2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
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.
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>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
34 #include <linux/vmalloc.h>
35 #include <linux/module.h>
36 #include <linux/mman.h>
37 #include <linux/highmem.h>
38 #include <linux/iommu.h>
39 #include <linux/intel-iommu.h>
40 #include <linux/cpufreq.h>
41 #include <linux/user-return-notifier.h>
42 #include <linux/srcu.h>
43 #include <linux/slab.h>
44 #include <linux/perf_event.h>
45 #include <linux/uaccess.h>
46 #include <linux/hash.h>
47 #include <trace/events/kvm.h>
49 #define CREATE_TRACE_POINTS
52 #include <asm/debugreg.h>
59 #include <asm/pvclock.h>
60 #include <asm/div64.h>
62 #define MAX_IO_MSRS 256
63 #define CR0_RESERVED_BITS \
64 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
65 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
66 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
67 #define CR4_RESERVED_BITS \
68 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
69 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
70 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
72 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
74 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
76 #define KVM_MAX_MCE_BANKS 32
77 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
80 * - enable syscall per default because its emulated by KVM
81 * - enable LME and LMA per default on 64 bit KVM
85 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
87 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
90 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
91 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
93 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
94 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
95 struct kvm_cpuid_entry2 __user *entries);
97 struct kvm_x86_ops *kvm_x86_ops;
98 EXPORT_SYMBOL_GPL(kvm_x86_ops);
101 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
103 #define KVM_NR_SHARED_MSRS 16
105 struct kvm_shared_msrs_global {
107 u32 msrs[KVM_NR_SHARED_MSRS];
110 struct kvm_shared_msrs {
111 struct user_return_notifier urn;
113 struct kvm_shared_msr_values {
116 } values[KVM_NR_SHARED_MSRS];
119 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
120 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
122 struct 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) },
158 u64 __read_mostly host_xcr0;
160 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
163 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
164 vcpu->arch.apf.gfns[i] = ~0;
167 static void kvm_on_user_return(struct user_return_notifier *urn)
170 struct kvm_shared_msrs *locals
171 = container_of(urn, struct kvm_shared_msrs, urn);
172 struct kvm_shared_msr_values *values;
174 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
175 values = &locals->values[slot];
176 if (values->host != values->curr) {
177 wrmsrl(shared_msrs_global.msrs[slot], values->host);
178 values->curr = values->host;
181 locals->registered = false;
182 user_return_notifier_unregister(urn);
185 static void shared_msr_update(unsigned slot, u32 msr)
187 struct kvm_shared_msrs *smsr;
190 smsr = &__get_cpu_var(shared_msrs);
191 /* only read, and nobody should modify it at this time,
192 * so don't need lock */
193 if (slot >= shared_msrs_global.nr) {
194 printk(KERN_ERR "kvm: invalid MSR slot!");
197 rdmsrl_safe(msr, &value);
198 smsr->values[slot].host = value;
199 smsr->values[slot].curr = value;
202 void kvm_define_shared_msr(unsigned slot, u32 msr)
204 if (slot >= shared_msrs_global.nr)
205 shared_msrs_global.nr = slot + 1;
206 shared_msrs_global.msrs[slot] = msr;
207 /* we need ensured the shared_msr_global have been updated */
210 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
212 static void kvm_shared_msr_cpu_online(void)
216 for (i = 0; i < shared_msrs_global.nr; ++i)
217 shared_msr_update(i, shared_msrs_global.msrs[i]);
220 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
222 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
224 if (((value ^ smsr->values[slot].curr) & mask) == 0)
226 smsr->values[slot].curr = value;
227 wrmsrl(shared_msrs_global.msrs[slot], value);
228 if (!smsr->registered) {
229 smsr->urn.on_user_return = kvm_on_user_return;
230 user_return_notifier_register(&smsr->urn);
231 smsr->registered = true;
234 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
236 static void drop_user_return_notifiers(void *ignore)
238 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
240 if (smsr->registered)
241 kvm_on_user_return(&smsr->urn);
244 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
246 if (irqchip_in_kernel(vcpu->kvm))
247 return vcpu->arch.apic_base;
249 return vcpu->arch.apic_base;
251 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
253 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
255 /* TODO: reserve bits check */
256 if (irqchip_in_kernel(vcpu->kvm))
257 kvm_lapic_set_base(vcpu, data);
259 vcpu->arch.apic_base = data;
261 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
263 #define EXCPT_BENIGN 0
264 #define EXCPT_CONTRIBUTORY 1
267 static int exception_class(int vector)
277 return EXCPT_CONTRIBUTORY;
284 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
285 unsigned nr, bool has_error, u32 error_code,
291 kvm_make_request(KVM_REQ_EVENT, vcpu);
293 if (!vcpu->arch.exception.pending) {
295 vcpu->arch.exception.pending = true;
296 vcpu->arch.exception.has_error_code = has_error;
297 vcpu->arch.exception.nr = nr;
298 vcpu->arch.exception.error_code = error_code;
299 vcpu->arch.exception.reinject = reinject;
303 /* to check exception */
304 prev_nr = vcpu->arch.exception.nr;
305 if (prev_nr == DF_VECTOR) {
306 /* triple fault -> shutdown */
307 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
310 class1 = exception_class(prev_nr);
311 class2 = exception_class(nr);
312 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
313 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
314 /* generate double fault per SDM Table 5-5 */
315 vcpu->arch.exception.pending = true;
316 vcpu->arch.exception.has_error_code = true;
317 vcpu->arch.exception.nr = DF_VECTOR;
318 vcpu->arch.exception.error_code = 0;
320 /* replace previous exception with a new one in a hope
321 that instruction re-execution will regenerate lost
326 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
328 kvm_multiple_exception(vcpu, nr, false, 0, false);
330 EXPORT_SYMBOL_GPL(kvm_queue_exception);
332 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
334 kvm_multiple_exception(vcpu, nr, false, 0, true);
336 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
338 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
341 kvm_inject_gp(vcpu, 0);
343 kvm_x86_ops->skip_emulated_instruction(vcpu);
345 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
347 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
349 ++vcpu->stat.pf_guest;
350 vcpu->arch.cr2 = fault->address;
351 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
354 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
356 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
357 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
359 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
362 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
364 kvm_make_request(KVM_REQ_EVENT, vcpu);
365 vcpu->arch.nmi_pending = 1;
367 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
369 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
371 kvm_multiple_exception(vcpu, nr, true, error_code, false);
373 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
375 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
377 kvm_multiple_exception(vcpu, nr, true, error_code, true);
379 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
382 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
383 * a #GP and return false.
385 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
387 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
389 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
392 EXPORT_SYMBOL_GPL(kvm_require_cpl);
395 * This function will be used to read from the physical memory of the currently
396 * running guest. The difference to kvm_read_guest_page is that this function
397 * can read from guest physical or from the guest's guest physical memory.
399 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
400 gfn_t ngfn, void *data, int offset, int len,
406 ngpa = gfn_to_gpa(ngfn);
407 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
408 if (real_gfn == UNMAPPED_GVA)
411 real_gfn = gpa_to_gfn(real_gfn);
413 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
415 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
417 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
418 void *data, int offset, int len, u32 access)
420 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
421 data, offset, len, access);
425 * Load the pae pdptrs. Return true is they are all valid.
427 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
429 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
430 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
433 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
435 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
436 offset * sizeof(u64), sizeof(pdpte),
437 PFERR_USER_MASK|PFERR_WRITE_MASK);
442 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
443 if (is_present_gpte(pdpte[i]) &&
444 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
451 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
452 __set_bit(VCPU_EXREG_PDPTR,
453 (unsigned long *)&vcpu->arch.regs_avail);
454 __set_bit(VCPU_EXREG_PDPTR,
455 (unsigned long *)&vcpu->arch.regs_dirty);
460 EXPORT_SYMBOL_GPL(load_pdptrs);
462 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
464 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
470 if (is_long_mode(vcpu) || !is_pae(vcpu))
473 if (!test_bit(VCPU_EXREG_PDPTR,
474 (unsigned long *)&vcpu->arch.regs_avail))
477 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
478 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
479 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
480 PFERR_USER_MASK | PFERR_WRITE_MASK);
483 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
489 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
491 unsigned long old_cr0 = kvm_read_cr0(vcpu);
492 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
493 X86_CR0_CD | X86_CR0_NW;
498 if (cr0 & 0xffffffff00000000UL)
502 cr0 &= ~CR0_RESERVED_BITS;
504 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
507 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
510 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
512 if ((vcpu->arch.efer & EFER_LME)) {
517 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
522 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
527 kvm_x86_ops->set_cr0(vcpu, cr0);
529 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
530 kvm_clear_async_pf_completion_queue(vcpu);
531 kvm_async_pf_hash_reset(vcpu);
534 if ((cr0 ^ old_cr0) & update_bits)
535 kvm_mmu_reset_context(vcpu);
538 EXPORT_SYMBOL_GPL(kvm_set_cr0);
540 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
542 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
544 EXPORT_SYMBOL_GPL(kvm_lmsw);
546 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
550 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
551 if (index != XCR_XFEATURE_ENABLED_MASK)
554 if (kvm_x86_ops->get_cpl(vcpu) != 0)
556 if (!(xcr0 & XSTATE_FP))
558 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
560 if (xcr0 & ~host_xcr0)
562 vcpu->arch.xcr0 = xcr0;
563 vcpu->guest_xcr0_loaded = 0;
567 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
569 if (__kvm_set_xcr(vcpu, index, xcr)) {
570 kvm_inject_gp(vcpu, 0);
575 EXPORT_SYMBOL_GPL(kvm_set_xcr);
577 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
579 struct kvm_cpuid_entry2 *best;
581 best = kvm_find_cpuid_entry(vcpu, 1, 0);
582 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
585 static void update_cpuid(struct kvm_vcpu *vcpu)
587 struct kvm_cpuid_entry2 *best;
589 best = kvm_find_cpuid_entry(vcpu, 1, 0);
593 /* Update OSXSAVE bit */
594 if (cpu_has_xsave && best->function == 0x1) {
595 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
596 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
597 best->ecx |= bit(X86_FEATURE_OSXSAVE);
601 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
603 unsigned long old_cr4 = kvm_read_cr4(vcpu);
604 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
606 if (cr4 & CR4_RESERVED_BITS)
609 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
612 if (is_long_mode(vcpu)) {
613 if (!(cr4 & X86_CR4_PAE))
615 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
616 && ((cr4 ^ old_cr4) & pdptr_bits)
617 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
621 if (cr4 & X86_CR4_VMXE)
624 kvm_x86_ops->set_cr4(vcpu, cr4);
626 if ((cr4 ^ old_cr4) & pdptr_bits)
627 kvm_mmu_reset_context(vcpu);
629 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
634 EXPORT_SYMBOL_GPL(kvm_set_cr4);
636 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
638 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
639 kvm_mmu_sync_roots(vcpu);
640 kvm_mmu_flush_tlb(vcpu);
644 if (is_long_mode(vcpu)) {
645 if (cr3 & CR3_L_MODE_RESERVED_BITS)
649 if (cr3 & CR3_PAE_RESERVED_BITS)
651 if (is_paging(vcpu) &&
652 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
656 * We don't check reserved bits in nonpae mode, because
657 * this isn't enforced, and VMware depends on this.
662 * Does the new cr3 value map to physical memory? (Note, we
663 * catch an invalid cr3 even in real-mode, because it would
664 * cause trouble later on when we turn on paging anyway.)
666 * A real CPU would silently accept an invalid cr3 and would
667 * attempt to use it - with largely undefined (and often hard
668 * to debug) behavior on the guest side.
670 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
672 vcpu->arch.cr3 = cr3;
673 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
674 vcpu->arch.mmu.new_cr3(vcpu);
677 EXPORT_SYMBOL_GPL(kvm_set_cr3);
679 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
681 if (cr8 & CR8_RESERVED_BITS)
683 if (irqchip_in_kernel(vcpu->kvm))
684 kvm_lapic_set_tpr(vcpu, cr8);
686 vcpu->arch.cr8 = cr8;
689 EXPORT_SYMBOL_GPL(kvm_set_cr8);
691 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
693 if (irqchip_in_kernel(vcpu->kvm))
694 return kvm_lapic_get_cr8(vcpu);
696 return vcpu->arch.cr8;
698 EXPORT_SYMBOL_GPL(kvm_get_cr8);
700 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
704 vcpu->arch.db[dr] = val;
705 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
706 vcpu->arch.eff_db[dr] = val;
709 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
713 if (val & 0xffffffff00000000ULL)
715 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
718 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
722 if (val & 0xffffffff00000000ULL)
724 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
725 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
726 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
727 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
735 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
739 res = __kvm_set_dr(vcpu, dr, val);
741 kvm_queue_exception(vcpu, UD_VECTOR);
743 kvm_inject_gp(vcpu, 0);
747 EXPORT_SYMBOL_GPL(kvm_set_dr);
749 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
753 *val = vcpu->arch.db[dr];
756 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
760 *val = vcpu->arch.dr6;
763 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
767 *val = vcpu->arch.dr7;
774 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
776 if (_kvm_get_dr(vcpu, dr, val)) {
777 kvm_queue_exception(vcpu, UD_VECTOR);
782 EXPORT_SYMBOL_GPL(kvm_get_dr);
785 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
786 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
788 * This list is modified at module load time to reflect the
789 * capabilities of the host cpu. This capabilities test skips MSRs that are
790 * kvm-specific. Those are put in the beginning of the list.
793 #define KVM_SAVE_MSRS_BEGIN 8
794 static u32 msrs_to_save[] = {
795 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
796 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
797 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
798 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
799 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
802 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
804 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
807 static unsigned num_msrs_to_save;
809 static u32 emulated_msrs[] = {
810 MSR_IA32_MISC_ENABLE,
815 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
817 u64 old_efer = vcpu->arch.efer;
819 if (efer & efer_reserved_bits)
823 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
826 if (efer & EFER_FFXSR) {
827 struct kvm_cpuid_entry2 *feat;
829 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
830 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
834 if (efer & EFER_SVME) {
835 struct kvm_cpuid_entry2 *feat;
837 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
838 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
843 efer |= vcpu->arch.efer & EFER_LMA;
845 kvm_x86_ops->set_efer(vcpu, efer);
847 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
849 /* Update reserved bits */
850 if ((efer ^ old_efer) & EFER_NX)
851 kvm_mmu_reset_context(vcpu);
856 void kvm_enable_efer_bits(u64 mask)
858 efer_reserved_bits &= ~mask;
860 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
864 * Writes msr value into into the appropriate "register".
865 * Returns 0 on success, non-0 otherwise.
866 * Assumes vcpu_load() was already called.
868 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
870 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
874 * Adapt set_msr() to msr_io()'s calling convention
876 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
878 return kvm_set_msr(vcpu, index, *data);
881 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
885 struct pvclock_wall_clock wc;
886 struct timespec boot;
891 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
896 ++version; /* first time write, random junk */
900 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
903 * The guest calculates current wall clock time by adding
904 * system time (updated by kvm_guest_time_update below) to the
905 * wall clock specified here. guest system time equals host
906 * system time for us, thus we must fill in host boot time here.
