projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
KVM: Add last_host_tsc tracking back to KVM
[linux-2.6-block.git] / arch / x86 / kvm / x86.c
... / ...
CommitLineData
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.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10 *
11 * Authors:
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
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
22#include <linux/kvm_host.h>
23#include "irq.h"
24#include "mmu.h"
25#include "i8254.h"
26#include "tss.h"
27#include "kvm_cache_regs.h"
28#include "x86.h"
29#include "cpuid.h"
30
31#include <linux/clocksource.h>
32#include <linux/interrupt.h>
33#include <linux/kvm.h>
34#include <linux/fs.h>
35#include <linux/vmalloc.h>
36#include <linux/module.h>
37#include <linux/mman.h>
38#include <linux/highmem.h>
39#include <linux/iommu.h>
40#include <linux/intel-iommu.h>
41#include <linux/cpufreq.h>
42#include <linux/user-return-notifier.h>
43#include <linux/srcu.h>
44#include <linux/slab.h>
45#include <linux/perf_event.h>
46#include <linux/uaccess.h>
47#include <linux/hash.h>
48#include <linux/pci.h>
49#include <trace/events/kvm.h>
50
51#define CREATE_TRACE_POINTS
52#include "trace.h"
53
54#include <asm/debugreg.h>
55#include <asm/msr.h>
56#include <asm/desc.h>
57#include <asm/mtrr.h>
58#include <asm/mce.h>
59#include <asm/i387.h>
60#include <asm/xcr.h>
61#include <asm/pvclock.h>
62#include <asm/div64.h>
63
64#define MAX_IO_MSRS 256
65#define KVM_MAX_MCE_BANKS 32
66#define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
67
68#define emul_to_vcpu(ctxt) \
69 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
70
71/* EFER defaults:
72 * - enable syscall per default because its emulated by KVM
73 * - enable LME and LMA per default on 64 bit KVM
74 */
75#ifdef CONFIG_X86_64
76static
77u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
78#else
79static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
80#endif
81
82#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
83#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
84
85static void update_cr8_intercept(struct kvm_vcpu *vcpu);
86static void process_nmi(struct kvm_vcpu *vcpu);
87
88struct kvm_x86_ops *kvm_x86_ops;
89EXPORT_SYMBOL_GPL(kvm_x86_ops);
90
91static bool ignore_msrs = 0;
92module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
93
94bool kvm_has_tsc_control;
95EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
96u32 kvm_max_guest_tsc_khz;
97EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
98
99/* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
100static u32 tsc_tolerance_ppm = 250;
101module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
102
103#define KVM_NR_SHARED_MSRS 16
104
105struct kvm_shared_msrs_global {
106 int nr;
107 u32 msrs[KVM_NR_SHARED_MSRS];
108};
109
110struct kvm_shared_msrs {
111 struct user_return_notifier urn;
112 bool registered;
113 struct kvm_shared_msr_values {
114 u64 host;
115 u64 curr;
116 } values[KVM_NR_SHARED_MSRS];
117};
118
119static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
120static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
121
122struct kvm_stats_debugfs_item debugfs_entries[] = {
123 { "pf_fixed", VCPU_STAT(pf_fixed) },
124 { "pf_guest", VCPU_STAT(pf_guest) },
125 { "tlb_flush", VCPU_STAT(tlb_flush) },
126 { "invlpg", VCPU_STAT(invlpg) },
127 { "exits", VCPU_STAT(exits) },
128 { "io_exits", VCPU_STAT(io_exits) },
129 { "mmio_exits", VCPU_STAT(mmio_exits) },
130 { "signal_exits", VCPU_STAT(signal_exits) },
131 { "irq_window", VCPU_STAT(irq_window_exits) },
132 { "nmi_window", VCPU_STAT(nmi_window_exits) },
133 { "halt_exits", VCPU_STAT(halt_exits) },
134 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
135 { "hypercalls", VCPU_STAT(hypercalls) },
136 { "request_irq", VCPU_STAT(request_irq_exits) },
137 { "irq_exits", VCPU_STAT(irq_exits) },
138 { "host_state_reload", VCPU_STAT(host_state_reload) },
139 { "efer_reload", VCPU_STAT(efer_reload) },
140 { "fpu_reload", VCPU_STAT(fpu_reload) },
141 { "insn_emulation", VCPU_STAT(insn_emulation) },
142 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
143 { "irq_injections", VCPU_STAT(irq_injections) },
144 { "nmi_injections", VCPU_STAT(nmi_injections) },
145 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
146 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
147 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
148 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
149 { "mmu_flooded", VM_STAT(mmu_flooded) },
150 { "mmu_recycled", VM_STAT(mmu_recycled) },
151 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
152 { "mmu_unsync", VM_STAT(mmu_unsync) },
153 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
154 { "largepages", VM_STAT(lpages) },
155 { NULL }
156};
157
158u64 __read_mostly host_xcr0;
159
160int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
161
162static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
163{
164 int i;
165 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
166 vcpu->arch.apf.gfns[i] = ~0;
167}
168
169static void kvm_on_user_return(struct user_return_notifier *urn)
170{
171 unsigned slot;
172 struct kvm_shared_msrs *locals
173 = container_of(urn, struct kvm_shared_msrs, urn);
174 struct kvm_shared_msr_values *values;
175
176 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
177 values = &locals->values[slot];
178 if (values->host != values->curr) {
179 wrmsrl(shared_msrs_global.msrs[slot], values->host);
180 values->curr = values->host;
181 }
182 }
183 locals->registered = false;
184 user_return_notifier_unregister(urn);
185}
186
187static void shared_msr_update(unsigned slot, u32 msr)
188{
189 struct kvm_shared_msrs *smsr;
190 u64 value;
191
192 smsr = &__get_cpu_var(shared_msrs);
193 /* only read, and nobody should modify it at this time,
194 * so don't need lock */
195 if (slot >= shared_msrs_global.nr) {
196 printk(KERN_ERR "kvm: invalid MSR slot!");
197 return;
198 }
199 rdmsrl_safe(msr, &value);
200 smsr->values[slot].host = value;
201 smsr->values[slot].curr = value;
202}
203
204void kvm_define_shared_msr(unsigned slot, u32 msr)
205{
206 if (slot >= shared_msrs_global.nr)
207 shared_msrs_global.nr = slot + 1;
208 shared_msrs_global.msrs[slot] = msr;
209 /* we need ensured the shared_msr_global have been updated */
210 smp_wmb();
211}
212EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
213
214static void kvm_shared_msr_cpu_online(void)
215{
216 unsigned i;
217
218 for (i = 0; i < shared_msrs_global.nr; ++i)
219 shared_msr_update(i, shared_msrs_global.msrs[i]);
220}
221
222void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
223{
224 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
225
226 if (((value ^ smsr->values[slot].curr) & mask) == 0)
227 return;
228 smsr->values[slot].curr = value;
229 wrmsrl(shared_msrs_global.msrs[slot], value);
230 if (!smsr->registered) {
231 smsr->urn.on_user_return = kvm_on_user_return;
232 user_return_notifier_register(&smsr->urn);
233 smsr->registered = true;
234 }
235}
236EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
237
238static void drop_user_return_notifiers(void *ignore)
239{
240 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
241
242 if (smsr->registered)
243 kvm_on_user_return(&smsr->urn);
244}
245
246u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
247{
248 if (irqchip_in_kernel(vcpu->kvm))
249 return vcpu->arch.apic_base;
250 else
251 return vcpu->arch.apic_base;
252}
253EXPORT_SYMBOL_GPL(kvm_get_apic_base);
254
255void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
256{
257 /* TODO: reserve bits check */
258 if (irqchip_in_kernel(vcpu->kvm))
259 kvm_lapic_set_base(vcpu, data);
260 else
261 vcpu->arch.apic_base = data;
262}
263EXPORT_SYMBOL_GPL(kvm_set_apic_base);
264
265#define EXCPT_BENIGN 0
266#define EXCPT_CONTRIBUTORY 1
267#define EXCPT_PF 2
268
269static int exception_class(int vector)
270{
271 switch (vector) {
272 case PF_VECTOR:
273 return EXCPT_PF;
274 case DE_VECTOR:
275 case TS_VECTOR:
276 case NP_VECTOR:
277 case SS_VECTOR:
278 case GP_VECTOR:
279 return EXCPT_CONTRIBUTORY;
280 default:
281 break;
282 }
283 return EXCPT_BENIGN;
284}
285
286static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
287 unsigned nr, bool has_error, u32 error_code,
288 bool reinject)
289{
290 u32 prev_nr;
291 int class1, class2;
292
293 kvm_make_request(KVM_REQ_EVENT, vcpu);
294
295 if (!vcpu->arch.exception.pending) {
296 queue:
297 vcpu->arch.exception.pending = true;
298 vcpu->arch.exception.has_error_code = has_error;
299 vcpu->arch.exception.nr = nr;
300 vcpu->arch.exception.error_code = error_code;
301 vcpu->arch.exception.reinject = reinject;
302 return;
303 }
304
305 /* to check exception */
306 prev_nr = vcpu->arch.exception.nr;
307 if (prev_nr == DF_VECTOR) {
308 /* triple fault -> shutdown */
309 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
310 return;
311 }
312 class1 = exception_class(prev_nr);
313 class2 = exception_class(nr);
314 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
315 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
316 /* generate double fault per SDM Table 5-5 */
317 vcpu->arch.exception.pending = true;
318 vcpu->arch.exception.has_error_code = true;
319 vcpu->arch.exception.nr = DF_VECTOR;
320 vcpu->arch.exception.error_code = 0;
321 } else
322 /* replace previous exception with a new one in a hope
323 that instruction re-execution will regenerate lost
324 exception */
325 goto queue;
326}
327
328void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
329{
330 kvm_multiple_exception(vcpu, nr, false, 0, false);
331}
332EXPORT_SYMBOL_GPL(kvm_queue_exception);
333
334void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
335{
336 kvm_multiple_exception(vcpu, nr, false, 0, true);
337}
338EXPORT_SYMBOL_GPL(kvm_requeue_exception);
339
340void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
341{
342 if (err)
343 kvm_inject_gp(vcpu, 0);
344 else
345 kvm_x86_ops->skip_emulated_instruction(vcpu);
346}
347EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
348
349void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
350{
351 ++vcpu->stat.pf_guest;
352 vcpu->arch.cr2 = fault->address;
353 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
354}
355EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
356
357void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
358{
359 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
360 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
361 else
362 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
363}
364
365void kvm_inject_nmi(struct kvm_vcpu *vcpu)
366{
367 atomic_inc(&vcpu->arch.nmi_queued);
368 kvm_make_request(KVM_REQ_NMI, vcpu);
369}
370EXPORT_SYMBOL_GPL(kvm_inject_nmi);
371
372void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
373{
374 kvm_multiple_exception(vcpu, nr, true, error_code, false);
375}
376EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
377
378void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
379{
380 kvm_multiple_exception(vcpu, nr, true, error_code, true);
381}
382EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
383
384/*
385 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
386 * a #GP and return false.
387 */
388bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
389{
390 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
391 return true;
392 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
393 return false;
394}
395EXPORT_SYMBOL_GPL(kvm_require_cpl);
396
397/*
398 * This function will be used to read from the physical memory of the currently
399 * running guest. The difference to kvm_read_guest_page is that this function
400 * can read from guest physical or from the guest's guest physical memory.
401 */
402int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
403 gfn_t ngfn, void *data, int offset, int len,
404 u32 access)
405{
406 gfn_t real_gfn;
407 gpa_t ngpa;
408
409 ngpa = gfn_to_gpa(ngfn);
410 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
411 if (real_gfn == UNMAPPED_GVA)
412 return -EFAULT;
413
414 real_gfn = gpa_to_gfn(real_gfn);
415
416 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
417}
418EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
419
420int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
421 void *data, int offset, int len, u32 access)
422{
423 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
424 data, offset, len, access);
425}
426
427/*
428 * Load the pae pdptrs. Return true is they are all valid.
429 */
430int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
431{
432 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
433 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
434 int i;
435 int ret;
436 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
437
438 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
439 offset * sizeof(u64), sizeof(pdpte),
440 PFERR_USER_MASK|PFERR_WRITE_MASK);
441 if (ret < 0) {
442 ret = 0;
443 goto out;
444 }
445 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
446 if (is_present_gpte(pdpte[i]) &&
447 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
448 ret = 0;
449 goto out;
450 }
451 }
452 ret = 1;
453
454 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
455 __set_bit(VCPU_EXREG_PDPTR,
456 (unsigned long *)&vcpu->arch.regs_avail);
457 __set_bit(VCPU_EXREG_PDPTR,
458 (unsigned long *)&vcpu->arch.regs_dirty);
459out:
460
461 return ret;
462}
463EXPORT_SYMBOL_GPL(load_pdptrs);
464
465static bool pdptrs_changed(struct kvm_vcpu *vcpu)
466{
467 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
468 bool changed = true;
469 int offset;
470 gfn_t gfn;
471 int r;
472
473 if (is_long_mode(vcpu) || !is_pae(vcpu))
474 return false;
475
476 if (!test_bit(VCPU_EXREG_PDPTR,
477 (unsigned long *)&vcpu->arch.regs_avail))
478 return true;
479
480 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
481 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
482 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
483 PFERR_USER_MASK | PFERR_WRITE_MASK);
484 if (r < 0)
485 goto out;
486 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
487out:
488
489 return changed;
490}
491
492int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
493{
494 unsigned long old_cr0 = kvm_read_cr0(vcpu);
495 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
496 X86_CR0_CD | X86_CR0_NW;
497
498 cr0 |= X86_CR0_ET;
499
500#ifdef CONFIG_X86_64
501 if (cr0 & 0xffffffff00000000UL)
502 return 1;
503#endif
504
505 cr0 &= ~CR0_RESERVED_BITS;
506
507 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
508 return 1;
509
510 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
511 return 1;
512
513 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
514#ifdef CONFIG_X86_64
515 if ((vcpu->arch.efer & EFER_LME)) {
516 int cs_db, cs_l;
517
518 if (!is_pae(vcpu))
519 return 1;
520 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
521 if (cs_l)
522 return 1;
523 } else
524#endif
525 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
526 kvm_read_cr3(vcpu)))
527 return 1;
528 }
529
530 kvm_x86_ops->set_cr0(vcpu, cr0);
531
532 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
533 kvm_clear_async_pf_completion_queue(vcpu);
534 kvm_async_pf_hash_reset(vcpu);
535 }
536
537 if ((cr0 ^ old_cr0) & update_bits)
538 kvm_mmu_reset_context(vcpu);
539 return 0;
540}
541EXPORT_SYMBOL_GPL(kvm_set_cr0);
542
543void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
544{
545 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
546}
547EXPORT_SYMBOL_GPL(kvm_lmsw);
548
549int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
550{
551 u64 xcr0;
552
553 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
554 if (index != XCR_XFEATURE_ENABLED_MASK)
555 return 1;
556 xcr0 = xcr;
557 if (kvm_x86_ops->get_cpl(vcpu) != 0)
558 return 1;
559 if (!(xcr0 & XSTATE_FP))
560 return 1;
561 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
562 return 1;
563 if (xcr0 & ~host_xcr0)
564 return 1;
565 vcpu->arch.xcr0 = xcr0;
566 vcpu->guest_xcr0_loaded = 0;
567 return 0;
568}
569
570int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
571{
572 if (__kvm_set_xcr(vcpu, index, xcr)) {
573 kvm_inject_gp(vcpu, 0);
574 return 1;
575 }
576 return 0;
577}
578EXPORT_SYMBOL_GPL(kvm_set_xcr);
579
580int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
581{
582 unsigned long old_cr4 = kvm_read_cr4(vcpu);
583 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
584 X86_CR4_PAE | X86_CR4_SMEP;
585 if (cr4 & CR4_RESERVED_BITS)
586 return 1;
587
588 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
589 return 1;
590
591 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
592 return 1;
593
594 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
595 return 1;
596
597 if (is_long_mode(vcpu)) {
598 if (!(cr4 & X86_CR4_PAE))
599 return 1;
600 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
601 && ((cr4 ^ old_cr4) & pdptr_bits)
602 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
603 kvm_read_cr3(vcpu)))
604 return 1;
605
606 if (kvm_x86_ops->set_cr4(vcpu, cr4))
607 return 1;
608
609 if ((cr4 ^ old_cr4) & pdptr_bits)
610 kvm_mmu_reset_context(vcpu);
611
612 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
613 kvm_update_cpuid(vcpu);
614
615 return 0;
616}
617EXPORT_SYMBOL_GPL(kvm_set_cr4);
618
619int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
620{
621 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
622 kvm_mmu_sync_roots(vcpu);
623 kvm_mmu_flush_tlb(vcpu);
624 return 0;
625 }
626
627 if (is_long_mode(vcpu)) {
628 if (cr3 & CR3_L_MODE_RESERVED_BITS)
629 return 1;
630 } else {
631 if (is_pae(vcpu)) {
632 if (cr3 & CR3_PAE_RESERVED_BITS)
633 return 1;
634 if (is_paging(vcpu) &&
635 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
636 return 1;
637 }
638 /*
639 * We don't check reserved bits in nonpae mode, because
640 * this isn't enforced, and VMware depends on this.
641 */
642 }
643
644 /*
645 * Does the new cr3 value map to physical memory? (Note, we
646 * catch an invalid cr3 even in real-mode, because it would
647 * cause trouble later on when we turn on paging anyway.)
648 *
649 * A real CPU would silently accept an invalid cr3 and would
650 * attempt to use it - with largely undefined (and often hard
651 * to debug) behavior on the guest side.
652 */
653 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
654 return 1;
655 vcpu->arch.cr3 = cr3;
656 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
657 vcpu->arch.mmu.new_cr3(vcpu);
658 return 0;
659}
660EXPORT_SYMBOL_GPL(kvm_set_cr3);
661
662int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
663{
664 if (cr8 & CR8_RESERVED_BITS)
665 return 1;
666 if (irqchip_in_kernel(vcpu->kvm))
667 kvm_lapic_set_tpr(vcpu, cr8);
668 else
669 vcpu->arch.cr8 = cr8;
670 return 0;
671}
672EXPORT_SYMBOL_GPL(kvm_set_cr8);
673
674unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
675{
676 if (irqchip_in_kernel(vcpu->kvm))
677 return kvm_lapic_get_cr8(vcpu);
678 else
679 return vcpu->arch.cr8;
680}
681EXPORT_SYMBOL_GPL(kvm_get_cr8);
682
683static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
684{
685 switch (dr) {
686 case 0 ... 3:
687 vcpu->arch.db[dr] = val;
688 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
689 vcpu->arch.eff_db[dr] = val;
690 break;
691 case 4:
692 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
693 return 1; /* #UD */
694 /* fall through */
695 case 6:
696 if (val & 0xffffffff00000000ULL)
697 return -1; /* #GP */
698 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
699 break;
700 case 5:
701 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
702 return 1; /* #UD */
703 /* fall through */
704 default: /* 7 */
705 if (val & 0xffffffff00000000ULL)
706 return -1; /* #GP */
707 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
708 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
709 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
710 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
711 }
712 break;
713 }
714
715 return 0;
716}
717
718int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
719{
720 int res;
721
722 res = __kvm_set_dr(vcpu, dr, val);
723 if (res > 0)
724 kvm_queue_exception(vcpu, UD_VECTOR);
725 else if (res < 0)
726 kvm_inject_gp(vcpu, 0);
727
728 return res;
729}
730EXPORT_SYMBOL_GPL(kvm_set_dr);
731
732static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
733{
734 switch (dr) {
735 case 0 ... 3:
736 *val = vcpu->arch.db[dr];
737 break;
738 case 4:
739 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
740 return 1;
741 /* fall through */
742 case 6:
743 *val = vcpu->arch.dr6;
744 break;
745 case 5:
746 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
747 return 1;
748 /* fall through */
749 default: /* 7 */
750 *val = vcpu->arch.dr7;
751 break;
752 }
753
754 return 0;
755}
756
757int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
758{
759 if (_kvm_get_dr(vcpu, dr, val)) {
760 kvm_queue_exception(vcpu, UD_VECTOR);
761 return 1;
762 }
763 return 0;
764}
765EXPORT_SYMBOL_GPL(kvm_get_dr);
766
767bool kvm_rdpmc(struct kvm_vcpu *vcpu)
768{
769 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
770 u64 data;
771 int err;
772
773 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
774 if (err)
775 return err;
776 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
777 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
778 return err;
779}
780EXPORT_SYMBOL_GPL(kvm_rdpmc);
781
782/*
783 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
784 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
785 *
786 * This list is modified at module load time to reflect the
787 * capabilities of the host cpu. This capabilities test skips MSRs that are
788 * kvm-specific. Those are put in the beginning of the list.
789 */
790
791#define KVM_SAVE_MSRS_BEGIN 9
792static u32 msrs_to_save[] = {
793 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
794 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
795 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
796 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
797 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
798 MSR_STAR,
799#ifdef CONFIG_X86_64
800 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
801#endif
802 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
803};
804
805static unsigned num_msrs_to_save;
806
807static u32 emulated_msrs[] = {
808 MSR_IA32_TSCDEADLINE,
809 MSR_IA32_MISC_ENABLE,
810 MSR_IA32_MCG_STATUS,
811 MSR_IA32_MCG_CTL,
812};
813
814static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
815{
816 u64 old_efer = vcpu->arch.efer;
817
818 if (efer & efer_reserved_bits)
819 return 1;
820
821 if (is_paging(vcpu)
822 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
823 return 1;
824
825 if (efer & EFER_FFXSR) {
826 struct kvm_cpuid_entry2 *feat;
827
828 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
829 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
830 return 1;
831 }
832
833 if (efer & EFER_SVME) {
834 struct kvm_cpuid_entry2 *feat;
835
836 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
837 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
838 return 1;
839 }
840
841 efer &= ~EFER_LMA;
842 efer |= vcpu->arch.efer & EFER_LMA;
843
844 kvm_x86_ops->set_efer(vcpu, efer);
845
846 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
847
848 /* Update reserved bits */
849 if ((efer ^ old_efer) & EFER_NX)
850 kvm_mmu_reset_context(vcpu);
851
852 return 0;
853}
854
855void kvm_enable_efer_bits(u64 mask)
856{
857 efer_reserved_bits &= ~mask;
858}
859EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
860
861
862/*
863 * Writes msr value into into the appropriate "register".
