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 KVM_MAX_MCE_BANKS 32
64 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
66 #define emul_to_vcpu(ctxt) \
67 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
70 * - enable syscall per default because its emulated by KVM
71 * - enable LME and LMA per default on 64 bit KVM
75 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
77 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
80 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
81 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
83 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
84 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
85 struct kvm_cpuid_entry2 __user *entries);
87 struct kvm_x86_ops *kvm_x86_ops;
88 EXPORT_SYMBOL_GPL(kvm_x86_ops);
91 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
93 bool kvm_has_tsc_control;
94 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
95 u32 kvm_max_guest_tsc_khz;
96 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
98 #define KVM_NR_SHARED_MSRS 16
100 struct kvm_shared_msrs_global {
102 u32 msrs[KVM_NR_SHARED_MSRS];
105 struct kvm_shared_msrs {
106 struct user_return_notifier urn;
108 struct kvm_shared_msr_values {
111 } values[KVM_NR_SHARED_MSRS];
114 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
115 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
117 struct kvm_stats_debugfs_item debugfs_entries[] = {
118 { "pf_fixed", VCPU_STAT(pf_fixed) },
119 { "pf_guest", VCPU_STAT(pf_guest) },
120 { "tlb_flush", VCPU_STAT(tlb_flush) },
121 { "invlpg", VCPU_STAT(invlpg) },
122 { "exits", VCPU_STAT(exits) },
123 { "io_exits", VCPU_STAT(io_exits) },
124 { "mmio_exits", VCPU_STAT(mmio_exits) },
125 { "signal_exits", VCPU_STAT(signal_exits) },
126 { "irq_window", VCPU_STAT(irq_window_exits) },
127 { "nmi_window", VCPU_STAT(nmi_window_exits) },
128 { "halt_exits", VCPU_STAT(halt_exits) },
129 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
130 { "hypercalls", VCPU_STAT(hypercalls) },
131 { "request_irq", VCPU_STAT(request_irq_exits) },
132 { "irq_exits", VCPU_STAT(irq_exits) },
133 { "host_state_reload", VCPU_STAT(host_state_reload) },
134 { "efer_reload", VCPU_STAT(efer_reload) },
135 { "fpu_reload", VCPU_STAT(fpu_reload) },
136 { "insn_emulation", VCPU_STAT(insn_emulation) },
137 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
138 { "irq_injections", VCPU_STAT(irq_injections) },
139 { "nmi_injections", VCPU_STAT(nmi_injections) },
140 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
141 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
142 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
143 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
144 { "mmu_flooded", VM_STAT(mmu_flooded) },
145 { "mmu_recycled", VM_STAT(mmu_recycled) },
146 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
147 { "mmu_unsync", VM_STAT(mmu_unsync) },
148 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
149 { "largepages", VM_STAT(lpages) },
153 u64 __read_mostly host_xcr0;
155 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
157 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
160 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
161 vcpu->arch.apf.gfns[i] = ~0;
164 static void kvm_on_user_return(struct user_return_notifier *urn)
167 struct kvm_shared_msrs *locals
168 = container_of(urn, struct kvm_shared_msrs, urn);
169 struct kvm_shared_msr_values *values;
171 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
172 values = &locals->values[slot];
173 if (values->host != values->curr) {
174 wrmsrl(shared_msrs_global.msrs[slot], values->host);
175 values->curr = values->host;
178 locals->registered = false;
179 user_return_notifier_unregister(urn);
182 static void shared_msr_update(unsigned slot, u32 msr)
184 struct kvm_shared_msrs *smsr;
187 smsr = &__get_cpu_var(shared_msrs);
188 /* only read, and nobody should modify it at this time,
189 * so don't need lock */
190 if (slot >= shared_msrs_global.nr) {
191 printk(KERN_ERR "kvm: invalid MSR slot!");
194 rdmsrl_safe(msr, &value);
195 smsr->values[slot].host = value;
196 smsr->values[slot].curr = value;
199 void kvm_define_shared_msr(unsigned slot, u32 msr)
201 if (slot >= shared_msrs_global.nr)
202 shared_msrs_global.nr = slot + 1;
203 shared_msrs_global.msrs[slot] = msr;
204 /* we need ensured the shared_msr_global have been updated */
207 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
209 static void kvm_shared_msr_cpu_online(void)
213 for (i = 0; i < shared_msrs_global.nr; ++i)
214 shared_msr_update(i, shared_msrs_global.msrs[i]);
217 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
219 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
221 if (((value ^ smsr->values[slot].curr) & mask) == 0)
223 smsr->values[slot].curr = value;
224 wrmsrl(shared_msrs_global.msrs[slot], value);
225 if (!smsr->registered) {
226 smsr->urn.on_user_return = kvm_on_user_return;
227 user_return_notifier_register(&smsr->urn);
228 smsr->registered = true;
231 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
233 static void drop_user_return_notifiers(void *ignore)
235 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
237 if (smsr->registered)
238 kvm_on_user_return(&smsr->urn);
241 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
243 if (irqchip_in_kernel(vcpu->kvm))
244 return vcpu->arch.apic_base;
246 return vcpu->arch.apic_base;
248 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
250 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
252 /* TODO: reserve bits check */
253 if (irqchip_in_kernel(vcpu->kvm))
254 kvm_lapic_set_base(vcpu, data);
256 vcpu->arch.apic_base = data;
258 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
260 #define EXCPT_BENIGN 0
261 #define EXCPT_CONTRIBUTORY 1
264 static int exception_class(int vector)
274 return EXCPT_CONTRIBUTORY;
281 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
282 unsigned nr, bool has_error, u32 error_code,
288 kvm_make_request(KVM_REQ_EVENT, vcpu);
290 if (!vcpu->arch.exception.pending) {
292 vcpu->arch.exception.pending = true;
293 vcpu->arch.exception.has_error_code = has_error;
294 vcpu->arch.exception.nr = nr;
295 vcpu->arch.exception.error_code = error_code;
296 vcpu->arch.exception.reinject = reinject;
300 /* to check exception */
301 prev_nr = vcpu->arch.exception.nr;
302 if (prev_nr == DF_VECTOR) {
303 /* triple fault -> shutdown */
304 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
307 class1 = exception_class(prev_nr);
308 class2 = exception_class(nr);
309 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
310 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
311 /* generate double fault per SDM Table 5-5 */
312 vcpu->arch.exception.pending = true;
313 vcpu->arch.exception.has_error_code = true;
314 vcpu->arch.exception.nr = DF_VECTOR;
315 vcpu->arch.exception.error_code = 0;
317 /* replace previous exception with a new one in a hope
318 that instruction re-execution will regenerate lost
323 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
325 kvm_multiple_exception(vcpu, nr, false, 0, false);
327 EXPORT_SYMBOL_GPL(kvm_queue_exception);
329 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
331 kvm_multiple_exception(vcpu, nr, false, 0, true);
333 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
335 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
338 kvm_inject_gp(vcpu, 0);
340 kvm_x86_ops->skip_emulated_instruction(vcpu);
342 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
344 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
346 ++vcpu->stat.pf_guest;
347 vcpu->arch.cr2 = fault->address;
348 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
350 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
352 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
354 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
355 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
357 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
360 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
362 kvm_make_request(KVM_REQ_EVENT, vcpu);
363 vcpu->arch.nmi_pending = 1;
365 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
367 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
369 kvm_multiple_exception(vcpu, nr, true, error_code, false);
371 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
373 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
375 kvm_multiple_exception(vcpu, nr, true, error_code, true);
377 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
380 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
381 * a #GP and return false.
383 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
385 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
387 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
390 EXPORT_SYMBOL_GPL(kvm_require_cpl);
393 * This function will be used to read from the physical memory of the currently
394 * running guest. The difference to kvm_read_guest_page is that this function
395 * can read from guest physical or from the guest's guest physical memory.
397 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
398 gfn_t ngfn, void *data, int offset, int len,
404 ngpa = gfn_to_gpa(ngfn);
405 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
406 if (real_gfn == UNMAPPED_GVA)
409 real_gfn = gpa_to_gfn(real_gfn);
411 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
413 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
415 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
416 void *data, int offset, int len, u32 access)
418 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
419 data, offset, len, access);
423 * Load the pae pdptrs. Return true is they are all valid.
425 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
427 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
428 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
431 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
433 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
434 offset * sizeof(u64), sizeof(pdpte),
435 PFERR_USER_MASK|PFERR_WRITE_MASK);
440 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
441 if (is_present_gpte(pdpte[i]) &&
442 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
449 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
450 __set_bit(VCPU_EXREG_PDPTR,
451 (unsigned long *)&vcpu->arch.regs_avail);
452 __set_bit(VCPU_EXREG_PDPTR,
453 (unsigned long *)&vcpu->arch.regs_dirty);
458 EXPORT_SYMBOL_GPL(load_pdptrs);
460 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
462 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
468 if (is_long_mode(vcpu) || !is_pae(vcpu))
471 if (!test_bit(VCPU_EXREG_PDPTR,
472 (unsigned long *)&vcpu->arch.regs_avail))
475 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
476 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
477 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
478 PFERR_USER_MASK | PFERR_WRITE_MASK);
481 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
487 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
489 unsigned long old_cr0 = kvm_read_cr0(vcpu);
490 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
491 X86_CR0_CD | X86_CR0_NW;
496 if (cr0 & 0xffffffff00000000UL)
500 cr0 &= ~CR0_RESERVED_BITS;
502 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
505 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
508 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
510 if ((vcpu->arch.efer & EFER_LME)) {
515 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
520 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
525 kvm_x86_ops->set_cr0(vcpu, cr0);
527 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
528 kvm_clear_async_pf_completion_queue(vcpu);
529 kvm_async_pf_hash_reset(vcpu);
532 if ((cr0 ^ old_cr0) & update_bits)
533 kvm_mmu_reset_context(vcpu);
536 EXPORT_SYMBOL_GPL(kvm_set_cr0);
538 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
540 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
542 EXPORT_SYMBOL_GPL(kvm_lmsw);
544 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
548 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
549 if (index != XCR_XFEATURE_ENABLED_MASK)
552 if (kvm_x86_ops->get_cpl(vcpu) != 0)
554 if (!(xcr0 & XSTATE_FP))
556 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
558 if (xcr0 & ~host_xcr0)
560 vcpu->arch.xcr0 = xcr0;
561 vcpu->guest_xcr0_loaded = 0;
565 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
567 if (__kvm_set_xcr(vcpu, index, xcr)) {
568 kvm_inject_gp(vcpu, 0);
573 EXPORT_SYMBOL_GPL(kvm_set_xcr);
575 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
577 struct kvm_cpuid_entry2 *best;
579 best = kvm_find_cpuid_entry(vcpu, 1, 0);
580 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
583 static bool guest_cpuid_has_smep(struct kvm_vcpu *vcpu)
585 struct kvm_cpuid_entry2 *best;
587 best = kvm_find_cpuid_entry(vcpu, 7, 0);
588 return best && (best->ebx & bit(X86_FEATURE_SMEP));
591 static bool guest_cpuid_has_fsgsbase(struct kvm_vcpu *vcpu)
593 struct kvm_cpuid_entry2 *best;
595 best = kvm_find_cpuid_entry(vcpu, 7, 0);
596 return best && (best->ebx & bit(X86_FEATURE_FSGSBASE));
599 static void update_cpuid(struct kvm_vcpu *vcpu)
601 struct kvm_cpuid_entry2 *best;
603 best = kvm_find_cpuid_entry(vcpu, 1, 0);
607 /* Update OSXSAVE bit */
608 if (cpu_has_xsave && best->function == 0x1) {
609 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
610 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
611 best->ecx |= bit(X86_FEATURE_OSXSAVE);
615 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
617 unsigned long old_cr4 = kvm_read_cr4(vcpu);
618 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
619 X86_CR4_PAE | X86_CR4_SMEP;
620 if (cr4 & CR4_RESERVED_BITS)
623 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
626 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
629 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
632 if (is_long_mode(vcpu)) {
633 if (!(cr4 & X86_CR4_PAE))
635 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
636 && ((cr4 ^ old_cr4) & pdptr_bits)
637 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
641 if (kvm_x86_ops->set_cr4(vcpu, cr4))
644 if ((cr4 ^ old_cr4) & pdptr_bits)
645 kvm_mmu_reset_context(vcpu);
647 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
652 EXPORT_SYMBOL_GPL(kvm_set_cr4);
654 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
656 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
657 kvm_mmu_sync_roots(vcpu);
658 kvm_mmu_flush_tlb(vcpu);
662 if (is_long_mode(vcpu)) {
663 if (cr3 & CR3_L_MODE_RESERVED_BITS)
667 if (cr3 & CR3_PAE_RESERVED_BITS)
669 if (is_paging(vcpu) &&
670 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
674 * We don't check reserved bits in nonpae mode, because
675 * this isn't enforced, and VMware depends on this.
680 * Does the new cr3 value map to physical memory? (Note, we
681 * catch an invalid cr3 even in real-mode, because it would
682 * cause trouble later on when we turn on paging anyway.)
684 * A real CPU would silently accept an invalid cr3 and would
685 * attempt to use it - with largely undefined (and often hard
686 * to debug) behavior on the guest side.
688 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
690 vcpu->arch.cr3 = cr3;
691 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
692 vcpu->arch.mmu.new_cr3(vcpu);
695 EXPORT_SYMBOL_GPL(kvm_set_cr3);
697 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
699 if (cr8 & CR8_RESERVED_BITS)
701 if (irqchip_in_kernel(vcpu->kvm))
702 kvm_lapic_set_tpr(vcpu, cr8);
704 vcpu->arch.cr8 = cr8;
707 EXPORT_SYMBOL_GPL(kvm_set_cr8);
709 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
711 if (irqchip_in_kernel(vcpu->kvm))
712 return kvm_lapic_get_cr8(vcpu);
714 return vcpu->arch.cr8;
716 EXPORT_SYMBOL_GPL(kvm_get_cr8);
718 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
722 vcpu->arch.db[dr] = val;
723 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
724 vcpu->arch.eff_db[dr] = val;
727 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
731 if (val & 0xffffffff00000000ULL)
733 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
736 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
740 if (val & 0xffffffff00000000ULL)
742 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
743 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
744 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
745 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
753 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
757 res = __kvm_set_dr(vcpu, dr, val);
759 kvm_queue_exception(vcpu, UD_VECTOR);
761 kvm_inject_gp(vcpu, 0);
765 EXPORT_SYMBOL_GPL(kvm_set_dr);
767 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
771 *val = vcpu->arch.db[dr];
774 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
778 *val = vcpu->arch.dr6;
781 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
785 *val = vcpu->arch.dr7;
792 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
794 if (_kvm_get_dr(vcpu, dr, val)) {
795 kvm_queue_exception(vcpu, UD_VECTOR);
800 EXPORT_SYMBOL_GPL(kvm_get_dr);
803 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
804 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
806 * This list is modified at module load time to reflect the
807 * capabilities of the host cpu. This capabilities test skips MSRs that are
808 * kvm-specific. Those are put in the beginning of the list.
811 #define KVM_SAVE_MSRS_BEGIN 9
812 static u32 msrs_to_save[] = {
813 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
814 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
815 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
816 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
817 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
820 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
822 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
825 static unsigned num_msrs_to_save;
827 static u32 emulated_msrs[] = {
828 MSR_IA32_MISC_ENABLE,
833 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
835 u64 old_efer = vcpu->arch.efer;
837 if (efer & efer_reserved_bits)
841 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
844 if (efer & EFER_FFXSR) {
845 struct kvm_cpuid_entry2 *feat;
847 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
848 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
852 if (efer & EFER_SVME) {
853 struct kvm_cpuid_entry2 *feat;
855 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
856 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
861 efer |= vcpu->arch.efer & EFER_LMA;
863 kvm_x86_ops->set_efer(vcpu, efer);
865 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
867 /* Update reserved bits */
868 if ((efer ^ old_efer) & EFER_NX)
869 kvm_mmu_reset_context(vcpu);
874 void kvm_enable_efer_bits(u64 mask)
876 efer_reserved_bits &= ~mask;
878 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
882 * Writes msr value into into the appropriate "register".
883 * Returns 0 on success, non-0 otherwise.
884 * Assumes vcpu_load() was already called.
886 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
888 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
892 * Adapt set_msr() to msr_io()'s calling convention
894 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
896 return kvm_set_msr(vcpu, index, *data);
899 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
903 struct pvclock_wall_clock wc;
904 struct timespec boot;
909 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
914 ++version; /* first time write, random junk */
918 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
921 * The guest calculates current wall clock time by adding
922 * system time (updated by kvm_guest_time_update below) to the
923 * wall clock specified here. guest system time equals host
924 * system time for us, thus we must fill in host boot time here.