910 wc.sec = boot.tv_sec;
911 wc.nsec = boot.tv_nsec;
912 wc.version = version;
914 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
917 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
920 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
922 uint32_t quotient, remainder;
924 /* Don't try to replace with do_div(), this one calculates
925 * "(dividend << 32) / divisor" */
927 : "=a" (quotient), "=d" (remainder)
928 : "0" (0), "1" (dividend), "r" (divisor) );
932 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
933 s8 *pshift, u32 *pmultiplier)
940 tps64 = base_khz * 1000LL;
941 scaled64 = scaled_khz * 1000LL;
942 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
947 tps32 = (uint32_t)tps64;
948 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
949 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
957 *pmultiplier = div_frac(scaled64, tps32);
959 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
960 __func__, base_khz, scaled_khz, shift, *pmultiplier);
963 static inline u64 get_kernel_ns(void)
967 WARN_ON(preemptible());
969 monotonic_to_bootbased(&ts);
970 return timespec_to_ns(&ts);
973 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
974 unsigned long max_tsc_khz;
976 static inline int kvm_tsc_changes_freq(void)
979 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
980 cpufreq_quick_get(cpu) != 0;
985 static inline u64 nsec_to_cycles(u64 nsec)
989 WARN_ON(preemptible());
990 if (kvm_tsc_changes_freq())
991 printk_once(KERN_WARNING
992 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
993 ret = nsec * __this_cpu_read(cpu_tsc_khz);
994 do_div(ret, USEC_PER_SEC);
998 static void kvm_arch_set_tsc_khz(struct kvm *kvm, u32 this_tsc_khz)
1000 /* Compute a scale to convert nanoseconds in TSC cycles */
1001 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1002 &kvm->arch.virtual_tsc_shift,
1003 &kvm->arch.virtual_tsc_mult);
1004 kvm->arch.virtual_tsc_khz = this_tsc_khz;
1007 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1009 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1010 vcpu->kvm->arch.virtual_tsc_mult,
1011 vcpu->kvm->arch.virtual_tsc_shift);
1012 tsc += vcpu->arch.last_tsc_write;
1016 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1018 struct kvm *kvm = vcpu->kvm;
1019 u64 offset, ns, elapsed;
1020 unsigned long flags;
1023 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1024 offset = data - native_read_tsc();
1025 ns = get_kernel_ns();
1026 elapsed = ns - kvm->arch.last_tsc_nsec;
1027 sdiff = data - kvm->arch.last_tsc_write;
1032 * Special case: close write to TSC within 5 seconds of
1033 * another CPU is interpreted as an attempt to synchronize
1034 * The 5 seconds is to accommodate host load / swapping as
1035 * well as any reset of TSC during the boot process.
1037 * In that case, for a reliable TSC, we can match TSC offsets,
1038 * or make a best guest using elapsed value.
1040 if (sdiff < nsec_to_cycles(5ULL * NSEC_PER_SEC) &&
1041 elapsed < 5ULL * NSEC_PER_SEC) {
1042 if (!check_tsc_unstable()) {
1043 offset = kvm->arch.last_tsc_offset;
1044 pr_debug("kvm: matched tsc offset for %llu\n", data);
1046 u64 delta = nsec_to_cycles(elapsed);
1048 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1050 ns = kvm->arch.last_tsc_nsec;
1052 kvm->arch.last_tsc_nsec = ns;
1053 kvm->arch.last_tsc_write = data;
1054 kvm->arch.last_tsc_offset = offset;
1055 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1056 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1058 /* Reset of TSC must disable overshoot protection below */
1059 vcpu->arch.hv_clock.tsc_timestamp = 0;
1060 vcpu->arch.last_tsc_write = data;
1061 vcpu->arch.last_tsc_nsec = ns;
1063 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1065 static int kvm_guest_time_update(struct kvm_vcpu *v)
1067 unsigned long flags;
1068 struct kvm_vcpu_arch *vcpu = &v->arch;
1070 unsigned long this_tsc_khz;
1071 s64 kernel_ns, max_kernel_ns;
1074 /* Keep irq disabled to prevent changes to the clock */
1075 local_irq_save(flags);
1076 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1077 kernel_ns = get_kernel_ns();
1078 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1080 if (unlikely(this_tsc_khz == 0)) {
1081 local_irq_restore(flags);
1082 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1087 * We may have to catch up the TSC to match elapsed wall clock
1088 * time for two reasons, even if kvmclock is used.
1089 * 1) CPU could have been running below the maximum TSC rate
1090 * 2) Broken TSC compensation resets the base at each VCPU
1091 * entry to avoid unknown leaps of TSC even when running
1092 * again on the same CPU. This may cause apparent elapsed
1093 * time to disappear, and the guest to stand still or run
1096 if (vcpu->tsc_catchup) {
1097 u64 tsc = compute_guest_tsc(v, kernel_ns);
1098 if (tsc > tsc_timestamp) {
1099 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1100 tsc_timestamp = tsc;
1104 local_irq_restore(flags);
1106 if (!vcpu->time_page)
1110 * Time as measured by the TSC may go backwards when resetting the base
1111 * tsc_timestamp. The reason for this is that the TSC resolution is
1112 * higher than the resolution of the other clock scales. Thus, many
1113 * possible measurments of the TSC correspond to one measurement of any
1114 * other clock, and so a spread of values is possible. This is not a
1115 * problem for the computation of the nanosecond clock; with TSC rates
1116 * around 1GHZ, there can only be a few cycles which correspond to one
1117 * nanosecond value, and any path through this code will inevitably
1118 * take longer than that. However, with the kernel_ns value itself,
1119 * the precision may be much lower, down to HZ granularity. If the
1120 * first sampling of TSC against kernel_ns ends in the low part of the
1121 * range, and the second in the high end of the range, we can get:
1123 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1125 * As the sampling errors potentially range in the thousands of cycles,
1126 * it is possible such a time value has already been observed by the
1127 * guest. To protect against this, we must compute the system time as
1128 * observed by the guest and ensure the new system time is greater.
1131 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1132 max_kernel_ns = vcpu->last_guest_tsc -
1133 vcpu->hv_clock.tsc_timestamp;
1134 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1135 vcpu->hv_clock.tsc_to_system_mul,
1136 vcpu->hv_clock.tsc_shift);
1137 max_kernel_ns += vcpu->last_kernel_ns;
1140 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1141 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1142 &vcpu->hv_clock.tsc_shift,
1143 &vcpu->hv_clock.tsc_to_system_mul);
1144 vcpu->hw_tsc_khz = this_tsc_khz;
1147 if (max_kernel_ns > kernel_ns)
1148 kernel_ns = max_kernel_ns;
1150 /* With all the info we got, fill in the values */
1151 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1152 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1153 vcpu->last_kernel_ns = kernel_ns;
1154 vcpu->last_guest_tsc = tsc_timestamp;
1155 vcpu->hv_clock.flags = 0;
1158 * The interface expects us to write an even number signaling that the
1159 * update is finished. Since the guest won't see the intermediate
1160 * state, we just increase by 2 at the end.
1162 vcpu->hv_clock.version += 2;
1164 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1166 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1167 sizeof(vcpu->hv_clock));
1169 kunmap_atomic(shared_kaddr, KM_USER0);
1171 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1175 static bool msr_mtrr_valid(unsigned msr)
1178 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1179 case MSR_MTRRfix64K_00000:
1180 case MSR_MTRRfix16K_80000:
1181 case MSR_MTRRfix16K_A0000:
1182 case MSR_MTRRfix4K_C0000:
1183 case MSR_MTRRfix4K_C8000:
1184 case MSR_MTRRfix4K_D0000:
1185 case MSR_MTRRfix4K_D8000:
1186 case MSR_MTRRfix4K_E0000:
1187 case MSR_MTRRfix4K_E8000:
1188 case MSR_MTRRfix4K_F0000:
1189 case MSR_MTRRfix4K_F8000:
1190 case MSR_MTRRdefType:
1191 case MSR_IA32_CR_PAT:
1199 static bool valid_pat_type(unsigned t)
1201 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1204 static bool valid_mtrr_type(unsigned t)
1206 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1209 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1213 if (!msr_mtrr_valid(msr))
1216 if (msr == MSR_IA32_CR_PAT) {
1217 for (i = 0; i < 8; i++)
1218 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1221 } else if (msr == MSR_MTRRdefType) {
1224 return valid_mtrr_type(data & 0xff);
1225 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1226 for (i = 0; i < 8 ; i++)
1227 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1232 /* variable MTRRs */
1233 return valid_mtrr_type(data & 0xff);
1236 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1238 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1240 if (!mtrr_valid(vcpu, msr, data))
1243 if (msr == MSR_MTRRdefType) {
1244 vcpu->arch.mtrr_state.def_type = data;
1245 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1246 } else if (msr == MSR_MTRRfix64K_00000)
1248 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1249 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1250 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1251 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1252 else if (msr == MSR_IA32_CR_PAT)
1253 vcpu->arch.pat = data;
1254 else { /* Variable MTRRs */
1255 int idx, is_mtrr_mask;
1258 idx = (msr - 0x200) / 2;
1259 is_mtrr_mask = msr - 0x200 - 2 * idx;
1262 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1265 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1269 kvm_mmu_reset_context(vcpu);
1273 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1275 u64 mcg_cap = vcpu->arch.mcg_cap;
1276 unsigned bank_num = mcg_cap & 0xff;
1279 case MSR_IA32_MCG_STATUS:
1280 vcpu->arch.mcg_status = data;
1282 case MSR_IA32_MCG_CTL:
1283 if (!(mcg_cap & MCG_CTL_P))
1285 if (data != 0 && data != ~(u64)0)
1287 vcpu->arch.mcg_ctl = data;
1290 if (msr >= MSR_IA32_MC0_CTL &&
1291 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1292 u32 offset = msr - MSR_IA32_MC0_CTL;
1293 /* only 0 or all 1s can be written to IA32_MCi_CTL
1294 * some Linux kernels though clear bit 10 in bank 4 to
1295 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1296 * this to avoid an uncatched #GP in the guest
1298 if ((offset & 0x3) == 0 &&
1299 data != 0 && (data | (1 << 10)) != ~(u64)0)
1301 vcpu->arch.mce_banks[offset] = data;
1309 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1311 struct kvm *kvm = vcpu->kvm;
1312 int lm = is_long_mode(vcpu);
1313 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1314 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1315 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1316 : kvm->arch.xen_hvm_config.blob_size_32;
1317 u32 page_num = data & ~PAGE_MASK;
1318 u64 page_addr = data & PAGE_MASK;
1323 if (page_num >= blob_size)
1326 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1330 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1332 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1341 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1343 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1346 static bool kvm_hv_msr_partition_wide(u32 msr)
1350 case HV_X64_MSR_GUEST_OS_ID:
1351 case HV_X64_MSR_HYPERCALL:
1359 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1361 struct kvm *kvm = vcpu->kvm;
1364 case HV_X64_MSR_GUEST_OS_ID:
1365 kvm->arch.hv_guest_os_id = data;
1366 /* setting guest os id to zero disables hypercall page */
1367 if (!kvm->arch.hv_guest_os_id)
1368 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1370 case HV_X64_MSR_HYPERCALL: {
1375 /* if guest os id is not set hypercall should remain disabled */
1376 if (!kvm->arch.hv_guest_os_id)
1378 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1379 kvm->arch.hv_hypercall = data;
1382 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1383 addr = gfn_to_hva(kvm, gfn);
1384 if (kvm_is_error_hva(addr))
1386 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1387 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1388 if (copy_to_user((void __user *)addr, instructions, 4))
1390 kvm->arch.hv_hypercall = data;
1394 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1395 "data 0x%llx\n", msr, data);
1401 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1404 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1407 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1408 vcpu->arch.hv_vapic = data;
1411 addr = gfn_to_hva(vcpu->kvm, data >>
1412 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1413 if (kvm_is_error_hva(addr))
1415 if (clear_user((void __user *)addr, PAGE_SIZE))
1417 vcpu->arch.hv_vapic = data;
1420 case HV_X64_MSR_EOI:
1421 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1422 case HV_X64_MSR_ICR:
1423 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1424 case HV_X64_MSR_TPR:
1425 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1427 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1428 "data 0x%llx\n", msr, data);
1435 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1437 gpa_t gpa = data & ~0x3f;
1439 /* Bits 2:5 are resrved, Should be zero */
1443 vcpu->arch.apf.msr_val = data;
1445 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1446 kvm_clear_async_pf_completion_queue(vcpu);
1447 kvm_async_pf_hash_reset(vcpu);
1451 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1454 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1455 kvm_async_pf_wakeup_all(vcpu);
1459 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1461 if (vcpu->arch.time_page) {
1462 kvm_release_page_dirty(vcpu->arch.time_page);
1463 vcpu->arch.time_page = NULL;
1467 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1471 return set_efer(vcpu, data);
1473 data &= ~(u64)0x40; /* ignore flush filter disable */
1474 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1476 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1481 case MSR_FAM10H_MMIO_CONF_BASE:
1483 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1488 case MSR_AMD64_NB_CFG:
1490 case MSR_IA32_DEBUGCTLMSR:
1492 /* We support the non-activated case already */
1494 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1495 /* Values other than LBR and BTF are vendor-specific,
1496 thus reserved and should throw a #GP */
1499 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1502 case MSR_IA32_UCODE_REV:
1503 case MSR_IA32_UCODE_WRITE:
1504 case MSR_VM_HSAVE_PA:
1505 case MSR_AMD64_PATCH_LOADER:
1507 case 0x200 ... 0x2ff:
1508 return set_msr_mtrr(vcpu, msr, data);
1509 case MSR_IA32_APICBASE:
1510 kvm_set_apic_base(vcpu, data);
1512 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1513 return kvm_x2apic_msr_write(vcpu, msr, data);
1514 case MSR_IA32_MISC_ENABLE:
1515 vcpu->arch.ia32_misc_enable_msr = data;
1517 case MSR_KVM_WALL_CLOCK_NEW:
1518 case MSR_KVM_WALL_CLOCK:
1519 vcpu->kvm->arch.wall_clock = data;
1520 kvm_write_wall_clock(vcpu->kvm, data);
1522 case MSR_KVM_SYSTEM_TIME_NEW:
1523 case MSR_KVM_SYSTEM_TIME: {
1524 kvmclock_reset(vcpu);
1526 vcpu->arch.time = data;
1527 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1529 /* we verify if the enable bit is set... */
1533 /* ...but clean it before doing the actual write */
1534 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1536 vcpu->arch.time_page =
1537 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1539 if (is_error_page(vcpu->arch.time_page)) {
1540 kvm_release_page_clean(vcpu->arch.time_page);
1541 vcpu->arch.time_page = NULL;
1545 case MSR_KVM_ASYNC_PF_EN:
1546 if (kvm_pv_enable_async_pf(vcpu, data))
1549 case MSR_IA32_MCG_CTL:
1550 case MSR_IA32_MCG_STATUS:
1551 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1552 return set_msr_mce(vcpu, msr, data);
1554 /* Performance counters are not protected by a CPUID bit,
1555 * so we should check all of them in the generic path for the sake of
1556 * cross vendor migration.
1557 * Writing a zero into the event select MSRs disables them,
1558 * which we perfectly emulate ;-). Any other value should be at least
1559 * reported, some guests depend on them.
1561 case MSR_P6_EVNTSEL0:
1562 case MSR_P6_EVNTSEL1:
1563 case MSR_K7_EVNTSEL0:
1564 case MSR_K7_EVNTSEL1:
1565 case MSR_K7_EVNTSEL2:
1566 case MSR_K7_EVNTSEL3:
1568 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1569 "0x%x data 0x%llx\n", msr, data);
1571 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1572 * so we ignore writes to make it happy.
1574 case MSR_P6_PERFCTR0:
1575 case MSR_P6_PERFCTR1:
1576 case MSR_K7_PERFCTR0:
1577 case MSR_K7_PERFCTR1:
1578 case MSR_K7_PERFCTR2:
1579 case MSR_K7_PERFCTR3:
1580 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1581 "0x%x data 0x%llx\n", msr, data);
1583 case MSR_K7_CLK_CTL:
1585 * Ignore all writes to this no longer documented MSR.
1586 * Writes are only relevant for old K7 processors,
1587 * all pre-dating SVM, but a recommended workaround from
1588 * AMD for these chips. It is possible to speicify the
1589 * affected processor models on the command line, hence
1590 * the need to ignore the workaround.
