]> bbs.cooldavid.org Git - net-next-2.6.git/blame - arch/x86/kvm/x86.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: Fix xsave and xcr save/restore memory leak
[net-next-2.6.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
221d059d 9 * Copyright 2010 Red Hat, Inc. and/or its affilates.
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>
aec51dc4 46#include <trace/events/kvm.h>
2ed152af 47
229456fc
MT		 48#define CREATE_TRACE_POINTS
49#include "trace.h"
043405e1 50
24f1e32c 51#include <asm/debugreg.h>
d825ed0a 52#include <asm/msr.h>
a5f61300 53#include <asm/desc.h>
0bed3b56 54#include <asm/mtrr.h>
890ca9ae 55#include <asm/mce.h>
7cf30855 56#include <asm/i387.h>
98918833 57#include <asm/xcr.h>
043405e1 58
313a3dc7 59#define MAX_IO_MSRS 256
a03490ed
CO		 60#define CR0_RESERVED_BITS \
61 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
62 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
63 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
64#define CR4_RESERVED_BITS \
65 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
66 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
67 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
2acf923e 68 | X86_CR4_OSXSAVE \
a03490ed
CO		 69 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
70
71#define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
890ca9ae
HY		 72
73#define KVM_MAX_MCE_BANKS 32
74#define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
75
50a37eb4
JR		 76/* EFER defaults:
77 * - enable syscall per default because its emulated by KVM
78 * - enable LME and LMA per default on 64 bit KVM
79 */
80#ifdef CONFIG_X86_64
81static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
82#else
83static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
84#endif
313a3dc7 85
ba1389b7
AK		 86#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
87#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
417bc304 88
cb142eb7 89static void update_cr8_intercept(struct kvm_vcpu *vcpu);
674eea0f
AK		 90static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
91 struct kvm_cpuid_entry2 __user *entries);
92
97896d04 93struct kvm_x86_ops *kvm_x86_ops;
5fdbf976 94EXPORT_SYMBOL_GPL(kvm_x86_ops);
97896d04 95
ed85c068
AP		 96int ignore_msrs = 0;
97module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
98
18863bdd
AK		 99#define KVM_NR_SHARED_MSRS 16
100
101struct kvm_shared_msrs_global {
102 int nr;
2bf78fa7 103 u32 msrs[KVM_NR_SHARED_MSRS];
18863bdd
AK		 104};
105
106struct kvm_shared_msrs {
107 struct user_return_notifier urn;
108 bool registered;
2bf78fa7
SY		 109 struct kvm_shared_msr_values {
110 u64 host;
111 u64 curr;
112 } values[KVM_NR_SHARED_MSRS];
18863bdd
AK		 113};
114
115static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
116static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
117
417bc304 118struct kvm_stats_debugfs_item debugfs_entries[] = {
ba1389b7
AK		 119 { "pf_fixed", VCPU_STAT(pf_fixed) },
120 { "pf_guest", VCPU_STAT(pf_guest) },
121 { "tlb_flush", VCPU_STAT(tlb_flush) },
122 { "invlpg", VCPU_STAT(invlpg) },
123 { "exits", VCPU_STAT(exits) },
124 { "io_exits", VCPU_STAT(io_exits) },
125 { "mmio_exits", VCPU_STAT(mmio_exits) },
126 { "signal_exits", VCPU_STAT(signal_exits) },
127 { "irq_window", VCPU_STAT(irq_window_exits) },
f08864b4 128 { "nmi_window", VCPU_STAT(nmi_window_exits) },
ba1389b7
AK		 129 { "halt_exits", VCPU_STAT(halt_exits) },
130 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
f11c3a8d 131 { "hypercalls", VCPU_STAT(hypercalls) },
ba1389b7
AK		 132 { "request_irq", VCPU_STAT(request_irq_exits) },
133 { "irq_exits", VCPU_STAT(irq_exits) },
134 { "host_state_reload", VCPU_STAT(host_state_reload) },
135 { "efer_reload", VCPU_STAT(efer_reload) },
136 { "fpu_reload", VCPU_STAT(fpu_reload) },
137 { "insn_emulation", VCPU_STAT(insn_emulation) },
138 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
fa89a817 139 { "irq_injections", VCPU_STAT(irq_injections) },
c4abb7c9 140 { "nmi_injections", VCPU_STAT(nmi_injections) },
4cee5764
AK		 141 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
142 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
143 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
144 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
145 { "mmu_flooded", VM_STAT(mmu_flooded) },
146 { "mmu_recycled", VM_STAT(mmu_recycled) },
dfc5aa00 147 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
4731d4c7 148 { "mmu_unsync", VM_STAT(mmu_unsync) },
0f74a24c 149 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
05da4558 150 { "largepages", VM_STAT(lpages) },
417bc304
HB		 151 { NULL }
152};
153
2acf923e
DC		 154u64 __read_mostly host_xcr0;
155
156static inline u32 bit(int bitno)
157{
158 return 1 << (bitno & 31);
159}
160
18863bdd
AK		 161static void kvm_on_user_return(struct user_return_notifier *urn)
162{
163 unsigned slot;
18863bdd
AK		 164 struct kvm_shared_msrs *locals
165 = container_of(urn, struct kvm_shared_msrs, urn);
2bf78fa7 166 struct kvm_shared_msr_values *values;
18863bdd
AK		 167
168 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
2bf78fa7
SY		 169 values = &locals->values[slot];
170 if (values->host != values->curr) {
171 wrmsrl(shared_msrs_global.msrs[slot], values->host);
172 values->curr = values->host;
18863bdd
AK		 173 }
174 }
175 locals->registered = false;
176 user_return_notifier_unregister(urn);
177}
178
2bf78fa7 179static void shared_msr_update(unsigned slot, u32 msr)
18863bdd 180{
2bf78fa7 181 struct kvm_shared_msrs *smsr;
18863bdd
AK		 182 u64 value;
183
2bf78fa7
SY		 184 smsr = &__get_cpu_var(shared_msrs);
185 /* only read, and nobody should modify it at this time,
186 * so don't need lock */
187 if (slot >= shared_msrs_global.nr) {
188 printk(KERN_ERR "kvm: invalid MSR slot!");
189 return;
190 }
191 rdmsrl_safe(msr, &value);
192 smsr->values[slot].host = value;
193 smsr->values[slot].curr = value;
194}
195
196void kvm_define_shared_msr(unsigned slot, u32 msr)
197{
18863bdd
AK		 198 if (slot >= shared_msrs_global.nr)
199 shared_msrs_global.nr = slot + 1;
2bf78fa7
SY		 200 shared_msrs_global.msrs[slot] = msr;
201 /* we need ensured the shared_msr_global have been updated */
202 smp_wmb();
18863bdd
AK		 203}
204EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
205
206static void kvm_shared_msr_cpu_online(void)
207{
208 unsigned i;
18863bdd
AK		 209
210 for (i = 0; i < shared_msrs_global.nr; ++i)
2bf78fa7 211 shared_msr_update(i, shared_msrs_global.msrs[i]);
18863bdd
AK		 212}
213
d5696725 214void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
18863bdd
AK		 215{
216 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
217
2bf78fa7 218 if (((value ^ smsr->values[slot].curr) & mask) == 0)
18863bdd 219 return;
2bf78fa7
SY		 220 smsr->values[slot].curr = value;
221 wrmsrl(shared_msrs_global.msrs[slot], value);
18863bdd
AK		 222 if (!smsr->registered) {
223 smsr->urn.on_user_return = kvm_on_user_return;
224 user_return_notifier_register(&smsr->urn);
225 smsr->registered = true;
226 }
227}
228EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
229
3548bab5
AK		 230static void drop_user_return_notifiers(void *ignore)
231{
232 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
233
234 if (smsr->registered)
235 kvm_on_user_return(&smsr->urn);
236}
237
6866b83e
CO		 238u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
239{
240 if (irqchip_in_kernel(vcpu->kvm))
ad312c7c 241 return vcpu->arch.apic_base;
6866b83e 242 else
ad312c7c 243 return vcpu->arch.apic_base;
6866b83e
CO		 244}
245EXPORT_SYMBOL_GPL(kvm_get_apic_base);
246
247void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
248{
249 /* TODO: reserve bits check */
250 if (irqchip_in_kernel(vcpu->kvm))
251 kvm_lapic_set_base(vcpu, data);
252 else
ad312c7c 253 vcpu->arch.apic_base = data;
6866b83e
CO		 254}
255EXPORT_SYMBOL_GPL(kvm_set_apic_base);
256
3fd28fce
ED		 257#define EXCPT_BENIGN 0
258#define EXCPT_CONTRIBUTORY 1
259#define EXCPT_PF 2
260
261static int exception_class(int vector)
262{
263 switch (vector) {
264 case PF_VECTOR:
265 return EXCPT_PF;
266 case DE_VECTOR:
267 case TS_VECTOR:
268 case NP_VECTOR:
269 case SS_VECTOR:
270 case GP_VECTOR:
271 return EXCPT_CONTRIBUTORY;
272 default:
273 break;
274 }
275 return EXCPT_BENIGN;
276}
277
278static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
ce7ddec4
JR		 279 unsigned nr, bool has_error, u32 error_code,
280 bool reinject)
3fd28fce
ED		 281{
282 u32 prev_nr;
283 int class1, class2;
284
285 if (!vcpu->arch.exception.pending) {
286 queue:
287 vcpu->arch.exception.pending = true;
288 vcpu->arch.exception.has_error_code = has_error;
289 vcpu->arch.exception.nr = nr;
290 vcpu->arch.exception.error_code = error_code;
3f0fd292 291 vcpu->arch.exception.reinject = reinject;
3fd28fce
ED		 292 return;
293 }
294
295 /* to check exception */
296 prev_nr = vcpu->arch.exception.nr;
297 if (prev_nr == DF_VECTOR) {
298 /* triple fault -> shutdown */
299 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
300 return;
301 }
302 class1 = exception_class(prev_nr);
303 class2 = exception_class(nr);
304 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
305 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
306 /* generate double fault per SDM Table 5-5 */
307 vcpu->arch.exception.pending = true;
308 vcpu->arch.exception.has_error_code = true;
309 vcpu->arch.exception.nr = DF_VECTOR;
310 vcpu->arch.exception.error_code = 0;
311 } else
312 /* replace previous exception with a new one in a hope
313 that instruction re-execution will regenerate lost
314 exception */
315 goto queue;
316}
317
298101da
AK		 318void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
319{
ce7ddec4 320 kvm_multiple_exception(vcpu, nr, false, 0, false);
298101da
AK		 321}
322EXPORT_SYMBOL_GPL(kvm_queue_exception);
323
ce7ddec4
JR		 324void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
325{
326 kvm_multiple_exception(vcpu, nr, false, 0, true);
327}
328EXPORT_SYMBOL_GPL(kvm_requeue_exception);
329
c3c91fee
AK		 330void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
331 u32 error_code)
332{
333 ++vcpu->stat.pf_guest;
ad312c7c 334 vcpu->arch.cr2 = addr;
c3c91fee
AK		 335 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
336}
337
3419ffc8
SY		 338void kvm_inject_nmi(struct kvm_vcpu *vcpu)
339{
340 vcpu->arch.nmi_pending = 1;
341}
342EXPORT_SYMBOL_GPL(kvm_inject_nmi);
343
298101da
AK		 344void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
345{
ce7ddec4 346 kvm_multiple_exception(vcpu, nr, true, error_code, false);
298101da
AK		 347}
348EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
349
ce7ddec4
JR		 350void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
351{
352 kvm_multiple_exception(vcpu, nr, true, error_code, true);
353}
354EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
355
0a79b009
AK		 356/*
357 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
358 * a #GP and return false.
359 */
360bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
298101da 361{
0a79b009
AK		 362 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
363 return true;
364 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
365 return false;
298101da 366}
0a79b009 367EXPORT_SYMBOL_GPL(kvm_require_cpl);
298101da 368
a03490ed
CO		 369/*
370 * Load the pae pdptrs. Return true is they are all valid.
371 */
372int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
373{
374 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
375 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
376 int i;
377 int ret;
ad312c7c 378 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
a03490ed 379
a03490ed
CO		 380 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
381 offset * sizeof(u64), sizeof(pdpte));
382 if (ret < 0) {
383 ret = 0;
384 goto out;
385 }
386 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
43a3795a 387 if (is_present_gpte(pdpte[i]) &&
20c466b5 388 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
a03490ed
CO		 389 ret = 0;
390 goto out;
391 }
392 }
393 ret = 1;
394
ad312c7c 395 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
6de4f3ad
AK		 396 __set_bit(VCPU_EXREG_PDPTR,
397 (unsigned long *)&vcpu->arch.regs_avail);
398 __set_bit(VCPU_EXREG_PDPTR,
399 (unsigned long *)&vcpu->arch.regs_dirty);
a03490ed 400out:
a03490ed
CO		 401
402 return ret;
403}
cc4b6871 404EXPORT_SYMBOL_GPL(load_pdptrs);
a03490ed 405
d835dfec
AK		 406static bool pdptrs_changed(struct kvm_vcpu *vcpu)
407{
ad312c7c 408 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
d835dfec
AK		 409 bool changed = true;
410 int r;
411
412 if (is_long_mode(vcpu) || !is_pae(vcpu))
413 return false;
414
6de4f3ad
AK		 415 if (!test_bit(VCPU_EXREG_PDPTR,
416 (unsigned long *)&vcpu->arch.regs_avail))
417 return true;
418
ad312c7c 419 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
d835dfec
AK		 420 if (r < 0)
421 goto out;
ad312c7c 422 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
d835dfec 423out:
d835dfec
AK		 424
425 return changed;
426}
427
49a9b07e 428int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
a03490ed 429{
aad82703
SY		 430 unsigned long old_cr0 = kvm_read_cr0(vcpu);
431 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
432 X86_CR0_CD | X86_CR0_NW;
433
f9a48e6a
AK		 434 cr0 |= X86_CR0_ET;
435
ab344828 436#ifdef CONFIG_X86_64
0f12244f
GN		 437 if (cr0 & 0xffffffff00000000UL)
438 return 1;
ab344828
GN		 439#endif
440
441 cr0 &= ~CR0_RESERVED_BITS;
a03490ed 442
0f12244f
GN		 443 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
444 return 1;
a03490ed 445
0f12244f
GN		 446 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
447 return 1;
a03490ed
CO		 448
449 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
450#ifdef CONFIG_X86_64
f6801dff 451 if ((vcpu->arch.efer & EFER_LME)) {
a03490ed
CO		 452 int cs_db, cs_l;
453
0f12244f
GN		 454 if (!is_pae(vcpu))
455 return 1;
a03490ed 456 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
0f12244f
GN		 457 if (cs_l)
458 return 1;
a03490ed
CO		 459 } else
460#endif
0f12244f
GN		 461 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3))
462 return 1;
a03490ed
CO		 463 }
464
465 kvm_x86_ops->set_cr0(vcpu, cr0);
a03490ed 466
aad82703
SY		 467 if ((cr0 ^ old_cr0) & update_bits)
468 kvm_mmu_reset_context(vcpu);
0f12244f
GN		 469 return 0;
470}
2d3ad1f4 471EXPORT_SYMBOL_GPL(kvm_set_cr0);
a03490ed 472
2d3ad1f4 473void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
a03490ed 474{
49a9b07e 475 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
a03490ed 476}
2d3ad1f4 477EXPORT_SYMBOL_GPL(kvm_lmsw);
a03490ed 478
2acf923e
DC		 479int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
480{
481 u64 xcr0;
482
483 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
484 if (index != XCR_XFEATURE_ENABLED_MASK)
485 return 1;
486 xcr0 = xcr;
487 if (kvm_x86_ops->get_cpl(vcpu) != 0)
488 return 1;
489 if (!(xcr0 & XSTATE_FP))
490 return 1;
491 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
492 return 1;
493 if (xcr0 & ~host_xcr0)
494 return 1;
495 vcpu->arch.xcr0 = xcr0;
496 vcpu->guest_xcr0_loaded = 0;
497 return 0;
498}
499
500int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
501{
502 if (__kvm_set_xcr(vcpu, index, xcr)) {
503 kvm_inject_gp(vcpu, 0);
504 return 1;
505 }
506 return 0;
507}
508EXPORT_SYMBOL_GPL(kvm_set_xcr);
509
510static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
511{
512 struct kvm_cpuid_entry2 *best;
513
514 best = kvm_find_cpuid_entry(vcpu, 1, 0);
515 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
516}
517
518static void update_cpuid(struct kvm_vcpu *vcpu)
519{
520 struct kvm_cpuid_entry2 *best;
521
522 best = kvm_find_cpuid_entry(vcpu, 1, 0);
523 if (!best)
524 return;
525
526 /* Update OSXSAVE bit */
527 if (cpu_has_xsave && best->function == 0x1) {
528 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
529 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
530 best->ecx |= bit(X86_FEATURE_OSXSAVE);
531 }
532}
533
a83b29c6 534int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
a03490ed 535{
fc78f519 536 unsigned long old_cr4 = kvm_read_cr4(vcpu);
a2edf57f
AK		 537 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
538
0f12244f
GN		 539 if (cr4 & CR4_RESERVED_BITS)
540 return 1;
a03490ed 541
2acf923e
DC		 542 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
543 return 1;
544
a03490ed 545 if (is_long_mode(vcpu)) {
0f12244f
GN		 546 if (!(cr4 & X86_CR4_PAE))
547 return 1;
a2edf57f
AK		 548 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
549 && ((cr4 ^ old_cr4) & pdptr_bits)
0f12244f
GN		 550 && !load_pdptrs(vcpu, vcpu->arch.cr3))
551 return 1;
552
553 if (cr4 & X86_CR4_VMXE)
554 return 1;
a03490ed 555
a03490ed 556 kvm_x86_ops->set_cr4(vcpu, cr4);
62ad0755 557
aad82703
SY		 558 if ((cr4 ^ old_cr4) & pdptr_bits)
559 kvm_mmu_reset_context(vcpu);
0f12244f 560
2acf923e
DC		 561 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
562 update_cpuid(vcpu);
563
0f12244f
GN		 564 return 0;
565}
2d3ad1f4 566EXPORT_SYMBOL_GPL(kvm_set_cr4);
a03490ed 567
2390218b 568int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
a03490ed 569{
ad312c7c 570 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
0ba73cda 571 kvm_mmu_sync_roots(vcpu);
d835dfec 572 kvm_mmu_flush_tlb(vcpu);
0f12244f 573 return 0;
d835dfec
AK		 574 }
575
a03490ed 576 if (is_long_mode(vcpu)) {
0f12244f
GN		 577 if (cr3 & CR3_L_MODE_RESERVED_BITS)
578 return 1;
a03490ed
CO		 579 } else {
580 if (is_pae(vcpu)) {
0f12244f
GN		 581 if (cr3 & CR3_PAE_RESERVED_BITS)
582 return 1;
583 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3))
584 return 1;
a03490ed
CO		 585 }
586 /*
587 * We don't check reserved bits in nonpae mode, because
588 * this isn't enforced, and VMware depends on this.
589 */
590 }
591
a03490ed
CO		 592 /*
593 * Does the new cr3 value map to physical memory? (Note, we
594 * catch an invalid cr3 even in real-mode, because it would
595 * cause trouble later on when we turn on paging anyway.)
596 *
597 * A real CPU would silently accept an invalid cr3 and would
598 * attempt to use it - with largely undefined (and often hard
599 * to debug) behavior on the guest side.
600 */
601 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
0f12244f
GN		 602 return 1;
603 vcpu->arch.cr3 = cr3;
604 vcpu->arch.mmu.new_cr3(vcpu);
605 return 0;
606}
2d3ad1f4 607EXPORT_SYMBOL_GPL(kvm_set_cr3);
a03490ed 608
0f12244f 609int __kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
a03490ed 610{
0f12244f
GN		 611 if (cr8 & CR8_RESERVED_BITS)
612 return 1;
a03490ed
CO		 613 if (irqchip_in_kernel(vcpu->kvm))
614 kvm_lapic_set_tpr(vcpu, cr8);
615 else
ad312c7c 616 vcpu->arch.cr8 = cr8;
0f12244f
GN		 617 return 0;
618}
619
620void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
621{
622 if (__kvm_set_cr8(vcpu, cr8))
623 kvm_inject_gp(vcpu, 0);
a03490ed 624}
2d3ad1f4 625EXPORT_SYMBOL_GPL(kvm_set_cr8);
a03490ed 626
2d3ad1f4 627unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
a03490ed
CO		 628{
629 if (irqchip_in_kernel(vcpu->kvm))
630 return kvm_lapic_get_cr8(vcpu);
631 else
ad312c7c 632 return vcpu->arch.cr8;
a03490ed 633}
2d3ad1f4 634EXPORT_SYMBOL_GPL(kvm_get_cr8);
a03490ed 635
338dbc97 636static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
020df079
GN		 637{
638 switch (dr) {
639 case 0 ... 3:
640 vcpu->arch.db[dr] = val;
641 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
642 vcpu->arch.eff_db[dr] = val;
643 break;
644 case 4:
338dbc97
GN		 645 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
646 return 1; /* #UD */
020df079
GN		 647 /* fall through */
648 case 6:
338dbc97
GN		 649 if (val & 0xffffffff00000000ULL)
650 return -1; /* #GP */
020df079
GN		 651 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
652 break;
653 case 5:
338dbc97
GN		 654 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
655 return 1; /* #UD */
020df079
GN		 656 /* fall through */
657 default: /* 7 */
338dbc97
GN		 658 if (val & 0xffffffff00000000ULL)
659 return -1; /* #GP */
020df079
GN		 660 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
661 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
662 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
663 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
664 }
665 break;
666 }
667
668 return 0;
669}
338dbc97
GN		 670
671int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
672{
673 int res;
674
675 res = __kvm_set_dr(vcpu, dr, val);
676 if (res > 0)
677 kvm_queue_exception(vcpu, UD_VECTOR);
678 else if (res < 0)
679 kvm_inject_gp(vcpu, 0);
680
681 return res;
682}
020df079
GN		 683EXPORT_SYMBOL_GPL(kvm_set_dr);
684
338dbc97 685static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
020df079
GN		 686{
687 switch (dr) {
688 case 0 ... 3:
689 *val = vcpu->arch.db[dr];
690 break;
691 case 4:
338dbc97 692 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
020df079 693 return 1;
020df079
GN		 694 /* fall through */
695 case 6:
696 *val = vcpu->arch.dr6;
697 break;
698 case 5:
338dbc97 699 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
020df079 700 return 1;
020df079
GN		 701 /* fall through */
702 default: /* 7 */
703 *val = vcpu->arch.dr7;
704 break;
705 }
706
707 return 0;
708}
338dbc97
GN		 709
710int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
711{
712 if (_kvm_get_dr(vcpu, dr, val)) {
713 kvm_queue_exception(vcpu, UD_VECTOR);
714 return 1;
715 }
716 return 0;
717}
020df079
GN		 718EXPORT_SYMBOL_GPL(kvm_get_dr);
719
043405e1
CO		 720/*
721 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
722 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
723 *
724 * This list is modified at module load time to reflect the
e3267cbb
GC		 725 * capabilities of the host cpu. This capabilities test skips MSRs that are
726 * kvm-specific. Those are put in the beginning of the list.