864 * Returns 0 on success, non-0 otherwise.
865 * Assumes vcpu_load() was already called.
866 */
867int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
868{
869 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
870}
871
872/*
873 * Adapt set_msr() to msr_io()'s calling convention
874 */
875static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
876{
877 return kvm_set_msr(vcpu, index, *data);
878}
879
880static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
881{
882 int version;
883 int r;
884 struct pvclock_wall_clock wc;
885 struct timespec boot;
886
887 if (!wall_clock)
888 return;
889
890 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
891 if (r)
892 return;
893
894 if (version & 1)
895 ++version; /* first time write, random junk */
896
897 ++version;
898
899 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
900
901 /*
902 * The guest calculates current wall clock time by adding
903 * system time (updated by kvm_guest_time_update below) to the
904 * wall clock specified here. guest system time equals host
905 * system time for us, thus we must fill in host boot time here.
906 */
907 getboottime(&boot);
908
909 wc.sec = boot.tv_sec;
910 wc.nsec = boot.tv_nsec;
911 wc.version = version;
912
913 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
914
915 version++;
916 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
917}
918
919static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
920{
921 uint32_t quotient, remainder;
922
923 /* Don't try to replace with do_div(), this one calculates
924 * "(dividend << 32) / divisor" */
925 __asm__ ( "divl %4"
926 : "=a" (quotient), "=d" (remainder)
927 : "0" (0), "1" (dividend), "r" (divisor) );
928 return quotient;
929}
930
931static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
932 s8 *pshift, u32 *pmultiplier)
933{
934 uint64_t scaled64;
935 int32_t shift = 0;
936 uint64_t tps64;
937 uint32_t tps32;
938
939 tps64 = base_khz * 1000LL;
940 scaled64 = scaled_khz * 1000LL;
941 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
942 tps64 >>= 1;
943 shift--;
944 }
945
946 tps32 = (uint32_t)tps64;
947 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
948 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
949 scaled64 >>= 1;
950 else
951 tps32 <<= 1;
952 shift++;
953 }
954
955 *pshift = shift;
956 *pmultiplier = div_frac(scaled64, tps32);
957
958 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
959 __func__, base_khz, scaled_khz, shift, *pmultiplier);
960}
961
962static inline u64 get_kernel_ns(void)
963{
964 struct timespec ts;
965
966 WARN_ON(preemptible());
967 ktime_get_ts(&ts);
968 monotonic_to_bootbased(&ts);
969 return timespec_to_ns(&ts);
970}
971
972static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
973unsigned long max_tsc_khz;
974
975static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
976{
977 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
978 vcpu->arch.virtual_tsc_shift);
979}
980
981static u32 adjust_tsc_khz(u32 khz, s32 ppm)
982{
983 u64 v = (u64)khz * (1000000 + ppm);
984 do_div(v, 1000000);
985 return v;
986}
987
988static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
989{
990 u32 thresh_lo, thresh_hi;
991 int use_scaling = 0;
992
993 /* Compute a scale to convert nanoseconds in TSC cycles */
994 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
995 &vcpu->arch.virtual_tsc_shift,
996 &vcpu->arch.virtual_tsc_mult);
997 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
998
999 /*
1000 * Compute the variation in TSC rate which is acceptable
1001 * within the range of tolerance and decide if the
1002 * rate being applied is within that bounds of the hardware
1003 * rate. If so, no scaling or compensation need be done.
1004 */
1005 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1006 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1007 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1008 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1009 use_scaling = 1;
1010 }
1011 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1012}
1013
1014static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1015{
1016 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1017 vcpu->arch.virtual_tsc_mult,
1018 vcpu->arch.virtual_tsc_shift);
1019 tsc += vcpu->arch.last_tsc_write;
1020 return tsc;
1021}
1022
1023void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1024{
1025 struct kvm *kvm = vcpu->kvm;
1026 u64 offset, ns, elapsed;
1027 unsigned long flags;
1028 s64 nsdiff;
1029
1030 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1031 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1032 ns = get_kernel_ns();
1033 elapsed = ns - kvm->arch.last_tsc_nsec;
1034
1035 /* n.b - signed multiplication and division required */
1036 nsdiff = data - kvm->arch.last_tsc_write;
1037#ifdef CONFIG_X86_64
1038 nsdiff = (nsdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1039#else
1040 /* do_div() only does unsigned */
1041 asm("idivl %2; xor %%edx, %%edx"
1042 : "=A"(nsdiff)
1043 : "A"(nsdiff * 1000), "rm"(vcpu->arch.virtual_tsc_khz));
1044#endif
1045 nsdiff -= elapsed;
1046 if (nsdiff < 0)
1047 nsdiff = -nsdiff;
1048
1049 /*
1050 * Special case: TSC write with a small delta (1 second) of virtual
1051 * cycle time against real time is interpreted as an attempt to
1052 * synchronize the CPU.
1053 *
1054 * For a reliable TSC, we can match TSC offsets, and for an unstable
1055 * TSC, we add elapsed time in this computation. We could let the
1056 * compensation code attempt to catch up if we fall behind, but
1057 * it's better to try to match offsets from the beginning.
1058 */
1059 if (nsdiff < NSEC_PER_SEC &&
1060 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1061 if (!check_tsc_unstable()) {
1062 offset = kvm->arch.last_tsc_offset;
1063 pr_debug("kvm: matched tsc offset for %llu\n", data);
1064 } else {
1065 u64 delta = nsec_to_cycles(vcpu, elapsed);
1066 data += delta;
1067 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1068 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1069 }
1070 }
1071 kvm->arch.last_tsc_nsec = ns;
1072 kvm->arch.last_tsc_write = data;
1073 kvm->arch.last_tsc_offset = offset;
1074 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1075 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1076 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1077
1078 /* Reset of TSC must disable overshoot protection below */
1079 vcpu->arch.hv_clock.tsc_timestamp = 0;
1080 vcpu->arch.last_tsc_write = data;
1081 vcpu->arch.last_tsc_nsec = ns;
1082 vcpu->arch.last_guest_tsc = data;
1083}
1084EXPORT_SYMBOL_GPL(kvm_write_tsc);
1085
1086static int kvm_guest_time_update(struct kvm_vcpu *v)
1087{
1088 unsigned long flags;
1089 struct kvm_vcpu_arch *vcpu = &v->arch;
1090 void *shared_kaddr;
1091 unsigned long this_tsc_khz;
1092 s64 kernel_ns, max_kernel_ns;
1093 u64 tsc_timestamp;
1094
1095 /* Keep irq disabled to prevent changes to the clock */
1096 local_irq_save(flags);
1097 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v);
1098 kernel_ns = get_kernel_ns();
1099 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1100 if (unlikely(this_tsc_khz == 0)) {
1101 local_irq_restore(flags);
1102 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1103 return 1;
1104 }
1105
1106 /*
1107 * We may have to catch up the TSC to match elapsed wall clock
1108 * time for two reasons, even if kvmclock is used.
1109 * 1) CPU could have been running below the maximum TSC rate
1110 * 2) Broken TSC compensation resets the base at each VCPU
1111 * entry to avoid unknown leaps of TSC even when running
1112 * again on the same CPU. This may cause apparent elapsed
1113 * time to disappear, and the guest to stand still or run
1114 * very slowly.
1115 */
1116 if (vcpu->tsc_catchup) {
1117 u64 tsc = compute_guest_tsc(v, kernel_ns);
1118 if (tsc > tsc_timestamp) {
1119 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1120 tsc_timestamp = tsc;
1121 }
1122 }
1123
1124 local_irq_restore(flags);
1125
1126 if (!vcpu->time_page)
1127 return 0;
1128
1129 /*
1130 * Time as measured by the TSC may go backwards when resetting the base
1131 * tsc_timestamp. The reason for this is that the TSC resolution is
1132 * higher than the resolution of the other clock scales. Thus, many
1133 * possible measurments of the TSC correspond to one measurement of any
1134 * other clock, and so a spread of values is possible. This is not a
1135 * problem for the computation of the nanosecond clock; with TSC rates
1136 * around 1GHZ, there can only be a few cycles which correspond to one
1137 * nanosecond value, and any path through this code will inevitably
1138 * take longer than that. However, with the kernel_ns value itself,
1139 * the precision may be much lower, down to HZ granularity. If the
1140 * first sampling of TSC against kernel_ns ends in the low part of the
1141 * range, and the second in the high end of the range, we can get:
1142 *
1143 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1144 *
1145 * As the sampling errors potentially range in the thousands of cycles,
1146 * it is possible such a time value has already been observed by the
1147 * guest. To protect against this, we must compute the system time as
1148 * observed by the guest and ensure the new system time is greater.
1149 */
1150 max_kernel_ns = 0;
1151 if (vcpu->hv_clock.tsc_timestamp) {
1152 max_kernel_ns = vcpu->last_guest_tsc -
1153 vcpu->hv_clock.tsc_timestamp;
1154 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1155 vcpu->hv_clock.tsc_to_system_mul,
1156 vcpu->hv_clock.tsc_shift);
1157 max_kernel_ns += vcpu->last_kernel_ns;
1158 }
1159
1160 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1161 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1162 &vcpu->hv_clock.tsc_shift,
1163 &vcpu->hv_clock.tsc_to_system_mul);
1164 vcpu->hw_tsc_khz = this_tsc_khz;
1165 }
1166
1167 if (max_kernel_ns > kernel_ns)
1168 kernel_ns = max_kernel_ns;
1169
1170 /* With all the info we got, fill in the values */
1171 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1172 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1173 vcpu->last_kernel_ns = kernel_ns;
1174 vcpu->last_guest_tsc = tsc_timestamp;
1175 vcpu->hv_clock.flags = 0;
1176
1177 /*
1178 * The interface expects us to write an even number signaling that the
1179 * update is finished. Since the guest won't see the intermediate
1180 * state, we just increase by 2 at the end.
1181 */
1182 vcpu->hv_clock.version += 2;
1183
1184 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1185
1186 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1187 sizeof(vcpu->hv_clock));
1188
1189 kunmap_atomic(shared_kaddr, KM_USER0);
1190
1191 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1192 return 0;
1193}
1194
1195static bool msr_mtrr_valid(unsigned msr)
1196{
1197 switch (msr) {
1198 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1199 case MSR_MTRRfix64K_00000:
1200 case MSR_MTRRfix16K_80000:
1201 case MSR_MTRRfix16K_A0000:
1202 case MSR_MTRRfix4K_C0000:
1203 case MSR_MTRRfix4K_C8000:
1204 case MSR_MTRRfix4K_D0000:
1205 case MSR_MTRRfix4K_D8000:
1206 case MSR_MTRRfix4K_E0000:
1207 case MSR_MTRRfix4K_E8000:
1208 case MSR_MTRRfix4K_F0000:
1209 case MSR_MTRRfix4K_F8000:
1210 case MSR_MTRRdefType:
1211 case MSR_IA32_CR_PAT:
1212 return true;
1213 case 0x2f8:
1214 return true;
1215 }
1216 return false;
1217}
1218
1219static bool valid_pat_type(unsigned t)
1220{
1221 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1222}
1223
1224static bool valid_mtrr_type(unsigned t)
1225{
1226 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1227}
1228
1229static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1230{
1231 int i;
1232
1233 if (!msr_mtrr_valid(msr))
1234 return false;
1235
1236 if (msr == MSR_IA32_CR_PAT) {
1237 for (i = 0; i < 8; i++)
1238 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1239 return false;
1240 return true;
1241 } else if (msr == MSR_MTRRdefType) {
1242 if (data & ~0xcff)
1243 return false;
1244 return valid_mtrr_type(data & 0xff);
1245 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1246 for (i = 0; i < 8 ; i++)
1247 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1248 return false;
1249 return true;
1250 }
1251
1252 /* variable MTRRs */
1253 return valid_mtrr_type(data & 0xff);
1254}
1255
1256static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1257{
1258 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1259
1260 if (!mtrr_valid(vcpu, msr, data))
1261 return 1;
1262
1263 if (msr == MSR_MTRRdefType) {
1264 vcpu->arch.mtrr_state.def_type = data;
1265 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1266 } else if (msr == MSR_MTRRfix64K_00000)
1267 p[0] = data;
1268 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1269 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1270 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1271 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1272 else if (msr == MSR_IA32_CR_PAT)
1273 vcpu->arch.pat = data;
1274 else { /* Variable MTRRs */
1275 int idx, is_mtrr_mask;
1276 u64 *pt;
1277
1278 idx = (msr - 0x200) / 2;
1279 is_mtrr_mask = msr - 0x200 - 2 * idx;
1280 if (!is_mtrr_mask)
1281 pt =
1282 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1283 else
1284 pt =
1285 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1286 *pt = data;
1287 }
1288
1289 kvm_mmu_reset_context(vcpu);
1290 return 0;
1291}
1292
1293static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1294{
1295 u64 mcg_cap = vcpu->arch.mcg_cap;
1296 unsigned bank_num = mcg_cap & 0xff;
1297
1298 switch (msr) {
1299 case MSR_IA32_MCG_STATUS:
1300 vcpu->arch.mcg_status = data;
1301 break;
1302 case MSR_IA32_MCG_CTL:
1303 if (!(mcg_cap & MCG_CTL_P))
1304 return 1;
1305 if (data != 0 && data != ~(u64)0)
1306 return -1;
1307 vcpu->arch.mcg_ctl = data;
1308 break;
1309 default:
1310 if (msr >= MSR_IA32_MC0_CTL &&
1311 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1312 u32 offset = msr - MSR_IA32_MC0_CTL;
1313 /* only 0 or all 1s can be written to IA32_MCi_CTL
1314 * some Linux kernels though clear bit 10 in bank 4 to
1315 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1316 * this to avoid an uncatched #GP in the guest
1317 */
1318 if ((offset & 0x3) == 0 &&
1319 data != 0 && (data | (1 << 10)) != ~(u64)0)
1320 return -1;
1321 vcpu->arch.mce_banks[offset] = data;
1322 break;
1323 }
1324 return 1;
1325 }
1326 return 0;
1327}
1328
1329static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1330{
1331 struct kvm *kvm = vcpu->kvm;
1332 int lm = is_long_mode(vcpu);
1333 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1334 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1335 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1336 : kvm->arch.xen_hvm_config.blob_size_32;
1337 u32 page_num = data & ~PAGE_MASK;
1338 u64 page_addr = data & PAGE_MASK;
1339 u8 *page;
1340 int r;
1341
1342 r = -E2BIG;
1343 if (page_num >= blob_size)
1344 goto out;
1345 r = -ENOMEM;
1346 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1347 if (IS_ERR(page)) {
1348 r = PTR_ERR(page);
1349 goto out;
1350 }
1351 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1352 goto out_free;
1353 r = 0;
1354out_free:
1355 kfree(page);
1356out:
1357 return r;
1358}
1359
1360static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1361{
1362 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1363}
1364
1365static bool kvm_hv_msr_partition_wide(u32 msr)
1366{
1367 bool r = false;
1368 switch (msr) {
1369 case HV_X64_MSR_GUEST_OS_ID:
1370 case HV_X64_MSR_HYPERCALL:
1371 r = true;
1372 break;
1373 }
1374
1375 return r;
1376}
1377
1378static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1379{
1380 struct kvm *kvm = vcpu->kvm;
1381
1382 switch (msr) {
1383 case HV_X64_MSR_GUEST_OS_ID:
1384 kvm->arch.hv_guest_os_id = data;
1385 /* setting guest os id to zero disables hypercall page */
1386 if (!kvm->arch.hv_guest_os_id)
1387 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1388 break;
1389 case HV_X64_MSR_HYPERCALL: {
1390 u64 gfn;
1391 unsigned long addr;
1392 u8 instructions[4];
1393
1394 /* if guest os id is not set hypercall should remain disabled */
1395 if (!kvm->arch.hv_guest_os_id)
1396 break;
1397 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1398 kvm->arch.hv_hypercall = data;
1399 break;
1400 }
1401 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1402 addr = gfn_to_hva(kvm, gfn);
1403 if (kvm_is_error_hva(addr))
1404 return 1;
1405 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1406 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1407 if (__copy_to_user((void __user *)addr, instructions, 4))
1408 return 1;
1409 kvm->arch.hv_hypercall = data;
1410 break;
1411 }
1412 default:
1413 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1414 "data 0x%llx\n", msr, data);
1415 return 1;
1416 }
1417 return 0;
1418}
1419
1420static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1421{
1422 switch (msr) {
1423 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1424 unsigned long addr;
1425
1426 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1427 vcpu->arch.hv_vapic = data;
1428 break;
1429 }
1430 addr = gfn_to_hva(vcpu->kvm, data >>
1431 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1432 if (kvm_is_error_hva(addr))
1433 return 1;
1434 if (__clear_user((void __user *)addr, PAGE_SIZE))
1435 return 1;
1436 vcpu->arch.hv_vapic = data;
1437 break;
1438 }
1439 case HV_X64_MSR_EOI:
1440 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1441 case HV_X64_MSR_ICR:
1442 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1443 case HV_X64_MSR_TPR:
1444 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1445 default:
1446 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1447 "data 0x%llx\n", msr, data);
1448 return 1;
1449 }
1450
1451 return 0;
1452}
1453
1454static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1455{
1456 gpa_t gpa = data & ~0x3f;
1457
1458 /* Bits 2:5 are resrved, Should be zero */
1459 if (data & 0x3c)
1460 return 1;
1461
1462 vcpu->arch.apf.msr_val = data;
1463
1464 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1465 kvm_clear_async_pf_completion_queue(vcpu);
1466 kvm_async_pf_hash_reset(vcpu);
1467 return 0;
1468 }
1469
1470 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1471 return 1;
1472
1473 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1474 kvm_async_pf_wakeup_all(vcpu);
1475 return 0;
1476}
1477
1478static void kvmclock_reset(struct kvm_vcpu *vcpu)
1479{
1480 if (vcpu->arch.time_page) {
1481 kvm_release_page_dirty(vcpu->arch.time_page);
1482 vcpu->arch.time_page = NULL;
1483 }
1484}
1485
1486static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1487{
1488 u64 delta;
1489
1490 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1491 return;
1492
1493 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1494 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1495 vcpu->arch.st.accum_steal = delta;
1496}
1497
1498static void record_steal_time(struct kvm_vcpu *vcpu)
1499{
1500 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1501 return;
1502
1503 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1504 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1505 return;
1506
1507 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1508 vcpu->arch.st.steal.version += 2;
1509 vcpu->arch.st.accum_steal = 0;
1510
1511 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1512 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1513}
1514
1515int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1516{
1517 bool pr = false;
1518
1519 switch (msr) {
1520 case MSR_EFER:
1521 return set_efer(vcpu, data);
1522 case MSR_K7_HWCR:
1523 data &= ~(u64)0x40; /* ignore flush filter disable */
1524 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1525 if (data != 0) {
1526 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1527 data);
1528 return 1;
1529 }
1530 break;
1531 case MSR_FAM10H_MMIO_CONF_BASE:
1532 if (data != 0) {
1533 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1534 "0x%llx\n", data);
1535 return 1;
1536 }
1537 break;
1538 case MSR_AMD64_NB_CFG:
1539 break;
1540 case MSR_IA32_DEBUGCTLMSR:
1541 if (!data) {
1542 /* We support the non-activated case already */
1543 break;
1544 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1545 /* Values other than LBR and BTF are vendor-specific,
1546 thus reserved and should throw a #GP */
1547 return 1;
1548 }
1549 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1550 __func__, data);
1551 break;
1552 case MSR_IA32_UCODE_REV:
1553 case MSR_IA32_UCODE_WRITE:
1554 case MSR_VM_HSAVE_PA:
1555 case MSR_AMD64_PATCH_LOADER:
1556 break;
1557 case 0x200 ... 0x2ff:
1558 return set_msr_mtrr(vcpu, msr, data);
1559 case MSR_IA32_APICBASE:
1560 kvm_set_apic_base(vcpu, data);
1561 break;
1562 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1563 return kvm_x2apic_msr_write(vcpu, msr, data);
1564 case MSR_IA32_TSCDEADLINE:
1565 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1566 break;
1567 case MSR_IA32_MISC_ENABLE:
1568 vcpu->arch.ia32_misc_enable_msr = data;
1569 break;
1570 case MSR_KVM_WALL_CLOCK_NEW:
1571 case MSR_KVM_WALL_CLOCK:
1572 vcpu->kvm->arch.wall_clock = data;
1573 kvm_write_wall_clock(vcpu->kvm, data);
1574 break;
1575 case MSR_KVM_SYSTEM_TIME_NEW:
1576 case MSR_KVM_SYSTEM_TIME: {
1577 kvmclock_reset(vcpu);
1578
1579 vcpu->arch.time = data;
1580 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1581
1582 /* we verify if the enable bit is set... */
1583 if (!(data & 1))
1584 break;
1585
1586 /* ...but clean it before doing the actual write */
1587 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1588
1589 vcpu->arch.time_page =
1590 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1591
1592 if (is_error_page(vcpu->arch.time_page)) {
1593 kvm_release_page_clean(vcpu->arch.time_page);
1594 vcpu->arch.time_page = NULL;
1595 }
1596 break;
1597 }
1598 case MSR_KVM_ASYNC_PF_EN:
1599 if (kvm_pv_enable_async_pf(vcpu, data))
1600 return 1;
1601 break;
1602 case MSR_KVM_STEAL_TIME:
1603
1604 if (unlikely(!sched_info_on()))
1605 return 1;
1606
1607 if (data & KVM_STEAL_RESERVED_MASK)
1608 return 1;
1609
1610 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1611 data & KVM_STEAL_VALID_BITS))
1612 return 1;
1613
1614 vcpu->arch.st.msr_val = data;
1615
1616 if (!(data & KVM_MSR_ENABLED))
1617 break;
1618
1619 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1620
1621 preempt_disable();
1622 accumulate_steal_time(vcpu);
1623 preempt_enable();
1624
1625 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1626
1627 break;
1628
1629 case MSR_IA32_MCG_CTL:
1630 case MSR_IA32_MCG_STATUS:
1631 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1632 return set_msr_mce(vcpu, msr, data);
1633
1634 /* Performance counters are not protected by a CPUID bit,
1635 * so we should check all of them in the generic path for the sake of
1636 * cross vendor migration.