1593 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1594 if (kvm_hv_msr_partition_wide(msr)) {
1596 mutex_lock(&vcpu->kvm->lock);
1597 r = set_msr_hyperv_pw(vcpu, msr, data);
1598 mutex_unlock(&vcpu->kvm->lock);
1601 return set_msr_hyperv(vcpu, msr, data);
1603 case MSR_IA32_BBL_CR_CTL3:
1604 /* Drop writes to this legacy MSR -- see rdmsr
1605 * counterpart for further detail.
1607 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1610 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1611 return xen_hvm_config(vcpu, data);
1613 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1617 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1624 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1628 * Reads an msr value (of 'msr_index') into 'pdata'.
1629 * Returns 0 on success, non-0 otherwise.
1630 * Assumes vcpu_load() was already called.
1632 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1634 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1637 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1639 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1641 if (!msr_mtrr_valid(msr))
1644 if (msr == MSR_MTRRdefType)
1645 *pdata = vcpu->arch.mtrr_state.def_type +
1646 (vcpu->arch.mtrr_state.enabled << 10);
1647 else if (msr == MSR_MTRRfix64K_00000)
1649 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1650 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1651 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1652 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1653 else if (msr == MSR_IA32_CR_PAT)
1654 *pdata = vcpu->arch.pat;
1655 else { /* Variable MTRRs */
1656 int idx, is_mtrr_mask;
1659 idx = (msr - 0x200) / 2;
1660 is_mtrr_mask = msr - 0x200 - 2 * idx;
1663 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1666 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1673 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1676 u64 mcg_cap = vcpu->arch.mcg_cap;
1677 unsigned bank_num = mcg_cap & 0xff;
1680 case MSR_IA32_P5_MC_ADDR:
1681 case MSR_IA32_P5_MC_TYPE:
1684 case MSR_IA32_MCG_CAP:
1685 data = vcpu->arch.mcg_cap;
1687 case MSR_IA32_MCG_CTL:
1688 if (!(mcg_cap & MCG_CTL_P))
1690 data = vcpu->arch.mcg_ctl;
1692 case MSR_IA32_MCG_STATUS:
1693 data = vcpu->arch.mcg_status;
1696 if (msr >= MSR_IA32_MC0_CTL &&
1697 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1698 u32 offset = msr - MSR_IA32_MC0_CTL;
1699 data = vcpu->arch.mce_banks[offset];
1708 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1711 struct kvm *kvm = vcpu->kvm;
1714 case HV_X64_MSR_GUEST_OS_ID:
1715 data = kvm->arch.hv_guest_os_id;
1717 case HV_X64_MSR_HYPERCALL:
1718 data = kvm->arch.hv_hypercall;
1721 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1729 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1734 case HV_X64_MSR_VP_INDEX: {
1737 kvm_for_each_vcpu(r, v, vcpu->kvm)
1742 case HV_X64_MSR_EOI:
1743 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1744 case HV_X64_MSR_ICR:
1745 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1746 case HV_X64_MSR_TPR:
1747 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1749 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1756 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1761 case MSR_IA32_PLATFORM_ID:
1762 case MSR_IA32_UCODE_REV:
1763 case MSR_IA32_EBL_CR_POWERON:
1764 case MSR_IA32_DEBUGCTLMSR:
1765 case MSR_IA32_LASTBRANCHFROMIP:
1766 case MSR_IA32_LASTBRANCHTOIP:
1767 case MSR_IA32_LASTINTFROMIP:
1768 case MSR_IA32_LASTINTTOIP:
1771 case MSR_VM_HSAVE_PA:
1772 case MSR_P6_PERFCTR0:
1773 case MSR_P6_PERFCTR1:
1774 case MSR_P6_EVNTSEL0:
1775 case MSR_P6_EVNTSEL1:
1776 case MSR_K7_EVNTSEL0:
1777 case MSR_K7_PERFCTR0:
1778 case MSR_K8_INT_PENDING_MSG:
1779 case MSR_AMD64_NB_CFG:
1780 case MSR_FAM10H_MMIO_CONF_BASE:
1784 data = 0x500 | KVM_NR_VAR_MTRR;
1786 case 0x200 ... 0x2ff:
1787 return get_msr_mtrr(vcpu, msr, pdata);
1788 case 0xcd: /* fsb frequency */
1792 * MSR_EBC_FREQUENCY_ID
1793 * Conservative value valid for even the basic CPU models.
1794 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1795 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1796 * and 266MHz for model 3, or 4. Set Core Clock
1797 * Frequency to System Bus Frequency Ratio to 1 (bits
1798 * 31:24) even though these are only valid for CPU
1799 * models > 2, however guests may end up dividing or
1800 * multiplying by zero otherwise.
1802 case MSR_EBC_FREQUENCY_ID:
1805 case MSR_IA32_APICBASE:
1806 data = kvm_get_apic_base(vcpu);
1808 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1809 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1811 case MSR_IA32_MISC_ENABLE:
1812 data = vcpu->arch.ia32_misc_enable_msr;
1814 case MSR_IA32_PERF_STATUS:
1815 /* TSC increment by tick */
1817 /* CPU multiplier */
1818 data |= (((uint64_t)4ULL) << 40);
1821 data = vcpu->arch.efer;
1823 case MSR_KVM_WALL_CLOCK:
1824 case MSR_KVM_WALL_CLOCK_NEW:
1825 data = vcpu->kvm->arch.wall_clock;
1827 case MSR_KVM_SYSTEM_TIME:
1828 case MSR_KVM_SYSTEM_TIME_NEW:
1829 data = vcpu->arch.time;
1831 case MSR_KVM_ASYNC_PF_EN:
1832 data = vcpu->arch.apf.msr_val;
1834 case MSR_IA32_P5_MC_ADDR:
1835 case MSR_IA32_P5_MC_TYPE:
1836 case MSR_IA32_MCG_CAP:
1837 case MSR_IA32_MCG_CTL:
1838 case MSR_IA32_MCG_STATUS:
1839 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1840 return get_msr_mce(vcpu, msr, pdata);
1841 case MSR_K7_CLK_CTL:
1843 * Provide expected ramp-up count for K7. All other
1844 * are set to zero, indicating minimum divisors for
1847 * This prevents guest kernels on AMD host with CPU
1848 * type 6, model 8 and higher from exploding due to
1849 * the rdmsr failing.
1853 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1854 if (kvm_hv_msr_partition_wide(msr)) {
1856 mutex_lock(&vcpu->kvm->lock);
1857 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1858 mutex_unlock(&vcpu->kvm->lock);
1861 return get_msr_hyperv(vcpu, msr, pdata);
1863 case MSR_IA32_BBL_CR_CTL3:
1864 /* This legacy MSR exists but isn't fully documented in current
1865 * silicon. It is however accessed by winxp in very narrow
1866 * scenarios where it sets bit #19, itself documented as
1867 * a "reserved" bit. Best effort attempt to source coherent
1868 * read data here should the balance of the register be
1869 * interpreted by the guest:
1871 * L2 cache control register 3: 64GB range, 256KB size,
1872 * enabled, latency 0x1, configured
1878 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1881 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1889 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1892 * Read or write a bunch of msrs. All parameters are kernel addresses.
1894 * @return number of msrs set successfully.
1896 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1897 struct kvm_msr_entry *entries,
1898 int (*do_msr)(struct kvm_vcpu *vcpu,
1899 unsigned index, u64 *data))
1903 idx = srcu_read_lock(&vcpu->kvm->srcu);
1904 for (i = 0; i < msrs->nmsrs; ++i)
1905 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1907 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1913 * Read or write a bunch of msrs. Parameters are user addresses.
1915 * @return number of msrs set successfully.
1917 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1918 int (*do_msr)(struct kvm_vcpu *vcpu,
1919 unsigned index, u64 *data),
1922 struct kvm_msrs msrs;
1923 struct kvm_msr_entry *entries;
1928 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1932 if (msrs.nmsrs >= MAX_IO_MSRS)
1936 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1937 entries = kmalloc(size, GFP_KERNEL);
1942 if (copy_from_user(entries, user_msrs->entries, size))
1945 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1950 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1961 int kvm_dev_ioctl_check_extension(long ext)
1966 case KVM_CAP_IRQCHIP:
1968 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1969 case KVM_CAP_SET_TSS_ADDR:
1970 case KVM_CAP_EXT_CPUID:
1971 case KVM_CAP_CLOCKSOURCE:
1973 case KVM_CAP_NOP_IO_DELAY:
1974 case KVM_CAP_MP_STATE:
1975 case KVM_CAP_SYNC_MMU:
1976 case KVM_CAP_USER_NMI:
1977 case KVM_CAP_REINJECT_CONTROL:
1978 case KVM_CAP_IRQ_INJECT_STATUS:
1979 case KVM_CAP_ASSIGN_DEV_IRQ:
1981 case KVM_CAP_IOEVENTFD:
1983 case KVM_CAP_PIT_STATE2:
1984 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1985 case KVM_CAP_XEN_HVM:
1986 case KVM_CAP_ADJUST_CLOCK:
1987 case KVM_CAP_VCPU_EVENTS:
1988 case KVM_CAP_HYPERV:
1989 case KVM_CAP_HYPERV_VAPIC:
1990 case KVM_CAP_HYPERV_SPIN:
1991 case KVM_CAP_PCI_SEGMENT:
1992 case KVM_CAP_DEBUGREGS:
1993 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1995 case KVM_CAP_ASYNC_PF:
1998 case KVM_CAP_COALESCED_MMIO:
1999 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2002 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2004 case KVM_CAP_NR_VCPUS:
2007 case KVM_CAP_NR_MEMSLOTS:
2008 r = KVM_MEMORY_SLOTS;
2010 case KVM_CAP_PV_MMU: /* obsolete */
2017 r = KVM_MAX_MCE_BANKS;
2030 long kvm_arch_dev_ioctl(struct file *filp,
2031 unsigned int ioctl, unsigned long arg)
2033 void __user *argp = (void __user *)arg;
2037 case KVM_GET_MSR_INDEX_LIST: {
2038 struct kvm_msr_list __user *user_msr_list = argp;
2039 struct kvm_msr_list msr_list;
2043 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2046 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2047 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2050 if (n < msr_list.nmsrs)
2053 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2054 num_msrs_to_save * sizeof(u32)))
2056 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2058 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2063 case KVM_GET_SUPPORTED_CPUID: {
2064 struct kvm_cpuid2 __user *cpuid_arg = argp;
2065 struct kvm_cpuid2 cpuid;
2068 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2070 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2071 cpuid_arg->entries);
2076 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2081 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2084 mce_cap = KVM_MCE_CAP_SUPPORTED;
2086 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2098 static void wbinvd_ipi(void *garbage)
2103 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2105 return vcpu->kvm->arch.iommu_domain &&
2106 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2109 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2111 /* Address WBINVD may be executed by guest */
2112 if (need_emulate_wbinvd(vcpu)) {
2113 if (kvm_x86_ops->has_wbinvd_exit())
2114 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2115 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2116 smp_call_function_single(vcpu->cpu,
2117 wbinvd_ipi, NULL, 1);
2120 kvm_x86_ops->vcpu_load(vcpu, cpu);
2121 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2122 /* Make sure TSC doesn't go backwards */
2123 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2124 native_read_tsc() - vcpu->arch.last_host_tsc;
2126 mark_tsc_unstable("KVM discovered backwards TSC");
2127 if (check_tsc_unstable()) {
2128 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2129 vcpu->arch.tsc_catchup = 1;
2131 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2132 if (vcpu->cpu != cpu)
2133 kvm_migrate_timers(vcpu);
2138 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2140 kvm_x86_ops->vcpu_put(vcpu);
2141 kvm_put_guest_fpu(vcpu);
2142 vcpu->arch.last_host_tsc = native_read_tsc();
2145 static int is_efer_nx(void)
2147 unsigned long long efer = 0;
2149 rdmsrl_safe(MSR_EFER, &efer);
2150 return efer & EFER_NX;
2153 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2156 struct kvm_cpuid_entry2 *e, *entry;
2159 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2160 e = &vcpu->arch.cpuid_entries[i];
2161 if (e->function == 0x80000001) {
2166 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2167 entry->edx &= ~(1 << 20);
2168 printk(KERN_INFO "kvm: guest NX capability removed\n");
2172 /* when an old userspace process fills a new kernel module */
2173 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2174 struct kvm_cpuid *cpuid,
2175 struct kvm_cpuid_entry __user *entries)
2178 struct kvm_cpuid_entry *cpuid_entries;
2181 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2184 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2188 if (copy_from_user(cpuid_entries, entries,
2189 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2191 for (i = 0; i < cpuid->nent; i++) {
2192 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2193 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2194 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2195 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2196 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2197 vcpu->arch.cpuid_entries[i].index = 0;
2198 vcpu->arch.cpuid_entries[i].flags = 0;
2199 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2200 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2201 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2203 vcpu->arch.cpuid_nent = cpuid->nent;
2204 cpuid_fix_nx_cap(vcpu);
2206 kvm_apic_set_version(vcpu);
2207 kvm_x86_ops->cpuid_update(vcpu);
2211 vfree(cpuid_entries);
2216 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2217 struct kvm_cpuid2 *cpuid,
2218 struct kvm_cpuid_entry2 __user *entries)
2223 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2226 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2227 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2229 vcpu->arch.cpuid_nent = cpuid->nent;
2230 kvm_apic_set_version(vcpu);
2231 kvm_x86_ops->cpuid_update(vcpu);
2239 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2240 struct kvm_cpuid2 *cpuid,
2241 struct kvm_cpuid_entry2 __user *entries)
2246 if (cpuid->nent < vcpu->arch.cpuid_nent)
2249 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2250 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2255 cpuid->nent = vcpu->arch.cpuid_nent;
2259 static void cpuid_mask(u32 *word, int wordnum)
2261 *word &= boot_cpu_data.x86_capability[wordnum];
2264 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2267 entry->function = function;
2268 entry->index = index;
2269 cpuid_count(entry->function, entry->index,
2270 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2274 #define F(x) bit(X86_FEATURE_##x)
2276 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2277 u32 index, int *nent, int maxnent)
2279 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2280 #ifdef CONFIG_X86_64
2281 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2283 unsigned f_lm = F(LM);
2285 unsigned f_gbpages = 0;
2288 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2291 const u32 kvm_supported_word0_x86_features =
2292 F(FPU) | F(VME) | F(DE) | F(PSE) |
2293 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2294 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2295 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2296 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2297 0 /* Reserved, DS, ACPI */ | F(MMX) |
2298 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2299 0 /* HTT, TM, Reserved, PBE */;
2300 /* cpuid 0x80000001.edx */
2301 const u32 kvm_supported_word1_x86_features =
2302 F(FPU) | F(VME) | F(DE) | F(PSE) |
2303 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2304 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2305 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2306 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2307 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2308 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2309 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2311 const u32 kvm_supported_word4_x86_features =
2312 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2313 0 /* DS-CPL, VMX, SMX, EST */ |
2314 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2315 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2316 0 /* Reserved, DCA */ | F(XMM4_1) |
2317 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2318 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2320 /* cpuid 0x80000001.ecx */
2321 const u32 kvm_supported_word6_x86_features =
2322 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2323 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2324 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2325 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2327 /* all calls to cpuid_count() should be made on the same cpu */
2329 do_cpuid_1_ent(entry, function, index);
2334 entry->eax = min(entry->eax, (u32)0xd);
2337 entry->edx &= kvm_supported_word0_x86_features;
2338 cpuid_mask(&entry->edx, 0);
2339 entry->ecx &= kvm_supported_word4_x86_features;
2340 cpuid_mask(&entry->ecx, 4);
2341 /* we support x2apic emulation even if host does not support
2342 * it since we emulate x2apic in software */
2343 entry->ecx |= F(X2APIC);
2345 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2346 * may return different values. This forces us to get_cpu() before
2347 * issuing the first command, and also to emulate this annoying behavior
2348 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2350 int t, times = entry->eax & 0xff;
2352 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2353 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2354 for (t = 1; t < times && *nent < maxnent; ++t) {
2355 do_cpuid_1_ent(&entry[t], function, 0);
2356 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2361 /* function 4 and 0xb have additional index. */
2365 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2366 /* read more entries until cache_type is zero */
2367 for (i = 1; *nent < maxnent; ++i) {