043405e1 727 */
e3267cbb 728
11c6bffa 729#define KVM_SAVE_MSRS_BEGIN 7
043405e1 730static u32 msrs_to_save[] = {
e3267cbb 731 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
11c6bffa 732 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
55cd8e5a 733 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
10388a07 734 HV_X64_MSR_APIC_ASSIST_PAGE,
043405e1
CO		 735 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
736 MSR_K6_STAR,
737#ifdef CONFIG_X86_64
738 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
739#endif
e3267cbb 740 MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
043405e1
CO		 741};
742
743static unsigned num_msrs_to_save;
744
745static u32 emulated_msrs[] = {
746 MSR_IA32_MISC_ENABLE,
747};
748
b69e8cae 749static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
15c4a640 750{
aad82703
SY		 751 u64 old_efer = vcpu->arch.efer;
752
b69e8cae
RJ		 753 if (efer & efer_reserved_bits)
754 return 1;
15c4a640
CO		 755
756 if (is_paging(vcpu)
b69e8cae
RJ		 757 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
758 return 1;
15c4a640 759
1b2fd70c
AG		 760 if (efer & EFER_FFXSR) {
761 struct kvm_cpuid_entry2 *feat;
762
763 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
b69e8cae
RJ		 764 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
765 return 1;
1b2fd70c
AG		 766 }
767
d8017474
AG		 768 if (efer & EFER_SVME) {
769 struct kvm_cpuid_entry2 *feat;
770
771 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
b69e8cae
RJ		 772 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
773 return 1;
d8017474
AG		 774 }
775
15c4a640 776 efer &= ~EFER_LMA;
f6801dff 777 efer |= vcpu->arch.efer & EFER_LMA;
15c4a640 778
a3d204e2
SY		 779 kvm_x86_ops->set_efer(vcpu, efer);
780
9645bb56
AK		 781 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
782 kvm_mmu_reset_context(vcpu);
b69e8cae 783
aad82703
SY		 784 /* Update reserved bits */
785 if ((efer ^ old_efer) & EFER_NX)
786 kvm_mmu_reset_context(vcpu);
787
b69e8cae 788 return 0;
15c4a640
CO		 789}
790
f2b4b7dd
JR		 791void kvm_enable_efer_bits(u64 mask)
792{
793 efer_reserved_bits &= ~mask;
794}
795EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
796
797
15c4a640
CO		 798/*
799 * Writes msr value into into the appropriate "register".
800 * Returns 0 on success, non-0 otherwise.
801 * Assumes vcpu_load() was already called.
802 */
803int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
804{
805 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
806}
807
313a3dc7
CO		 808/*
809 * Adapt set_msr() to msr_io()'s calling convention
810 */
811static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
812{
813 return kvm_set_msr(vcpu, index, *data);
814}
815
18068523
GOC		 816static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
817{
9ed3c444
AK		 818 int version;
819 int r;
50d0a0f9 820 struct pvclock_wall_clock wc;
923de3cf 821 struct timespec boot;
18068523
GOC		 822
823 if (!wall_clock)
824 return;
825
9ed3c444
AK		 826 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
827 if (r)
828 return;
829
830 if (version & 1)
831 ++version; /* first time write, random junk */
832
833 ++version;
18068523 834
18068523
GOC		 835 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
836
50d0a0f9
GH		 837 /*
838 * The guest calculates current wall clock time by adding
839 * system time (updated by kvm_write_guest_time below) to the
840 * wall clock specified here. guest system time equals host
841 * system time for us, thus we must fill in host boot time here.
842 */
923de3cf 843 getboottime(&boot);
50d0a0f9
GH		 844
845 wc.sec = boot.tv_sec;
846 wc.nsec = boot.tv_nsec;
847 wc.version = version;
18068523
GOC		 848
849 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
850
851 version++;
852 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
18068523
GOC		 853}
854
50d0a0f9
GH		 855static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
856{
857 uint32_t quotient, remainder;
858
859 /* Don't try to replace with do_div(), this one calculates
860 * "(dividend << 32) / divisor" */
861 __asm__ ( "divl %4"
862 : "=a" (quotient), "=d" (remainder)
863 : "0" (0), "1" (dividend), "r" (divisor) );
864 return quotient;
865}
866
867static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
868{
869 uint64_t nsecs = 1000000000LL;
870 int32_t shift = 0;
871 uint64_t tps64;
872 uint32_t tps32;
873
874 tps64 = tsc_khz * 1000LL;
875 while (tps64 > nsecs*2) {
876 tps64 >>= 1;
877 shift--;
878 }
879
880 tps32 = (uint32_t)tps64;
881 while (tps32 <= (uint32_t)nsecs) {
882 tps32 <<= 1;
883 shift++;
884 }
885
886 hv_clock->tsc_shift = shift;
887 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
888
889 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
80a914dc 890 __func__, tsc_khz, hv_clock->tsc_shift,
50d0a0f9
GH		 891 hv_clock->tsc_to_system_mul);
892}
893
c8076604
GH		 894static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
895
18068523
GOC		 896static void kvm_write_guest_time(struct kvm_vcpu *v)
897{
898 struct timespec ts;
899 unsigned long flags;
900 struct kvm_vcpu_arch *vcpu = &v->arch;
901 void *shared_kaddr;
463656c0 902 unsigned long this_tsc_khz;
18068523
GOC		 903
904 if ((!vcpu->time_page))
905 return;
906
463656c0
AK		 907 this_tsc_khz = get_cpu_var(cpu_tsc_khz);
908 if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
909 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
910 vcpu->hv_clock_tsc_khz = this_tsc_khz;
50d0a0f9 911 }
463656c0 912 put_cpu_var(cpu_tsc_khz);
50d0a0f9 913
18068523
GOC		 914 /* Keep irq disabled to prevent changes to the clock */
915 local_irq_save(flags);
af24a4e4 916 kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
18068523 917 ktime_get_ts(&ts);
923de3cf 918 monotonic_to_bootbased(&ts);
18068523
GOC		 919 local_irq_restore(flags);
920
921 /* With all the info we got, fill in the values */
922
923 vcpu->hv_clock.system_time = ts.tv_nsec +
afbcf7ab
GC		 924 (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
925
371bcf64
GC		 926 vcpu->hv_clock.flags = 0;
927
18068523
GOC		 928 /*
929 * The interface expects us to write an even number signaling that the
930 * update is finished. Since the guest won't see the intermediate
50d0a0f9 931 * state, we just increase by 2 at the end.
18068523 932 */
50d0a0f9 933 vcpu->hv_clock.version += 2;
18068523
GOC		 934
935 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
936
937 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
50d0a0f9 938 sizeof(vcpu->hv_clock));
18068523
GOC		 939
940 kunmap_atomic(shared_kaddr, KM_USER0);
941
942 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
943}
944
c8076604
GH		 945static int kvm_request_guest_time_update(struct kvm_vcpu *v)
946{
947 struct kvm_vcpu_arch *vcpu = &v->arch;
948
949 if (!vcpu->time_page)
950 return 0;
951 set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
952 return 1;
953}
954
9ba075a6
AK		 955static bool msr_mtrr_valid(unsigned msr)
956{
957 switch (msr) {
958 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
959 case MSR_MTRRfix64K_00000:
960 case MSR_MTRRfix16K_80000:
961 case MSR_MTRRfix16K_A0000:
962 case MSR_MTRRfix4K_C0000:
963 case MSR_MTRRfix4K_C8000:
964 case MSR_MTRRfix4K_D0000:
965 case MSR_MTRRfix4K_D8000:
966 case MSR_MTRRfix4K_E0000:
967 case MSR_MTRRfix4K_E8000:
968 case MSR_MTRRfix4K_F0000:
969 case MSR_MTRRfix4K_F8000:
970 case MSR_MTRRdefType:
971 case MSR_IA32_CR_PAT:
972 return true;
973 case 0x2f8:
974 return true;
975 }
976 return false;
977}
978
d6289b93
MT		 979static bool valid_pat_type(unsigned t)
980{
981 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
982}
983
984static bool valid_mtrr_type(unsigned t)
985{
986 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
987}
988
989static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
990{
991 int i;
992
993 if (!msr_mtrr_valid(msr))
994 return false;
995
996 if (msr == MSR_IA32_CR_PAT) {
997 for (i = 0; i < 8; i++)
998 if (!valid_pat_type((data >> (i * 8)) & 0xff))
999 return false;
1000 return true;
1001 } else if (msr == MSR_MTRRdefType) {
1002 if (data & ~0xcff)
1003 return false;
1004 return valid_mtrr_type(data & 0xff);
1005 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1006 for (i = 0; i < 8 ; i++)
1007 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1008 return false;
1009 return true;
1010 }
1011
1012 /* variable MTRRs */
1013 return valid_mtrr_type(data & 0xff);
1014}
1015
9ba075a6
AK		 1016static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1017{
0bed3b56
SY		 1018 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1019
d6289b93 1020 if (!mtrr_valid(vcpu, msr, data))
9ba075a6
AK		 1021 return 1;
1022
0bed3b56
SY		 1023 if (msr == MSR_MTRRdefType) {
1024 vcpu->arch.mtrr_state.def_type = data;
1025 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1026 } else if (msr == MSR_MTRRfix64K_00000)
1027 p[0] = data;
1028 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1029 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1030 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1031 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1032 else if (msr == MSR_IA32_CR_PAT)
1033 vcpu->arch.pat = data;
1034 else { /* Variable MTRRs */
1035 int idx, is_mtrr_mask;
1036 u64 *pt;
1037
1038 idx = (msr - 0x200) / 2;
1039 is_mtrr_mask = msr - 0x200 - 2 * idx;
1040 if (!is_mtrr_mask)
1041 pt =
1042 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1043 else
1044 pt =
1045 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1046 *pt = data;
1047 }
1048
1049 kvm_mmu_reset_context(vcpu);
9ba075a6
AK		 1050 return 0;
1051}
15c4a640 1052
890ca9ae 1053static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
15c4a640 1054{
890ca9ae
HY		 1055 u64 mcg_cap = vcpu->arch.mcg_cap;
1056 unsigned bank_num = mcg_cap & 0xff;
1057
15c4a640 1058 switch (msr) {
15c4a640 1059 case MSR_IA32_MCG_STATUS:
890ca9ae 1060 vcpu->arch.mcg_status = data;
15c4a640 1061 break;
c7ac679c 1062 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 1063 if (!(mcg_cap & MCG_CTL_P))
1064 return 1;
1065 if (data != 0 && data != ~(u64)0)
1066 return -1;
1067 vcpu->arch.mcg_ctl = data;
1068 break;
1069 default:
1070 if (msr >= MSR_IA32_MC0_CTL &&
1071 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1072 u32 offset = msr - MSR_IA32_MC0_CTL;
114be429
AP		 1073 /* only 0 or all 1s can be written to IA32_MCi_CTL
1074 * some Linux kernels though clear bit 10 in bank 4 to
1075 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1076 * this to avoid an uncatched #GP in the guest
1077 */
890ca9ae 1078 if ((offset & 0x3) == 0 &&
114be429 1079 data != 0 && (data | (1 << 10)) != ~(u64)0)
890ca9ae
HY		 1080 return -1;
1081 vcpu->arch.mce_banks[offset] = data;
1082 break;
1083 }
1084 return 1;
1085 }
1086 return 0;
1087}
1088
ffde22ac
ES		 1089static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1090{
1091 struct kvm *kvm = vcpu->kvm;
1092 int lm = is_long_mode(vcpu);
1093 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1094 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1095 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1096 : kvm->arch.xen_hvm_config.blob_size_32;
1097 u32 page_num = data & ~PAGE_MASK;
1098 u64 page_addr = data & PAGE_MASK;
1099 u8 *page;
1100 int r;
1101
1102 r = -E2BIG;
1103 if (page_num >= blob_size)
1104 goto out;
1105 r = -ENOMEM;
1106 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1107 if (!page)
1108 goto out;
1109 r = -EFAULT;
1110 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1111 goto out_free;
1112 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1113 goto out_free;
1114 r = 0;
1115out_free:
1116 kfree(page);
1117out:
1118 return r;
1119}
1120
55cd8e5a
GN		 1121static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1122{
1123 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1124}
1125
1126static bool kvm_hv_msr_partition_wide(u32 msr)
1127{
1128 bool r = false;
1129 switch (msr) {
1130 case HV_X64_MSR_GUEST_OS_ID:
1131 case HV_X64_MSR_HYPERCALL:
1132 r = true;
1133 break;
1134 }
1135
1136 return r;
1137}
1138
1139static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1140{
1141 struct kvm *kvm = vcpu->kvm;
1142
1143 switch (msr) {
1144 case HV_X64_MSR_GUEST_OS_ID:
1145 kvm->arch.hv_guest_os_id = data;
1146 /* setting guest os id to zero disables hypercall page */
1147 if (!kvm->arch.hv_guest_os_id)
1148 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1149 break;
1150 case HV_X64_MSR_HYPERCALL: {
1151 u64 gfn;
1152 unsigned long addr;
1153 u8 instructions[4];
1154
1155 /* if guest os id is not set hypercall should remain disabled */
1156 if (!kvm->arch.hv_guest_os_id)
1157 break;
1158 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1159 kvm->arch.hv_hypercall = data;
1160 break;
1161 }
1162 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1163 addr = gfn_to_hva(kvm, gfn);
1164 if (kvm_is_error_hva(addr))
1165 return 1;
1166 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1167 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1168 if (copy_to_user((void __user *)addr, instructions, 4))
1169 return 1;
1170 kvm->arch.hv_hypercall = data;
1171 break;
1172 }
1173 default:
1174 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1175 "data 0x%llx\n", msr, data);
1176 return 1;
1177 }
1178 return 0;
1179}
1180
1181static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1182{
10388a07
GN		 1183 switch (msr) {
1184 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1185 unsigned long addr;
55cd8e5a 1186
10388a07
GN		 1187 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1188 vcpu->arch.hv_vapic = data;
1189 break;
1190 }
1191 addr = gfn_to_hva(vcpu->kvm, data >>
1192 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1193 if (kvm_is_error_hva(addr))
1194 return 1;
1195 if (clear_user((void __user *)addr, PAGE_SIZE))
1196 return 1;
1197 vcpu->arch.hv_vapic = data;
1198 break;
1199 }
1200 case HV_X64_MSR_EOI:
1201 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1202 case HV_X64_MSR_ICR:
1203 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1204 case HV_X64_MSR_TPR:
1205 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1206 default:
1207 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1208 "data 0x%llx\n", msr, data);
1209 return 1;
1210 }
1211
1212 return 0;
55cd8e5a
GN		 1213}
1214
15c4a640
CO		 1215int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1216{
1217 switch (msr) {
15c4a640 1218 case MSR_EFER:
b69e8cae 1219 return set_efer(vcpu, data);
8f1589d9
AP		 1220 case MSR_K7_HWCR:
1221 data &= ~(u64)0x40; /* ignore flush filter disable */
82494028 1222 data &= ~(u64)0x100; /* ignore ignne emulation enable */
8f1589d9
AP		 1223 if (data != 0) {
1224 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1225 data);
1226 return 1;
1227 }
15c4a640 1228 break;
f7c6d140
AP		 1229 case MSR_FAM10H_MMIO_CONF_BASE:
1230 if (data != 0) {
1231 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1232 "0x%llx\n", data);
1233 return 1;
1234 }
15c4a640 1235 break;
c323c0e5 1236 case MSR_AMD64_NB_CFG:
c7ac679c 1237 break;
b5e2fec0
AG		 1238 case MSR_IA32_DEBUGCTLMSR:
1239 if (!data) {
1240 /* We support the non-activated case already */
1241 break;
1242 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1243 /* Values other than LBR and BTF are vendor-specific,
1244 thus reserved and should throw a #GP */
1245 return 1;
1246 }
1247 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1248 __func__, data);
1249 break;
15c4a640
CO		 1250 case MSR_IA32_UCODE_REV:
1251 case MSR_IA32_UCODE_WRITE:
61a6bd67 1252 case MSR_VM_HSAVE_PA:
6098ca93 1253 case MSR_AMD64_PATCH_LOADER:
15c4a640 1254 break;
9ba075a6
AK		 1255 case 0x200 ... 0x2ff:
1256 return set_msr_mtrr(vcpu, msr, data);
15c4a640
CO		 1257 case MSR_IA32_APICBASE:
1258 kvm_set_apic_base(vcpu, data);
1259 break;
0105d1a5
GN		 1260 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1261 return kvm_x2apic_msr_write(vcpu, msr, data);
15c4a640 1262 case MSR_IA32_MISC_ENABLE:
ad312c7c 1263 vcpu->arch.ia32_misc_enable_msr = data;
15c4a640 1264 break;
11c6bffa 1265 case MSR_KVM_WALL_CLOCK_NEW:
18068523
GOC		 1266 case MSR_KVM_WALL_CLOCK:
1267 vcpu->kvm->arch.wall_clock = data;
1268 kvm_write_wall_clock(vcpu->kvm, data);
1269 break;
11c6bffa 1270 case MSR_KVM_SYSTEM_TIME_NEW:
18068523
GOC		 1271 case MSR_KVM_SYSTEM_TIME: {
1272 if (vcpu->arch.time_page) {
1273 kvm_release_page_dirty(vcpu->arch.time_page);
1274 vcpu->arch.time_page = NULL;
1275 }
1276
1277 vcpu->arch.time = data;
1278
1279 /* we verify if the enable bit is set... */
1280 if (!(data & 1))
1281 break;
1282
1283 /* ...but clean it before doing the actual write */
1284 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1285
18068523
GOC		 1286 vcpu->arch.time_page =
1287 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
18068523
GOC		 1288
1289 if (is_error_page(vcpu->arch.time_page)) {
1290 kvm_release_page_clean(vcpu->arch.time_page);
1291 vcpu->arch.time_page = NULL;
1292 }
1293
c8076604 1294 kvm_request_guest_time_update(vcpu);
18068523
GOC		 1295 break;
1296 }
890ca9ae
HY		 1297 case MSR_IA32_MCG_CTL:
1298 case MSR_IA32_MCG_STATUS:
1299 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1300 return set_msr_mce(vcpu, msr, data);
71db6023
AP		 1301
1302 /* Performance counters are not protected by a CPUID bit,
1303 * so we should check all of them in the generic path for the sake of
1304 * cross vendor migration.
1305 * Writing a zero into the event select MSRs disables them,
1306 * which we perfectly emulate ;-). Any other value should be at least
1307 * reported, some guests depend on them.
1308 */
1309 case MSR_P6_EVNTSEL0:
1310 case MSR_P6_EVNTSEL1:
1311 case MSR_K7_EVNTSEL0:
1312 case MSR_K7_EVNTSEL1:
1313 case MSR_K7_EVNTSEL2:
1314 case MSR_K7_EVNTSEL3:
1315 if (data != 0)
1316 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1317 "0x%x data 0x%llx\n", msr, data);
1318 break;
1319 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1320 * so we ignore writes to make it happy.
1321 */
1322 case MSR_P6_PERFCTR0:
1323 case MSR_P6_PERFCTR1:
1324 case MSR_K7_PERFCTR0:
1325 case MSR_K7_PERFCTR1:
1326 case MSR_K7_PERFCTR2:
1327 case MSR_K7_PERFCTR3:
1328 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1329 "0x%x data 0x%llx\n", msr, data);
1330 break;
55cd8e5a
GN		 1331 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1332 if (kvm_hv_msr_partition_wide(msr)) {
1333 int r;
1334 mutex_lock(&vcpu->kvm->lock);
1335 r = set_msr_hyperv_pw(vcpu, msr, data);
1336 mutex_unlock(&vcpu->kvm->lock);
1337 return r;
1338 } else
1339 return set_msr_hyperv(vcpu, msr, data);
1340 break;
15c4a640 1341 default:
ffde22ac
ES		 1342 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1343 return xen_hvm_config(vcpu, data);
ed85c068
AP		 1344 if (!ignore_msrs) {
1345 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1346 msr, data);
1347 return 1;
1348 } else {
1349 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1350 msr, data);
1351 break;
1352 }
15c4a640
CO		 1353 }
1354 return 0;
1355}
1356EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1357
1358
1359/*
1360 * Reads an msr value (of 'msr_index') into 'pdata'.
1361 * Returns 0 on success, non-0 otherwise.
1362 * Assumes vcpu_load() was already called.
1363 */
1364int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1365{
1366 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1367}
1368
9ba075a6
AK		 1369static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1370{
0bed3b56
SY		 1371 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1372
9ba075a6
AK		 1373 if (!msr_mtrr_valid(msr))
1374 return 1;
1375
0bed3b56
SY		 1376 if (msr == MSR_MTRRdefType)
1377 *pdata = vcpu->arch.mtrr_state.def_type +
1378 (vcpu->arch.mtrr_state.enabled << 10);
1379 else if (msr == MSR_MTRRfix64K_00000)
1380 *pdata = p[0];
1381 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1382 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1383 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1384 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1385 else if (msr == MSR_IA32_CR_PAT)
1386 *pdata = vcpu->arch.pat;
1387 else { /* Variable MTRRs */
1388 int idx, is_mtrr_mask;
1389 u64 *pt;
1390
1391 idx = (msr - 0x200) / 2;
1392 is_mtrr_mask = msr - 0x200 - 2 * idx;
1393 if (!is_mtrr_mask)
1394 pt =
1395 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1396 else
1397 pt =
1398 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1399 *pdata = *pt;
1400 }
1401
9ba075a6
AK		 1402 return 0;
1403}
1404
890ca9ae 1405static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
15c4a640
CO		 1406{
1407 u64 data;
890ca9ae
HY		 1408 u64 mcg_cap = vcpu->arch.mcg_cap;
1409 unsigned bank_num = mcg_cap & 0xff;
15c4a640
CO		 1410
1411 switch (msr) {
15c4a640
CO		 1412 case MSR_IA32_P5_MC_ADDR:
1413 case MSR_IA32_P5_MC_TYPE:
890ca9ae
HY		 1414 data = 0;
1415 break;
15c4a640 1416 case MSR_IA32_MCG_CAP:
890ca9ae
HY		 1417 data = vcpu->arch.mcg_cap;
1418 break;
c7ac679c 1419 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 1420 if (!(mcg_cap & MCG_CTL_P))
1421 return 1;
1422 data = vcpu->arch.mcg_ctl;
1423 break;
1424 case MSR_IA32_MCG_STATUS:
1425 data = vcpu->arch.mcg_status;
1426 break;
1427 default:
1428 if (msr >= MSR_IA32_MC0_CTL &&
1429 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1430 u32 offset = msr - MSR_IA32_MC0_CTL;
1431 data = vcpu->arch.mce_banks[offset];
1432 break;
1433 }
1434 return 1;
1435 }
1436 *pdata = data;
1437 return 0;
1438}
1439
55cd8e5a
GN		 1440static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1441{
1442 u64 data = 0;
1443 struct kvm *kvm = vcpu->kvm;
1444
1445 switch (msr) {
1446 case HV_X64_MSR_GUEST_OS_ID:
1447 data = kvm->arch.hv_guest_os_id;
1448 break;
1449 case HV_X64_MSR_HYPERCALL:
1450 data = kvm->arch.hv_hypercall;
1451 break;
1452 default:
1453 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1454 return 1;
1455 }
1456
1457 *pdata = data;
1458 return 0;
1459}
1460
1461static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1462{
1463 u64 data = 0;
1464
1465 switch (msr) {
1466 case HV_X64_MSR_VP_INDEX: {
1467 int r;
1468 struct kvm_vcpu *v;
1469 kvm_for_each_vcpu(r, v, vcpu->kvm)
1470 if (v == vcpu)
1471 data = r;
1472 break;
1473 }
10388a07
GN		 1474 case HV_X64_MSR_EOI:
1475 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1476 case HV_X64_MSR_ICR:
1477 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1478 case HV_X64_MSR_TPR:
1479 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
55cd8e5a
GN		 1480 default:
1481 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1482 return 1;
1483 }
1484 *pdata = data;
1485 return 0;
1486}
1487
890ca9ae
HY		 1488int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1489{
1490 u64 data;
1491
1492 switch (msr) {
890ca9ae 1493 case MSR_IA32_PLATFORM_ID:
15c4a640 1494 case MSR_IA32_UCODE_REV:
15c4a640 1495 case MSR_IA32_EBL_CR_POWERON:
b5e2fec0
AG		 1496 case MSR_IA32_DEBUGCTLMSR:
1497 case MSR_IA32_LASTBRANCHFROMIP:
1498 case MSR_IA32_LASTBRANCHTOIP:
1499 case MSR_IA32_LASTINTFROMIP:
1500 case MSR_IA32_LASTINTTOIP:
60af2ecd
JSR		 1501 case MSR_K8_SYSCFG:
1502 case MSR_K7_HWCR:
61a6bd67 1503 case MSR_VM_HSAVE_PA:
1f3ee616
AS		 1504 case MSR_P6_PERFCTR0:
1505 case MSR_P6_PERFCTR1:
7fe29e0f
AS		 1506 case MSR_P6_EVNTSEL0:
1507 case MSR_P6_EVNTSEL1:
9e699624 1508 case MSR_K7_EVNTSEL0:
1f3ee616 1509 case MSR_K7_PERFCTR0:
1fdbd48c 1510 case MSR_K8_INT_PENDING_MSG:
c323c0e5 1511 case MSR_AMD64_NB_CFG:
f7c6d140 1512 case MSR_FAM10H_MMIO_CONF_BASE:
15c4a640
CO		 1513 data = 0;
1514 break;
9ba075a6
AK		 1515 case MSR_MTRRcap:
1516 data = 0x500 | KVM_NR_VAR_MTRR;
1517 break;
1518 case 0x200 ... 0x2ff:
1519 return get_msr_mtrr(vcpu, msr, pdata);
15c4a640
CO		 1520 case 0xcd: /* fsb frequency */
1521 data = 3;
1522 break;
1523 case MSR_IA32_APICBASE:
1524 data = kvm_get_apic_base(vcpu);
1525 break;
0105d1a5
GN		 1526 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1527 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1528 break;
15c4a640 1529 case MSR_IA32_MISC_ENABLE:
ad312c7c 1530 data = vcpu->arch.ia32_misc_enable_msr;
15c4a640 1531 break;
847f0ad8
AG		 1532 case MSR_IA32_PERF_STATUS:
1533 /* TSC increment by tick */
1534 data = 1000ULL;
1535 /* CPU multiplier */
1536 data |= (((uint64_t)4ULL) << 40);
1537 break;
15c4a640 1538 case MSR_EFER:
f6801dff 1539 data = vcpu->arch.efer;
15c4a640 1540 break;
18068523 1541 case MSR_KVM_WALL_CLOCK:
11c6bffa 1542 case MSR_KVM_WALL_CLOCK_NEW:
18068523
GOC		 1543 data = vcpu->kvm->arch.wall_clock;
1544 break;
1545 case MSR_KVM_SYSTEM_TIME:
11c6bffa 1546 case MSR_KVM_SYSTEM_TIME_NEW:
18068523
GOC		 1547 data = vcpu->arch.time;
1548 break;
890ca9ae
HY		 1549 case MSR_IA32_P5_MC_ADDR:
1550 case MSR_IA32_P5_MC_TYPE:
1551 case MSR_IA32_MCG_CAP:
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 get_msr_mce(vcpu, msr, pdata);
55cd8e5a
GN		 1556 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1557 if (kvm_hv_msr_partition_wide(msr)) {
1558 int r;
1559 mutex_lock(&vcpu->kvm->lock);
1560 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1561 mutex_unlock(&vcpu->kvm->lock);
1562 return r;
1563 } else
1564 return get_msr_hyperv(vcpu, msr, pdata);
1565 break;
15c4a640 1566 default:
ed85c068
AP		 1567 if (!ignore_msrs) {
1568 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1569 return 1;
1570 } else {
1571 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1572 data = 0;
1573 }
1574 break;
15c4a640
CO		 1575 }
1576 *pdata = data;
1577 return 0;
1578}
1579EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1580
313a3dc7
CO		 1581/*
1582 * Read or write a bunch of msrs. All parameters are kernel addresses.