928 wc.sec = boot.tv_sec;
929 wc.nsec = boot.tv_nsec;
930 wc.version = version;
932 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
935 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
938 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
940 uint32_t quotient, remainder;
942 /* Don't try to replace with do_div(), this one calculates
943 * "(dividend << 32) / divisor" */
945 : "=a" (quotient), "=d" (remainder)
946 : "0" (0), "1" (dividend), "r" (divisor) );
950 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
951 s8 *pshift, u32 *pmultiplier)
958 tps64 = base_khz * 1000LL;
959 scaled64 = scaled_khz * 1000LL;
960 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
965 tps32 = (uint32_t)tps64;
966 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
967 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
975 *pmultiplier = div_frac(scaled64, tps32);
977 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
978 __func__, base_khz, scaled_khz, shift, *pmultiplier);
981 static inline u64 get_kernel_ns(void)
985 WARN_ON(preemptible());
987 monotonic_to_bootbased(&ts);
988 return timespec_to_ns(&ts);
991 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
992 unsigned long max_tsc_khz;
994 static inline int kvm_tsc_changes_freq(void)
997 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
998 cpufreq_quick_get(cpu) != 0;
1003 static u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
1005 if (vcpu->arch.virtual_tsc_khz)
1006 return vcpu->arch.virtual_tsc_khz;
1008 return __this_cpu_read(cpu_tsc_khz);
1011 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1015 WARN_ON(preemptible());
1016 if (kvm_tsc_changes_freq())
1017 printk_once(KERN_WARNING
1018 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
1019 ret = nsec * vcpu_tsc_khz(vcpu);
1020 do_div(ret, USEC_PER_SEC);
1024 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1026 /* Compute a scale to convert nanoseconds in TSC cycles */
1027 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1028 &vcpu->arch.tsc_catchup_shift,
1029 &vcpu->arch.tsc_catchup_mult);
1032 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1034 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1035 vcpu->arch.tsc_catchup_mult,
1036 vcpu->arch.tsc_catchup_shift);
1037 tsc += vcpu->arch.last_tsc_write;
1041 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1043 struct kvm *kvm = vcpu->kvm;
1044 u64 offset, ns, elapsed;
1045 unsigned long flags;
1048 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1049 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1050 ns = get_kernel_ns();
1051 elapsed = ns - kvm->arch.last_tsc_nsec;
1052 sdiff = data - kvm->arch.last_tsc_write;
1057 * Special case: close write to TSC within 5 seconds of
1058 * another CPU is interpreted as an attempt to synchronize
1059 * The 5 seconds is to accommodate host load / swapping as
1060 * well as any reset of TSC during the boot process.
1062 * In that case, for a reliable TSC, we can match TSC offsets,
1063 * or make a best guest using elapsed value.
1065 if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1066 elapsed < 5ULL * NSEC_PER_SEC) {
1067 if (!check_tsc_unstable()) {
1068 offset = kvm->arch.last_tsc_offset;
1069 pr_debug("kvm: matched tsc offset for %llu\n", data);
1071 u64 delta = nsec_to_cycles(vcpu, elapsed);
1073 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1075 ns = kvm->arch.last_tsc_nsec;
1077 kvm->arch.last_tsc_nsec = ns;
1078 kvm->arch.last_tsc_write = data;
1079 kvm->arch.last_tsc_offset = offset;
1080 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1081 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1083 /* Reset of TSC must disable overshoot protection below */
1084 vcpu->arch.hv_clock.tsc_timestamp = 0;
1085 vcpu->arch.last_tsc_write = data;
1086 vcpu->arch.last_tsc_nsec = ns;
1088 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1090 static int kvm_guest_time_update(struct kvm_vcpu *v)
1092 unsigned long flags;
1093 struct kvm_vcpu_arch *vcpu = &v->arch;
1095 unsigned long this_tsc_khz;
1096 s64 kernel_ns, max_kernel_ns;
1099 /* Keep irq disabled to prevent changes to the clock */
1100 local_irq_save(flags);
1101 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1102 kernel_ns = get_kernel_ns();
1103 this_tsc_khz = vcpu_tsc_khz(v);
1104 if (unlikely(this_tsc_khz == 0)) {
1105 local_irq_restore(flags);
1106 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1111 * We may have to catch up the TSC to match elapsed wall clock
1112 * time for two reasons, even if kvmclock is used.
1113 * 1) CPU could have been running below the maximum TSC rate
1114 * 2) Broken TSC compensation resets the base at each VCPU
1115 * entry to avoid unknown leaps of TSC even when running
1116 * again on the same CPU. This may cause apparent elapsed
1117 * time to disappear, and the guest to stand still or run
1120 if (vcpu->tsc_catchup) {
1121 u64 tsc = compute_guest_tsc(v, kernel_ns);
1122 if (tsc > tsc_timestamp) {
1123 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1124 tsc_timestamp = tsc;
1128 local_irq_restore(flags);
1130 if (!vcpu->time_page)
1134 * Time as measured by the TSC may go backwards when resetting the base
1135 * tsc_timestamp. The reason for this is that the TSC resolution is
1136 * higher than the resolution of the other clock scales. Thus, many
1137 * possible measurments of the TSC correspond to one measurement of any
1138 * other clock, and so a spread of values is possible. This is not a
1139 * problem for the computation of the nanosecond clock; with TSC rates
1140 * around 1GHZ, there can only be a few cycles which correspond to one
1141 * nanosecond value, and any path through this code will inevitably
1142 * take longer than that. However, with the kernel_ns value itself,
1143 * the precision may be much lower, down to HZ granularity. If the
1144 * first sampling of TSC against kernel_ns ends in the low part of the
1145 * range, and the second in the high end of the range, we can get:
1147 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1149 * As the sampling errors potentially range in the thousands of cycles,
1150 * it is possible such a time value has already been observed by the
1151 * guest. To protect against this, we must compute the system time as
1152 * observed by the guest and ensure the new system time is greater.
1155 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1156 max_kernel_ns = vcpu->last_guest_tsc -
1157 vcpu->hv_clock.tsc_timestamp;
1158 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1159 vcpu->hv_clock.tsc_to_system_mul,
1160 vcpu->hv_clock.tsc_shift);
1161 max_kernel_ns += vcpu->last_kernel_ns;
1164 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1165 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1166 &vcpu->hv_clock.tsc_shift,
1167 &vcpu->hv_clock.tsc_to_system_mul);
1168 vcpu->hw_tsc_khz = this_tsc_khz;
1171 if (max_kernel_ns > kernel_ns)
1172 kernel_ns = max_kernel_ns;
1174 /* With all the info we got, fill in the values */
1175 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1176 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1177 vcpu->last_kernel_ns = kernel_ns;
1178 vcpu->last_guest_tsc = tsc_timestamp;
1179 vcpu->hv_clock.flags = 0;
1182 * The interface expects us to write an even number signaling that the
1183 * update is finished. Since the guest won't see the intermediate
1184 * state, we just increase by 2 at the end.
1186 vcpu->hv_clock.version += 2;
1188 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1190 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1191 sizeof(vcpu->hv_clock));
1193 kunmap_atomic(shared_kaddr, KM_USER0);
1195 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1199 static bool msr_mtrr_valid(unsigned msr)
1202 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1203 case MSR_MTRRfix64K_00000:
1204 case MSR_MTRRfix16K_80000:
1205 case MSR_MTRRfix16K_A0000:
1206 case MSR_MTRRfix4K_C0000:
1207 case MSR_MTRRfix4K_C8000:
1208 case MSR_MTRRfix4K_D0000:
1209 case MSR_MTRRfix4K_D8000:
1210 case MSR_MTRRfix4K_E0000:
1211 case MSR_MTRRfix4K_E8000:
1212 case MSR_MTRRfix4K_F0000:
1213 case MSR_MTRRfix4K_F8000:
1214 case MSR_MTRRdefType:
1215 case MSR_IA32_CR_PAT:
1223 static bool valid_pat_type(unsigned t)
1225 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1228 static bool valid_mtrr_type(unsigned t)
1230 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1233 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1237 if (!msr_mtrr_valid(msr))
1240 if (msr == MSR_IA32_CR_PAT) {
1241 for (i = 0; i < 8; i++)
1242 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1245 } else if (msr == MSR_MTRRdefType) {
1248 return valid_mtrr_type(data & 0xff);
1249 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1250 for (i = 0; i < 8 ; i++)
1251 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1256 /* variable MTRRs */
1257 return valid_mtrr_type(data & 0xff);
1260 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1262 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1264 if (!mtrr_valid(vcpu, msr, data))
1267 if (msr == MSR_MTRRdefType) {
1268 vcpu->arch.mtrr_state.def_type = data;
1269 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1270 } else if (msr == MSR_MTRRfix64K_00000)
1272 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1273 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1274 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1275 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1276 else if (msr == MSR_IA32_CR_PAT)
1277 vcpu->arch.pat = data;
1278 else { /* Variable MTRRs */
1279 int idx, is_mtrr_mask;
1282 idx = (msr - 0x200) / 2;
1283 is_mtrr_mask = msr - 0x200 - 2 * idx;
1286 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1289 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1293 kvm_mmu_reset_context(vcpu);
1297 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1299 u64 mcg_cap = vcpu->arch.mcg_cap;
1300 unsigned bank_num = mcg_cap & 0xff;
1303 case MSR_IA32_MCG_STATUS:
1304 vcpu->arch.mcg_status = data;
1306 case MSR_IA32_MCG_CTL:
1307 if (!(mcg_cap & MCG_CTL_P))
1309 if (data != 0 && data != ~(u64)0)
1311 vcpu->arch.mcg_ctl = data;
1314 if (msr >= MSR_IA32_MC0_CTL &&
1315 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1316 u32 offset = msr - MSR_IA32_MC0_CTL;
1317 /* only 0 or all 1s can be written to IA32_MCi_CTL
1318 * some Linux kernels though clear bit 10 in bank 4 to
1319 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1320 * this to avoid an uncatched #GP in the guest
1322 if ((offset & 0x3) == 0 &&
1323 data != 0 && (data | (1 << 10)) != ~(u64)0)
1325 vcpu->arch.mce_banks[offset] = data;
1333 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1335 struct kvm *kvm = vcpu->kvm;
1336 int lm = is_long_mode(vcpu);
1337 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1338 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1339 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1340 : kvm->arch.xen_hvm_config.blob_size_32;
1341 u32 page_num = data & ~PAGE_MASK;
1342 u64 page_addr = data & PAGE_MASK;
1347 if (page_num >= blob_size)
1350 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1354 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1356 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1365 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1367 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1370 static bool kvm_hv_msr_partition_wide(u32 msr)
1374 case HV_X64_MSR_GUEST_OS_ID:
1375 case HV_X64_MSR_HYPERCALL:
1383 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1385 struct kvm *kvm = vcpu->kvm;
1388 case HV_X64_MSR_GUEST_OS_ID:
1389 kvm->arch.hv_guest_os_id = data;
1390 /* setting guest os id to zero disables hypercall page */
1391 if (!kvm->arch.hv_guest_os_id)
1392 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1394 case HV_X64_MSR_HYPERCALL: {
1399 /* if guest os id is not set hypercall should remain disabled */
1400 if (!kvm->arch.hv_guest_os_id)
1402 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1403 kvm->arch.hv_hypercall = data;
1406 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1407 addr = gfn_to_hva(kvm, gfn);
1408 if (kvm_is_error_hva(addr))
1410 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1411 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1412 if (__copy_to_user((void __user *)addr, instructions, 4))
1414 kvm->arch.hv_hypercall = data;
1418 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1419 "data 0x%llx\n", msr, data);
1425 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1428 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1431 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1432 vcpu->arch.hv_vapic = data;
1435 addr = gfn_to_hva(vcpu->kvm, data >>
1436 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1437 if (kvm_is_error_hva(addr))
1439 if (__clear_user((void __user *)addr, PAGE_SIZE))
1441 vcpu->arch.hv_vapic = data;
1444 case HV_X64_MSR_EOI:
1445 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1446 case HV_X64_MSR_ICR:
1447 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1448 case HV_X64_MSR_TPR:
1449 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1451 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1452 "data 0x%llx\n", msr, data);
1459 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1461 gpa_t gpa = data & ~0x3f;
1463 /* Bits 2:5 are resrved, Should be zero */
1467 vcpu->arch.apf.msr_val = data;
1469 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1470 kvm_clear_async_pf_completion_queue(vcpu);
1471 kvm_async_pf_hash_reset(vcpu);
1475 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1478 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1479 kvm_async_pf_wakeup_all(vcpu);
1483 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1485 if (vcpu->arch.time_page) {
1486 kvm_release_page_dirty(vcpu->arch.time_page);
1487 vcpu->arch.time_page = NULL;
1491 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1495 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1498 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1499 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1500 vcpu->arch.st.accum_steal = delta;
1503 static void record_steal_time(struct kvm_vcpu *vcpu)
1505 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1508 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1509 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1512 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1513 vcpu->arch.st.steal.version += 2;
1514 vcpu->arch.st.accum_steal = 0;
1516 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1517 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1520 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1524 return set_efer(vcpu, data);
1526 data &= ~(u64)0x40; /* ignore flush filter disable */
1527 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1529 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1534 case MSR_FAM10H_MMIO_CONF_BASE:
1536 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1541 case MSR_AMD64_NB_CFG:
1543 case MSR_IA32_DEBUGCTLMSR:
1545 /* We support the non-activated case already */
1547 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1548 /* Values other than LBR and BTF are vendor-specific,
1549 thus reserved and should throw a #GP */
1552 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1555 case MSR_IA32_UCODE_REV:
1556 case MSR_IA32_UCODE_WRITE:
1557 case MSR_VM_HSAVE_PA:
1558 case MSR_AMD64_PATCH_LOADER:
1560 case 0x200 ... 0x2ff:
1561 return set_msr_mtrr(vcpu, msr, data);
1562 case MSR_IA32_APICBASE:
1563 kvm_set_apic_base(vcpu, data);
1565 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1566 return kvm_x2apic_msr_write(vcpu, msr, data);
1567 case MSR_IA32_MISC_ENABLE:
1568 vcpu->arch.ia32_misc_enable_msr = data;
1570 case MSR_KVM_WALL_CLOCK_NEW:
1571 case MSR_KVM_WALL_CLOCK:
1572 vcpu->kvm->arch.wall_clock = data;
1573 kvm_write_wall_clock(vcpu->kvm, data);
1575 case MSR_KVM_SYSTEM_TIME_NEW:
1576 case MSR_KVM_SYSTEM_TIME: {
1577 kvmclock_reset(vcpu);
1579 vcpu->arch.time = data;
1580 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1582 /* we verify if the enable bit is set... */
1586 /* ...but clean it before doing the actual write */
1587 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1589 vcpu->arch.time_page =
1590 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1592 if (is_error_page(vcpu->arch.time_page)) {
1593 kvm_release_page_clean(vcpu->arch.time_page);
1594 vcpu->arch.time_page = NULL;
1598 case MSR_KVM_ASYNC_PF_EN:
1599 if (kvm_pv_enable_async_pf(vcpu, data))
1602 case MSR_KVM_STEAL_TIME:
1604 if (unlikely(!sched_info_on()))
1607 if (data & KVM_STEAL_RESERVED_MASK)
1610 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1611 data & KVM_STEAL_VALID_BITS))
1614 vcpu->arch.st.msr_val = data;
1616 if (!(data & KVM_MSR_ENABLED))
1619 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1622 accumulate_steal_time(vcpu);
1625 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1629 case MSR_IA32_MCG_CTL:
1630 case MSR_IA32_MCG_STATUS:
1631 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1632 return set_msr_mce(vcpu, msr, data);
1634 /* Performance counters are not protected by a CPUID bit,
1635 * so we should check all of them in the generic path for the sake of
1636 * cross vendor migration.
1637 * Writing a zero into the event select MSRs disables them,
1638 * which we perfectly emulate ;-). Any other value should be at least
1639 * reported, some guests depend on them.
1641 case MSR_P6_EVNTSEL0:
1642 case MSR_P6_EVNTSEL1:
1643 case MSR_K7_EVNTSEL0:
1644 case MSR_K7_EVNTSEL1:
1645 case MSR_K7_EVNTSEL2:
1646 case MSR_K7_EVNTSEL3:
1648 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1649 "0x%x data 0x%llx\n", msr, data);
1651 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1652 * so we ignore writes to make it happy.