1637 * Writing a zero into the event select MSRs disables them,
1638 * which we perfectly emulate ;-). Any other value should be at least
1639 * reported, some guests depend on them.
1640 */
1641 case MSR_K7_EVNTSEL0:
1642 case MSR_K7_EVNTSEL1:
1643 case MSR_K7_EVNTSEL2:
1644 case MSR_K7_EVNTSEL3:
1645 if (data != 0)
1646 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1647 "0x%x data 0x%llx\n", msr, data);
1648 break;
1649 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1650 * so we ignore writes to make it happy.
1651 */
1652 case MSR_K7_PERFCTR0:
1653 case MSR_K7_PERFCTR1:
1654 case MSR_K7_PERFCTR2:
1655 case MSR_K7_PERFCTR3:
1656 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1657 "0x%x data 0x%llx\n", msr, data);
1658 break;
1659 case MSR_P6_PERFCTR0:
1660 case MSR_P6_PERFCTR1:
1661 pr = true;
1662 case MSR_P6_EVNTSEL0:
1663 case MSR_P6_EVNTSEL1:
1664 if (kvm_pmu_msr(vcpu, msr))
1665 return kvm_pmu_set_msr(vcpu, msr, data);
1666
1667 if (pr || data != 0)
1668 pr_unimpl(vcpu, "disabled perfctr wrmsr: "
1669 "0x%x data 0x%llx\n", msr, data);
1670 break;
1671 case MSR_K7_CLK_CTL:
1672 /*
1673 * Ignore all writes to this no longer documented MSR.
1674 * Writes are only relevant for old K7 processors,
1675 * all pre-dating SVM, but a recommended workaround from
1676 * AMD for these chips. It is possible to speicify the
1677 * affected processor models on the command line, hence
1678 * the need to ignore the workaround.
1679 */
1680 break;
1681 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1682 if (kvm_hv_msr_partition_wide(msr)) {
1683 int r;
1684 mutex_lock(&vcpu->kvm->lock);
1685 r = set_msr_hyperv_pw(vcpu, msr, data);
1686 mutex_unlock(&vcpu->kvm->lock);
1687 return r;
1688 } else
1689 return set_msr_hyperv(vcpu, msr, data);
1690 break;
1691 case MSR_IA32_BBL_CR_CTL3:
1692 /* Drop writes to this legacy MSR -- see rdmsr
1693 * counterpart for further detail.
1694 */
1695 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1696 break;
1697 case MSR_AMD64_OSVW_ID_LENGTH:
1698 if (!guest_cpuid_has_osvw(vcpu))
1699 return 1;
1700 vcpu->arch.osvw.length = data;
1701 break;
1702 case MSR_AMD64_OSVW_STATUS:
1703 if (!guest_cpuid_has_osvw(vcpu))
1704 return 1;
1705 vcpu->arch.osvw.status = data;
1706 break;
1707 default:
1708 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1709 return xen_hvm_config(vcpu, data);
1710 if (kvm_pmu_msr(vcpu, msr))
1711 return kvm_pmu_set_msr(vcpu, msr, data);
1712 if (!ignore_msrs) {
1713 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1714 msr, data);
1715 return 1;
1716 } else {
1717 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1718 msr, data);
1719 break;
1720 }
1721 }
1722 return 0;
1723}
1724EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1725
1726
1727/*
1728 * Reads an msr value (of 'msr_index') into 'pdata'.
1729 * Returns 0 on success, non-0 otherwise.
1730 * Assumes vcpu_load() was already called.
1731 */
1732int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1733{
1734 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1735}
1736
1737static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1738{
1739 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1740
1741 if (!msr_mtrr_valid(msr))
1742 return 1;
1743
1744 if (msr == MSR_MTRRdefType)
1745 *pdata = vcpu->arch.mtrr_state.def_type +
1746 (vcpu->arch.mtrr_state.enabled << 10);
1747 else if (msr == MSR_MTRRfix64K_00000)
1748 *pdata = p[0];
1749 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1750 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1751 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1752 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1753 else if (msr == MSR_IA32_CR_PAT)
1754 *pdata = vcpu->arch.pat;
1755 else { /* Variable MTRRs */
1756 int idx, is_mtrr_mask;
1757 u64 *pt;
1758
1759 idx = (msr - 0x200) / 2;
1760 is_mtrr_mask = msr - 0x200 - 2 * idx;
1761 if (!is_mtrr_mask)
1762 pt =
1763 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1764 else
1765 pt =
1766 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1767 *pdata = *pt;
1768 }
1769
1770 return 0;
1771}
1772
1773static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1774{
1775 u64 data;
1776 u64 mcg_cap = vcpu->arch.mcg_cap;
1777 unsigned bank_num = mcg_cap & 0xff;
1778
1779 switch (msr) {
1780 case MSR_IA32_P5_MC_ADDR:
1781 case MSR_IA32_P5_MC_TYPE:
1782 data = 0;
1783 break;
1784 case MSR_IA32_MCG_CAP:
1785 data = vcpu->arch.mcg_cap;
1786 break;
1787 case MSR_IA32_MCG_CTL:
1788 if (!(mcg_cap & MCG_CTL_P))
1789 return 1;
1790 data = vcpu->arch.mcg_ctl;
1791 break;
1792 case MSR_IA32_MCG_STATUS:
1793 data = vcpu->arch.mcg_status;
1794 break;
1795 default:
1796 if (msr >= MSR_IA32_MC0_CTL &&
1797 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1798 u32 offset = msr - MSR_IA32_MC0_CTL;
1799 data = vcpu->arch.mce_banks[offset];
1800 break;
1801 }
1802 return 1;
1803 }
1804 *pdata = data;
1805 return 0;
1806}
1807
1808static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1809{
1810 u64 data = 0;
1811 struct kvm *kvm = vcpu->kvm;
1812
1813 switch (msr) {
1814 case HV_X64_MSR_GUEST_OS_ID:
1815 data = kvm->arch.hv_guest_os_id;
1816 break;
1817 case HV_X64_MSR_HYPERCALL:
1818 data = kvm->arch.hv_hypercall;
1819 break;
1820 default:
1821 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1822 return 1;
1823 }
1824
1825 *pdata = data;
1826 return 0;
1827}
1828
1829static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1830{
1831 u64 data = 0;
1832
1833 switch (msr) {
1834 case HV_X64_MSR_VP_INDEX: {
1835 int r;
1836 struct kvm_vcpu *v;
1837 kvm_for_each_vcpu(r, v, vcpu->kvm)
1838 if (v == vcpu)
1839 data = r;
1840 break;
1841 }
1842 case HV_X64_MSR_EOI:
1843 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1844 case HV_X64_MSR_ICR:
1845 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1846 case HV_X64_MSR_TPR:
1847 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1848 case HV_X64_MSR_APIC_ASSIST_PAGE:
1849 data = vcpu->arch.hv_vapic;
1850 break;
1851 default:
1852 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1853 return 1;
1854 }
1855 *pdata = data;
1856 return 0;
1857}
1858
1859int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1860{
1861 u64 data;
1862
1863 switch (msr) {
1864 case MSR_IA32_PLATFORM_ID:
1865 case MSR_IA32_EBL_CR_POWERON:
1866 case MSR_IA32_DEBUGCTLMSR:
1867 case MSR_IA32_LASTBRANCHFROMIP:
1868 case MSR_IA32_LASTBRANCHTOIP:
1869 case MSR_IA32_LASTINTFROMIP:
1870 case MSR_IA32_LASTINTTOIP:
1871 case MSR_K8_SYSCFG:
1872 case MSR_K7_HWCR:
1873 case MSR_VM_HSAVE_PA:
1874 case MSR_K7_EVNTSEL0:
1875 case MSR_K7_PERFCTR0:
1876 case MSR_K8_INT_PENDING_MSG:
1877 case MSR_AMD64_NB_CFG:
1878 case MSR_FAM10H_MMIO_CONF_BASE:
1879 data = 0;
1880 break;
1881 case MSR_P6_PERFCTR0:
1882 case MSR_P6_PERFCTR1:
1883 case MSR_P6_EVNTSEL0:
1884 case MSR_P6_EVNTSEL1:
1885 if (kvm_pmu_msr(vcpu, msr))
1886 return kvm_pmu_get_msr(vcpu, msr, pdata);
1887 data = 0;
1888 break;
1889 case MSR_IA32_UCODE_REV:
1890 data = 0x100000000ULL;
1891 break;
1892 case MSR_MTRRcap:
1893 data = 0x500 | KVM_NR_VAR_MTRR;
1894 break;
1895 case 0x200 ... 0x2ff:
1896 return get_msr_mtrr(vcpu, msr, pdata);
1897 case 0xcd: /* fsb frequency */
1898 data = 3;
1899 break;
1900 /*
1901 * MSR_EBC_FREQUENCY_ID
1902 * Conservative value valid for even the basic CPU models.
1903 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1904 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1905 * and 266MHz for model 3, or 4. Set Core Clock
1906 * Frequency to System Bus Frequency Ratio to 1 (bits
1907 * 31:24) even though these are only valid for CPU
1908 * models > 2, however guests may end up dividing or
1909 * multiplying by zero otherwise.
1910 */
1911 case MSR_EBC_FREQUENCY_ID:
1912 data = 1 << 24;
1913 break;
1914 case MSR_IA32_APICBASE:
1915 data = kvm_get_apic_base(vcpu);
1916 break;
1917 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1918 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1919 break;
1920 case MSR_IA32_TSCDEADLINE:
1921 data = kvm_get_lapic_tscdeadline_msr(vcpu);
1922 break;
1923 case MSR_IA32_MISC_ENABLE:
1924 data = vcpu->arch.ia32_misc_enable_msr;
1925 break;
1926 case MSR_IA32_PERF_STATUS:
1927 /* TSC increment by tick */
1928 data = 1000ULL;
1929 /* CPU multiplier */
1930 data |= (((uint64_t)4ULL) << 40);
1931 break;
1932 case MSR_EFER:
1933 data = vcpu->arch.efer;
1934 break;
1935 case MSR_KVM_WALL_CLOCK:
1936 case MSR_KVM_WALL_CLOCK_NEW:
1937 data = vcpu->kvm->arch.wall_clock;
1938 break;
1939 case MSR_KVM_SYSTEM_TIME:
1940 case MSR_KVM_SYSTEM_TIME_NEW:
1941 data = vcpu->arch.time;
1942 break;
1943 case MSR_KVM_ASYNC_PF_EN:
1944 data = vcpu->arch.apf.msr_val;
1945 break;
1946 case MSR_KVM_STEAL_TIME:
1947 data = vcpu->arch.st.msr_val;
1948 break;
1949 case MSR_IA32_P5_MC_ADDR:
1950 case MSR_IA32_P5_MC_TYPE:
1951 case MSR_IA32_MCG_CAP:
1952 case MSR_IA32_MCG_CTL:
1953 case MSR_IA32_MCG_STATUS:
1954 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1955 return get_msr_mce(vcpu, msr, pdata);
1956 case MSR_K7_CLK_CTL:
1957 /*
1958 * Provide expected ramp-up count for K7. All other
1959 * are set to zero, indicating minimum divisors for
1960 * every field.
1961 *
1962 * This prevents guest kernels on AMD host with CPU
1963 * type 6, model 8 and higher from exploding due to
1964 * the rdmsr failing.
1965 */
1966 data = 0x20000000;
1967 break;
1968 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1969 if (kvm_hv_msr_partition_wide(msr)) {
1970 int r;
1971 mutex_lock(&vcpu->kvm->lock);
1972 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1973 mutex_unlock(&vcpu->kvm->lock);
1974 return r;
1975 } else
1976 return get_msr_hyperv(vcpu, msr, pdata);
1977 break;
1978 case MSR_IA32_BBL_CR_CTL3:
1979 /* This legacy MSR exists but isn't fully documented in current
1980 * silicon. It is however accessed by winxp in very narrow
1981 * scenarios where it sets bit #19, itself documented as
1982 * a "reserved" bit. Best effort attempt to source coherent
1983 * read data here should the balance of the register be
1984 * interpreted by the guest:
1985 *
1986 * L2 cache control register 3: 64GB range, 256KB size,
1987 * enabled, latency 0x1, configured
1988 */
1989 data = 0xbe702111;
1990 break;
1991 case MSR_AMD64_OSVW_ID_LENGTH:
1992 if (!guest_cpuid_has_osvw(vcpu))
1993 return 1;
1994 data = vcpu->arch.osvw.length;
1995 break;
1996 case MSR_AMD64_OSVW_STATUS:
1997 if (!guest_cpuid_has_osvw(vcpu))
1998 return 1;
1999 data = vcpu->arch.osvw.status;
2000 break;
2001 default:
2002 if (kvm_pmu_msr(vcpu, msr))
2003 return kvm_pmu_get_msr(vcpu, msr, pdata);
2004 if (!ignore_msrs) {
2005 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2006 return 1;
2007 } else {
2008 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2009 data = 0;
2010 }
2011 break;
2012 }
2013 *pdata = data;
2014 return 0;
2015}
2016EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2017
2018/*
2019 * Read or write a bunch of msrs. All parameters are kernel addresses.
2020 *
2021 * @return number of msrs set successfully.
2022 */
2023static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2024 struct kvm_msr_entry *entries,
2025 int (*do_msr)(struct kvm_vcpu *vcpu,
2026 unsigned index, u64 *data))
2027{
2028 int i, idx;
2029
2030 idx = srcu_read_lock(&vcpu->kvm->srcu);
2031 for (i = 0; i < msrs->nmsrs; ++i)
2032 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2033 break;
2034 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2035
2036 return i;
2037}
2038
2039/*
2040 * Read or write a bunch of msrs. Parameters are user addresses.
2041 *
2042 * @return number of msrs set successfully.
2043 */
2044static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2045 int (*do_msr)(struct kvm_vcpu *vcpu,
2046 unsigned index, u64 *data),
2047 int writeback)
2048{
2049 struct kvm_msrs msrs;
2050 struct kvm_msr_entry *entries;
2051 int r, n;
2052 unsigned size;
2053
2054 r = -EFAULT;
2055 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2056 goto out;
2057
2058 r = -E2BIG;
2059 if (msrs.nmsrs >= MAX_IO_MSRS)
2060 goto out;
2061
2062 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2063 entries = memdup_user(user_msrs->entries, size);
2064 if (IS_ERR(entries)) {
2065 r = PTR_ERR(entries);
2066 goto out;
2067 }
2068
2069 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2070 if (r < 0)
2071 goto out_free;
2072
2073 r = -EFAULT;
2074 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2075 goto out_free;
2076
2077 r = n;
2078
2079out_free:
2080 kfree(entries);
2081out:
2082 return r;
2083}
2084
2085int kvm_dev_ioctl_check_extension(long ext)
2086{
2087 int r;
2088
2089 switch (ext) {
2090 case KVM_CAP_IRQCHIP:
2091 case KVM_CAP_HLT:
2092 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2093 case KVM_CAP_SET_TSS_ADDR:
2094 case KVM_CAP_EXT_CPUID:
2095 case KVM_CAP_CLOCKSOURCE:
2096 case KVM_CAP_PIT:
2097 case KVM_CAP_NOP_IO_DELAY:
2098 case KVM_CAP_MP_STATE:
2099 case KVM_CAP_SYNC_MMU:
2100 case KVM_CAP_USER_NMI:
2101 case KVM_CAP_REINJECT_CONTROL:
2102 case KVM_CAP_IRQ_INJECT_STATUS:
2103 case KVM_CAP_ASSIGN_DEV_IRQ:
2104 case KVM_CAP_IRQFD:
2105 case KVM_CAP_IOEVENTFD:
2106 case KVM_CAP_PIT2:
2107 case KVM_CAP_PIT_STATE2:
2108 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2109 case KVM_CAP_XEN_HVM:
2110 case KVM_CAP_ADJUST_CLOCK:
2111 case KVM_CAP_VCPU_EVENTS:
2112 case KVM_CAP_HYPERV:
2113 case KVM_CAP_HYPERV_VAPIC:
2114 case KVM_CAP_HYPERV_SPIN:
2115 case KVM_CAP_PCI_SEGMENT:
2116 case KVM_CAP_DEBUGREGS:
2117 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2118 case KVM_CAP_XSAVE:
2119 case KVM_CAP_ASYNC_PF:
2120 case KVM_CAP_GET_TSC_KHZ:
2121 r = 1;
2122 break;
2123 case KVM_CAP_COALESCED_MMIO:
2124 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2125 break;
2126 case KVM_CAP_VAPIC:
2127 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2128 break;
2129 case KVM_CAP_NR_VCPUS:
2130 r = KVM_SOFT_MAX_VCPUS;
2131 break;
2132 case KVM_CAP_MAX_VCPUS:
2133 r = KVM_MAX_VCPUS;
2134 break;
2135 case KVM_CAP_NR_MEMSLOTS:
2136 r = KVM_MEMORY_SLOTS;
2137 break;
2138 case KVM_CAP_PV_MMU: /* obsolete */
2139 r = 0;
2140 break;
2141 case KVM_CAP_IOMMU:
2142 r = iommu_present(&pci_bus_type);
2143 break;
2144 case KVM_CAP_MCE:
2145 r = KVM_MAX_MCE_BANKS;
2146 break;
2147 case KVM_CAP_XCRS:
2148 r = cpu_has_xsave;
2149 break;
2150 case KVM_CAP_TSC_CONTROL:
2151 r = kvm_has_tsc_control;
2152 break;
2153 case KVM_CAP_TSC_DEADLINE_TIMER:
2154 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2155 break;
2156 default:
2157 r = 0;
2158 break;
2159 }
2160 return r;
2161
2162}
2163
2164long kvm_arch_dev_ioctl(struct file *filp,
2165 unsigned int ioctl, unsigned long arg)
2166{
2167 void __user *argp = (void __user *)arg;
2168 long r;
2169
2170 switch (ioctl) {
2171 case KVM_GET_MSR_INDEX_LIST: {
2172 struct kvm_msr_list __user *user_msr_list = argp;
2173 struct kvm_msr_list msr_list;
2174 unsigned n;
2175
2176 r = -EFAULT;
2177 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2178 goto out;
2179 n = msr_list.nmsrs;
2180 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2181 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2182 goto out;
2183 r = -E2BIG;
2184 if (n < msr_list.nmsrs)
2185 goto out;
2186 r = -EFAULT;
2187 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2188 num_msrs_to_save * sizeof(u32)))
2189 goto out;
2190 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2191 &emulated_msrs,
2192 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2193 goto out;
2194 r = 0;
2195 break;
2196 }
2197 case KVM_GET_SUPPORTED_CPUID: {
2198 struct kvm_cpuid2 __user *cpuid_arg = argp;
2199 struct kvm_cpuid2 cpuid;
2200
2201 r = -EFAULT;
2202 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2203 goto out;
2204 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2205 cpuid_arg->entries);
2206 if (r)
2207 goto out;
2208
2209 r = -EFAULT;
2210 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2211 goto out;
2212 r = 0;
2213 break;
2214 }
2215 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2216 u64 mce_cap;
2217
2218 mce_cap = KVM_MCE_CAP_SUPPORTED;
2219 r = -EFAULT;
2220 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2221 goto out;
2222 r = 0;
2223 break;
2224 }
2225 default:
2226 r = -EINVAL;
2227 }
2228out:
2229 return r;
2230}
2231
2232static void wbinvd_ipi(void *garbage)
2233{
2234 wbinvd();
2235}
2236
2237static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2238{
2239 return vcpu->kvm->arch.iommu_domain &&
2240 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2241}
2242
2243void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2244{
2245 /* Address WBINVD may be executed by guest */
2246 if (need_emulate_wbinvd(vcpu)) {
2247 if (kvm_x86_ops->has_wbinvd_exit())
2248 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2249 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2250 smp_call_function_single(vcpu->cpu,
2251 wbinvd_ipi, NULL, 1);
2252 }
2253
2254 kvm_x86_ops->vcpu_load(vcpu, cpu);
2255 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2256 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2257 native_read_tsc() - vcpu->arch.last_host_tsc;
2258 if (tsc_delta < 0)
2259 mark_tsc_unstable("KVM discovered backwards TSC");
2260 if (check_tsc_unstable()) {
2261 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2262 vcpu->arch.last_guest_tsc);
2263 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2264 vcpu->arch.tsc_catchup = 1;
2265 }
2266 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2267 if (vcpu->cpu != cpu)
2268 kvm_migrate_timers(vcpu);
2269 vcpu->cpu = cpu;
2270 }
2271
2272 accumulate_steal_time(vcpu);
2273 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2274}
2275
2276void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2277{
2278 kvm_x86_ops->vcpu_put(vcpu);
2279 kvm_put_guest_fpu(vcpu);
2280 vcpu->arch.last_host_tsc = native_read_tsc();
2281}
2282
2283static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2284 struct kvm_lapic_state *s)
2285{
2286 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2287
2288 return 0;
2289}
2290
2291static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2292 struct kvm_lapic_state *s)
2293{
2294 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2295 kvm_apic_post_state_restore(vcpu);
2296 update_cr8_intercept(vcpu);
2297
2298 return 0;
2299}
2300
2301static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2302 struct kvm_interrupt *irq)
2303{
2304 if (irq->irq < 0 || irq->irq >= 256)
2305 return -EINVAL;
2306 if (irqchip_in_kernel(vcpu->kvm))
2307 return -ENXIO;
2308
2309 kvm_queue_interrupt(vcpu, irq->irq, false);
2310 kvm_make_request(KVM_REQ_EVENT, vcpu);
2311
2312 return 0;
2313}
2314
2315static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2316{
2317 kvm_inject_nmi(vcpu);
2318
2319 return 0;
2320}
2321
2322static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2323 struct kvm_tpr_access_ctl *tac)
2324{
2325 if (tac->flags)
2326 return -EINVAL;
2327 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2328 return 0;
2329}
2330
2331static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2332 u64 mcg_cap)
2333{
2334 int r;
2335 unsigned bank_num = mcg_cap & 0xff, bank;
2336
2337 r = -EINVAL;
2338 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2339 goto out;
2340 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2341 goto out;
2342 r = 0;
2343 vcpu->arch.mcg_cap = mcg_cap;
2344 /* Init IA32_MCG_CTL to all 1s */
2345 if (mcg_cap & MCG_CTL_P)
2346 vcpu->arch.mcg_ctl = ~(u64)0;
2347 /* Init IA32_MCi_CTL to all 1s */
2348 for (bank = 0; bank < bank_num; bank++)
2349 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2350out:
2351 return r;
2352}
2353
2354static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2355 struct kvm_x86_mce *mce)
2356{
2357 u64 mcg_cap = vcpu->arch.mcg_cap;
2358 unsigned bank_num = mcg_cap & 0xff;
2359 u64 *banks = vcpu->arch.mce_banks;
2360
2361 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2362 return -EINVAL;
2363 /*
2364 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2365 * reporting is disabled
2366 */
2367 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2368 vcpu->arch.mcg_ctl != ~(u64)0)
2369 return 0;
2370 banks += 4 * mce->bank;
2371 /*
2372 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2373 * reporting is disabled for the bank
2374 */
2375 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2376 return 0;
2377 if (mce->status & MCI_STATUS_UC) {
2378 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2379 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2380 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2381 return 0;
2382 }
2383 if (banks[1] & MCI_STATUS_VAL)
