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:
1561 case 0x200 ... 0x2ff:
1562 return set_msr_mtrr(vcpu, msr, data);
1563 case MSR_IA32_APICBASE:
1564 kvm_set_apic_base(vcpu, data);
1566 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1567 return kvm_x2apic_msr_write(vcpu, msr, data);
1568 case MSR_IA32_MISC_ENABLE:
1569 vcpu->arch.ia32_misc_enable_msr = data;
1571 case MSR_KVM_WALL_CLOCK_NEW:
1572 case MSR_KVM_WALL_CLOCK:
1573 vcpu->kvm->arch.wall_clock = data;
1574 kvm_write_wall_clock(vcpu->kvm, data);
1576 case MSR_KVM_SYSTEM_TIME_NEW:
1577 case MSR_KVM_SYSTEM_TIME: {
1578 kvmclock_reset(vcpu);
1580 vcpu->arch.time = data;
1581 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1583 /* we verify if the enable bit is set... */
1587 /* ...but clean it before doing the actual write */
1588 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1590 vcpu->arch.time_page =
1591 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1593 if (is_error_page(vcpu->arch.time_page)) {
1594 kvm_release_page_clean(vcpu->arch.time_page);
1595 vcpu->arch.time_page = NULL;
1599 case MSR_KVM_ASYNC_PF_EN:
1600 if (kvm_pv_enable_async_pf(vcpu, data))
1603 case MSR_KVM_STEAL_TIME:
1605 if (unlikely(!sched_info_on()))
1608 if (data & KVM_STEAL_RESERVED_MASK)
1611 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1612 data & KVM_STEAL_VALID_BITS))
1615 vcpu->arch.st.msr_val = data;
1617 if (!(data & KVM_MSR_ENABLED))
1620 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1623 accumulate_steal_time(vcpu);
1626 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1630 case MSR_IA32_MCG_CTL:
1631 case MSR_IA32_MCG_STATUS:
1632 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1633 return set_msr_mce(vcpu, msr, data);
1635 /* Performance counters are not protected by a CPUID bit,
1636 * so we should check all of them in the generic path for the sake of
1637 * cross vendor migration.
1638 * Writing a zero into the event select MSRs disables them,
1639 * which we perfectly emulate ;-). Any other value should be at least
1640 * reported, some guests depend on them.
1642 case MSR_P6_EVNTSEL0:
1643 case MSR_P6_EVNTSEL1:
1644 case MSR_K7_EVNTSEL0:
1645 case MSR_K7_EVNTSEL1:
1646 case MSR_K7_EVNTSEL2:
1647 case MSR_K7_EVNTSEL3:
1649 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1650 "0x%x data 0x%llx\n", msr, data);
1652 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1653 * so we ignore writes to make it happy.
1655 case MSR_P6_PERFCTR0:
1656 case MSR_P6_PERFCTR1:
1657 case MSR_K7_PERFCTR0:
1658 case MSR_K7_PERFCTR1:
1659 case MSR_K7_PERFCTR2:
1660 case MSR_K7_PERFCTR3:
1661 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1662 "0x%x data 0x%llx\n", msr, data);
1664 case MSR_K7_CLK_CTL:
1666 * Ignore all writes to this no longer documented MSR.
1667 * Writes are only relevant for old K7 processors,
1668 * all pre-dating SVM, but a recommended workaround from
1669 * AMD for these chips. It is possible to speicify the
1670 * affected processor models on the command line, hence
1671 * the need to ignore the workaround.
1674 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1675 if (kvm_hv_msr_partition_wide(msr)) {
1677 mutex_lock(&vcpu->kvm->lock);
1678 r = set_msr_hyperv_pw(vcpu, msr, data);
1679 mutex_unlock(&vcpu->kvm->lock);
1682 return set_msr_hyperv(vcpu, msr, data);
1684 case MSR_IA32_BBL_CR_CTL3:
1685 /* Drop writes to this legacy MSR -- see rdmsr
1686 * counterpart for further detail.
1688 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1691 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1692 return xen_hvm_config(vcpu, data);
1694 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1698 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1705 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1709 * Reads an msr value (of 'msr_index') into 'pdata'.
1710 * Returns 0 on success, non-0 otherwise.
1711 * Assumes vcpu_load() was already called.
1713 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1715 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1718 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1720 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1722 if (!msr_mtrr_valid(msr))
1725 if (msr == MSR_MTRRdefType)
1726 *pdata = vcpu->arch.mtrr_state.def_type +
1727 (vcpu->arch.mtrr_state.enabled << 10);
1728 else if (msr == MSR_MTRRfix64K_00000)
1730 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1731 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1732 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1733 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1734 else if (msr == MSR_IA32_CR_PAT)
1735 *pdata = vcpu->arch.pat;
1736 else { /* Variable MTRRs */
1737 int idx, is_mtrr_mask;
1740 idx = (msr - 0x200) / 2;
1741 is_mtrr_mask = msr - 0x200 - 2 * idx;
1744 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1747 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1754 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1757 u64 mcg_cap = vcpu->arch.mcg_cap;
1758 unsigned bank_num = mcg_cap & 0xff;
1761 case MSR_IA32_P5_MC_ADDR:
1762 case MSR_IA32_P5_MC_TYPE:
1765 case MSR_IA32_MCG_CAP:
1766 data = vcpu->arch.mcg_cap;
1768 case MSR_IA32_MCG_CTL:
1769 if (!(mcg_cap & MCG_CTL_P))
1771 data = vcpu->arch.mcg_ctl;
1773 case MSR_IA32_MCG_STATUS:
1774 data = vcpu->arch.mcg_status;
1777 if (msr >= MSR_IA32_MC0_CTL &&
1778 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1779 u32 offset = msr - MSR_IA32_MC0_CTL;
1780 data = vcpu->arch.mce_banks[offset];
1789 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1792 struct kvm *kvm = vcpu->kvm;
1795 case HV_X64_MSR_GUEST_OS_ID:
1796 data = kvm->arch.hv_guest_os_id;
1798 case HV_X64_MSR_HYPERCALL:
1799 data = kvm->arch.hv_hypercall;
1802 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1810 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1815 case HV_X64_MSR_VP_INDEX: {
1818 kvm_for_each_vcpu(r, v, vcpu->kvm)
1823 case HV_X64_MSR_EOI:
1824 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1825 case HV_X64_MSR_ICR:
1826 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1827 case HV_X64_MSR_TPR:
1828 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1830 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1837 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1842 case MSR_IA32_PLATFORM_ID:
1843 case MSR_IA32_UCODE_REV:
1844 case MSR_IA32_EBL_CR_POWERON:
1845 case MSR_IA32_DEBUGCTLMSR:
1846 case MSR_IA32_LASTBRANCHFROMIP:
1847 case MSR_IA32_LASTBRANCHTOIP:
1848 case MSR_IA32_LASTINTFROMIP:
1849 case MSR_IA32_LASTINTTOIP:
1852 case MSR_VM_HSAVE_PA:
1853 case MSR_P6_PERFCTR0:
1854 case MSR_P6_PERFCTR1:
1855 case MSR_P6_EVNTSEL0:
1856 case MSR_P6_EVNTSEL1:
1857 case MSR_K7_EVNTSEL0:
1858 case MSR_K7_PERFCTR0:
1859 case MSR_K8_INT_PENDING_MSG:
1860 case MSR_AMD64_NB_CFG:
1861 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:
2093 case KVM_CAP_NR_MEMSLOTS:
2094 r = KVM_MEMORY_SLOTS;
2096 case KVM_CAP_PV_MMU: /* obsolete */
2103 r = KVM_MAX_MCE_BANKS;
2108 case KVM_CAP_TSC_CONTROL:
2109 r = kvm_has_tsc_control;
2119 long kvm_arch_dev_ioctl(struct file *filp,
2120 unsigned int ioctl, unsigned long arg)
2122 void __user *argp = (void __user *)arg;
2126 case KVM_GET_MSR_INDEX_LIST: {
2127 struct kvm_msr_list __user *user_msr_list = argp;
2128 struct kvm_msr_list msr_list;
2132 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2135 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2136 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2139 if (n < msr_list.nmsrs)
2142 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2143 num_msrs_to_save * sizeof(u32)))
2145 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2147 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2152 case KVM_GET_SUPPORTED_CPUID: {
2153 struct kvm_cpuid2 __user *cpuid_arg = argp;
2154 struct kvm_cpuid2 cpuid;
2157 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2159 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2160 cpuid_arg->entries);
2165 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2170 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2173 mce_cap = KVM_MCE_CAP_SUPPORTED;
2175 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2187 static void wbinvd_ipi(void *garbage)
2192 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2194 return vcpu->kvm->arch.iommu_domain &&
2195 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2198 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2200 /* Address WBINVD may be executed by guest */
2201 if (need_emulate_wbinvd(vcpu)) {
2202 if (kvm_x86_ops->has_wbinvd_exit())
2203 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2204 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2205 smp_call_function_single(vcpu->cpu,
2206 wbinvd_ipi, NULL, 1);
2209 kvm_x86_ops->vcpu_load(vcpu, cpu);
2210 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2211 /* Make sure TSC doesn't go backwards */
2215 kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
2216 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2217 tsc - vcpu->arch.last_guest_tsc;
2220 mark_tsc_unstable("KVM discovered backwards TSC");
2221 if (check_tsc_unstable()) {
2222 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2223 vcpu->arch.tsc_catchup = 1;
2225 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2226 if (vcpu->cpu != cpu)
2227 kvm_migrate_timers(vcpu);
2231 accumulate_steal_time(vcpu);
2232 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2235 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2237 kvm_x86_ops->vcpu_put(vcpu);
2238 kvm_put_guest_fpu(vcpu);
2239 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
2242 static int is_efer_nx(void)
2244 unsigned long long efer = 0;
2246 rdmsrl_safe(MSR_EFER, &efer);
2247 return efer & EFER_NX;
2250 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2253 struct kvm_cpuid_entry2 *e, *entry;
2256 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2257 e = &vcpu->arch.cpuid_entries[i];
2258 if (e->function == 0x80000001) {
2263 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2264 entry->edx &= ~(1 << 20);
2265 printk(KERN_INFO "kvm: guest NX capability removed\n");
2269 /* when an old userspace process fills a new kernel module */
2270 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2271 struct kvm_cpuid *cpuid,
2272 struct kvm_cpuid_entry __user *entries)
2275 struct kvm_cpuid_entry *cpuid_entries;
2278 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2281 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2285 if (copy_from_user(cpuid_entries, entries,
2286 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2288 for (i = 0; i < cpuid->nent; i++) {
2289 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2290 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2291 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2292 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2293 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2294 vcpu->arch.cpuid_entries[i].index = 0;
2295 vcpu->arch.cpuid_entries[i].flags = 0;
2296 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2297 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2298 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2300 vcpu->arch.cpuid_nent = cpuid->nent;
2301 cpuid_fix_nx_cap(vcpu);
2303 kvm_apic_set_version(vcpu);
2304 kvm_x86_ops->cpuid_update(vcpu);
2308 vfree(cpuid_entries);
2313 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2314 struct kvm_cpuid2 *cpuid,
2315 struct kvm_cpuid_entry2 __user *entries)