2368 cache_type = entry[i - 1].eax & 0x1f;
2371 do_cpuid_1_ent(&entry[i], function, i);
2373 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2381 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2382 /* read more entries until level_type is zero */
2383 for (i = 1; *nent < maxnent; ++i) {
2384 level_type = entry[i - 1].ecx & 0xff00;
2387 do_cpuid_1_ent(&entry[i], function, i);
2389 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2397 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2398 for (i = 1; *nent < maxnent && i < 64; ++i) {
2399 if (entry[i].eax == 0)
2401 do_cpuid_1_ent(&entry[i], function, i);
2403 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2408 case KVM_CPUID_SIGNATURE: {
2409 char signature[12] = "KVMKVMKVM\0\0";
2410 u32 *sigptr = (u32 *)signature;
2412 entry->ebx = sigptr[0];
2413 entry->ecx = sigptr[1];
2414 entry->edx = sigptr[2];
2417 case KVM_CPUID_FEATURES:
2418 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2419 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2420 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2421 (1 << KVM_FEATURE_ASYNC_PF) |
2422 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2428 entry->eax = min(entry->eax, 0x8000001a);
2431 entry->edx &= kvm_supported_word1_x86_features;
2432 cpuid_mask(&entry->edx, 1);
2433 entry->ecx &= kvm_supported_word6_x86_features;
2434 cpuid_mask(&entry->ecx, 6);
2438 kvm_x86_ops->set_supported_cpuid(function, entry);
2445 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2446 struct kvm_cpuid_entry2 __user *entries)
2448 struct kvm_cpuid_entry2 *cpuid_entries;
2449 int limit, nent = 0, r = -E2BIG;
2452 if (cpuid->nent < 1)
2454 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2455 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2457 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2461 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2462 limit = cpuid_entries[0].eax;
2463 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2464 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2465 &nent, cpuid->nent);
2467 if (nent >= cpuid->nent)
2470 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2471 limit = cpuid_entries[nent - 1].eax;
2472 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2473 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2474 &nent, cpuid->nent);
2479 if (nent >= cpuid->nent)
2482 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2486 if (nent >= cpuid->nent)
2489 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2493 if (nent >= cpuid->nent)
2497 if (copy_to_user(entries, cpuid_entries,
2498 nent * sizeof(struct kvm_cpuid_entry2)))
2504 vfree(cpuid_entries);
2509 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2510 struct kvm_lapic_state *s)
2512 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2517 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2518 struct kvm_lapic_state *s)
2520 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2521 kvm_apic_post_state_restore(vcpu);
2522 update_cr8_intercept(vcpu);
2527 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2528 struct kvm_interrupt *irq)
2530 if (irq->irq < 0 || irq->irq >= 256)
2532 if (irqchip_in_kernel(vcpu->kvm))
2535 kvm_queue_interrupt(vcpu, irq->irq, false);
2536 kvm_make_request(KVM_REQ_EVENT, vcpu);
2541 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2543 kvm_inject_nmi(vcpu);
2548 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2549 struct kvm_tpr_access_ctl *tac)
2553 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2557 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2561 unsigned bank_num = mcg_cap & 0xff, bank;
2564 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2566 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2569 vcpu->arch.mcg_cap = mcg_cap;
2570 /* Init IA32_MCG_CTL to all 1s */
2571 if (mcg_cap & MCG_CTL_P)
2572 vcpu->arch.mcg_ctl = ~(u64)0;
2573 /* Init IA32_MCi_CTL to all 1s */
2574 for (bank = 0; bank < bank_num; bank++)
2575 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2580 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2581 struct kvm_x86_mce *mce)
2583 u64 mcg_cap = vcpu->arch.mcg_cap;
2584 unsigned bank_num = mcg_cap & 0xff;
2585 u64 *banks = vcpu->arch.mce_banks;
2587 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2590 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2591 * reporting is disabled
2593 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2594 vcpu->arch.mcg_ctl != ~(u64)0)
2596 banks += 4 * mce->bank;
2598 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2599 * reporting is disabled for the bank
2601 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2603 if (mce->status & MCI_STATUS_UC) {
2604 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2605 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2606 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2609 if (banks[1] & MCI_STATUS_VAL)
2610 mce->status |= MCI_STATUS_OVER;
2611 banks[2] = mce->addr;
2612 banks[3] = mce->misc;
2613 vcpu->arch.mcg_status = mce->mcg_status;
2614 banks[1] = mce->status;
2615 kvm_queue_exception(vcpu, MC_VECTOR);
2616 } else if (!(banks[1] & MCI_STATUS_VAL)
2617 || !(banks[1] & MCI_STATUS_UC)) {
2618 if (banks[1] & MCI_STATUS_VAL)
2619 mce->status |= MCI_STATUS_OVER;
2620 banks[2] = mce->addr;
2621 banks[3] = mce->misc;
2622 banks[1] = mce->status;
2624 banks[1] |= MCI_STATUS_OVER;
2628 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2629 struct kvm_vcpu_events *events)
2631 events->exception.injected =
2632 vcpu->arch.exception.pending &&
2633 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2634 events->exception.nr = vcpu->arch.exception.nr;
2635 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2636 events->exception.pad = 0;
2637 events->exception.error_code = vcpu->arch.exception.error_code;
2639 events->interrupt.injected =
2640 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2641 events->interrupt.nr = vcpu->arch.interrupt.nr;
2642 events->interrupt.soft = 0;
2643 events->interrupt.shadow =
2644 kvm_x86_ops->get_interrupt_shadow(vcpu,
2645 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2647 events->nmi.injected = vcpu->arch.nmi_injected;
2648 events->nmi.pending = vcpu->arch.nmi_pending;
2649 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2650 events->nmi.pad = 0;
2652 events->sipi_vector = vcpu->arch.sipi_vector;
2654 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2655 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2656 | KVM_VCPUEVENT_VALID_SHADOW);
2657 memset(&events->reserved, 0, sizeof(events->reserved));
2660 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2661 struct kvm_vcpu_events *events)
2663 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2664 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2665 | KVM_VCPUEVENT_VALID_SHADOW))
2668 vcpu->arch.exception.pending = events->exception.injected;
2669 vcpu->arch.exception.nr = events->exception.nr;
2670 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2671 vcpu->arch.exception.error_code = events->exception.error_code;
2673 vcpu->arch.interrupt.pending = events->interrupt.injected;
2674 vcpu->arch.interrupt.nr = events->interrupt.nr;
2675 vcpu->arch.interrupt.soft = events->interrupt.soft;
2676 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2677 kvm_x86_ops->set_interrupt_shadow(vcpu,
2678 events->interrupt.shadow);
2680 vcpu->arch.nmi_injected = events->nmi.injected;
2681 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2682 vcpu->arch.nmi_pending = events->nmi.pending;
2683 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2685 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2686 vcpu->arch.sipi_vector = events->sipi_vector;
2688 kvm_make_request(KVM_REQ_EVENT, vcpu);
2693 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2694 struct kvm_debugregs *dbgregs)
2696 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2697 dbgregs->dr6 = vcpu->arch.dr6;
2698 dbgregs->dr7 = vcpu->arch.dr7;
2700 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2703 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2704 struct kvm_debugregs *dbgregs)
2709 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2710 vcpu->arch.dr6 = dbgregs->dr6;
2711 vcpu->arch.dr7 = dbgregs->dr7;
2716 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2717 struct kvm_xsave *guest_xsave)
2720 memcpy(guest_xsave->region,
2721 &vcpu->arch.guest_fpu.state->xsave,
2724 memcpy(guest_xsave->region,
2725 &vcpu->arch.guest_fpu.state->fxsave,
2726 sizeof(struct i387_fxsave_struct));
2727 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2732 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2733 struct kvm_xsave *guest_xsave)
2736 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2739 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2740 guest_xsave->region, xstate_size);
2742 if (xstate_bv & ~XSTATE_FPSSE)
2744 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2745 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2750 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2751 struct kvm_xcrs *guest_xcrs)
2753 if (!cpu_has_xsave) {
2754 guest_xcrs->nr_xcrs = 0;
2758 guest_xcrs->nr_xcrs = 1;
2759 guest_xcrs->flags = 0;
2760 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2761 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2764 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2765 struct kvm_xcrs *guest_xcrs)
2772 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2775 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2776 /* Only support XCR0 currently */
2777 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2778 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2779 guest_xcrs->xcrs[0].value);
2787 long kvm_arch_vcpu_ioctl(struct file *filp,
2788 unsigned int ioctl, unsigned long arg)
2790 struct kvm_vcpu *vcpu = filp->private_data;
2791 void __user *argp = (void __user *)arg;
2794 struct kvm_lapic_state *lapic;
2795 struct kvm_xsave *xsave;
2796 struct kvm_xcrs *xcrs;
2802 case KVM_GET_LAPIC: {
2804 if (!vcpu->arch.apic)
2806 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2811 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2815 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2820 case KVM_SET_LAPIC: {
2822 if (!vcpu->arch.apic)
2824 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2829 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2831 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2837 case KVM_INTERRUPT: {
2838 struct kvm_interrupt irq;
2841 if (copy_from_user(&irq, argp, sizeof irq))
2843 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2850 r = kvm_vcpu_ioctl_nmi(vcpu);
2856 case KVM_SET_CPUID: {
2857 struct kvm_cpuid __user *cpuid_arg = argp;
2858 struct kvm_cpuid cpuid;
2861 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2863 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2868 case KVM_SET_CPUID2: {
2869 struct kvm_cpuid2 __user *cpuid_arg = argp;
2870 struct kvm_cpuid2 cpuid;
2873 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2875 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2876 cpuid_arg->entries);
2881 case KVM_GET_CPUID2: {
2882 struct kvm_cpuid2 __user *cpuid_arg = argp;
2883 struct kvm_cpuid2 cpuid;
2886 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2888 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2889 cpuid_arg->entries);
2893 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2899 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2902 r = msr_io(vcpu, argp, do_set_msr, 0);
2904 case KVM_TPR_ACCESS_REPORTING: {
2905 struct kvm_tpr_access_ctl tac;
2908 if (copy_from_user(&tac, argp, sizeof tac))
2910 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2914 if (copy_to_user(argp, &tac, sizeof tac))
2919 case KVM_SET_VAPIC_ADDR: {
2920 struct kvm_vapic_addr va;
2923 if (!irqchip_in_kernel(vcpu->kvm))
2926 if (copy_from_user(&va, argp, sizeof va))
2929 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2932 case KVM_X86_SETUP_MCE: {
2936 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2938 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2941 case KVM_X86_SET_MCE: {
2942 struct kvm_x86_mce mce;
2945 if (copy_from_user(&mce, argp, sizeof mce))
2947 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2950 case KVM_GET_VCPU_EVENTS: {
2951 struct kvm_vcpu_events events;
2953 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2956 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2961 case KVM_SET_VCPU_EVENTS: {
2962 struct kvm_vcpu_events events;
2965 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2968 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2971 case KVM_GET_DEBUGREGS: {
2972 struct kvm_debugregs dbgregs;
2974 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2977 if (copy_to_user(argp, &dbgregs,
2978 sizeof(struct kvm_debugregs)))
2983 case KVM_SET_DEBUGREGS: {
2984 struct kvm_debugregs dbgregs;
2987 if (copy_from_user(&dbgregs, argp,
2988 sizeof(struct kvm_debugregs)))
2991 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2994 case KVM_GET_XSAVE: {
2995 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3000 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3003 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3008 case KVM_SET_XSAVE: {
3009 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3015 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3018 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3021 case KVM_GET_XCRS: {
3022 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3027 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3030 if (copy_to_user(argp, u.xcrs,
3031 sizeof(struct kvm_xcrs)))
3036 case KVM_SET_XCRS: {
3037 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3043 if (copy_from_user(u.xcrs, argp,
3044 sizeof(struct kvm_xcrs)))
3047 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3058 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3062 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3064 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3068 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3071 kvm->arch.ept_identity_map_addr = ident_addr;
3075 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3076 u32 kvm_nr_mmu_pages)
3078 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3081 mutex_lock(&kvm->slots_lock);
3082 spin_lock(&kvm->mmu_lock);
3084 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3085 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3087 spin_unlock(&kvm->mmu_lock);
3088 mutex_unlock(&kvm->slots_lock);
3092 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3094 return kvm->arch.n_max_mmu_pages;
3097 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3102 switch (chip->chip_id) {
3103 case KVM_IRQCHIP_PIC_MASTER:
3104 memcpy(&chip->chip.pic,
3105 &pic_irqchip(kvm)->pics[0],
3106 sizeof(struct kvm_pic_state));
3108 case KVM_IRQCHIP_PIC_SLAVE:
3109 memcpy(&chip->chip.pic,
3110 &pic_irqchip(kvm)->pics[1],
3111 sizeof(struct kvm_pic_state));
3113 case KVM_IRQCHIP_IOAPIC:
3114 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3123 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3128 switch (chip->chip_id) {
3129 case KVM_IRQCHIP_PIC_MASTER:
3130 spin_lock(&pic_irqchip(kvm)->lock);
3131 memcpy(&pic_irqchip(kvm)->pics[0],
3133 sizeof(struct kvm_pic_state));
3134 spin_unlock(&pic_irqchip(kvm)->lock);
3136 case KVM_IRQCHIP_PIC_SLAVE:
3137 spin_lock(&pic_irqchip(kvm)->lock);
3138 memcpy(&pic_irqchip(kvm)->pics[1],
3140 sizeof(struct kvm_pic_state));
3141 spin_unlock(&pic_irqchip(kvm)->lock);
3143 case KVM_IRQCHIP_IOAPIC:
3144 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3150 kvm_pic_update_irq(pic_irqchip(kvm));
3154 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3158 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3159 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3160 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3164 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3168 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3169 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3170 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3171 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3175 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3179 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3180 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3181 sizeof(ps->channels));
3182 ps->flags = kvm->arch.vpit->pit_state.flags;
3183 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3184 memset(&ps->reserved, 0, sizeof(ps->reserved));
3188 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3190 int r = 0, start = 0;
3191 u32 prev_legacy, cur_legacy;
3192 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3193 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3194 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3195 if (!prev_legacy && cur_legacy)
3197 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3198 sizeof(kvm->arch.vpit->pit_state.channels));
3199 kvm->arch.vpit->pit_state.flags = ps->flags;
3200 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3201 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3205 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3206 struct kvm_reinject_control *control)
3208 if (!kvm->arch.vpit)
3210 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3211 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3212 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3217 * Get (and clear) the dirty memory log for a memory slot.