2384 mce->status |= MCI_STATUS_OVER;
2385 banks[2] = mce->addr;
2386 banks[3] = mce->misc;
2387 vcpu->arch.mcg_status = mce->mcg_status;
2388 banks[1] = mce->status;
2389 kvm_queue_exception(vcpu, MC_VECTOR);
2390 } else if (!(banks[1] & MCI_STATUS_VAL)
2391 || !(banks[1] & MCI_STATUS_UC)) {
2392 if (banks[1] & MCI_STATUS_VAL)
2393 mce->status |= MCI_STATUS_OVER;
2394 banks[2] = mce->addr;
2395 banks[3] = mce->misc;
2396 banks[1] = mce->status;
2397 } else
2398 banks[1] |= MCI_STATUS_OVER;
2399 return 0;
2400}
2401
2402static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2403 struct kvm_vcpu_events *events)
2404{
2405 process_nmi(vcpu);
2406 events->exception.injected =
2407 vcpu->arch.exception.pending &&
2408 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2409 events->exception.nr = vcpu->arch.exception.nr;
2410 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2411 events->exception.pad = 0;
2412 events->exception.error_code = vcpu->arch.exception.error_code;
2413
2414 events->interrupt.injected =
2415 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2416 events->interrupt.nr = vcpu->arch.interrupt.nr;
2417 events->interrupt.soft = 0;
2418 events->interrupt.shadow =
2419 kvm_x86_ops->get_interrupt_shadow(vcpu,
2420 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2421
2422 events->nmi.injected = vcpu->arch.nmi_injected;
2423 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2424 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2425 events->nmi.pad = 0;
2426
2427 events->sipi_vector = vcpu->arch.sipi_vector;
2428
2429 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2430 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2431 | KVM_VCPUEVENT_VALID_SHADOW);
2432 memset(&events->reserved, 0, sizeof(events->reserved));
2433}
2434
2435static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2436 struct kvm_vcpu_events *events)
2437{
2438 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2439 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2440 | KVM_VCPUEVENT_VALID_SHADOW))
2441 return -EINVAL;
2442
2443 process_nmi(vcpu);
2444 vcpu->arch.exception.pending = events->exception.injected;
2445 vcpu->arch.exception.nr = events->exception.nr;
2446 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2447 vcpu->arch.exception.error_code = events->exception.error_code;
2448
2449 vcpu->arch.interrupt.pending = events->interrupt.injected;
2450 vcpu->arch.interrupt.nr = events->interrupt.nr;
2451 vcpu->arch.interrupt.soft = events->interrupt.soft;
2452 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2453 kvm_x86_ops->set_interrupt_shadow(vcpu,
2454 events->interrupt.shadow);
2455
2456 vcpu->arch.nmi_injected = events->nmi.injected;
2457 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2458 vcpu->arch.nmi_pending = events->nmi.pending;
2459 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2460
2461 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2462 vcpu->arch.sipi_vector = events->sipi_vector;
2463
2464 kvm_make_request(KVM_REQ_EVENT, vcpu);
2465
2466 return 0;
2467}
2468
2469static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2470 struct kvm_debugregs *dbgregs)
2471{
2472 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2473 dbgregs->dr6 = vcpu->arch.dr6;
2474 dbgregs->dr7 = vcpu->arch.dr7;
2475 dbgregs->flags = 0;
2476 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2477}
2478
2479static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2480 struct kvm_debugregs *dbgregs)
2481{
2482 if (dbgregs->flags)
2483 return -EINVAL;
2484
2485 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2486 vcpu->arch.dr6 = dbgregs->dr6;
2487 vcpu->arch.dr7 = dbgregs->dr7;
2488
2489 return 0;
2490}
2491
2492static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2493 struct kvm_xsave *guest_xsave)
2494{
2495 if (cpu_has_xsave)
2496 memcpy(guest_xsave->region,
2497 &vcpu->arch.guest_fpu.state->xsave,
2498 xstate_size);
2499 else {
2500 memcpy(guest_xsave->region,
2501 &vcpu->arch.guest_fpu.state->fxsave,
2502 sizeof(struct i387_fxsave_struct));
2503 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2504 XSTATE_FPSSE;
2505 }
2506}
2507
2508static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2509 struct kvm_xsave *guest_xsave)
2510{
2511 u64 xstate_bv =
2512 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2513
2514 if (cpu_has_xsave)
2515 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2516 guest_xsave->region, xstate_size);
2517 else {
2518 if (xstate_bv & ~XSTATE_FPSSE)
2519 return -EINVAL;
2520 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2521 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2522 }
2523 return 0;
2524}
2525
2526static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2527 struct kvm_xcrs *guest_xcrs)
2528{
2529 if (!cpu_has_xsave) {
2530 guest_xcrs->nr_xcrs = 0;
2531 return;
2532 }
2533
2534 guest_xcrs->nr_xcrs = 1;
2535 guest_xcrs->flags = 0;
2536 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2537 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2538}
2539
2540static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2541 struct kvm_xcrs *guest_xcrs)
2542{
2543 int i, r = 0;
2544
2545 if (!cpu_has_xsave)
2546 return -EINVAL;
2547
2548 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2549 return -EINVAL;
2550
2551 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2552 /* Only support XCR0 currently */
2553 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2554 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2555 guest_xcrs->xcrs[0].value);
2556 break;
2557 }
2558 if (r)
2559 r = -EINVAL;
2560 return r;
2561}
2562
2563long kvm_arch_vcpu_ioctl(struct file *filp,
2564 unsigned int ioctl, unsigned long arg)
2565{
2566 struct kvm_vcpu *vcpu = filp->private_data;
2567 void __user *argp = (void __user *)arg;
2568 int r;
2569 union {
2570 struct kvm_lapic_state *lapic;
2571 struct kvm_xsave *xsave;
2572 struct kvm_xcrs *xcrs;
2573 void *buffer;
2574 } u;
2575
2576 u.buffer = NULL;
2577 switch (ioctl) {
2578 case KVM_GET_LAPIC: {
2579 r = -EINVAL;
2580 if (!vcpu->arch.apic)
2581 goto out;
2582 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2583
2584 r = -ENOMEM;
2585 if (!u.lapic)
2586 goto out;
2587 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2588 if (r)
2589 goto out;
2590 r = -EFAULT;
2591 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2592 goto out;
2593 r = 0;
2594 break;
2595 }
2596 case KVM_SET_LAPIC: {
2597 r = -EINVAL;
2598 if (!vcpu->arch.apic)
2599 goto out;
2600 u.lapic = memdup_user(argp, sizeof(*u.lapic));
2601 if (IS_ERR(u.lapic)) {
2602 r = PTR_ERR(u.lapic);
2603 goto out;
2604 }
2605
2606 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2607 if (r)
2608 goto out;
2609 r = 0;
2610 break;
2611 }
2612 case KVM_INTERRUPT: {
2613 struct kvm_interrupt irq;
2614
2615 r = -EFAULT;
2616 if (copy_from_user(&irq, argp, sizeof irq))
2617 goto out;
2618 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2619 if (r)
2620 goto out;
2621 r = 0;
2622 break;
2623 }
2624 case KVM_NMI: {
2625 r = kvm_vcpu_ioctl_nmi(vcpu);
2626 if (r)
2627 goto out;
2628 r = 0;
2629 break;
2630 }
2631 case KVM_SET_CPUID: {
2632 struct kvm_cpuid __user *cpuid_arg = argp;
2633 struct kvm_cpuid cpuid;
2634
2635 r = -EFAULT;
2636 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2637 goto out;
2638 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2639 if (r)
2640 goto out;
2641 break;
2642 }
2643 case KVM_SET_CPUID2: {
2644 struct kvm_cpuid2 __user *cpuid_arg = argp;
2645 struct kvm_cpuid2 cpuid;
2646
2647 r = -EFAULT;
2648 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2649 goto out;
2650 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2651 cpuid_arg->entries);
2652 if (r)
2653 goto out;
2654 break;
2655 }
2656 case KVM_GET_CPUID2: {
2657 struct kvm_cpuid2 __user *cpuid_arg = argp;
2658 struct kvm_cpuid2 cpuid;
2659
2660 r = -EFAULT;
2661 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2662 goto out;
2663 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2664 cpuid_arg->entries);
2665 if (r)
2666 goto out;
2667 r = -EFAULT;
2668 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2669 goto out;
2670 r = 0;
2671 break;
2672 }
2673 case KVM_GET_MSRS:
2674 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2675 break;
2676 case KVM_SET_MSRS:
2677 r = msr_io(vcpu, argp, do_set_msr, 0);
2678 break;
2679 case KVM_TPR_ACCESS_REPORTING: {
2680 struct kvm_tpr_access_ctl tac;
2681
2682 r = -EFAULT;
2683 if (copy_from_user(&tac, argp, sizeof tac))
2684 goto out;
2685 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2686 if (r)
2687 goto out;
2688 r = -EFAULT;
2689 if (copy_to_user(argp, &tac, sizeof tac))
2690 goto out;
2691 r = 0;
2692 break;
2693 };
2694 case KVM_SET_VAPIC_ADDR: {
2695 struct kvm_vapic_addr va;
2696
2697 r = -EINVAL;
2698 if (!irqchip_in_kernel(vcpu->kvm))
2699 goto out;
2700 r = -EFAULT;
2701 if (copy_from_user(&va, argp, sizeof va))
2702 goto out;
2703 r = 0;
2704 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2705 break;
2706 }
2707 case KVM_X86_SETUP_MCE: {
2708 u64 mcg_cap;
2709
2710 r = -EFAULT;
2711 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2712 goto out;
2713 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2714 break;
2715 }
2716 case KVM_X86_SET_MCE: {
2717 struct kvm_x86_mce mce;
2718
2719 r = -EFAULT;
2720 if (copy_from_user(&mce, argp, sizeof mce))
2721 goto out;
2722 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2723 break;
2724 }
2725 case KVM_GET_VCPU_EVENTS: {
2726 struct kvm_vcpu_events events;
2727
2728 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2729
2730 r = -EFAULT;
2731 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2732 break;
2733 r = 0;
2734 break;
2735 }
2736 case KVM_SET_VCPU_EVENTS: {
2737 struct kvm_vcpu_events events;
2738
2739 r = -EFAULT;
2740 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2741 break;
2742
2743 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2744 break;
2745 }
2746 case KVM_GET_DEBUGREGS: {
2747 struct kvm_debugregs dbgregs;
2748
2749 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2750
2751 r = -EFAULT;
2752 if (copy_to_user(argp, &dbgregs,
2753 sizeof(struct kvm_debugregs)))
2754 break;
2755 r = 0;
2756 break;
2757 }
2758 case KVM_SET_DEBUGREGS: {
2759 struct kvm_debugregs dbgregs;
2760
2761 r = -EFAULT;
2762 if (copy_from_user(&dbgregs, argp,
2763 sizeof(struct kvm_debugregs)))
2764 break;
2765
2766 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2767 break;
2768 }
2769 case KVM_GET_XSAVE: {
2770 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2771 r = -ENOMEM;
2772 if (!u.xsave)
2773 break;
2774
2775 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2776
2777 r = -EFAULT;
2778 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2779 break;
2780 r = 0;
2781 break;
2782 }
2783 case KVM_SET_XSAVE: {
2784 u.xsave = memdup_user(argp, sizeof(*u.xsave));
2785 if (IS_ERR(u.xsave)) {
2786 r = PTR_ERR(u.xsave);
2787 goto out;
2788 }
2789
2790 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2791 break;
2792 }
2793 case KVM_GET_XCRS: {
2794 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2795 r = -ENOMEM;
2796 if (!u.xcrs)
2797 break;
2798
2799 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2800
2801 r = -EFAULT;
2802 if (copy_to_user(argp, u.xcrs,
2803 sizeof(struct kvm_xcrs)))
2804 break;
2805 r = 0;
2806 break;
2807 }
2808 case KVM_SET_XCRS: {
2809 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
2810 if (IS_ERR(u.xcrs)) {
2811 r = PTR_ERR(u.xcrs);
2812 goto out;
2813 }
2814
2815 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2816 break;
2817 }
2818 case KVM_SET_TSC_KHZ: {
2819 u32 user_tsc_khz;
2820
2821 r = -EINVAL;
2822 user_tsc_khz = (u32)arg;
2823
2824 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
2825 goto out;
2826
2827 if (user_tsc_khz == 0)
2828 user_tsc_khz = tsc_khz;
2829
2830 kvm_set_tsc_khz(vcpu, user_tsc_khz);
2831
2832 r = 0;
2833 goto out;
2834 }
2835 case KVM_GET_TSC_KHZ: {
2836 r = vcpu->arch.virtual_tsc_khz;
2837 goto out;
2838 }
2839 default:
2840 r = -EINVAL;
2841 }
2842out:
2843 kfree(u.buffer);
2844 return r;
2845}
2846
2847int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2848{
2849 return VM_FAULT_SIGBUS;
2850}
2851
2852static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2853{
2854 int ret;
2855
2856 if (addr > (unsigned int)(-3 * PAGE_SIZE))
2857 return -1;
2858 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2859 return ret;
2860}
2861
2862static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2863 u64 ident_addr)
2864{
2865 kvm->arch.ept_identity_map_addr = ident_addr;
2866 return 0;
2867}
2868
2869static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2870 u32 kvm_nr_mmu_pages)
2871{
2872 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2873 return -EINVAL;
2874
2875 mutex_lock(&kvm->slots_lock);
2876 spin_lock(&kvm->mmu_lock);
2877
2878 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2879 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2880
2881 spin_unlock(&kvm->mmu_lock);
2882 mutex_unlock(&kvm->slots_lock);
2883 return 0;
2884}
2885
2886static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2887{
2888 return kvm->arch.n_max_mmu_pages;
2889}
2890
2891static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2892{
2893 int r;
2894
2895 r = 0;
2896 switch (chip->chip_id) {
2897 case KVM_IRQCHIP_PIC_MASTER:
2898 memcpy(&chip->chip.pic,
2899 &pic_irqchip(kvm)->pics[0],
2900 sizeof(struct kvm_pic_state));
2901 break;
2902 case KVM_IRQCHIP_PIC_SLAVE:
2903 memcpy(&chip->chip.pic,
2904 &pic_irqchip(kvm)->pics[1],
2905 sizeof(struct kvm_pic_state));
2906 break;
2907 case KVM_IRQCHIP_IOAPIC:
2908 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2909 break;
2910 default:
2911 r = -EINVAL;
2912 break;
2913 }
2914 return r;
2915}
2916
2917static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2918{
2919 int r;
2920
2921 r = 0;
2922 switch (chip->chip_id) {
2923 case KVM_IRQCHIP_PIC_MASTER:
2924 spin_lock(&pic_irqchip(kvm)->lock);
2925 memcpy(&pic_irqchip(kvm)->pics[0],
2926 &chip->chip.pic,
2927 sizeof(struct kvm_pic_state));
2928 spin_unlock(&pic_irqchip(kvm)->lock);
2929 break;
2930 case KVM_IRQCHIP_PIC_SLAVE:
2931 spin_lock(&pic_irqchip(kvm)->lock);
2932 memcpy(&pic_irqchip(kvm)->pics[1],
2933 &chip->chip.pic,
2934 sizeof(struct kvm_pic_state));
2935 spin_unlock(&pic_irqchip(kvm)->lock);
2936 break;
2937 case KVM_IRQCHIP_IOAPIC:
2938 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2939 break;
2940 default:
2941 r = -EINVAL;
2942 break;
2943 }
2944 kvm_pic_update_irq(pic_irqchip(kvm));
2945 return r;
2946}
2947
2948static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2949{
2950 int r = 0;
2951
2952 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2953 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2954 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2955 return r;
2956}
2957
2958static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2959{
2960 int r = 0;
2961
2962 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2963 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2964 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2965 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2966 return r;
2967}
2968
2969static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2970{
2971 int r = 0;
2972
2973 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2974 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2975 sizeof(ps->channels));
2976 ps->flags = kvm->arch.vpit->pit_state.flags;
2977 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2978 memset(&ps->reserved, 0, sizeof(ps->reserved));
2979 return r;
2980}
2981
2982static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2983{
2984 int r = 0, start = 0;
2985 u32 prev_legacy, cur_legacy;
2986 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2987 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2988 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2989 if (!prev_legacy && cur_legacy)
2990 start = 1;
2991 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2992 sizeof(kvm->arch.vpit->pit_state.channels));
2993 kvm->arch.vpit->pit_state.flags = ps->flags;
2994 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2995 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2996 return r;
2997}
2998
2999static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3000 struct kvm_reinject_control *control)
3001{
3002 if (!kvm->arch.vpit)
3003 return -ENXIO;
3004 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3005 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3006 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3007 return 0;
3008}
3009
3010/**
3011 * write_protect_slot - write protect a slot for dirty logging
3012 * @kvm: the kvm instance
3013 * @memslot: the slot we protect
3014 * @dirty_bitmap: the bitmap indicating which pages are dirty
3015 * @nr_dirty_pages: the number of dirty pages
3016 *
3017 * We have two ways to find all sptes to protect:
3018 * 1. Use kvm_mmu_slot_remove_write_access() which walks all shadow pages and
3019 * checks ones that have a spte mapping a page in the slot.
3020 * 2. Use kvm_mmu_rmap_write_protect() for each gfn found in the bitmap.
3021 *
3022 * Generally speaking, if there are not so many dirty pages compared to the
3023 * number of shadow pages, we should use the latter.
3024 *
3025 * Note that letting others write into a page marked dirty in the old bitmap
3026 * by using the remaining tlb entry is not a problem. That page will become
3027 * write protected again when we flush the tlb and then be reported dirty to
3028 * the user space by copying the old bitmap.
3029 */
3030static void write_protect_slot(struct kvm *kvm,
3031 struct kvm_memory_slot *memslot,
3032 unsigned long *dirty_bitmap,
3033 unsigned long nr_dirty_pages)
3034{
3035 /* Not many dirty pages compared to # of shadow pages. */
3036 if (nr_dirty_pages < kvm->arch.n_used_mmu_pages) {
3037 unsigned long gfn_offset;
3038
3039 for_each_set_bit(gfn_offset, dirty_bitmap, memslot->npages) {
3040 unsigned long gfn = memslot->base_gfn + gfn_offset;
3041
3042 spin_lock(&kvm->mmu_lock);
3043 kvm_mmu_rmap_write_protect(kvm, gfn, memslot);
3044 spin_unlock(&kvm->mmu_lock);
3045 }
3046 kvm_flush_remote_tlbs(kvm);
3047 } else {
3048 spin_lock(&kvm->mmu_lock);
3049 kvm_mmu_slot_remove_write_access(kvm, memslot->id);
3050 spin_unlock(&kvm->mmu_lock);
3051 }
3052}
3053
3054/*
3055 * Get (and clear) the dirty memory log for a memory slot.
3056 */
3057int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3058 struct kvm_dirty_log *log)
3059{
3060 int r;
3061 struct kvm_memory_slot *memslot;
3062 unsigned long n, nr_dirty_pages;
3063
3064 mutex_lock(&kvm->slots_lock);
3065
3066 r = -EINVAL;
3067 if (log->slot >= KVM_MEMORY_SLOTS)
3068 goto out;
3069
3070 memslot = id_to_memslot(kvm->memslots, log->slot);
3071 r = -ENOENT;
3072 if (!memslot->dirty_bitmap)
3073 goto out;
3074
3075 n = kvm_dirty_bitmap_bytes(memslot);
3076 nr_dirty_pages = memslot->nr_dirty_pages;
3077
3078 /* If nothing is dirty, don't bother messing with page tables. */
3079 if (nr_dirty_pages) {
3080 struct kvm_memslots *slots, *old_slots;
3081 unsigned long *dirty_bitmap, *dirty_bitmap_head;
3082
3083 dirty_bitmap = memslot->dirty_bitmap;
3084 dirty_bitmap_head = memslot->dirty_bitmap_head;
3085 if (dirty_bitmap == dirty_bitmap_head)
3086 dirty_bitmap_head += n / sizeof(long);
3087 memset(dirty_bitmap_head, 0, n);
3088
3089 r = -ENOMEM;
3090 slots = kmemdup(kvm->memslots, sizeof(*kvm->memslots), GFP_KERNEL);
3091 if (!slots)
3092 goto out;
3093
3094 memslot = id_to_memslot(slots, log->slot);
3095 memslot->nr_dirty_pages = 0;
3096 memslot->dirty_bitmap = dirty_bitmap_head;
3097 update_memslots(slots, NULL);
3098
3099 old_slots = kvm->memslots;
3100 rcu_assign_pointer(kvm->memslots, slots);
3101 synchronize_srcu_expedited(&kvm->srcu);
3102 kfree(old_slots);
3103
3104 write_protect_slot(kvm, memslot, dirty_bitmap, nr_dirty_pages);
3105
3106 r = -EFAULT;
3107 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3108 goto out;
3109 } else {
3110 r = -EFAULT;
3111 if (clear_user(log->dirty_bitmap, n))
3112 goto out;
3113 }
3114
3115 r = 0;
3116out:
3117 mutex_unlock(&kvm->slots_lock);
3118 return r;
3119}
3120
3121long kvm_arch_vm_ioctl(struct file *filp,
3122 unsigned int ioctl, unsigned long arg)
3123{
3124 struct kvm *kvm = filp->private_data;
3125 void __user *argp = (void __user *)arg;
3126 int r = -ENOTTY;
3127 /*
3128 * This union makes it completely explicit to gcc-3.x
3129 * that these two variables' stack usage should be
3130 * combined, not added together.