2320 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2323 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2324 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2326 vcpu->arch.cpuid_nent = cpuid->nent;
2327 kvm_apic_set_version(vcpu);
2328 kvm_x86_ops->cpuid_update(vcpu);
2336 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2337 struct kvm_cpuid2 *cpuid,
2338 struct kvm_cpuid_entry2 __user *entries)
2343 if (cpuid->nent < vcpu->arch.cpuid_nent)
2346 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2347 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2352 cpuid->nent = vcpu->arch.cpuid_nent;
2356 static void cpuid_mask(u32 *word, int wordnum)
2358 *word &= boot_cpu_data.x86_capability[wordnum];
2361 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2364 entry->function = function;
2365 entry->index = index;
2366 cpuid_count(entry->function, entry->index,
2367 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2371 static bool supported_xcr0_bit(unsigned bit)
2373 u64 mask = ((u64)1 << bit);
2375 return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
2378 #define F(x) bit(X86_FEATURE_##x)
2380 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2381 u32 index, int *nent, int maxnent)
2383 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2384 #ifdef CONFIG_X86_64
2385 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2387 unsigned f_lm = F(LM);
2389 unsigned f_gbpages = 0;
2392 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2395 const u32 kvm_supported_word0_x86_features =
2396 F(FPU) | F(VME) | F(DE) | F(PSE) |
2397 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2398 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2399 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2400 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2401 0 /* Reserved, DS, ACPI */ | F(MMX) |
2402 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2403 0 /* HTT, TM, Reserved, PBE */;
2404 /* cpuid 0x80000001.edx */
2405 const u32 kvm_supported_word1_x86_features =
2406 F(FPU) | F(VME) | F(DE) | F(PSE) |
2407 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2408 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2409 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2410 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2411 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2412 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2413 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2415 const u32 kvm_supported_word4_x86_features =
2416 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64 */ | F(MWAIT) |
2417 0 /* DS-CPL, VMX, SMX, EST */ |
2418 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2419 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2420 0 /* Reserved, DCA */ | F(XMM4_1) |
2421 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2422 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2423 F(F16C) | F(RDRAND);
2424 /* cpuid 0x80000001.ecx */
2425 const u32 kvm_supported_word6_x86_features =
2426 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2427 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2428 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2429 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2431 /* cpuid 0xC0000001.edx */
2432 const u32 kvm_supported_word5_x86_features =
2433 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2434 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2438 const u32 kvm_supported_word9_x86_features =
2439 F(SMEP) | F(FSGSBASE) | F(ERMS);
2441 /* all calls to cpuid_count() should be made on the same cpu */
2443 do_cpuid_1_ent(entry, function, index);
2448 entry->eax = min(entry->eax, (u32)0xd);
2451 entry->edx &= kvm_supported_word0_x86_features;
2452 cpuid_mask(&entry->edx, 0);
2453 entry->ecx &= kvm_supported_word4_x86_features;
2454 cpuid_mask(&entry->ecx, 4);
2455 /* we support x2apic emulation even if host does not support
2456 * it since we emulate x2apic in software */
2457 entry->ecx |= F(X2APIC);
2459 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2460 * may return different values. This forces us to get_cpu() before
2461 * issuing the first command, and also to emulate this annoying behavior
2462 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2464 int t, times = entry->eax & 0xff;
2466 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2467 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2468 for (t = 1; t < times && *nent < maxnent; ++t) {
2469 do_cpuid_1_ent(&entry[t], function, 0);
2470 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2475 /* function 4 has additional index. */
2479 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2480 /* read more entries until cache_type is zero */
2481 for (i = 1; *nent < maxnent; ++i) {
2482 cache_type = entry[i - 1].eax & 0x1f;
2485 do_cpuid_1_ent(&entry[i], function, i);
2487 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2493 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2494 /* Mask ebx against host capbability word 9 */
2496 entry->ebx &= kvm_supported_word9_x86_features;
2497 cpuid_mask(&entry->ebx, 9);
2507 /* function 0xb has additional index. */
2511 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2512 /* read more entries until level_type is zero */
2513 for (i = 1; *nent < maxnent; ++i) {
2514 level_type = entry[i - 1].ecx & 0xff00;
2517 do_cpuid_1_ent(&entry[i], function, i);
2519 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2527 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2528 for (idx = 1, i = 1; *nent < maxnent && idx < 64; ++idx) {
2529 do_cpuid_1_ent(&entry[i], function, idx);
2530 if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
2533 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2539 case KVM_CPUID_SIGNATURE: {
2540 char signature[12] = "KVMKVMKVM\0\0";
2541 u32 *sigptr = (u32 *)signature;
2543 entry->ebx = sigptr[0];
2544 entry->ecx = sigptr[1];
2545 entry->edx = sigptr[2];
2548 case KVM_CPUID_FEATURES:
2549 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2550 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2551 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2552 (1 << KVM_FEATURE_ASYNC_PF) |
2553 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2555 if (sched_info_on())
2556 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
2563 entry->eax = min(entry->eax, 0x8000001a);
2566 entry->edx &= kvm_supported_word1_x86_features;
2567 cpuid_mask(&entry->edx, 1);
2568 entry->ecx &= kvm_supported_word6_x86_features;
2569 cpuid_mask(&entry->ecx, 6);
2572 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
2573 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
2574 unsigned phys_as = entry->eax & 0xff;
2577 g_phys_as = phys_as;
2578 entry->eax = g_phys_as | (virt_as << 8);
2579 entry->ebx = entry->edx = 0;
2583 entry->ecx = entry->edx = 0;
2589 /*Add support for Centaur's CPUID instruction*/
2591 /*Just support up to 0xC0000004 now*/
2592 entry->eax = min(entry->eax, 0xC0000004);
2595 entry->edx &= kvm_supported_word5_x86_features;
2596 cpuid_mask(&entry->edx, 5);
2598 case 3: /* Processor serial number */
2599 case 5: /* MONITOR/MWAIT */
2600 case 6: /* Thermal management */
2601 case 0xA: /* Architectural Performance Monitoring */
2602 case 0x80000007: /* Advanced power management */
2607 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
2611 kvm_x86_ops->set_supported_cpuid(function, entry);
2618 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2619 struct kvm_cpuid_entry2 __user *entries)
2621 struct kvm_cpuid_entry2 *cpuid_entries;
2622 int limit, nent = 0, r = -E2BIG;
2625 if (cpuid->nent < 1)
2627 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2628 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2630 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2634 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2635 limit = cpuid_entries[0].eax;
2636 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2637 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2638 &nent, cpuid->nent);
2640 if (nent >= cpuid->nent)
2643 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2644 limit = cpuid_entries[nent - 1].eax;
2645 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2646 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2647 &nent, cpuid->nent);
2652 if (nent >= cpuid->nent)
2655 /* Add support for Centaur's CPUID instruction. */
2656 if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2657 do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2658 &nent, cpuid->nent);
2661 if (nent >= cpuid->nent)
2664 limit = cpuid_entries[nent - 1].eax;
2665 for (func = 0xC0000001;
2666 func <= limit && nent < cpuid->nent; ++func)
2667 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2668 &nent, cpuid->nent);
2671 if (nent >= cpuid->nent)
2675 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2679 if (nent >= cpuid->nent)
2682 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2686 if (nent >= cpuid->nent)
2690 if (copy_to_user(entries, cpuid_entries,
2691 nent * sizeof(struct kvm_cpuid_entry2)))
2697 vfree(cpuid_entries);
2702 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2703 struct kvm_lapic_state *s)
2705 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2710 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2711 struct kvm_lapic_state *s)
2713 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2714 kvm_apic_post_state_restore(vcpu);
2715 update_cr8_intercept(vcpu);
2720 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2721 struct kvm_interrupt *irq)
2723 if (irq->irq < 0 || irq->irq >= 256)
2725 if (irqchip_in_kernel(vcpu->kvm))
2728 kvm_queue_interrupt(vcpu, irq->irq, false);
2729 kvm_make_request(KVM_REQ_EVENT, vcpu);
2734 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2736 kvm_inject_nmi(vcpu);
2741 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2742 struct kvm_tpr_access_ctl *tac)
2746 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2750 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2754 unsigned bank_num = mcg_cap & 0xff, bank;
2757 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2759 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2762 vcpu->arch.mcg_cap = mcg_cap;
2763 /* Init IA32_MCG_CTL to all 1s */
2764 if (mcg_cap & MCG_CTL_P)
2765 vcpu->arch.mcg_ctl = ~(u64)0;
2766 /* Init IA32_MCi_CTL to all 1s */
2767 for (bank = 0; bank < bank_num; bank++)
2768 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2773 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2774 struct kvm_x86_mce *mce)
2776 u64 mcg_cap = vcpu->arch.mcg_cap;
2777 unsigned bank_num = mcg_cap & 0xff;
2778 u64 *banks = vcpu->arch.mce_banks;
2780 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2783 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2784 * reporting is disabled
2786 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2787 vcpu->arch.mcg_ctl != ~(u64)0)
2789 banks += 4 * mce->bank;
2791 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2792 * reporting is disabled for the bank
2794 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2796 if (mce->status & MCI_STATUS_UC) {
2797 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2798 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2799 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2802 if (banks[1] & MCI_STATUS_VAL)
2803 mce->status |= MCI_STATUS_OVER;
2804 banks[2] = mce->addr;
2805 banks[3] = mce->misc;
2806 vcpu->arch.mcg_status = mce->mcg_status;