1583 *
1584 * @return number of msrs set successfully.
1585 */
1586static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1587 struct kvm_msr_entry *entries,
1588 int (*do_msr)(struct kvm_vcpu *vcpu,
1589 unsigned index, u64 *data))
1590{
f656ce01 1591 int i, idx;
313a3dc7 1592
f656ce01 1593 idx = srcu_read_lock(&vcpu->kvm->srcu);
313a3dc7
CO		 1594 for (i = 0; i < msrs->nmsrs; ++i)
1595 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1596 break;
f656ce01 1597 srcu_read_unlock(&vcpu->kvm->srcu, idx);
313a3dc7 1598
313a3dc7
CO		 1599 return i;
1600}
1601
1602/*
1603 * Read or write a bunch of msrs. Parameters are user addresses.
1604 *
1605 * @return number of msrs set successfully.
1606 */
1607static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1608 int (*do_msr)(struct kvm_vcpu *vcpu,
1609 unsigned index, u64 *data),
1610 int writeback)
1611{
1612 struct kvm_msrs msrs;
1613 struct kvm_msr_entry *entries;
1614 int r, n;
1615 unsigned size;
1616
1617 r = -EFAULT;
1618 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1619 goto out;
1620
1621 r = -E2BIG;
1622 if (msrs.nmsrs >= MAX_IO_MSRS)
1623 goto out;
1624
1625 r = -ENOMEM;
1626 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
7a73c028 1627 entries = kmalloc(size, GFP_KERNEL);
313a3dc7
CO		 1628 if (!entries)
1629 goto out;
1630
1631 r = -EFAULT;
1632 if (copy_from_user(entries, user_msrs->entries, size))
1633 goto out_free;
1634
1635 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1636 if (r < 0)
1637 goto out_free;
1638
1639 r = -EFAULT;
1640 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1641 goto out_free;
1642
1643 r = n;
1644
1645out_free:
7a73c028 1646 kfree(entries);
313a3dc7
CO		 1647out:
1648 return r;
1649}
1650
018d00d2
ZX		 1651int kvm_dev_ioctl_check_extension(long ext)
1652{
1653 int r;
1654
1655 switch (ext) {
1656 case KVM_CAP_IRQCHIP:
1657 case KVM_CAP_HLT:
1658 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
018d00d2 1659 case KVM_CAP_SET_TSS_ADDR:
07716717 1660 case KVM_CAP_EXT_CPUID:
c8076604 1661 case KVM_CAP_CLOCKSOURCE:
7837699f 1662 case KVM_CAP_PIT:
a28e4f5a 1663 case KVM_CAP_NOP_IO_DELAY:
62d9f0db 1664 case KVM_CAP_MP_STATE:
ed848624 1665 case KVM_CAP_SYNC_MMU:
52d939a0 1666 case KVM_CAP_REINJECT_CONTROL:
4925663a 1667 case KVM_CAP_IRQ_INJECT_STATUS:
e56d532f 1668 case KVM_CAP_ASSIGN_DEV_IRQ:
721eecbf 1669 case KVM_CAP_IRQFD:
d34e6b17 1670 case KVM_CAP_IOEVENTFD:
c5ff41ce 1671 case KVM_CAP_PIT2:
e9f42757 1672 case KVM_CAP_PIT_STATE2:
b927a3ce 1673 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
ffde22ac 1674 case KVM_CAP_XEN_HVM:
afbcf7ab 1675 case KVM_CAP_ADJUST_CLOCK:
3cfc3092 1676 case KVM_CAP_VCPU_EVENTS:
55cd8e5a 1677 case KVM_CAP_HYPERV:
10388a07 1678 case KVM_CAP_HYPERV_VAPIC:
c25bc163 1679 case KVM_CAP_HYPERV_SPIN:
ab9f4ecb 1680 case KVM_CAP_PCI_SEGMENT:
a1efbe77 1681 case KVM_CAP_DEBUGREGS:
d2be1651 1682 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2d5b5a66 1683 case KVM_CAP_XSAVE:
018d00d2
ZX		 1684 r = 1;
1685 break;
542472b5
LV		 1686 case KVM_CAP_COALESCED_MMIO:
1687 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1688 break;
774ead3a
AK		 1689 case KVM_CAP_VAPIC:
1690 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1691 break;
f725230a
AK		 1692 case KVM_CAP_NR_VCPUS:
1693 r = KVM_MAX_VCPUS;
1694 break;
a988b910
AK		 1695 case KVM_CAP_NR_MEMSLOTS:
1696 r = KVM_MEMORY_SLOTS;
1697 break;
a68a6a72
MT		 1698 case KVM_CAP_PV_MMU: /* obsolete */
1699 r = 0;
2f333bcb 1700 break;
62c476c7 1701 case KVM_CAP_IOMMU:
19de40a8 1702 r = iommu_found();
62c476c7 1703 break;
890ca9ae
HY		 1704 case KVM_CAP_MCE:
1705 r = KVM_MAX_MCE_BANKS;
1706 break;
2d5b5a66
SY		 1707 case KVM_CAP_XCRS:
1708 r = cpu_has_xsave;
1709 break;
018d00d2
ZX		 1710 default:
1711 r = 0;
1712 break;
1713 }
1714 return r;
1715
1716}
1717
043405e1
CO		 1718long kvm_arch_dev_ioctl(struct file *filp,
1719 unsigned int ioctl, unsigned long arg)
1720{
1721 void __user *argp = (void __user *)arg;
1722 long r;
1723
1724 switch (ioctl) {
1725 case KVM_GET_MSR_INDEX_LIST: {
1726 struct kvm_msr_list __user *user_msr_list = argp;
1727 struct kvm_msr_list msr_list;
1728 unsigned n;
1729
1730 r = -EFAULT;
1731 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1732 goto out;
1733 n = msr_list.nmsrs;
1734 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1735 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1736 goto out;
1737 r = -E2BIG;
e125e7b6 1738 if (n < msr_list.nmsrs)
043405e1
CO		 1739 goto out;
1740 r = -EFAULT;
1741 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1742 num_msrs_to_save * sizeof(u32)))
1743 goto out;
e125e7b6 1744 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
043405e1
CO		 1745 &emulated_msrs,
1746 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1747 goto out;
1748 r = 0;
1749 break;
1750 }
674eea0f
AK		 1751 case KVM_GET_SUPPORTED_CPUID: {
1752 struct kvm_cpuid2 __user *cpuid_arg = argp;
1753 struct kvm_cpuid2 cpuid;
1754
1755 r = -EFAULT;
1756 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1757 goto out;
1758 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
19355475 1759 cpuid_arg->entries);
674eea0f
AK		 1760 if (r)
1761 goto out;
1762
1763 r = -EFAULT;
1764 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1765 goto out;
1766 r = 0;
1767 break;
1768 }
890ca9ae
HY		 1769 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1770 u64 mce_cap;
1771
1772 mce_cap = KVM_MCE_CAP_SUPPORTED;
1773 r = -EFAULT;
1774 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1775 goto out;
1776 r = 0;
1777 break;
1778 }
043405e1
CO		 1779 default:
1780 r = -EINVAL;
1781 }
1782out:
1783 return r;
1784}
1785
313a3dc7
CO		 1786void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1787{
1788 kvm_x86_ops->vcpu_load(vcpu, cpu);
6b7d7e76
ZA		 1789 if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1790 unsigned long khz = cpufreq_quick_get(cpu);
1791 if (!khz)
1792 khz = tsc_khz;
1793 per_cpu(cpu_tsc_khz, cpu) = khz;
1794 }
c8076604 1795 kvm_request_guest_time_update(vcpu);
313a3dc7
CO		 1796}
1797
1798void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1799{
02daab21 1800 kvm_x86_ops->vcpu_put(vcpu);
1c11e713 1801 kvm_put_guest_fpu(vcpu);
313a3dc7
CO		 1802}
1803
07716717 1804static int is_efer_nx(void)
313a3dc7 1805{
e286e86e 1806 unsigned long long efer = 0;
313a3dc7 1807
e286e86e 1808 rdmsrl_safe(MSR_EFER, &efer);
07716717
DK		 1809 return efer & EFER_NX;
1810}
1811
1812static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1813{
1814 int i;
1815 struct kvm_cpuid_entry2 *e, *entry;
1816
313a3dc7 1817 entry = NULL;
ad312c7c
ZX		 1818 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1819 e = &vcpu->arch.cpuid_entries[i];
313a3dc7
CO		 1820 if (e->function == 0x80000001) {
1821 entry = e;
1822 break;
1823 }
1824 }
07716717 1825 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
313a3dc7
CO		 1826 entry->edx &= ~(1 << 20);
1827 printk(KERN_INFO "kvm: guest NX capability removed\n");
1828 }
1829}
1830
07716717 1831/* when an old userspace process fills a new kernel module */
313a3dc7
CO		 1832static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1833 struct kvm_cpuid *cpuid,
1834 struct kvm_cpuid_entry __user *entries)
07716717
DK		 1835{
1836 int r, i;
1837 struct kvm_cpuid_entry *cpuid_entries;
1838
1839 r = -E2BIG;
1840 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1841 goto out;
1842 r = -ENOMEM;
1843 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1844 if (!cpuid_entries)
1845 goto out;
1846 r = -EFAULT;
1847 if (copy_from_user(cpuid_entries, entries,
1848 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1849 goto out_free;
1850 for (i = 0; i < cpuid->nent; i++) {
ad312c7c
ZX		 1851 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1852 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1853 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1854 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1855 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1856 vcpu->arch.cpuid_entries[i].index = 0;
1857 vcpu->arch.cpuid_entries[i].flags = 0;
1858 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1859 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1860 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1861 }
1862 vcpu->arch.cpuid_nent = cpuid->nent;
07716717
DK		 1863 cpuid_fix_nx_cap(vcpu);
1864 r = 0;
fc61b800 1865 kvm_apic_set_version(vcpu);
0e851880 1866 kvm_x86_ops->cpuid_update(vcpu);
2acf923e 1867 update_cpuid(vcpu);
07716717
DK		 1868
1869out_free:
1870 vfree(cpuid_entries);
1871out:
1872 return r;
1873}
1874
1875static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
19355475
AS		 1876 struct kvm_cpuid2 *cpuid,
1877 struct kvm_cpuid_entry2 __user *entries)
313a3dc7
CO		 1878{
1879 int r;
1880
1881 r = -E2BIG;
1882 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1883 goto out;
1884 r = -EFAULT;
ad312c7c 1885 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
07716717 1886 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
313a3dc7 1887 goto out;
ad312c7c 1888 vcpu->arch.cpuid_nent = cpuid->nent;
fc61b800 1889 kvm_apic_set_version(vcpu);
0e851880 1890 kvm_x86_ops->cpuid_update(vcpu);
2acf923e 1891 update_cpuid(vcpu);
313a3dc7
CO		 1892 return 0;
1893
1894out:
1895 return r;
1896}
1897
07716717 1898static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
19355475
AS		 1899 struct kvm_cpuid2 *cpuid,
1900 struct kvm_cpuid_entry2 __user *entries)
07716717
DK		 1901{
1902 int r;
1903
1904 r = -E2BIG;
ad312c7c 1905 if (cpuid->nent < vcpu->arch.cpuid_nent)
07716717
DK		 1906 goto out;
1907 r = -EFAULT;
ad312c7c 1908 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
19355475 1909 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
07716717
DK		 1910 goto out;
1911 return 0;
1912
1913out:
ad312c7c 1914 cpuid->nent = vcpu->arch.cpuid_nent;
07716717
DK		 1915 return r;
1916}
1917
07716717 1918static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
19355475 1919 u32 index)
07716717
DK		 1920{
1921 entry->function = function;
1922 entry->index = index;
1923 cpuid_count(entry->function, entry->index,
19355475 1924 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
07716717
DK		 1925 entry->flags = 0;
1926}
1927
7faa4ee1
AK		 1928#define F(x) bit(X86_FEATURE_##x)
1929
07716717
DK		 1930static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1931 u32 index, int *nent, int maxnent)
1932{
7faa4ee1 1933 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
07716717 1934#ifdef CONFIG_X86_64
17cc3935
SY		 1935 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1936 ? F(GBPAGES) : 0;
7faa4ee1
AK		 1937 unsigned f_lm = F(LM);
1938#else
17cc3935 1939 unsigned f_gbpages = 0;
7faa4ee1 1940 unsigned f_lm = 0;
07716717 1941#endif
4e47c7a6 1942 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
7faa4ee1
AK		 1943
1944 /* cpuid 1.edx */
1945 const u32 kvm_supported_word0_x86_features =
1946 F(FPU) | F(VME) | F(DE) | F(PSE) |
1947 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1948 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1949 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1950 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1951 0 /* Reserved, DS, ACPI */ | F(MMX) |
1952 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1953 0 /* HTT, TM, Reserved, PBE */;
1954 /* cpuid 0x80000001.edx */
1955 const u32 kvm_supported_word1_x86_features =
1956 F(FPU) | F(VME) | F(DE) | F(PSE) |
1957 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1958 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1959 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1960 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1961 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
4e47c7a6 1962 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
7faa4ee1
AK		 1963 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1964 /* cpuid 1.ecx */
1965 const u32 kvm_supported_word4_x86_features =
d149c731
AK		 1966 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1967 0 /* DS-CPL, VMX, SMX, EST */ |
1968 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1969 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1970 0 /* Reserved, DCA */ | F(XMM4_1) |
0105d1a5 1971 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2acf923e 1972 0 /* Reserved, AES */ | F(XSAVE) | 0 /* OSXSAVE */;
7faa4ee1 1973 /* cpuid 0x80000001.ecx */
07716717 1974 const u32 kvm_supported_word6_x86_features =
7faa4ee1
AK		 1975 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1976 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1977 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1978 0 /* SKINIT */ | 0 /* WDT */;
07716717 1979
19355475 1980 /* all calls to cpuid_count() should be made on the same cpu */
07716717
DK		 1981 get_cpu();
1982 do_cpuid_1_ent(entry, function, index);
1983 ++*nent;
1984
1985 switch (function) {
1986 case 0:
2acf923e 1987 entry->eax = min(entry->eax, (u32)0xd);
07716717
DK		 1988 break;
1989 case 1:
1990 entry->edx &= kvm_supported_word0_x86_features;
7faa4ee1 1991 entry->ecx &= kvm_supported_word4_x86_features;
0d1de2d9
GN		 1992 /* we support x2apic emulation even if host does not support
1993 * it since we emulate x2apic in software */
1994 entry->ecx |= F(X2APIC);
07716717
DK		 1995 break;
1996 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1997 * may return different values. This forces us to get_cpu() before
1998 * issuing the first command, and also to emulate this annoying behavior
1999 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2000 case 2: {
2001 int t, times = entry->eax & 0xff;
2002
2003 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
0fdf8e59 2004 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
07716717
DK		 2005 for (t = 1; t < times && *nent < maxnent; ++t) {
2006 do_cpuid_1_ent(&entry[t], function, 0);
2007 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2008 ++*nent;
2009 }
2010 break;
2011 }
2012 /* function 4 and 0xb have additional index. */
2013 case 4: {
14af3f3c 2014 int i, cache_type;
07716717
DK		 2015
2016 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2017 /* read more entries until cache_type is zero */
14af3f3c
HH		 2018 for (i = 1; *nent < maxnent; ++i) {
2019 cache_type = entry[i - 1].eax & 0x1f;
07716717
DK		 2020 if (!cache_type)
2021 break;
14af3f3c
HH		 2022 do_cpuid_1_ent(&entry[i], function, i);
2023 entry[i].flags |=
07716717
DK		 2024 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2025 ++*nent;
2026 }
2027 break;
2028 }
2029 case 0xb: {
14af3f3c 2030 int i, level_type;
07716717
DK		 2031
2032 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2033 /* read more entries until level_type is zero */
14af3f3c 2034 for (i = 1; *nent < maxnent; ++i) {
0853d2c1 2035 level_type = entry[i - 1].ecx & 0xff00;
07716717
DK		 2036 if (!level_type)
2037 break;
14af3f3c
HH		 2038 do_cpuid_1_ent(&entry[i], function, i);
2039 entry[i].flags |=
07716717
DK		 2040 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2041 ++*nent;
2042 }
2043 break;
2044 }
2acf923e
DC		 2045 case 0xd: {
2046 int i;
2047
2048 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2049 for (i = 1; *nent < maxnent; ++i) {
2050 if (entry[i - 1].eax == 0 && i != 2)
2051 break;
2052 do_cpuid_1_ent(&entry[i], function, i);
2053 entry[i].flags |=
2054 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2055 ++*nent;
2056 }
2057 break;
2058 }
84478c82
GC		 2059 case KVM_CPUID_SIGNATURE: {
2060 char signature[12] = "KVMKVMKVM\0\0";
2061 u32 *sigptr = (u32 *)signature;
2062 entry->eax = 0;
2063 entry->ebx = sigptr[0];
2064 entry->ecx = sigptr[1];
2065 entry->edx = sigptr[2];
2066 break;
2067 }
2068 case KVM_CPUID_FEATURES:
2069 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2070 (1 << KVM_FEATURE_NOP_IO_DELAY) |
371bcf64
GC		 2071 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2072 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
84478c82
GC		 2073 entry->ebx = 0;
2074 entry->ecx = 0;
2075 entry->edx = 0;
2076 break;
07716717
DK		 2077 case 0x80000000:
2078 entry->eax = min(entry->eax, 0x8000001a);
2079 break;
2080 case 0x80000001:
2081 entry->edx &= kvm_supported_word1_x86_features;
2082 entry->ecx &= kvm_supported_word6_x86_features;
2083 break;
2084 }
d4330ef2
JR		 2085
2086 kvm_x86_ops->set_supported_cpuid(function, entry);
2087
07716717
DK		 2088 put_cpu();
2089}
2090
7faa4ee1
AK		 2091#undef F
2092
674eea0f 2093static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
19355475 2094 struct kvm_cpuid_entry2 __user *entries)
07716717
DK		 2095{
2096 struct kvm_cpuid_entry2 *cpuid_entries;
2097 int limit, nent = 0, r = -E2BIG;
2098 u32 func;
2099
2100 if (cpuid->nent < 1)
2101 goto out;
6a544355
AK		 2102 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2103 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
07716717
DK		 2104 r = -ENOMEM;
2105 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2106 if (!cpuid_entries)
2107 goto out;
2108
2109 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2110 limit = cpuid_entries[0].eax;
2111 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2112 do_cpuid_ent(&cpuid_entries[nent], func, 0,
19355475 2113 &nent, cpuid->nent);
07716717
DK		 2114 r = -E2BIG;
2115 if (nent >= cpuid->nent)
2116 goto out_free;
2117
2118 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2119 limit = cpuid_entries[nent - 1].eax;
2120 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2121 do_cpuid_ent(&cpuid_entries[nent], func, 0,
19355475 2122 &nent, cpuid->nent);
84478c82
GC		 2123
2124
2125
2126 r = -E2BIG;
2127 if (nent >= cpuid->nent)
2128 goto out_free;
2129
2130 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2131 cpuid->nent);
2132
2133 r = -E2BIG;
2134 if (nent >= cpuid->nent)
2135 goto out_free;
2136
2137 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2138 cpuid->nent);
2139
cb007648
MM		 2140 r = -E2BIG;
2141 if (nent >= cpuid->nent)
2142 goto out_free;
2143
07716717
DK		 2144 r = -EFAULT;
2145 if (copy_to_user(entries, cpuid_entries,
19355475 2146 nent * sizeof(struct kvm_cpuid_entry2)))
07716717
DK		 2147 goto out_free;
2148 cpuid->nent = nent;
2149 r = 0;
2150
2151out_free:
2152 vfree(cpuid_entries);
2153out:
2154 return r;
2155}
2156
313a3dc7
CO		 2157static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2158 struct kvm_lapic_state *s)
2159{
ad312c7c 2160 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
313a3dc7
CO		 2161
2162 return 0;
2163}
2164
2165static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2166 struct kvm_lapic_state *s)
2167{
ad312c7c 2168 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
313a3dc7 2169 kvm_apic_post_state_restore(vcpu);
cb142eb7 2170 update_cr8_intercept(vcpu);
313a3dc7
CO		 2171
2172 return 0;
2173}
2174
f77bc6a4
ZX		 2175static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2176 struct kvm_interrupt *irq)
2177{
2178 if (irq->irq < 0 || irq->irq >= 256)
2179 return -EINVAL;
2180 if (irqchip_in_kernel(vcpu->kvm))
2181 return -ENXIO;
f77bc6a4 2182
66fd3f7f 2183 kvm_queue_interrupt(vcpu, irq->irq, false);
f77bc6a4 2184
f77bc6a4
ZX		 2185 return 0;
2186}
2187
c4abb7c9
JK		 2188static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2189{
c4abb7c9 2190 kvm_inject_nmi(vcpu);
c4abb7c9
JK		 2191
2192 return 0;
2193}
2194
b209749f
AK		 2195static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2196 struct kvm_tpr_access_ctl *tac)
2197{
2198 if (tac->flags)
2199 return -EINVAL;
2200 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2201 return 0;
2202}
2203
890ca9ae
HY		 2204static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2205 u64 mcg_cap)
2206{
2207 int r;
2208 unsigned bank_num = mcg_cap & 0xff, bank;
2209
2210 r = -EINVAL;
a9e38c3e 2211 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
890ca9ae
HY		 2212 goto out;
2213 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2214 goto out;
2215 r = 0;
2216 vcpu->arch.mcg_cap = mcg_cap;
2217 /* Init IA32_MCG_CTL to all 1s */
2218 if (mcg_cap & MCG_CTL_P)
2219 vcpu->arch.mcg_ctl = ~(u64)0;
2220 /* Init IA32_MCi_CTL to all 1s */
2221 for (bank = 0; bank < bank_num; bank++)
2222 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2223out:
2224 return r;
2225}
2226
2227static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2228 struct kvm_x86_mce *mce)
2229{
2230 u64 mcg_cap = vcpu->arch.mcg_cap;
2231 unsigned bank_num = mcg_cap & 0xff;
2232 u64 *banks = vcpu->arch.mce_banks;
2233
2234 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2235 return -EINVAL;
2236 /*
2237 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2238 * reporting is disabled
2239 */
2240 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2241 vcpu->arch.mcg_ctl != ~(u64)0)
2242 return 0;
2243 banks += 4 * mce->bank;
2244 /*
2245 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2246 * reporting is disabled for the bank
2247 */
2248 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2249 return 0;
2250 if (mce->status & MCI_STATUS_UC) {
2251 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
fc78f519 2252 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
890ca9ae
HY		 2253 printk(KERN_DEBUG "kvm: set_mce: "
2254 "injects mce exception while "
2255 "previous one is in progress!\n");