1654 case MSR_P6_PERFCTR0:
1655 case MSR_P6_PERFCTR1:
1656 case MSR_K7_PERFCTR0:
1657 case MSR_K7_PERFCTR1:
1658 case MSR_K7_PERFCTR2:
1659 case MSR_K7_PERFCTR3:
1660 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1661 "0x%x data 0x%llx\n", msr, data);
1663 case MSR_K7_CLK_CTL:
1665 * Ignore all writes to this no longer documented MSR.
1666 * Writes are only relevant for old K7 processors,
1667 * all pre-dating SVM, but a recommended workaround from
1668 * AMD for these chips. It is possible to speicify the
1669 * affected processor models on the command line, hence
1670 * the need to ignore the workaround.
1673 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1674 if (kvm_hv_msr_partition_wide(msr)) {
1676 mutex_lock(&vcpu->kvm->lock);
1677 r = set_msr_hyperv_pw(vcpu, msr, data);
1678 mutex_unlock(&vcpu->kvm->lock);
1681 return set_msr_hyperv(vcpu, msr, data);
1683 case MSR_IA32_BBL_CR_CTL3:
1684 /* Drop writes to this legacy MSR -- see rdmsr
1685 * counterpart for further detail.
1687 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1690 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1691 return xen_hvm_config(vcpu, data);
1693 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1697 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1704 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1708 * Reads an msr value (of 'msr_index') into 'pdata'.
1709 * Returns 0 on success, non-0 otherwise.
1710 * Assumes vcpu_load() was already called.
1712 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1714 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1717 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1719 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1721 if (!msr_mtrr_valid(msr))
1724 if (msr == MSR_MTRRdefType)
1725 *pdata = vcpu->arch.mtrr_state.def_type +
1726 (vcpu->arch.mtrr_state.enabled << 10);
1727 else if (msr == MSR_MTRRfix64K_00000)
1729 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1730 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1731 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1732 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1733 else if (msr == MSR_IA32_CR_PAT)
1734 *pdata = vcpu->arch.pat;
1735 else { /* Variable MTRRs */
1736 int idx, is_mtrr_mask;
1739 idx = (msr - 0x200) / 2;
1740 is_mtrr_mask = msr - 0x200 - 2 * idx;
1743 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1746 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1753 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1756 u64 mcg_cap = vcpu->arch.mcg_cap;
1757 unsigned bank_num = mcg_cap & 0xff;
1760 case MSR_IA32_P5_MC_ADDR:
1761 case MSR_IA32_P5_MC_TYPE:
1764 case MSR_IA32_MCG_CAP:
1765 data = vcpu->arch.mcg_cap;
1767 case MSR_IA32_MCG_CTL:
1768 if (!(mcg_cap & MCG_CTL_P))
1770 data = vcpu->arch.mcg_ctl;
1772 case MSR_IA32_MCG_STATUS:
1773 data = vcpu->arch.mcg_status;
1776 if (msr >= MSR_IA32_MC0_CTL &&
1777 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1778 u32 offset = msr - MSR_IA32_MC0_CTL;
1779 data = vcpu->arch.mce_banks[offset];
1788 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1791 struct kvm *kvm = vcpu->kvm;
1794 case HV_X64_MSR_GUEST_OS_ID:
1795 data = kvm->arch.hv_guest_os_id;
1797 case HV_X64_MSR_HYPERCALL:
1798 data = kvm->arch.hv_hypercall;
1801 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1809 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1814 case HV_X64_MSR_VP_INDEX: {
1817 kvm_for_each_vcpu(r, v, vcpu->kvm)
1822 case HV_X64_MSR_EOI:
1823 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1824 case HV_X64_MSR_ICR:
1825 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1826 case HV_X64_MSR_TPR:
1827 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1828 case HV_X64_MSR_APIC_ASSIST_PAGE:
1829 return vcpu->arch.hv_vapic;
1831 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1838 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1843 case MSR_IA32_PLATFORM_ID:
1844 case MSR_IA32_UCODE_REV:
1845 case MSR_IA32_EBL_CR_POWERON:
1846 case MSR_IA32_DEBUGCTLMSR:
1847 case MSR_IA32_LASTBRANCHFROMIP:
1848 case MSR_IA32_LASTBRANCHTOIP:
1849 case MSR_IA32_LASTINTFROMIP:
1850 case MSR_IA32_LASTINTTOIP:
1853 case MSR_VM_HSAVE_PA:
1854 case MSR_P6_PERFCTR0:
1855 case MSR_P6_PERFCTR1:
1856 case MSR_P6_EVNTSEL0:
1857 case MSR_P6_EVNTSEL1:
1858 case MSR_K7_EVNTSEL0:
1859 case MSR_K7_PERFCTR0:
1860 case MSR_K8_INT_PENDING_MSG:
1861 case MSR_AMD64_NB_CFG:
1862 case MSR_FAM10H_MMIO_CONF_BASE:
1866 data = 0x500 | KVM_NR_VAR_MTRR;
1868 case 0x200 ... 0x2ff:
1869 return get_msr_mtrr(vcpu, msr, pdata);
1870 case 0xcd: /* fsb frequency */
1874 * MSR_EBC_FREQUENCY_ID
1875 * Conservative value valid for even the basic CPU models.
1876 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1877 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1878 * and 266MHz for model 3, or 4. Set Core Clock
1879 * Frequency to System Bus Frequency Ratio to 1 (bits
1880 * 31:24) even though these are only valid for CPU
1881 * models > 2, however guests may end up dividing or
1882 * multiplying by zero otherwise.
1884 case MSR_EBC_FREQUENCY_ID:
1887 case MSR_IA32_APICBASE:
1888 data = kvm_get_apic_base(vcpu);
1890 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1891 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1893 case MSR_IA32_MISC_ENABLE:
1894 data = vcpu->arch.ia32_misc_enable_msr;
1896 case MSR_IA32_PERF_STATUS:
1897 /* TSC increment by tick */
1899 /* CPU multiplier */
1900 data |= (((uint64_t)4ULL) << 40);
1903 data = vcpu->arch.efer;
1905 case MSR_KVM_WALL_CLOCK:
1906 case MSR_KVM_WALL_CLOCK_NEW:
1907 data = vcpu->kvm->arch.wall_clock;
1909 case MSR_KVM_SYSTEM_TIME:
1910 case MSR_KVM_SYSTEM_TIME_NEW:
1911 data = vcpu->arch.time;
1913 case MSR_KVM_ASYNC_PF_EN:
1914 data = vcpu->arch.apf.msr_val;
1916 case MSR_KVM_STEAL_TIME:
1917 data = vcpu->arch.st.msr_val;
1919 case MSR_IA32_P5_MC_ADDR:
1920 case MSR_IA32_P5_MC_TYPE:
1921 case MSR_IA32_MCG_CAP:
1922 case MSR_IA32_MCG_CTL:
1923 case MSR_IA32_MCG_STATUS:
1924 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1925 return get_msr_mce(vcpu, msr, pdata);
1926 case MSR_K7_CLK_CTL:
1928 * Provide expected ramp-up count for K7. All other
1929 * are set to zero, indicating minimum divisors for
1932 * This prevents guest kernels on AMD host with CPU
1933 * type 6, model 8 and higher from exploding due to
1934 * the rdmsr failing.
1938 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1939 if (kvm_hv_msr_partition_wide(msr)) {
1941 mutex_lock(&vcpu->kvm->lock);
1942 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1943 mutex_unlock(&vcpu->kvm->lock);
1946 return get_msr_hyperv(vcpu, msr, pdata);
1948 case MSR_IA32_BBL_CR_CTL3:
1949 /* This legacy MSR exists but isn't fully documented in current
1950 * silicon. It is however accessed by winxp in very narrow
1951 * scenarios where it sets bit #19, itself documented as
1952 * a "reserved" bit. Best effort attempt to source coherent
1953 * read data here should the balance of the register be
1954 * interpreted by the guest:
1956 * L2 cache control register 3: 64GB range, 256KB size,
1957 * enabled, latency 0x1, configured
1963 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1966 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1974 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1977 * Read or write a bunch of msrs. All parameters are kernel addresses.
1979 * @return number of msrs set successfully.
1981 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1982 struct kvm_msr_entry *entries,
1983 int (*do_msr)(struct kvm_vcpu *vcpu,
1984 unsigned index, u64 *data))
1988 idx = srcu_read_lock(&vcpu->kvm->srcu);
1989 for (i = 0; i < msrs->nmsrs; ++i)
1990 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1992 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1998 * Read or write a bunch of msrs. Parameters are user addresses.
2000 * @return number of msrs set successfully.
2002 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2003 int (*do_msr)(struct kvm_vcpu *vcpu,
2004 unsigned index, u64 *data),
2007 struct kvm_msrs msrs;
2008 struct kvm_msr_entry *entries;
2013 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2017 if (msrs.nmsrs >= MAX_IO_MSRS)
2021 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2022 entries = kmalloc(size, GFP_KERNEL);
2027 if (copy_from_user(entries, user_msrs->entries, size))
2030 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2035 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2046 int kvm_dev_ioctl_check_extension(long ext)
2051 case KVM_CAP_IRQCHIP:
2053 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2054 case KVM_CAP_SET_TSS_ADDR:
2055 case KVM_CAP_EXT_CPUID:
2056 case KVM_CAP_CLOCKSOURCE:
2058 case KVM_CAP_NOP_IO_DELAY:
2059 case KVM_CAP_MP_STATE:
2060 case KVM_CAP_SYNC_MMU:
2061 case KVM_CAP_USER_NMI:
2062 case KVM_CAP_REINJECT_CONTROL:
2063 case KVM_CAP_IRQ_INJECT_STATUS:
2064 case KVM_CAP_ASSIGN_DEV_IRQ:
2066 case KVM_CAP_IOEVENTFD:
2068 case KVM_CAP_PIT_STATE2:
2069 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2070 case KVM_CAP_XEN_HVM:
2071 case KVM_CAP_ADJUST_CLOCK:
2072 case KVM_CAP_VCPU_EVENTS:
2073 case KVM_CAP_HYPERV:
2074 case KVM_CAP_HYPERV_VAPIC:
2075 case KVM_CAP_HYPERV_SPIN:
2076 case KVM_CAP_PCI_SEGMENT:
2077 case KVM_CAP_DEBUGREGS:
2078 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2080 case KVM_CAP_ASYNC_PF:
2081 case KVM_CAP_GET_TSC_KHZ:
2084 case KVM_CAP_COALESCED_MMIO:
2085 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2088 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2090 case KVM_CAP_NR_VCPUS:
2091 r = KVM_SOFT_MAX_VCPUS;
2093 case KVM_CAP_MAX_VCPUS:
2096 case KVM_CAP_NR_MEMSLOTS:
2097 r = KVM_MEMORY_SLOTS;
2099 case KVM_CAP_PV_MMU: /* obsolete */
2106 r = KVM_MAX_MCE_BANKS;
2111 case KVM_CAP_TSC_CONTROL:
2112 r = kvm_has_tsc_control;
2122 long kvm_arch_dev_ioctl(struct file *filp,
2123 unsigned int ioctl, unsigned long arg)
2125 void __user *argp = (void __user *)arg;
2129 case KVM_GET_MSR_INDEX_LIST: {
2130 struct kvm_msr_list __user *user_msr_list = argp;
2131 struct kvm_msr_list msr_list;
2135 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2138 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2139 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2142 if (n < msr_list.nmsrs)
2145 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2146 num_msrs_to_save * sizeof(u32)))
2148 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2150 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2155 case KVM_GET_SUPPORTED_CPUID: {
2156 struct kvm_cpuid2 __user *cpuid_arg = argp;
2157 struct kvm_cpuid2 cpuid;
2160 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2162 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2163 cpuid_arg->entries);
2168 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2173 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2176 mce_cap = KVM_MCE_CAP_SUPPORTED;
2178 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2190 static void wbinvd_ipi(void *garbage)
2195 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2197 return vcpu->kvm->arch.iommu_domain &&
2198 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2201 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2203 /* Address WBINVD may be executed by guest */
2204 if (need_emulate_wbinvd(vcpu)) {
2205 if (kvm_x86_ops->has_wbinvd_exit())
2206 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2207 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2208 smp_call_function_single(vcpu->cpu,
2209 wbinvd_ipi, NULL, 1);
2212 kvm_x86_ops->vcpu_load(vcpu, cpu);
2213 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2214 /* Make sure TSC doesn't go backwards */
2218 kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
2219 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2220 tsc - vcpu->arch.last_guest_tsc;
2223 mark_tsc_unstable("KVM discovered backwards TSC");
2224 if (check_tsc_unstable()) {
2225 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2226 vcpu->arch.tsc_catchup = 1;
2228 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2229 if (vcpu->cpu != cpu)
2230 kvm_migrate_timers(vcpu);
2234 accumulate_steal_time(vcpu);
2235 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2238 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2240 kvm_x86_ops->vcpu_put(vcpu);
2241 kvm_put_guest_fpu(vcpu);
2242 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
2245 static int is_efer_nx(void)
2247 unsigned long long efer = 0;
2249 rdmsrl_safe(MSR_EFER, &efer);
2250 return efer & EFER_NX;
2253 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2256 struct kvm_cpuid_entry2 *e, *entry;
2259 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2260 e = &vcpu->arch.cpuid_entries[i];
2261 if (e->function == 0x80000001) {
2266 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2267 entry->edx &= ~(1 << 20);
2268 printk(KERN_INFO "kvm: guest NX capability removed\n");
2272 /* when an old userspace process fills a new kernel module */
2273 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2274 struct kvm_cpuid *cpuid,
2275 struct kvm_cpuid_entry __user *entries)
2278 struct kvm_cpuid_entry *cpuid_entries;
2281 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2284 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2288 if (copy_from_user(cpuid_entries, entries,
2289 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2291 for (i = 0; i < cpuid->nent; i++) {
2292 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2293 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2294 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2295 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2296 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2297 vcpu->arch.cpuid_entries[i].index = 0;
2298 vcpu->arch.cpuid_entries[i].flags = 0;
2299 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2300 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2301 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2303 vcpu->arch.cpuid_nent = cpuid->nent;
2304 cpuid_fix_nx_cap(vcpu);
2306 kvm_apic_set_version(vcpu);
2307 kvm_x86_ops->cpuid_update(vcpu);
2311 vfree(cpuid_entries);
2316 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2317 struct kvm_cpuid2 *cpuid,
2318 struct kvm_cpuid_entry2 __user *entries)
2323 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2326 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2327 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2329 vcpu->arch.cpuid_nent = cpuid->nent;
2330 kvm_apic_set_version(vcpu);
2331 kvm_x86_ops->cpuid_update(vcpu);
2339 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2340 struct kvm_cpuid2 *cpuid,
2341 struct kvm_cpuid_entry2 __user *entries)
2346 if (cpuid->nent < vcpu->arch.cpuid_nent)
2349 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2350 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2355 cpuid->nent = vcpu->arch.cpuid_nent;
2359 static void cpuid_mask(u32 *word, int wordnum)
2361 *word &= boot_cpu_data.x86_capability[wordnum];
2364 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2367 entry->function = function;
2368 entry->index = index;
2369 cpuid_count(entry->function, entry->index,
2370 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2374 static bool supported_xcr0_bit(unsigned bit)
2376 u64 mask = ((u64)1 << bit);
2378 return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
2381 #define F(x) bit(X86_FEATURE_##x)
2383 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2384 u32 index, int *nent, int maxnent)
2386 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2387 #ifdef CONFIG_X86_64
2388 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2390 unsigned f_lm = F(LM);
2392 unsigned f_gbpages = 0;
2395 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2398 const u32 kvm_supported_word0_x86_features =
2399 F(FPU) | F(VME) | F(DE) | F(PSE) |
2400 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2401 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2402 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2403 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2404 0 /* Reserved, DS, ACPI */ | F(MMX) |
2405 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2406 0 /* HTT, TM, Reserved, PBE */;
2407 /* cpuid 0x80000001.edx */
2408 const u32 kvm_supported_word1_x86_features =
2409 F(FPU) | F(VME) | F(DE) | F(PSE) |
2410 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2411 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2412 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2413 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2414 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2415 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2416 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2418 const u32 kvm_supported_word4_x86_features =
2419 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2420 0 /* DS-CPL, VMX, SMX, EST */ |
2421 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2422 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2423 0 /* Reserved, DCA */ | F(XMM4_1) |
2424 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2425 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2426 F(F16C) | F(RDRAND);
2427 /* cpuid 0x80000001.ecx */
2428 const u32 kvm_supported_word6_x86_features =
2429 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2430 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2431 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2432 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2434 /* cpuid 0xC0000001.edx */
2435 const u32 kvm_supported_word5_x86_features =
2436 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2437 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2441 const u32 kvm_supported_word9_x86_features =
2442 F(SMEP) | F(FSGSBASE) | F(ERMS);
2444 /* all calls to cpuid_count() should be made on the same cpu */
2446 do_cpuid_1_ent(entry, function, index);
2451 entry->eax = min(entry->eax, (u32)0xd);
2454 entry->edx &= kvm_supported_word0_x86_features;
2455 cpuid_mask(&entry->edx, 0);
2456 entry->ecx &= kvm_supported_word4_x86_features;
2457 cpuid_mask(&entry->ecx, 4);
2458 /* we support x2apic emulation even if host does not support
2459 * it since we emulate x2apic in software */
2460 entry->ecx |= F(X2APIC);
2462 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2463 * may return different values. This forces us to get_cpu() before
2464 * issuing the first command, and also to emulate this annoying behavior
2465 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2467 int t, times = entry->eax & 0xff;
2469 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2470 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2471 for (t = 1; t < times && *nent < maxnent; ++t) {
2472 do_cpuid_1_ent(&entry[t], function, 0);
2473 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2478 /* function 4 has additional index. */
2482 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2483 /* read more entries until cache_type is zero */
2484 for (i = 1; *nent < maxnent; ++i) {
2485 cache_type = entry[i - 1].eax & 0x1f;
2488 do_cpuid_1_ent(&entry[i], function, i);
2490 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2496 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2497 /* Mask ebx against host capbability word 9 */