2807 banks[1] = mce->status;
2808 kvm_queue_exception(vcpu, MC_VECTOR);
2809 } else if (!(banks[1] & MCI_STATUS_VAL)
2810 || !(banks[1] & MCI_STATUS_UC)) {
2811 if (banks[1] & MCI_STATUS_VAL)
2812 mce->status |= MCI_STATUS_OVER;
2813 banks[2] = mce->addr;
2814 banks[3] = mce->misc;
2815 banks[1] = mce->status;
2817 banks[1] |= MCI_STATUS_OVER;
2821 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2822 struct kvm_vcpu_events *events)
2824 events->exception.injected =
2825 vcpu->arch.exception.pending &&
2826 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2827 events->exception.nr = vcpu->arch.exception.nr;
2828 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2829 events->exception.pad = 0;
2830 events->exception.error_code = vcpu->arch.exception.error_code;
2832 events->interrupt.injected =
2833 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2834 events->interrupt.nr = vcpu->arch.interrupt.nr;
2835 events->interrupt.soft = 0;
2836 events->interrupt.shadow =
2837 kvm_x86_ops->get_interrupt_shadow(vcpu,
2838 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2840 events->nmi.injected = vcpu->arch.nmi_injected;
2841 events->nmi.pending = vcpu->arch.nmi_pending;
2842 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2843 events->nmi.pad = 0;
2845 events->sipi_vector = vcpu->arch.sipi_vector;
2847 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2848 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2849 | KVM_VCPUEVENT_VALID_SHADOW);
2850 memset(&events->reserved, 0, sizeof(events->reserved));
2853 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2854 struct kvm_vcpu_events *events)
2856 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2857 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2858 | KVM_VCPUEVENT_VALID_SHADOW))
2861 vcpu->arch.exception.pending = events->exception.injected;
2862 vcpu->arch.exception.nr = events->exception.nr;
2863 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2864 vcpu->arch.exception.error_code = events->exception.error_code;
2866 vcpu->arch.interrupt.pending = events->interrupt.injected;
2867 vcpu->arch.interrupt.nr = events->interrupt.nr;
2868 vcpu->arch.interrupt.soft = events->interrupt.soft;
2869 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2870 kvm_x86_ops->set_interrupt_shadow(vcpu,
2871 events->interrupt.shadow);
2873 vcpu->arch.nmi_injected = events->nmi.injected;
2874 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2875 vcpu->arch.nmi_pending = events->nmi.pending;
2876 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2878 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2879 vcpu->arch.sipi_vector = events->sipi_vector;
2881 kvm_make_request(KVM_REQ_EVENT, vcpu);
2886 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2887 struct kvm_debugregs *dbgregs)
2889 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2890 dbgregs->dr6 = vcpu->arch.dr6;
2891 dbgregs->dr7 = vcpu->arch.dr7;
2893 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2896 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2897 struct kvm_debugregs *dbgregs)
2902 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2903 vcpu->arch.dr6 = dbgregs->dr6;
2904 vcpu->arch.dr7 = dbgregs->dr7;
2909 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2910 struct kvm_xsave *guest_xsave)
2913 memcpy(guest_xsave->region,
2914 &vcpu->arch.guest_fpu.state->xsave,
2917 memcpy(guest_xsave->region,
2918 &vcpu->arch.guest_fpu.state->fxsave,
2919 sizeof(struct i387_fxsave_struct));
2920 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2925 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2926 struct kvm_xsave *guest_xsave)
2929 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2932 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2933 guest_xsave->region, xstate_size);
2935 if (xstate_bv & ~XSTATE_FPSSE)
2937 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2938 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2943 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2944 struct kvm_xcrs *guest_xcrs)
2946 if (!cpu_has_xsave) {
2947 guest_xcrs->nr_xcrs = 0;
2951 guest_xcrs->nr_xcrs = 1;
2952 guest_xcrs->flags = 0;
2953 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2954 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2957 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2958 struct kvm_xcrs *guest_xcrs)
2965 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2968 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2969 /* Only support XCR0 currently */
2970 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2971 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2972 guest_xcrs->xcrs[0].value);
2980 long kvm_arch_vcpu_ioctl(struct file *filp,
2981 unsigned int ioctl, unsigned long arg)
2983 struct kvm_vcpu *vcpu = filp->private_data;
2984 void __user *argp = (void __user *)arg;
2987 struct kvm_lapic_state *lapic;
2988 struct kvm_xsave *xsave;
2989 struct kvm_xcrs *xcrs;
2995 case KVM_GET_LAPIC: {
2997 if (!vcpu->arch.apic)
2999 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3004 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3008 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3013 case KVM_SET_LAPIC: {
3015 if (!vcpu->arch.apic)
3017 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3022 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
3024 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3030 case KVM_INTERRUPT: {
3031 struct kvm_interrupt irq;
3034 if (copy_from_user(&irq, argp, sizeof irq))
3036 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3043 r = kvm_vcpu_ioctl_nmi(vcpu);
3049 case KVM_SET_CPUID: {
3050 struct kvm_cpuid __user *cpuid_arg = argp;
3051 struct kvm_cpuid cpuid;
3054 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3056 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3061 case KVM_SET_CPUID2: {
3062 struct kvm_cpuid2 __user *cpuid_arg = argp;
3063 struct kvm_cpuid2 cpuid;
3066 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3068 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3069 cpuid_arg->entries);
3074 case KVM_GET_CPUID2: {
3075 struct kvm_cpuid2 __user *cpuid_arg = argp;
3076 struct kvm_cpuid2 cpuid;
3079 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3081 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3082 cpuid_arg->entries);
3086 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3092 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3095 r = msr_io(vcpu, argp, do_set_msr, 0);
3097 case KVM_TPR_ACCESS_REPORTING: {
3098 struct kvm_tpr_access_ctl tac;
3101 if (copy_from_user(&tac, argp, sizeof tac))
3103 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3107 if (copy_to_user(argp, &tac, sizeof tac))
3112 case KVM_SET_VAPIC_ADDR: {
3113 struct kvm_vapic_addr va;
3116 if (!irqchip_in_kernel(vcpu->kvm))
3119 if (copy_from_user(&va, argp, sizeof va))
3122 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3125 case KVM_X86_SETUP_MCE: {
3129 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3131 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3134 case KVM_X86_SET_MCE: {
3135 struct kvm_x86_mce mce;
3138 if (copy_from_user(&mce, argp, sizeof mce))
3140 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3143 case KVM_GET_VCPU_EVENTS: {
3144 struct kvm_vcpu_events events;
3146 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3149 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3154 case KVM_SET_VCPU_EVENTS: {
3155 struct kvm_vcpu_events events;
3158 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3161 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3164 case KVM_GET_DEBUGREGS: {
3165 struct kvm_debugregs dbgregs;
3167 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3170 if (copy_to_user(argp, &dbgregs,
3171 sizeof(struct kvm_debugregs)))
3176 case KVM_SET_DEBUGREGS: {
3177 struct kvm_debugregs dbgregs;
3180 if (copy_from_user(&dbgregs, argp,
3181 sizeof(struct kvm_debugregs)))
3184 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3187 case KVM_GET_XSAVE: {
3188 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3193 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3196 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3201 case KVM_SET_XSAVE: {
3202 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3208 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3211 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3214 case KVM_GET_XCRS: {
3215 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3220 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3223 if (copy_to_user(argp, u.xcrs,
3224 sizeof(struct kvm_xcrs)))
3229 case KVM_SET_XCRS: {
3230 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3236 if (copy_from_user(u.xcrs, argp,
3237 sizeof(struct kvm_xcrs)))
3240 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3243 case KVM_SET_TSC_KHZ: {
3247 if (!kvm_has_tsc_control)
3250 user_tsc_khz = (u32)arg;
3252 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3255 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3260 case KVM_GET_TSC_KHZ: {
3262 if (check_tsc_unstable())
3265 r = vcpu_tsc_khz(vcpu);
3277 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3281 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3283 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3287 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3290 kvm->arch.ept_identity_map_addr = ident_addr;
3294 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3295 u32 kvm_nr_mmu_pages)
3297 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3300 mutex_lock(&kvm->slots_lock);
3301 spin_lock(&kvm->mmu_lock);
3303 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3304 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3306 spin_unlock(&kvm->mmu_lock);
3307 mutex_unlock(&kvm->slots_lock);
3311 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3313 return kvm->arch.n_max_mmu_pages;
3316 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3321 switch (chip->chip_id) {
3322 case KVM_IRQCHIP_PIC_MASTER:
3323 memcpy(&chip->chip.pic,
3324 &pic_irqchip(kvm)->pics[0],
3325 sizeof(struct kvm_pic_state));
3327 case KVM_IRQCHIP_PIC_SLAVE:
3328 memcpy(&chip->chip.pic,
3329 &pic_irqchip(kvm)->pics[1],
3330 sizeof(struct kvm_pic_state));
3332 case KVM_IRQCHIP_IOAPIC:
3333 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3342 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3347 switch (chip->chip_id) {
3348 case KVM_IRQCHIP_PIC_MASTER:
3349 spin_lock(&pic_irqchip(kvm)->lock);
3350 memcpy(&pic_irqchip(kvm)->pics[0],
3352 sizeof(struct kvm_pic_state));
3353 spin_unlock(&pic_irqchip(kvm)->lock);
3355 case KVM_IRQCHIP_PIC_SLAVE:
3356 spin_lock(&pic_irqchip(kvm)->lock);
3357 memcpy(&pic_irqchip(kvm)->pics[1],
3359 sizeof(struct kvm_pic_state));
3360 spin_unlock(&pic_irqchip(kvm)->lock);
3362 case KVM_IRQCHIP_IOAPIC:
3363 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3369 kvm_pic_update_irq(pic_irqchip(kvm));
3373 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3377 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3378 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3379 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3383 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3387 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3388 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3389 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3390 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3394 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3398 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3399 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3400 sizeof(ps->channels));