2256 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2257 return 0;
2258 }
2259 if (banks[1] & MCI_STATUS_VAL)
2260 mce->status |= MCI_STATUS_OVER;
2261 banks[2] = mce->addr;
2262 banks[3] = mce->misc;
2263 vcpu->arch.mcg_status = mce->mcg_status;
2264 banks[1] = mce->status;
2265 kvm_queue_exception(vcpu, MC_VECTOR);
2266 } else if (!(banks[1] & MCI_STATUS_VAL)
2267 || !(banks[1] & MCI_STATUS_UC)) {
2268 if (banks[1] & MCI_STATUS_VAL)
2269 mce->status |= MCI_STATUS_OVER;
2270 banks[2] = mce->addr;
2271 banks[3] = mce->misc;
2272 banks[1] = mce->status;
2273 } else
2274 banks[1] |= MCI_STATUS_OVER;
2275 return 0;
2276}
2277
3cfc3092
JK		 2278static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2279 struct kvm_vcpu_events *events)
2280{
03b82a30
JK		 2281 events->exception.injected =
2282 vcpu->arch.exception.pending &&
2283 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3cfc3092
JK		 2284 events->exception.nr = vcpu->arch.exception.nr;
2285 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2286 events->exception.error_code = vcpu->arch.exception.error_code;
2287
03b82a30
JK		 2288 events->interrupt.injected =
2289 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3cfc3092 2290 events->interrupt.nr = vcpu->arch.interrupt.nr;
03b82a30 2291 events->interrupt.soft = 0;
48005f64
JK		 2292 events->interrupt.shadow =
2293 kvm_x86_ops->get_interrupt_shadow(vcpu,
2294 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
3cfc3092
JK		 2295
2296 events->nmi.injected = vcpu->arch.nmi_injected;
2297 events->nmi.pending = vcpu->arch.nmi_pending;
2298 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2299
2300 events->sipi_vector = vcpu->arch.sipi_vector;
2301
dab4b911 2302 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64
JK		 2303 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2304 | KVM_VCPUEVENT_VALID_SHADOW);
3cfc3092
JK		 2305}
2306
2307static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2308 struct kvm_vcpu_events *events)
2309{
dab4b911 2310 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64
JK		 2311 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2312 | KVM_VCPUEVENT_VALID_SHADOW))
3cfc3092
JK		 2313 return -EINVAL;
2314
3cfc3092
JK		 2315 vcpu->arch.exception.pending = events->exception.injected;
2316 vcpu->arch.exception.nr = events->exception.nr;
2317 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2318 vcpu->arch.exception.error_code = events->exception.error_code;
2319
2320 vcpu->arch.interrupt.pending = events->interrupt.injected;
2321 vcpu->arch.interrupt.nr = events->interrupt.nr;
2322 vcpu->arch.interrupt.soft = events->interrupt.soft;
2323 if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2324 kvm_pic_clear_isr_ack(vcpu->kvm);
48005f64
JK		 2325 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2326 kvm_x86_ops->set_interrupt_shadow(vcpu,
2327 events->interrupt.shadow);
3cfc3092
JK		 2328
2329 vcpu->arch.nmi_injected = events->nmi.injected;
dab4b911
JK		 2330 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2331 vcpu->arch.nmi_pending = events->nmi.pending;
3cfc3092
JK		 2332 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2333
dab4b911
JK		 2334 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2335 vcpu->arch.sipi_vector = events->sipi_vector;
3cfc3092 2336
3cfc3092
JK		 2337 return 0;
2338}
2339
a1efbe77
JK		 2340static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2341 struct kvm_debugregs *dbgregs)
2342{
a1efbe77
JK		 2343 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2344 dbgregs->dr6 = vcpu->arch.dr6;
2345 dbgregs->dr7 = vcpu->arch.dr7;
2346 dbgregs->flags = 0;
a1efbe77
JK		 2347}
2348
2349static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2350 struct kvm_debugregs *dbgregs)
2351{
2352 if (dbgregs->flags)
2353 return -EINVAL;
2354
a1efbe77
JK		 2355 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2356 vcpu->arch.dr6 = dbgregs->dr6;
2357 vcpu->arch.dr7 = dbgregs->dr7;
2358
a1efbe77
JK		 2359 return 0;
2360}
2361
2d5b5a66
SY		 2362static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2363 struct kvm_xsave *guest_xsave)
2364{
2365 if (cpu_has_xsave)
2366 memcpy(guest_xsave->region,
2367 &vcpu->arch.guest_fpu.state->xsave,
2368 sizeof(struct xsave_struct));
2369 else {
2370 memcpy(guest_xsave->region,
2371 &vcpu->arch.guest_fpu.state->fxsave,
2372 sizeof(struct i387_fxsave_struct));
2373 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2374 XSTATE_FPSSE;
2375 }
2376}
2377
2378static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2379 struct kvm_xsave *guest_xsave)
2380{
2381 u64 xstate_bv =
2382 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2383
2384 if (cpu_has_xsave)
2385 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2386 guest_xsave->region, sizeof(struct xsave_struct));
2387 else {
2388 if (xstate_bv & ~XSTATE_FPSSE)
2389 return -EINVAL;
2390 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2391 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2392 }
2393 return 0;
2394}
2395
2396static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2397 struct kvm_xcrs *guest_xcrs)
2398{
2399 if (!cpu_has_xsave) {
2400 guest_xcrs->nr_xcrs = 0;
2401 return;
2402 }
2403
2404 guest_xcrs->nr_xcrs = 1;
2405 guest_xcrs->flags = 0;
2406 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2407 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2408}
2409
2410static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2411 struct kvm_xcrs *guest_xcrs)
2412{
2413 int i, r = 0;
2414
2415 if (!cpu_has_xsave)
2416 return -EINVAL;
2417
2418 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2419 return -EINVAL;
2420
2421 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2422 /* Only support XCR0 currently */
2423 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2424 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2425 guest_xcrs->xcrs[0].value);
2426 break;
2427 }
2428 if (r)
2429 r = -EINVAL;
2430 return r;
2431}
2432
313a3dc7
CO		 2433long kvm_arch_vcpu_ioctl(struct file *filp,
2434 unsigned int ioctl, unsigned long arg)
2435{
2436 struct kvm_vcpu *vcpu = filp->private_data;
2437 void __user *argp = (void __user *)arg;
2438 int r;
b772ff36 2439 struct kvm_lapic_state *lapic = NULL;
a1a005f3
AK		 2440 struct kvm_xsave *xsave = NULL;
2441 struct kvm_xcrs *xcrs = NULL;
313a3dc7
CO		 2442
2443 switch (ioctl) {
2444 case KVM_GET_LAPIC: {
2204ae3c
MT		 2445 r = -EINVAL;
2446 if (!vcpu->arch.apic)
2447 goto out;
b772ff36 2448 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
313a3dc7 2449
b772ff36
DH		 2450 r = -ENOMEM;
2451 if (!lapic)
2452 goto out;
2453 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
313a3dc7
CO		 2454 if (r)
2455 goto out;
2456 r = -EFAULT;
b772ff36 2457 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
313a3dc7
CO		 2458 goto out;
2459 r = 0;
2460 break;
2461 }
2462 case KVM_SET_LAPIC: {
2204ae3c
MT		 2463 r = -EINVAL;
2464 if (!vcpu->arch.apic)
2465 goto out;
b772ff36
DH		 2466 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2467 r = -ENOMEM;
2468 if (!lapic)
2469 goto out;
313a3dc7 2470 r = -EFAULT;
b772ff36 2471 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
313a3dc7 2472 goto out;
b772ff36 2473 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
313a3dc7
CO		 2474 if (r)
2475 goto out;
2476 r = 0;
2477 break;
2478 }
f77bc6a4
ZX		 2479 case KVM_INTERRUPT: {
2480 struct kvm_interrupt irq;
2481
2482 r = -EFAULT;
2483 if (copy_from_user(&irq, argp, sizeof irq))
2484 goto out;
2485 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2486 if (r)
2487 goto out;
2488 r = 0;
2489 break;
2490 }
c4abb7c9
JK		 2491 case KVM_NMI: {
2492 r = kvm_vcpu_ioctl_nmi(vcpu);
2493 if (r)
2494 goto out;
2495 r = 0;
2496 break;
2497 }
313a3dc7
CO		 2498 case KVM_SET_CPUID: {
2499 struct kvm_cpuid __user *cpuid_arg = argp;
2500 struct kvm_cpuid cpuid;
2501
2502 r = -EFAULT;
2503 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2504 goto out;
2505 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2506 if (r)
2507 goto out;
2508 break;
2509 }
07716717
DK		 2510 case KVM_SET_CPUID2: {
2511 struct kvm_cpuid2 __user *cpuid_arg = argp;
2512 struct kvm_cpuid2 cpuid;
2513
2514 r = -EFAULT;
2515 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2516 goto out;
2517 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
19355475 2518 cpuid_arg->entries);
07716717
DK		 2519 if (r)
2520 goto out;
2521 break;
2522 }
2523 case KVM_GET_CPUID2: {
2524 struct kvm_cpuid2 __user *cpuid_arg = argp;
2525 struct kvm_cpuid2 cpuid;
2526
2527 r = -EFAULT;
2528 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2529 goto out;
2530 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
19355475 2531 cpuid_arg->entries);
07716717
DK		 2532 if (r)
2533 goto out;
2534 r = -EFAULT;
2535 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2536 goto out;
2537 r = 0;
2538 break;
2539 }
313a3dc7
CO		 2540 case KVM_GET_MSRS:
2541 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2542 break;
2543 case KVM_SET_MSRS:
2544 r = msr_io(vcpu, argp, do_set_msr, 0);
2545 break;
b209749f
AK		 2546 case KVM_TPR_ACCESS_REPORTING: {
2547 struct kvm_tpr_access_ctl tac;
2548
2549 r = -EFAULT;
2550 if (copy_from_user(&tac, argp, sizeof tac))
2551 goto out;
2552 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2553 if (r)
2554 goto out;
2555 r = -EFAULT;
2556 if (copy_to_user(argp, &tac, sizeof tac))
2557 goto out;
2558 r = 0;
2559 break;
2560 };
b93463aa
AK		 2561 case KVM_SET_VAPIC_ADDR: {
2562 struct kvm_vapic_addr va;
2563
2564 r = -EINVAL;
2565 if (!irqchip_in_kernel(vcpu->kvm))
2566 goto out;
2567 r = -EFAULT;
2568 if (copy_from_user(&va, argp, sizeof va))
2569 goto out;
2570 r = 0;
2571 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2572 break;
2573 }
890ca9ae
HY		 2574 case KVM_X86_SETUP_MCE: {
2575 u64 mcg_cap;
2576
2577 r = -EFAULT;
2578 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2579 goto out;
2580 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2581 break;
2582 }
2583 case KVM_X86_SET_MCE: {
2584 struct kvm_x86_mce mce;
2585
2586 r = -EFAULT;
2587 if (copy_from_user(&mce, argp, sizeof mce))
2588 goto out;
2589 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2590 break;
2591 }
3cfc3092
JK		 2592 case KVM_GET_VCPU_EVENTS: {
2593 struct kvm_vcpu_events events;
2594
2595 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2596
2597 r = -EFAULT;
2598 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2599 break;
2600 r = 0;
2601 break;
2602 }
2603 case KVM_SET_VCPU_EVENTS: {
2604 struct kvm_vcpu_events events;
2605
2606 r = -EFAULT;
2607 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2608 break;
2609
2610 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2611 break;
2612 }
a1efbe77
JK		 2613 case KVM_GET_DEBUGREGS: {
2614 struct kvm_debugregs dbgregs;
2615
2616 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2617
2618 r = -EFAULT;
2619 if (copy_to_user(argp, &dbgregs,
2620 sizeof(struct kvm_debugregs)))
2621 break;
2622 r = 0;
2623 break;
2624 }
2625 case KVM_SET_DEBUGREGS: {
2626 struct kvm_debugregs dbgregs;
2627
2628 r = -EFAULT;
2629 if (copy_from_user(&dbgregs, argp,
2630 sizeof(struct kvm_debugregs)))
2631 break;
2632
2633 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2634 break;
2635 }
2d5b5a66 2636 case KVM_GET_XSAVE: {
2d5b5a66
SY		 2637 xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2638 r = -ENOMEM;
2639 if (!xsave)
2640 break;
2641
2642 kvm_vcpu_ioctl_x86_get_xsave(vcpu, xsave);
2643
2644 r = -EFAULT;
2645 if (copy_to_user(argp, xsave, sizeof(struct kvm_xsave)))
2646 break;
2647 r = 0;
2648 break;
2649 }
2650 case KVM_SET_XSAVE: {
2d5b5a66
SY		 2651 xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2652 r = -ENOMEM;
2653 if (!xsave)
2654 break;
2655
2656 r = -EFAULT;
2657 if (copy_from_user(xsave, argp, sizeof(struct kvm_xsave)))
2658 break;
2659
2660 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, xsave);
2661 break;
2662 }
2663 case KVM_GET_XCRS: {
2d5b5a66
SY		 2664 xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2665 r = -ENOMEM;
2666 if (!xcrs)
2667 break;
2668
2669 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, xcrs);
2670
2671 r = -EFAULT;
2672 if (copy_to_user(argp, xcrs,
2673 sizeof(struct kvm_xcrs)))
2674 break;
2675 r = 0;
2676 break;
2677 }
2678 case KVM_SET_XCRS: {
2d5b5a66
SY		 2679 xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2680 r = -ENOMEM;
2681 if (!xcrs)
2682 break;
2683
2684 r = -EFAULT;
2685 if (copy_from_user(xcrs, argp,
2686 sizeof(struct kvm_xcrs)))
2687 break;
2688
2689 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, xcrs);
2690 break;
2691 }
313a3dc7
CO		 2692 default:
2693 r = -EINVAL;
2694 }
2695out:
7a6ce84c 2696 kfree(lapic);
a1a005f3
AK		 2697 kfree(xsave);
2698 kfree(xcrs);
313a3dc7
CO		 2699 return r;
2700}
2701
1fe779f8
CO		 2702static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2703{
2704 int ret;
2705
2706 if (addr > (unsigned int)(-3 * PAGE_SIZE))
2707 return -1;
2708 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2709 return ret;
2710}
2711
b927a3ce
SY		 2712static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2713 u64 ident_addr)
2714{
2715 kvm->arch.ept_identity_map_addr = ident_addr;
2716 return 0;
2717}
2718
1fe779f8
CO		 2719static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2720 u32 kvm_nr_mmu_pages)
2721{
2722 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2723 return -EINVAL;
2724
79fac95e 2725 mutex_lock(&kvm->slots_lock);
7c8a83b7 2726 spin_lock(&kvm->mmu_lock);
1fe779f8
CO		 2727
2728 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
f05e70ac 2729 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1fe779f8 2730
7c8a83b7 2731 spin_unlock(&kvm->mmu_lock);
79fac95e 2732 mutex_unlock(&kvm->slots_lock);
1fe779f8
CO		 2733 return 0;
2734}
2735
2736static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2737{
f05e70ac 2738 return kvm->arch.n_alloc_mmu_pages;
1fe779f8
CO		 2739}
2740
a983fb23
MT		 2741gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2742{
2743 int i;
2744 struct kvm_mem_alias *alias;
2745 struct kvm_mem_aliases *aliases;
2746
90d83dc3 2747 aliases = kvm_aliases(kvm);
a983fb23
MT		 2748
2749 for (i = 0; i < aliases->naliases; ++i) {
2750 alias = &aliases->aliases[i];
2751 if (alias->flags & KVM_ALIAS_INVALID)
2752 continue;
2753 if (gfn >= alias->base_gfn
2754 && gfn < alias->base_gfn + alias->npages)
2755 return alias->target_gfn + gfn - alias->base_gfn;
2756 }
2757 return gfn;
2758}
2759
e9f85cde
ZX		 2760gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2761{
2762 int i;
2763 struct kvm_mem_alias *alias;
a983fb23
MT		 2764 struct kvm_mem_aliases *aliases;
2765
90d83dc3 2766 aliases = kvm_aliases(kvm);
e9f85cde 2767
fef9cce0
MT		 2768 for (i = 0; i < aliases->naliases; ++i) {
2769 alias = &aliases->aliases[i];
e9f85cde
ZX		 2770 if (gfn >= alias->base_gfn
2771 && gfn < alias->base_gfn + alias->npages)
2772 return alias->target_gfn + gfn - alias->base_gfn;
2773 }
2774 return gfn;
2775}
2776
1fe779f8
CO		 2777/*
2778 * Set a new alias region. Aliases map a portion of physical memory into
2779 * another portion. This is useful for memory windows, for example the PC
2780 * VGA region.
2781 */
2782static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2783 struct kvm_memory_alias *alias)
2784{
2785 int r, n;
2786 struct kvm_mem_alias *p;
a983fb23 2787 struct kvm_mem_aliases *aliases, *old_aliases;
1fe779f8
CO		 2788
2789 r = -EINVAL;
2790 /* General sanity checks */
2791 if (alias->memory_size & (PAGE_SIZE - 1))
2792 goto out;
2793 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2794 goto out;
2795 if (alias->slot >= KVM_ALIAS_SLOTS)
2796 goto out;
2797 if (alias->guest_phys_addr + alias->memory_size
2798 < alias->guest_phys_addr)
2799 goto out;
2800 if (alias->target_phys_addr + alias->memory_size
2801 < alias->target_phys_addr)
2802 goto out;
2803
a983fb23
MT		 2804 r = -ENOMEM;
2805 aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2806 if (!aliases)
2807 goto out;
2808
79fac95e 2809 mutex_lock(&kvm->slots_lock);
1fe779f8 2810
a983fb23
MT		 2811 /* invalidate any gfn reference in case of deletion/shrinking */
2812 memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2813 aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2814 old_aliases = kvm->arch.aliases;
2815 rcu_assign_pointer(kvm->arch.aliases, aliases);
2816 synchronize_srcu_expedited(&kvm->srcu);
2817 kvm_mmu_zap_all(kvm);
2818 kfree(old_aliases);
2819
2820 r = -ENOMEM;
2821 aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2822 if (!aliases)
2823 goto out_unlock;
2824
2825 memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
fef9cce0
MT		 2826
2827 p = &aliases->aliases[alias->slot];
1fe779f8
CO		 2828 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2829 p->npages = alias->memory_size >> PAGE_SHIFT;
2830 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
a983fb23 2831 p->flags &= ~(KVM_ALIAS_INVALID);
1fe779f8
CO		 2832
2833 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
fef9cce0 2834 if (aliases->aliases[n - 1].npages)
1fe779f8 2835 break;
fef9cce0 2836 aliases->naliases = n;
1fe779f8 2837
a983fb23
MT		 2838 old_aliases = kvm->arch.aliases;
2839 rcu_assign_pointer(kvm->arch.aliases, aliases);
2840 synchronize_srcu_expedited(&kvm->srcu);
2841 kfree(old_aliases);
2842 r = 0;
1fe779f8 2843
a983fb23 2844out_unlock:
79fac95e 2845 mutex_unlock(&kvm->slots_lock);
1fe779f8
CO		 2846out:
2847 return r;
2848}
2849
2850static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2851{
2852 int r;
2853
2854 r = 0;
2855 switch (chip->chip_id) {
2856 case KVM_IRQCHIP_PIC_MASTER:
2857 memcpy(&chip->chip.pic,
2858 &pic_irqchip(kvm)->pics[0],
2859 sizeof(struct kvm_pic_state));
2860 break;
2861 case KVM_IRQCHIP_PIC_SLAVE:
2862 memcpy(&chip->chip.pic,
2863 &pic_irqchip(kvm)->pics[1],
2864 sizeof(struct kvm_pic_state));
2865 break;
2866 case KVM_IRQCHIP_IOAPIC:
eba0226b 2867 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 2868 break;
2869 default:
2870 r = -EINVAL;
2871 break;
2872 }
2873 return r;
2874}
2875
2876static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2877{
2878 int r;
2879
2880 r = 0;
2881 switch (chip->chip_id) {
2882 case KVM_IRQCHIP_PIC_MASTER:
fa8273e9 2883 raw_spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 2884 memcpy(&pic_irqchip(kvm)->pics[0],
2885 &chip->chip.pic,
2886 sizeof(struct kvm_pic_state));
fa8273e9 2887 raw_spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 2888 break;
2889 case KVM_IRQCHIP_PIC_SLAVE:
fa8273e9 2890 raw_spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 2891 memcpy(&pic_irqchip(kvm)->pics[1],
2892 &chip->chip.pic,
2893 sizeof(struct kvm_pic_state));
fa8273e9 2894 raw_spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 2895 break;
2896 case KVM_IRQCHIP_IOAPIC:
eba0226b 2897 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 2898 break;
2899 default:
2900 r = -EINVAL;
2901 break;
2902 }
2903 kvm_pic_update_irq(pic_irqchip(kvm));
2904 return r;
2905}
2906
e0f63cb9
SY		 2907static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2908{
2909 int r = 0;
2910
894a9c55 2911 mutex_lock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9 2912 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
894a9c55 2913 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9
SY		 2914 return r;
2915}
2916
2917static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2918{
2919 int r = 0;
2920
894a9c55 2921 mutex_lock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9 2922 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
e9f42757
BK		 2923 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2924 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2925 return r;
2926}
2927
2928static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2929{
2930 int r = 0;
2931
2932 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2933 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2934 sizeof(ps->channels));
2935 ps->flags = kvm->arch.vpit->pit_state.flags;
2936 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2937 return r;
2938}
2939
2940static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2941{
2942 int r = 0, start = 0;
2943 u32 prev_legacy, cur_legacy;
2944 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2945 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2946 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2947 if (!prev_legacy && cur_legacy)
2948 start = 1;
2949 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2950 sizeof(kvm->arch.vpit->pit_state.channels));
2951 kvm->arch.vpit->pit_state.flags = ps->flags;
2952 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
894a9c55 2953 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
e0f63cb9
SY		 2954 return r;
2955}
2956
52d939a0
MT		 2957static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2958 struct kvm_reinject_control *control)
2959{
2960 if (!kvm->arch.vpit)
2961 return -ENXIO;
894a9c55 2962 mutex_lock(&kvm->arch.vpit->pit_state.lock);
52d939a0 2963 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
894a9c55 2964 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
52d939a0
MT		 2965 return 0;
2966}
2967
5bb064dc
ZX		 2968/*
2969 * Get (and clear) the dirty memory log for a memory slot.