3131 */
3132 union {
3133 struct kvm_pit_state ps;
3134 struct kvm_pit_state2 ps2;
3135 struct kvm_pit_config pit_config;
3136 } u;
3137
3138 switch (ioctl) {
3139 case KVM_SET_TSS_ADDR:
3140 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3141 if (r < 0)
3142 goto out;
3143 break;
3144 case KVM_SET_IDENTITY_MAP_ADDR: {
3145 u64 ident_addr;
3146
3147 r = -EFAULT;
3148 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3149 goto out;
3150 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3151 if (r < 0)
3152 goto out;
3153 break;
3154 }
3155 case KVM_SET_NR_MMU_PAGES:
3156 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3157 if (r)
3158 goto out;
3159 break;
3160 case KVM_GET_NR_MMU_PAGES:
3161 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3162 break;
3163 case KVM_CREATE_IRQCHIP: {
3164 struct kvm_pic *vpic;
3165
3166 mutex_lock(&kvm->lock);
3167 r = -EEXIST;
3168 if (kvm->arch.vpic)
3169 goto create_irqchip_unlock;
3170 r = -ENOMEM;
3171 vpic = kvm_create_pic(kvm);
3172 if (vpic) {
3173 r = kvm_ioapic_init(kvm);
3174 if (r) {
3175 mutex_lock(&kvm->slots_lock);
3176 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3177 &vpic->dev_master);
3178 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3179 &vpic->dev_slave);
3180 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3181 &vpic->dev_eclr);
3182 mutex_unlock(&kvm->slots_lock);
3183 kfree(vpic);
3184 goto create_irqchip_unlock;
3185 }
3186 } else
3187 goto create_irqchip_unlock;
3188 smp_wmb();
3189 kvm->arch.vpic = vpic;
3190 smp_wmb();
3191 r = kvm_setup_default_irq_routing(kvm);
3192 if (r) {
3193 mutex_lock(&kvm->slots_lock);
3194 mutex_lock(&kvm->irq_lock);
3195 kvm_ioapic_destroy(kvm);
3196 kvm_destroy_pic(kvm);
3197 mutex_unlock(&kvm->irq_lock);
3198 mutex_unlock(&kvm->slots_lock);
3199 }
3200 create_irqchip_unlock:
3201 mutex_unlock(&kvm->lock);
3202 break;
3203 }
3204 case KVM_CREATE_PIT:
3205 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3206 goto create_pit;
3207 case KVM_CREATE_PIT2:
3208 r = -EFAULT;
3209 if (copy_from_user(&u.pit_config, argp,
3210 sizeof(struct kvm_pit_config)))
3211 goto out;
3212 create_pit:
3213 mutex_lock(&kvm->slots_lock);
3214 r = -EEXIST;
3215 if (kvm->arch.vpit)
3216 goto create_pit_unlock;
3217 r = -ENOMEM;
3218 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3219 if (kvm->arch.vpit)
3220 r = 0;
3221 create_pit_unlock:
3222 mutex_unlock(&kvm->slots_lock);
3223 break;
3224 case KVM_IRQ_LINE_STATUS:
3225 case KVM_IRQ_LINE: {
3226 struct kvm_irq_level irq_event;
3227
3228 r = -EFAULT;
3229 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3230 goto out;
3231 r = -ENXIO;
3232 if (irqchip_in_kernel(kvm)) {
3233 __s32 status;
3234 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3235 irq_event.irq, irq_event.level);
3236 if (ioctl == KVM_IRQ_LINE_STATUS) {
3237 r = -EFAULT;
3238 irq_event.status = status;
3239 if (copy_to_user(argp, &irq_event,
3240 sizeof irq_event))
3241 goto out;
3242 }
3243 r = 0;
3244 }
3245 break;
3246 }
3247 case KVM_GET_IRQCHIP: {
3248 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3249 struct kvm_irqchip *chip;
3250
3251 chip = memdup_user(argp, sizeof(*chip));
3252 if (IS_ERR(chip)) {
3253 r = PTR_ERR(chip);
3254 goto out;
3255 }
3256
3257 r = -ENXIO;
3258 if (!irqchip_in_kernel(kvm))
3259 goto get_irqchip_out;
3260 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3261 if (r)
3262 goto get_irqchip_out;
3263 r = -EFAULT;
3264 if (copy_to_user(argp, chip, sizeof *chip))
3265 goto get_irqchip_out;
3266 r = 0;
3267 get_irqchip_out:
3268 kfree(chip);
3269 if (r)
3270 goto out;
3271 break;
3272 }
3273 case KVM_SET_IRQCHIP: {
3274 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3275 struct kvm_irqchip *chip;
3276
3277 chip = memdup_user(argp, sizeof(*chip));
3278 if (IS_ERR(chip)) {
3279 r = PTR_ERR(chip);
3280 goto out;
3281 }
3282
3283 r = -ENXIO;
3284 if (!irqchip_in_kernel(kvm))
3285 goto set_irqchip_out;
3286 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3287 if (r)
3288 goto set_irqchip_out;
3289 r = 0;
3290 set_irqchip_out:
3291 kfree(chip);
3292 if (r)
3293 goto out;
3294 break;
3295 }
3296 case KVM_GET_PIT: {
3297 r = -EFAULT;
3298 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3299 goto out;
3300 r = -ENXIO;
3301 if (!kvm->arch.vpit)
3302 goto out;
3303 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3304 if (r)
3305 goto out;
3306 r = -EFAULT;
3307 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3308 goto out;
3309 r = 0;
3310 break;
3311 }
3312 case KVM_SET_PIT: {
3313 r = -EFAULT;
3314 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3315 goto out;
3316 r = -ENXIO;
3317 if (!kvm->arch.vpit)
3318 goto out;
3319 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3320 if (r)
3321 goto out;
3322 r = 0;
3323 break;
3324 }
3325 case KVM_GET_PIT2: {
3326 r = -ENXIO;
3327 if (!kvm->arch.vpit)
3328 goto out;
3329 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3330 if (r)
3331 goto out;
3332 r = -EFAULT;
3333 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3334 goto out;
3335 r = 0;
3336 break;
3337 }
3338 case KVM_SET_PIT2: {
3339 r = -EFAULT;
3340 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3341 goto out;
3342 r = -ENXIO;
3343 if (!kvm->arch.vpit)
3344 goto out;
3345 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3346 if (r)
3347 goto out;
3348 r = 0;
3349 break;
3350 }
3351 case KVM_REINJECT_CONTROL: {
3352 struct kvm_reinject_control control;
3353 r = -EFAULT;
3354 if (copy_from_user(&control, argp, sizeof(control)))
3355 goto out;
3356 r = kvm_vm_ioctl_reinject(kvm, &control);
3357 if (r)
3358 goto out;
3359 r = 0;
3360 break;
3361 }
3362 case KVM_XEN_HVM_CONFIG: {
3363 r = -EFAULT;
3364 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3365 sizeof(struct kvm_xen_hvm_config)))
3366 goto out;
3367 r = -EINVAL;
3368 if (kvm->arch.xen_hvm_config.flags)
3369 goto out;
3370 r = 0;
3371 break;
3372 }
3373 case KVM_SET_CLOCK: {
3374 struct kvm_clock_data user_ns;
3375 u64 now_ns;
3376 s64 delta;
3377
3378 r = -EFAULT;
3379 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3380 goto out;
3381
3382 r = -EINVAL;
3383 if (user_ns.flags)
3384 goto out;
3385
3386 r = 0;
3387 local_irq_disable();
3388 now_ns = get_kernel_ns();
3389 delta = user_ns.clock - now_ns;
3390 local_irq_enable();
3391 kvm->arch.kvmclock_offset = delta;
3392 break;
3393 }
3394 case KVM_GET_CLOCK: {
3395 struct kvm_clock_data user_ns;
3396 u64 now_ns;
3397
3398 local_irq_disable();
3399 now_ns = get_kernel_ns();
3400 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3401 local_irq_enable();
3402 user_ns.flags = 0;
3403 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3404
3405 r = -EFAULT;
3406 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3407 goto out;
3408 r = 0;
3409 break;
3410 }
3411
3412 default:
3413 ;
3414 }
3415out:
3416 return r;
3417}
3418
3419static void kvm_init_msr_list(void)
3420{
3421 u32 dummy[2];
3422 unsigned i, j;
3423
3424 /* skip the first msrs in the list. KVM-specific */
3425 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3426 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3427 continue;
3428 if (j < i)
3429 msrs_to_save[j] = msrs_to_save[i];
3430 j++;
3431 }
3432 num_msrs_to_save = j;
3433}
3434
3435static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3436 const void *v)
3437{
3438 int handled = 0;
3439 int n;
3440
3441 do {
3442 n = min(len, 8);
3443 if (!(vcpu->arch.apic &&
3444 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3445 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3446 break;
3447 handled += n;
3448 addr += n;
3449 len -= n;
3450 v += n;
3451 } while (len);
3452
3453 return handled;
3454}
3455
3456static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3457{
3458 int handled = 0;
3459 int n;
3460
3461 do {
3462 n = min(len, 8);
3463 if (!(vcpu->arch.apic &&
3464 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3465 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3466 break;
3467 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3468 handled += n;
3469 addr += n;
3470 len -= n;
3471 v += n;
3472 } while (len);
3473
3474 return handled;
3475}
3476
3477static void kvm_set_segment(struct kvm_vcpu *vcpu,
3478 struct kvm_segment *var, int seg)
3479{
3480 kvm_x86_ops->set_segment(vcpu, var, seg);
3481}
3482
3483void kvm_get_segment(struct kvm_vcpu *vcpu,
3484 struct kvm_segment *var, int seg)
3485{
3486 kvm_x86_ops->get_segment(vcpu, var, seg);
3487}
3488
3489gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3490{
3491 gpa_t t_gpa;
3492 struct x86_exception exception;
3493
3494 BUG_ON(!mmu_is_nested(vcpu));
3495
3496 /* NPT walks are always user-walks */
3497 access |= PFERR_USER_MASK;
3498 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3499
3500 return t_gpa;
3501}
3502
3503gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3504 struct x86_exception *exception)
3505{
3506 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3507 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3508}
3509
3510 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3511 struct x86_exception *exception)
3512{
3513 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3514 access |= PFERR_FETCH_MASK;
3515 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3516}
3517
3518gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3519 struct x86_exception *exception)
3520{
3521 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3522 access |= PFERR_WRITE_MASK;
3523 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3524}
3525
3526/* uses this to access any guest's mapped memory without checking CPL */
3527gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3528 struct x86_exception *exception)
3529{
3530 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3531}
3532
3533static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3534 struct kvm_vcpu *vcpu, u32 access,
3535 struct x86_exception *exception)
3536{
3537 void *data = val;
3538 int r = X86EMUL_CONTINUE;
3539
3540 while (bytes) {
3541 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3542 exception);
3543 unsigned offset = addr & (PAGE_SIZE-1);
3544 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3545 int ret;
3546
3547 if (gpa == UNMAPPED_GVA)
3548 return X86EMUL_PROPAGATE_FAULT;
3549 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3550 if (ret < 0) {
3551 r = X86EMUL_IO_NEEDED;
3552 goto out;
3553 }
3554
3555 bytes -= toread;
3556 data += toread;
3557 addr += toread;
3558 }
3559out:
3560 return r;
3561}
3562
3563/* used for instruction fetching */
3564static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3565 gva_t addr, void *val, unsigned int bytes,
3566 struct x86_exception *exception)
3567{
3568 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3569 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3570
3571 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3572 access | PFERR_FETCH_MASK,
3573 exception);
3574}
3575
3576int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3577 gva_t addr, void *val, unsigned int bytes,
3578 struct x86_exception *exception)
3579{
3580 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3581 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3582
3583 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3584 exception);
3585}
3586EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3587
3588static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3589 gva_t addr, void *val, unsigned int bytes,
3590 struct x86_exception *exception)
3591{
3592 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3593 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3594}
3595
3596int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3597 gva_t addr, void *val,
3598 unsigned int bytes,
3599 struct x86_exception *exception)
3600{
3601 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3602 void *data = val;
3603 int r = X86EMUL_CONTINUE;
3604
3605 while (bytes) {
3606 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3607 PFERR_WRITE_MASK,
3608 exception);
3609 unsigned offset = addr & (PAGE_SIZE-1);
3610 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3611 int ret;
3612
3613 if (gpa == UNMAPPED_GVA)
3614 return X86EMUL_PROPAGATE_FAULT;
3615 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3616 if (ret < 0) {
3617 r = X86EMUL_IO_NEEDED;
3618 goto out;
3619 }
3620
3621 bytes -= towrite;
3622 data += towrite;
3623 addr += towrite;
3624 }
3625out:
3626 return r;
3627}
3628EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
3629
3630static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
3631 gpa_t *gpa, struct x86_exception *exception,
3632 bool write)
3633{
3634 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3635
3636 if (vcpu_match_mmio_gva(vcpu, gva) &&
3637 check_write_user_access(vcpu, write, access,
3638 vcpu->arch.access)) {
3639 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
3640 (gva & (PAGE_SIZE - 1));
3641 trace_vcpu_match_mmio(gva, *gpa, write, false);
3642 return 1;
3643 }
3644
3645 if (write)
3646 access |= PFERR_WRITE_MASK;
3647
3648 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3649
3650 if (*gpa == UNMAPPED_GVA)
3651 return -1;
3652
3653 /* For APIC access vmexit */
3654 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3655 return 1;
3656
3657 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
3658 trace_vcpu_match_mmio(gva, *gpa, write, true);
3659 return 1;
3660 }
3661
3662 return 0;
3663}
3664
3665int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3666 const void *val, int bytes)
3667{
3668 int ret;
3669
3670 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3671 if (ret < 0)
3672 return 0;
3673 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
3674 return 1;
3675}
3676
3677struct read_write_emulator_ops {
3678 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
3679 int bytes);
3680 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
3681 void *val, int bytes);
3682 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
3683 int bytes, void *val);
3684 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
3685 void *val, int bytes);
3686 bool write;
3687};
3688
3689static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
3690{
3691 if (vcpu->mmio_read_completed) {
3692 memcpy(val, vcpu->mmio_data, bytes);
3693 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3694 vcpu->mmio_phys_addr, *(u64 *)val);
3695 vcpu->mmio_read_completed = 0;
3696 return 1;
3697 }
3698
3699 return 0;
3700}
3701
3702static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
3703 void *val, int bytes)
3704{
3705 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
3706}
3707
3708static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
3709 void *val, int bytes)
3710{
3711 return emulator_write_phys(vcpu, gpa, val, bytes);
3712}
3713
3714static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
3715{
3716 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3717 return vcpu_mmio_write(vcpu, gpa, bytes, val);
3718}
3719
3720static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
3721 void *val, int bytes)
3722{
3723 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3724 return X86EMUL_IO_NEEDED;
3725}
3726
3727static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
3728 void *val, int bytes)
3729{
3730 memcpy(vcpu->mmio_data, val, bytes);
3731 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3732 return X86EMUL_CONTINUE;
3733}
3734
3735static struct read_write_emulator_ops read_emultor = {
3736 .read_write_prepare = read_prepare,
3737 .read_write_emulate = read_emulate,
3738 .read_write_mmio = vcpu_mmio_read,
3739 .read_write_exit_mmio = read_exit_mmio,
3740};
3741
3742static struct read_write_emulator_ops write_emultor = {
3743 .read_write_emulate = write_emulate,
3744 .read_write_mmio = write_mmio,
3745 .read_write_exit_mmio = write_exit_mmio,
3746 .write = true,
3747};
3748
3749static int emulator_read_write_onepage(unsigned long addr, void *val,
3750 unsigned int bytes,
3751 struct x86_exception *exception,
3752 struct kvm_vcpu *vcpu,
3753 struct read_write_emulator_ops *ops)
3754{
3755 gpa_t gpa;
3756 int handled, ret;
3757 bool write = ops->write;
3758
3759 if (ops->read_write_prepare &&
3760 ops->read_write_prepare(vcpu, val, bytes))
3761 return X86EMUL_CONTINUE;
3762
3763 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
3764
3765 if (ret < 0)
3766 return X86EMUL_PROPAGATE_FAULT;
3767
3768 /* For APIC access vmexit */
3769 if (ret)
3770 goto mmio;
3771
3772 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
3773 return X86EMUL_CONTINUE;
3774
3775mmio:
3776 /*
3777 * Is this MMIO handled locally?