3401 ps->flags = kvm->arch.vpit->pit_state.flags;
3402 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3403 memset(&ps->reserved, 0, sizeof(ps->reserved));
3407 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3409 int r = 0, start = 0;
3410 u32 prev_legacy, cur_legacy;
3411 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3412 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3413 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3414 if (!prev_legacy && cur_legacy)
3416 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3417 sizeof(kvm->arch.vpit->pit_state.channels));
3418 kvm->arch.vpit->pit_state.flags = ps->flags;
3419 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3420 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3424 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3425 struct kvm_reinject_control *control)
3427 if (!kvm->arch.vpit)
3429 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3430 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3431 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3436 * Get (and clear) the dirty memory log for a memory slot.
3438 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3439 struct kvm_dirty_log *log)
3442 struct kvm_memory_slot *memslot;
3444 unsigned long is_dirty = 0;
3446 mutex_lock(&kvm->slots_lock);
3449 if (log->slot >= KVM_MEMORY_SLOTS)
3452 memslot = &kvm->memslots->memslots[log->slot];
3454 if (!memslot->dirty_bitmap)
3457 n = kvm_dirty_bitmap_bytes(memslot);
3459 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3460 is_dirty = memslot->dirty_bitmap[i];
3462 /* If nothing is dirty, don't bother messing with page tables. */
3464 struct kvm_memslots *slots, *old_slots;
3465 unsigned long *dirty_bitmap;
3467 dirty_bitmap = memslot->dirty_bitmap_head;
3468 if (memslot->dirty_bitmap == dirty_bitmap)
3469 dirty_bitmap += n / sizeof(long);
3470 memset(dirty_bitmap, 0, n);
3473 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3476 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3477 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3478 slots->generation++;
3480 old_slots = kvm->memslots;
3481 rcu_assign_pointer(kvm->memslots, slots);
3482 synchronize_srcu_expedited(&kvm->srcu);
3483 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3486 spin_lock(&kvm->mmu_lock);
3487 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3488 spin_unlock(&kvm->mmu_lock);
3491 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3495 if (clear_user(log->dirty_bitmap, n))
3501 mutex_unlock(&kvm->slots_lock);
3505 long kvm_arch_vm_ioctl(struct file *filp,
3506 unsigned int ioctl, unsigned long arg)
3508 struct kvm *kvm = filp->private_data;
3509 void __user *argp = (void __user *)arg;
3512 * This union makes it completely explicit to gcc-3.x
3513 * that these two variables' stack usage should be
3514 * combined, not added together.
3517 struct kvm_pit_state ps;
3518 struct kvm_pit_state2 ps2;
3519 struct kvm_pit_config pit_config;
3523 case KVM_SET_TSS_ADDR:
3524 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3528 case KVM_SET_IDENTITY_MAP_ADDR: {
3532 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3534 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3539 case KVM_SET_NR_MMU_PAGES:
3540 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3544 case KVM_GET_NR_MMU_PAGES:
3545 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3547 case KVM_CREATE_IRQCHIP: {
3548 struct kvm_pic *vpic;
3550 mutex_lock(&kvm->lock);
3553 goto create_irqchip_unlock;
3555 vpic = kvm_create_pic(kvm);
3557 r = kvm_ioapic_init(kvm);
3559 mutex_lock(&kvm->slots_lock);
3560 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3562 mutex_unlock(&kvm->slots_lock);
3564 goto create_irqchip_unlock;
3567 goto create_irqchip_unlock;
3569 kvm->arch.vpic = vpic;
3571 r = kvm_setup_default_irq_routing(kvm);
3573 mutex_lock(&kvm->slots_lock);
3574 mutex_lock(&kvm->irq_lock);
3575 kvm_ioapic_destroy(kvm);
3576 kvm_destroy_pic(kvm);
3577 mutex_unlock(&kvm->irq_lock);
3578 mutex_unlock(&kvm->slots_lock);
3580 create_irqchip_unlock:
3581 mutex_unlock(&kvm->lock);
3584 case KVM_CREATE_PIT:
3585 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3587 case KVM_CREATE_PIT2:
3589 if (copy_from_user(&u.pit_config, argp,
3590 sizeof(struct kvm_pit_config)))
3593 mutex_lock(&kvm->slots_lock);
3596 goto create_pit_unlock;
3598 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3602 mutex_unlock(&kvm->slots_lock);
3604 case KVM_IRQ_LINE_STATUS:
3605 case KVM_IRQ_LINE: {
3606 struct kvm_irq_level irq_event;
3609 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3612 if (irqchip_in_kernel(kvm)) {
3614 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3615 irq_event.irq, irq_event.level);
3616 if (ioctl == KVM_IRQ_LINE_STATUS) {
3618 irq_event.status = status;
3619 if (copy_to_user(argp, &irq_event,
3627 case KVM_GET_IRQCHIP: {
3628 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3629 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3635 if (copy_from_user(chip, argp, sizeof *chip))
3636 goto get_irqchip_out;
3638 if (!irqchip_in_kernel(kvm))
3639 goto get_irqchip_out;
3640 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3642 goto get_irqchip_out;
3644 if (copy_to_user(argp, chip, sizeof *chip))
3645 goto get_irqchip_out;
3653 case KVM_SET_IRQCHIP: {
3654 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3655 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3661 if (copy_from_user(chip, argp, sizeof *chip))
3662 goto set_irqchip_out;
3664 if (!irqchip_in_kernel(kvm))
3665 goto set_irqchip_out;
3666 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3668 goto set_irqchip_out;
3678 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3681 if (!kvm->arch.vpit)
3683 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3687 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3694 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3697 if (!kvm->arch.vpit)
3699 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3705 case KVM_GET_PIT2: {
3707 if (!kvm->arch.vpit)
3709 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3713 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3718 case KVM_SET_PIT2: {
3720 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3723 if (!kvm->arch.vpit)
3725 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3731 case KVM_REINJECT_CONTROL: {
3732 struct kvm_reinject_control control;
3734 if (copy_from_user(&control, argp, sizeof(control)))
3736 r = kvm_vm_ioctl_reinject(kvm, &control);
3742 case KVM_XEN_HVM_CONFIG: {
3744 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3745 sizeof(struct kvm_xen_hvm_config)))
3748 if (kvm->arch.xen_hvm_config.flags)
3753 case KVM_SET_CLOCK: {
3754 struct kvm_clock_data user_ns;
3759 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3767 local_irq_disable();
3768 now_ns = get_kernel_ns();
3769 delta = user_ns.clock - now_ns;
3771 kvm->arch.kvmclock_offset = delta;
3774 case KVM_GET_CLOCK: {
3775 struct kvm_clock_data user_ns;
3778 local_irq_disable();
3779 now_ns = get_kernel_ns();
3780 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3783 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3786 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3799 static void kvm_init_msr_list(void)
3804 /* skip the first msrs in the list. KVM-specific */
3805 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3806 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3809 msrs_to_save[j] = msrs_to_save[i];
3812 num_msrs_to_save = j;
3815 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3823 if (!(vcpu->arch.apic &&
3824 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3825 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3836 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3843 if (!(vcpu->arch.apic &&
3844 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3845 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3847 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3857 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3858 struct kvm_segment *var, int seg)
3860 kvm_x86_ops->set_segment(vcpu, var, seg);
3863 void kvm_get_segment(struct kvm_vcpu *vcpu,
3864 struct kvm_segment *var, int seg)
3866 kvm_x86_ops->get_segment(vcpu, var, seg);
3869 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3874 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3877 struct x86_exception exception;
3879 BUG_ON(!mmu_is_nested(vcpu));
3881 /* NPT walks are always user-walks */
3882 access |= PFERR_USER_MASK;
3883 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3888 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3889 struct x86_exception *exception)
3891 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3892 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3895 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3896 struct x86_exception *exception)
3898 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3899 access |= PFERR_FETCH_MASK;
3900 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3903 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3904 struct x86_exception *exception)
3906 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3907 access |= PFERR_WRITE_MASK;
3908 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3911 /* uses this to access any guest's mapped memory without checking CPL */
3912 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3913 struct x86_exception *exception)
3915 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3918 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3919 struct kvm_vcpu *vcpu, u32 access,
3920 struct x86_exception *exception)
3923 int r = X86EMUL_CONTINUE;
3926 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3928 unsigned offset = addr & (PAGE_SIZE-1);
3929 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3932 if (gpa == UNMAPPED_GVA)
3933 return X86EMUL_PROPAGATE_FAULT;
3934 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3936 r = X86EMUL_IO_NEEDED;
3948 /* used for instruction fetching */
3949 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3950 gva_t addr, void *val, unsigned int bytes,
3951 struct x86_exception *exception)
3953 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3954 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3956 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3957 access | PFERR_FETCH_MASK,
3961 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3962 gva_t addr, void *val, unsigned int bytes,
3963 struct x86_exception *exception)
3965 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3966 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3968 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3971 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3973 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3974 gva_t addr, void *val, unsigned int bytes,
3975 struct x86_exception *exception)
3977 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3978 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3981 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3982 gva_t addr, void *val,
3984 struct x86_exception *exception)
3986 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3988 int r = X86EMUL_CONTINUE;
3991 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3994 unsigned offset = addr & (PAGE_SIZE-1);
3995 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3998 if (gpa == UNMAPPED_GVA)