2499 entry->ebx &= kvm_supported_word9_x86_features;
2500 cpuid_mask(&entry->ebx, 9);
2510 /* function 0xb has additional index. */
2514 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2515 /* read more entries until level_type is zero */
2516 for (i = 1; *nent < maxnent; ++i) {
2517 level_type = entry[i - 1].ecx & 0xff00;
2520 do_cpuid_1_ent(&entry[i], function, i);
2522 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2530 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2531 for (idx = 1, i = 1; *nent < maxnent && idx < 64; ++idx) {
2532 do_cpuid_1_ent(&entry[i], function, idx);
2533 if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
2536 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2542 case KVM_CPUID_SIGNATURE: {
2543 char signature[12] = "KVMKVMKVM\0\0";
2544 u32 *sigptr = (u32 *)signature;
2546 entry->ebx = sigptr[0];
2547 entry->ecx = sigptr[1];
2548 entry->edx = sigptr[2];
2551 case KVM_CPUID_FEATURES:
2552 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2553 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2554 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2555 (1 << KVM_FEATURE_ASYNC_PF) |
2556 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2558 if (sched_info_on())
2559 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
2566 entry->eax = min(entry->eax, 0x8000001a);
2569 entry->edx &= kvm_supported_word1_x86_features;
2570 cpuid_mask(&entry->edx, 1);
2571 entry->ecx &= kvm_supported_word6_x86_features;
2572 cpuid_mask(&entry->ecx, 6);
2575 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
2576 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
2577 unsigned phys_as = entry->eax & 0xff;
2580 g_phys_as = phys_as;
2581 entry->eax = g_phys_as | (virt_as << 8);
2582 entry->ebx = entry->edx = 0;
2586 entry->ecx = entry->edx = 0;
2592 /*Add support for Centaur's CPUID instruction*/
2594 /*Just support up to 0xC0000004 now*/
2595 entry->eax = min(entry->eax, 0xC0000004);
2598 entry->edx &= kvm_supported_word5_x86_features;
2599 cpuid_mask(&entry->edx, 5);
2601 case 3: /* Processor serial number */
2602 case 5: /* MONITOR/MWAIT */
2603 case 6: /* Thermal management */
2604 case 0xA: /* Architectural Performance Monitoring */
2605 case 0x80000007: /* Advanced power management */
2610 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
2614 kvm_x86_ops->set_supported_cpuid(function, entry);
2621 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2622 struct kvm_cpuid_entry2 __user *entries)
2624 struct kvm_cpuid_entry2 *cpuid_entries;
2625 int limit, nent = 0, r = -E2BIG;
2628 if (cpuid->nent < 1)
2630 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2631 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2633 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2637 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2638 limit = cpuid_entries[0].eax;
2639 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2640 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2641 &nent, cpuid->nent);
2643 if (nent >= cpuid->nent)
2646 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2647 limit = cpuid_entries[nent - 1].eax;
2648 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2649 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2650 &nent, cpuid->nent);
2655 if (nent >= cpuid->nent)
2658 /* Add support for Centaur's CPUID instruction. */
2659 if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2660 do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2661 &nent, cpuid->nent);
2664 if (nent >= cpuid->nent)
2667 limit = cpuid_entries[nent - 1].eax;
2668 for (func = 0xC0000001;
2669 func <= limit && nent < cpuid->nent; ++func)
2670 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2671 &nent, cpuid->nent);
2674 if (nent >= cpuid->nent)
2678 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2682 if (nent >= cpuid->nent)
2685 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2689 if (nent >= cpuid->nent)
2693 if (copy_to_user(entries, cpuid_entries,
2694 nent * sizeof(struct kvm_cpuid_entry2)))
2700 vfree(cpuid_entries);
2705 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2706 struct kvm_lapic_state *s)
2708 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2713 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2714 struct kvm_lapic_state *s)
2716 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2717 kvm_apic_post_state_restore(vcpu);
2718 update_cr8_intercept(vcpu);
2723 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2724 struct kvm_interrupt *irq)
2726 if (irq->irq < 0 || irq->irq >= 256)
2728 if (irqchip_in_kernel(vcpu->kvm))
2731 kvm_queue_interrupt(vcpu, irq->irq, false);
2732 kvm_make_request(KVM_REQ_EVENT, vcpu);
2737 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2739 kvm_inject_nmi(vcpu);
2744 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2745 struct kvm_tpr_access_ctl *tac)
2749 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2753 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2757 unsigned bank_num = mcg_cap & 0xff, bank;
2760 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2762 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2765 vcpu->arch.mcg_cap = mcg_cap;
2766 /* Init IA32_MCG_CTL to all 1s */
2767 if (mcg_cap & MCG_CTL_P)
2768 vcpu->arch.mcg_ctl = ~(u64)0;
2769 /* Init IA32_MCi_CTL to all 1s */
2770 for (bank = 0; bank < bank_num; bank++)
2771 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2776 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2777 struct kvm_x86_mce *mce)
2779 u64 mcg_cap = vcpu->arch.mcg_cap;
2780 unsigned bank_num = mcg_cap & 0xff;
2781 u64 *banks = vcpu->arch.mce_banks;
2783 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2786 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2787 * reporting is disabled
2789 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2790 vcpu->arch.mcg_ctl != ~(u64)0)
2792 banks += 4 * mce->bank;
2794 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2795 * reporting is disabled for the bank
2797 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2799 if (mce->status & MCI_STATUS_UC) {
2800 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2801 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2802 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2805 if (banks[1] & MCI_STATUS_VAL)
2806 mce->status |= MCI_STATUS_OVER;
2807 banks[2] = mce->addr;
2808 banks[3] = mce->misc;
2809 vcpu->arch.mcg_status = mce->mcg_status;
2810 banks[1] = mce->status;
2811 kvm_queue_exception(vcpu, MC_VECTOR);
2812 } else if (!(banks[1] & MCI_STATUS_VAL)
2813 || !(banks[1] & MCI_STATUS_UC)) {
2814 if (banks[1] & MCI_STATUS_VAL)
2815 mce->status |= MCI_STATUS_OVER;
2816 banks[2] = mce->addr;
2817 banks[3] = mce->misc;
2818 banks[1] = mce->status;
2820 banks[1] |= MCI_STATUS_OVER;
2824 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2825 struct kvm_vcpu_events *events)
2827 events->exception.injected =
2828 vcpu->arch.exception.pending &&
2829 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2830 events->exception.nr = vcpu->arch.exception.nr;
2831 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2832 events->exception.pad = 0;
2833 events->exception.error_code = vcpu->arch.exception.error_code;
2835 events->interrupt.injected =
2836 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2837 events->interrupt.nr = vcpu->arch.interrupt.nr;
2838 events->interrupt.soft = 0;
2839 events->interrupt.shadow =
2840 kvm_x86_ops->get_interrupt_shadow(vcpu,
2841 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2843 events->nmi.injected = vcpu->arch.nmi_injected;
2844 events->nmi.pending = vcpu->arch.nmi_pending;
2845 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2846 events->nmi.pad = 0;
2848 events->sipi_vector = vcpu->arch.sipi_vector;
2850 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2851 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2852 | KVM_VCPUEVENT_VALID_SHADOW);
2853 memset(&events->reserved, 0, sizeof(events->reserved));
2856 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2857 struct kvm_vcpu_events *events)
2859 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2860 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2861 | KVM_VCPUEVENT_VALID_SHADOW))
2864 vcpu->arch.exception.pending = events->exception.injected;
2865 vcpu->arch.exception.nr = events->exception.nr;
2866 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2867 vcpu->arch.exception.error_code = events->exception.error_code;
2869 vcpu->arch.interrupt.pending = events->interrupt.injected;
2870 vcpu->arch.interrupt.nr = events->interrupt.nr;
2871 vcpu->arch.interrupt.soft = events->interrupt.soft;
2872 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2873 kvm_x86_ops->set_interrupt_shadow(vcpu,
2874 events->interrupt.shadow);
2876 vcpu->arch.nmi_injected = events->nmi.injected;
2877 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2878 vcpu->arch.nmi_pending = events->nmi.pending;
2879 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2881 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2882 vcpu->arch.sipi_vector = events->sipi_vector;
2884 kvm_make_request(KVM_REQ_EVENT, vcpu);
2889 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2890 struct kvm_debugregs *dbgregs)
2892 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2893 dbgregs->dr6 = vcpu->arch.dr6;
2894 dbgregs->dr7 = vcpu->arch.dr7;
2896 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2899 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2900 struct kvm_debugregs *dbgregs)
2905 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2906 vcpu->arch.dr6 = dbgregs->dr6;
2907 vcpu->arch.dr7 = dbgregs->dr7;
2912 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2913 struct kvm_xsave *guest_xsave)
2916 memcpy(guest_xsave->region,
2917 &vcpu->arch.guest_fpu.state->xsave,
2920 memcpy(guest_xsave->region,
2921 &vcpu->arch.guest_fpu.state->fxsave,
2922 sizeof(struct i387_fxsave_struct));
2923 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2928 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2929 struct kvm_xsave *guest_xsave)
2932 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2935 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2936 guest_xsave->region, xstate_size);
2938 if (xstate_bv & ~XSTATE_FPSSE)
2940 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2941 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2946 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2947 struct kvm_xcrs *guest_xcrs)
2949 if (!cpu_has_xsave) {
2950 guest_xcrs->nr_xcrs = 0;
2954 guest_xcrs->nr_xcrs = 1;
2955 guest_xcrs->flags = 0;
2956 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2957 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2960 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2961 struct kvm_xcrs *guest_xcrs)
2968 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2971 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2972 /* Only support XCR0 currently */
2973 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2974 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2975 guest_xcrs->xcrs[0].value);
2983 long kvm_arch_vcpu_ioctl(struct file *filp,
2984 unsigned int ioctl, unsigned long arg)
2986 struct kvm_vcpu *vcpu = filp->private_data;
2987 void __user *argp = (void __user *)arg;
2990 struct kvm_lapic_state *lapic;
2991 struct kvm_xsave *xsave;
2992 struct kvm_xcrs *xcrs;
2998 case KVM_GET_LAPIC: {
3000 if (!vcpu->arch.apic)
3002 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3007 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3011 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3016 case KVM_SET_LAPIC: {
3018 if (!vcpu->arch.apic)
3020 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3025 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
3027 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3033 case KVM_INTERRUPT: {
3034 struct kvm_interrupt irq;
3037 if (copy_from_user(&irq, argp, sizeof irq))
3039 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3046 r = kvm_vcpu_ioctl_nmi(vcpu);
3052 case KVM_SET_CPUID: {
3053 struct kvm_cpuid __user *cpuid_arg = argp;
3054 struct kvm_cpuid cpuid;
3057 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3059 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3064 case KVM_SET_CPUID2: {
3065 struct kvm_cpuid2 __user *cpuid_arg = argp;
3066 struct kvm_cpuid2 cpuid;
3069 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3071 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3072 cpuid_arg->entries);
3077 case KVM_GET_CPUID2: {
3078 struct kvm_cpuid2 __user *cpuid_arg = argp;
3079 struct kvm_cpuid2 cpuid;
3082 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3084 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3085 cpuid_arg->entries);
3089 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3095 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3098 r = msr_io(vcpu, argp, do_set_msr, 0);
3100 case KVM_TPR_ACCESS_REPORTING: {
3101 struct kvm_tpr_access_ctl tac;
3104 if (copy_from_user(&tac, argp, sizeof tac))
3106 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3110 if (copy_to_user(argp, &tac, sizeof tac))
3115 case KVM_SET_VAPIC_ADDR: {
3116 struct kvm_vapic_addr va;
3119 if (!irqchip_in_kernel(vcpu->kvm))
3122 if (copy_from_user(&va, argp, sizeof va))
3125 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3128 case KVM_X86_SETUP_MCE: {
3132 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3134 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3137 case KVM_X86_SET_MCE: {
3138 struct kvm_x86_mce mce;
3141 if (copy_from_user(&mce, argp, sizeof mce))
3143 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3146 case KVM_GET_VCPU_EVENTS: {
3147 struct kvm_vcpu_events events;
3149 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3152 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3157 case KVM_SET_VCPU_EVENTS: {
3158 struct kvm_vcpu_events events;
3161 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3164 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3167 case KVM_GET_DEBUGREGS: {
3168 struct kvm_debugregs dbgregs;
3170 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3173 if (copy_to_user(argp, &dbgregs,
3174 sizeof(struct kvm_debugregs)))
3179 case KVM_SET_DEBUGREGS: {
3180 struct kvm_debugregs dbgregs;
3183 if (copy_from_user(&dbgregs, argp,
3184 sizeof(struct kvm_debugregs)))
3187 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3190 case KVM_GET_XSAVE: {
3191 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3196 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3199 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3204 case KVM_SET_XSAVE: {
3205 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3211 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3214 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3217 case KVM_GET_XCRS: {
3218 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3223 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3226 if (copy_to_user(argp, u.xcrs,
3227 sizeof(struct kvm_xcrs)))
3232 case KVM_SET_XCRS: {
3233 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3239 if (copy_from_user(u.xcrs, argp,
3240 sizeof(struct kvm_xcrs)))
3243 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3246 case KVM_SET_TSC_KHZ: {
3250 if (!kvm_has_tsc_control)
3253 user_tsc_khz = (u32)arg;
3255 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3258 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3263 case KVM_GET_TSC_KHZ: {
3265 if (check_tsc_unstable())
3268 r = vcpu_tsc_khz(vcpu);
3280 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3284 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3286 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3290 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3293 kvm->arch.ept_identity_map_addr = ident_addr;
3297 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3298 u32 kvm_nr_mmu_pages)
3300 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3303 mutex_lock(&kvm->slots_lock);
3304 spin_lock(&kvm->mmu_lock);
3306 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3307 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3309 spin_unlock(&kvm->mmu_lock);
3310 mutex_unlock(&kvm->slots_lock);
3314 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3316 return kvm->arch.n_max_mmu_pages;
3319 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3324 switch (chip->chip_id) {
3325 case KVM_IRQCHIP_PIC_MASTER:
3326 memcpy(&chip->chip.pic,
3327 &pic_irqchip(kvm)->pics[0],
3328 sizeof(struct kvm_pic_state));
3330 case KVM_IRQCHIP_PIC_SLAVE:
3331 memcpy(&chip->chip.pic,
3332 &pic_irqchip(kvm)->pics[1],
3333 sizeof(struct kvm_pic_state));
3335 case KVM_IRQCHIP_IOAPIC:
3336 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3345 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3350 switch (chip->chip_id) {
3351 case KVM_IRQCHIP_PIC_MASTER:
3352 spin_lock(&pic_irqchip(kvm)->lock);
3353 memcpy(&pic_irqchip(kvm)->pics[0],
3355 sizeof(struct kvm_pic_state));
3356 spin_unlock(&pic_irqchip(kvm)->lock);
3358 case KVM_IRQCHIP_PIC_SLAVE:
3359 spin_lock(&pic_irqchip(kvm)->lock);
3360 memcpy(&pic_irqchip(kvm)->pics[1],
3362 sizeof(struct kvm_pic_state));
3363 spin_unlock(&pic_irqchip(kvm)->lock);
3365 case KVM_IRQCHIP_IOAPIC:
3366 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3372 kvm_pic_update_irq(pic_irqchip(kvm));
3376 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3380 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3381 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3382 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3386 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3390 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3391 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3392 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3393 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3397 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3401 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3402 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3403 sizeof(ps->channels));
3404 ps->flags = kvm->arch.vpit->pit_state.flags;
3405 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3406 memset(&ps->reserved, 0, sizeof(ps->reserved));
3410 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3412 int r = 0, start = 0;
3413 u32 prev_legacy, cur_legacy;
3414 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3415 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3416 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3417 if (!prev_legacy && cur_legacy)
3419 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3420 sizeof(kvm->arch.vpit->pit_state.channels));
3421 kvm->arch.vpit->pit_state.flags = ps->flags;
3422 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3423 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3427 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3428 struct kvm_reinject_control *control)
3430 if (!kvm->arch.vpit)
3432 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3433 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3434 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3439 * Get (and clear) the dirty memory log for a memory slot.