3219 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3220 struct kvm_dirty_log *log)
3223 struct kvm_memory_slot *memslot;
3225 unsigned long is_dirty = 0;
3227 mutex_lock(&kvm->slots_lock);
3230 if (log->slot >= KVM_MEMORY_SLOTS)
3233 memslot = &kvm->memslots->memslots[log->slot];
3235 if (!memslot->dirty_bitmap)
3238 n = kvm_dirty_bitmap_bytes(memslot);
3240 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3241 is_dirty = memslot->dirty_bitmap[i];
3243 /* If nothing is dirty, don't bother messing with page tables. */
3245 struct kvm_memslots *slots, *old_slots;
3246 unsigned long *dirty_bitmap;
3248 dirty_bitmap = memslot->dirty_bitmap_head;
3249 if (memslot->dirty_bitmap == dirty_bitmap)
3250 dirty_bitmap += n / sizeof(long);
3251 memset(dirty_bitmap, 0, n);
3254 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3257 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3258 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3259 slots->generation++;
3261 old_slots = kvm->memslots;
3262 rcu_assign_pointer(kvm->memslots, slots);
3263 synchronize_srcu_expedited(&kvm->srcu);
3264 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3267 spin_lock(&kvm->mmu_lock);
3268 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3269 spin_unlock(&kvm->mmu_lock);
3272 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3276 if (clear_user(log->dirty_bitmap, n))
3282 mutex_unlock(&kvm->slots_lock);
3286 long kvm_arch_vm_ioctl(struct file *filp,
3287 unsigned int ioctl, unsigned long arg)
3289 struct kvm *kvm = filp->private_data;
3290 void __user *argp = (void __user *)arg;
3293 * This union makes it completely explicit to gcc-3.x
3294 * that these two variables' stack usage should be
3295 * combined, not added together.
3298 struct kvm_pit_state ps;
3299 struct kvm_pit_state2 ps2;
3300 struct kvm_pit_config pit_config;
3304 case KVM_SET_TSS_ADDR:
3305 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3309 case KVM_SET_IDENTITY_MAP_ADDR: {
3313 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3315 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3320 case KVM_SET_NR_MMU_PAGES:
3321 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3325 case KVM_GET_NR_MMU_PAGES:
3326 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3328 case KVM_CREATE_IRQCHIP: {
3329 struct kvm_pic *vpic;
3331 mutex_lock(&kvm->lock);
3334 goto create_irqchip_unlock;
3336 vpic = kvm_create_pic(kvm);
3338 r = kvm_ioapic_init(kvm);
3340 mutex_lock(&kvm->slots_lock);
3341 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3343 mutex_unlock(&kvm->slots_lock);
3345 goto create_irqchip_unlock;
3348 goto create_irqchip_unlock;
3350 kvm->arch.vpic = vpic;
3352 r = kvm_setup_default_irq_routing(kvm);
3354 mutex_lock(&kvm->slots_lock);
3355 mutex_lock(&kvm->irq_lock);
3356 kvm_ioapic_destroy(kvm);
3357 kvm_destroy_pic(kvm);
3358 mutex_unlock(&kvm->irq_lock);
3359 mutex_unlock(&kvm->slots_lock);
3361 create_irqchip_unlock:
3362 mutex_unlock(&kvm->lock);
3365 case KVM_CREATE_PIT:
3366 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3368 case KVM_CREATE_PIT2:
3370 if (copy_from_user(&u.pit_config, argp,
3371 sizeof(struct kvm_pit_config)))
3374 mutex_lock(&kvm->slots_lock);
3377 goto create_pit_unlock;
3379 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3383 mutex_unlock(&kvm->slots_lock);
3385 case KVM_IRQ_LINE_STATUS:
3386 case KVM_IRQ_LINE: {
3387 struct kvm_irq_level irq_event;
3390 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3393 if (irqchip_in_kernel(kvm)) {
3395 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3396 irq_event.irq, irq_event.level);
3397 if (ioctl == KVM_IRQ_LINE_STATUS) {
3399 irq_event.status = status;
3400 if (copy_to_user(argp, &irq_event,
3408 case KVM_GET_IRQCHIP: {
3409 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3410 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3416 if (copy_from_user(chip, argp, sizeof *chip))
3417 goto get_irqchip_out;
3419 if (!irqchip_in_kernel(kvm))
3420 goto get_irqchip_out;
3421 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3423 goto get_irqchip_out;
3425 if (copy_to_user(argp, chip, sizeof *chip))
3426 goto get_irqchip_out;
3434 case KVM_SET_IRQCHIP: {
3435 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3436 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3442 if (copy_from_user(chip, argp, sizeof *chip))
3443 goto set_irqchip_out;
3445 if (!irqchip_in_kernel(kvm))
3446 goto set_irqchip_out;
3447 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3449 goto set_irqchip_out;
3459 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3462 if (!kvm->arch.vpit)
3464 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3468 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3475 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3478 if (!kvm->arch.vpit)
3480 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3486 case KVM_GET_PIT2: {
3488 if (!kvm->arch.vpit)
3490 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3494 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3499 case KVM_SET_PIT2: {
3501 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3504 if (!kvm->arch.vpit)
3506 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3512 case KVM_REINJECT_CONTROL: {
3513 struct kvm_reinject_control control;
3515 if (copy_from_user(&control, argp, sizeof(control)))
3517 r = kvm_vm_ioctl_reinject(kvm, &control);
3523 case KVM_XEN_HVM_CONFIG: {
3525 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3526 sizeof(struct kvm_xen_hvm_config)))
3529 if (kvm->arch.xen_hvm_config.flags)
3534 case KVM_SET_CLOCK: {
3535 struct kvm_clock_data user_ns;
3540 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3548 local_irq_disable();
3549 now_ns = get_kernel_ns();
3550 delta = user_ns.clock - now_ns;
3552 kvm->arch.kvmclock_offset = delta;
3555 case KVM_GET_CLOCK: {
3556 struct kvm_clock_data user_ns;
3559 local_irq_disable();
3560 now_ns = get_kernel_ns();
3561 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3564 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3567 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3580 static void kvm_init_msr_list(void)
3585 /* skip the first msrs in the list. KVM-specific */
3586 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3587 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3590 msrs_to_save[j] = msrs_to_save[i];
3593 num_msrs_to_save = j;
3596 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3604 if (!(vcpu->arch.apic &&
3605 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3606 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3617 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3624 if (!(vcpu->arch.apic &&
3625 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3626 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3628 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3638 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3639 struct kvm_segment *var, int seg)
3641 kvm_x86_ops->set_segment(vcpu, var, seg);
3644 void kvm_get_segment(struct kvm_vcpu *vcpu,
3645 struct kvm_segment *var, int seg)
3647 kvm_x86_ops->get_segment(vcpu, var, seg);
3650 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3655 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3658 struct x86_exception exception;
3660 BUG_ON(!mmu_is_nested(vcpu));
3662 /* NPT walks are always user-walks */
3663 access |= PFERR_USER_MASK;
3664 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3669 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3670 struct x86_exception *exception)
3672 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3673 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3676 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3677 struct x86_exception *exception)
3679 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3680 access |= PFERR_FETCH_MASK;
3681 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3684 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3685 struct x86_exception *exception)
3687 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3688 access |= PFERR_WRITE_MASK;
3689 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3692 /* uses this to access any guest's mapped memory without checking CPL */
3693 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3694 struct x86_exception *exception)
3696 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3699 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3700 struct kvm_vcpu *vcpu, u32 access,
3701 struct x86_exception *exception)
3704 int r = X86EMUL_CONTINUE;
3707 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3709 unsigned offset = addr & (PAGE_SIZE-1);
3710 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3713 if (gpa == UNMAPPED_GVA)
3714 return X86EMUL_PROPAGATE_FAULT;
3715 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3717 r = X86EMUL_IO_NEEDED;
3729 /* used for instruction fetching */
3730 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3731 struct kvm_vcpu *vcpu,
3732 struct x86_exception *exception)
3734 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3735 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3736 access | PFERR_FETCH_MASK,
3740 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3741 struct kvm_vcpu *vcpu,
3742 struct x86_exception *exception)
3744 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3745 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3749 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3750 struct kvm_vcpu *vcpu,
3751 struct x86_exception *exception)
3753 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3756 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3758 struct kvm_vcpu *vcpu,
3759 struct x86_exception *exception)
3762 int r = X86EMUL_CONTINUE;
3765 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3768 unsigned offset = addr & (PAGE_SIZE-1);
3769 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3772 if (gpa == UNMAPPED_GVA)
3773 return X86EMUL_PROPAGATE_FAULT;
3774 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3776 r = X86EMUL_IO_NEEDED;
3788 static int emulator_read_emulated(unsigned long addr,
3791 struct x86_exception *exception,
3792 struct kvm_vcpu *vcpu)
3797 if (vcpu->mmio_read_completed) {
3798 memcpy(val, vcpu->mmio_data, bytes);
3799 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3800 vcpu->mmio_phys_addr, *(u64 *)val);
3801 vcpu->mmio_read_completed = 0;
3802 return X86EMUL_CONTINUE;
3805 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, exception);
3807 if (gpa == UNMAPPED_GVA)
3808 return X86EMUL_PROPAGATE_FAULT;
3810 /* For APIC access vmexit */
3811 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3814 if (kvm_read_guest_virt(addr, val, bytes, vcpu, exception)
3815 == X86EMUL_CONTINUE)
3816 return X86EMUL_CONTINUE;
3820 * Is this MMIO handled locally?
3822 handled = vcpu_mmio_read(vcpu, gpa, bytes, val);
3824 if (handled == bytes)
3825 return X86EMUL_CONTINUE;
3831 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3833 vcpu->mmio_needed = 1;
3834 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3835 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3836 vcpu->mmio_size = bytes;
3837 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3838 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3839 vcpu->mmio_index = 0;
3841 return X86EMUL_IO_NEEDED;
3844 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3845 const void *val, int bytes)
3849 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3852 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3856 static int emulator_write_emulated_onepage(unsigned long addr,
3859 struct x86_exception *exception,
3860 struct kvm_vcpu *vcpu)
3865 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, exception);
3867 if (gpa == UNMAPPED_GVA)
3868 return X86EMUL_PROPAGATE_FAULT;
3870 /* For APIC access vmexit */
3871 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3874 if (emulator_write_phys(vcpu, gpa, val, bytes))
3875 return X86EMUL_CONTINUE;
3878 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3880 * Is this MMIO handled locally?
3882 handled = vcpu_mmio_write(vcpu, gpa, bytes, val);
3883 if (handled == bytes)
3884 return X86EMUL_CONTINUE;
3890 vcpu->mmio_needed = 1;
3891 memcpy(vcpu->mmio_data, val, bytes);
3892 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3893 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3894 vcpu->mmio_size = bytes;
3895 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3896 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3897 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3898 vcpu->mmio_index = 0;
3900 return X86EMUL_CONTINUE;
3903 int emulator_write_emulated(unsigned long addr,
3906 struct x86_exception *exception,
3907 struct kvm_vcpu *vcpu)
3909 /* Crossing a page boundary? */
3910 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3913 now = -addr & ~PAGE_MASK;
3914 rc = emulator_write_emulated_onepage(addr, val, now, exception,
3916 if (rc != X86EMUL_CONTINUE)
3922 return emulator_write_emulated_onepage(addr, val, bytes, exception,
3926 #define CMPXCHG_TYPE(t, ptr, old, new) \
3927 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3929 #ifdef CONFIG_X86_64
3930 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3932 # define CMPXCHG64(ptr, old, new) \
3933 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3936 static int emulator_cmpxchg_emulated(unsigned long addr,
3940 struct x86_exception *exception,
3941 struct kvm_vcpu *vcpu)
3948 /* guests cmpxchg8b have to be emulated atomically */
3949 if (bytes > 8 || (bytes & (bytes - 1)))
3952 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3954 if (gpa == UNMAPPED_GVA ||
3955 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3958 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3961 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3962 if (is_error_page(page)) {
3963 kvm_release_page_clean(page);
3967 kaddr = kmap_atomic(page, KM_USER0);
3968 kaddr += offset_in_page(gpa);
3971 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3974 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3977 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3980 exchanged = CMPXCHG64(kaddr, old, new);
3985 kunmap_atomic(kaddr, KM_USER0);
3986 kvm_release_page_dirty(page);
3989 return X86EMUL_CMPXCHG_FAILED;
3991 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3993 return X86EMUL_CONTINUE;
3996 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3998 return emulator_write_emulated(addr, new, bytes, exception, vcpu);
4001 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4003 /* TODO: String I/O for in kernel device */
4006 if (vcpu->arch.pio.in)
4007 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4008 vcpu->arch.pio.size, pd);
4010 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4011 vcpu->arch.pio.port, vcpu->arch.pio.size,
4017 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
4018 unsigned int count, struct kvm_vcpu *vcpu)
4020 if (vcpu->arch.pio.count)
4023 trace_kvm_pio(0, port, size, count);
4025 vcpu->arch.pio.port = port;
4026 vcpu->arch.pio.in = 1;
4027 vcpu->arch.pio.count = count;
4028 vcpu->arch.pio.size = size;
4030 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4032 memcpy(val, vcpu->arch.pio_data, size * count);
4033 vcpu->arch.pio.count = 0;
4037 vcpu->run->exit_reason = KVM_EXIT_IO;
4038 vcpu->run->io.direction = KVM_EXIT_IO_IN;
4039 vcpu->run->io.size = size;
4040 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4041 vcpu->run->io.count = count;
4042 vcpu->run->io.port = port;
4047 static int emulator_pio_out_emulated(int size, unsigned short port,
4048 const void *val, unsigned int count,
4049 struct kvm_vcpu *vcpu)
4051 trace_kvm_pio(1, port, size, count);
4053 vcpu->arch.pio.port = port;
4054 vcpu->arch.pio.in = 0;
4055 vcpu->arch.pio.count = count;
4056 vcpu->arch.pio.size = size;
4058 memcpy(vcpu->arch.pio_data, val, size * count);
4060 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4061 vcpu->arch.pio.count = 0;
4065 vcpu->run->exit_reason = KVM_EXIT_IO;
4066 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4067 vcpu->run->io.size = size;
4068 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4069 vcpu->run->io.count = count;
4070 vcpu->run->io.port = port;
4075 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4077 return kvm_x86_ops->get_segment_base(vcpu, seg);
4080 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
4082 kvm_mmu_invlpg(vcpu, address);
4083 return X86EMUL_CONTINUE;
4086 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4088 if (!need_emulate_wbinvd(vcpu))
4089 return X86EMUL_CONTINUE;
4091 if (kvm_x86_ops->has_wbinvd_exit()) {
4092 int cpu = get_cpu();
4094 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4095 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4096 wbinvd_ipi, NULL, 1);
4098 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4101 return X86EMUL_CONTINUE;
4103 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4105 int emulate_clts(struct kvm_vcpu *vcpu)
4107 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4108 kvm_x86_ops->fpu_activate(vcpu);
4109 return X86EMUL_CONTINUE;
4112 int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
4114 return _kvm_get_dr(vcpu, dr, dest);
4117 int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
4120 return __kvm_set_dr(vcpu, dr, value);
4123 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4125 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4128 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
4130 unsigned long value;
4134 value = kvm_read_cr0(vcpu);
4137 value = vcpu->arch.cr2;
4140 value = kvm_read_cr3(vcpu);
4143 value = kvm_read_cr4(vcpu);
4146 value = kvm_get_cr8(vcpu);
4149 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4156 static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
4162 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4165 vcpu->arch.cr2 = val;
4168 res = kvm_set_cr3(vcpu, val);
4171 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4174 res = kvm_set_cr8(vcpu, val);
4177 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4184 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
4186 return kvm_x86_ops->get_cpl(vcpu);
4189 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4191 kvm_x86_ops->get_gdt(vcpu, dt);
4194 static void emulator_get_idt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4196 kvm_x86_ops->get_idt(vcpu, dt);
4199 static unsigned long emulator_get_cached_segment_base(int seg,
4200 struct kvm_vcpu *vcpu)
4202 return get_segment_base(vcpu, seg);
4205 static bool emulator_get_cached_descriptor(struct desc_struct *desc, u32 *base3,
4206 int seg, struct kvm_vcpu *vcpu)
4208 struct kvm_segment var;
4210 kvm_get_segment(vcpu, &var, seg);
4217 set_desc_limit(desc, var.limit);
4218 set_desc_base(desc, (unsigned long)var.base);
4219 #ifdef CONFIG_X86_64
4221 *base3 = var.base >> 32;
4223 desc->type = var.type;
4225 desc->dpl = var.dpl;
4226 desc->p = var.present;
4227 desc->avl = var.avl;
4235 static void emulator_set_cached_descriptor(struct desc_struct *desc, u32 base3,
4236 int seg, struct kvm_vcpu *vcpu)
4238 struct kvm_segment var;
4240 /* needed to preserve selector */
4241 kvm_get_segment(vcpu, &var, seg);
4243 var.base = get_desc_base(desc);
4244 #ifdef CONFIG_X86_64
4245 var.base |= ((u64)base3) << 32;
4247 var.limit = get_desc_limit(desc);
4249 var.limit = (var.limit << 12) | 0xfff;
4250 var.type = desc->type;
4251 var.present = desc->p;
4252 var.dpl = desc->dpl;
4257 var.avl = desc->avl;
4258 var.present = desc->p;
4259 var.unusable = !var.present;
4262 kvm_set_segment(vcpu, &var, seg);
4266 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
4268 struct kvm_segment kvm_seg;
4270 kvm_get_segment(vcpu, &kvm_seg, seg);
4271 return kvm_seg.selector;
4274 static void emulator_set_segment_selector(u16 sel, int seg,
4275 struct kvm_vcpu *vcpu)
4277 struct kvm_segment kvm_seg;
4279 kvm_get_segment(vcpu, &kvm_seg, seg);
4280 kvm_seg.selector = sel;
4281 kvm_set_segment(vcpu, &kvm_seg, seg);
4284 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4287 kvm_load_guest_fpu(ctxt->vcpu);
4289 * CR0.TS may reference the host fpu state, not the guest fpu state,
4290 * so it may be clear at this point.