3999 return X86EMUL_PROPAGATE_FAULT;
4000 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4002 r = X86EMUL_IO_NEEDED;
4013 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4015 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4016 gpa_t *gpa, struct x86_exception *exception,
4019 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4021 if (vcpu_match_mmio_gva(vcpu, gva) &&
4022 check_write_user_access(vcpu, write, access,
4023 vcpu->arch.access)) {
4024 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4025 (gva & (PAGE_SIZE - 1));
4026 trace_vcpu_match_mmio(gva, *gpa, write, false);
4031 access |= PFERR_WRITE_MASK;
4033 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4035 if (*gpa == UNMAPPED_GVA)
4038 /* For APIC access vmexit */
4039 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4042 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4043 trace_vcpu_match_mmio(gva, *gpa, write, true);
4050 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4054 struct x86_exception *exception)
4056 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4060 if (vcpu->mmio_read_completed) {
4061 memcpy(val, vcpu->mmio_data, bytes);
4062 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4063 vcpu->mmio_phys_addr, *(u64 *)val);
4064 vcpu->mmio_read_completed = 0;
4065 return X86EMUL_CONTINUE;
4068 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, false);
4071 return X86EMUL_PROPAGATE_FAULT;
4076 if (kvm_read_guest_virt(ctxt, addr, val, bytes, exception)
4077 == X86EMUL_CONTINUE)
4078 return X86EMUL_CONTINUE;
4082 * Is this MMIO handled locally?
4084 handled = vcpu_mmio_read(vcpu, gpa, bytes, val);
4086 if (handled == bytes)
4087 return X86EMUL_CONTINUE;
4093 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4095 vcpu->mmio_needed = 1;
4096 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4097 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
4098 vcpu->mmio_size = bytes;
4099 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
4100 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
4101 vcpu->mmio_index = 0;
4103 return X86EMUL_IO_NEEDED;
4106 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4107 const void *val, int bytes)
4111 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4114 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
4118 static int emulator_write_emulated_onepage(unsigned long addr,
4121 struct x86_exception *exception,
4122 struct kvm_vcpu *vcpu)
4127 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, true);
4130 return X86EMUL_PROPAGATE_FAULT;
4132 /* For APIC access vmexit */
4136 if (emulator_write_phys(vcpu, gpa, val, bytes))
4137 return X86EMUL_CONTINUE;
4140 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4142 * Is this MMIO handled locally?
4144 handled = vcpu_mmio_write(vcpu, gpa, bytes, val);
4145 if (handled == bytes)
4146 return X86EMUL_CONTINUE;
4152 vcpu->mmio_needed = 1;
4153 memcpy(vcpu->mmio_data, val, bytes);
4154 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4155 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
4156 vcpu->mmio_size = bytes;
4157 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
4158 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
4159 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
4160 vcpu->mmio_index = 0;
4162 return X86EMUL_CONTINUE;
4165 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4169 struct x86_exception *exception)
4171 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4173 /* Crossing a page boundary? */
4174 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4177 now = -addr & ~PAGE_MASK;
4178 rc = emulator_write_emulated_onepage(addr, val, now, exception,
4180 if (rc != X86EMUL_CONTINUE)
4186 return emulator_write_emulated_onepage(addr, val, bytes, exception,
4190 #define CMPXCHG_TYPE(t, ptr, old, new) \
4191 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4193 #ifdef CONFIG_X86_64
4194 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4196 # define CMPXCHG64(ptr, old, new) \
4197 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4200 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4205 struct x86_exception *exception)
4207 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4213 /* guests cmpxchg8b have to be emulated atomically */
4214 if (bytes > 8 || (bytes & (bytes - 1)))
4217 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4219 if (gpa == UNMAPPED_GVA ||
4220 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4223 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4226 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4227 if (is_error_page(page)) {
4228 kvm_release_page_clean(page);
4232 kaddr = kmap_atomic(page, KM_USER0);
4233 kaddr += offset_in_page(gpa);
4236 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4239 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4242 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4245 exchanged = CMPXCHG64(kaddr, old, new);
4250 kunmap_atomic(kaddr, KM_USER0);
4251 kvm_release_page_dirty(page);
4254 return X86EMUL_CMPXCHG_FAILED;
4256 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4258 return X86EMUL_CONTINUE;
4261 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4263 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4266 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4268 /* TODO: String I/O for in kernel device */
4271 if (vcpu->arch.pio.in)
4272 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4273 vcpu->arch.pio.size, pd);
4275 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4276 vcpu->arch.pio.port, vcpu->arch.pio.size,
4282 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4283 int size, unsigned short port, void *val,
4286 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4288 if (vcpu->arch.pio.count)
4291 trace_kvm_pio(0, port, size, count);
4293 vcpu->arch.pio.port = port;
4294 vcpu->arch.pio.in = 1;
4295 vcpu->arch.pio.count = count;
4296 vcpu->arch.pio.size = size;
4298 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4300 memcpy(val, vcpu->arch.pio_data, size * count);
4301 vcpu->arch.pio.count = 0;
4305 vcpu->run->exit_reason = KVM_EXIT_IO;
4306 vcpu->run->io.direction = KVM_EXIT_IO_IN;
4307 vcpu->run->io.size = size;
4308 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4309 vcpu->run->io.count = count;
4310 vcpu->run->io.port = port;
4315 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4316 int size, unsigned short port,
4317 const void *val, unsigned int count)
4319 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4321 trace_kvm_pio(1, port, size, count);
4323 vcpu->arch.pio.port = port;
4324 vcpu->arch.pio.in = 0;
4325 vcpu->arch.pio.count = count;
4326 vcpu->arch.pio.size = size;
4328 memcpy(vcpu->arch.pio_data, val, size * count);
4330 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4331 vcpu->arch.pio.count = 0;
4335 vcpu->run->exit_reason = KVM_EXIT_IO;
4336 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4337 vcpu->run->io.size = size;
4338 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4339 vcpu->run->io.count = count;
4340 vcpu->run->io.port = port;
4345 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4347 return kvm_x86_ops->get_segment_base(vcpu, seg);
4350 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4352 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4355 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4357 if (!need_emulate_wbinvd(vcpu))
4358 return X86EMUL_CONTINUE;
4360 if (kvm_x86_ops->has_wbinvd_exit()) {
4361 int cpu = get_cpu();
4363 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4364 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4365 wbinvd_ipi, NULL, 1);
4367 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4370 return X86EMUL_CONTINUE;
4372 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4374 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4376 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4379 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4381 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4384 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4387 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4390 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4392 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4395 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4397 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4398 unsigned long value;
4402 value = kvm_read_cr0(vcpu);
4405 value = vcpu->arch.cr2;
4408 value = kvm_read_cr3(vcpu);
4411 value = kvm_read_cr4(vcpu);
4414 value = kvm_get_cr8(vcpu);
4417 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4424 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4426 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4431 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4434 vcpu->arch.cr2 = val;
4437 res = kvm_set_cr3(vcpu, val);
4440 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4443 res = kvm_set_cr8(vcpu, val);
4446 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4453 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4455 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4458 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4460 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4463 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4465 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4468 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4470 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4473 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4475 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4478 static unsigned long emulator_get_cached_segment_base(
4479 struct x86_emulate_ctxt *ctxt, int seg)
4481 return get_segment_base(emul_to_vcpu(ctxt), seg);
4484 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4485 struct desc_struct *desc, u32 *base3,
4488 struct kvm_segment var;
4490 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4491 *selector = var.selector;
4498 set_desc_limit(desc, var.limit);
4499 set_desc_base(desc, (unsigned long)var.base);
4500 #ifdef CONFIG_X86_64
4502 *base3 = var.base >> 32;
4504 desc->type = var.type;
4506 desc->dpl = var.dpl;
4507 desc->p = var.present;
4508 desc->avl = var.avl;
4516 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4517 struct desc_struct *desc, u32 base3,
4520 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4521 struct kvm_segment var;
4523 var.selector = selector;
4524 var.base = get_desc_base(desc);
4525 #ifdef CONFIG_X86_64
4526 var.base |= ((u64)base3) << 32;
4528 var.limit = get_desc_limit(desc);
4530 var.limit = (var.limit << 12) | 0xfff;
4531 var.type = desc->type;
4532 var.present = desc->p;
4533 var.dpl = desc->dpl;
4538 var.avl = desc->avl;
4539 var.present = desc->p;
4540 var.unusable = !var.present;
4543 kvm_set_segment(vcpu, &var, seg);
4547 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4548 u32 msr_index, u64 *pdata)
4550 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4553 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4554 u32 msr_index, u64 data)
4556 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4559 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4561 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4564 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4567 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4569 * CR0.TS may reference the host fpu state, not the guest fpu state,
4570 * so it may be clear at this point.