3441 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3442 struct kvm_dirty_log *log)
3445 struct kvm_memory_slot *memslot;
3447 unsigned long is_dirty = 0;
3449 mutex_lock(&kvm->slots_lock);
3452 if (log->slot >= KVM_MEMORY_SLOTS)
3455 memslot = &kvm->memslots->memslots[log->slot];
3457 if (!memslot->dirty_bitmap)
3460 n = kvm_dirty_bitmap_bytes(memslot);
3462 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3463 is_dirty = memslot->dirty_bitmap[i];
3465 /* If nothing is dirty, don't bother messing with page tables. */
3467 struct kvm_memslots *slots, *old_slots;
3468 unsigned long *dirty_bitmap;
3470 dirty_bitmap = memslot->dirty_bitmap_head;
3471 if (memslot->dirty_bitmap == dirty_bitmap)
3472 dirty_bitmap += n / sizeof(long);
3473 memset(dirty_bitmap, 0, n);
3476 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3479 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3480 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3481 slots->generation++;
3483 old_slots = kvm->memslots;
3484 rcu_assign_pointer(kvm->memslots, slots);
3485 synchronize_srcu_expedited(&kvm->srcu);
3486 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3489 spin_lock(&kvm->mmu_lock);
3490 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3491 spin_unlock(&kvm->mmu_lock);
3494 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3498 if (clear_user(log->dirty_bitmap, n))
3504 mutex_unlock(&kvm->slots_lock);
3508 long kvm_arch_vm_ioctl(struct file *filp,
3509 unsigned int ioctl, unsigned long arg)
3511 struct kvm *kvm = filp->private_data;
3512 void __user *argp = (void __user *)arg;
3515 * This union makes it completely explicit to gcc-3.x
3516 * that these two variables' stack usage should be
3517 * combined, not added together.
3520 struct kvm_pit_state ps;
3521 struct kvm_pit_state2 ps2;
3522 struct kvm_pit_config pit_config;
3526 case KVM_SET_TSS_ADDR:
3527 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3531 case KVM_SET_IDENTITY_MAP_ADDR: {
3535 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3537 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3542 case KVM_SET_NR_MMU_PAGES:
3543 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3547 case KVM_GET_NR_MMU_PAGES:
3548 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3550 case KVM_CREATE_IRQCHIP: {
3551 struct kvm_pic *vpic;
3553 mutex_lock(&kvm->lock);
3556 goto create_irqchip_unlock;
3558 vpic = kvm_create_pic(kvm);
3560 r = kvm_ioapic_init(kvm);
3562 mutex_lock(&kvm->slots_lock);
3563 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3565 mutex_unlock(&kvm->slots_lock);
3567 goto create_irqchip_unlock;
3570 goto create_irqchip_unlock;
3572 kvm->arch.vpic = vpic;
3574 r = kvm_setup_default_irq_routing(kvm);
3576 mutex_lock(&kvm->slots_lock);
3577 mutex_lock(&kvm->irq_lock);
3578 kvm_ioapic_destroy(kvm);
3579 kvm_destroy_pic(kvm);
3580 mutex_unlock(&kvm->irq_lock);
3581 mutex_unlock(&kvm->slots_lock);
3583 create_irqchip_unlock:
3584 mutex_unlock(&kvm->lock);
3587 case KVM_CREATE_PIT:
3588 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3590 case KVM_CREATE_PIT2:
3592 if (copy_from_user(&u.pit_config, argp,
3593 sizeof(struct kvm_pit_config)))
3596 mutex_lock(&kvm->slots_lock);
3599 goto create_pit_unlock;
3601 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3605 mutex_unlock(&kvm->slots_lock);
3607 case KVM_IRQ_LINE_STATUS:
3608 case KVM_IRQ_LINE: {
3609 struct kvm_irq_level irq_event;
3612 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3615 if (irqchip_in_kernel(kvm)) {
3617 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3618 irq_event.irq, irq_event.level);
3619 if (ioctl == KVM_IRQ_LINE_STATUS) {
3621 irq_event.status = status;
3622 if (copy_to_user(argp, &irq_event,
3630 case KVM_GET_IRQCHIP: {
3631 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3632 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3638 if (copy_from_user(chip, argp, sizeof *chip))
3639 goto get_irqchip_out;
3641 if (!irqchip_in_kernel(kvm))
3642 goto get_irqchip_out;
3643 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3645 goto get_irqchip_out;
3647 if (copy_to_user(argp, chip, sizeof *chip))
3648 goto get_irqchip_out;
3656 case KVM_SET_IRQCHIP: {
3657 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3658 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3664 if (copy_from_user(chip, argp, sizeof *chip))
3665 goto set_irqchip_out;
3667 if (!irqchip_in_kernel(kvm))
3668 goto set_irqchip_out;
3669 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3671 goto set_irqchip_out;
3681 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3684 if (!kvm->arch.vpit)
3686 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3690 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3697 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3700 if (!kvm->arch.vpit)
3702 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3708 case KVM_GET_PIT2: {
3710 if (!kvm->arch.vpit)
3712 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3716 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3721 case KVM_SET_PIT2: {
3723 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3726 if (!kvm->arch.vpit)
3728 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3734 case KVM_REINJECT_CONTROL: {
3735 struct kvm_reinject_control control;
3737 if (copy_from_user(&control, argp, sizeof(control)))
3739 r = kvm_vm_ioctl_reinject(kvm, &control);
3745 case KVM_XEN_HVM_CONFIG: {
3747 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3748 sizeof(struct kvm_xen_hvm_config)))
3751 if (kvm->arch.xen_hvm_config.flags)
3756 case KVM_SET_CLOCK: {
3757 struct kvm_clock_data user_ns;
3762 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3770 local_irq_disable();
3771 now_ns = get_kernel_ns();
3772 delta = user_ns.clock - now_ns;
3774 kvm->arch.kvmclock_offset = delta;
3777 case KVM_GET_CLOCK: {
3778 struct kvm_clock_data user_ns;
3781 local_irq_disable();
3782 now_ns = get_kernel_ns();
3783 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3786 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3789 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3802 static void kvm_init_msr_list(void)
3807 /* skip the first msrs in the list. KVM-specific */
3808 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3809 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3812 msrs_to_save[j] = msrs_to_save[i];
3815 num_msrs_to_save = j;
3818 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3826 if (!(vcpu->arch.apic &&
3827 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3828 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3839 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3846 if (!(vcpu->arch.apic &&
3847 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3848 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3850 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3860 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3861 struct kvm_segment *var, int seg)
3863 kvm_x86_ops->set_segment(vcpu, var, seg);
3866 void kvm_get_segment(struct kvm_vcpu *vcpu,
3867 struct kvm_segment *var, int seg)
3869 kvm_x86_ops->get_segment(vcpu, var, seg);
3872 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3877 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3880 struct x86_exception exception;
3882 BUG_ON(!mmu_is_nested(vcpu));
3884 /* NPT walks are always user-walks */
3885 access |= PFERR_USER_MASK;
3886 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3891 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3892 struct x86_exception *exception)
3894 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3895 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3898 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3899 struct x86_exception *exception)
3901 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3902 access |= PFERR_FETCH_MASK;
3903 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3906 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3907 struct x86_exception *exception)
3909 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3910 access |= PFERR_WRITE_MASK;
3911 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3914 /* uses this to access any guest's mapped memory without checking CPL */
3915 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3916 struct x86_exception *exception)
3918 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3921 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3922 struct kvm_vcpu *vcpu, u32 access,
3923 struct x86_exception *exception)
3926 int r = X86EMUL_CONTINUE;
3929 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3931 unsigned offset = addr & (PAGE_SIZE-1);
3932 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3935 if (gpa == UNMAPPED_GVA)
3936 return X86EMUL_PROPAGATE_FAULT;
3937 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3939 r = X86EMUL_IO_NEEDED;
3951 /* used for instruction fetching */
3952 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3953 gva_t addr, void *val, unsigned int bytes,
3954 struct x86_exception *exception)
3956 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3957 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3959 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3960 access | PFERR_FETCH_MASK,
3964 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3965 gva_t addr, void *val, unsigned int bytes,
3966 struct x86_exception *exception)
3968 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3969 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3971 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3974 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3976 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3977 gva_t addr, void *val, unsigned int bytes,
3978 struct x86_exception *exception)
3980 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3981 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3984 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3985 gva_t addr, void *val,
3987 struct x86_exception *exception)
3989 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3991 int r = X86EMUL_CONTINUE;
3994 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3997 unsigned offset = addr & (PAGE_SIZE-1);
3998 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4001 if (gpa == UNMAPPED_GVA)
4002 return X86EMUL_PROPAGATE_FAULT;
4003 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4005 r = X86EMUL_IO_NEEDED;
4016 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4018 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4019 gpa_t *gpa, struct x86_exception *exception,
4022 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4024 if (vcpu_match_mmio_gva(vcpu, gva) &&
4025 check_write_user_access(vcpu, write, access,
4026 vcpu->arch.access)) {
4027 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4028 (gva & (PAGE_SIZE - 1));
4029 trace_vcpu_match_mmio(gva, *gpa, write, false);
4034 access |= PFERR_WRITE_MASK;
4036 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4038 if (*gpa == UNMAPPED_GVA)
4041 /* For APIC access vmexit */
4042 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4045 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4046 trace_vcpu_match_mmio(gva, *gpa, write, true);
4053 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4054 const void *val, int bytes)
4058 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4061 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
4065 struct read_write_emulator_ops {
4066 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4068 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4069 void *val, int bytes);
4070 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4071 int bytes, void *val);
4072 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4073 void *val, int bytes);
4077 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4079 if (vcpu->mmio_read_completed) {
4080 memcpy(val, vcpu->mmio_data, bytes);
4081 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4082 vcpu->mmio_phys_addr, *(u64 *)val);
4083 vcpu->mmio_read_completed = 0;
4090 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4091 void *val, int bytes)
4093 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4096 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4097 void *val, int bytes)
4099 return emulator_write_phys(vcpu, gpa, val, bytes);
4102 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4104 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4105 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4108 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4109 void *val, int bytes)
4111 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4112 return X86EMUL_IO_NEEDED;
4115 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4116 void *val, int bytes)
4118 memcpy(vcpu->mmio_data, val, bytes);
4119 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
4120 return X86EMUL_CONTINUE;
4123 static struct read_write_emulator_ops read_emultor = {
4124 .read_write_prepare = read_prepare,
4125 .read_write_emulate = read_emulate,
4126 .read_write_mmio = vcpu_mmio_read,
4127 .read_write_exit_mmio = read_exit_mmio,
4130 static struct read_write_emulator_ops write_emultor = {
4131 .read_write_emulate = write_emulate,
4132 .read_write_mmio = write_mmio,
4133 .read_write_exit_mmio = write_exit_mmio,
4137 static int emulator_read_write_onepage(unsigned long addr, void *val,
4139 struct x86_exception *exception,
4140 struct kvm_vcpu *vcpu,
4141 struct read_write_emulator_ops *ops)
4145 bool write = ops->write;
4147 if (ops->read_write_prepare &&
4148 ops->read_write_prepare(vcpu, val, bytes))
4149 return X86EMUL_CONTINUE;
4151 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4154 return X86EMUL_PROPAGATE_FAULT;
4156 /* For APIC access vmexit */
4160 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4161 return X86EMUL_CONTINUE;
4165 * Is this MMIO handled locally?
4167 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4168 if (handled == bytes)
4169 return X86EMUL_CONTINUE;
4175 vcpu->mmio_needed = 1;
4176 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4177 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
4178 vcpu->mmio_size = bytes;
4179 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
4180 vcpu->run->mmio.is_write = vcpu->mmio_is_write = write;
4181 vcpu->mmio_index = 0;
4183 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4186 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4187 void *val, unsigned int bytes,
4188 struct x86_exception *exception,
4189 struct read_write_emulator_ops *ops)
4191 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4193 /* Crossing a page boundary? */
4194 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4197 now = -addr & ~PAGE_MASK;
4198 rc = emulator_read_write_onepage(addr, val, now, exception,
4201 if (rc != X86EMUL_CONTINUE)
4208 return emulator_read_write_onepage(addr, val, bytes, exception,
4212 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4216 struct x86_exception *exception)
4218 return emulator_read_write(ctxt, addr, val, bytes,
4219 exception, &read_emultor);
4222 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4226 struct x86_exception *exception)
4228 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4229 exception, &write_emultor);
4232 #define CMPXCHG_TYPE(t, ptr, old, new) \
4233 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4235 #ifdef CONFIG_X86_64
4236 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4238 # define CMPXCHG64(ptr, old, new) \
4239 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4242 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4247 struct x86_exception *exception)
4249 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4255 /* guests cmpxchg8b have to be emulated atomically */
4256 if (bytes > 8 || (bytes & (bytes - 1)))
4259 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4261 if (gpa == UNMAPPED_GVA ||
4262 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4265 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4268 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4269 if (is_error_page(page)) {
4270 kvm_release_page_clean(page);
4274 kaddr = kmap_atomic(page, KM_USER0);
4275 kaddr += offset_in_page(gpa);
4278 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4281 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4284 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4287 exchanged = CMPXCHG64(kaddr, old, new);
4292 kunmap_atomic(kaddr, KM_USER0);
4293 kvm_release_page_dirty(page);
4296 return X86EMUL_CMPXCHG_FAILED;
4298 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4300 return X86EMUL_CONTINUE;
4303 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4305 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4308 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4310 /* TODO: String I/O for in kernel device */
4313 if (vcpu->arch.pio.in)
4314 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4315 vcpu->arch.pio.size, pd);
4317 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4318 vcpu->arch.pio.port, vcpu->arch.pio.size,
4324 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4325 int size, unsigned short port, void *val,
4328 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4330 if (vcpu->arch.pio.count)
4333 trace_kvm_pio(0, port, size, count);
4335 vcpu->arch.pio.port = port;
4336 vcpu->arch.pio.in = 1;
4337 vcpu->arch.pio.count = count;
4338 vcpu->arch.pio.size = size;
4340 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4342 memcpy(val, vcpu->arch.pio_data, size * count);
4343 vcpu->arch.pio.count = 0;
4347 vcpu->run->exit_reason = KVM_EXIT_IO;
4348 vcpu->run->io.direction = KVM_EXIT_IO_IN;
4349 vcpu->run->io.size = size;
4350 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4351 vcpu->run->io.count = count;
4352 vcpu->run->io.port = port;
4357 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4358 int size, unsigned short port,
4359 const void *val, unsigned int count)
4361 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4363 trace_kvm_pio(1, port, size, count);
4365 vcpu->arch.pio.port = port;
4366 vcpu->arch.pio.in = 0;
4367 vcpu->arch.pio.count = count;
4368 vcpu->arch.pio.size = size;
4370 memcpy(vcpu->arch.pio_data, val, size * count);
4372 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4373 vcpu->arch.pio.count = 0;
4377 vcpu->run->exit_reason = KVM_EXIT_IO;
4378 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4379 vcpu->run->io.size = size;
4380 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4381 vcpu->run->io.count = count;
4382 vcpu->run->io.port = port;
4387 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4389 return kvm_x86_ops->get_segment_base(vcpu, seg);
4392 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4394 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4397 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4399 if (!need_emulate_wbinvd(vcpu))
4400 return X86EMUL_CONTINUE;
4402 if (kvm_x86_ops->has_wbinvd_exit()) {
4403 int cpu = get_cpu();
4405 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4406 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4407 wbinvd_ipi, NULL, 1);
4409 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4412 return X86EMUL_CONTINUE;
4414 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4416 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4418 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4421 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4423 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4426 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4429 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4432 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4434 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4437 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4439 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4440 unsigned long value;
4444 value = kvm_read_cr0(vcpu);
4447 value = vcpu->arch.cr2;
4450 value = kvm_read_cr3(vcpu);
4453 value = kvm_read_cr4(vcpu);
4456 value = kvm_get_cr8(vcpu);
4459 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4466 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4468 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4473 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4476 vcpu->arch.cr2 = val;
4479 res = kvm_set_cr3(vcpu, val);
4482 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4485 res = kvm_set_cr8(vcpu, val);
4488 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4495 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4497 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4500 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4502 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4505 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4507 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4510 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4512 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4515 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4517 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4520 static unsigned long emulator_get_cached_segment_base(
4521 struct x86_emulate_ctxt *ctxt, int seg)
4523 return get_segment_base(emul_to_vcpu(ctxt), seg);
4526 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4527 struct desc_struct *desc, u32 *base3,
4530 struct kvm_segment var;
4532 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4533 *selector = var.selector;
4540 set_desc_limit(desc, var.limit);
4541 set_desc_base(desc, (unsigned long)var.base);
4542 #ifdef CONFIG_X86_64
4544 *base3 = var.base >> 32;
4546 desc->type = var.type;
4548 desc->dpl = var.dpl;
4549 desc->p = var.present;
4550 desc->avl = var.avl;
4558 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4559 struct desc_struct *desc, u32 base3,
4562 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4563 struct kvm_segment var;
4565 var.selector = selector;
4566 var.base = get_desc_base(desc);
4567 #ifdef CONFIG_X86_64
4568 var.base |= ((u64)base3) << 32;
4570 var.limit = get_desc_limit(desc);
4572 var.limit = (var.limit << 12) | 0xfff;
4573 var.type = desc->type;
4574 var.present = desc->p;
4575 var.dpl = desc->dpl;
4580 var.avl = desc->avl;
4581 var.present = desc->p;
4582 var.unusable = !var.present;
4585 kvm_set_segment(vcpu, &var, seg);
4589 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4590 u32 msr_index, u64 *pdata)
4592 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4595 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4596 u32 msr_index, u64 data)
4598 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4601 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4603 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4606 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4609 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4611 * CR0.TS may reference the host fpu state, not the guest fpu state,
4612 * so it may be clear at this point.