3778 */
3779 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
3780 if (handled == bytes)
3781 return X86EMUL_CONTINUE;
3782
3783 gpa += handled;
3784 bytes -= handled;
3785 val += handled;
3786
3787 vcpu->mmio_needed = 1;
3788 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3789 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3790 vcpu->mmio_size = bytes;
3791 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3792 vcpu->run->mmio.is_write = vcpu->mmio_is_write = write;
3793 vcpu->mmio_index = 0;
3794
3795 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
3796}
3797
3798int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
3799 void *val, unsigned int bytes,
3800 struct x86_exception *exception,
3801 struct read_write_emulator_ops *ops)
3802{
3803 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3804
3805 /* Crossing a page boundary? */
3806 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3807 int rc, now;
3808
3809 now = -addr & ~PAGE_MASK;
3810 rc = emulator_read_write_onepage(addr, val, now, exception,
3811 vcpu, ops);
3812
3813 if (rc != X86EMUL_CONTINUE)
3814 return rc;
3815 addr += now;
3816 val += now;
3817 bytes -= now;
3818 }
3819
3820 return emulator_read_write_onepage(addr, val, bytes, exception,
3821 vcpu, ops);
3822}
3823
3824static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
3825 unsigned long addr,
3826 void *val,
3827 unsigned int bytes,
3828 struct x86_exception *exception)
3829{
3830 return emulator_read_write(ctxt, addr, val, bytes,
3831 exception, &read_emultor);
3832}
3833
3834int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
3835 unsigned long addr,
3836 const void *val,
3837 unsigned int bytes,
3838 struct x86_exception *exception)
3839{
3840 return emulator_read_write(ctxt, addr, (void *)val, bytes,
3841 exception, &write_emultor);
3842}
3843
3844#define CMPXCHG_TYPE(t, ptr, old, new) \
3845 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3846
3847#ifdef CONFIG_X86_64
3848# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3849#else
3850# define CMPXCHG64(ptr, old, new) \
3851 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3852#endif
3853
3854static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
3855 unsigned long addr,
3856 const void *old,
3857 const void *new,
3858 unsigned int bytes,
3859 struct x86_exception *exception)
3860{
3861 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3862 gpa_t gpa;
3863 struct page *page;
3864 char *kaddr;
3865 bool exchanged;
3866
3867 /* guests cmpxchg8b have to be emulated atomically */
3868 if (bytes > 8 || (bytes & (bytes - 1)))
3869 goto emul_write;
3870
3871 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3872
3873 if (gpa == UNMAPPED_GVA ||
3874 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3875 goto emul_write;
3876
3877 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3878 goto emul_write;
3879
3880 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3881 if (is_error_page(page)) {
3882 kvm_release_page_clean(page);
3883 goto emul_write;
3884 }
3885
3886 kaddr = kmap_atomic(page, KM_USER0);
3887 kaddr += offset_in_page(gpa);
3888 switch (bytes) {
3889 case 1:
3890 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3891 break;
3892 case 2:
3893 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3894 break;
3895 case 4:
3896 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3897 break;
3898 case 8:
3899 exchanged = CMPXCHG64(kaddr, old, new);
3900 break;
3901 default:
3902 BUG();
3903 }
3904 kunmap_atomic(kaddr, KM_USER0);
3905 kvm_release_page_dirty(page);
3906
3907 if (!exchanged)
3908 return X86EMUL_CMPXCHG_FAILED;
3909
3910 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
3911
3912 return X86EMUL_CONTINUE;
3913
3914emul_write:
3915 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3916
3917 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
3918}
3919
3920static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3921{
3922 /* TODO: String I/O for in kernel device */
3923 int r;
3924
3925 if (vcpu->arch.pio.in)
3926 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3927 vcpu->arch.pio.size, pd);
3928 else
3929 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3930 vcpu->arch.pio.port, vcpu->arch.pio.size,
3931 pd);
3932 return r;
3933}
3934
3935static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
3936 unsigned short port, void *val,
3937 unsigned int count, bool in)
3938{
3939 trace_kvm_pio(!in, port, size, count);
3940
3941 vcpu->arch.pio.port = port;
3942 vcpu->arch.pio.in = in;
3943 vcpu->arch.pio.count = count;
3944 vcpu->arch.pio.size = size;
3945
3946 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3947 vcpu->arch.pio.count = 0;
3948 return 1;
3949 }
3950
3951 vcpu->run->exit_reason = KVM_EXIT_IO;
3952 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3953 vcpu->run->io.size = size;
3954 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3955 vcpu->run->io.count = count;
3956 vcpu->run->io.port = port;
3957
3958 return 0;
3959}
3960
3961static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
3962 int size, unsigned short port, void *val,
3963 unsigned int count)
3964{
3965 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3966 int ret;
3967
3968 if (vcpu->arch.pio.count)
3969 goto data_avail;
3970
3971 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
3972 if (ret) {
3973data_avail:
3974 memcpy(val, vcpu->arch.pio_data, size * count);
3975 vcpu->arch.pio.count = 0;
3976 return 1;
3977 }
3978
3979 return 0;
3980}
3981
3982static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
3983 int size, unsigned short port,
3984 const void *val, unsigned int count)
3985{
3986 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3987
3988 memcpy(vcpu->arch.pio_data, val, size * count);
3989 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
3990}
3991
3992static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3993{
3994 return kvm_x86_ops->get_segment_base(vcpu, seg);
3995}
3996
3997static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
3998{
3999 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4000}
4001
4002int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4003{
4004 if (!need_emulate_wbinvd(vcpu))
4005 return X86EMUL_CONTINUE;
4006
4007 if (kvm_x86_ops->has_wbinvd_exit()) {
4008 int cpu = get_cpu();
4009
4010 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4011 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4012 wbinvd_ipi, NULL, 1);
4013 put_cpu();
4014 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4015 } else
4016 wbinvd();
4017 return X86EMUL_CONTINUE;
4018}
4019EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4020
4021static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4022{
4023 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4024}
4025
4026int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4027{
4028 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4029}
4030
4031int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4032{
4033
4034 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4035}
4036
4037static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4038{
4039 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4040}
4041
4042static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4043{
4044 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4045 unsigned long value;
4046
4047 switch (cr) {
4048 case 0:
4049 value = kvm_read_cr0(vcpu);
4050 break;
4051 case 2:
4052 value = vcpu->arch.cr2;
4053 break;
4054 case 3:
4055 value = kvm_read_cr3(vcpu);
4056 break;
4057 case 4:
4058 value = kvm_read_cr4(vcpu);
4059 break;
4060 case 8:
4061 value = kvm_get_cr8(vcpu);
4062 break;
4063 default:
4064 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4065 return 0;
4066 }
4067
4068 return value;
4069}
4070
4071static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4072{
4073 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4074 int res = 0;
4075
4076 switch (cr) {
4077 case 0:
4078 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4079 break;
4080 case 2:
4081 vcpu->arch.cr2 = val;
4082 break;
4083 case 3:
4084 res = kvm_set_cr3(vcpu, val);
4085 break;
4086 case 4:
4087 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4088 break;
4089 case 8:
4090 res = kvm_set_cr8(vcpu, val);
4091 break;
4092 default:
4093 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4094 res = -1;
4095 }
4096
4097 return res;
4098}
4099
4100static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4101{
4102 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4103}
4104
4105static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4106{
4107 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4108}
4109
4110static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4111{
4112 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4113}
4114
4115static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4116{
4117 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4118}
4119
4120static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4121{
4122 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4123}
4124
4125static unsigned long emulator_get_cached_segment_base(
4126 struct x86_emulate_ctxt *ctxt, int seg)
4127{
4128 return get_segment_base(emul_to_vcpu(ctxt), seg);
4129}
4130
4131static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4132 struct desc_struct *desc, u32 *base3,
4133 int seg)
4134{
4135 struct kvm_segment var;
4136
4137 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4138 *selector = var.selector;
4139
4140 if (var.unusable)
4141 return false;
4142
4143 if (var.g)
4144 var.limit >>= 12;
4145 set_desc_limit(desc, var.limit);
4146 set_desc_base(desc, (unsigned long)var.base);
4147#ifdef CONFIG_X86_64
4148 if (base3)
4149 *base3 = var.base >> 32;
4150#endif
4151 desc->type = var.type;
4152 desc->s = var.s;
4153 desc->dpl = var.dpl;
4154 desc->p = var.present;
4155 desc->avl = var.avl;
4156 desc->l = var.l;
4157 desc->d = var.db;
4158 desc->g = var.g;
4159
4160 return true;
4161}
4162
4163static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4164 struct desc_struct *desc, u32 base3,
4165 int seg)
4166{
4167 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4168 struct kvm_segment var;
4169
4170 var.selector = selector;
4171 var.base = get_desc_base(desc);
4172#ifdef CONFIG_X86_64
4173 var.base |= ((u64)base3) << 32;
4174#endif
4175 var.limit = get_desc_limit(desc);
4176 if (desc->g)
4177 var.limit = (var.limit << 12) | 0xfff;
4178 var.type = desc->type;
4179 var.present = desc->p;
4180 var.dpl = desc->dpl;
4181 var.db = desc->d;
4182 var.s = desc->s;
4183 var.l = desc->l;
4184 var.g = desc->g;
4185 var.avl = desc->avl;
4186 var.present = desc->p;
4187 var.unusable = !var.present;
4188 var.padding = 0;
4189
4190 kvm_set_segment(vcpu, &var, seg);
4191 return;
4192}
4193
4194static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4195 u32 msr_index, u64 *pdata)
4196{
4197 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4198}
4199
4200static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4201 u32 msr_index, u64 data)
4202{
4203 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4204}
4205
4206static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4207 u32 pmc, u64 *pdata)
4208{
4209 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4210}
4211
4212static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4213{
4214 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4215}
4216
4217static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4218{
4219 preempt_disable();
4220 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4221 /*
4222 * CR0.TS may reference the host fpu state, not the guest fpu state,
4223 * so it may be clear at this point.
4224 */
4225 clts();
4226}
4227
4228static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4229{
4230 preempt_enable();
4231}
4232
4233static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4234 struct x86_instruction_info *info,
4235 enum x86_intercept_stage stage)
4236{
4237 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4238}
4239
4240static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4241 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4242{
4243 struct kvm_cpuid_entry2 *cpuid = NULL;
4244
4245 if (eax && ecx)
4246 cpuid = kvm_find_cpuid_entry(emul_to_vcpu(ctxt),
4247 *eax, *ecx);
4248
4249 if (cpuid) {
4250 *eax = cpuid->eax;
4251 *ecx = cpuid->ecx;
4252 if (ebx)
4253 *ebx = cpuid->ebx;
4254 if (edx)
4255 *edx = cpuid->edx;
4256 return true;
4257 }
4258
4259 return false;
4260}
4261
4262static struct x86_emulate_ops emulate_ops = {
4263 .read_std = kvm_read_guest_virt_system,
4264 .write_std = kvm_write_guest_virt_system,
4265 .fetch = kvm_fetch_guest_virt,
4266 .read_emulated = emulator_read_emulated,
4267 .write_emulated = emulator_write_emulated,
4268 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4269 .invlpg = emulator_invlpg,
4270 .pio_in_emulated = emulator_pio_in_emulated,
4271 .pio_out_emulated = emulator_pio_out_emulated,
4272 .get_segment = emulator_get_segment,
4273 .set_segment = emulator_set_segment,
4274 .get_cached_segment_base = emulator_get_cached_segment_base,
4275 .get_gdt = emulator_get_gdt,
4276 .get_idt = emulator_get_idt,
4277 .set_gdt = emulator_set_gdt,
4278 .set_idt = emulator_set_idt,
4279 .get_cr = emulator_get_cr,
4280 .set_cr = emulator_set_cr,
4281 .cpl = emulator_get_cpl,
4282 .get_dr = emulator_get_dr,
4283 .set_dr = emulator_set_dr,
4284 .set_msr = emulator_set_msr,
4285 .get_msr = emulator_get_msr,
4286 .read_pmc = emulator_read_pmc,
4287 .halt = emulator_halt,
4288 .wbinvd = emulator_wbinvd,
4289 .fix_hypercall = emulator_fix_hypercall,
4290 .get_fpu = emulator_get_fpu,
4291 .put_fpu = emulator_put_fpu,
4292 .intercept = emulator_intercept,
4293 .get_cpuid = emulator_get_cpuid,
4294};
4295
4296static void cache_all_regs(struct kvm_vcpu *vcpu)
4297{
4298 kvm_register_read(vcpu, VCPU_REGS_RAX);
4299 kvm_register_read(vcpu, VCPU_REGS_RSP);
4300 kvm_register_read(vcpu, VCPU_REGS_RIP);
4301 vcpu->arch.regs_dirty = ~0;
4302}
4303
4304static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4305{
4306 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4307 /*
4308 * an sti; sti; sequence only disable interrupts for the first
4309 * instruction. So, if the last instruction, be it emulated or
4310 * not, left the system with the INT_STI flag enabled, it
4311 * means that the last instruction is an sti. We should not
4312 * leave the flag on in this case. The same goes for mov ss
4313 */
4314 if (!(int_shadow & mask))
4315 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4316}
4317
4318static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4319{
4320 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4321 if (ctxt->exception.vector == PF_VECTOR)
4322 kvm_propagate_fault(vcpu, &ctxt->exception);
4323 else if (ctxt->exception.error_code_valid)
4324 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4325 ctxt->exception.error_code);
4326 else
4327 kvm_queue_exception(vcpu, ctxt->exception.vector);
4328}
4329
4330static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
4331 const unsigned long *regs)
4332{
4333 memset(&ctxt->twobyte, 0,
4334 (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4335 memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
4336
4337 ctxt->fetch.start = 0;
4338 ctxt->fetch.end = 0;
4339 ctxt->io_read.pos = 0;
4340 ctxt->io_read.end = 0;
4341 ctxt->mem_read.pos = 0;
4342 ctxt->mem_read.end = 0;
4343}
4344
4345static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4346{
4347 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4348 int cs_db, cs_l;
4349
4350 /*
4351 * TODO: fix emulate.c to use guest_read/write_register
4352 * instead of direct ->regs accesses, can save hundred cycles
4353 * on Intel for instructions that don't read/change RSP, for
4354 * for example.
4355 */
4356 cache_all_regs(vcpu);
4357
4358 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4359
4360 ctxt->eflags = kvm_get_rflags(vcpu);
4361 ctxt->eip = kvm_rip_read(vcpu);
4362 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4363 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4364 cs_l ? X86EMUL_MODE_PROT64 :
4365 cs_db ? X86EMUL_MODE_PROT32 :
4366 X86EMUL_MODE_PROT16;
4367 ctxt->guest_mode = is_guest_mode(vcpu);
4368
4369 init_decode_cache(ctxt, vcpu->arch.regs);
4370 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4371}
4372
4373int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4374{
4375 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4376 int ret;
4377
4378 init_emulate_ctxt(vcpu);
4379
4380 ctxt->op_bytes = 2;
4381 ctxt->ad_bytes = 2;
4382 ctxt->_eip = ctxt->eip + inc_eip;
4383 ret = emulate_int_real(ctxt, irq);
4384
4385 if (ret != X86EMUL_CONTINUE)
4386 return EMULATE_FAIL;
4387
4388 ctxt->eip = ctxt->_eip;
4389 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4390 kvm_rip_write(vcpu, ctxt->eip);
4391 kvm_set_rflags(vcpu, ctxt->eflags);
4392
4393 if (irq == NMI_VECTOR)
4394 vcpu->arch.nmi_pending = 0;
4395 else
4396 vcpu->arch.interrupt.pending = false;
4397
4398 return EMULATE_DONE;
4399}
4400EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4401
4402static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4403{
4404 int r = EMULATE_DONE;
4405
4406 ++vcpu->stat.insn_emulation_fail;
4407 trace_kvm_emulate_insn_failed(vcpu);
4408 if (!is_guest_mode(vcpu)) {
4409 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4410 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4411 vcpu->run->internal.ndata = 0;
4412 r = EMULATE_FAIL;
4413 }
4414 kvm_queue_exception(vcpu, UD_VECTOR);
4415
4416 return r;
4417}
4418
4419static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4420{
4421 gpa_t gpa;
4422
4423 if (tdp_enabled)
4424 return false;
4425
4426 /*
4427 * if emulation was due to access to shadowed page table
4428 * and it failed try to unshadow page and re-entetr the
4429 * guest to let CPU execute the instruction.
4430 */
4431 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4432 return true;
4433
4434 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4435
4436 if (gpa == UNMAPPED_GVA)
4437 return true; /* let cpu generate fault */
4438
4439 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4440 return true;
4441
4442 return false;
4443}
4444
4445static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4446 unsigned long cr2, int emulation_type)
4447{
4448 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4449 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4450
4451 last_retry_eip = vcpu->arch.last_retry_eip;
4452 last_retry_addr = vcpu->arch.last_retry_addr;
4453
4454 /*
4455 * If the emulation is caused by #PF and it is non-page_table
4456 * writing instruction, it means the VM-EXIT is caused by shadow
4457 * page protected, we can zap the shadow page and retry this
4458 * instruction directly.
4459 *
4460 * Note: if the guest uses a non-page-table modifying instruction
4461 * on the PDE that points to the instruction, then we will unmap
4462 * the instruction and go to an infinite loop. So, we cache the
4463 * last retried eip and the last fault address, if we meet the eip
4464 * and the address again, we can break out of the potential infinite
4465 * loop.
4466 */
4467 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4468
4469 if (!(emulation_type & EMULTYPE_RETRY))
4470 return false;
4471
4472 if (x86_page_table_writing_insn(ctxt))
4473 return false;
4474
4475 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4476 return false;
4477
4478 vcpu->arch.last_retry_eip = ctxt->eip;
4479 vcpu->arch.last_retry_addr = cr2;
4480
4481 if (!vcpu->arch.mmu.direct_map)
4482 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4483
4484 kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4485
4486 return true;
4487}
4488
4489int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4490 unsigned long cr2,
4491 int emulation_type,
4492 void *insn,
4493 int insn_len)
4494{
4495 int r;
4496 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4497 bool writeback = true;
4498
4499 kvm_clear_exception_queue(vcpu);
4500
4501 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4502 init_emulate_ctxt(vcpu);
4503 ctxt->interruptibility = 0;
4504 ctxt->have_exception = false;
4505 ctxt->perm_ok = false;
4506
4507 ctxt->only_vendor_specific_insn
4508 = emulation_type & EMULTYPE_TRAP_UD;
4509
4510 r = x86_decode_insn(ctxt, insn, insn_len);
4511
4512 trace_kvm_emulate_insn_start(vcpu);
4513 ++vcpu->stat.insn_emulation;
4514 if (r != EMULATION_OK) {
4515 if (emulation_type & EMULTYPE_TRAP_UD)
4516 return EMULATE_FAIL;
4517 if (reexecute_instruction(vcpu, cr2))
4518 return EMULATE_DONE;
4519 if (emulation_type & EMULTYPE_SKIP)
4520 return EMULATE_FAIL;
4521 return handle_emulation_failure(vcpu);
4522 }
4523 }
4524
4525 if (emulation_type & EMULTYPE_SKIP) {
4526 kvm_rip_write(vcpu, ctxt->_eip);
4527 return EMULATE_DONE;
4528 }
4529
4530 if (retry_instruction(ctxt, cr2, emulation_type))
4531 return EMULATE_DONE;
4532
4533 /* this is needed for vmware backdoor interface to work since it
4534 changes registers values during IO operation */
4535 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4536 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4537 memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
4538 }
4539
4540restart:
4541 r = x86_emulate_insn(ctxt);
4542
4543 if (r == EMULATION_INTERCEPTED)
4544 return EMULATE_DONE;
4545
4546 if (r == EMULATION_FAILED) {
4547 if (reexecute_instruction(vcpu, cr2))
4548 return EMULATE_DONE;
4549
4550 return handle_emulation_failure(vcpu);
4551 }
4552
4553 if (ctxt->have_exception) {
4554 inject_emulated_exception(vcpu);
4555 r = EMULATE_DONE;
4556 } else if (vcpu->arch.pio.count) {
4557 if (!vcpu->arch.pio.in)
4558 vcpu->arch.pio.count = 0;
4559 else
4560 writeback = false;
4561 r = EMULATE_DO_MMIO;
4562 } else if (vcpu->mmio_needed) {
4563 if (!vcpu->mmio_is_write)
4564 writeback = false;
4565 r = EMULATE_DO_MMIO;
4566 } else if (r == EMULATION_RESTART)
4567 goto restart;
4568 else
4569 r = EMULATE_DONE;
4570
4571 if (writeback) {
4572 toggle_interruptibility(vcpu, ctxt->interruptibility);
4573 kvm_set_rflags(vcpu, ctxt->eflags);
4574 kvm_make_request(KVM_REQ_EVENT, vcpu);
4575 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4576 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4577 kvm_rip_write(vcpu, ctxt->eip);
4578 } else
4579 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4580
4581 return r;
4582}
4583EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4584
4585int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4586{
4587 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4588 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4589 size, port, &val, 1);
4590 /* do not return to emulator after return from userspace */
4591 vcpu->arch.pio.count = 0;
4592 return ret;
4593}
4594EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4595
4596static void tsc_bad(void *info)
4597{
4598 __this_cpu_write(cpu_tsc_khz, 0);
4599}
4600
4601static void tsc_khz_changed(void *data)
4602{
4603 struct cpufreq_freqs *freq = data;
4604 unsigned long khz = 0;
4605
4606 if (data)
4607 khz = freq->new;
4608 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4609 khz = cpufreq_quick_get(raw_smp_processor_id());
4610 if (!khz)
4611 khz = tsc_khz;
4612 __this_cpu_write(cpu_tsc_khz, khz);
4613}
4614
4615static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4616 void *data)
4617{
4618 struct cpufreq_freqs *freq = data;
4619 struct kvm *kvm;
4620 struct kvm_vcpu *vcpu;
4621 int i, send_ipi = 0;
4622
4623 /*
4624 * We allow guests to temporarily run on slowing clocks,
4625 * provided we notify them after, or to run on accelerating
4626 * clocks, provided we notify them before. Thus time never
4627 * goes backwards.
4628 *
4629 * However, we have a problem. We can't atomically update
4630 * the frequency of a given CPU from this function; it is
4631 * merely a notifier, which can be called from any CPU.
4632 * Changing the TSC frequency at arbitrary points in time
4633 * requires a recomputation of local variables related to
4634 * the TSC for each VCPU. We must flag these local variables
4635 * to be updated and be sure the update takes place with the
4636 * new frequency before any guests proceed.
4637 *
4638 * Unfortunately, the combination of hotplug CPU and frequency
4639 * change creates an intractable locking scenario; the order
4640 * of when these callouts happen is undefined with respect to
4641 * CPU hotplug, and they can race with each other. As such,
4642 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4643 * undefined; you can actually have a CPU frequency change take
4644 * place in between the computation of X and the setting of the
4645 * variable. To protect against this problem, all updates of
4646 * the per_cpu tsc_khz variable are done in an interrupt
4647 * protected IPI, and all callers wishing to update the value
4648 * must wait for a synchronous IPI to complete (which is trivial
4649 * if the caller is on the CPU already). This establishes the
4650 * necessary total order on variable updates.
4651 *
4652 * Note that because a guest time update may take place
4653 * anytime after the setting of the VCPU's request bit, the
4654 * correct TSC value must be set before the request. However,
4655 * to ensure the update actually makes it to any guest which
4656 * starts running in hardware virtualization between the set
4657 * and the acquisition of the spinlock, we must also ping the
4658 * CPU after setting the request bit.
4659 *
4660 */
4661
4662 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4663 return 0;
4664 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4665 return 0;
4666
4667 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4668
4669 raw_spin_lock(&kvm_lock);
4670 list_for_each_entry(kvm, &vm_list, vm_list) {
4671 kvm_for_each_vcpu(i, vcpu, kvm) {
4672 if (vcpu->cpu != freq->cpu)
4673 continue;
4674 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4675 if (vcpu->cpu != smp_processor_id())
4676 send_ipi = 1;
4677 }
4678 }
4679 raw_spin_unlock(&kvm_lock);
4680
4681 if (freq->old < freq->new && send_ipi) {
4682 /*
4683 * We upscale the frequency. Must make the guest
4684 * doesn't see old kvmclock values while running with
4685 * the new frequency, otherwise we risk the guest sees
4686 * time go backwards.
4687 *
4688 * In case we update the frequency for another cpu
4689 * (which might be in guest context) send an interrupt
4690 * to kick the cpu out of guest context. Next time
4691 * guest context is entered kvmclock will be updated,
4692 * so the guest will not see stale values.
4693 */
4694 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4695 }
4696 return 0;
4697}
4698
4699static struct notifier_block kvmclock_cpufreq_notifier_block = {
4700 .notifier_call = kvmclock_cpufreq_notifier
4701};
4702
4703static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4704 unsigned long action, void *hcpu)
4705{
4706 unsigned int cpu = (unsigned long)hcpu;
4707
4708 switch (action) {
4709 case CPU_ONLINE:
4710 case CPU_DOWN_FAILED:
4711 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4712 break;
4713 case CPU_DOWN_PREPARE:
4714 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4715 break;
4716 }
4717 return NOTIFY_OK;
4718}
4719
4720static struct notifier_block kvmclock_cpu_notifier_block = {
4721 .notifier_call = kvmclock_cpu_notifier,
4722 .priority = -INT_MAX
4723};
4724
4725static void kvm_timer_init(void)
4726{
4727 int cpu;
4728
4729 max_tsc_khz = tsc_khz;
4730 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4731 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4732#ifdef CONFIG_CPU_FREQ
4733 struct cpufreq_policy policy;
4734 memset(&policy, 0, sizeof(policy));
4735 cpu = get_cpu();
4736 cpufreq_get_policy(&policy, cpu);
4737 if (policy.cpuinfo.max_freq)
4738 max_tsc_khz = policy.cpuinfo.max_freq;
4739 put_cpu();
4740#endif
4741 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4742 CPUFREQ_TRANSITION_NOTIFIER);
4743 }
4744 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4745 for_each_online_cpu(cpu)
4746 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4747}
4748
4749static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4750
4751int kvm_is_in_guest(void)
4752{
4753 return __this_cpu_read(current_vcpu) != NULL;
4754}
4755
4756static int kvm_is_user_mode(void)
4757{
4758 int user_mode = 3;
4759
4760 if (__this_cpu_read(current_vcpu))
4761 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
4762
4763 return user_mode != 0;
4764}
4765
4766static unsigned long kvm_get_guest_ip(void)
4767{
4768 unsigned long ip = 0;
4769
4770 if (__this_cpu_read(current_vcpu))
4771 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
4772
4773 return ip;
4774}
4775
4776static struct perf_guest_info_callbacks kvm_guest_cbs = {
4777 .is_in_guest = kvm_is_in_guest,
4778 .is_user_mode = kvm_is_user_mode,
4779 .get_guest_ip = kvm_get_guest_ip,
4780};
4781
4782void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4783{
4784 __this_cpu_write(current_vcpu, vcpu);
4785}
4786EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4787
4788void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4789{
4790 __this_cpu_write(current_vcpu, NULL);
4791}
4792EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4793
4794static void kvm_set_mmio_spte_mask(void)
4795{
4796 u64 mask;
4797 int maxphyaddr = boot_cpu_data.x86_phys_bits;
4798
4799 /*
4800 * Set the reserved bits and the present bit of an paging-structure
4801 * entry to generate page fault with PFER.RSV = 1.