4575 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4580 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4581 struct x86_instruction_info *info,
4582 enum x86_intercept_stage stage)
4584 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4587 static struct x86_emulate_ops emulate_ops = {
4588 .read_std = kvm_read_guest_virt_system,
4589 .write_std = kvm_write_guest_virt_system,
4590 .fetch = kvm_fetch_guest_virt,
4591 .read_emulated = emulator_read_emulated,
4592 .write_emulated = emulator_write_emulated,
4593 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4594 .invlpg = emulator_invlpg,
4595 .pio_in_emulated = emulator_pio_in_emulated,
4596 .pio_out_emulated = emulator_pio_out_emulated,
4597 .get_segment = emulator_get_segment,
4598 .set_segment = emulator_set_segment,
4599 .get_cached_segment_base = emulator_get_cached_segment_base,
4600 .get_gdt = emulator_get_gdt,
4601 .get_idt = emulator_get_idt,
4602 .set_gdt = emulator_set_gdt,
4603 .set_idt = emulator_set_idt,
4604 .get_cr = emulator_get_cr,
4605 .set_cr = emulator_set_cr,
4606 .cpl = emulator_get_cpl,
4607 .get_dr = emulator_get_dr,
4608 .set_dr = emulator_set_dr,
4609 .set_msr = emulator_set_msr,
4610 .get_msr = emulator_get_msr,
4611 .halt = emulator_halt,
4612 .wbinvd = emulator_wbinvd,
4613 .fix_hypercall = emulator_fix_hypercall,
4614 .get_fpu = emulator_get_fpu,
4615 .put_fpu = emulator_put_fpu,
4616 .intercept = emulator_intercept,
4619 static void cache_all_regs(struct kvm_vcpu *vcpu)
4621 kvm_register_read(vcpu, VCPU_REGS_RAX);
4622 kvm_register_read(vcpu, VCPU_REGS_RSP);
4623 kvm_register_read(vcpu, VCPU_REGS_RIP);
4624 vcpu->arch.regs_dirty = ~0;
4627 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4629 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4631 * an sti; sti; sequence only disable interrupts for the first
4632 * instruction. So, if the last instruction, be it emulated or
4633 * not, left the system with the INT_STI flag enabled, it
4634 * means that the last instruction is an sti. We should not
4635 * leave the flag on in this case. The same goes for mov ss
4637 if (!(int_shadow & mask))
4638 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4641 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4643 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4644 if (ctxt->exception.vector == PF_VECTOR)
4645 kvm_propagate_fault(vcpu, &ctxt->exception);
4646 else if (ctxt->exception.error_code_valid)
4647 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4648 ctxt->exception.error_code);
4650 kvm_queue_exception(vcpu, ctxt->exception.vector);
4653 static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
4654 const unsigned long *regs)
4656 memset(&ctxt->twobyte, 0,
4657 (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4658 memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
4660 ctxt->fetch.start = 0;
4661 ctxt->fetch.end = 0;
4662 ctxt->io_read.pos = 0;
4663 ctxt->io_read.end = 0;
4664 ctxt->mem_read.pos = 0;
4665 ctxt->mem_read.end = 0;
4668 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4670 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4674 * TODO: fix emulate.c to use guest_read/write_register
4675 * instead of direct ->regs accesses, can save hundred cycles
4676 * on Intel for instructions that don't read/change RSP, for
4679 cache_all_regs(vcpu);
4681 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4683 ctxt->eflags = kvm_get_rflags(vcpu);
4684 ctxt->eip = kvm_rip_read(vcpu);
4685 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4686 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4687 cs_l ? X86EMUL_MODE_PROT64 :
4688 cs_db ? X86EMUL_MODE_PROT32 :
4689 X86EMUL_MODE_PROT16;
4690 ctxt->guest_mode = is_guest_mode(vcpu);
4692 init_decode_cache(ctxt, vcpu->arch.regs);
4693 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4696 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4698 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4701 init_emulate_ctxt(vcpu);
4705 ctxt->_eip = ctxt->eip + inc_eip;
4706 ret = emulate_int_real(ctxt, irq);
4708 if (ret != X86EMUL_CONTINUE)
4709 return EMULATE_FAIL;
4711 ctxt->eip = ctxt->_eip;
4712 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4713 kvm_rip_write(vcpu, ctxt->eip);
4714 kvm_set_rflags(vcpu, ctxt->eflags);
4716 if (irq == NMI_VECTOR)
4717 vcpu->arch.nmi_pending = false;
4719 vcpu->arch.interrupt.pending = false;
4721 return EMULATE_DONE;
4723 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4725 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4727 int r = EMULATE_DONE;
4729 ++vcpu->stat.insn_emulation_fail;
4730 trace_kvm_emulate_insn_failed(vcpu);
4731 if (!is_guest_mode(vcpu)) {
4732 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4733 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4734 vcpu->run->internal.ndata = 0;
4737 kvm_queue_exception(vcpu, UD_VECTOR);
4742 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4750 * if emulation was due to access to shadowed page table
4751 * and it failed try to unshadow page and re-entetr the
4752 * guest to let CPU execute the instruction.
4754 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4757 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4759 if (gpa == UNMAPPED_GVA)
4760 return true; /* let cpu generate fault */
4762 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4768 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4775 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4776 bool writeback = true;
4778 kvm_clear_exception_queue(vcpu);
4780 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4781 init_emulate_ctxt(vcpu);
4782 ctxt->interruptibility = 0;
4783 ctxt->have_exception = false;
4784 ctxt->perm_ok = false;
4786 ctxt->only_vendor_specific_insn
4787 = emulation_type & EMULTYPE_TRAP_UD;
4789 r = x86_decode_insn(ctxt, insn, insn_len);
4791 trace_kvm_emulate_insn_start(vcpu);
4792 ++vcpu->stat.insn_emulation;
4794 if (emulation_type & EMULTYPE_TRAP_UD)
4795 return EMULATE_FAIL;
4796 if (reexecute_instruction(vcpu, cr2))
4797 return EMULATE_DONE;
4798 if (emulation_type & EMULTYPE_SKIP)
4799 return EMULATE_FAIL;
4800 return handle_emulation_failure(vcpu);
4804 if (emulation_type & EMULTYPE_SKIP) {
4805 kvm_rip_write(vcpu, ctxt->_eip);
4806 return EMULATE_DONE;
4809 /* this is needed for vmware backdoor interface to work since it
4810 changes registers values during IO operation */
4811 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4812 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4813 memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
4817 r = x86_emulate_insn(ctxt);
4819 if (r == EMULATION_INTERCEPTED)
4820 return EMULATE_DONE;
4822 if (r == EMULATION_FAILED) {
4823 if (reexecute_instruction(vcpu, cr2))
4824 return EMULATE_DONE;
4826 return handle_emulation_failure(vcpu);
4829 if (ctxt->have_exception) {
4830 inject_emulated_exception(vcpu);
4832 } else if (vcpu->arch.pio.count) {
4833 if (!vcpu->arch.pio.in)
4834 vcpu->arch.pio.count = 0;
4837 r = EMULATE_DO_MMIO;
4838 } else if (vcpu->mmio_needed) {
4839 if (!vcpu->mmio_is_write)
4841 r = EMULATE_DO_MMIO;
4842 } else if (r == EMULATION_RESTART)
4848 toggle_interruptibility(vcpu, ctxt->interruptibility);
4849 kvm_set_rflags(vcpu, ctxt->eflags);
4850 kvm_make_request(KVM_REQ_EVENT, vcpu);
4851 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4852 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4853 kvm_rip_write(vcpu, ctxt->eip);
4855 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4859 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4861 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4863 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4864 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4865 size, port, &val, 1);
4866 /* do not return to emulator after return from userspace */
4867 vcpu->arch.pio.count = 0;
4870 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4872 static void tsc_bad(void *info)
4874 __this_cpu_write(cpu_tsc_khz, 0);
4877 static void tsc_khz_changed(void *data)
4879 struct cpufreq_freqs *freq = data;
4880 unsigned long khz = 0;
4884 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4885 khz = cpufreq_quick_get(raw_smp_processor_id());
4888 __this_cpu_write(cpu_tsc_khz, khz);
4891 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4894 struct cpufreq_freqs *freq = data;
4896 struct kvm_vcpu *vcpu;
4897 int i, send_ipi = 0;
4900 * We allow guests to temporarily run on slowing clocks,
4901 * provided we notify them after, or to run on accelerating
4902 * clocks, provided we notify them before. Thus time never
4905 * However, we have a problem. We can't atomically update
4906 * the frequency of a given CPU from this function; it is
4907 * merely a notifier, which can be called from any CPU.