2970 */
2971int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2972 struct kvm_dirty_log *log)
2973{
87bf6e7d 2974 int r, i;
5bb064dc 2975 struct kvm_memory_slot *memslot;
87bf6e7d 2976 unsigned long n;
b050b015 2977 unsigned long is_dirty = 0;
5bb064dc 2978
79fac95e 2979 mutex_lock(&kvm->slots_lock);
5bb064dc 2980
b050b015
MT		 2981 r = -EINVAL;
2982 if (log->slot >= KVM_MEMORY_SLOTS)
2983 goto out;
2984
2985 memslot = &kvm->memslots->memslots[log->slot];
2986 r = -ENOENT;
2987 if (!memslot->dirty_bitmap)
2988 goto out;
2989
87bf6e7d 2990 n = kvm_dirty_bitmap_bytes(memslot);
b050b015 2991
b050b015
MT		 2992 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2993 is_dirty = memslot->dirty_bitmap[i];
5bb064dc
ZX		 2994
2995 /* If nothing is dirty, don't bother messing with page tables. */
2996 if (is_dirty) {
b050b015 2997 struct kvm_memslots *slots, *old_slots;
914ebccd 2998 unsigned long *dirty_bitmap;
b050b015 2999
7c8a83b7 3000 spin_lock(&kvm->mmu_lock);
5bb064dc 3001 kvm_mmu_slot_remove_write_access(kvm, log->slot);
7c8a83b7 3002 spin_unlock(&kvm->mmu_lock);
b050b015 3003
914ebccd
TY		 3004 r = -ENOMEM;
3005 dirty_bitmap = vmalloc(n);
3006 if (!dirty_bitmap)
3007 goto out;
3008 memset(dirty_bitmap, 0, n);
b050b015 3009
914ebccd
TY		 3010 r = -ENOMEM;
3011 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3012 if (!slots) {
3013 vfree(dirty_bitmap);
3014 goto out;
3015 }
b050b015
MT		 3016 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3017 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3018
3019 old_slots = kvm->memslots;
3020 rcu_assign_pointer(kvm->memslots, slots);
3021 synchronize_srcu_expedited(&kvm->srcu);
3022 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3023 kfree(old_slots);
914ebccd
TY		 3024
3025 r = -EFAULT;
3026 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n)) {
3027 vfree(dirty_bitmap);
3028 goto out;
3029 }
3030 vfree(dirty_bitmap);
3031 } else {
3032 r = -EFAULT;
3033 if (clear_user(log->dirty_bitmap, n))
3034 goto out;
5bb064dc 3035 }
b050b015 3036
5bb064dc
ZX		 3037 r = 0;
3038out:
79fac95e 3039 mutex_unlock(&kvm->slots_lock);
5bb064dc
ZX		 3040 return r;
3041}
3042
1fe779f8
CO		 3043long kvm_arch_vm_ioctl(struct file *filp,
3044 unsigned int ioctl, unsigned long arg)
3045{
3046 struct kvm *kvm = filp->private_data;
3047 void __user *argp = (void __user *)arg;
367e1319 3048 int r = -ENOTTY;
f0d66275
DH		 3049 /*
3050 * This union makes it completely explicit to gcc-3.x
3051 * that these two variables' stack usage should be
3052 * combined, not added together.
3053 */
3054 union {
3055 struct kvm_pit_state ps;
e9f42757 3056 struct kvm_pit_state2 ps2;
f0d66275 3057 struct kvm_memory_alias alias;
c5ff41ce 3058 struct kvm_pit_config pit_config;
f0d66275 3059 } u;
1fe779f8
CO		 3060
3061 switch (ioctl) {
3062 case KVM_SET_TSS_ADDR:
3063 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3064 if (r < 0)
3065 goto out;
3066 break;
b927a3ce
SY		 3067 case KVM_SET_IDENTITY_MAP_ADDR: {
3068 u64 ident_addr;
3069
3070 r = -EFAULT;
3071 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3072 goto out;
3073 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3074 if (r < 0)
3075 goto out;
3076 break;
3077 }
1fe779f8
CO		 3078 case KVM_SET_MEMORY_REGION: {
3079 struct kvm_memory_region kvm_mem;
3080 struct kvm_userspace_memory_region kvm_userspace_mem;
3081
3082 r = -EFAULT;
3083 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
3084 goto out;
3085 kvm_userspace_mem.slot = kvm_mem.slot;
3086 kvm_userspace_mem.flags = kvm_mem.flags;
3087 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
3088 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
3089 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
3090 if (r)
3091 goto out;
3092 break;
3093 }
3094 case KVM_SET_NR_MMU_PAGES:
3095 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3096 if (r)
3097 goto out;
3098 break;
3099 case KVM_GET_NR_MMU_PAGES:
3100 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3101 break;
f0d66275 3102 case KVM_SET_MEMORY_ALIAS:
1fe779f8 3103 r = -EFAULT;
f0d66275 3104 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
1fe779f8 3105 goto out;
f0d66275 3106 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
1fe779f8
CO		 3107 if (r)
3108 goto out;
3109 break;
3ddea128
MT		 3110 case KVM_CREATE_IRQCHIP: {
3111 struct kvm_pic *vpic;
3112
3113 mutex_lock(&kvm->lock);
3114 r = -EEXIST;
3115 if (kvm->arch.vpic)
3116 goto create_irqchip_unlock;
1fe779f8 3117 r = -ENOMEM;
3ddea128
MT		 3118 vpic = kvm_create_pic(kvm);
3119 if (vpic) {
1fe779f8
CO		 3120 r = kvm_ioapic_init(kvm);
3121 if (r) {
72bb2fcd
WY		 3122 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3123 &vpic->dev);
3ddea128
MT		 3124 kfree(vpic);
3125 goto create_irqchip_unlock;
1fe779f8
CO		 3126 }
3127 } else
3ddea128
MT		 3128 goto create_irqchip_unlock;
3129 smp_wmb();
3130 kvm->arch.vpic = vpic;
3131 smp_wmb();
399ec807
AK		 3132 r = kvm_setup_default_irq_routing(kvm);
3133 if (r) {
3ddea128 3134 mutex_lock(&kvm->irq_lock);
72bb2fcd
WY		 3135 kvm_ioapic_destroy(kvm);
3136 kvm_destroy_pic(kvm);
3ddea128 3137 mutex_unlock(&kvm->irq_lock);
399ec807 3138 }
3ddea128
MT		 3139 create_irqchip_unlock:
3140 mutex_unlock(&kvm->lock);
1fe779f8 3141 break;
3ddea128 3142 }
7837699f 3143 case KVM_CREATE_PIT:
c5ff41ce
JK		 3144 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3145 goto create_pit;
3146 case KVM_CREATE_PIT2:
3147 r = -EFAULT;
3148 if (copy_from_user(&u.pit_config, argp,
3149 sizeof(struct kvm_pit_config)))
3150 goto out;
3151 create_pit:
79fac95e 3152 mutex_lock(&kvm->slots_lock);
269e05e4
AK		 3153 r = -EEXIST;
3154 if (kvm->arch.vpit)
3155 goto create_pit_unlock;
7837699f 3156 r = -ENOMEM;
c5ff41ce 3157 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
7837699f
SY		 3158 if (kvm->arch.vpit)
3159 r = 0;
269e05e4 3160 create_pit_unlock:
79fac95e 3161 mutex_unlock(&kvm->slots_lock);
7837699f 3162 break;
4925663a 3163 case KVM_IRQ_LINE_STATUS:
1fe779f8
CO		 3164 case KVM_IRQ_LINE: {
3165 struct kvm_irq_level irq_event;
3166
3167 r = -EFAULT;
3168 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3169 goto out;
160d2f6c 3170 r = -ENXIO;
1fe779f8 3171 if (irqchip_in_kernel(kvm)) {
4925663a 3172 __s32 status;
4925663a
GN		 3173 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3174 irq_event.irq, irq_event.level);
4925663a 3175 if (ioctl == KVM_IRQ_LINE_STATUS) {
160d2f6c 3176 r = -EFAULT;
4925663a
GN		 3177 irq_event.status = status;
3178 if (copy_to_user(argp, &irq_event,
3179 sizeof irq_event))
3180 goto out;
3181 }
1fe779f8
CO		 3182 r = 0;
3183 }
3184 break;
3185 }
3186 case KVM_GET_IRQCHIP: {
3187 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
f0d66275 3188 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1fe779f8 3189
f0d66275
DH		 3190 r = -ENOMEM;
3191 if (!chip)
1fe779f8 3192 goto out;
f0d66275
DH		 3193 r = -EFAULT;
3194 if (copy_from_user(chip, argp, sizeof *chip))
3195 goto get_irqchip_out;
1fe779f8
CO		 3196 r = -ENXIO;
3197 if (!irqchip_in_kernel(kvm))
f0d66275
DH		 3198 goto get_irqchip_out;
3199 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1fe779f8 3200 if (r)
f0d66275 3201 goto get_irqchip_out;
1fe779f8 3202 r = -EFAULT;
f0d66275
DH		 3203 if (copy_to_user(argp, chip, sizeof *chip))
3204 goto get_irqchip_out;
1fe779f8 3205 r = 0;
f0d66275
DH		 3206 get_irqchip_out:
3207 kfree(chip);
3208 if (r)
3209 goto out;
1fe779f8
CO		 3210 break;
3211 }
3212 case KVM_SET_IRQCHIP: {
3213 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
f0d66275 3214 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
1fe779f8 3215
f0d66275
DH		 3216 r = -ENOMEM;
3217 if (!chip)
1fe779f8 3218 goto out;
f0d66275
DH		 3219 r = -EFAULT;
3220 if (copy_from_user(chip, argp, sizeof *chip))
3221 goto set_irqchip_out;
1fe779f8
CO		 3222 r = -ENXIO;
3223 if (!irqchip_in_kernel(kvm))
f0d66275
DH		 3224 goto set_irqchip_out;
3225 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1fe779f8 3226 if (r)
f0d66275 3227 goto set_irqchip_out;
1fe779f8 3228 r = 0;
f0d66275
DH		 3229 set_irqchip_out:
3230 kfree(chip);
3231 if (r)
3232 goto out;
1fe779f8
CO		 3233 break;
3234 }
e0f63cb9 3235 case KVM_GET_PIT: {
e0f63cb9 3236 r = -EFAULT;
f0d66275 3237 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 3238 goto out;
3239 r = -ENXIO;
3240 if (!kvm->arch.vpit)
3241 goto out;
f0d66275 3242 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
e0f63cb9
SY		 3243 if (r)
3244 goto out;
3245 r = -EFAULT;
f0d66275 3246 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 3247 goto out;
3248 r = 0;
3249 break;
3250 }
3251 case KVM_SET_PIT: {
e0f63cb9 3252 r = -EFAULT;
f0d66275 3253 if (copy_from_user(&u.ps, argp, sizeof u.ps))
e0f63cb9
SY		 3254 goto out;
3255 r = -ENXIO;
3256 if (!kvm->arch.vpit)
3257 goto out;
f0d66275 3258 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
e0f63cb9
SY		 3259 if (r)
3260 goto out;
3261 r = 0;
3262 break;
3263 }
e9f42757
BK		 3264 case KVM_GET_PIT2: {
3265 r = -ENXIO;
3266 if (!kvm->arch.vpit)
3267 goto out;
3268 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3269 if (r)
3270 goto out;
3271 r = -EFAULT;
3272 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3273 goto out;
3274 r = 0;
3275 break;
3276 }
3277 case KVM_SET_PIT2: {
3278 r = -EFAULT;
3279 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3280 goto out;
3281 r = -ENXIO;
3282 if (!kvm->arch.vpit)
3283 goto out;
3284 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3285 if (r)
3286 goto out;
3287 r = 0;
3288 break;
3289 }
52d939a0
MT		 3290 case KVM_REINJECT_CONTROL: {
3291 struct kvm_reinject_control control;
3292 r = -EFAULT;
3293 if (copy_from_user(&control, argp, sizeof(control)))
3294 goto out;
3295 r = kvm_vm_ioctl_reinject(kvm, &control);
3296 if (r)
3297 goto out;
3298 r = 0;
3299 break;
3300 }
ffde22ac
ES		 3301 case KVM_XEN_HVM_CONFIG: {
3302 r = -EFAULT;
3303 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3304 sizeof(struct kvm_xen_hvm_config)))
3305 goto out;
3306 r = -EINVAL;
3307 if (kvm->arch.xen_hvm_config.flags)
3308 goto out;
3309 r = 0;
3310 break;
3311 }
afbcf7ab
GC		 3312 case KVM_SET_CLOCK: {
3313 struct timespec now;
3314 struct kvm_clock_data user_ns;
3315 u64 now_ns;
3316 s64 delta;
3317
3318 r = -EFAULT;
3319 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3320 goto out;
3321
3322 r = -EINVAL;
3323 if (user_ns.flags)
3324 goto out;
3325
3326 r = 0;
3327 ktime_get_ts(&now);
3328 now_ns = timespec_to_ns(&now);
3329 delta = user_ns.clock - now_ns;
3330 kvm->arch.kvmclock_offset = delta;
3331 break;
3332 }
3333 case KVM_GET_CLOCK: {
3334 struct timespec now;
3335 struct kvm_clock_data user_ns;
3336 u64 now_ns;
3337
3338 ktime_get_ts(&now);
3339 now_ns = timespec_to_ns(&now);
3340 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3341 user_ns.flags = 0;
3342
3343 r = -EFAULT;
3344 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3345 goto out;
3346 r = 0;
3347 break;
3348 }
3349
1fe779f8
CO		 3350 default:
3351 ;
3352 }
3353out:
3354 return r;
3355}
3356
a16b043c 3357static void kvm_init_msr_list(void)
043405e1
CO		 3358{
3359 u32 dummy[2];
3360 unsigned i, j;
3361
e3267cbb
GC		 3362 /* skip the first msrs in the list. KVM-specific */
3363 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
043405e1
CO		 3364 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3365 continue;
3366 if (j < i)
3367 msrs_to_save[j] = msrs_to_save[i];
3368 j++;
3369 }
3370 num_msrs_to_save = j;
3371}
3372
bda9020e
MT		 3373static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3374 const void *v)
bbd9b64e 3375{
bda9020e
MT		 3376 if (vcpu->arch.apic &&
3377 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3378 return 0;
bbd9b64e 3379
e93f8a0f 3380 return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
bbd9b64e
CO		 3381}
3382
bda9020e 3383static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
bbd9b64e 3384{
bda9020e
MT		 3385 if (vcpu->arch.apic &&
3386 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3387 return 0;
bbd9b64e 3388
e93f8a0f 3389 return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
bbd9b64e
CO		 3390}
3391
2dafc6c2
GN		 3392static void kvm_set_segment(struct kvm_vcpu *vcpu,
3393 struct kvm_segment *var, int seg)
3394{
3395 kvm_x86_ops->set_segment(vcpu, var, seg);
3396}
3397
3398void kvm_get_segment(struct kvm_vcpu *vcpu,
3399 struct kvm_segment *var, int seg)
3400{
3401 kvm_x86_ops->get_segment(vcpu, var, seg);
3402}
3403
1871c602
GN		 3404gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3405{
3406 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3407 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3408}
3409
3410 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3411{
3412 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3413 access |= PFERR_FETCH_MASK;
3414 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3415}
3416
3417gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3418{
3419 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3420 access |= PFERR_WRITE_MASK;
3421 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3422}
3423
3424/* uses this to access any guest's mapped memory without checking CPL */
3425gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3426{
3427 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, 0, error);
3428}
3429
3430static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3431 struct kvm_vcpu *vcpu, u32 access,
3432 u32 *error)
bbd9b64e
CO		 3433{
3434 void *data = val;
10589a46 3435 int r = X86EMUL_CONTINUE;
bbd9b64e
CO		 3436
3437 while (bytes) {
1871c602 3438 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, access, error);
bbd9b64e 3439 unsigned offset = addr & (PAGE_SIZE-1);
77c2002e 3440 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
bbd9b64e
CO		 3441 int ret;
3442
10589a46
MT		 3443 if (gpa == UNMAPPED_GVA) {
3444 r = X86EMUL_PROPAGATE_FAULT;
3445 goto out;
3446 }
77c2002e 3447 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
10589a46 3448 if (ret < 0) {
c3cd7ffa 3449 r = X86EMUL_IO_NEEDED;
10589a46
MT		 3450 goto out;
3451 }
bbd9b64e 3452
77c2002e
IE		 3453 bytes -= toread;
3454 data += toread;
3455 addr += toread;
bbd9b64e 3456 }
10589a46 3457out:
10589a46 3458 return r;
bbd9b64e 3459}
77c2002e 3460
1871c602
GN		 3461/* used for instruction fetching */
3462static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3463 struct kvm_vcpu *vcpu, u32 *error)
3464{
3465 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3466 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3467 access | PFERR_FETCH_MASK, error);
3468}
3469
3470static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3471 struct kvm_vcpu *vcpu, u32 *error)
3472{
3473 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3474 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3475 error);
3476}
3477
3478static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3479 struct kvm_vcpu *vcpu, u32 *error)
3480{
3481 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3482}
3483
7972995b 3484static int kvm_write_guest_virt_system(gva_t addr, void *val,
2dafc6c2 3485 unsigned int bytes,
7972995b 3486 struct kvm_vcpu *vcpu,
2dafc6c2 3487 u32 *error)
77c2002e
IE		 3488{
3489 void *data = val;
3490 int r = X86EMUL_CONTINUE;
3491
3492 while (bytes) {
7972995b
GN		 3493 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr,
3494 PFERR_WRITE_MASK, error);
77c2002e
IE		 3495 unsigned offset = addr & (PAGE_SIZE-1);
3496 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3497 int ret;
3498
3499 if (gpa == UNMAPPED_GVA) {
3500 r = X86EMUL_PROPAGATE_FAULT;
3501 goto out;
3502 }
3503 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3504 if (ret < 0) {
c3cd7ffa 3505 r = X86EMUL_IO_NEEDED;
77c2002e
IE		 3506 goto out;
3507 }
3508
3509 bytes -= towrite;
3510 data += towrite;
3511 addr += towrite;
3512 }
3513out:
3514 return r;
3515}
3516
bbd9b64e
CO		 3517static int emulator_read_emulated(unsigned long addr,
3518 void *val,
3519 unsigned int bytes,
8fe681e9 3520 unsigned int *error_code,
bbd9b64e
CO		 3521 struct kvm_vcpu *vcpu)
3522{
bbd9b64e
CO		 3523 gpa_t gpa;
3524
3525 if (vcpu->mmio_read_completed) {
3526 memcpy(val, vcpu->mmio_data, bytes);
aec51dc4
AK		 3527 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3528 vcpu->mmio_phys_addr, *(u64 *)val);
bbd9b64e
CO		 3529 vcpu->mmio_read_completed = 0;
3530 return X86EMUL_CONTINUE;
3531 }
3532
8fe681e9 3533 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, error_code);
1871c602 3534
8fe681e9 3535 if (gpa == UNMAPPED_GVA)
1871c602 3536 return X86EMUL_PROPAGATE_FAULT;
bbd9b64e
CO		 3537
3538 /* For APIC access vmexit */
3539 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3540 goto mmio;
3541
1871c602 3542 if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
77c2002e 3543 == X86EMUL_CONTINUE)
bbd9b64e 3544 return X86EMUL_CONTINUE;
bbd9b64e
CO		 3545
3546mmio:
3547 /*
3548 * Is this MMIO handled locally?
3549 */
aec51dc4
AK		 3550 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3551 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
bbd9b64e
CO		 3552 return X86EMUL_CONTINUE;
3553 }
aec51dc4
AK		 3554
3555 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
bbd9b64e
CO		 3556
3557 vcpu->mmio_needed = 1;
411c35b7
GN		 3558 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3559 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3560 vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3561 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
bbd9b64e 3562
c3cd7ffa 3563 return X86EMUL_IO_NEEDED;
bbd9b64e
CO		 3564}
3565
3200f405 3566int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
9f811285 3567 const void *val, int bytes)
bbd9b64e
CO		 3568{
3569 int ret;
3570
3571 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
9f811285 3572 if (ret < 0)
bbd9b64e 3573 return 0;
ad218f85 3574 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
bbd9b64e
CO		 3575 return 1;
3576}
3577
3578static int emulator_write_emulated_onepage(unsigned long addr,
3579 const void *val,
3580 unsigned int bytes,
8fe681e9 3581 unsigned int *error_code,
bbd9b64e
CO		 3582 struct kvm_vcpu *vcpu)
3583{
10589a46
MT		 3584 gpa_t gpa;
3585
8fe681e9 3586 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error_code);
bbd9b64e 3587
8fe681e9 3588 if (gpa == UNMAPPED_GVA)
bbd9b64e 3589 return X86EMUL_PROPAGATE_FAULT;
bbd9b64e
CO		 3590
3591 /* For APIC access vmexit */
3592 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3593 goto mmio;
3594
3595 if (emulator_write_phys(vcpu, gpa, val, bytes))
3596 return X86EMUL_CONTINUE;
3597
3598mmio:
aec51dc4 3599 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
bbd9b64e
CO		 3600 /*
3601 * Is this MMIO handled locally?