4295 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4300 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4301 enum x86_intercept intercept,
4302 enum x86_intercept_stage stage)
4304 return X86EMUL_CONTINUE;
4307 static struct x86_emulate_ops emulate_ops = {
4308 .read_std = kvm_read_guest_virt_system,
4309 .write_std = kvm_write_guest_virt_system,
4310 .fetch = kvm_fetch_guest_virt,
4311 .read_emulated = emulator_read_emulated,
4312 .write_emulated = emulator_write_emulated,
4313 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4314 .pio_in_emulated = emulator_pio_in_emulated,
4315 .pio_out_emulated = emulator_pio_out_emulated,
4316 .get_cached_descriptor = emulator_get_cached_descriptor,
4317 .set_cached_descriptor = emulator_set_cached_descriptor,
4318 .get_segment_selector = emulator_get_segment_selector,
4319 .set_segment_selector = emulator_set_segment_selector,
4320 .get_cached_segment_base = emulator_get_cached_segment_base,
4321 .get_gdt = emulator_get_gdt,
4322 .get_idt = emulator_get_idt,
4323 .get_cr = emulator_get_cr,
4324 .set_cr = emulator_set_cr,
4325 .cpl = emulator_get_cpl,
4326 .get_dr = emulator_get_dr,
4327 .set_dr = emulator_set_dr,
4328 .set_msr = kvm_set_msr,
4329 .get_msr = kvm_get_msr,
4330 .get_fpu = emulator_get_fpu,
4331 .put_fpu = emulator_put_fpu,
4332 .intercept = emulator_intercept,
4335 static void cache_all_regs(struct kvm_vcpu *vcpu)
4337 kvm_register_read(vcpu, VCPU_REGS_RAX);
4338 kvm_register_read(vcpu, VCPU_REGS_RSP);
4339 kvm_register_read(vcpu, VCPU_REGS_RIP);
4340 vcpu->arch.regs_dirty = ~0;
4343 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4345 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4347 * an sti; sti; sequence only disable interrupts for the first
4348 * instruction. So, if the last instruction, be it emulated or
4349 * not, left the system with the INT_STI flag enabled, it
4350 * means that the last instruction is an sti. We should not
4351 * leave the flag on in this case. The same goes for mov ss
4353 if (!(int_shadow & mask))
4354 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4357 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4359 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4360 if (ctxt->exception.vector == PF_VECTOR)
4361 kvm_propagate_fault(vcpu, &ctxt->exception);
4362 else if (ctxt->exception.error_code_valid)
4363 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4364 ctxt->exception.error_code);
4366 kvm_queue_exception(vcpu, ctxt->exception.vector);
4369 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4371 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4374 cache_all_regs(vcpu);
4376 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4378 vcpu->arch.emulate_ctxt.vcpu = vcpu;
4379 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
4380 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4381 vcpu->arch.emulate_ctxt.mode =
4382 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4383 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4384 ? X86EMUL_MODE_VM86 : cs_l
4385 ? X86EMUL_MODE_PROT64 : cs_db
4386 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4387 vcpu->arch.emulate_ctxt.guest_mode = is_guest_mode(vcpu);
4388 memset(c, 0, sizeof(struct decode_cache));
4389 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4392 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq)
4394 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4397 init_emulate_ctxt(vcpu);
4399 vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4400 vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4401 vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip;
4402 ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4404 if (ret != X86EMUL_CONTINUE)
4405 return EMULATE_FAIL;
4407 vcpu->arch.emulate_ctxt.eip = c->eip;
4408 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4409 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4410 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4412 if (irq == NMI_VECTOR)
4413 vcpu->arch.nmi_pending = false;
4415 vcpu->arch.interrupt.pending = false;
4417 return EMULATE_DONE;
4419 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4421 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4423 int r = EMULATE_DONE;
4425 ++vcpu->stat.insn_emulation_fail;
4426 trace_kvm_emulate_insn_failed(vcpu);
4427 if (!is_guest_mode(vcpu)) {
4428 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4429 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4430 vcpu->run->internal.ndata = 0;
4433 kvm_queue_exception(vcpu, UD_VECTOR);
4438 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4446 * if emulation was due to access to shadowed page table
4447 * and it failed try to unshadow page and re-entetr the
4448 * guest to let CPU execute the instruction.
4450 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4453 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4455 if (gpa == UNMAPPED_GVA)
4456 return true; /* let cpu generate fault */
4458 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4464 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4471 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4473 kvm_clear_exception_queue(vcpu);
4474 vcpu->arch.mmio_fault_cr2 = cr2;
4476 * TODO: fix emulate.c to use guest_read/write_register
4477 * instead of direct ->regs accesses, can save hundred cycles
4478 * on Intel for instructions that don't read/change RSP, for
4481 cache_all_regs(vcpu);
4483 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4484 init_emulate_ctxt(vcpu);
4485 vcpu->arch.emulate_ctxt.interruptibility = 0;
4486 vcpu->arch.emulate_ctxt.have_exception = false;
4487 vcpu->arch.emulate_ctxt.perm_ok = false;
4489 vcpu->arch.emulate_ctxt.only_vendor_specific_insn
4490 = emulation_type & EMULTYPE_TRAP_UD;
4492 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, insn, insn_len);
4494 trace_kvm_emulate_insn_start(vcpu);
4495 ++vcpu->stat.insn_emulation;
4497 if (emulation_type & EMULTYPE_TRAP_UD)
4498 return EMULATE_FAIL;
4499 if (reexecute_instruction(vcpu, cr2))
4500 return EMULATE_DONE;
4501 if (emulation_type & EMULTYPE_SKIP)
4502 return EMULATE_FAIL;
4503 return handle_emulation_failure(vcpu);
4507 if (emulation_type & EMULTYPE_SKIP) {
4508 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4509 return EMULATE_DONE;
4512 /* this is needed for vmware backdor interface to work since it
4513 changes registers values during IO operation */
4514 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4517 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4519 if (r == EMULATION_INTERCEPTED)
4520 return EMULATE_DONE;
4522 if (r == EMULATION_FAILED) {
4523 if (reexecute_instruction(vcpu, cr2))
4524 return EMULATE_DONE;
4526 return handle_emulation_failure(vcpu);
4529 if (vcpu->arch.emulate_ctxt.have_exception) {
4530 inject_emulated_exception(vcpu);
4532 } else if (vcpu->arch.pio.count) {
4533 if (!vcpu->arch.pio.in)
4534 vcpu->arch.pio.count = 0;
4535 r = EMULATE_DO_MMIO;
4536 } else if (vcpu->mmio_needed)
4537 r = EMULATE_DO_MMIO;
4538 else if (r == EMULATION_RESTART)
4543 toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility);
4544 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4545 kvm_make_request(KVM_REQ_EVENT, vcpu);
4546 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4547 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4551 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4553 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4555 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4556 int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
4557 /* do not return to emulator after return from userspace */
4558 vcpu->arch.pio.count = 0;
4561 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4563 static void tsc_bad(void *info)
4565 __this_cpu_write(cpu_tsc_khz, 0);
4568 static void tsc_khz_changed(void *data)
4570 struct cpufreq_freqs *freq = data;
4571 unsigned long khz = 0;
4575 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4576 khz = cpufreq_quick_get(raw_smp_processor_id());
4579 __this_cpu_write(cpu_tsc_khz, khz);
4582 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4585 struct cpufreq_freqs *freq = data;
4587 struct kvm_vcpu *vcpu;
4588 int i, send_ipi = 0;
4591 * We allow guests to temporarily run on slowing clocks,
4592 * provided we notify them after, or to run on accelerating
4593 * clocks, provided we notify them before. Thus time never
4596 * However, we have a problem. We can't atomically update
4597 * the frequency of a given CPU from this function; it is
4598 * merely a notifier, which can be called from any CPU.
4599 * Changing the TSC frequency at arbitrary points in time
4600 * requires a recomputation of local variables related to
4601 * the TSC for each VCPU. We must flag these local variables
4602 * to be updated and be sure the update takes place with the
4603 * new frequency before any guests proceed.
4605 * Unfortunately, the combination of hotplug CPU and frequency
4606 * change creates an intractable locking scenario; the order
4607 * of when these callouts happen is undefined with respect to
4608 * CPU hotplug, and they can race with each other. As such,
4609 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4610 * undefined; you can actually have a CPU frequency change take
4611 * place in between the computation of X and the setting of the
4612 * variable. To protect against this problem, all updates of
4613 * the per_cpu tsc_khz variable are done in an interrupt
4614 * protected IPI, and all callers wishing to update the value
4615 * must wait for a synchronous IPI to complete (which is trivial
4616 * if the caller is on the CPU already). This establishes the
4617 * necessary total order on variable updates.
4619 * Note that because a guest time update may take place
4620 * anytime after the setting of the VCPU's request bit, the
4621 * correct TSC value must be set before the request. However,
4622 * to ensure the update actually makes it to any guest which
4623 * starts running in hardware virtualization between the set
4624 * and the acquisition of the spinlock, we must also ping the
4625 * CPU after setting the request bit.
4629 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4631 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4634 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4636 raw_spin_lock(&kvm_lock);
4637 list_for_each_entry(kvm, &vm_list, vm_list) {
4638 kvm_for_each_vcpu(i, vcpu, kvm) {
4639 if (vcpu->cpu != freq->cpu)
4641 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4642 if (vcpu->cpu != smp_processor_id())
4646 raw_spin_unlock(&kvm_lock);
4648 if (freq->old < freq->new && send_ipi) {
4650 * We upscale the frequency. Must make the guest
4651 * doesn't see old kvmclock values while running with
4652 * the new frequency, otherwise we risk the guest sees
4653 * time go backwards.
4655 * In case we update the frequency for another cpu
4656 * (which might be in guest context) send an interrupt
4657 * to kick the cpu out of guest context. Next time
4658 * guest context is entered kvmclock will be updated,
4659 * so the guest will not see stale values.
4661 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4666 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4667 .notifier_call = kvmclock_cpufreq_notifier
4670 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4671 unsigned long action, void *hcpu)
4673 unsigned int cpu = (unsigned long)hcpu;
4677 case CPU_DOWN_FAILED:
4678 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4680 case CPU_DOWN_PREPARE:
4681 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4687 static struct notifier_block kvmclock_cpu_notifier_block = {
4688 .notifier_call = kvmclock_cpu_notifier,
4689 .priority = -INT_MAX
4692 static void kvm_timer_init(void)
4696 max_tsc_khz = tsc_khz;
4697 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4698 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4699 #ifdef CONFIG_CPU_FREQ
4700 struct cpufreq_policy policy;
4701 memset(&policy, 0, sizeof(policy));
4703 cpufreq_get_policy(&policy, cpu);
4704 if (policy.cpuinfo.max_freq)
4705 max_tsc_khz = policy.cpuinfo.max_freq;
4708 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4709 CPUFREQ_TRANSITION_NOTIFIER);
4711 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4712 for_each_online_cpu(cpu)
4713 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4716 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4718 static int kvm_is_in_guest(void)
4720 return percpu_read(current_vcpu) != NULL;
4723 static int kvm_is_user_mode(void)
4727 if (percpu_read(current_vcpu))
4728 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4730 return user_mode != 0;
4733 static unsigned long kvm_get_guest_ip(void)
4735 unsigned long ip = 0;
4737 if (percpu_read(current_vcpu))
4738 ip = kvm_rip_read(percpu_read(current_vcpu));
4743 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4744 .is_in_guest = kvm_is_in_guest,
4745 .is_user_mode = kvm_is_user_mode,
4746 .get_guest_ip = kvm_get_guest_ip,
4749 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4751 percpu_write(current_vcpu, vcpu);
4753 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4755 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4757 percpu_write(current_vcpu, NULL);
4759 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4761 int kvm_arch_init(void *opaque)
4764 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4767 printk(KERN_ERR "kvm: already loaded the other module\n");
4772 if (!ops->cpu_has_kvm_support()) {
4773 printk(KERN_ERR "kvm: no hardware support\n");
4777 if (ops->disabled_by_bios()) {
4778 printk(KERN_ERR "kvm: disabled by bios\n");
4783 r = kvm_mmu_module_init();
4787 kvm_init_msr_list();
4790 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4791 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4792 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4796 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4799 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4807 void kvm_arch_exit(void)
4809 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4811 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4812 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4813 CPUFREQ_TRANSITION_NOTIFIER);
4814 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4816 kvm_mmu_module_exit();
4819 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4821 ++vcpu->stat.halt_exits;
4822 if (irqchip_in_kernel(vcpu->kvm)) {
4823 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4826 vcpu->run->exit_reason = KVM_EXIT_HLT;
4830 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4832 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4835 if (is_long_mode(vcpu))
4838 return a0 | ((gpa_t)a1 << 32);
4841 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4843 u64 param, ingpa, outgpa, ret;
4844 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4845 bool fast, longmode;
4849 * hypercall generates UD from non zero cpl and real mode
4852 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4853 kvm_queue_exception(vcpu, UD_VECTOR);
4857 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4858 longmode = is_long_mode(vcpu) && cs_l == 1;
4861 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4862 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4863 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4864 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4865 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4866 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4868 #ifdef CONFIG_X86_64
4870 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4871 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4872 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4876 code = param & 0xffff;
4877 fast = (param >> 16) & 0x1;
4878 rep_cnt = (param >> 32) & 0xfff;
4879 rep_idx = (param >> 48) & 0xfff;
4881 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4884 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4885 kvm_vcpu_on_spin(vcpu);
4888 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4892 ret = res | (((u64)rep_done & 0xfff) << 32);
4894 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4896 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4897 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4903 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4905 unsigned long nr, a0, a1, a2, a3, ret;
4908 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4909 return kvm_hv_hypercall(vcpu);
4911 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4912 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4913 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4914 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4915 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4917 trace_kvm_hypercall(nr, a0, a1, a2, a3);
4919 if (!is_long_mode(vcpu)) {
4927 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4933 case KVM_HC_VAPIC_POLL_IRQ:
4937 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4944 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4945 ++vcpu->stat.hypercalls;
4948 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4950 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4952 char instruction[3];
4953 unsigned long rip = kvm_rip_read(vcpu);
4956 * Blow out the MMU to ensure that no other VCPU has an active mapping
4957 * to ensure that the updated hypercall appears atomically across all
4960 kvm_mmu_zap_all(vcpu->kvm);
4962 kvm_x86_ops->patch_hypercall(vcpu, instruction);
4964 return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
4967 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4969 struct desc_ptr dt = { limit, base };
4971 kvm_x86_ops->set_gdt(vcpu, &dt);
4974 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4976 struct desc_ptr dt = { limit, base };
4978 kvm_x86_ops->set_idt(vcpu, &dt);
4981 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4983 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4984 int j, nent = vcpu->arch.cpuid_nent;
4986 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4987 /* when no next entry is found, the current entry[i] is reselected */
4988 for (j = i + 1; ; j = (j + 1) % nent) {
4989 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4990 if (ej->function == e->function) {
4991 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4995 return 0; /* silence gcc, even though control never reaches here */
4998 /* find an entry with matching function, matching index (if needed), and that
4999 * should be read next (if it's stateful) */
5000 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5001 u32 function, u32 index)
5003 if (e->function != function)
5005 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5007 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5008 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5013 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5014 u32 function, u32 index)
5017 struct kvm_cpuid_entry2 *best = NULL;
5019 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5020 struct kvm_cpuid_entry2 *e;
5022 e = &vcpu->arch.cpuid_entries[i];
5023 if (is_matching_cpuid_entry(e, function, index)) {
5024 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5025 move_to_next_stateful_cpuid_entry(vcpu, i);
5032 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5034 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5036 struct kvm_cpuid_entry2 *best;
5038 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5039 if (!best || best->eax < 0x80000008)
5041 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5043 return best->eax & 0xff;
5049 * If no match is found, check whether we exceed the vCPU's limit
5050 * and return the content of the highest valid _standard_ leaf instead.