4908 * Changing the TSC frequency at arbitrary points in time
4909 * requires a recomputation of local variables related to
4910 * the TSC for each VCPU. We must flag these local variables
4911 * to be updated and be sure the update takes place with the
4912 * new frequency before any guests proceed.
4914 * Unfortunately, the combination of hotplug CPU and frequency
4915 * change creates an intractable locking scenario; the order
4916 * of when these callouts happen is undefined with respect to
4917 * CPU hotplug, and they can race with each other. As such,
4918 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4919 * undefined; you can actually have a CPU frequency change take
4920 * place in between the computation of X and the setting of the
4921 * variable. To protect against this problem, all updates of
4922 * the per_cpu tsc_khz variable are done in an interrupt
4923 * protected IPI, and all callers wishing to update the value
4924 * must wait for a synchronous IPI to complete (which is trivial
4925 * if the caller is on the CPU already). This establishes the
4926 * necessary total order on variable updates.
4928 * Note that because a guest time update may take place
4929 * anytime after the setting of the VCPU's request bit, the
4930 * correct TSC value must be set before the request. However,
4931 * to ensure the update actually makes it to any guest which
4932 * starts running in hardware virtualization between the set
4933 * and the acquisition of the spinlock, we must also ping the
4934 * CPU after setting the request bit.
4938 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4940 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4943 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4945 raw_spin_lock(&kvm_lock);
4946 list_for_each_entry(kvm, &vm_list, vm_list) {
4947 kvm_for_each_vcpu(i, vcpu, kvm) {
4948 if (vcpu->cpu != freq->cpu)
4950 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4951 if (vcpu->cpu != smp_processor_id())
4955 raw_spin_unlock(&kvm_lock);
4957 if (freq->old < freq->new && send_ipi) {
4959 * We upscale the frequency. Must make the guest
4960 * doesn't see old kvmclock values while running with
4961 * the new frequency, otherwise we risk the guest sees
4962 * time go backwards.
4964 * In case we update the frequency for another cpu
4965 * (which might be in guest context) send an interrupt
4966 * to kick the cpu out of guest context. Next time
4967 * guest context is entered kvmclock will be updated,
4968 * so the guest will not see stale values.
4970 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4975 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4976 .notifier_call = kvmclock_cpufreq_notifier
4979 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4980 unsigned long action, void *hcpu)
4982 unsigned int cpu = (unsigned long)hcpu;
4986 case CPU_DOWN_FAILED:
4987 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4989 case CPU_DOWN_PREPARE:
4990 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4996 static struct notifier_block kvmclock_cpu_notifier_block = {
4997 .notifier_call = kvmclock_cpu_notifier,
4998 .priority = -INT_MAX
5001 static void kvm_timer_init(void)
5005 max_tsc_khz = tsc_khz;
5006 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5007 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5008 #ifdef CONFIG_CPU_FREQ
5009 struct cpufreq_policy policy;
5010 memset(&policy, 0, sizeof(policy));
5012 cpufreq_get_policy(&policy, cpu);
5013 if (policy.cpuinfo.max_freq)
5014 max_tsc_khz = policy.cpuinfo.max_freq;
5017 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5018 CPUFREQ_TRANSITION_NOTIFIER);
5020 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5021 for_each_online_cpu(cpu)
5022 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5025 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5027 static int kvm_is_in_guest(void)
5029 return percpu_read(current_vcpu) != NULL;
5032 static int kvm_is_user_mode(void)
5036 if (percpu_read(current_vcpu))
5037 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
5039 return user_mode != 0;
5042 static unsigned long kvm_get_guest_ip(void)
5044 unsigned long ip = 0;
5046 if (percpu_read(current_vcpu))
5047 ip = kvm_rip_read(percpu_read(current_vcpu));
5052 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5053 .is_in_guest = kvm_is_in_guest,
5054 .is_user_mode = kvm_is_user_mode,
5055 .get_guest_ip = kvm_get_guest_ip,
5058 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5060 percpu_write(current_vcpu, vcpu);
5062 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5064 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5066 percpu_write(current_vcpu, NULL);
5068 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5070 static void kvm_set_mmio_spte_mask(void)
5073 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5076 * Set the reserved bits and the present bit of an paging-structure
5077 * entry to generate page fault with PFER.RSV = 1.
5079 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5082 #ifdef CONFIG_X86_64
5084 * If reserved bit is not supported, clear the present bit to disable
5087 if (maxphyaddr == 52)
5091 kvm_mmu_set_mmio_spte_mask(mask);
5094 int kvm_arch_init(void *opaque)
5097 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5100 printk(KERN_ERR "kvm: already loaded the other module\n");
5105 if (!ops->cpu_has_kvm_support()) {
5106 printk(KERN_ERR "kvm: no hardware support\n");
5110 if (ops->disabled_by_bios()) {
5111 printk(KERN_ERR "kvm: disabled by bios\n");
5116 r = kvm_mmu_module_init();
5120 kvm_set_mmio_spte_mask();
5121 kvm_init_msr_list();
5124 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5125 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5129 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5132 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5140 void kvm_arch_exit(void)
5142 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5144 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5145 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5146 CPUFREQ_TRANSITION_NOTIFIER);
5147 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5149 kvm_mmu_module_exit();
5152 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5154 ++vcpu->stat.halt_exits;
5155 if (irqchip_in_kernel(vcpu->kvm)) {
5156 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5159 vcpu->run->exit_reason = KVM_EXIT_HLT;
5163 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5165 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
5168 if (is_long_mode(vcpu))
5171 return a0 | ((gpa_t)a1 << 32);
5174 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5176 u64 param, ingpa, outgpa, ret;
5177 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5178 bool fast, longmode;
5182 * hypercall generates UD from non zero cpl and real mode
5185 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5186 kvm_queue_exception(vcpu, UD_VECTOR);
5190 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5191 longmode = is_long_mode(vcpu) && cs_l == 1;
5194 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5195 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5196 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5197 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5198 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5199 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5201 #ifdef CONFIG_X86_64
5203 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5204 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5205 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5209 code = param & 0xffff;
5210 fast = (param >> 16) & 0x1;
5211 rep_cnt = (param >> 32) & 0xfff;
5212 rep_idx = (param >> 48) & 0xfff;
5214 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5217 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5218 kvm_vcpu_on_spin(vcpu);
5221 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5225 ret = res | (((u64)rep_done & 0xfff) << 32);
5227 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5229 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5230 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5236 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5238 unsigned long nr, a0, a1, a2, a3, ret;
5241 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5242 return kvm_hv_hypercall(vcpu);
5244 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5245 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5246 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5247 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5248 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5250 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5252 if (!is_long_mode(vcpu)) {
5260 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5266 case KVM_HC_VAPIC_POLL_IRQ:
5270 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5277 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5278 ++vcpu->stat.hypercalls;
5281 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5283 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5285 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5286 char instruction[3];
5287 unsigned long rip = kvm_rip_read(vcpu);
5290 * Blow out the MMU to ensure that no other VCPU has an active mapping
5291 * to ensure that the updated hypercall appears atomically across all
5294 kvm_mmu_zap_all(vcpu->kvm);
5296 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5298 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5301 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5303 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5304 int j, nent = vcpu->arch.cpuid_nent;
5306 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5307 /* when no next entry is found, the current entry[i] is reselected */
5308 for (j = i + 1; ; j = (j + 1) % nent) {
5309 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5310 if (ej->function == e->function) {
5311 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5315 return 0; /* silence gcc, even though control never reaches here */
5318 /* find an entry with matching function, matching index (if needed), and that
5319 * should be read next (if it's stateful) */
5320 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5321 u32 function, u32 index)
5323 if (e->function != function)
5325 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5327 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5328 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5333 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5334 u32 function, u32 index)
5337 struct kvm_cpuid_entry2 *best = NULL;
5339 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5340 struct kvm_cpuid_entry2 *e;
5342 e = &vcpu->arch.cpuid_entries[i];
5343 if (is_matching_cpuid_entry(e, function, index)) {
5344 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5345 move_to_next_stateful_cpuid_entry(vcpu, i);
5352 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5354 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5356 struct kvm_cpuid_entry2 *best;
5358 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5359 if (!best || best->eax < 0x80000008)
5361 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5363 return best->eax & 0xff;
5369 * If no match is found, check whether we exceed the vCPU's limit
5370 * and return the content of the highest valid _standard_ leaf instead.