4802 */
4803 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
4804 mask |= 1ull;
4805
4806#ifdef CONFIG_X86_64
4807 /*
4808 * If reserved bit is not supported, clear the present bit to disable
4809 * mmio page fault.
4810 */
4811 if (maxphyaddr == 52)
4812 mask &= ~1ull;
4813#endif
4814
4815 kvm_mmu_set_mmio_spte_mask(mask);
4816}
4817
4818int kvm_arch_init(void *opaque)
4819{
4820 int r;
4821 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4822
4823 if (kvm_x86_ops) {
4824 printk(KERN_ERR "kvm: already loaded the other module\n");
4825 r = -EEXIST;
4826 goto out;
4827 }
4828
4829 if (!ops->cpu_has_kvm_support()) {
4830 printk(KERN_ERR "kvm: no hardware support\n");
4831 r = -EOPNOTSUPP;
4832 goto out;
4833 }
4834 if (ops->disabled_by_bios()) {
4835 printk(KERN_ERR "kvm: disabled by bios\n");
4836 r = -EOPNOTSUPP;
4837 goto out;
4838 }
4839
4840 r = kvm_mmu_module_init();
4841 if (r)
4842 goto out;
4843
4844 kvm_set_mmio_spte_mask();
4845 kvm_init_msr_list();
4846
4847 kvm_x86_ops = ops;
4848 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4849 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4850
4851 kvm_timer_init();
4852
4853 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4854
4855 if (cpu_has_xsave)
4856 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4857
4858 return 0;
4859
4860out:
4861 return r;
4862}
4863
4864void kvm_arch_exit(void)
4865{
4866 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4867
4868 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4869 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4870 CPUFREQ_TRANSITION_NOTIFIER);
4871 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4872 kvm_x86_ops = NULL;
4873 kvm_mmu_module_exit();
4874}
4875
4876int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4877{
4878 ++vcpu->stat.halt_exits;
4879 if (irqchip_in_kernel(vcpu->kvm)) {
4880 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4881 return 1;
4882 } else {
4883 vcpu->run->exit_reason = KVM_EXIT_HLT;
4884 return 0;
4885 }
4886}
4887EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4888
4889int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4890{
4891 u64 param, ingpa, outgpa, ret;
4892 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4893 bool fast, longmode;
4894 int cs_db, cs_l;
4895
4896 /*
4897 * hypercall generates UD from non zero cpl and real mode
4898 * per HYPER-V spec
4899 */
4900 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4901 kvm_queue_exception(vcpu, UD_VECTOR);
4902 return 0;
4903 }
4904
4905 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4906 longmode = is_long_mode(vcpu) && cs_l == 1;
4907
4908 if (!longmode) {
4909 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4910 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4911 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4912 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4913 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4914 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4915 }
4916#ifdef CONFIG_X86_64
4917 else {
4918 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4919 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4920 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4921 }
4922#endif
4923
4924 code = param & 0xffff;
4925 fast = (param >> 16) & 0x1;
4926 rep_cnt = (param >> 32) & 0xfff;
4927 rep_idx = (param >> 48) & 0xfff;
4928
4929 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4930
4931 switch (code) {
4932 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4933 kvm_vcpu_on_spin(vcpu);
4934 break;
4935 default:
4936 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4937 break;
4938 }
4939
4940 ret = res | (((u64)rep_done & 0xfff) << 32);
4941 if (longmode) {
4942 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4943 } else {
4944 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4945 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4946 }
4947
4948 return 1;
4949}
4950
4951int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4952{
4953 unsigned long nr, a0, a1, a2, a3, ret;
4954 int r = 1;
4955
4956 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4957 return kvm_hv_hypercall(vcpu);
4958
4959 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4960 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4961 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4962 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4963 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4964
4965 trace_kvm_hypercall(nr, a0, a1, a2, a3);
4966
4967 if (!is_long_mode(vcpu)) {
4968 nr &= 0xFFFFFFFF;
4969 a0 &= 0xFFFFFFFF;
4970 a1 &= 0xFFFFFFFF;
4971 a2 &= 0xFFFFFFFF;
4972 a3 &= 0xFFFFFFFF;
4973 }
4974
4975 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4976 ret = -KVM_EPERM;
4977 goto out;
4978 }
4979
4980 switch (nr) {
4981 case KVM_HC_VAPIC_POLL_IRQ:
4982 ret = 0;
4983 break;
4984 default:
4985 ret = -KVM_ENOSYS;
4986 break;
4987 }
4988out:
4989 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4990 ++vcpu->stat.hypercalls;
4991 return r;
4992}
4993EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4994
4995int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
4996{
4997 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4998 char instruction[3];
4999 unsigned long rip = kvm_rip_read(vcpu);
5000
5001 /*
5002 * Blow out the MMU to ensure that no other VCPU has an active mapping
5003 * to ensure that the updated hypercall appears atomically across all
5004 * VCPUs.
5005 */
5006 kvm_mmu_zap_all(vcpu->kvm);
5007
5008 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5009
5010 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5011}
5012
5013/*
5014 * Check if userspace requested an interrupt window, and that the
5015 * interrupt window is open.
5016 *
5017 * No need to exit to userspace if we already have an interrupt queued.
5018 */
5019static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5020{
5021 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5022 vcpu->run->request_interrupt_window &&
5023 kvm_arch_interrupt_allowed(vcpu));
5024}
5025
5026static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5027{
5028 struct kvm_run *kvm_run = vcpu->run;
5029
5030 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5031 kvm_run->cr8 = kvm_get_cr8(vcpu);
5032 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5033 if (irqchip_in_kernel(vcpu->kvm))
5034 kvm_run->ready_for_interrupt_injection = 1;
5035 else
5036 kvm_run->ready_for_interrupt_injection =
5037 kvm_arch_interrupt_allowed(vcpu) &&
5038 !kvm_cpu_has_interrupt(vcpu) &&
5039 !kvm_event_needs_reinjection(vcpu);
5040}
5041
5042static void vapic_enter(struct kvm_vcpu *vcpu)
5043{
5044 struct kvm_lapic *apic = vcpu->arch.apic;
5045 struct page *page;
5046
5047 if (!apic || !apic->vapic_addr)
5048 return;
5049
5050 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5051
5052 vcpu->arch.apic->vapic_page = page;
5053}
5054
5055static void vapic_exit(struct kvm_vcpu *vcpu)
5056{
5057 struct kvm_lapic *apic = vcpu->arch.apic;
5058 int idx;
5059
5060 if (!apic || !apic->vapic_addr)
5061 return;
5062
5063 idx = srcu_read_lock(&vcpu->kvm->srcu);
5064 kvm_release_page_dirty(apic->vapic_page);
5065 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5066 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5067}
5068
5069static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5070{
5071 int max_irr, tpr;
5072
5073 if (!kvm_x86_ops->update_cr8_intercept)
5074 return;
5075
5076 if (!vcpu->arch.apic)
5077 return;
5078
5079 if (!vcpu->arch.apic->vapic_addr)
5080 max_irr = kvm_lapic_find_highest_irr(vcpu);
5081 else
5082 max_irr = -1;
5083
5084 if (max_irr != -1)
5085 max_irr >>= 4;
5086
5087 tpr = kvm_lapic_get_cr8(vcpu);
5088
5089 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5090}
5091
5092static void inject_pending_event(struct kvm_vcpu *vcpu)
5093{
5094 /* try to reinject previous events if any */
5095 if (vcpu->arch.exception.pending) {
5096 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5097 vcpu->arch.exception.has_error_code,
5098 vcpu->arch.exception.error_code);
5099 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5100 vcpu->arch.exception.has_error_code,
5101 vcpu->arch.exception.error_code,
5102 vcpu->arch.exception.reinject);
5103 return;
5104 }
5105
5106 if (vcpu->arch.nmi_injected) {
5107 kvm_x86_ops->set_nmi(vcpu);
5108 return;
5109 }
5110
5111 if (vcpu->arch.interrupt.pending) {
5112 kvm_x86_ops->set_irq(vcpu);
5113 return;
5114 }
5115
5116 /* try to inject new event if pending */
5117 if (vcpu->arch.nmi_pending) {
5118 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5119 --vcpu->arch.nmi_pending;
5120 vcpu->arch.nmi_injected = true;
5121 kvm_x86_ops->set_nmi(vcpu);
5122 }
5123 } else if (kvm_cpu_has_interrupt(vcpu)) {
5124 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5125 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5126 false);
5127 kvm_x86_ops->set_irq(vcpu);
5128 }
5129 }
5130}
5131
5132static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5133{
5134 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5135 !vcpu->guest_xcr0_loaded) {
5136 /* kvm_set_xcr() also depends on this */
5137 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5138 vcpu->guest_xcr0_loaded = 1;
5139 }
5140}
5141
5142static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5143{
5144 if (vcpu->guest_xcr0_loaded) {
5145 if (vcpu->arch.xcr0 != host_xcr0)
5146 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5147 vcpu->guest_xcr0_loaded = 0;
5148 }
5149}
5150
5151static void process_nmi(struct kvm_vcpu *vcpu)
5152{
5153 unsigned limit = 2;
5154
5155 /*
5156 * x86 is limited to one NMI running, and one NMI pending after it.
5157 * If an NMI is already in progress, limit further NMIs to just one.
5158 * Otherwise, allow two (and we'll inject the first one immediately).
5159 */
5160 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5161 limit = 1;
5162
5163 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5164 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5165 kvm_make_request(KVM_REQ_EVENT, vcpu);
5166}
5167
5168static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5169{
5170 int r;
5171 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5172 vcpu->run->request_interrupt_window;
5173 bool req_immediate_exit = 0;
5174
5175 if (vcpu->requests) {
5176 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5177 kvm_mmu_unload(vcpu);
5178 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5179 __kvm_migrate_timers(vcpu);
5180 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5181 r = kvm_guest_time_update(vcpu);
5182 if (unlikely(r))
5183 goto out;
5184 }
5185 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5186 kvm_mmu_sync_roots(vcpu);
5187 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5188 kvm_x86_ops->tlb_flush(vcpu);
5189 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5190 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5191 r = 0;
5192 goto out;
5193 }
5194 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5195 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5196 r = 0;
5197 goto out;
5198 }
5199 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5200 vcpu->fpu_active = 0;
5201 kvm_x86_ops->fpu_deactivate(vcpu);
5202 }
5203 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5204 /* Page is swapped out. Do synthetic halt */
5205 vcpu->arch.apf.halted = true;
5206 r = 1;
5207 goto out;
5208 }
5209 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5210 record_steal_time(vcpu);
5211 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5212 process_nmi(vcpu);
5213 req_immediate_exit =
5214 kvm_check_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
5215 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5216 kvm_handle_pmu_event(vcpu);
5217 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5218 kvm_deliver_pmi(vcpu);
5219 }
5220
5221 r = kvm_mmu_reload(vcpu);
5222 if (unlikely(r))
5223 goto out;
5224
5225 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5226 inject_pending_event(vcpu);
5227
5228 /* enable NMI/IRQ window open exits if needed */
5229 if (vcpu->arch.nmi_pending)
5230 kvm_x86_ops->enable_nmi_window(vcpu);
5231 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5232 kvm_x86_ops->enable_irq_window(vcpu);
5233
5234 if (kvm_lapic_enabled(vcpu)) {
5235 update_cr8_intercept(vcpu);
5236 kvm_lapic_sync_to_vapic(vcpu);
5237 }
5238 }
5239
5240 preempt_disable();
5241
5242 kvm_x86_ops->prepare_guest_switch(vcpu);
5243 if (vcpu->fpu_active)
5244 kvm_load_guest_fpu(vcpu);
5245 kvm_load_guest_xcr0(vcpu);
5246
5247 vcpu->mode = IN_GUEST_MODE;
5248
5249 /* We should set ->mode before check ->requests,
5250 * see the comment in make_all_cpus_request.
5251 */
5252 smp_mb();
5253
5254 local_irq_disable();
5255
5256 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5257 || need_resched() || signal_pending(current)) {
5258 vcpu->mode = OUTSIDE_GUEST_MODE;
5259 smp_wmb();
5260 local_irq_enable();
5261 preempt_enable();
5262 kvm_x86_ops->cancel_injection(vcpu);
5263 r = 1;
5264 goto out;
5265 }
5266
5267 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5268
5269 if (req_immediate_exit)
5270 smp_send_reschedule(vcpu->cpu);
5271
5272 kvm_guest_enter();
5273
5274 if (unlikely(vcpu->arch.switch_db_regs)) {
5275 set_debugreg(0, 7);
5276 set_debugreg(vcpu->arch.eff_db[0], 0);
5277 set_debugreg(vcpu->arch.eff_db[1], 1);
5278 set_debugreg(vcpu->arch.eff_db[2], 2);
5279 set_debugreg(vcpu->arch.eff_db[3], 3);
5280 }
5281
5282 trace_kvm_entry(vcpu->vcpu_id);
5283 kvm_x86_ops->run(vcpu);
5284
5285 /*
5286 * If the guest has used debug registers, at least dr7
5287 * will be disabled while returning to the host.
5288 * If we don't have active breakpoints in the host, we don't
5289 * care about the messed up debug address registers. But if
5290 * we have some of them active, restore the old state.
5291 */
5292 if (hw_breakpoint_active())
5293 hw_breakpoint_restore();
5294
5295 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
5296
5297 vcpu->mode = OUTSIDE_GUEST_MODE;
5298 smp_wmb();
5299 local_irq_enable();
5300
5301 ++vcpu->stat.exits;
5302
5303 /*
5304 * We must have an instruction between local_irq_enable() and
5305 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5306 * the interrupt shadow. The stat.exits increment will do nicely.
5307 * But we need to prevent reordering, hence this barrier():
5308 */
5309 barrier();
5310
5311 kvm_guest_exit();
5312
5313 preempt_enable();
5314
5315 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5316
5317 /*
5318 * Profile KVM exit RIPs:
5319 */
5320 if (unlikely(prof_on == KVM_PROFILING)) {
5321 unsigned long rip = kvm_rip_read(vcpu);
5322 profile_hit(KVM_PROFILING, (void *)rip);
5323 }
5324
5325 if (unlikely(vcpu->arch.tsc_always_catchup))
5326 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5327
5328 kvm_lapic_sync_from_vapic(vcpu);
5329
5330 r = kvm_x86_ops->handle_exit(vcpu);
5331out:
5332 return r;
5333}
5334
5335
5336static int __vcpu_run(struct kvm_vcpu *vcpu)
5337{
5338 int r;
5339 struct kvm *kvm = vcpu->kvm;
5340
5341 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5342 pr_debug("vcpu %d received sipi with vector # %x\n",
5343 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5344 kvm_lapic_reset(vcpu);
5345 r = kvm_arch_vcpu_reset(vcpu);
5346 if (r)
5347 return r;
5348 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5349 }
5350
5351 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5352 vapic_enter(vcpu);
5353
5354 r = 1;
5355 while (r > 0) {
5356 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5357 !vcpu->arch.apf.halted)
5358 r = vcpu_enter_guest(vcpu);
5359 else {
5360 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5361 kvm_vcpu_block(vcpu);
5362 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5363 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5364 {
5365 switch(vcpu->arch.mp_state) {
5366 case KVM_MP_STATE_HALTED:
5367 vcpu->arch.mp_state =
5368 KVM_MP_STATE_RUNNABLE;
5369 case KVM_MP_STATE_RUNNABLE:
5370 vcpu->arch.apf.halted = false;
5371 break;
5372 case KVM_MP_STATE_SIPI_RECEIVED:
5373 default:
5374 r = -EINTR;
5375 break;
5376 }
5377 }
5378 }
5379
5380 if (r <= 0)
5381 break;
5382
5383 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5384 if (kvm_cpu_has_pending_timer(vcpu))
5385 kvm_inject_pending_timer_irqs(vcpu);
5386
5387 if (dm_request_for_irq_injection(vcpu)) {
5388 r = -EINTR;
5389 vcpu->run->exit_reason = KVM_EXIT_INTR;
5390 ++vcpu->stat.request_irq_exits;
5391 }
5392
5393 kvm_check_async_pf_completion(vcpu);
5394
5395 if (signal_pending(current)) {
5396 r = -EINTR;
5397 vcpu->run->exit_reason = KVM_EXIT_INTR;
5398 ++vcpu->stat.signal_exits;
5399 }
5400 if (need_resched()) {
5401 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5402 kvm_resched(vcpu);
5403 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5404 }
5405 }
5406
5407 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5408
5409 vapic_exit(vcpu);
5410
5411 return r;
5412}
5413
5414static int complete_mmio(struct kvm_vcpu *vcpu)
5415{
5416 struct kvm_run *run = vcpu->run;
5417 int r;
5418
5419 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5420 return 1;
5421
5422 if (vcpu->mmio_needed) {
5423 vcpu->mmio_needed = 0;
5424 if (!vcpu->mmio_is_write)
5425 memcpy(vcpu->mmio_data + vcpu->mmio_index,
5426 run->mmio.data, 8);
5427 vcpu->mmio_index += 8;
5428 if (vcpu->mmio_index < vcpu->mmio_size) {
5429 run->exit_reason = KVM_EXIT_MMIO;
5430 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5431 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5432 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5433 run->mmio.is_write = vcpu->mmio_is_write;
5434 vcpu->mmio_needed = 1;
5435 return 0;
5436 }
5437 if (vcpu->mmio_is_write)
5438 return 1;
5439 vcpu->mmio_read_completed = 1;
5440 }
5441 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5442 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5443 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5444 if (r != EMULATE_DONE)
5445 return 0;
5446 return 1;
5447}
5448
5449int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5450{
5451 int r;
5452 sigset_t sigsaved;
5453
5454 if (!tsk_used_math(current) && init_fpu(current))
5455 return -ENOMEM;
5456
5457 if (vcpu->sigset_active)
5458 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5459
5460 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5461 kvm_vcpu_block(vcpu);
5462 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5463 r = -EAGAIN;
5464 goto out;
5465 }
5466
5467 /* re-sync apic's tpr */
5468 if (!irqchip_in_kernel(vcpu->kvm)) {
5469 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5470 r = -EINVAL;
5471 goto out;
5472 }
5473 }
5474
5475 r = complete_mmio(vcpu);
5476 if (r <= 0)
5477 goto out;
5478
5479 r = __vcpu_run(vcpu);
5480
5481out:
5482 post_kvm_run_save(vcpu);
5483 if (vcpu->sigset_active)
5484 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5485
5486 return r;
5487}
5488
5489int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5490{
5491 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5492 /*
5493 * We are here if userspace calls get_regs() in the middle of
5494 * instruction emulation. Registers state needs to be copied
5495 * back from emulation context to vcpu. Usrapace shouldn't do
5496 * that usually, but some bad designed PV devices (vmware
5497 * backdoor interface) need this to work
5498 */
5499 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5500 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5501 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5502 }
5503 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5504 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5505 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5506 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5507 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5508 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5509 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5510 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5511#ifdef CONFIG_X86_64
5512 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5513 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5514 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5515 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5516 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5517 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5518 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5519 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5520#endif
5521
5522 regs->rip = kvm_rip_read(vcpu);
5523 regs->rflags = kvm_get_rflags(vcpu);
5524
5525 return 0;
5526}
5527
5528int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5529{
5530 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5531 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5532
5533 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5534 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5535 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5536 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5537 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5538 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5539 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5540 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5541#ifdef CONFIG_X86_64
5542 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5543 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5544 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5545 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5546 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5547 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5548 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5549 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5550#endif
5551
5552 kvm_rip_write(vcpu, regs->rip);
5553 kvm_set_rflags(vcpu, regs->rflags);
5554
5555 vcpu->arch.exception.pending = false;
5556
5557 kvm_make_request(KVM_REQ_EVENT, vcpu);
5558
5559 return 0;
5560}
5561
5562void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5563{
5564 struct kvm_segment cs;
5565
5566 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5567 *db = cs.db;
5568 *l = cs.l;
5569}
5570EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5571
5572int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5573 struct kvm_sregs *sregs)
5574{
5575 struct desc_ptr dt;
5576
5577 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5578 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5579 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5580 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5581 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5582 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5583
5584 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5585 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5586
5587 kvm_x86_ops->get_idt(vcpu, &dt);
5588 sregs->idt.limit = dt.size;
5589 sregs->idt.base = dt.address;
5590 kvm_x86_ops->get_gdt(vcpu, &dt);
5591 sregs->gdt.limit = dt.size;
5592 sregs->gdt.base = dt.address;
5593
5594 sregs->cr0 = kvm_read_cr0(vcpu);
5595 sregs->cr2 = vcpu->arch.cr2;
5596 sregs->cr3 = kvm_read_cr3(vcpu);
5597 sregs->cr4 = kvm_read_cr4(vcpu);
5598 sregs->cr8 = kvm_get_cr8(vcpu);
5599 sregs->efer = vcpu->arch.efer;
5600 sregs->apic_base = kvm_get_apic_base(vcpu);
5601
5602 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5603
5604 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5605 set_bit(vcpu->arch.interrupt.nr,
5606 (unsigned long *)sregs->interrupt_bitmap);
5607
5608 return 0;
5609}
5610
5611int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5612 struct kvm_mp_state *mp_state)
5613{
5614 mp_state->mp_state = vcpu->arch.mp_state;
5615 return 0;
5616}
5617
5618int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5619 struct kvm_mp_state *mp_state)
5620{
5621 vcpu->arch.mp_state = mp_state->mp_state;
5622 kvm_make_request(KVM_REQ_EVENT, vcpu);
5623 return 0;
5624}
5625
5626int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5627 bool has_error_code, u32 error_code)
5628{
5629 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5630 int ret;
5631
5632 init_emulate_ctxt(vcpu);
5633
5634 ret = emulator_task_switch(ctxt, tss_selector, reason,
5635 has_error_code, error_code);
5636
5637 if (ret)
5638 return EMULATE_FAIL;
5639
5640 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5641 kvm_rip_write(vcpu, ctxt->eip);
5642 kvm_set_rflags(vcpu, ctxt->eflags);
5643 kvm_make_request(KVM_REQ_EVENT, vcpu);
5644 return EMULATE_DONE;
5645}
5646EXPORT_SYMBOL_GPL(kvm_task_switch);
5647
5648int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5649 struct kvm_sregs *sregs)
5650{
5651 int mmu_reset_needed = 0;
5652 int pending_vec, max_bits, idx;
5653 struct desc_ptr dt;
5654
5655 dt.size = sregs->idt.limit;
5656 dt.address = sregs->idt.base;
5657 kvm_x86_ops->set_idt(vcpu, &dt);
5658 dt.size = sregs->gdt.limit;
5659 dt.address = sregs->gdt.base;
5660 kvm_x86_ops->set_gdt(vcpu, &dt);
5661
5662 vcpu->arch.cr2 = sregs->cr2;
5663 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5664 vcpu->arch.cr3 = sregs->cr3;
5665 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5666
5667 kvm_set_cr8(vcpu, sregs->cr8);
5668
5669 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5670 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5671 kvm_set_apic_base(vcpu, sregs->apic_base);
5672
5673 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5674 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5675 vcpu->arch.cr0 = sregs->cr0;
5676
5677 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5678 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5679 if (sregs->cr4 & X86_CR4_OSXSAVE)
5680 kvm_update_cpuid(vcpu);
5681
5682 idx = srcu_read_lock(&vcpu->kvm->srcu);
5683 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5684 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5685 mmu_reset_needed = 1;
5686 }
5687 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5688
5689 if (mmu_reset_needed)
5690 kvm_mmu_reset_context(vcpu);
5691
5692 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5693 pending_vec = find_first_bit(
5694 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5695 if (pending_vec < max_bits) {
5696 kvm_queue_interrupt(vcpu, pending_vec, false);
5697 pr_debug("Set back pending irq %d\n", pending_vec);
5698 }
5699
5700 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5701 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5702 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5703 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5704 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5705 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5706
5707 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5708 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5709
5710 update_cr8_intercept(vcpu);
5711
5712 /* Older userspace won't unhalt the vcpu on reset. */
5713 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5714 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5715 !is_protmode(vcpu))
5716 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5717
5718 kvm_make_request(KVM_REQ_EVENT, vcpu);
5719
5720 return 0;
5721}
5722
5723int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5724 struct kvm_guest_debug *dbg)
5725{
5726 unsigned long rflags;
5727 int i, r;
5728
5729 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5730 r = -EBUSY;
5731 if (vcpu->arch.exception.pending)
5732 goto out;
5733 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5734 kvm_queue_exception(vcpu, DB_VECTOR);
5735 else
5736 kvm_queue_exception(vcpu, BP_VECTOR);
5737 }
5738
5739 /*
5740 * Read rflags as long as potentially injected trace flags are still
5741 * filtered out.