4617 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4622 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4623 struct x86_instruction_info *info,
4624 enum x86_intercept_stage stage)
4626 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4629 static struct x86_emulate_ops emulate_ops = {
4630 .read_std = kvm_read_guest_virt_system,
4631 .write_std = kvm_write_guest_virt_system,
4632 .fetch = kvm_fetch_guest_virt,
4633 .read_emulated = emulator_read_emulated,
4634 .write_emulated = emulator_write_emulated,
4635 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4636 .invlpg = emulator_invlpg,
4637 .pio_in_emulated = emulator_pio_in_emulated,
4638 .pio_out_emulated = emulator_pio_out_emulated,
4639 .get_segment = emulator_get_segment,
4640 .set_segment = emulator_set_segment,
4641 .get_cached_segment_base = emulator_get_cached_segment_base,
4642 .get_gdt = emulator_get_gdt,
4643 .get_idt = emulator_get_idt,
4644 .set_gdt = emulator_set_gdt,
4645 .set_idt = emulator_set_idt,
4646 .get_cr = emulator_get_cr,
4647 .set_cr = emulator_set_cr,
4648 .cpl = emulator_get_cpl,
4649 .get_dr = emulator_get_dr,
4650 .set_dr = emulator_set_dr,
4651 .set_msr = emulator_set_msr,
4652 .get_msr = emulator_get_msr,
4653 .halt = emulator_halt,
4654 .wbinvd = emulator_wbinvd,
4655 .fix_hypercall = emulator_fix_hypercall,
4656 .get_fpu = emulator_get_fpu,
4657 .put_fpu = emulator_put_fpu,
4658 .intercept = emulator_intercept,
4661 static void cache_all_regs(struct kvm_vcpu *vcpu)
4663 kvm_register_read(vcpu, VCPU_REGS_RAX);
4664 kvm_register_read(vcpu, VCPU_REGS_RSP);
4665 kvm_register_read(vcpu, VCPU_REGS_RIP);
4666 vcpu->arch.regs_dirty = ~0;
4669 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4671 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4673 * an sti; sti; sequence only disable interrupts for the first
4674 * instruction. So, if the last instruction, be it emulated or
4675 * not, left the system with the INT_STI flag enabled, it
4676 * means that the last instruction is an sti. We should not
4677 * leave the flag on in this case. The same goes for mov ss
4679 if (!(int_shadow & mask))
4680 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4683 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4685 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4686 if (ctxt->exception.vector == PF_VECTOR)
4687 kvm_propagate_fault(vcpu, &ctxt->exception);
4688 else if (ctxt->exception.error_code_valid)
4689 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4690 ctxt->exception.error_code);
4692 kvm_queue_exception(vcpu, ctxt->exception.vector);
4695 static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
4696 const unsigned long *regs)
4698 memset(&ctxt->twobyte, 0,
4699 (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4700 memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
4702 ctxt->fetch.start = 0;
4703 ctxt->fetch.end = 0;
4704 ctxt->io_read.pos = 0;
4705 ctxt->io_read.end = 0;
4706 ctxt->mem_read.pos = 0;
4707 ctxt->mem_read.end = 0;
4710 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4712 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4716 * TODO: fix emulate.c to use guest_read/write_register
4717 * instead of direct ->regs accesses, can save hundred cycles
4718 * on Intel for instructions that don't read/change RSP, for
4721 cache_all_regs(vcpu);
4723 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4725 ctxt->eflags = kvm_get_rflags(vcpu);
4726 ctxt->eip = kvm_rip_read(vcpu);
4727 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4728 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4729 cs_l ? X86EMUL_MODE_PROT64 :
4730 cs_db ? X86EMUL_MODE_PROT32 :
4731 X86EMUL_MODE_PROT16;
4732 ctxt->guest_mode = is_guest_mode(vcpu);
4734 init_decode_cache(ctxt, vcpu->arch.regs);
4735 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4738 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4740 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4743 init_emulate_ctxt(vcpu);
4747 ctxt->_eip = ctxt->eip + inc_eip;
4748 ret = emulate_int_real(ctxt, irq);
4750 if (ret != X86EMUL_CONTINUE)
4751 return EMULATE_FAIL;
4753 ctxt->eip = ctxt->_eip;
4754 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4755 kvm_rip_write(vcpu, ctxt->eip);
4756 kvm_set_rflags(vcpu, ctxt->eflags);
4758 if (irq == NMI_VECTOR)
4759 vcpu->arch.nmi_pending = false;
4761 vcpu->arch.interrupt.pending = false;
4763 return EMULATE_DONE;
4765 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4767 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4769 int r = EMULATE_DONE;
4771 ++vcpu->stat.insn_emulation_fail;
4772 trace_kvm_emulate_insn_failed(vcpu);
4773 if (!is_guest_mode(vcpu)) {
4774 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4775 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4776 vcpu->run->internal.ndata = 0;
4779 kvm_queue_exception(vcpu, UD_VECTOR);
4784 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4792 * if emulation was due to access to shadowed page table
4793 * and it failed try to unshadow page and re-entetr the
4794 * guest to let CPU execute the instruction.
4796 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4799 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4801 if (gpa == UNMAPPED_GVA)
4802 return true; /* let cpu generate fault */
4804 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4810 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4817 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4818 bool writeback = true;
4820 kvm_clear_exception_queue(vcpu);
4822 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4823 init_emulate_ctxt(vcpu);
4824 ctxt->interruptibility = 0;
4825 ctxt->have_exception = false;
4826 ctxt->perm_ok = false;
4828 ctxt->only_vendor_specific_insn
4829 = emulation_type & EMULTYPE_TRAP_UD;
4831 r = x86_decode_insn(ctxt, insn, insn_len);
4833 trace_kvm_emulate_insn_start(vcpu);
4834 ++vcpu->stat.insn_emulation;
4836 if (emulation_type & EMULTYPE_TRAP_UD)
4837 return EMULATE_FAIL;
4838 if (reexecute_instruction(vcpu, cr2))
4839 return EMULATE_DONE;
4840 if (emulation_type & EMULTYPE_SKIP)
4841 return EMULATE_FAIL;
4842 return handle_emulation_failure(vcpu);
4846 if (emulation_type & EMULTYPE_SKIP) {
4847 kvm_rip_write(vcpu, ctxt->_eip);
4848 return EMULATE_DONE;
4851 /* this is needed for vmware backdoor interface to work since it
4852 changes registers values during IO operation */
4853 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4854 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4855 memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
4859 r = x86_emulate_insn(ctxt);
4861 if (r == EMULATION_INTERCEPTED)
4862 return EMULATE_DONE;
4864 if (r == EMULATION_FAILED) {
4865 if (reexecute_instruction(vcpu, cr2))
4866 return EMULATE_DONE;
4868 return handle_emulation_failure(vcpu);
4871 if (ctxt->have_exception) {
4872 inject_emulated_exception(vcpu);
4874 } else if (vcpu->arch.pio.count) {
4875 if (!vcpu->arch.pio.in)
4876 vcpu->arch.pio.count = 0;
4879 r = EMULATE_DO_MMIO;
4880 } else if (vcpu->mmio_needed) {
4881 if (!vcpu->mmio_is_write)
4883 r = EMULATE_DO_MMIO;
4884 } else if (r == EMULATION_RESTART)
4890 toggle_interruptibility(vcpu, ctxt->interruptibility);
4891 kvm_set_rflags(vcpu, ctxt->eflags);
4892 kvm_make_request(KVM_REQ_EVENT, vcpu);
4893 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4894 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4895 kvm_rip_write(vcpu, ctxt->eip);
4897 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4901 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4903 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4905 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4906 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4907 size, port, &val, 1);
4908 /* do not return to emulator after return from userspace */
4909 vcpu->arch.pio.count = 0;
4912 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4914 static void tsc_bad(void *info)
4916 __this_cpu_write(cpu_tsc_khz, 0);
4919 static void tsc_khz_changed(void *data)
4921 struct cpufreq_freqs *freq = data;
4922 unsigned long khz = 0;
4926 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4927 khz = cpufreq_quick_get(raw_smp_processor_id());
4930 __this_cpu_write(cpu_tsc_khz, khz);
4933 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4936 struct cpufreq_freqs *freq = data;
4938 struct kvm_vcpu *vcpu;
4939 int i, send_ipi = 0;
4942 * We allow guests to temporarily run on slowing clocks,
4943 * provided we notify them after, or to run on accelerating
4944 * clocks, provided we notify them before. Thus time never
4947 * However, we have a problem. We can't atomically update
4948 * the frequency of a given CPU from this function; it is
4949 * merely a notifier, which can be called from any CPU.
4950 * Changing the TSC frequency at arbitrary points in time
4951 * requires a recomputation of local variables related to
4952 * the TSC for each VCPU. We must flag these local variables
4953 * to be updated and be sure the update takes place with the
4954 * new frequency before any guests proceed.
4956 * Unfortunately, the combination of hotplug CPU and frequency
4957 * change creates an intractable locking scenario; the order
4958 * of when these callouts happen is undefined with respect to
4959 * CPU hotplug, and they can race with each other. As such,
4960 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4961 * undefined; you can actually have a CPU frequency change take
4962 * place in between the computation of X and the setting of the
4963 * variable. To protect against this problem, all updates of
4964 * the per_cpu tsc_khz variable are done in an interrupt
4965 * protected IPI, and all callers wishing to update the value
4966 * must wait for a synchronous IPI to complete (which is trivial
4967 * if the caller is on the CPU already). This establishes the
4968 * necessary total order on variable updates.
4970 * Note that because a guest time update may take place
4971 * anytime after the setting of the VCPU's request bit, the
4972 * correct TSC value must be set before the request. However,
4973 * to ensure the update actually makes it to any guest which
4974 * starts running in hardware virtualization between the set
4975 * and the acquisition of the spinlock, we must also ping the
4976 * CPU after setting the request bit.
4980 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4982 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4985 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4987 raw_spin_lock(&kvm_lock);
4988 list_for_each_entry(kvm, &vm_list, vm_list) {
4989 kvm_for_each_vcpu(i, vcpu, kvm) {
4990 if (vcpu->cpu != freq->cpu)
4992 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4993 if (vcpu->cpu != smp_processor_id())
4997 raw_spin_unlock(&kvm_lock);
4999 if (freq->old < freq->new && send_ipi) {
5001 * We upscale the frequency. Must make the guest
5002 * doesn't see old kvmclock values while running with
5003 * the new frequency, otherwise we risk the guest sees
5004 * time go backwards.
5006 * In case we update the frequency for another cpu
5007 * (which might be in guest context) send an interrupt
5008 * to kick the cpu out of guest context. Next time
5009 * guest context is entered kvmclock will be updated,
5010 * so the guest will not see stale values.
5012 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5017 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5018 .notifier_call = kvmclock_cpufreq_notifier
5021 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5022 unsigned long action, void *hcpu)
5024 unsigned int cpu = (unsigned long)hcpu;
5028 case CPU_DOWN_FAILED:
5029 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5031 case CPU_DOWN_PREPARE:
5032 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5038 static struct notifier_block kvmclock_cpu_notifier_block = {
5039 .notifier_call = kvmclock_cpu_notifier,
5040 .priority = -INT_MAX
5043 static void kvm_timer_init(void)
5047 max_tsc_khz = tsc_khz;
5048 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5049 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5050 #ifdef CONFIG_CPU_FREQ
5051 struct cpufreq_policy policy;
5052 memset(&policy, 0, sizeof(policy));
5054 cpufreq_get_policy(&policy, cpu);
5055 if (policy.cpuinfo.max_freq)
5056 max_tsc_khz = policy.cpuinfo.max_freq;
5059 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5060 CPUFREQ_TRANSITION_NOTIFIER);
5062 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5063 for_each_online_cpu(cpu)
5064 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5067 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5069 static int kvm_is_in_guest(void)
5071 return percpu_read(current_vcpu) != NULL;
5074 static int kvm_is_user_mode(void)
5078 if (percpu_read(current_vcpu))
5079 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
5081 return user_mode != 0;
5084 static unsigned long kvm_get_guest_ip(void)
5086 unsigned long ip = 0;
5088 if (percpu_read(current_vcpu))
5089 ip = kvm_rip_read(percpu_read(current_vcpu));
5094 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5095 .is_in_guest = kvm_is_in_guest,
5096 .is_user_mode = kvm_is_user_mode,
5097 .get_guest_ip = kvm_get_guest_ip,
5100 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5102 percpu_write(current_vcpu, vcpu);
5104 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5106 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5108 percpu_write(current_vcpu, NULL);
5110 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5112 static void kvm_set_mmio_spte_mask(void)
5115 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5118 * Set the reserved bits and the present bit of an paging-structure
5119 * entry to generate page fault with PFER.RSV = 1.