3602 */
bda9020e 3603 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
bbd9b64e 3604 return X86EMUL_CONTINUE;
bbd9b64e
CO		 3605
3606 vcpu->mmio_needed = 1;
411c35b7
GN		 3607 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3608 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3609 vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3610 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3611 memcpy(vcpu->run->mmio.data, val, bytes);
bbd9b64e
CO		 3612
3613 return X86EMUL_CONTINUE;
3614}
3615
3616int emulator_write_emulated(unsigned long addr,
8f6abd06
GN		 3617 const void *val,
3618 unsigned int bytes,
8fe681e9 3619 unsigned int *error_code,
8f6abd06 3620 struct kvm_vcpu *vcpu)
bbd9b64e
CO		 3621{
3622 /* Crossing a page boundary? */
3623 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3624 int rc, now;
3625
3626 now = -addr & ~PAGE_MASK;
8fe681e9
GN		 3627 rc = emulator_write_emulated_onepage(addr, val, now, error_code,
3628 vcpu);
bbd9b64e
CO		 3629 if (rc != X86EMUL_CONTINUE)
3630 return rc;
3631 addr += now;
3632 val += now;
3633 bytes -= now;
3634 }
8fe681e9
GN		 3635 return emulator_write_emulated_onepage(addr, val, bytes, error_code,
3636 vcpu);
bbd9b64e 3637}
bbd9b64e 3638
daea3e73
AK		 3639#define CMPXCHG_TYPE(t, ptr, old, new) \
3640 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3641
3642#ifdef CONFIG_X86_64
3643# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3644#else
3645# define CMPXCHG64(ptr, old, new) \
9749a6c0 3646 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
daea3e73
AK		 3647#endif
3648
bbd9b64e
CO		 3649static int emulator_cmpxchg_emulated(unsigned long addr,
3650 const void *old,
3651 const void *new,
3652 unsigned int bytes,
8fe681e9 3653 unsigned int *error_code,
bbd9b64e
CO		 3654 struct kvm_vcpu *vcpu)
3655{
daea3e73
AK		 3656 gpa_t gpa;
3657 struct page *page;
3658 char *kaddr;
3659 bool exchanged;
2bacc55c 3660
daea3e73
AK		 3661 /* guests cmpxchg8b have to be emulated atomically */
3662 if (bytes > 8 || (bytes & (bytes - 1)))
3663 goto emul_write;
10589a46 3664
daea3e73 3665 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
2bacc55c 3666
daea3e73
AK		 3667 if (gpa == UNMAPPED_GVA ||
3668 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3669 goto emul_write;
2bacc55c 3670
daea3e73
AK		 3671 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3672 goto emul_write;
72dc67a6 3673
daea3e73 3674 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
72dc67a6 3675
daea3e73
AK		 3676 kaddr = kmap_atomic(page, KM_USER0);
3677 kaddr += offset_in_page(gpa);
3678 switch (bytes) {
3679 case 1:
3680 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3681 break;
3682 case 2:
3683 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3684 break;
3685 case 4:
3686 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3687 break;
3688 case 8:
3689 exchanged = CMPXCHG64(kaddr, old, new);
3690 break;
3691 default:
3692 BUG();
2bacc55c 3693 }
daea3e73
AK		 3694 kunmap_atomic(kaddr, KM_USER0);
3695 kvm_release_page_dirty(page);
3696
3697 if (!exchanged)
3698 return X86EMUL_CMPXCHG_FAILED;
3699
8f6abd06
GN		 3700 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3701
3702 return X86EMUL_CONTINUE;
4a5f48f6 3703
3200f405 3704emul_write:
daea3e73 3705 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2bacc55c 3706
8fe681e9 3707 return emulator_write_emulated(addr, new, bytes, error_code, vcpu);
bbd9b64e
CO		 3708}
3709
cf8f70bf
GN		 3710static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3711{
3712 /* TODO: String I/O for in kernel device */
3713 int r;
3714
3715 if (vcpu->arch.pio.in)
3716 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3717 vcpu->arch.pio.size, pd);
3718 else
3719 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3720 vcpu->arch.pio.port, vcpu->arch.pio.size,
3721 pd);
3722 return r;
3723}
3724
3725
3726static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
3727 unsigned int count, struct kvm_vcpu *vcpu)
3728{
7972995b 3729 if (vcpu->arch.pio.count)
cf8f70bf
GN		 3730 goto data_avail;
3731
3732 trace_kvm_pio(1, port, size, 1);
3733
3734 vcpu->arch.pio.port = port;
3735 vcpu->arch.pio.in = 1;
7972995b 3736 vcpu->arch.pio.count = count;
cf8f70bf
GN		 3737 vcpu->arch.pio.size = size;
3738
3739 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3740 data_avail:
3741 memcpy(val, vcpu->arch.pio_data, size * count);
7972995b 3742 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 3743 return 1;
3744 }
3745
3746 vcpu->run->exit_reason = KVM_EXIT_IO;
3747 vcpu->run->io.direction = KVM_EXIT_IO_IN;
3748 vcpu->run->io.size = size;
3749 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3750 vcpu->run->io.count = count;
3751 vcpu->run->io.port = port;
3752
3753 return 0;
3754}
3755
3756static int emulator_pio_out_emulated(int size, unsigned short port,
3757 const void *val, unsigned int count,
3758 struct kvm_vcpu *vcpu)
3759{
3760 trace_kvm_pio(0, port, size, 1);
3761
3762 vcpu->arch.pio.port = port;
3763 vcpu->arch.pio.in = 0;
7972995b 3764 vcpu->arch.pio.count = count;
cf8f70bf
GN		 3765 vcpu->arch.pio.size = size;
3766
3767 memcpy(vcpu->arch.pio_data, val, size * count);
3768
3769 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
7972995b 3770 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 3771 return 1;
3772 }
3773
3774 vcpu->run->exit_reason = KVM_EXIT_IO;
3775 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
3776 vcpu->run->io.size = size;
3777 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3778 vcpu->run->io.count = count;
3779 vcpu->run->io.port = port;
3780
3781 return 0;
3782}
3783
bbd9b64e
CO		 3784static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3785{
3786 return kvm_x86_ops->get_segment_base(vcpu, seg);
3787}
3788
3789int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3790{
a7052897 3791 kvm_mmu_invlpg(vcpu, address);
bbd9b64e
CO		 3792 return X86EMUL_CONTINUE;
3793}
3794
3795int emulate_clts(struct kvm_vcpu *vcpu)
3796{
4d4ec087 3797 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
6b52d186 3798 kvm_x86_ops->fpu_activate(vcpu);
bbd9b64e
CO		 3799 return X86EMUL_CONTINUE;
3800}
3801
35aa5375 3802int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
bbd9b64e 3803{
338dbc97 3804 return _kvm_get_dr(vcpu, dr, dest);
bbd9b64e
CO		 3805}
3806
35aa5375 3807int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
bbd9b64e 3808{
338dbc97
GN		 3809
3810 return __kvm_set_dr(vcpu, dr, value);
bbd9b64e
CO		 3811}
3812
52a46617 3813static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5fdbf976 3814{
52a46617 3815 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5fdbf976
MT		 3816}
3817
52a46617 3818static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
bbd9b64e 3819{
52a46617
GN		 3820 unsigned long value;
3821
3822 switch (cr) {
3823 case 0:
3824 value = kvm_read_cr0(vcpu);
3825 break;
3826 case 2:
3827 value = vcpu->arch.cr2;
3828 break;
3829 case 3:
3830 value = vcpu->arch.cr3;
3831 break;
3832 case 4:
3833 value = kvm_read_cr4(vcpu);
3834 break;
3835 case 8:
3836 value = kvm_get_cr8(vcpu);
3837 break;
3838 default:
3839 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3840 return 0;
3841 }
3842
3843 return value;
3844}
3845
0f12244f 3846static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
52a46617 3847{
0f12244f
GN		 3848 int res = 0;
3849
52a46617
GN		 3850 switch (cr) {
3851 case 0:
49a9b07e 3852 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
52a46617
GN		 3853 break;
3854 case 2:
3855 vcpu->arch.cr2 = val;
3856 break;
3857 case 3:
2390218b 3858 res = kvm_set_cr3(vcpu, val);
52a46617
GN		 3859 break;
3860 case 4:
a83b29c6 3861 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
52a46617
GN		 3862 break;
3863 case 8:
0f12244f 3864 res = __kvm_set_cr8(vcpu, val & 0xfUL);
52a46617
GN		 3865 break;
3866 default:
3867 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
0f12244f 3868 res = -1;
52a46617 3869 }
0f12244f
GN		 3870
3871 return res;
52a46617
GN		 3872}
3873
9c537244
GN		 3874static int emulator_get_cpl(struct kvm_vcpu *vcpu)
3875{
3876 return kvm_x86_ops->get_cpl(vcpu);
3877}
3878
2dafc6c2
GN		 3879static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
3880{
3881 kvm_x86_ops->get_gdt(vcpu, dt);
3882}
3883
5951c442
GN		 3884static unsigned long emulator_get_cached_segment_base(int seg,
3885 struct kvm_vcpu *vcpu)
3886{
3887 return get_segment_base(vcpu, seg);
3888}
3889
2dafc6c2
GN		 3890static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
3891 struct kvm_vcpu *vcpu)
3892{
3893 struct kvm_segment var;
3894
3895 kvm_get_segment(vcpu, &var, seg);
3896
3897 if (var.unusable)
3898 return false;
3899
3900 if (var.g)
3901 var.limit >>= 12;
3902 set_desc_limit(desc, var.limit);
3903 set_desc_base(desc, (unsigned long)var.base);
3904 desc->type = var.type;
3905 desc->s = var.s;
3906 desc->dpl = var.dpl;
3907 desc->p = var.present;
3908 desc->avl = var.avl;
3909 desc->l = var.l;
3910 desc->d = var.db;
3911 desc->g = var.g;
3912
3913 return true;
3914}
3915
3916static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg,
3917 struct kvm_vcpu *vcpu)
3918{
3919 struct kvm_segment var;
3920
3921 /* needed to preserve selector */
3922 kvm_get_segment(vcpu, &var, seg);
3923
3924 var.base = get_desc_base(desc);
3925 var.limit = get_desc_limit(desc);
3926 if (desc->g)
3927 var.limit = (var.limit << 12) | 0xfff;
3928 var.type = desc->type;
3929 var.present = desc->p;
3930 var.dpl = desc->dpl;
3931 var.db = desc->d;
3932 var.s = desc->s;
3933 var.l = desc->l;
3934 var.g = desc->g;
3935 var.avl = desc->avl;
3936 var.present = desc->p;
3937 var.unusable = !var.present;
3938 var.padding = 0;
3939
3940 kvm_set_segment(vcpu, &var, seg);
3941 return;
3942}
3943
3944static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
3945{
3946 struct kvm_segment kvm_seg;
3947
3948 kvm_get_segment(vcpu, &kvm_seg, seg);
3949 return kvm_seg.selector;
3950}
3951
3952static void emulator_set_segment_selector(u16 sel, int seg,
3953 struct kvm_vcpu *vcpu)
3954{
3955 struct kvm_segment kvm_seg;
3956
3957 kvm_get_segment(vcpu, &kvm_seg, seg);
3958 kvm_seg.selector = sel;
3959 kvm_set_segment(vcpu, &kvm_seg, seg);
3960}
3961
14af3f3c 3962static struct x86_emulate_ops emulate_ops = {
1871c602 3963 .read_std = kvm_read_guest_virt_system,
2dafc6c2 3964 .write_std = kvm_write_guest_virt_system,
1871c602 3965 .fetch = kvm_fetch_guest_virt,
bbd9b64e
CO		 3966 .read_emulated = emulator_read_emulated,
3967 .write_emulated = emulator_write_emulated,
3968 .cmpxchg_emulated = emulator_cmpxchg_emulated,
cf8f70bf
GN		 3969 .pio_in_emulated = emulator_pio_in_emulated,
3970 .pio_out_emulated = emulator_pio_out_emulated,
2dafc6c2
GN		 3971 .get_cached_descriptor = emulator_get_cached_descriptor,
3972 .set_cached_descriptor = emulator_set_cached_descriptor,
3973 .get_segment_selector = emulator_get_segment_selector,
3974 .set_segment_selector = emulator_set_segment_selector,
5951c442 3975 .get_cached_segment_base = emulator_get_cached_segment_base,
2dafc6c2 3976 .get_gdt = emulator_get_gdt,
52a46617
GN		 3977 .get_cr = emulator_get_cr,
3978 .set_cr = emulator_set_cr,
9c537244 3979 .cpl = emulator_get_cpl,
35aa5375
GN		 3980 .get_dr = emulator_get_dr,
3981 .set_dr = emulator_set_dr,
3fb1b5db
GN		 3982 .set_msr = kvm_set_msr,
3983 .get_msr = kvm_get_msr,
bbd9b64e
CO		 3984};
3985
5fdbf976
MT		 3986static void cache_all_regs(struct kvm_vcpu *vcpu)
3987{
3988 kvm_register_read(vcpu, VCPU_REGS_RAX);
3989 kvm_register_read(vcpu, VCPU_REGS_RSP);
3990 kvm_register_read(vcpu, VCPU_REGS_RIP);
3991 vcpu->arch.regs_dirty = ~0;
3992}
3993
95cb2295
GN		 3994static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
3995{
3996 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
3997 /*
3998 * an sti; sti; sequence only disable interrupts for the first
3999 * instruction. So, if the last instruction, be it emulated or
4000 * not, left the system with the INT_STI flag enabled, it
4001 * means that the last instruction is an sti. We should not
4002 * leave the flag on in this case. The same goes for mov ss
4003 */
4004 if (!(int_shadow & mask))
4005 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4006}
4007
54b8486f
GN		 4008static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4009{
4010 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4011 if (ctxt->exception == PF_VECTOR)
4012 kvm_inject_page_fault(vcpu, ctxt->cr2, ctxt->error_code);
4013 else if (ctxt->error_code_valid)
4014 kvm_queue_exception_e(vcpu, ctxt->exception, ctxt->error_code);
4015 else
4016 kvm_queue_exception(vcpu, ctxt->exception);
4017}
4018
6d77dbfc
GN		 4019static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4020{
6d77dbfc
GN		 4021 ++vcpu->stat.insn_emulation_fail;
4022 trace_kvm_emulate_insn_failed(vcpu);
4023 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4024 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4025 vcpu->run->internal.ndata = 0;
4026 kvm_queue_exception(vcpu, UD_VECTOR);
4027 return EMULATE_FAIL;
4028}
4029
bbd9b64e 4030int emulate_instruction(struct kvm_vcpu *vcpu,
bbd9b64e
CO		 4031 unsigned long cr2,
4032 u16 error_code,
571008da 4033 int emulation_type)
bbd9b64e 4034{
95cb2295 4035 int r;
4d2179e1 4036 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
bbd9b64e 4037
26eef70c 4038 kvm_clear_exception_queue(vcpu);
ad312c7c 4039 vcpu->arch.mmio_fault_cr2 = cr2;
5fdbf976 4040 /*
56e82318 4041 * TODO: fix emulate.c to use guest_read/write_register
5fdbf976
MT		 4042 * instead of direct ->regs accesses, can save hundred cycles
4043 * on Intel for instructions that don't read/change RSP, for
4044 * for example.
4045 */
4046 cache_all_regs(vcpu);
bbd9b64e 4047
571008da 4048 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
bbd9b64e
CO		 4049 int cs_db, cs_l;
4050 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4051
ad312c7c 4052 vcpu->arch.emulate_ctxt.vcpu = vcpu;
83bf0002 4053 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
063db061 4054 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
ad312c7c 4055 vcpu->arch.emulate_ctxt.mode =
a0044755 4056 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
ad312c7c 4057 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
a0044755 4058 ? X86EMUL_MODE_VM86 : cs_l
bbd9b64e
CO		 4059 ? X86EMUL_MODE_PROT64 : cs_db
4060 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4d2179e1
GN		 4061 memset(c, 0, sizeof(struct decode_cache));
4062 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
95cb2295 4063 vcpu->arch.emulate_ctxt.interruptibility = 0;
54b8486f 4064 vcpu->arch.emulate_ctxt.exception = -1;
bbd9b64e 4065
ad312c7c 4066 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
e46479f8 4067 trace_kvm_emulate_insn_start(vcpu);
571008da 4068
0cb5762e
AP		 4069 /* Only allow emulation of specific instructions on #UD
4070 * (namely VMMCALL, sysenter, sysexit, syscall)*/
0cb5762e
AP		 4071 if (emulation_type & EMULTYPE_TRAP_UD) {
4072 if (!c->twobyte)
4073 return EMULATE_FAIL;
4074 switch (c->b) {
4075 case 0x01: /* VMMCALL */
4076 if (c->modrm_mod != 3 || c->modrm_rm != 1)
4077 return EMULATE_FAIL;
4078 break;
4079 case 0x34: /* sysenter */
4080 case 0x35: /* sysexit */
4081 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4082 return EMULATE_FAIL;
4083 break;
4084 case 0x05: /* syscall */
4085 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4086 return EMULATE_FAIL;
4087 break;
4088 default:
4089 return EMULATE_FAIL;
4090 }
4091
4092 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
4093 return EMULATE_FAIL;
4094 }
571008da 4095
f2b5756b 4096 ++vcpu->stat.insn_emulation;
bbd9b64e
CO		 4097 if (r) {
4098 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
4099 return EMULATE_DONE;
6d77dbfc
GN		 4100 if (emulation_type & EMULTYPE_SKIP)
4101 return EMULATE_FAIL;
4102 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 4103 }
4104 }
4105
ba8afb6b
GN		 4106 if (emulation_type & EMULTYPE_SKIP) {
4107 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4108 return EMULATE_DONE;
4109 }
4110
4d2179e1
GN		 4111 /* this is needed for vmware backdor interface to work since it
4112 changes registers values during IO operation */
4113 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4114
5cd21917 4115restart:
ad312c7c 4116 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
bbd9b64e 4117
c3cd7ffa
GN		 4118 if (r) { /* emulation failed */
4119 /*
4120 * if emulation was due to access to shadowed page table
4121 * and it failed try to unshadow page and re-entetr the
4122 * guest to let CPU execute the instruction.
4123 */
4124 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
4125 return EMULATE_DONE;
4126
6d77dbfc 4127 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 4128 }
4129
95cb2295 4130 toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility);
ef050dc0 4131 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4d2179e1 4132 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
95c55886 4133 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
3457e419 4134
54b8486f
GN		 4135 if (vcpu->arch.emulate_ctxt.exception >= 0) {
4136 inject_emulated_exception(vcpu);
4137 return EMULATE_DONE;
4138 }
4139
3457e419
GN		 4140 if (vcpu->arch.pio.count) {
4141 if (!vcpu->arch.pio.in)
4142 vcpu->arch.pio.count = 0;
4143 return EMULATE_DO_MMIO;
4144 }
4145
4146 if (vcpu->mmio_needed) {
4147 if (vcpu->mmio_is_write)
4148 vcpu->mmio_needed = 0;
4149 return EMULATE_DO_MMIO;
4150 }
4151
5cd21917
GN		 4152 if (vcpu->arch.emulate_ctxt.restart)
4153 goto restart;
f850e2e6 4154
bbd9b64e 4155 return EMULATE_DONE;
de7d789a 4156}
bbd9b64e 4157EXPORT_SYMBOL_GPL(emulate_instruction);
de7d789a 4158
cf8f70bf 4159int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
de7d789a 4160{
cf8f70bf
GN		 4161 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4162 int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
4163 /* do not return to emulator after return from userspace */
7972995b 4164 vcpu->arch.pio.count = 0;
de7d789a
CO		 4165 return ret;
4166}
cf8f70bf 4167EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
de7d789a 4168
c8076604
GH		 4169static void bounce_off(void *info)
4170{
4171 /* nothing */
4172}
4173
c8076604
GH		 4174static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4175 void *data)
4176{
4177 struct cpufreq_freqs *freq = data;
4178 struct kvm *kvm;
4179 struct kvm_vcpu *vcpu;
4180 int i, send_ipi = 0;
4181
c8076604
GH		 4182 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4183 return 0;
4184 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4185 return 0;
0cca7907 4186 per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
c8076604
GH		 4187
4188 spin_lock(&kvm_lock);
4189 list_for_each_entry(kvm, &vm_list, vm_list) {
988a2cae 4190 kvm_for_each_vcpu(i, vcpu, kvm) {
c8076604
GH		 4191 if (vcpu->cpu != freq->cpu)
4192 continue;
4193 if (!kvm_request_guest_time_update(vcpu))
4194 continue;
4195 if (vcpu->cpu != smp_processor_id())
4196 send_ipi++;
4197 }
4198 }
4199 spin_unlock(&kvm_lock);
4200
4201 if (freq->old < freq->new && send_ipi) {
4202 /*
4203 * We upscale the frequency. Must make the guest
4204 * doesn't see old kvmclock values while running with
4205 * the new frequency, otherwise we risk the guest sees
4206 * time go backwards.
4207 *
4208 * In case we update the frequency for another cpu
4209 * (which might be in guest context) send an interrupt
4210 * to kick the cpu out of guest context. Next time
4211 * guest context is entered kvmclock will be updated,
4212 * so the guest will not see stale values.
4213 */
4214 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
4215 }
4216 return 0;
4217}
4218
4219static struct notifier_block kvmclock_cpufreq_notifier_block = {
4220 .notifier_call = kvmclock_cpufreq_notifier
4221};
4222
b820cc0c
ZA		 4223static void kvm_timer_init(void)
4224{
4225 int cpu;
4226
b820cc0c 4227 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
b820cc0c
ZA		 4228 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4229 CPUFREQ_TRANSITION_NOTIFIER);
6b7d7e76
ZA		 4230 for_each_online_cpu(cpu) {
4231 unsigned long khz = cpufreq_get(cpu);
4232 if (!khz)
4233 khz = tsc_khz;
4234 per_cpu(cpu_tsc_khz, cpu) = khz;
4235 }
0cca7907
ZA		 4236 } else {
4237 for_each_possible_cpu(cpu)
4238 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
b820cc0c
ZA		 4239 }
4240}
4241
ff9d07a0
ZY		 4242static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4243
4244static int kvm_is_in_guest(void)
4245{
4246 return percpu_read(current_vcpu) != NULL;
4247}
4248
4249static int kvm_is_user_mode(void)
4250{
4251 int user_mode = 3;
dcf46b94 4252
ff9d07a0
ZY		 4253 if (percpu_read(current_vcpu))
4254 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
dcf46b94 4255
ff9d07a0
ZY		 4256 return user_mode != 0;
4257}
4258
4259static unsigned long kvm_get_guest_ip(void)
4260{
4261 unsigned long ip = 0;
dcf46b94 4262
ff9d07a0
ZY		 4263 if (percpu_read(current_vcpu))
4264 ip = kvm_rip_read(percpu_read(current_vcpu));
dcf46b94 4265
ff9d07a0
ZY		 4266 return ip;
4267}
4268
4269static struct perf_guest_info_callbacks kvm_guest_cbs = {
4270 .is_in_guest = kvm_is_in_guest,
4271 .is_user_mode = kvm_is_user_mode,
4272 .get_guest_ip = kvm_get_guest_ip,
4273};
4274
4275void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4276{
4277 percpu_write(current_vcpu, vcpu);
4278}
4279EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4280
4281void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4282{
4283 percpu_write(current_vcpu, NULL);
4284}
4285EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4286
f8c16bba 4287int kvm_arch_init(void *opaque)
043405e1 4288{
b820cc0c 4289 int r;
f8c16bba
ZX		 4290 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4291
f8c16bba
ZX		 4292 if (kvm_x86_ops) {
4293 printk(KERN_ERR "kvm: already loaded the other module\n");
56c6d28a
ZX		 4294 r = -EEXIST;
4295 goto out;
f8c16bba
ZX		 4296 }
4297
4298 if (!ops->cpu_has_kvm_support()) {
4299 printk(KERN_ERR "kvm: no hardware support\n");
56c6d28a
ZX		 4300 r = -EOPNOTSUPP;
4301 goto out;
f8c16bba
ZX		 4302 }
4303 if (ops->disabled_by_bios()) {
4304 printk(KERN_ERR "kvm: disabled by bios\n");
56c6d28a
ZX		 4305 r = -EOPNOTSUPP;
4306 goto out;
f8c16bba
ZX		 4307 }
4308
97db56ce
AK		 4309 r = kvm_mmu_module_init();
4310 if (r)
4311 goto out;
4312
4313 kvm_init_msr_list();
4314
f8c16bba 4315 kvm_x86_ops = ops;
56c6d28a 4316 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
7b52345e
SY		 4317 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
4318 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4b12f0de 4319 PT_DIRTY_MASK, PT64_NX_MASK, 0);
c8076604 4320
b820cc0c 4321 kvm_timer_init();
c8076604 4322
ff9d07a0
ZY		 4323 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4324
2acf923e
DC		 4325 if (cpu_has_xsave)
4326 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4327
f8c16bba 4328 return 0;
56c6d28a
ZX		 4329
4330out:
56c6d28a 4331 return r;
043405e1 4332}
8776e519 4333
f8c16bba
ZX		 4334void kvm_arch_exit(void)
4335{
ff9d07a0
ZY		 4336 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4337
888d256e
JK		 4338 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4339 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4340 CPUFREQ_TRANSITION_NOTIFIER);
f8c16bba 4341 kvm_x86_ops = NULL;
56c6d28a
ZX		 4342 kvm_mmu_module_exit();
4343}
f8c16bba 4344
8776e519
HB		 4345int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4346{
4347 ++vcpu->stat.halt_exits;
4348 if (irqchip_in_kernel(vcpu->kvm)) {
a4535290 4349 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
8776e519
HB		 4350 return 1;
4351 } else {
4352 vcpu->run->exit_reason = KVM_EXIT_HLT;
4353 return 0;
4354 }
4355}
4356EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4357
2f333bcb
MT		 4358static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4359 unsigned long a1)
4360{
4361 if (is_long_mode(vcpu))
4362 return a0;
4363 else
4364 return a0 | ((gpa_t)a1 << 32);
4365}
4366
55cd8e5a
GN		 4367int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4368{
4369 u64 param, ingpa, outgpa, ret;
4370 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4371 bool fast, longmode;
4372 int cs_db, cs_l;
4373
4374 /*
4375 * hypercall generates UD from non zero cpl and real mode
4376 * per HYPER-V spec
4377 */
3eeb3288 4378 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
55cd8e5a
GN		 4379 kvm_queue_exception(vcpu, UD_VECTOR);
4380 return 0;
4381 }
4382
4383 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4384 longmode = is_long_mode(vcpu) && cs_l == 1;
4385
4386 if (!longmode) {
ccd46936
GN		 4387 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4388 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4389 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4390 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4391 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4392 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
55cd8e5a
GN		 4393 }
4394#ifdef CONFIG_X86_64
4395 else {
4396 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4397 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4398 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4399 }
4400#endif
4401
4402 code = param & 0xffff;
4403 fast = (param >> 16) & 0x1;
4404 rep_cnt = (param >> 32) & 0xfff;
4405 rep_idx = (param >> 48) & 0xfff;
4406
4407 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4408
c25bc163
GN		 4409 switch (code) {
4410 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4411 kvm_vcpu_on_spin(vcpu);
4412 break;
4413 default:
4414 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4415 break;
4416 }
55cd8e5a
GN		 4417
4418 ret = res | (((u64)rep_done & 0xfff) << 32);
4419 if (longmode) {
4420 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4421 } else {
4422 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4423 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4424 }
4425
4426 return 1;
4427}
4428
8776e519
HB		 4429int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4430{
4431 unsigned long nr, a0, a1, a2, a3, ret;
2f333bcb 4432 int r = 1;
8776e519 4433
55cd8e5a
GN		 4434 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4435 return kvm_hv_hypercall(vcpu);
4436
5fdbf976
MT		 4437 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4438 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4439 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4440 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4441 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
8776e519 4442
229456fc 4443 trace_kvm_hypercall(nr, a0, a1, a2, a3);
2714d1d3 4444
8776e519
HB		 4445 if (!is_long_mode(vcpu)) {
4446 nr &= 0xFFFFFFFF;
4447 a0 &= 0xFFFFFFFF;
4448 a1 &= 0xFFFFFFFF;
4449 a2 &= 0xFFFFFFFF;
4450 a3 &= 0xFFFFFFFF;
4451 }
4452
07708c4a
JK		 4453 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4454 ret = -KVM_EPERM;
4455 goto out;
4456 }
4457
8776e519 4458 switch (nr) {
b93463aa
AK		 4459 case KVM_HC_VAPIC_POLL_IRQ:
4460 ret = 0;
4461 break;
2f333bcb
MT		 4462 case KVM_HC_MMU_OP:
4463 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4464 break;
8776e519
HB		 4465 default:
4466 ret = -KVM_ENOSYS;
4467 break;
4468 }
07708c4a 4469out:
5fdbf976 4470 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
f11c3a8d 4471 ++vcpu->stat.hypercalls;
2f333bcb 4472 return r;
8776e519
HB		 4473}
4474EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4475
4476int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4477{
4478 char instruction[3];
5fdbf976 4479 unsigned long rip = kvm_rip_read(vcpu);
8776e519 4480
8776e519
HB		 4481 /*
4482 * Blow out the MMU to ensure that no other VCPU has an active mapping
4483 * to ensure that the updated hypercall appears atomically across all
4484 * VCPUs.