5742 */
5743 rflags = kvm_get_rflags(vcpu);
5744
5745 vcpu->guest_debug = dbg->control;
5746 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5747 vcpu->guest_debug = 0;
5748
5749 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5750 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5751 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5752 vcpu->arch.switch_db_regs =
5753 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5754 } else {
5755 for (i = 0; i < KVM_NR_DB_REGS; i++)
5756 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5757 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5758 }
5759
5760 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5761 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5762 get_segment_base(vcpu, VCPU_SREG_CS);
5763
5764 /*
5765 * Trigger an rflags update that will inject or remove the trace
5766 * flags.
5767 */
5768 kvm_set_rflags(vcpu, rflags);
5769
5770 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5771
5772 r = 0;
5773
5774out:
5775
5776 return r;
5777}
5778
5779/*
5780 * Translate a guest virtual address to a guest physical address.
5781 */
5782int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5783 struct kvm_translation *tr)
5784{
5785 unsigned long vaddr = tr->linear_address;
5786 gpa_t gpa;
5787 int idx;
5788
5789 idx = srcu_read_lock(&vcpu->kvm->srcu);
5790 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5791 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5792 tr->physical_address = gpa;
5793 tr->valid = gpa != UNMAPPED_GVA;
5794 tr->writeable = 1;
5795 tr->usermode = 0;
5796
5797 return 0;
5798}
5799
5800int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5801{
5802 struct i387_fxsave_struct *fxsave =
5803 &vcpu->arch.guest_fpu.state->fxsave;
5804
5805 memcpy(fpu->fpr, fxsave->st_space, 128);
5806 fpu->fcw = fxsave->cwd;
5807 fpu->fsw = fxsave->swd;
5808 fpu->ftwx = fxsave->twd;
5809 fpu->last_opcode = fxsave->fop;
5810 fpu->last_ip = fxsave->rip;
5811 fpu->last_dp = fxsave->rdp;
5812 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5813
5814 return 0;
5815}
5816
5817int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5818{
5819 struct i387_fxsave_struct *fxsave =
5820 &vcpu->arch.guest_fpu.state->fxsave;
5821
5822 memcpy(fxsave->st_space, fpu->fpr, 128);
5823 fxsave->cwd = fpu->fcw;
5824 fxsave->swd = fpu->fsw;
5825 fxsave->twd = fpu->ftwx;
5826 fxsave->fop = fpu->last_opcode;
5827 fxsave->rip = fpu->last_ip;
5828 fxsave->rdp = fpu->last_dp;
5829 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5830
5831 return 0;
5832}
5833
5834int fx_init(struct kvm_vcpu *vcpu)
5835{
5836 int err;
5837
5838 err = fpu_alloc(&vcpu->arch.guest_fpu);
5839 if (err)
5840 return err;
5841
5842 fpu_finit(&vcpu->arch.guest_fpu);
5843
5844 /*
5845 * Ensure guest xcr0 is valid for loading
5846 */
5847 vcpu->arch.xcr0 = XSTATE_FP;
5848
5849 vcpu->arch.cr0 |= X86_CR0_ET;
5850
5851 return 0;
5852}
5853EXPORT_SYMBOL_GPL(fx_init);
5854
5855static void fx_free(struct kvm_vcpu *vcpu)
5856{
5857 fpu_free(&vcpu->arch.guest_fpu);
5858}
5859
5860void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5861{
5862 if (vcpu->guest_fpu_loaded)
5863 return;
5864
5865 /*
5866 * Restore all possible states in the guest,
5867 * and assume host would use all available bits.
5868 * Guest xcr0 would be loaded later.
5869 */
5870 kvm_put_guest_xcr0(vcpu);
5871 vcpu->guest_fpu_loaded = 1;
5872 unlazy_fpu(current);
5873 fpu_restore_checking(&vcpu->arch.guest_fpu);
5874 trace_kvm_fpu(1);
5875}
5876
5877void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5878{
5879 kvm_put_guest_xcr0(vcpu);
5880
5881 if (!vcpu->guest_fpu_loaded)
5882 return;
5883
5884 vcpu->guest_fpu_loaded = 0;
5885 fpu_save_init(&vcpu->arch.guest_fpu);
5886 ++vcpu->stat.fpu_reload;
5887 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5888 trace_kvm_fpu(0);
5889}
5890
5891void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5892{
5893 kvmclock_reset(vcpu);
5894
5895 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5896 fx_free(vcpu);
5897 kvm_x86_ops->vcpu_free(vcpu);
5898}
5899
5900struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5901 unsigned int id)
5902{
5903 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5904 printk_once(KERN_WARNING
5905 "kvm: SMP vm created on host with unstable TSC; "
5906 "guest TSC will not be reliable\n");
5907 return kvm_x86_ops->vcpu_create(kvm, id);
5908}
5909
5910int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5911{
5912 int r;
5913
5914 vcpu->arch.mtrr_state.have_fixed = 1;
5915 vcpu_load(vcpu);
5916 r = kvm_arch_vcpu_reset(vcpu);
5917 if (r == 0)
5918 r = kvm_mmu_setup(vcpu);
5919 vcpu_put(vcpu);
5920
5921 return r;
5922}
5923
5924void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5925{
5926 vcpu->arch.apf.msr_val = 0;
5927
5928 vcpu_load(vcpu);
5929 kvm_mmu_unload(vcpu);
5930 vcpu_put(vcpu);
5931
5932 fx_free(vcpu);
5933 kvm_x86_ops->vcpu_free(vcpu);
5934}
5935
5936int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5937{
5938 atomic_set(&vcpu->arch.nmi_queued, 0);
5939 vcpu->arch.nmi_pending = 0;
5940 vcpu->arch.nmi_injected = false;
5941
5942 vcpu->arch.switch_db_regs = 0;
5943 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5944 vcpu->arch.dr6 = DR6_FIXED_1;
5945 vcpu->arch.dr7 = DR7_FIXED_1;
5946
5947 kvm_make_request(KVM_REQ_EVENT, vcpu);
5948 vcpu->arch.apf.msr_val = 0;
5949 vcpu->arch.st.msr_val = 0;
5950
5951 kvmclock_reset(vcpu);
5952
5953 kvm_clear_async_pf_completion_queue(vcpu);
5954 kvm_async_pf_hash_reset(vcpu);
5955 vcpu->arch.apf.halted = false;
5956
5957 kvm_pmu_reset(vcpu);
5958
5959 return kvm_x86_ops->vcpu_reset(vcpu);
5960}
5961
5962int kvm_arch_hardware_enable(void *garbage)
5963{
5964 struct kvm *kvm;
5965 struct kvm_vcpu *vcpu;
5966 int i;
5967
5968 kvm_shared_msr_cpu_online();
5969 list_for_each_entry(kvm, &vm_list, vm_list)
5970 kvm_for_each_vcpu(i, vcpu, kvm)
5971 if (vcpu->cpu == smp_processor_id())
5972 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5973 return kvm_x86_ops->hardware_enable(garbage);
5974}
5975
5976void kvm_arch_hardware_disable(void *garbage)
5977{
5978 kvm_x86_ops->hardware_disable(garbage);
5979 drop_user_return_notifiers(garbage);
5980}
5981
5982int kvm_arch_hardware_setup(void)
5983{
5984 return kvm_x86_ops->hardware_setup();
5985}
5986
5987void kvm_arch_hardware_unsetup(void)
5988{
5989 kvm_x86_ops->hardware_unsetup();
5990}
5991
5992void kvm_arch_check_processor_compat(void *rtn)
5993{
5994 kvm_x86_ops->check_processor_compatibility(rtn);
5995}
5996
5997int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5998{
5999 struct page *page;
6000 struct kvm *kvm;
6001 int r;
6002
6003 BUG_ON(vcpu->kvm == NULL);
6004 kvm = vcpu->kvm;
6005
6006 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6007 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6008 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6009 else
6010 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6011
6012 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6013 if (!page) {
6014 r = -ENOMEM;
6015 goto fail;
6016 }
6017 vcpu->arch.pio_data = page_address(page);
6018
6019 kvm_set_tsc_khz(vcpu, max_tsc_khz);
6020
6021 r = kvm_mmu_create(vcpu);
6022 if (r < 0)
6023 goto fail_free_pio_data;
6024
6025 if (irqchip_in_kernel(kvm)) {
6026 r = kvm_create_lapic(vcpu);
6027 if (r < 0)
6028 goto fail_mmu_destroy;
6029 }
6030
6031 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6032 GFP_KERNEL);
6033 if (!vcpu->arch.mce_banks) {
6034 r = -ENOMEM;
6035 goto fail_free_lapic;
6036 }
6037 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6038
6039 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6040 goto fail_free_mce_banks;
6041
6042 kvm_async_pf_hash_reset(vcpu);
6043 kvm_pmu_init(vcpu);
6044
6045 return 0;
6046fail_free_mce_banks:
6047 kfree(vcpu->arch.mce_banks);
6048fail_free_lapic:
6049 kvm_free_lapic(vcpu);
6050fail_mmu_destroy:
6051 kvm_mmu_destroy(vcpu);
6052fail_free_pio_data:
6053 free_page((unsigned long)vcpu->arch.pio_data);
6054fail:
6055 return r;
6056}
6057
6058void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6059{
6060 int idx;
6061
6062 kvm_pmu_destroy(vcpu);
6063 kfree(vcpu->arch.mce_banks);
6064 kvm_free_lapic(vcpu);
6065 idx = srcu_read_lock(&vcpu->kvm->srcu);
6066 kvm_mmu_destroy(vcpu);
6067 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6068 free_page((unsigned long)vcpu->arch.pio_data);
6069}
6070
6071int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
6072{
6073 if (type)
6074 return -EINVAL;
6075
6076 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6077 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6078
6079 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6080 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6081
6082 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6083
6084 return 0;
6085}
6086
6087static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6088{
6089 vcpu_load(vcpu);
6090 kvm_mmu_unload(vcpu);
6091 vcpu_put(vcpu);
6092}
6093
6094static void kvm_free_vcpus(struct kvm *kvm)
6095{
6096 unsigned int i;
6097 struct kvm_vcpu *vcpu;
6098
6099 /*
6100 * Unpin any mmu pages first.
6101 */
6102 kvm_for_each_vcpu(i, vcpu, kvm) {
6103 kvm_clear_async_pf_completion_queue(vcpu);
6104 kvm_unload_vcpu_mmu(vcpu);
6105 }
6106 kvm_for_each_vcpu(i, vcpu, kvm)
6107 kvm_arch_vcpu_free(vcpu);
6108
6109 mutex_lock(&kvm->lock);
6110 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6111 kvm->vcpus[i] = NULL;
6112
6113 atomic_set(&kvm->online_vcpus, 0);
6114 mutex_unlock(&kvm->lock);
6115}
6116
6117void kvm_arch_sync_events(struct kvm *kvm)
6118{
6119 kvm_free_all_assigned_devices(kvm);
6120 kvm_free_pit(kvm);
6121}
6122
6123void kvm_arch_destroy_vm(struct kvm *kvm)
6124{
6125 kvm_iommu_unmap_guest(kvm);
6126 kfree(kvm->arch.vpic);
6127 kfree(kvm->arch.vioapic);
6128 kvm_free_vcpus(kvm);
6129 if (kvm->arch.apic_access_page)
6130 put_page(kvm->arch.apic_access_page);
6131 if (kvm->arch.ept_identity_pagetable)
6132 put_page(kvm->arch.ept_identity_pagetable);
6133}
6134
6135int kvm_arch_prepare_memory_region(struct kvm *kvm,
6136 struct kvm_memory_slot *memslot,
6137 struct kvm_memory_slot old,
6138 struct kvm_userspace_memory_region *mem,
6139 int user_alloc)
6140{
6141 int npages = memslot->npages;
6142 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6143
6144 /* Prevent internal slot pages from being moved by fork()/COW. */
6145 if (memslot->id >= KVM_MEMORY_SLOTS)
6146 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6147
6148 /*To keep backward compatibility with older userspace,
6149 *x86 needs to hanlde !user_alloc case.
6150 */
6151 if (!user_alloc) {
6152 if (npages && !old.rmap) {
6153 unsigned long userspace_addr;
6154
6155 down_write(&current->mm->mmap_sem);
6156 userspace_addr = do_mmap(NULL, 0,
6157 npages * PAGE_SIZE,
6158 PROT_READ | PROT_WRITE,
6159 map_flags,
6160 0);
6161 up_write(&current->mm->mmap_sem);
6162
6163 if (IS_ERR((void *)userspace_addr))
6164 return PTR_ERR((void *)userspace_addr);
6165
6166 memslot->userspace_addr = userspace_addr;
6167 }
6168 }
6169
6170
6171 return 0;
6172}
6173
6174void kvm_arch_commit_memory_region(struct kvm *kvm,
6175 struct kvm_userspace_memory_region *mem,
6176 struct kvm_memory_slot old,
6177 int user_alloc)
6178{
6179
6180 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6181
6182 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6183 int ret;
6184
6185 down_write(&current->mm->mmap_sem);
6186 ret = do_munmap(current->mm, old.userspace_addr,
6187 old.npages * PAGE_SIZE);
6188 up_write(&current->mm->mmap_sem);
6189 if (ret < 0)
6190 printk(KERN_WARNING
6191 "kvm_vm_ioctl_set_memory_region: "
6192 "failed to munmap memory\n");
6193 }
6194
6195 if (!kvm->arch.n_requested_mmu_pages)
6196 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6197
6198 spin_lock(&kvm->mmu_lock);
6199 if (nr_mmu_pages)
6200 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6201 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6202 spin_unlock(&kvm->mmu_lock);
6203}
6204
6205void kvm_arch_flush_shadow(struct kvm *kvm)
6206{
6207 kvm_mmu_zap_all(kvm);
6208 kvm_reload_remote_mmus(kvm);
6209}
6210
6211int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6212{
6213 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6214 !vcpu->arch.apf.halted)
6215 || !list_empty_careful(&vcpu->async_pf.done)
6216 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6217 || atomic_read(&vcpu->arch.nmi_queued) ||
6218 (kvm_arch_interrupt_allowed(vcpu) &&
6219 kvm_cpu_has_interrupt(vcpu));
6220}
6221
6222void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6223{
6224 int me;
6225 int cpu = vcpu->cpu;
6226
6227 if (waitqueue_active(&vcpu->wq)) {
6228 wake_up_interruptible(&vcpu->wq);
6229 ++vcpu->stat.halt_wakeup;
6230 }
6231
6232 me = get_cpu();
6233 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6234 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6235 smp_send_reschedule(cpu);
6236 put_cpu();
6237}
6238
6239int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6240{
6241 return kvm_x86_ops->interrupt_allowed(vcpu);
6242}
6243
6244bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6245{
6246 unsigned long current_rip = kvm_rip_read(vcpu) +
6247 get_segment_base(vcpu, VCPU_SREG_CS);
6248
6249 return current_rip == linear_rip;
6250}
6251EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6252
6253unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6254{
6255 unsigned long rflags;
6256
6257 rflags = kvm_x86_ops->get_rflags(vcpu);
6258 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6259 rflags &= ~X86_EFLAGS_TF;
6260 return rflags;
6261}
6262EXPORT_SYMBOL_GPL(kvm_get_rflags);
6263
6264void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6265{
6266 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6267 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6268 rflags |= X86_EFLAGS_TF;
6269 kvm_x86_ops->set_rflags(vcpu, rflags);
6270 kvm_make_request(KVM_REQ_EVENT, vcpu);
6271}
6272EXPORT_SYMBOL_GPL(kvm_set_rflags);
6273
6274void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6275{
6276 int r;
6277
6278 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6279 is_error_page(work->page))
6280 return;
6281
6282 r = kvm_mmu_reload(vcpu);
6283 if (unlikely(r))
6284 return;
6285
6286 if (!vcpu->arch.mmu.direct_map &&
6287 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6288 return;
6289
6290 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6291}
6292
6293static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6294{
6295 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6296}
6297
6298static inline u32 kvm_async_pf_next_probe(u32 key)
6299{
6300 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6301}
6302
6303static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6304{
6305 u32 key = kvm_async_pf_hash_fn(gfn);
6306
6307 while (vcpu->arch.apf.gfns[key] != ~0)
6308 key = kvm_async_pf_next_probe(key);
6309
6310 vcpu->arch.apf.gfns[key] = gfn;
6311}
6312
6313static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6314{
6315 int i;
6316 u32 key = kvm_async_pf_hash_fn(gfn);
6317
6318 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6319 (vcpu->arch.apf.gfns[key] != gfn &&
6320 vcpu->arch.apf.gfns[key] != ~0); i++)
6321 key = kvm_async_pf_next_probe(key);
6322
6323 return key;
6324}
6325
6326bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6327{
6328 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6329}
6330
6331static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6332{
6333 u32 i, j, k;
6334
6335 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6336 while (true) {
6337 vcpu->arch.apf.gfns[i] = ~0;
6338 do {
6339 j = kvm_async_pf_next_probe(j);
6340 if (vcpu->arch.apf.gfns[j] == ~0)
6341 return;
6342 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6343 /*
6344 * k lies cyclically in ]i,j]
6345 * | i.k.j |
6346 * |....j i.k.| or |.k..j i...|
6347 */
6348 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6349 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6350 i = j;
6351 }
6352}
6353
6354static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6355{
6356
6357 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6358 sizeof(val));
6359}
6360
6361void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6362 struct kvm_async_pf *work)
6363{
6364 struct x86_exception fault;
6365
6366 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6367 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6368
6369 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6370 (vcpu->arch.apf.send_user_only &&
6371 kvm_x86_ops->get_cpl(vcpu) == 0))
6372 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6373 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6374 fault.vector = PF_VECTOR;
6375 fault.error_code_valid = true;
6376 fault.error_code = 0;
6377 fault.nested_page_fault = false;
6378 fault.address = work->arch.token;
6379 kvm_inject_page_fault(vcpu, &fault);
6380 }
6381}
6382
6383void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6384 struct kvm_async_pf *work)
6385{
6386 struct x86_exception fault;
6387
6388 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6389 if (is_error_page(work->page))
6390 work->arch.token = ~0; /* broadcast wakeup */
6391 else
6392 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6393
6394 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6395 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6396 fault.vector = PF_VECTOR;
6397 fault.error_code_valid = true;
6398 fault.error_code = 0;
6399 fault.nested_page_fault = false;
6400 fault.address = work->arch.token;
6401 kvm_inject_page_fault(vcpu, &fault);
6402 }
6403 vcpu->arch.apf.halted = false;
6404}
6405
6406bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6407{
6408 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6409 return true;
6410 else
6411 return !kvm_event_needs_reinjection(vcpu) &&
6412 kvm_x86_ops->interrupt_allowed(vcpu);
6413}
6414
6415EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6416EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6417EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6418EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6419EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6420EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6421EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6422EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6423EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6424EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6425EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6426EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
Linux 6.x block layer and io_uring tree(s)