5121 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5124 #ifdef CONFIG_X86_64
5126 * If reserved bit is not supported, clear the present bit to disable
5129 if (maxphyaddr == 52)
5133 kvm_mmu_set_mmio_spte_mask(mask);
5136 int kvm_arch_init(void *opaque)
5139 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5142 printk(KERN_ERR "kvm: already loaded the other module\n");
5147 if (!ops->cpu_has_kvm_support()) {
5148 printk(KERN_ERR "kvm: no hardware support\n");
5152 if (ops->disabled_by_bios()) {
5153 printk(KERN_ERR "kvm: disabled by bios\n");
5158 r = kvm_mmu_module_init();
5162 kvm_set_mmio_spte_mask();
5163 kvm_init_msr_list();
5166 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5167 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5171 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5174 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5182 void kvm_arch_exit(void)
5184 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5186 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5187 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5188 CPUFREQ_TRANSITION_NOTIFIER);
5189 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5191 kvm_mmu_module_exit();
5194 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5196 ++vcpu->stat.halt_exits;
5197 if (irqchip_in_kernel(vcpu->kvm)) {
5198 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5201 vcpu->run->exit_reason = KVM_EXIT_HLT;
5205 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5207 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
5210 if (is_long_mode(vcpu))
5213 return a0 | ((gpa_t)a1 << 32);
5216 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5218 u64 param, ingpa, outgpa, ret;
5219 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5220 bool fast, longmode;
5224 * hypercall generates UD from non zero cpl and real mode
5227 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5228 kvm_queue_exception(vcpu, UD_VECTOR);
5232 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5233 longmode = is_long_mode(vcpu) && cs_l == 1;
5236 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5237 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5238 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5239 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5240 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5241 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5243 #ifdef CONFIG_X86_64
5245 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5246 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5247 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5251 code = param & 0xffff;
5252 fast = (param >> 16) & 0x1;
5253 rep_cnt = (param >> 32) & 0xfff;
5254 rep_idx = (param >> 48) & 0xfff;
5256 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5259 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5260 kvm_vcpu_on_spin(vcpu);
5263 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5267 ret = res | (((u64)rep_done & 0xfff) << 32);
5269 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5271 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5272 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5278 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5280 unsigned long nr, a0, a1, a2, a3, ret;
5283 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5284 return kvm_hv_hypercall(vcpu);
5286 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5287 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5288 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5289 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5290 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5292 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5294 if (!is_long_mode(vcpu)) {
5302 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5308 case KVM_HC_VAPIC_POLL_IRQ:
5312 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5319 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5320 ++vcpu->stat.hypercalls;
5323 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5325 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5327 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5328 char instruction[3];
5329 unsigned long rip = kvm_rip_read(vcpu);
5332 * Blow out the MMU to ensure that no other VCPU has an active mapping
5333 * to ensure that the updated hypercall appears atomically across all
5336 kvm_mmu_zap_all(vcpu->kvm);
5338 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5340 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5343 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5345 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5346 int j, nent = vcpu->arch.cpuid_nent;
5348 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5349 /* when no next entry is found, the current entry[i] is reselected */
5350 for (j = i + 1; ; j = (j + 1) % nent) {
5351 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5352 if (ej->function == e->function) {
5353 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5357 return 0; /* silence gcc, even though control never reaches here */
5360 /* find an entry with matching function, matching index (if needed), and that
5361 * should be read next (if it's stateful) */
5362 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5363 u32 function, u32 index)
5365 if (e->function != function)
5367 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5369 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5370 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5375 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5376 u32 function, u32 index)
5379 struct kvm_cpuid_entry2 *best = NULL;
5381 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5382 struct kvm_cpuid_entry2 *e;
5384 e = &vcpu->arch.cpuid_entries[i];
5385 if (is_matching_cpuid_entry(e, function, index)) {
5386 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5387 move_to_next_stateful_cpuid_entry(vcpu, i);
5394 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5396 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5398 struct kvm_cpuid_entry2 *best;
5400 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5401 if (!best || best->eax < 0x80000008)
5403 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5405 return best->eax & 0xff;
5411 * If no match is found, check whether we exceed the vCPU's limit
5412 * and return the content of the highest valid _standard_ leaf instead.
5413 * This is to satisfy the CPUID specification.
5415 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5416 u32 function, u32 index)
5418 struct kvm_cpuid_entry2 *maxlevel;
5420 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5421 if (!maxlevel || maxlevel->eax >= function)
5423 if (function & 0x80000000) {
5424 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5428 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5431 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5433 u32 function, index;
5434 struct kvm_cpuid_entry2 *best;
5436 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5437 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5438 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5439 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5440 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5441 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5442 best = kvm_find_cpuid_entry(vcpu, function, index);
5445 best = check_cpuid_limit(vcpu, function, index);
5448 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5449 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5450 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5451 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5453 kvm_x86_ops->skip_emulated_instruction(vcpu);
5454 trace_kvm_cpuid(function,
5455 kvm_register_read(vcpu, VCPU_REGS_RAX),
5456 kvm_register_read(vcpu, VCPU_REGS_RBX),
5457 kvm_register_read(vcpu, VCPU_REGS_RCX),
5458 kvm_register_read(vcpu, VCPU_REGS_RDX));
5460 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5463 * Check if userspace requested an interrupt window, and that the
5464 * interrupt window is open.
5466 * No need to exit to userspace if we already have an interrupt queued.
5468 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5470 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5471 vcpu->run->request_interrupt_window &&
5472 kvm_arch_interrupt_allowed(vcpu));
5475 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5477 struct kvm_run *kvm_run = vcpu->run;
5479 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5480 kvm_run->cr8 = kvm_get_cr8(vcpu);
5481 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5482 if (irqchip_in_kernel(vcpu->kvm))
5483 kvm_run->ready_for_interrupt_injection = 1;
5485 kvm_run->ready_for_interrupt_injection =
5486 kvm_arch_interrupt_allowed(vcpu) &&
5487 !kvm_cpu_has_interrupt(vcpu) &&
5488 !kvm_event_needs_reinjection(vcpu);
5491 static void vapic_enter(struct kvm_vcpu *vcpu)
5493 struct kvm_lapic *apic = vcpu->arch.apic;
5496 if (!apic || !apic->vapic_addr)
5499 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5501 vcpu->arch.apic->vapic_page = page;
5504 static void vapic_exit(struct kvm_vcpu *vcpu)
5506 struct kvm_lapic *apic = vcpu->arch.apic;
5509 if (!apic || !apic->vapic_addr)
5512 idx = srcu_read_lock(&vcpu->kvm->srcu);
5513 kvm_release_page_dirty(apic->vapic_page);
5514 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5515 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5518 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5522 if (!kvm_x86_ops->update_cr8_intercept)
5525 if (!vcpu->arch.apic)
5528 if (!vcpu->arch.apic->vapic_addr)
5529 max_irr = kvm_lapic_find_highest_irr(vcpu);
5536 tpr = kvm_lapic_get_cr8(vcpu);
5538 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5541 static void inject_pending_event(struct kvm_vcpu *vcpu)
5543 /* try to reinject previous events if any */
5544 if (vcpu->arch.exception.pending) {
5545 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5546 vcpu->arch.exception.has_error_code,
5547 vcpu->arch.exception.error_code);
5548 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5549 vcpu->arch.exception.has_error_code,
5550 vcpu->arch.exception.error_code,
5551 vcpu->arch.exception.reinject);
5555 if (vcpu->arch.nmi_injected) {
5556 kvm_x86_ops->set_nmi(vcpu);
5560 if (vcpu->arch.interrupt.pending) {
5561 kvm_x86_ops->set_irq(vcpu);
5565 /* try to inject new event if pending */
5566 if (vcpu->arch.nmi_pending) {
5567 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5568 vcpu->arch.nmi_pending = false;
5569 vcpu->arch.nmi_injected = true;
5570 kvm_x86_ops->set_nmi(vcpu);
5572 } else if (kvm_cpu_has_interrupt(vcpu)) {
5573 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5574 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5576 kvm_x86_ops->set_irq(vcpu);
5581 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5583 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5584 !vcpu->guest_xcr0_loaded) {
5585 /* kvm_set_xcr() also depends on this */
5586 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5587 vcpu->guest_xcr0_loaded = 1;
5591 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5593 if (vcpu->guest_xcr0_loaded) {
5594 if (vcpu->arch.xcr0 != host_xcr0)
5595 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5596 vcpu->guest_xcr0_loaded = 0;
5600 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5604 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5605 vcpu->run->request_interrupt_window;
5607 if (vcpu->requests) {
5608 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5609 kvm_mmu_unload(vcpu);
5610 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5611 __kvm_migrate_timers(vcpu);
5612 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5613 r = kvm_guest_time_update(vcpu);
5617 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5618 kvm_mmu_sync_roots(vcpu);
5619 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5620 kvm_x86_ops->tlb_flush(vcpu);
5621 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5622 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5626 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5627 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5631 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5632 vcpu->fpu_active = 0;
5633 kvm_x86_ops->fpu_deactivate(vcpu);
5635 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5636 /* Page is swapped out. Do synthetic halt */
5637 vcpu->arch.apf.halted = true;
5641 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5642 record_steal_time(vcpu);
5646 r = kvm_mmu_reload(vcpu);
5651 * An NMI can be injected between local nmi_pending read and
5652 * vcpu->arch.nmi_pending read inside inject_pending_event().
5653 * But in that case, KVM_REQ_EVENT will be set, which makes
5654 * the race described above benign.
5656 nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5658 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5659 inject_pending_event(vcpu);
5661 /* enable NMI/IRQ window open exits if needed */
5663 kvm_x86_ops->enable_nmi_window(vcpu);
5664 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5665 kvm_x86_ops->enable_irq_window(vcpu);
5667 if (kvm_lapic_enabled(vcpu)) {
5668 update_cr8_intercept(vcpu);
5669 kvm_lapic_sync_to_vapic(vcpu);
5675 kvm_x86_ops->prepare_guest_switch(vcpu);
5676 if (vcpu->fpu_active)
5677 kvm_load_guest_fpu(vcpu);
5678 kvm_load_guest_xcr0(vcpu);
5680 vcpu->mode = IN_GUEST_MODE;
5682 /* We should set ->mode before check ->requests,
5683 * see the comment in make_all_cpus_request.
5687 local_irq_disable();
5689 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5690 || need_resched() || signal_pending(current)) {
5691 vcpu->mode = OUTSIDE_GUEST_MODE;
5695 kvm_x86_ops->cancel_injection(vcpu);
5700 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5704 if (unlikely(vcpu->arch.switch_db_regs)) {
5706 set_debugreg(vcpu->arch.eff_db[0], 0);
5707 set_debugreg(vcpu->arch.eff_db[1], 1);
5708 set_debugreg(vcpu->arch.eff_db[2], 2);
5709 set_debugreg(vcpu->arch.eff_db[3], 3);
5712 trace_kvm_entry(vcpu->vcpu_id);
5713 kvm_x86_ops->run(vcpu);
5716 * If the guest has used debug registers, at least dr7
5717 * will be disabled while returning to the host.
5718 * If we don't have active breakpoints in the host, we don't
5719 * care about the messed up debug address registers. But if
5720 * we have some of them active, restore the old state.
5722 if (hw_breakpoint_active())
5723 hw_breakpoint_restore();
5725 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5727 vcpu->mode = OUTSIDE_GUEST_MODE;
5734 * We must have an instruction between local_irq_enable() and
5735 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5736 * the interrupt shadow. The stat.exits increment will do nicely.
5737 * But we need to prevent reordering, hence this barrier():
5745 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5748 * Profile KVM exit RIPs:
5750 if (unlikely(prof_on == KVM_PROFILING)) {
5751 unsigned long rip = kvm_rip_read(vcpu);
5752 profile_hit(KVM_PROFILING, (void *)rip);
5756 kvm_lapic_sync_from_vapic(vcpu);
5758 r = kvm_x86_ops->handle_exit(vcpu);
5764 static int __vcpu_run(struct kvm_vcpu *vcpu)
5767 struct kvm *kvm = vcpu->kvm;
5769 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5770 pr_debug("vcpu %d received sipi with vector # %x\n",
5771 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5772 kvm_lapic_reset(vcpu);
5773 r = kvm_arch_vcpu_reset(vcpu);
5776 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5779 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5784 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5785 !vcpu->arch.apf.halted)
5786 r = vcpu_enter_guest(vcpu);
5788 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5789 kvm_vcpu_block(vcpu);
5790 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5791 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5793 switch(vcpu->arch.mp_state) {
5794 case KVM_MP_STATE_HALTED:
5795 vcpu->arch.mp_state =
5796 KVM_MP_STATE_RUNNABLE;
5797 case KVM_MP_STATE_RUNNABLE:
5798 vcpu->arch.apf.halted = false;
5800 case KVM_MP_STATE_SIPI_RECEIVED:
5811 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5812 if (kvm_cpu_has_pending_timer(vcpu))
5813 kvm_inject_pending_timer_irqs(vcpu);
5815 if (dm_request_for_irq_injection(vcpu)) {
5817 vcpu->run->exit_reason = KVM_EXIT_INTR;
5818 ++vcpu->stat.request_irq_exits;
5821 kvm_check_async_pf_completion(vcpu);
5823 if (signal_pending(current)) {
5825 vcpu->run->exit_reason = KVM_EXIT_INTR;
5826 ++vcpu->stat.signal_exits;
5828 if (need_resched()) {
5829 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5831 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5835 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5842 static int complete_mmio(struct kvm_vcpu *vcpu)
5844 struct kvm_run *run = vcpu->run;
5847 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5850 if (vcpu->mmio_needed) {
5851 vcpu->mmio_needed = 0;
5852 if (!vcpu->mmio_is_write)
5853 memcpy(vcpu->mmio_data + vcpu->mmio_index,
5855 vcpu->mmio_index += 8;
5856 if (vcpu->mmio_index < vcpu->mmio_size) {
5857 run->exit_reason = KVM_EXIT_MMIO;
5858 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5859 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5860 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5861 run->mmio.is_write = vcpu->mmio_is_write;
5862 vcpu->mmio_needed = 1;
5865 if (vcpu->mmio_is_write)
5867 vcpu->mmio_read_completed = 1;
5869 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5870 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5871 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5872 if (r != EMULATE_DONE)
5877 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5882 if (!tsk_used_math(current) && init_fpu(current))
5885 if (vcpu->sigset_active)
5886 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5888 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5889 kvm_vcpu_block(vcpu);
5890 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5895 /* re-sync apic's tpr */
5896 if (!irqchip_in_kernel(vcpu->kvm)) {
5897 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5903 r = complete_mmio(vcpu);
5907 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5908 kvm_register_write(vcpu, VCPU_REGS_RAX,
5909 kvm_run->hypercall.ret);
5911 r = __vcpu_run(vcpu);
5914 post_kvm_run_save(vcpu);
5915 if (vcpu->sigset_active)
5916 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5921 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5923 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5925 * We are here if userspace calls get_regs() in the middle of
5926 * instruction emulation. Registers state needs to be copied
5927 * back from emulation context to vcpu. Usrapace shouldn't do
5928 * that usually, but some bad designed PV devices (vmware
5929 * backdoor interface) need this to work
5931 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5932 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5933 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5935 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5936 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5937 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5938 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5939 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5940 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5941 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5942 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5943 #ifdef CONFIG_X86_64
5944 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5945 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5946 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5947 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5948 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5949 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5950 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5951 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5954 regs->rip = kvm_rip_read(vcpu);
5955 regs->rflags = kvm_get_rflags(vcpu);
5960 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5962 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5963 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5965 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5966 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5967 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5968 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5969 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5970 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5971 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5972 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5973 #ifdef CONFIG_X86_64
5974 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5975 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5976 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5977 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5978 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5979 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5980 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5981 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5984 kvm_rip_write(vcpu, regs->rip);
5985 kvm_set_rflags(vcpu, regs->rflags);
5987 vcpu->arch.exception.pending = false;
5989 kvm_make_request(KVM_REQ_EVENT, vcpu);
5994 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5996 struct kvm_segment cs;
5998 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6002 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6004 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6005 struct kvm_sregs *sregs)
6009 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6010 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6011 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6012 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6013 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6014 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6016 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6017 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6019 kvm_x86_ops->get_idt(vcpu, &dt);
6020 sregs->idt.limit = dt.size;
6021 sregs->idt.base = dt.address;
6022 kvm_x86_ops->get_gdt(vcpu, &dt);
6023 sregs->gdt.limit = dt.size;
6024 sregs->gdt.base = dt.address;
6026 sregs->cr0 = kvm_read_cr0(vcpu);
6027 sregs->cr2 = vcpu->arch.cr2;
6028 sregs->cr3 = kvm_read_cr3(vcpu);
6029 sregs->cr4 = kvm_read_cr4(vcpu);
6030 sregs->cr8 = kvm_get_cr8(vcpu);
6031 sregs->efer = vcpu->arch.efer;
6032 sregs->apic_base = kvm_get_apic_base(vcpu);
6034 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6036 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6037 set_bit(vcpu->arch.interrupt.nr,
6038 (unsigned long *)sregs->interrupt_bitmap);
6043 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6044 struct kvm_mp_state *mp_state)
6046 mp_state->mp_state = vcpu->arch.mp_state;
6050 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6051 struct kvm_mp_state *mp_state)
6053 vcpu->arch.mp_state = mp_state->mp_state;
6054 kvm_make_request(KVM_REQ_EVENT, vcpu);
6058 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
6059 bool has_error_code, u32 error_code)
6061 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6064 init_emulate_ctxt(vcpu);
6066 ret = emulator_task_switch(ctxt, tss_selector, reason,
6067 has_error_code, error_code);
6070 return EMULATE_FAIL;
6072 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
6073 kvm_rip_write(vcpu, ctxt->eip);
6074 kvm_set_rflags(vcpu, ctxt->eflags);
6075 kvm_make_request(KVM_REQ_EVENT, vcpu);
6076 return EMULATE_DONE;
6078 EXPORT_SYMBOL_GPL(kvm_task_switch);
6080 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6081 struct kvm_sregs *sregs)
6083 int mmu_reset_needed = 0;
6084 int pending_vec, max_bits, idx;
6087 dt.size = sregs->idt.limit;
6088 dt.address = sregs->idt.base;
6089 kvm_x86_ops->set_idt(vcpu, &dt);
6090 dt.size = sregs->gdt.limit;
6091 dt.address = sregs->gdt.base;
6092 kvm_x86_ops->set_gdt(vcpu, &dt);
6094 vcpu->arch.cr2 = sregs->cr2;
6095 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6096 vcpu->arch.cr3 = sregs->cr3;
6097 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6099 kvm_set_cr8(vcpu, sregs->cr8);
6101 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6102 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6103 kvm_set_apic_base(vcpu, sregs->apic_base);
6105 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6106 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6107 vcpu->arch.cr0 = sregs->cr0;
6109 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6110 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6111 if (sregs->cr4 & X86_CR4_OSXSAVE)
6114 idx = srcu_read_lock(&vcpu->kvm->srcu);
6115 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6116 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6117 mmu_reset_needed = 1;
6119 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6121 if (mmu_reset_needed)
6122 kvm_mmu_reset_context(vcpu);
6124 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
6125 pending_vec = find_first_bit(
6126 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6127 if (pending_vec < max_bits) {
6128 kvm_queue_interrupt(vcpu, pending_vec, false);
6129 pr_debug("Set back pending irq %d\n", pending_vec);
6132 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6133 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6134 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6135 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6136 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6137 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6139 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6140 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6142 update_cr8_intercept(vcpu);
6144 /* Older userspace won't unhalt the vcpu on reset. */
6145 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6146 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6148 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6150 kvm_make_request(KVM_REQ_EVENT, vcpu);
6155 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6156 struct kvm_guest_debug *dbg)
6158 unsigned long rflags;
6161 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6163 if (vcpu->arch.exception.pending)
6165 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6166 kvm_queue_exception(vcpu, DB_VECTOR);
6168 kvm_queue_exception(vcpu, BP_VECTOR);
6172 * Read rflags as long as potentially injected trace flags are still
6175 rflags = kvm_get_rflags(vcpu);
6177 vcpu->guest_debug = dbg->control;
6178 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6179 vcpu->guest_debug = 0;
6181 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6182 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6183 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6184 vcpu->arch.switch_db_regs =
6185 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
6187 for (i = 0; i < KVM_NR_DB_REGS; i++)
6188 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6189 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
6192 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6193 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6194 get_segment_base(vcpu, VCPU_SREG_CS);
6197 * Trigger an rflags update that will inject or remove the trace
6200 kvm_set_rflags(vcpu, rflags);
6202 kvm_x86_ops->set_guest_debug(vcpu, dbg);
6212 * Translate a guest virtual address to a guest physical address.