5051 * This is to satisfy the CPUID specification.
5053 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5054 u32 function, u32 index)
5056 struct kvm_cpuid_entry2 *maxlevel;
5058 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5059 if (!maxlevel || maxlevel->eax >= function)
5061 if (function & 0x80000000) {
5062 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5066 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5069 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5071 u32 function, index;
5072 struct kvm_cpuid_entry2 *best;
5074 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5075 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5076 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5077 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5078 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5079 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5080 best = kvm_find_cpuid_entry(vcpu, function, index);
5083 best = check_cpuid_limit(vcpu, function, index);
5086 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5087 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5088 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5089 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5091 kvm_x86_ops->skip_emulated_instruction(vcpu);
5092 trace_kvm_cpuid(function,
5093 kvm_register_read(vcpu, VCPU_REGS_RAX),
5094 kvm_register_read(vcpu, VCPU_REGS_RBX),
5095 kvm_register_read(vcpu, VCPU_REGS_RCX),
5096 kvm_register_read(vcpu, VCPU_REGS_RDX));
5098 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5101 * Check if userspace requested an interrupt window, and that the
5102 * interrupt window is open.
5104 * No need to exit to userspace if we already have an interrupt queued.
5106 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5108 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5109 vcpu->run->request_interrupt_window &&
5110 kvm_arch_interrupt_allowed(vcpu));
5113 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5115 struct kvm_run *kvm_run = vcpu->run;
5117 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5118 kvm_run->cr8 = kvm_get_cr8(vcpu);
5119 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5120 if (irqchip_in_kernel(vcpu->kvm))
5121 kvm_run->ready_for_interrupt_injection = 1;
5123 kvm_run->ready_for_interrupt_injection =
5124 kvm_arch_interrupt_allowed(vcpu) &&
5125 !kvm_cpu_has_interrupt(vcpu) &&
5126 !kvm_event_needs_reinjection(vcpu);
5129 static void vapic_enter(struct kvm_vcpu *vcpu)
5131 struct kvm_lapic *apic = vcpu->arch.apic;
5134 if (!apic || !apic->vapic_addr)
5137 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5139 vcpu->arch.apic->vapic_page = page;
5142 static void vapic_exit(struct kvm_vcpu *vcpu)
5144 struct kvm_lapic *apic = vcpu->arch.apic;
5147 if (!apic || !apic->vapic_addr)
5150 idx = srcu_read_lock(&vcpu->kvm->srcu);
5151 kvm_release_page_dirty(apic->vapic_page);
5152 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5153 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5156 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5160 if (!kvm_x86_ops->update_cr8_intercept)
5163 if (!vcpu->arch.apic)
5166 if (!vcpu->arch.apic->vapic_addr)
5167 max_irr = kvm_lapic_find_highest_irr(vcpu);
5174 tpr = kvm_lapic_get_cr8(vcpu);
5176 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5179 static void inject_pending_event(struct kvm_vcpu *vcpu)
5181 /* try to reinject previous events if any */
5182 if (vcpu->arch.exception.pending) {
5183 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5184 vcpu->arch.exception.has_error_code,
5185 vcpu->arch.exception.error_code);
5186 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5187 vcpu->arch.exception.has_error_code,
5188 vcpu->arch.exception.error_code,
5189 vcpu->arch.exception.reinject);
5193 if (vcpu->arch.nmi_injected) {
5194 kvm_x86_ops->set_nmi(vcpu);
5198 if (vcpu->arch.interrupt.pending) {
5199 kvm_x86_ops->set_irq(vcpu);
5203 /* try to inject new event if pending */
5204 if (vcpu->arch.nmi_pending) {
5205 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5206 vcpu->arch.nmi_pending = false;
5207 vcpu->arch.nmi_injected = true;
5208 kvm_x86_ops->set_nmi(vcpu);
5210 } else if (kvm_cpu_has_interrupt(vcpu)) {
5211 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5212 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5214 kvm_x86_ops->set_irq(vcpu);
5219 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5221 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5222 !vcpu->guest_xcr0_loaded) {
5223 /* kvm_set_xcr() also depends on this */
5224 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5225 vcpu->guest_xcr0_loaded = 1;
5229 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5231 if (vcpu->guest_xcr0_loaded) {
5232 if (vcpu->arch.xcr0 != host_xcr0)
5233 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5234 vcpu->guest_xcr0_loaded = 0;
5238 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5242 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5243 vcpu->run->request_interrupt_window;
5245 if (vcpu->requests) {
5246 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5247 kvm_mmu_unload(vcpu);
5248 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5249 __kvm_migrate_timers(vcpu);
5250 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5251 r = kvm_guest_time_update(vcpu);
5255 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5256 kvm_mmu_sync_roots(vcpu);
5257 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5258 kvm_x86_ops->tlb_flush(vcpu);
5259 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5260 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5264 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5265 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5269 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5270 vcpu->fpu_active = 0;
5271 kvm_x86_ops->fpu_deactivate(vcpu);
5273 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5274 /* Page is swapped out. Do synthetic halt */
5275 vcpu->arch.apf.halted = true;
5281 r = kvm_mmu_reload(vcpu);
5286 * An NMI can be injected between local nmi_pending read and
5287 * vcpu->arch.nmi_pending read inside inject_pending_event().
5288 * But in that case, KVM_REQ_EVENT will be set, which makes
5289 * the race described above benign.
5291 nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5293 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5294 inject_pending_event(vcpu);
5296 /* enable NMI/IRQ window open exits if needed */
5298 kvm_x86_ops->enable_nmi_window(vcpu);
5299 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5300 kvm_x86_ops->enable_irq_window(vcpu);
5302 if (kvm_lapic_enabled(vcpu)) {
5303 update_cr8_intercept(vcpu);
5304 kvm_lapic_sync_to_vapic(vcpu);
5310 kvm_x86_ops->prepare_guest_switch(vcpu);
5311 if (vcpu->fpu_active)
5312 kvm_load_guest_fpu(vcpu);
5313 kvm_load_guest_xcr0(vcpu);
5315 vcpu->mode = IN_GUEST_MODE;
5317 /* We should set ->mode before check ->requests,
5318 * see the comment in make_all_cpus_request.
5322 local_irq_disable();
5324 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5325 || need_resched() || signal_pending(current)) {
5326 vcpu->mode = OUTSIDE_GUEST_MODE;
5330 kvm_x86_ops->cancel_injection(vcpu);
5335 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5339 if (unlikely(vcpu->arch.switch_db_regs)) {
5341 set_debugreg(vcpu->arch.eff_db[0], 0);
5342 set_debugreg(vcpu->arch.eff_db[1], 1);
5343 set_debugreg(vcpu->arch.eff_db[2], 2);
5344 set_debugreg(vcpu->arch.eff_db[3], 3);
5347 trace_kvm_entry(vcpu->vcpu_id);
5348 kvm_x86_ops->run(vcpu);
5351 * If the guest has used debug registers, at least dr7
5352 * will be disabled while returning to the host.
5353 * If we don't have active breakpoints in the host, we don't
5354 * care about the messed up debug address registers. But if
5355 * we have some of them active, restore the old state.
5357 if (hw_breakpoint_active())
5358 hw_breakpoint_restore();
5360 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5362 vcpu->mode = OUTSIDE_GUEST_MODE;
5369 * We must have an instruction between local_irq_enable() and
5370 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5371 * the interrupt shadow. The stat.exits increment will do nicely.
5372 * But we need to prevent reordering, hence this barrier():
5380 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5383 * Profile KVM exit RIPs:
5385 if (unlikely(prof_on == KVM_PROFILING)) {
5386 unsigned long rip = kvm_rip_read(vcpu);
5387 profile_hit(KVM_PROFILING, (void *)rip);
5391 kvm_lapic_sync_from_vapic(vcpu);
5393 r = kvm_x86_ops->handle_exit(vcpu);
5399 static int __vcpu_run(struct kvm_vcpu *vcpu)
5402 struct kvm *kvm = vcpu->kvm;
5404 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5405 pr_debug("vcpu %d received sipi with vector # %x\n",
5406 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5407 kvm_lapic_reset(vcpu);
5408 r = kvm_arch_vcpu_reset(vcpu);
5411 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5414 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5419 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5420 !vcpu->arch.apf.halted)
5421 r = vcpu_enter_guest(vcpu);
5423 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5424 kvm_vcpu_block(vcpu);
5425 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5426 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5428 switch(vcpu->arch.mp_state) {
5429 case KVM_MP_STATE_HALTED:
5430 vcpu->arch.mp_state =
5431 KVM_MP_STATE_RUNNABLE;
5432 case KVM_MP_STATE_RUNNABLE:
5433 vcpu->arch.apf.halted = false;
5435 case KVM_MP_STATE_SIPI_RECEIVED:
5446 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5447 if (kvm_cpu_has_pending_timer(vcpu))
5448 kvm_inject_pending_timer_irqs(vcpu);
5450 if (dm_request_for_irq_injection(vcpu)) {
5452 vcpu->run->exit_reason = KVM_EXIT_INTR;
5453 ++vcpu->stat.request_irq_exits;
5456 kvm_check_async_pf_completion(vcpu);
5458 if (signal_pending(current)) {
5460 vcpu->run->exit_reason = KVM_EXIT_INTR;
5461 ++vcpu->stat.signal_exits;
5463 if (need_resched()) {
5464 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5466 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5470 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5477 static int complete_mmio(struct kvm_vcpu *vcpu)
5479 struct kvm_run *run = vcpu->run;
5482 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5485 if (vcpu->mmio_needed) {
5486 vcpu->mmio_needed = 0;
5487 if (!vcpu->mmio_is_write)
5488 memcpy(vcpu->mmio_data, run->mmio.data, 8);
5489 vcpu->mmio_index += 8;
5490 if (vcpu->mmio_index < vcpu->mmio_size) {
5491 run->exit_reason = KVM_EXIT_MMIO;
5492 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5493 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5494 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5495 run->mmio.is_write = vcpu->mmio_is_write;
5496 vcpu->mmio_needed = 1;
5499 if (vcpu->mmio_is_write)
5501 vcpu->mmio_read_completed = 1;
5503 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5504 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5505 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5506 if (r != EMULATE_DONE)
5511 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5516 if (!tsk_used_math(current) && init_fpu(current))
5519 if (vcpu->sigset_active)
5520 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5522 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5523 kvm_vcpu_block(vcpu);
5524 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5529 /* re-sync apic's tpr */
5530 if (!irqchip_in_kernel(vcpu->kvm)) {
5531 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5537 r = complete_mmio(vcpu);
5541 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5542 kvm_register_write(vcpu, VCPU_REGS_RAX,
5543 kvm_run->hypercall.ret);
5545 r = __vcpu_run(vcpu);
5548 post_kvm_run_save(vcpu);
5549 if (vcpu->sigset_active)
5550 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5555 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5557 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5558 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5559 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5560 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5561 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5562 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5563 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5564 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5565 #ifdef CONFIG_X86_64
5566 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5567 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5568 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5569 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5570 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5571 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5572 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5573 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5576 regs->rip = kvm_rip_read(vcpu);
5577 regs->rflags = kvm_get_rflags(vcpu);
5582 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5584 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5585 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5586 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5587 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5588 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5589 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5590 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5591 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5592 #ifdef CONFIG_X86_64
5593 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5594 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5595 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5596 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5597 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5598 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5599 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5600 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5603 kvm_rip_write(vcpu, regs->rip);
5604 kvm_set_rflags(vcpu, regs->rflags);
5606 vcpu->arch.exception.pending = false;
5608 kvm_make_request(KVM_REQ_EVENT, vcpu);
5613 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5615 struct kvm_segment cs;
5617 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5621 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5623 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5624 struct kvm_sregs *sregs)
5628 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5629 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5630 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5631 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5632 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5633 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5635 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5636 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5638 kvm_x86_ops->get_idt(vcpu, &dt);
5639 sregs->idt.limit = dt.size;
5640 sregs->idt.base = dt.address;
5641 kvm_x86_ops->get_gdt(vcpu, &dt);
5642 sregs->gdt.limit = dt.size;
5643 sregs->gdt.base = dt.address;
5645 sregs->cr0 = kvm_read_cr0(vcpu);
5646 sregs->cr2 = vcpu->arch.cr2;
5647 sregs->cr3 = kvm_read_cr3(vcpu);
5648 sregs->cr4 = kvm_read_cr4(vcpu);
5649 sregs->cr8 = kvm_get_cr8(vcpu);
5650 sregs->efer = vcpu->arch.efer;
5651 sregs->apic_base = kvm_get_apic_base(vcpu);
5653 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5655 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5656 set_bit(vcpu->arch.interrupt.nr,
5657 (unsigned long *)sregs->interrupt_bitmap);
5662 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5663 struct kvm_mp_state *mp_state)
5665 mp_state->mp_state = vcpu->arch.mp_state;
5669 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5670 struct kvm_mp_state *mp_state)
5672 vcpu->arch.mp_state = mp_state->mp_state;
5673 kvm_make_request(KVM_REQ_EVENT, vcpu);
5677 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5678 bool has_error_code, u32 error_code)
5680 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5683 init_emulate_ctxt(vcpu);
5685 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5686 tss_selector, reason, has_error_code,
5690 return EMULATE_FAIL;
5692 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5693 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5694 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5695 kvm_make_request(KVM_REQ_EVENT, vcpu);
5696 return EMULATE_DONE;
5698 EXPORT_SYMBOL_GPL(kvm_task_switch);
5700 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5701 struct kvm_sregs *sregs)
5703 int mmu_reset_needed = 0;
5704 int pending_vec, max_bits, idx;
5707 dt.size = sregs->idt.limit;
5708 dt.address = sregs->idt.base;
5709 kvm_x86_ops->set_idt(vcpu, &dt);
5710 dt.size = sregs->gdt.limit;
5711 dt.address = sregs->gdt.base;
5712 kvm_x86_ops->set_gdt(vcpu, &dt);
5714 vcpu->arch.cr2 = sregs->cr2;
5715 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5716 vcpu->arch.cr3 = sregs->cr3;
5717 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5719 kvm_set_cr8(vcpu, sregs->cr8);
5721 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5722 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5723 kvm_set_apic_base(vcpu, sregs->apic_base);
5725 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5726 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5727 vcpu->arch.cr0 = sregs->cr0;
5729 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5730 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5731 if (sregs->cr4 & X86_CR4_OSXSAVE)
5734 idx = srcu_read_lock(&vcpu->kvm->srcu);
5735 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5736 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5737 mmu_reset_needed = 1;
5739 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5741 if (mmu_reset_needed)
5742 kvm_mmu_reset_context(vcpu);
5744 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5745 pending_vec = find_first_bit(
5746 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5747 if (pending_vec < max_bits) {
5748 kvm_queue_interrupt(vcpu, pending_vec, false);
5749 pr_debug("Set back pending irq %d\n", pending_vec);
5752 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5753 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5754 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5755 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5756 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5757 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5759 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5760 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5762 update_cr8_intercept(vcpu);
5764 /* Older userspace won't unhalt the vcpu on reset. */
5765 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5766 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5768 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5770 kvm_make_request(KVM_REQ_EVENT, vcpu);
5775 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5776 struct kvm_guest_debug *dbg)
5778 unsigned long rflags;
5781 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5783 if (vcpu->arch.exception.pending)
5785 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5786 kvm_queue_exception(vcpu, DB_VECTOR);
5788 kvm_queue_exception(vcpu, BP_VECTOR);
5792 * Read rflags as long as potentially injected trace flags are still
5795 rflags = kvm_get_rflags(vcpu);
5797 vcpu->guest_debug = dbg->control;
5798 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5799 vcpu->guest_debug = 0;
5801 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5802 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5803 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5804 vcpu->arch.switch_db_regs =
5805 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5807 for (i = 0; i < KVM_NR_DB_REGS; i++)
5808 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5809 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5812 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5813 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5814 get_segment_base(vcpu, VCPU_SREG_CS);
5817 * Trigger an rflags update that will inject or remove the trace
5820 kvm_set_rflags(vcpu, rflags);
5822 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5832 * Translate a guest virtual address to a guest physical address.