4485 */
4486 kvm_mmu_zap_all(vcpu->kvm);
4487
8776e519 4488 kvm_x86_ops->patch_hypercall(vcpu, instruction);
8776e519 4489
8fe681e9 4490 return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
8776e519
HB		 4491}
4492
8776e519
HB		 4493void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4494{
89a27f4d 4495 struct desc_ptr dt = { limit, base };
8776e519
HB		 4496
4497 kvm_x86_ops->set_gdt(vcpu, &dt);
4498}
4499
4500void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4501{
89a27f4d 4502 struct desc_ptr dt = { limit, base };
8776e519
HB		 4503
4504 kvm_x86_ops->set_idt(vcpu, &dt);
4505}
4506
07716717
DK		 4507static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4508{
ad312c7c
ZX		 4509 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4510 int j, nent = vcpu->arch.cpuid_nent;
07716717
DK		 4511
4512 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4513 /* when no next entry is found, the current entry[i] is reselected */
0fdf8e59 4514 for (j = i + 1; ; j = (j + 1) % nent) {
ad312c7c 4515 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
07716717
DK		 4516 if (ej->function == e->function) {
4517 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4518 return j;
4519 }
4520 }
4521 return 0; /* silence gcc, even though control never reaches here */
4522}
4523
4524/* find an entry with matching function, matching index (if needed), and that
4525 * should be read next (if it's stateful) */
4526static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4527 u32 function, u32 index)
4528{
4529 if (e->function != function)
4530 return 0;
4531 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4532 return 0;
4533 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
19355475 4534 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
07716717
DK		 4535 return 0;
4536 return 1;
4537}
4538
d8017474
AG		 4539struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4540 u32 function, u32 index)
8776e519
HB		 4541{
4542 int i;
d8017474 4543 struct kvm_cpuid_entry2 *best = NULL;
8776e519 4544
ad312c7c 4545 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
d8017474
AG		 4546 struct kvm_cpuid_entry2 *e;
4547
ad312c7c 4548 e = &vcpu->arch.cpuid_entries[i];
07716717
DK		 4549 if (is_matching_cpuid_entry(e, function, index)) {
4550 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4551 move_to_next_stateful_cpuid_entry(vcpu, i);
8776e519
HB		 4552 best = e;
4553 break;
4554 }
4555 /*
4556 * Both basic or both extended?
4557 */
4558 if (((e->function ^ function) & 0x80000000) == 0)
4559 if (!best || e->function > best->function)
4560 best = e;
4561 }
d8017474
AG		 4562 return best;
4563}
0e851880 4564EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
d8017474 4565
82725b20
DE		 4566int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4567{
4568 struct kvm_cpuid_entry2 *best;
4569
f7a71197
AK		 4570 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
4571 if (!best || best->eax < 0x80000008)
4572 goto not_found;
82725b20
DE		 4573 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4574 if (best)
4575 return best->eax & 0xff;
f7a71197 4576not_found:
82725b20
DE		 4577 return 36;
4578}
4579
d8017474
AG		 4580void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4581{
4582 u32 function, index;
4583 struct kvm_cpuid_entry2 *best;
4584
4585 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4586 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4587 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4588 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4589 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4590 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4591 best = kvm_find_cpuid_entry(vcpu, function, index);
8776e519 4592 if (best) {
5fdbf976
MT		 4593 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4594 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4595 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4596 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
8776e519 4597 }
8776e519 4598 kvm_x86_ops->skip_emulated_instruction(vcpu);
229456fc
MT		 4599 trace_kvm_cpuid(function,
4600 kvm_register_read(vcpu, VCPU_REGS_RAX),
4601 kvm_register_read(vcpu, VCPU_REGS_RBX),
4602 kvm_register_read(vcpu, VCPU_REGS_RCX),
4603 kvm_register_read(vcpu, VCPU_REGS_RDX));
8776e519
HB		 4604}
4605EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
d0752060 4606
b6c7a5dc
HB		 4607/*
4608 * Check if userspace requested an interrupt window, and that the
4609 * interrupt window is open.
4610 *
4611 * No need to exit to userspace if we already have an interrupt queued.
4612 */
851ba692 4613static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
b6c7a5dc 4614{
8061823a 4615 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
851ba692 4616 vcpu->run->request_interrupt_window &&
5df56646 4617 kvm_arch_interrupt_allowed(vcpu));
b6c7a5dc
HB		 4618}
4619
851ba692 4620static void post_kvm_run_save(struct kvm_vcpu *vcpu)
b6c7a5dc 4621{
851ba692
AK		 4622 struct kvm_run *kvm_run = vcpu->run;
4623
91586a3b 4624 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2d3ad1f4 4625 kvm_run->cr8 = kvm_get_cr8(vcpu);
b6c7a5dc 4626 kvm_run->apic_base = kvm_get_apic_base(vcpu);
4531220b 4627 if (irqchip_in_kernel(vcpu->kvm))
b6c7a5dc 4628 kvm_run->ready_for_interrupt_injection = 1;
4531220b 4629 else
b6c7a5dc 4630 kvm_run->ready_for_interrupt_injection =
fa9726b0
GN		 4631 kvm_arch_interrupt_allowed(vcpu) &&
4632 !kvm_cpu_has_interrupt(vcpu) &&
4633 !kvm_event_needs_reinjection(vcpu);
b6c7a5dc
HB		 4634}
4635
b93463aa
AK		 4636static void vapic_enter(struct kvm_vcpu *vcpu)
4637{
4638 struct kvm_lapic *apic = vcpu->arch.apic;
4639 struct page *page;
4640
4641 if (!apic || !apic->vapic_addr)
4642 return;
4643
4644 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
72dc67a6
IE		 4645
4646 vcpu->arch.apic->vapic_page = page;
b93463aa
AK		 4647}
4648
4649static void vapic_exit(struct kvm_vcpu *vcpu)
4650{
4651 struct kvm_lapic *apic = vcpu->arch.apic;
f656ce01 4652 int idx;
b93463aa
AK		 4653
4654 if (!apic || !apic->vapic_addr)
4655 return;
4656
f656ce01 4657 idx = srcu_read_lock(&vcpu->kvm->srcu);
b93463aa
AK		 4658 kvm_release_page_dirty(apic->vapic_page);
4659 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
f656ce01 4660 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b93463aa
AK		 4661}
4662
95ba8273
GN		 4663static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4664{
4665 int max_irr, tpr;
4666
4667 if (!kvm_x86_ops->update_cr8_intercept)
4668 return;
4669
88c808fd
AK		 4670 if (!vcpu->arch.apic)
4671 return;
4672
8db3baa2
GN		 4673 if (!vcpu->arch.apic->vapic_addr)
4674 max_irr = kvm_lapic_find_highest_irr(vcpu);
4675 else
4676 max_irr = -1;
95ba8273
GN		 4677
4678 if (max_irr != -1)
4679 max_irr >>= 4;
4680
4681 tpr = kvm_lapic_get_cr8(vcpu);
4682
4683 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4684}
4685
851ba692 4686static void inject_pending_event(struct kvm_vcpu *vcpu)
95ba8273
GN		 4687{
4688 /* try to reinject previous events if any */
b59bb7bd 4689 if (vcpu->arch.exception.pending) {
5c1c85d0
AK		 4690 trace_kvm_inj_exception(vcpu->arch.exception.nr,
4691 vcpu->arch.exception.has_error_code,
4692 vcpu->arch.exception.error_code);
b59bb7bd
GN		 4693 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4694 vcpu->arch.exception.has_error_code,
ce7ddec4
JR		 4695 vcpu->arch.exception.error_code,
4696 vcpu->arch.exception.reinject);
b59bb7bd
GN		 4697 return;
4698 }
4699
95ba8273
GN		 4700 if (vcpu->arch.nmi_injected) {
4701 kvm_x86_ops->set_nmi(vcpu);
4702 return;
4703 }
4704
4705 if (vcpu->arch.interrupt.pending) {
66fd3f7f 4706 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 4707 return;
4708 }
4709
4710 /* try to inject new event if pending */
4711 if (vcpu->arch.nmi_pending) {
4712 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4713 vcpu->arch.nmi_pending = false;
4714 vcpu->arch.nmi_injected = true;
4715 kvm_x86_ops->set_nmi(vcpu);
4716 }
4717 } else if (kvm_cpu_has_interrupt(vcpu)) {
4718 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
66fd3f7f
GN		 4719 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4720 false);
4721 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 4722 }
4723 }
4724}
4725
2acf923e
DC		 4726static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
4727{
4728 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
4729 !vcpu->guest_xcr0_loaded) {
4730 /* kvm_set_xcr() also depends on this */
4731 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
4732 vcpu->guest_xcr0_loaded = 1;
4733 }
4734}
4735
4736static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
4737{
4738 if (vcpu->guest_xcr0_loaded) {
4739 if (vcpu->arch.xcr0 != host_xcr0)
4740 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
4741 vcpu->guest_xcr0_loaded = 0;
4742 }
4743}
4744
851ba692 4745static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
b6c7a5dc
HB		 4746{
4747 int r;
6a8b1d13 4748 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
851ba692 4749 vcpu->run->request_interrupt_window;
b6c7a5dc 4750
2e53d63a
MT		 4751 if (vcpu->requests)
4752 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4753 kvm_mmu_unload(vcpu);
4754
b6c7a5dc
HB		 4755 r = kvm_mmu_reload(vcpu);
4756 if (unlikely(r))
4757 goto out;
4758
2f52d58c
AK		 4759 if (vcpu->requests) {
4760 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
2f599714 4761 __kvm_migrate_timers(vcpu);
c8076604
GH		 4762 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4763 kvm_write_guest_time(vcpu);
4731d4c7
MT		 4764 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4765 kvm_mmu_sync_roots(vcpu);
d4acf7e7
MT		 4766 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4767 kvm_x86_ops->tlb_flush(vcpu);
b93463aa
AK		 4768 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4769 &vcpu->requests)) {
851ba692 4770 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
b93463aa
AK		 4771 r = 0;
4772 goto out;
4773 }
71c4dfaf 4774 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
851ba692 4775 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
71c4dfaf
JR		 4776 r = 0;
4777 goto out;
4778 }
02daab21
AK		 4779 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4780 vcpu->fpu_active = 0;
4781 kvm_x86_ops->fpu_deactivate(vcpu);
4782 }
2f52d58c 4783 }
b93463aa 4784
b6c7a5dc
HB		 4785 preempt_disable();
4786
4787 kvm_x86_ops->prepare_guest_switch(vcpu);
2608d7a1
AK		 4788 if (vcpu->fpu_active)
4789 kvm_load_guest_fpu(vcpu);
2acf923e 4790 kvm_load_guest_xcr0(vcpu);
b6c7a5dc 4791
d94e1dc9
AK		 4792 atomic_set(&vcpu->guest_mode, 1);
4793 smp_wmb();
b6c7a5dc 4794
d94e1dc9 4795 local_irq_disable();
32f88400 4796
d94e1dc9
AK		 4797 if (!atomic_read(&vcpu->guest_mode) || vcpu->requests
4798 || need_resched() || signal_pending(current)) {
4799 atomic_set(&vcpu->guest_mode, 0);
4800 smp_wmb();
6c142801
AK		 4801 local_irq_enable();
4802 preempt_enable();
4803 r = 1;
4804 goto out;
4805 }
4806
851ba692 4807 inject_pending_event(vcpu);
b6c7a5dc 4808
6a8b1d13
GN		 4809 /* enable NMI/IRQ window open exits if needed */
4810 if (vcpu->arch.nmi_pending)
4811 kvm_x86_ops->enable_nmi_window(vcpu);
4812 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4813 kvm_x86_ops->enable_irq_window(vcpu);
4814
95ba8273 4815 if (kvm_lapic_enabled(vcpu)) {
8db3baa2
GN		 4816 update_cr8_intercept(vcpu);
4817 kvm_lapic_sync_to_vapic(vcpu);
95ba8273 4818 }
b93463aa 4819
f656ce01 4820 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
3200f405 4821
b6c7a5dc
HB		 4822 kvm_guest_enter();
4823
42dbaa5a 4824 if (unlikely(vcpu->arch.switch_db_regs)) {
42dbaa5a
JK		 4825 set_debugreg(0, 7);
4826 set_debugreg(vcpu->arch.eff_db[0], 0);
4827 set_debugreg(vcpu->arch.eff_db[1], 1);
4828 set_debugreg(vcpu->arch.eff_db[2], 2);
4829 set_debugreg(vcpu->arch.eff_db[3], 3);
4830 }
b6c7a5dc 4831
229456fc 4832 trace_kvm_entry(vcpu->vcpu_id);
851ba692 4833 kvm_x86_ops->run(vcpu);
b6c7a5dc 4834
24f1e32c
FW		 4835 /*
4836 * If the guest has used debug registers, at least dr7
4837 * will be disabled while returning to the host.
4838 * If we don't have active breakpoints in the host, we don't
4839 * care about the messed up debug address registers. But if
4840 * we have some of them active, restore the old state.
4841 */
59d8eb53 4842 if (hw_breakpoint_active())
24f1e32c 4843 hw_breakpoint_restore();
42dbaa5a 4844
d94e1dc9
AK		 4845 atomic_set(&vcpu->guest_mode, 0);
4846 smp_wmb();
b6c7a5dc
HB		 4847 local_irq_enable();
4848
4849 ++vcpu->stat.exits;
4850
4851 /*
4852 * We must have an instruction between local_irq_enable() and
4853 * kvm_guest_exit(), so the timer interrupt isn't delayed by
4854 * the interrupt shadow. The stat.exits increment will do nicely.
4855 * But we need to prevent reordering, hence this barrier():
4856 */
4857 barrier();
4858
4859 kvm_guest_exit();
4860
4861 preempt_enable();
4862
f656ce01 4863 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3200f405 4864
b6c7a5dc
HB		 4865 /*
4866 * Profile KVM exit RIPs:
4867 */
4868 if (unlikely(prof_on == KVM_PROFILING)) {
5fdbf976
MT		 4869 unsigned long rip = kvm_rip_read(vcpu);
4870 profile_hit(KVM_PROFILING, (void *)rip);
b6c7a5dc
HB		 4871 }
4872
298101da 4873
b93463aa
AK		 4874 kvm_lapic_sync_from_vapic(vcpu);
4875
851ba692 4876 r = kvm_x86_ops->handle_exit(vcpu);
d7690175
MT		 4877out:
4878 return r;
4879}
b6c7a5dc 4880
09cec754 4881
851ba692 4882static int __vcpu_run(struct kvm_vcpu *vcpu)
d7690175
MT		 4883{
4884 int r;
f656ce01 4885 struct kvm *kvm = vcpu->kvm;
d7690175
MT		 4886
4887 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
1b10bf31
JK		 4888 pr_debug("vcpu %d received sipi with vector # %x\n",
4889 vcpu->vcpu_id, vcpu->arch.sipi_vector);
d7690175 4890 kvm_lapic_reset(vcpu);
5f179287 4891 r = kvm_arch_vcpu_reset(vcpu);
d7690175
MT		 4892 if (r)
4893 return r;
4894 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
b6c7a5dc
HB		 4895 }
4896
f656ce01 4897 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175
MT		 4898 vapic_enter(vcpu);
4899
4900 r = 1;
4901 while (r > 0) {
af2152f5 4902 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
851ba692 4903 r = vcpu_enter_guest(vcpu);
d7690175 4904 else {
f656ce01 4905 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
d7690175 4906 kvm_vcpu_block(vcpu);
f656ce01 4907 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 4908 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
09cec754
GN		 4909 {
4910 switch(vcpu->arch.mp_state) {
4911 case KVM_MP_STATE_HALTED:
d7690175 4912 vcpu->arch.mp_state =
09cec754
GN		 4913 KVM_MP_STATE_RUNNABLE;
4914 case KVM_MP_STATE_RUNNABLE:
4915 break;
4916 case KVM_MP_STATE_SIPI_RECEIVED:
4917 default:
4918 r = -EINTR;
4919 break;
4920 }
4921 }
d7690175
MT		 4922 }
4923
09cec754
GN		 4924 if (r <= 0)
4925 break;
4926
4927 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4928 if (kvm_cpu_has_pending_timer(vcpu))
4929 kvm_inject_pending_timer_irqs(vcpu);
4930
851ba692 4931 if (dm_request_for_irq_injection(vcpu)) {
09cec754 4932 r = -EINTR;
851ba692 4933 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754
GN		 4934 ++vcpu->stat.request_irq_exits;
4935 }
4936 if (signal_pending(current)) {
4937 r = -EINTR;
851ba692 4938 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754
GN		 4939 ++vcpu->stat.signal_exits;
4940 }
4941 if (need_resched()) {
f656ce01 4942 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
09cec754 4943 kvm_resched(vcpu);
f656ce01 4944 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 4945 }
b6c7a5dc
HB		 4946 }
4947
f656ce01 4948 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
b6c7a5dc 4949
b93463aa
AK		 4950 vapic_exit(vcpu);
4951
b6c7a5dc
HB		 4952 return r;
4953}
4954
4955int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4956{
4957 int r;
4958 sigset_t sigsaved;
4959
ac9f6dc0
AK		 4960 if (vcpu->sigset_active)
4961 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4962
a4535290 4963 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
b6c7a5dc 4964 kvm_vcpu_block(vcpu);
d7690175 4965 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
ac9f6dc0
AK		 4966 r = -EAGAIN;
4967 goto out;
b6c7a5dc
HB		 4968 }
4969
b6c7a5dc
HB		 4970 /* re-sync apic's tpr */
4971 if (!irqchip_in_kernel(vcpu->kvm))
2d3ad1f4 4972 kvm_set_cr8(vcpu, kvm_run->cr8);
b6c7a5dc 4973
92bf9748
GN		 4974 if (vcpu->arch.pio.count || vcpu->mmio_needed ||
4975 vcpu->arch.emulate_ctxt.restart) {
4976 if (vcpu->mmio_needed) {
4977 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4978 vcpu->mmio_read_completed = 1;
4979 vcpu->mmio_needed = 0;
b6c7a5dc 4980 }
f656ce01 4981 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5cd21917 4982 r = emulate_instruction(vcpu, 0, 0, EMULTYPE_NO_DECODE);
f656ce01 4983 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6d77dbfc 4984 if (r != EMULATE_DONE) {
b6c7a5dc
HB		 4985 r = 0;
4986 goto out;
4987 }
4988 }
5fdbf976
MT		 4989 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4990 kvm_register_write(vcpu, VCPU_REGS_RAX,
4991 kvm_run->hypercall.ret);
b6c7a5dc 4992
851ba692 4993 r = __vcpu_run(vcpu);
b6c7a5dc
HB		 4994
4995out:
f1d86e46 4996 post_kvm_run_save(vcpu);
b6c7a5dc
HB		 4997 if (vcpu->sigset_active)
4998 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4999
b6c7a5dc
HB		 5000 return r;
5001}
5002
5003int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5004{
5fdbf976
MT		 5005 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5006 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5007 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5008 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5009 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5010 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5011 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5012 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
b6c7a5dc 5013#ifdef CONFIG_X86_64
5fdbf976
MT		 5014 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5015 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5016 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5017 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5018 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5019 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5020 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5021 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
b6c7a5dc
HB		 5022#endif
5023
5fdbf976 5024 regs->rip = kvm_rip_read(vcpu);
91586a3b 5025 regs->rflags = kvm_get_rflags(vcpu);
b6c7a5dc 5026
b6c7a5dc
HB		 5027 return 0;
5028}
5029
5030int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5031{
5fdbf976
MT		 5032 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5033 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5034 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5035 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5036 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5037 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5038 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5039 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
b6c7a5dc 5040#ifdef CONFIG_X86_64
5fdbf976
MT		 5041 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5042 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5043 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5044 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5045 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5046 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5047 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5048 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
b6c7a5dc
HB		 5049#endif
5050
5fdbf976 5051 kvm_rip_write(vcpu, regs->rip);
91586a3b 5052 kvm_set_rflags(vcpu, regs->rflags);
b6c7a5dc 5053
b4f14abd
JK		 5054 vcpu->arch.exception.pending = false;
5055
b6c7a5dc
HB		 5056 return 0;
5057}
5058
b6c7a5dc
HB		 5059void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5060{
5061 struct kvm_segment cs;
5062
3e6e0aab 5063 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
b6c7a5dc
HB		 5064 *db = cs.db;
5065 *l = cs.l;
5066}
5067EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5068
5069int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5070 struct kvm_sregs *sregs)
5071{
89a27f4d 5072 struct desc_ptr dt;
b6c7a5dc 5073
3e6e0aab
GT		 5074 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5075 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5076 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5077 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5078 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5079 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 5080
3e6e0aab
GT		 5081 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5082 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc
HB		 5083
5084 kvm_x86_ops->get_idt(vcpu, &dt);
89a27f4d
GN		 5085 sregs->idt.limit = dt.size;
5086 sregs->idt.base = dt.address;
b6c7a5dc 5087 kvm_x86_ops->get_gdt(vcpu, &dt);
89a27f4d
GN		 5088 sregs->gdt.limit = dt.size;
5089 sregs->gdt.base = dt.address;
b6c7a5dc 5090
4d4ec087 5091 sregs->cr0 = kvm_read_cr0(vcpu);
ad312c7c
ZX		 5092 sregs->cr2 = vcpu->arch.cr2;
5093 sregs->cr3 = vcpu->arch.cr3;
fc78f519 5094 sregs->cr4 = kvm_read_cr4(vcpu);
2d3ad1f4 5095 sregs->cr8 = kvm_get_cr8(vcpu);
f6801dff 5096 sregs->efer = vcpu->arch.efer;
b6c7a5dc
HB		 5097 sregs->apic_base = kvm_get_apic_base(vcpu);
5098
923c61bb 5099 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
b6c7a5dc 5100
36752c9b 5101 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
14d0bc1f
GN		 5102 set_bit(vcpu->arch.interrupt.nr,
5103 (unsigned long *)sregs->interrupt_bitmap);
16d7a191 5104
b6c7a5dc
HB		 5105 return 0;
5106}
5107
62d9f0db
MT		 5108int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5109 struct kvm_mp_state *mp_state)
5110{
62d9f0db 5111 mp_state->mp_state = vcpu->arch.mp_state;
62d9f0db
MT		 5112 return 0;
5113}
5114
5115int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5116 struct kvm_mp_state *mp_state)
5117{
62d9f0db 5118 vcpu->arch.mp_state = mp_state->mp_state;
62d9f0db
MT		 5119 return 0;
5120}
5121
e269fb21
JK		 5122int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5123 bool has_error_code, u32 error_code)
b6c7a5dc 5124{
4d2179e1 5125 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
ceffb459
GN		 5126 int cs_db, cs_l, ret;
5127 cache_all_regs(vcpu);
37817f29 5128
ceffb459 5129 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
e01c2426 5130
ceffb459
GN		 5131 vcpu->arch.emulate_ctxt.vcpu = vcpu;
5132 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
5133 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
5134 vcpu->arch.emulate_ctxt.mode =
5135 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5136 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
5137 ? X86EMUL_MODE_VM86 : cs_l
5138 ? X86EMUL_MODE_PROT64 : cs_db
5139 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4d2179e1
GN		 5140 memset(c, 0, sizeof(struct decode_cache));
5141 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
c697518a 5142
ceffb459 5143 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt, &emulate_ops,
e269fb21
JK		 5144 tss_selector, reason, has_error_code,
5145 error_code);
c697518a 5146
c697518a 5147 if (ret)
19d04437 5148 return EMULATE_FAIL;
37817f29 5149
4d2179e1 5150 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
95c55886 5151 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
19d04437
GN		 5152 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5153 return EMULATE_DONE;
37817f29
IE		 5154}
5155EXPORT_SYMBOL_GPL(kvm_task_switch);
5156
b6c7a5dc
HB		 5157int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5158 struct kvm_sregs *sregs)
5159{
5160 int mmu_reset_needed = 0;
923c61bb 5161 int pending_vec, max_bits;
89a27f4d 5162 struct desc_ptr dt;
b6c7a5dc 5163
89a27f4d
GN		 5164 dt.size = sregs->idt.limit;
5165 dt.address = sregs->idt.base;
b6c7a5dc 5166 kvm_x86_ops->set_idt(vcpu, &dt);
89a27f4d
GN		 5167 dt.size = sregs->gdt.limit;
5168 dt.address = sregs->gdt.base;
b6c7a5dc
HB		 5169 kvm_x86_ops->set_gdt(vcpu, &dt);
5170
ad312c7c
ZX		 5171 vcpu->arch.cr2 = sregs->cr2;
5172 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
dc7e795e 5173 vcpu->arch.cr3 = sregs->cr3;
b6c7a5dc 5174
2d3ad1f4 5175 kvm_set_cr8(vcpu, sregs->cr8);
b6c7a5dc 5176
f6801dff 5177 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
b6c7a5dc 5178 kvm_x86_ops->set_efer(vcpu, sregs->efer);
b6c7a5dc
HB		 5179 kvm_set_apic_base(vcpu, sregs->apic_base);
5180
4d4ec087 5181 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
b6c7a5dc 5182 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
d7306163 5183 vcpu->arch.cr0 = sregs->cr0;
b6c7a5dc 5184
fc78f519 5185 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
b6c7a5dc 5186 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7c93be44 5187 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
ad312c7c 5188 load_pdptrs(vcpu, vcpu->arch.cr3);
7c93be44
MT		 5189 mmu_reset_needed = 1;
5190 }
b6c7a5dc
HB		 5191
5192 if (mmu_reset_needed)
5193 kvm_mmu_reset_context(vcpu);
5194
923c61bb
GN		 5195 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5196 pending_vec = find_first_bit(
5197 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5198 if (pending_vec < max_bits) {
66fd3f7f 5199 kvm_queue_interrupt(vcpu, pending_vec, false);
923c61bb
GN		 5200 pr_debug("Set back pending irq %d\n", pending_vec);
5201 if (irqchip_in_kernel(vcpu->kvm))
5202 kvm_pic_clear_isr_ack(vcpu->kvm);
b6c7a5dc
HB		 5203 }
5204
3e6e0aab
GT		 5205 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5206 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5207 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5208 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5209 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5210 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 5211
3e6e0aab
GT		 5212 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5213 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc 5214
5f0269f5
ME		 5215 update_cr8_intercept(vcpu);
5216
9c3e4aab 5217 /* Older userspace won't unhalt the vcpu on reset. */
c5af89b6 5218 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9c3e4aab 5219 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3eeb3288 5220 !is_protmode(vcpu))
9c3e4aab
MT		 5221 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5222
b6c7a5dc
HB		 5223 return 0;
5224}
5225
d0bfb940
JK		 5226int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5227 struct kvm_guest_debug *dbg)
b6c7a5dc 5228{
355be0b9 5229 unsigned long rflags;
ae675ef0 5230 int i, r;
b6c7a5dc 5231
4f926bf2
JK		 5232 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5233 r = -EBUSY;
5234 if (vcpu->arch.exception.pending)
2122ff5e 5235 goto out;
4f926bf2
JK		 5236 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5237 kvm_queue_exception(vcpu, DB_VECTOR);
5238 else
5239 kvm_queue_exception(vcpu, BP_VECTOR);
5240 }
5241
91586a3b
JK		 5242 /*
5243 * Read rflags as long as potentially injected trace flags are still
5244 * filtered out.