6214 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6215 struct kvm_translation *tr)
6217 unsigned long vaddr = tr->linear_address;
6221 idx = srcu_read_lock(&vcpu->kvm->srcu);
6222 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6223 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6224 tr->physical_address = gpa;
6225 tr->valid = gpa != UNMAPPED_GVA;
6232 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6234 struct i387_fxsave_struct *fxsave =
6235 &vcpu->arch.guest_fpu.state->fxsave;
6237 memcpy(fpu->fpr, fxsave->st_space, 128);
6238 fpu->fcw = fxsave->cwd;
6239 fpu->fsw = fxsave->swd;
6240 fpu->ftwx = fxsave->twd;
6241 fpu->last_opcode = fxsave->fop;
6242 fpu->last_ip = fxsave->rip;
6243 fpu->last_dp = fxsave->rdp;
6244 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6249 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6251 struct i387_fxsave_struct *fxsave =
6252 &vcpu->arch.guest_fpu.state->fxsave;
6254 memcpy(fxsave->st_space, fpu->fpr, 128);
6255 fxsave->cwd = fpu->fcw;
6256 fxsave->swd = fpu->fsw;
6257 fxsave->twd = fpu->ftwx;
6258 fxsave->fop = fpu->last_opcode;
6259 fxsave->rip = fpu->last_ip;
6260 fxsave->rdp = fpu->last_dp;
6261 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6266 int fx_init(struct kvm_vcpu *vcpu)
6270 err = fpu_alloc(&vcpu->arch.guest_fpu);
6274 fpu_finit(&vcpu->arch.guest_fpu);
6277 * Ensure guest xcr0 is valid for loading
6279 vcpu->arch.xcr0 = XSTATE_FP;
6281 vcpu->arch.cr0 |= X86_CR0_ET;
6285 EXPORT_SYMBOL_GPL(fx_init);
6287 static void fx_free(struct kvm_vcpu *vcpu)
6289 fpu_free(&vcpu->arch.guest_fpu);
6292 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6294 if (vcpu->guest_fpu_loaded)
6298 * Restore all possible states in the guest,
6299 * and assume host would use all available bits.
6300 * Guest xcr0 would be loaded later.
6302 kvm_put_guest_xcr0(vcpu);
6303 vcpu->guest_fpu_loaded = 1;
6304 unlazy_fpu(current);
6305 fpu_restore_checking(&vcpu->arch.guest_fpu);
6309 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6311 kvm_put_guest_xcr0(vcpu);
6313 if (!vcpu->guest_fpu_loaded)
6316 vcpu->guest_fpu_loaded = 0;
6317 fpu_save_init(&vcpu->arch.guest_fpu);
6318 ++vcpu->stat.fpu_reload;
6319 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6323 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6325 kvmclock_reset(vcpu);
6327 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6329 kvm_x86_ops->vcpu_free(vcpu);
6332 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6335 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6336 printk_once(KERN_WARNING
6337 "kvm: SMP vm created on host with unstable TSC; "
6338 "guest TSC will not be reliable\n");
6339 return kvm_x86_ops->vcpu_create(kvm, id);
6342 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6346 vcpu->arch.mtrr_state.have_fixed = 1;
6348 r = kvm_arch_vcpu_reset(vcpu);
6350 r = kvm_mmu_setup(vcpu);
6356 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6358 vcpu->arch.apf.msr_val = 0;
6361 kvm_mmu_unload(vcpu);
6365 kvm_x86_ops->vcpu_free(vcpu);
6368 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6370 vcpu->arch.nmi_pending = false;
6371 vcpu->arch.nmi_injected = false;
6373 vcpu->arch.switch_db_regs = 0;
6374 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6375 vcpu->arch.dr6 = DR6_FIXED_1;
6376 vcpu->arch.dr7 = DR7_FIXED_1;
6378 kvm_make_request(KVM_REQ_EVENT, vcpu);
6379 vcpu->arch.apf.msr_val = 0;
6380 vcpu->arch.st.msr_val = 0;
6382 kvmclock_reset(vcpu);
6384 kvm_clear_async_pf_completion_queue(vcpu);
6385 kvm_async_pf_hash_reset(vcpu);
6386 vcpu->arch.apf.halted = false;
6388 return kvm_x86_ops->vcpu_reset(vcpu);
6391 int kvm_arch_hardware_enable(void *garbage)
6394 struct kvm_vcpu *vcpu;
6397 kvm_shared_msr_cpu_online();
6398 list_for_each_entry(kvm, &vm_list, vm_list)
6399 kvm_for_each_vcpu(i, vcpu, kvm)
6400 if (vcpu->cpu == smp_processor_id())
6401 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6402 return kvm_x86_ops->hardware_enable(garbage);
6405 void kvm_arch_hardware_disable(void *garbage)
6407 kvm_x86_ops->hardware_disable(garbage);
6408 drop_user_return_notifiers(garbage);
6411 int kvm_arch_hardware_setup(void)
6413 return kvm_x86_ops->hardware_setup();
6416 void kvm_arch_hardware_unsetup(void)
6418 kvm_x86_ops->hardware_unsetup();
6421 void kvm_arch_check_processor_compat(void *rtn)
6423 kvm_x86_ops->check_processor_compatibility(rtn);
6426 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6432 BUG_ON(vcpu->kvm == NULL);
6435 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6436 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6437 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6438 vcpu->arch.mmu.translate_gpa = translate_gpa;
6439 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6440 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6441 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6443 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6445 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6450 vcpu->arch.pio_data = page_address(page);
6452 kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6454 r = kvm_mmu_create(vcpu);
6456 goto fail_free_pio_data;
6458 if (irqchip_in_kernel(kvm)) {
6459 r = kvm_create_lapic(vcpu);
6461 goto fail_mmu_destroy;
6464 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6466 if (!vcpu->arch.mce_banks) {
6468 goto fail_free_lapic;
6470 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6472 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6473 goto fail_free_mce_banks;
6475 kvm_async_pf_hash_reset(vcpu);
6478 fail_free_mce_banks:
6479 kfree(vcpu->arch.mce_banks);
6481 kvm_free_lapic(vcpu);
6483 kvm_mmu_destroy(vcpu);
6485 free_page((unsigned long)vcpu->arch.pio_data);
6490 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6494 kfree(vcpu->arch.mce_banks);
6495 kvm_free_lapic(vcpu);
6496 idx = srcu_read_lock(&vcpu->kvm->srcu);
6497 kvm_mmu_destroy(vcpu);
6498 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6499 free_page((unsigned long)vcpu->arch.pio_data);
6502 int kvm_arch_init_vm(struct kvm *kvm)
6504 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6505 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6507 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6508 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6510 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6515 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6518 kvm_mmu_unload(vcpu);
6522 static void kvm_free_vcpus(struct kvm *kvm)
6525 struct kvm_vcpu *vcpu;
6528 * Unpin any mmu pages first.
6530 kvm_for_each_vcpu(i, vcpu, kvm) {
6531 kvm_clear_async_pf_completion_queue(vcpu);
6532 kvm_unload_vcpu_mmu(vcpu);
6534 kvm_for_each_vcpu(i, vcpu, kvm)
6535 kvm_arch_vcpu_free(vcpu);
6537 mutex_lock(&kvm->lock);
6538 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6539 kvm->vcpus[i] = NULL;
6541 atomic_set(&kvm->online_vcpus, 0);
6542 mutex_unlock(&kvm->lock);
6545 void kvm_arch_sync_events(struct kvm *kvm)
6547 kvm_free_all_assigned_devices(kvm);
6551 void kvm_arch_destroy_vm(struct kvm *kvm)
6553 kvm_iommu_unmap_guest(kvm);
6554 kfree(kvm->arch.vpic);
6555 kfree(kvm->arch.vioapic);
6556 kvm_free_vcpus(kvm);
6557 if (kvm->arch.apic_access_page)
6558 put_page(kvm->arch.apic_access_page);
6559 if (kvm->arch.ept_identity_pagetable)
6560 put_page(kvm->arch.ept_identity_pagetable);
6563 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6564 struct kvm_memory_slot *memslot,
6565 struct kvm_memory_slot old,
6566 struct kvm_userspace_memory_region *mem,
6569 int npages = memslot->npages;
6570 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6572 /* Prevent internal slot pages from being moved by fork()/COW. */
6573 if (memslot->id >= KVM_MEMORY_SLOTS)
6574 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6576 /*To keep backward compatibility with older userspace,
6577 *x86 needs to hanlde !user_alloc case.
6580 if (npages && !old.rmap) {
6581 unsigned long userspace_addr;
6583 down_write(¤t->mm->mmap_sem);
6584 userspace_addr = do_mmap(NULL, 0,
6586 PROT_READ | PROT_WRITE,
6589 up_write(¤t->mm->mmap_sem);
6591 if (IS_ERR((void *)userspace_addr))
6592 return PTR_ERR((void *)userspace_addr);
6594 memslot->userspace_addr = userspace_addr;
6602 void kvm_arch_commit_memory_region(struct kvm *kvm,
6603 struct kvm_userspace_memory_region *mem,
6604 struct kvm_memory_slot old,
6608 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6610 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6613 down_write(¤t->mm->mmap_sem);
6614 ret = do_munmap(current->mm, old.userspace_addr,
6615 old.npages * PAGE_SIZE);
6616 up_write(¤t->mm->mmap_sem);
6619 "kvm_vm_ioctl_set_memory_region: "
6620 "failed to munmap memory\n");
6623 if (!kvm->arch.n_requested_mmu_pages)
6624 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6626 spin_lock(&kvm->mmu_lock);
6628 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6629 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6630 spin_unlock(&kvm->mmu_lock);
6633 void kvm_arch_flush_shadow(struct kvm *kvm)
6635 kvm_mmu_zap_all(kvm);
6636 kvm_reload_remote_mmus(kvm);
6639 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6641 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6642 !vcpu->arch.apf.halted)
6643 || !list_empty_careful(&vcpu->async_pf.done)
6644 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6645 || vcpu->arch.nmi_pending ||
6646 (kvm_arch_interrupt_allowed(vcpu) &&
6647 kvm_cpu_has_interrupt(vcpu));
6650 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6653 int cpu = vcpu->cpu;
6655 if (waitqueue_active(&vcpu->wq)) {
6656 wake_up_interruptible(&vcpu->wq);
6657 ++vcpu->stat.halt_wakeup;
6661 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6662 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6663 smp_send_reschedule(cpu);
6667 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6669 return kvm_x86_ops->interrupt_allowed(vcpu);
6672 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6674 unsigned long current_rip = kvm_rip_read(vcpu) +
6675 get_segment_base(vcpu, VCPU_SREG_CS);
6677 return current_rip == linear_rip;
6679 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6681 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6683 unsigned long rflags;
6685 rflags = kvm_x86_ops->get_rflags(vcpu);
6686 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6687 rflags &= ~X86_EFLAGS_TF;
6690 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6692 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6694 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6695 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6696 rflags |= X86_EFLAGS_TF;
6697 kvm_x86_ops->set_rflags(vcpu, rflags);
6698 kvm_make_request(KVM_REQ_EVENT, vcpu);
6700 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6702 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6706 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6707 is_error_page(work->page))
6710 r = kvm_mmu_reload(vcpu);
6714 if (!vcpu->arch.mmu.direct_map &&
6715 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6718 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6721 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6723 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6726 static inline u32 kvm_async_pf_next_probe(u32 key)
6728 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6731 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6733 u32 key = kvm_async_pf_hash_fn(gfn);
6735 while (vcpu->arch.apf.gfns[key] != ~0)
6736 key = kvm_async_pf_next_probe(key);
6738 vcpu->arch.apf.gfns[key] = gfn;
6741 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6744 u32 key = kvm_async_pf_hash_fn(gfn);
6746 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6747 (vcpu->arch.apf.gfns[key] != gfn &&
6748 vcpu->arch.apf.gfns[key] != ~0); i++)
6749 key = kvm_async_pf_next_probe(key);
6754 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6756 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6759 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6763 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6765 vcpu->arch.apf.gfns[i] = ~0;
6767 j = kvm_async_pf_next_probe(j);
6768 if (vcpu->arch.apf.gfns[j] == ~0)
6770 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6772 * k lies cyclically in ]i,j]
6774 * |....j i.k.| or |.k..j i...|
6776 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6777 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6782 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6785 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6789 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6790 struct kvm_async_pf *work)
6792 struct x86_exception fault;
6794 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6795 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6797 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6798 (vcpu->arch.apf.send_user_only &&
6799 kvm_x86_ops->get_cpl(vcpu) == 0))
6800 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6801 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6802 fault.vector = PF_VECTOR;
6803 fault.error_code_valid = true;
6804 fault.error_code = 0;
6805 fault.nested_page_fault = false;
6806 fault.address = work->arch.token;
6807 kvm_inject_page_fault(vcpu, &fault);
6811 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6812 struct kvm_async_pf *work)
6814 struct x86_exception fault;
6816 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6817 if (is_error_page(work->page))
6818 work->arch.token = ~0; /* broadcast wakeup */
6820 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6822 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6823 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6824 fault.vector = PF_VECTOR;
6825 fault.error_code_valid = true;
6826 fault.error_code = 0;
6827 fault.nested_page_fault = false;
6828 fault.address = work->arch.token;
6829 kvm_inject_page_fault(vcpu, &fault);
6831 vcpu->arch.apf.halted = false;
6834 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6836 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6839 return !kvm_event_needs_reinjection(vcpu) &&
6840 kvm_x86_ops->interrupt_allowed(vcpu);
6843 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6844 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6845 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6846 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6847 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6848 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6849 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6850 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6851 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6852 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6853 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6854 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);