5834 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5835 struct kvm_translation *tr)
5837 unsigned long vaddr = tr->linear_address;
5841 idx = srcu_read_lock(&vcpu->kvm->srcu);
5842 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5843 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5844 tr->physical_address = gpa;
5845 tr->valid = gpa != UNMAPPED_GVA;
5852 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5854 struct i387_fxsave_struct *fxsave =
5855 &vcpu->arch.guest_fpu.state->fxsave;
5857 memcpy(fpu->fpr, fxsave->st_space, 128);
5858 fpu->fcw = fxsave->cwd;
5859 fpu->fsw = fxsave->swd;
5860 fpu->ftwx = fxsave->twd;
5861 fpu->last_opcode = fxsave->fop;
5862 fpu->last_ip = fxsave->rip;
5863 fpu->last_dp = fxsave->rdp;
5864 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5869 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5871 struct i387_fxsave_struct *fxsave =
5872 &vcpu->arch.guest_fpu.state->fxsave;
5874 memcpy(fxsave->st_space, fpu->fpr, 128);
5875 fxsave->cwd = fpu->fcw;
5876 fxsave->swd = fpu->fsw;
5877 fxsave->twd = fpu->ftwx;
5878 fxsave->fop = fpu->last_opcode;
5879 fxsave->rip = fpu->last_ip;
5880 fxsave->rdp = fpu->last_dp;
5881 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5886 int fx_init(struct kvm_vcpu *vcpu)
5890 err = fpu_alloc(&vcpu->arch.guest_fpu);
5894 fpu_finit(&vcpu->arch.guest_fpu);
5897 * Ensure guest xcr0 is valid for loading
5899 vcpu->arch.xcr0 = XSTATE_FP;
5901 vcpu->arch.cr0 |= X86_CR0_ET;
5905 EXPORT_SYMBOL_GPL(fx_init);
5907 static void fx_free(struct kvm_vcpu *vcpu)
5909 fpu_free(&vcpu->arch.guest_fpu);
5912 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5914 if (vcpu->guest_fpu_loaded)
5918 * Restore all possible states in the guest,
5919 * and assume host would use all available bits.
5920 * Guest xcr0 would be loaded later.
5922 kvm_put_guest_xcr0(vcpu);
5923 vcpu->guest_fpu_loaded = 1;
5924 unlazy_fpu(current);
5925 fpu_restore_checking(&vcpu->arch.guest_fpu);
5929 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5931 kvm_put_guest_xcr0(vcpu);
5933 if (!vcpu->guest_fpu_loaded)
5936 vcpu->guest_fpu_loaded = 0;
5937 fpu_save_init(&vcpu->arch.guest_fpu);
5938 ++vcpu->stat.fpu_reload;
5939 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5943 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5945 kvmclock_reset(vcpu);
5947 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5949 kvm_x86_ops->vcpu_free(vcpu);
5952 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5955 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5956 printk_once(KERN_WARNING
5957 "kvm: SMP vm created on host with unstable TSC; "
5958 "guest TSC will not be reliable\n");
5959 return kvm_x86_ops->vcpu_create(kvm, id);
5962 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5966 vcpu->arch.mtrr_state.have_fixed = 1;
5968 r = kvm_arch_vcpu_reset(vcpu);
5970 r = kvm_mmu_setup(vcpu);
5977 kvm_x86_ops->vcpu_free(vcpu);
5981 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5983 vcpu->arch.apf.msr_val = 0;
5986 kvm_mmu_unload(vcpu);
5990 kvm_x86_ops->vcpu_free(vcpu);
5993 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5995 vcpu->arch.nmi_pending = false;
5996 vcpu->arch.nmi_injected = false;
5998 vcpu->arch.switch_db_regs = 0;
5999 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6000 vcpu->arch.dr6 = DR6_FIXED_1;
6001 vcpu->arch.dr7 = DR7_FIXED_1;
6003 kvm_make_request(KVM_REQ_EVENT, vcpu);
6004 vcpu->arch.apf.msr_val = 0;
6006 kvmclock_reset(vcpu);
6008 kvm_clear_async_pf_completion_queue(vcpu);
6009 kvm_async_pf_hash_reset(vcpu);
6010 vcpu->arch.apf.halted = false;
6012 return kvm_x86_ops->vcpu_reset(vcpu);
6015 int kvm_arch_hardware_enable(void *garbage)
6018 struct kvm_vcpu *vcpu;
6021 kvm_shared_msr_cpu_online();
6022 list_for_each_entry(kvm, &vm_list, vm_list)
6023 kvm_for_each_vcpu(i, vcpu, kvm)
6024 if (vcpu->cpu == smp_processor_id())
6025 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6026 return kvm_x86_ops->hardware_enable(garbage);
6029 void kvm_arch_hardware_disable(void *garbage)
6031 kvm_x86_ops->hardware_disable(garbage);
6032 drop_user_return_notifiers(garbage);
6035 int kvm_arch_hardware_setup(void)
6037 return kvm_x86_ops->hardware_setup();
6040 void kvm_arch_hardware_unsetup(void)
6042 kvm_x86_ops->hardware_unsetup();
6045 void kvm_arch_check_processor_compat(void *rtn)
6047 kvm_x86_ops->check_processor_compatibility(rtn);
6050 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6056 BUG_ON(vcpu->kvm == NULL);
6059 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6060 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6061 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6062 vcpu->arch.mmu.translate_gpa = translate_gpa;
6063 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6064 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6065 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6067 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6069 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6074 vcpu->arch.pio_data = page_address(page);
6076 if (!kvm->arch.virtual_tsc_khz)
6077 kvm_arch_set_tsc_khz(kvm, max_tsc_khz);
6079 r = kvm_mmu_create(vcpu);
6081 goto fail_free_pio_data;
6083 if (irqchip_in_kernel(kvm)) {
6084 r = kvm_create_lapic(vcpu);
6086 goto fail_mmu_destroy;
6089 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6091 if (!vcpu->arch.mce_banks) {
6093 goto fail_free_lapic;
6095 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6097 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6098 goto fail_free_mce_banks;
6100 kvm_async_pf_hash_reset(vcpu);
6103 fail_free_mce_banks:
6104 kfree(vcpu->arch.mce_banks);
6106 kvm_free_lapic(vcpu);
6108 kvm_mmu_destroy(vcpu);
6110 free_page((unsigned long)vcpu->arch.pio_data);
6115 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6119 kfree(vcpu->arch.mce_banks);
6120 kvm_free_lapic(vcpu);
6121 idx = srcu_read_lock(&vcpu->kvm->srcu);
6122 kvm_mmu_destroy(vcpu);
6123 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6124 free_page((unsigned long)vcpu->arch.pio_data);
6127 int kvm_arch_init_vm(struct kvm *kvm)
6129 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6130 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6132 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6133 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6135 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6140 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6143 kvm_mmu_unload(vcpu);
6147 static void kvm_free_vcpus(struct kvm *kvm)
6150 struct kvm_vcpu *vcpu;
6153 * Unpin any mmu pages first.
6155 kvm_for_each_vcpu(i, vcpu, kvm) {
6156 kvm_clear_async_pf_completion_queue(vcpu);
6157 kvm_unload_vcpu_mmu(vcpu);
6159 kvm_for_each_vcpu(i, vcpu, kvm)
6160 kvm_arch_vcpu_free(vcpu);
6162 mutex_lock(&kvm->lock);
6163 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6164 kvm->vcpus[i] = NULL;
6166 atomic_set(&kvm->online_vcpus, 0);
6167 mutex_unlock(&kvm->lock);
6170 void kvm_arch_sync_events(struct kvm *kvm)
6172 kvm_free_all_assigned_devices(kvm);
6176 void kvm_arch_destroy_vm(struct kvm *kvm)
6178 kvm_iommu_unmap_guest(kvm);
6179 kfree(kvm->arch.vpic);
6180 kfree(kvm->arch.vioapic);
6181 kvm_free_vcpus(kvm);
6182 if (kvm->arch.apic_access_page)
6183 put_page(kvm->arch.apic_access_page);
6184 if (kvm->arch.ept_identity_pagetable)
6185 put_page(kvm->arch.ept_identity_pagetable);
6188 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6189 struct kvm_memory_slot *memslot,
6190 struct kvm_memory_slot old,
6191 struct kvm_userspace_memory_region *mem,
6194 int npages = memslot->npages;
6195 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6197 /* Prevent internal slot pages from being moved by fork()/COW. */
6198 if (memslot->id >= KVM_MEMORY_SLOTS)
6199 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6201 /*To keep backward compatibility with older userspace,
6202 *x86 needs to hanlde !user_alloc case.
6205 if (npages && !old.rmap) {
6206 unsigned long userspace_addr;
6208 down_write(¤t->mm->mmap_sem);
6209 userspace_addr = do_mmap(NULL, 0,
6211 PROT_READ | PROT_WRITE,
6214 up_write(¤t->mm->mmap_sem);
6216 if (IS_ERR((void *)userspace_addr))
6217 return PTR_ERR((void *)userspace_addr);
6219 memslot->userspace_addr = userspace_addr;
6227 void kvm_arch_commit_memory_region(struct kvm *kvm,
6228 struct kvm_userspace_memory_region *mem,
6229 struct kvm_memory_slot old,
6233 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6235 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6238 down_write(¤t->mm->mmap_sem);
6239 ret = do_munmap(current->mm, old.userspace_addr,
6240 old.npages * PAGE_SIZE);
6241 up_write(¤t->mm->mmap_sem);
6244 "kvm_vm_ioctl_set_memory_region: "
6245 "failed to munmap memory\n");
6248 if (!kvm->arch.n_requested_mmu_pages)
6249 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6251 spin_lock(&kvm->mmu_lock);
6253 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6254 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6255 spin_unlock(&kvm->mmu_lock);
6258 void kvm_arch_flush_shadow(struct kvm *kvm)
6260 kvm_mmu_zap_all(kvm);
6261 kvm_reload_remote_mmus(kvm);
6264 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6266 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6267 !vcpu->arch.apf.halted)
6268 || !list_empty_careful(&vcpu->async_pf.done)
6269 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6270 || vcpu->arch.nmi_pending ||
6271 (kvm_arch_interrupt_allowed(vcpu) &&
6272 kvm_cpu_has_interrupt(vcpu));
6275 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6278 int cpu = vcpu->cpu;
6280 if (waitqueue_active(&vcpu->wq)) {
6281 wake_up_interruptible(&vcpu->wq);
6282 ++vcpu->stat.halt_wakeup;
6286 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6287 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6288 smp_send_reschedule(cpu);
6292 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6294 return kvm_x86_ops->interrupt_allowed(vcpu);
6297 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6299 unsigned long current_rip = kvm_rip_read(vcpu) +
6300 get_segment_base(vcpu, VCPU_SREG_CS);
6302 return current_rip == linear_rip;
6304 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6306 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6308 unsigned long rflags;
6310 rflags = kvm_x86_ops->get_rflags(vcpu);
6311 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6312 rflags &= ~X86_EFLAGS_TF;
6315 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6317 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6319 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6320 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6321 rflags |= X86_EFLAGS_TF;
6322 kvm_x86_ops->set_rflags(vcpu, rflags);
6323 kvm_make_request(KVM_REQ_EVENT, vcpu);
6325 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6327 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6331 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6332 is_error_page(work->page))
6335 r = kvm_mmu_reload(vcpu);
6339 if (!vcpu->arch.mmu.direct_map &&
6340 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6343 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6346 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6348 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6351 static inline u32 kvm_async_pf_next_probe(u32 key)
6353 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6356 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6358 u32 key = kvm_async_pf_hash_fn(gfn);
6360 while (vcpu->arch.apf.gfns[key] != ~0)
6361 key = kvm_async_pf_next_probe(key);
6363 vcpu->arch.apf.gfns[key] = gfn;
6366 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6369 u32 key = kvm_async_pf_hash_fn(gfn);
6371 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6372 (vcpu->arch.apf.gfns[key] != gfn &&
6373 vcpu->arch.apf.gfns[key] != ~0); i++)
6374 key = kvm_async_pf_next_probe(key);
6379 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6381 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6384 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6388 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6390 vcpu->arch.apf.gfns[i] = ~0;
6392 j = kvm_async_pf_next_probe(j);
6393 if (vcpu->arch.apf.gfns[j] == ~0)
6395 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6397 * k lies cyclically in ]i,j]
6399 * |....j i.k.| or |.k..j i...|
6401 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6402 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6407 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6410 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6414 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6415 struct kvm_async_pf *work)
6417 struct x86_exception fault;
6419 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6420 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6422 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6423 (vcpu->arch.apf.send_user_only &&
6424 kvm_x86_ops->get_cpl(vcpu) == 0))
6425 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6426 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6427 fault.vector = PF_VECTOR;
6428 fault.error_code_valid = true;
6429 fault.error_code = 0;
6430 fault.nested_page_fault = false;
6431 fault.address = work->arch.token;
6432 kvm_inject_page_fault(vcpu, &fault);
6436 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6437 struct kvm_async_pf *work)
6439 struct x86_exception fault;
6441 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6442 if (is_error_page(work->page))
6443 work->arch.token = ~0; /* broadcast wakeup */
6445 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6447 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6448 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6449 fault.vector = PF_VECTOR;
6450 fault.error_code_valid = true;
6451 fault.error_code = 0;
6452 fault.nested_page_fault = false;
6453 fault.address = work->arch.token;
6454 kvm_inject_page_fault(vcpu, &fault);
6456 vcpu->arch.apf.halted = false;
6459 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6461 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6464 return !kvm_event_needs_reinjection(vcpu) &&
6465 kvm_x86_ops->interrupt_allowed(vcpu);
6468 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6469 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6470 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6471 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6472 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6473 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6474 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6475 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6476 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6477 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6478 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6479 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);