5245 */
5246 rflags = kvm_get_rflags(vcpu);
355be0b9
JK		 5247
5248 vcpu->guest_debug = dbg->control;
5249 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5250 vcpu->guest_debug = 0;
5251
5252 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
ae675ef0
JK		 5253 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5254 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5255 vcpu->arch.switch_db_regs =
5256 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5257 } else {
5258 for (i = 0; i < KVM_NR_DB_REGS; i++)
5259 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5260 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5261 }
5262
f92653ee
JK		 5263 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5264 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5265 get_segment_base(vcpu, VCPU_SREG_CS);
94fe45da 5266
91586a3b
JK		 5267 /*
5268 * Trigger an rflags update that will inject or remove the trace
5269 * flags.
5270 */
5271 kvm_set_rflags(vcpu, rflags);
b6c7a5dc 5272
355be0b9 5273 kvm_x86_ops->set_guest_debug(vcpu, dbg);
b6c7a5dc 5274
4f926bf2 5275 r = 0;
d0bfb940 5276
2122ff5e 5277out:
b6c7a5dc
HB		 5278
5279 return r;
5280}
5281
8b006791
ZX		 5282/*
5283 * Translate a guest virtual address to a guest physical address.
5284 */
5285int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5286 struct kvm_translation *tr)
5287{
5288 unsigned long vaddr = tr->linear_address;
5289 gpa_t gpa;
f656ce01 5290 int idx;
8b006791 5291
f656ce01 5292 idx = srcu_read_lock(&vcpu->kvm->srcu);
1871c602 5293 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
f656ce01 5294 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8b006791
ZX		 5295 tr->physical_address = gpa;
5296 tr->valid = gpa != UNMAPPED_GVA;
5297 tr->writeable = 1;
5298 tr->usermode = 0;
8b006791
ZX		 5299
5300 return 0;
5301}
5302
d0752060
HB		 5303int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5304{
98918833
SY		 5305 struct i387_fxsave_struct *fxsave =
5306 &vcpu->arch.guest_fpu.state->fxsave;
d0752060 5307
d0752060
HB		 5308 memcpy(fpu->fpr, fxsave->st_space, 128);
5309 fpu->fcw = fxsave->cwd;
5310 fpu->fsw = fxsave->swd;
5311 fpu->ftwx = fxsave->twd;
5312 fpu->last_opcode = fxsave->fop;
5313 fpu->last_ip = fxsave->rip;
5314 fpu->last_dp = fxsave->rdp;
5315 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5316
d0752060
HB		 5317 return 0;
5318}
5319
5320int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5321{
98918833
SY		 5322 struct i387_fxsave_struct *fxsave =
5323 &vcpu->arch.guest_fpu.state->fxsave;
d0752060 5324
d0752060
HB		 5325 memcpy(fxsave->st_space, fpu->fpr, 128);
5326 fxsave->cwd = fpu->fcw;
5327 fxsave->swd = fpu->fsw;
5328 fxsave->twd = fpu->ftwx;
5329 fxsave->fop = fpu->last_opcode;
5330 fxsave->rip = fpu->last_ip;
5331 fxsave->rdp = fpu->last_dp;
5332 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5333
d0752060
HB		 5334 return 0;
5335}
5336
10ab25cd 5337int fx_init(struct kvm_vcpu *vcpu)
d0752060 5338{
10ab25cd
JK		 5339 int err;
5340
5341 err = fpu_alloc(&vcpu->arch.guest_fpu);
5342 if (err)
5343 return err;
5344
98918833 5345 fpu_finit(&vcpu->arch.guest_fpu);
d0752060 5346
2acf923e
DC		 5347 /*
5348 * Ensure guest xcr0 is valid for loading
5349 */
5350 vcpu->arch.xcr0 = XSTATE_FP;
5351
ad312c7c 5352 vcpu->arch.cr0 |= X86_CR0_ET;
10ab25cd
JK		 5353
5354 return 0;
d0752060
HB		 5355}
5356EXPORT_SYMBOL_GPL(fx_init);
5357
98918833
SY		 5358static void fx_free(struct kvm_vcpu *vcpu)
5359{
5360 fpu_free(&vcpu->arch.guest_fpu);
5361}
5362
d0752060
HB		 5363void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5364{
2608d7a1 5365 if (vcpu->guest_fpu_loaded)
d0752060
HB		 5366 return;
5367
2acf923e
DC		 5368 /*
5369 * Restore all possible states in the guest,
5370 * and assume host would use all available bits.
5371 * Guest xcr0 would be loaded later.
5372 */
5373 kvm_put_guest_xcr0(vcpu);
d0752060 5374 vcpu->guest_fpu_loaded = 1;
7cf30855 5375 unlazy_fpu(current);
98918833 5376 fpu_restore_checking(&vcpu->arch.guest_fpu);
0c04851c 5377 trace_kvm_fpu(1);
d0752060 5378}
d0752060
HB		 5379
5380void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5381{
2acf923e
DC		 5382 kvm_put_guest_xcr0(vcpu);
5383
d0752060
HB		 5384 if (!vcpu->guest_fpu_loaded)
5385 return;
5386
5387 vcpu->guest_fpu_loaded = 0;
98918833 5388 fpu_save_init(&vcpu->arch.guest_fpu);
f096ed85 5389 ++vcpu->stat.fpu_reload;
02daab21 5390 set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
0c04851c 5391 trace_kvm_fpu(0);
d0752060 5392}
e9b11c17
ZX		 5393
5394void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5395{
7f1ea208
JR		 5396 if (vcpu->arch.time_page) {
5397 kvm_release_page_dirty(vcpu->arch.time_page);
5398 vcpu->arch.time_page = NULL;
5399 }
5400
98918833 5401 fx_free(vcpu);
e9b11c17
ZX		 5402 kvm_x86_ops->vcpu_free(vcpu);
5403}
5404
5405struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5406 unsigned int id)
5407{
26e5215f
AK		 5408 return kvm_x86_ops->vcpu_create(kvm, id);
5409}
e9b11c17 5410
26e5215f
AK		 5411int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5412{
5413 int r;
e9b11c17 5414
0bed3b56 5415 vcpu->arch.mtrr_state.have_fixed = 1;
e9b11c17
ZX		 5416 vcpu_load(vcpu);
5417 r = kvm_arch_vcpu_reset(vcpu);
5418 if (r == 0)
5419 r = kvm_mmu_setup(vcpu);
5420 vcpu_put(vcpu);
5421 if (r < 0)
5422 goto free_vcpu;
5423
26e5215f 5424 return 0;
e9b11c17
ZX		 5425free_vcpu:
5426 kvm_x86_ops->vcpu_free(vcpu);
26e5215f 5427 return r;
e9b11c17
ZX		 5428}
5429
d40ccc62 5430void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
e9b11c17
ZX		 5431{
5432 vcpu_load(vcpu);
5433 kvm_mmu_unload(vcpu);
5434 vcpu_put(vcpu);
5435
98918833 5436 fx_free(vcpu);
e9b11c17
ZX		 5437 kvm_x86_ops->vcpu_free(vcpu);
5438}
5439
5440int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5441{
448fa4a9
JK		 5442 vcpu->arch.nmi_pending = false;
5443 vcpu->arch.nmi_injected = false;
5444
42dbaa5a
JK		 5445 vcpu->arch.switch_db_regs = 0;
5446 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5447 vcpu->arch.dr6 = DR6_FIXED_1;
5448 vcpu->arch.dr7 = DR7_FIXED_1;
5449
e9b11c17
ZX		 5450 return kvm_x86_ops->vcpu_reset(vcpu);
5451}
5452
10474ae8 5453int kvm_arch_hardware_enable(void *garbage)
e9b11c17 5454{
0cca7907
ZA		 5455 /*
5456 * Since this may be called from a hotplug notifcation,
5457 * we can't get the CPU frequency directly.
5458 */
5459 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5460 int cpu = raw_smp_processor_id();
5461 per_cpu(cpu_tsc_khz, cpu) = 0;
5462 }
18863bdd
AK		 5463
5464 kvm_shared_msr_cpu_online();
5465
10474ae8 5466 return kvm_x86_ops->hardware_enable(garbage);
e9b11c17
ZX		 5467}
5468
5469void kvm_arch_hardware_disable(void *garbage)
5470{
5471 kvm_x86_ops->hardware_disable(garbage);
3548bab5 5472 drop_user_return_notifiers(garbage);
e9b11c17
ZX		 5473}
5474
5475int kvm_arch_hardware_setup(void)
5476{
5477 return kvm_x86_ops->hardware_setup();
5478}
5479
5480void kvm_arch_hardware_unsetup(void)
5481{
5482 kvm_x86_ops->hardware_unsetup();
5483}
5484
5485void kvm_arch_check_processor_compat(void *rtn)
5486{
5487 kvm_x86_ops->check_processor_compatibility(rtn);
5488}
5489
5490int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5491{
5492 struct page *page;
5493 struct kvm *kvm;
5494 int r;
5495
5496 BUG_ON(vcpu->kvm == NULL);
5497 kvm = vcpu->kvm;
5498
ad312c7c 5499 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
c5af89b6 5500 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
a4535290 5501 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
e9b11c17 5502 else
a4535290 5503 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
e9b11c17
ZX		 5504
5505 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5506 if (!page) {
5507 r = -ENOMEM;
5508 goto fail;
5509 }
ad312c7c 5510 vcpu->arch.pio_data = page_address(page);
e9b11c17
ZX		 5511
5512 r = kvm_mmu_create(vcpu);
5513 if (r < 0)
5514 goto fail_free_pio_data;
5515
5516 if (irqchip_in_kernel(kvm)) {
5517 r = kvm_create_lapic(vcpu);
5518 if (r < 0)
5519 goto fail_mmu_destroy;
5520 }
5521
890ca9ae
HY		 5522 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5523 GFP_KERNEL);
5524 if (!vcpu->arch.mce_banks) {
5525 r = -ENOMEM;
443c39bc 5526 goto fail_free_lapic;
890ca9ae
HY		 5527 }
5528 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5529
e9b11c17 5530 return 0;
443c39bc
WY		 5531fail_free_lapic:
5532 kvm_free_lapic(vcpu);
e9b11c17
ZX		 5533fail_mmu_destroy:
5534 kvm_mmu_destroy(vcpu);
5535fail_free_pio_data:
ad312c7c 5536 free_page((unsigned long)vcpu->arch.pio_data);
e9b11c17
ZX		 5537fail:
5538 return r;
5539}
5540
5541void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5542{
f656ce01
MT		 5543 int idx;
5544
36cb93fd 5545 kfree(vcpu->arch.mce_banks);
e9b11c17 5546 kvm_free_lapic(vcpu);
f656ce01 5547 idx = srcu_read_lock(&vcpu->kvm->srcu);
e9b11c17 5548 kvm_mmu_destroy(vcpu);
f656ce01 5549 srcu_read_unlock(&vcpu->kvm->srcu, idx);
ad312c7c 5550 free_page((unsigned long)vcpu->arch.pio_data);
e9b11c17 5551}
d19a9cd2
ZX		 5552
5553struct kvm *kvm_arch_create_vm(void)
5554{
5555 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5556
5557 if (!kvm)
5558 return ERR_PTR(-ENOMEM);
5559
fef9cce0
MT		 5560 kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5561 if (!kvm->arch.aliases) {
5562 kfree(kvm);
5563 return ERR_PTR(-ENOMEM);
5564 }
5565
f05e70ac 5566 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4d5c5d0f 5567 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
d19a9cd2 5568
5550af4d
SY		 5569 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5570 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5571
53f658b3
MT		 5572 rdtscll(kvm->arch.vm_init_tsc);
5573
d19a9cd2
ZX		 5574 return kvm;
5575}
5576
5577static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5578{
5579 vcpu_load(vcpu);
5580 kvm_mmu_unload(vcpu);
5581 vcpu_put(vcpu);
5582}
5583
5584static void kvm_free_vcpus(struct kvm *kvm)
5585{
5586 unsigned int i;
988a2cae 5587 struct kvm_vcpu *vcpu;
d19a9cd2
ZX		 5588
5589 /*
5590 * Unpin any mmu pages first.
5591 */
988a2cae
GN		 5592 kvm_for_each_vcpu(i, vcpu, kvm)
5593 kvm_unload_vcpu_mmu(vcpu);
5594 kvm_for_each_vcpu(i, vcpu, kvm)
5595 kvm_arch_vcpu_free(vcpu);
5596
5597 mutex_lock(&kvm->lock);
5598 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5599 kvm->vcpus[i] = NULL;
d19a9cd2 5600
988a2cae
GN		 5601 atomic_set(&kvm->online_vcpus, 0);
5602 mutex_unlock(&kvm->lock);
d19a9cd2
ZX		 5603}
5604
ad8ba2cd
SY		 5605void kvm_arch_sync_events(struct kvm *kvm)
5606{
ba4cef31 5607 kvm_free_all_assigned_devices(kvm);
ad8ba2cd
SY		 5608}
5609
d19a9cd2
ZX		 5610void kvm_arch_destroy_vm(struct kvm *kvm)
5611{
6eb55818 5612 kvm_iommu_unmap_guest(kvm);
7837699f 5613 kvm_free_pit(kvm);
d7deeeb0
ZX		 5614 kfree(kvm->arch.vpic);
5615 kfree(kvm->arch.vioapic);
d19a9cd2
ZX		 5616 kvm_free_vcpus(kvm);
5617 kvm_free_physmem(kvm);
3d45830c
AK		 5618 if (kvm->arch.apic_access_page)
5619 put_page(kvm->arch.apic_access_page);
b7ebfb05
SY		 5620 if (kvm->arch.ept_identity_pagetable)
5621 put_page(kvm->arch.ept_identity_pagetable);
64749204 5622 cleanup_srcu_struct(&kvm->srcu);
fef9cce0 5623 kfree(kvm->arch.aliases);
d19a9cd2
ZX		 5624 kfree(kvm);
5625}
0de10343 5626
f7784b8e
MT		 5627int kvm_arch_prepare_memory_region(struct kvm *kvm,
5628 struct kvm_memory_slot *memslot,
0de10343 5629 struct kvm_memory_slot old,
f7784b8e 5630 struct kvm_userspace_memory_region *mem,
0de10343
ZX		 5631 int user_alloc)
5632{
f7784b8e 5633 int npages = memslot->npages;
0de10343
ZX		 5634
5635 /*To keep backward compatibility with older userspace,
5636 *x86 needs to hanlde !user_alloc case.
5637 */
5638 if (!user_alloc) {
5639 if (npages && !old.rmap) {
604b38ac
AA		 5640 unsigned long userspace_addr;
5641
72dc67a6 5642 down_write(&current->mm->mmap_sem);
604b38ac
AA		 5643 userspace_addr = do_mmap(NULL, 0,
5644 npages * PAGE_SIZE,
5645 PROT_READ | PROT_WRITE,
acee3c04 5646 MAP_PRIVATE | MAP_ANONYMOUS,
604b38ac 5647 0);
72dc67a6 5648 up_write(&current->mm->mmap_sem);
0de10343 5649
604b38ac
AA		 5650 if (IS_ERR((void *)userspace_addr))
5651 return PTR_ERR((void *)userspace_addr);
5652
604b38ac 5653 memslot->userspace_addr = userspace_addr;
0de10343
ZX		 5654 }
5655 }
5656
f7784b8e
MT		 5657
5658 return 0;
5659}
5660
5661void kvm_arch_commit_memory_region(struct kvm *kvm,
5662 struct kvm_userspace_memory_region *mem,
5663 struct kvm_memory_slot old,
5664 int user_alloc)
5665{
5666
5667 int npages = mem->memory_size >> PAGE_SHIFT;
5668
5669 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5670 int ret;
5671
5672 down_write(&current->mm->mmap_sem);
5673 ret = do_munmap(current->mm, old.userspace_addr,
5674 old.npages * PAGE_SIZE);
5675 up_write(&current->mm->mmap_sem);
5676 if (ret < 0)
5677 printk(KERN_WARNING
5678 "kvm_vm_ioctl_set_memory_region: "
5679 "failed to munmap memory\n");
5680 }
5681
7c8a83b7 5682 spin_lock(&kvm->mmu_lock);
f05e70ac 5683 if (!kvm->arch.n_requested_mmu_pages) {
0de10343
ZX		 5684 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5685 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5686 }
5687
5688 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7c8a83b7 5689 spin_unlock(&kvm->mmu_lock);
0de10343 5690}
1d737c8a 5691
34d4cb8f
MT		 5692void kvm_arch_flush_shadow(struct kvm *kvm)
5693{
5694 kvm_mmu_zap_all(kvm);
8986ecc0 5695 kvm_reload_remote_mmus(kvm);
34d4cb8f
MT		 5696}
5697
1d737c8a
ZX		 5698int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5699{
a4535290 5700 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
a1b37100
GN		 5701 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5702 || vcpu->arch.nmi_pending ||
5703 (kvm_arch_interrupt_allowed(vcpu) &&
5704 kvm_cpu_has_interrupt(vcpu));
1d737c8a 5705}
5736199a 5706
5736199a
ZX		 5707void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5708{
32f88400
MT		 5709 int me;
5710 int cpu = vcpu->cpu;
5736199a
ZX		 5711
5712 if (waitqueue_active(&vcpu->wq)) {
5713 wake_up_interruptible(&vcpu->wq);
5714 ++vcpu->stat.halt_wakeup;
5715 }
32f88400
MT		 5716
5717 me = get_cpu();
5718 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
d94e1dc9 5719 if (atomic_xchg(&vcpu->guest_mode, 0))
32f88400 5720 smp_send_reschedule(cpu);
e9571ed5 5721 put_cpu();
5736199a 5722}
78646121
GN		 5723
5724int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5725{
5726 return kvm_x86_ops->interrupt_allowed(vcpu);
5727}
229456fc 5728
f92653ee
JK		 5729bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
5730{
5731 unsigned long current_rip = kvm_rip_read(vcpu) +
5732 get_segment_base(vcpu, VCPU_SREG_CS);
5733
5734 return current_rip == linear_rip;
5735}
5736EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
5737
94fe45da
JK		 5738unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5739{
5740 unsigned long rflags;
5741
5742 rflags = kvm_x86_ops->get_rflags(vcpu);
5743 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
c310bac5 5744 rflags &= ~X86_EFLAGS_TF;
94fe45da
JK		 5745 return rflags;
5746}
5747EXPORT_SYMBOL_GPL(kvm_get_rflags);
5748
5749void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5750{
5751 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
f92653ee 5752 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
c310bac5 5753 rflags |= X86_EFLAGS_TF;
94fe45da
JK		 5754 kvm_x86_ops->set_rflags(vcpu, rflags);
5755}
5756EXPORT_SYMBOL_GPL(kvm_set_rflags);
5757
229456fc
MT		 5758EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5759EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5760EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5761EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5762EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
0ac406de 5763EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
d8cabddf 5764EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
17897f36 5765EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
236649de 5766EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
ec1ff790 5767EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
532a46b9 5768EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
2e554e8d 5769EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
Clone of davem's net-next-2.6 tree