip == linear_rip;
6409}
6410EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6411
94fe45da
JK		 6412unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6413{
6414 unsigned long rflags;
6415
6416 rflags = kvm_x86_ops->get_rflags(vcpu);
6417 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
c310bac5 6418 rflags &= ~X86_EFLAGS_TF;
94fe45da
JK		 6419 return rflags;
6420}
6421EXPORT_SYMBOL_GPL(kvm_get_rflags);
6422
6423void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6424{
6425 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
f92653ee 6426 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
c310bac5 6427 rflags |= X86_EFLAGS_TF;
94fe45da 6428 kvm_x86_ops->set_rflags(vcpu, rflags);
3842d135 6429 kvm_make_request(KVM_REQ_EVENT, vcpu);
94fe45da
JK		 6430}
6431EXPORT_SYMBOL_GPL(kvm_set_rflags);
6432
56028d08
GN		 6433void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6434{
6435 int r;
6436
fb67e14f 6437 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
c4806acd 6438 is_error_page(work->page))
56028d08
GN		 6439 return;
6440
6441 r = kvm_mmu_reload(vcpu);
6442 if (unlikely(r))
6443 return;
6444
fb67e14f
XG		 6445 if (!vcpu->arch.mmu.direct_map &&
6446 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6447 return;
6448
56028d08
GN		 6449 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6450}
6451
af585b92
GN		 6452static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6453{
6454 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6455}
6456
6457static inline u32 kvm_async_pf_next_probe(u32 key)
6458{
6459 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6460}
6461
6462static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6463{
6464 u32 key = kvm_async_pf_hash_fn(gfn);
6465
6466 while (vcpu->arch.apf.gfns[key] != ~0)
6467 key = kvm_async_pf_next_probe(key);
6468
6469 vcpu->arch.apf.gfns[key] = gfn;
6470}
6471
6472static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6473{
6474 int i;
6475 u32 key = kvm_async_pf_hash_fn(gfn);
6476
6477 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
c7d28c24
XG		 6478 (vcpu->arch.apf.gfns[key] != gfn &&
6479 vcpu->arch.apf.gfns[key] != ~0); i++)
af585b92
GN		 6480 key = kvm_async_pf_next_probe(key);
6481
6482 return key;
6483}
6484
6485bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6486{
6487 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6488}
6489
6490static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6491{
6492 u32 i, j, k;
6493
6494 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6495 while (true) {
6496 vcpu->arch.apf.gfns[i] = ~0;
6497 do {
6498 j = kvm_async_pf_next_probe(j);
6499 if (vcpu->arch.apf.gfns[j] == ~0)
6500 return;
6501 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6502 /*
6503 * k lies cyclically in ]i,j]
6504 * | i.k.j |
6505 * |....j i.k.| or |.k..j i...|
6506 */
6507 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6508 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6509 i = j;
6510 }
6511}
6512
7c90705b
GN		 6513static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6514{
6515
6516 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6517 sizeof(val));
6518}
6519
af585b92
GN		 6520void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6521 struct kvm_async_pf *work)
6522{
6389ee94
AK		 6523 struct x86_exception fault;
6524
7c90705b 6525 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
af585b92 6526 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7c90705b
GN		 6527
6528 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
fc5f06fa
GN		 6529 (vcpu->arch.apf.send_user_only &&
6530 kvm_x86_ops->get_cpl(vcpu) == 0))
7c90705b
GN		 6531 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6532 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6389ee94
AK		 6533 fault.vector = PF_VECTOR;
6534 fault.error_code_valid = true;
6535 fault.error_code = 0;
6536 fault.nested_page_fault = false;
6537 fault.address = work->arch.token;
6538 kvm_inject_page_fault(vcpu, &fault);
7c90705b 6539 }
af585b92
GN		 6540}
6541
6542void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6543 struct kvm_async_pf *work)
6544{
6389ee94
AK		 6545 struct x86_exception fault;
6546
7c90705b
GN		 6547 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6548 if (is_error_page(work->page))
6549 work->arch.token = ~0; /* broadcast wakeup */
6550 else
6551 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6552
6553 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6554 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6389ee94
AK		 6555 fault.vector = PF_VECTOR;
6556 fault.error_code_valid = true;
6557 fault.error_code = 0;
6558 fault.nested_page_fault = false;
6559 fault.address = work->arch.token;
6560 kvm_inject_page_fault(vcpu, &fault);
7c90705b 6561 }
e6d53e3b 6562 vcpu->arch.apf.halted = false;
7c90705b
GN		 6563}
6564
6565bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6566{
6567 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6568 return true;
6569 else
6570 return !kvm_event_needs_reinjection(vcpu) &&
6571 kvm_x86_ops->interrupt_allowed(vcpu);
af585b92
GN		 6572}
6573
229456fc
MT		 6574EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6575EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6576EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6577EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6578EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
0ac406de 6579EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
d8cabddf 6580EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
17897f36 6581EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
236649de 6582EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
ec1ff790 6583EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
532a46b9 6584EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
2e554e8d 6585EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
Linux kernel - Deliverables
This page took 0.926628 seconds and 5 git commands to generate.