5371 * This is to satisfy the CPUID specification.
5373 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5374 u32 function, u32 index)
5376 struct kvm_cpuid_entry2 *maxlevel;
5378 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5379 if (!maxlevel || maxlevel->eax >= function)
5381 if (function & 0x80000000) {
5382 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5386 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5389 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5391 u32 function, index;
5392 struct kvm_cpuid_entry2 *best;
5394 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5395 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5396 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5397 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5398 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5399 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5400 best = kvm_find_cpuid_entry(vcpu, function, index);
5403 best = check_cpuid_limit(vcpu, function, index);
5406 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5407 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5408 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5409 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5411 kvm_x86_ops->skip_emulated_instruction(vcpu);
5412 trace_kvm_cpuid(function,
5413 kvm_register_read(vcpu, VCPU_REGS_RAX),
5414 kvm_register_read(vcpu, VCPU_REGS_RBX),
5415 kvm_register_read(vcpu, VCPU_REGS_RCX),
5416 kvm_register_read(vcpu, VCPU_REGS_RDX));
5418 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5421 * Check if userspace requested an interrupt window, and that the
5422 * interrupt window is open.
5424 * No need to exit to userspace if we already have an interrupt queued.
5426 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5428 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5429 vcpu->run->request_interrupt_window &&
5430 kvm_arch_interrupt_allowed(vcpu));
5433 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5435 struct kvm_run *kvm_run = vcpu->run;
5437 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5438 kvm_run->cr8 = kvm_get_cr8(vcpu);
5439 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5440 if (irqchip_in_kernel(vcpu->kvm))
5441 kvm_run->ready_for_interrupt_injection = 1;
5443 kvm_run->ready_for_interrupt_injection =
5444 kvm_arch_interrupt_allowed(vcpu) &&
5445 !kvm_cpu_has_interrupt(vcpu) &&
5446 !kvm_event_needs_reinjection(vcpu);
5449 static void vapic_enter(struct kvm_vcpu *vcpu)
5451 struct kvm_lapic *apic = vcpu->arch.apic;
5454 if (!apic || !apic->vapic_addr)
5457 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5459 vcpu->arch.apic->vapic_page = page;
5462 static void vapic_exit(struct kvm_vcpu *vcpu)
5464 struct kvm_lapic *apic = vcpu->arch.apic;
5467 if (!apic || !apic->vapic_addr)
5470 idx = srcu_read_lock(&vcpu->kvm->srcu);
5471 kvm_release_page_dirty(apic->vapic_page);
5472 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5473 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5476 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5480 if (!kvm_x86_ops->update_cr8_intercept)
5483 if (!vcpu->arch.apic)
5486 if (!vcpu->arch.apic->vapic_addr)
5487 max_irr = kvm_lapic_find_highest_irr(vcpu);
5494 tpr = kvm_lapic_get_cr8(vcpu);
5496 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5499 static void inject_pending_event(struct kvm_vcpu *vcpu)
5501 /* try to reinject previous events if any */
5502 if (vcpu->arch.exception.pending) {
5503 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5504 vcpu->arch.exception.has_error_code,
5505 vcpu->arch.exception.error_code);
5506 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5507 vcpu->arch.exception.has_error_code,
5508 vcpu->arch.exception.error_code,
5509 vcpu->arch.exception.reinject);
5513 if (vcpu->arch.nmi_injected) {
5514 kvm_x86_ops->set_nmi(vcpu);
5518 if (vcpu->arch.interrupt.pending) {
5519 kvm_x86_ops->set_irq(vcpu);
5523 /* try to inject new event if pending */
5524 if (vcpu->arch.nmi_pending) {
5525 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5526 vcpu->arch.nmi_pending = false;
5527 vcpu->arch.nmi_injected = true;
5528 kvm_x86_ops->set_nmi(vcpu);
5530 } else if (kvm_cpu_has_interrupt(vcpu)) {
5531 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5532 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5534 kvm_x86_ops->set_irq(vcpu);
5539 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5541 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5542 !vcpu->guest_xcr0_loaded) {
5543 /* kvm_set_xcr() also depends on this */
5544 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5545 vcpu->guest_xcr0_loaded = 1;
5549 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5551 if (vcpu->guest_xcr0_loaded) {
5552 if (vcpu->arch.xcr0 != host_xcr0)
5553 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5554 vcpu->guest_xcr0_loaded = 0;
5558 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5562 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5563 vcpu->run->request_interrupt_window;
5565 if (vcpu->requests) {
5566 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5567 kvm_mmu_unload(vcpu);
5568 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5569 __kvm_migrate_timers(vcpu);
5570 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5571 r = kvm_guest_time_update(vcpu);
5575 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5576 kvm_mmu_sync_roots(vcpu);
5577 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5578 kvm_x86_ops->tlb_flush(vcpu);
5579 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5580 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5584 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5585 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5589 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5590 vcpu->fpu_active = 0;
5591 kvm_x86_ops->fpu_deactivate(vcpu);
5593 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5594 /* Page is swapped out. Do synthetic halt */
5595 vcpu->arch.apf.halted = true;
5599 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5600 record_steal_time(vcpu);
5604 r = kvm_mmu_reload(vcpu);
5609 * An NMI can be injected between local nmi_pending read and
5610 * vcpu->arch.nmi_pending read inside inject_pending_event().
5611 * But in that case, KVM_REQ_EVENT will be set, which makes
5612 * the race described above benign.
5614 nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5616 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5617 inject_pending_event(vcpu);
5619 /* enable NMI/IRQ window open exits if needed */
5621 kvm_x86_ops->enable_nmi_window(vcpu);
5622 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5623 kvm_x86_ops->enable_irq_window(vcpu);
5625 if (kvm_lapic_enabled(vcpu)) {
5626 update_cr8_intercept(vcpu);
5627 kvm_lapic_sync_to_vapic(vcpu);
5633 kvm_x86_ops->prepare_guest_switch(vcpu);
5634 if (vcpu->fpu_active)
5635 kvm_load_guest_fpu(vcpu);
5636 kvm_load_guest_xcr0(vcpu);
5638 vcpu->mode = IN_GUEST_MODE;
5640 /* We should set ->mode before check ->requests,
5641 * see the comment in make_all_cpus_request.
5645 local_irq_disable();
5647 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5648 || need_resched() || signal_pending(current)) {
5649 vcpu->mode = OUTSIDE_GUEST_MODE;
5653 kvm_x86_ops->cancel_injection(vcpu);
5658 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5662 if (unlikely(vcpu->arch.switch_db_regs)) {
5664 set_debugreg(vcpu->arch.eff_db[0], 0);
5665 set_debugreg(vcpu->arch.eff_db[1], 1);
5666 set_debugreg(vcpu->arch.eff_db[2], 2);
5667 set_debugreg(vcpu->arch.eff_db[3], 3);
5670 trace_kvm_entry(vcpu->vcpu_id);
5671 kvm_x86_ops->run(vcpu);
5674 * If the guest has used debug registers, at least dr7
5675 * will be disabled while returning to the host.
5676 * If we don't have active breakpoints in the host, we don't
5677 * care about the messed up debug address registers. But if
5678 * we have some of them active, restore the old state.
5680 if (hw_breakpoint_active())
5681 hw_breakpoint_restore();
5683 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5685 vcpu->mode = OUTSIDE_GUEST_MODE;
5692 * We must have an instruction between local_irq_enable() and
5693 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5694 * the interrupt shadow. The stat.exits increment will do nicely.
5695 * But we need to prevent reordering, hence this barrier():
5703 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5706 * Profile KVM exit RIPs:
5708 if (unlikely(prof_on == KVM_PROFILING)) {
5709 unsigned long rip = kvm_rip_read(vcpu);
5710 profile_hit(KVM_PROFILING, (void *)rip);
5714 kvm_lapic_sync_from_vapic(vcpu);
5716 r = kvm_x86_ops->handle_exit(vcpu);
5722 static int __vcpu_run(struct kvm_vcpu *vcpu)
5725 struct kvm *kvm = vcpu->kvm;
5727 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5728 pr_debug("vcpu %d received sipi with vector # %x\n",
5729 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5730 kvm_lapic_reset(vcpu);
5731 r = kvm_arch_vcpu_reset(vcpu);
5734 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5737 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5742 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5743 !vcpu->arch.apf.halted)
5744 r = vcpu_enter_guest(vcpu);
5746 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5747 kvm_vcpu_block(vcpu);
5748 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5749 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5751 switch(vcpu->arch.mp_state) {
5752 case KVM_MP_STATE_HALTED:
5753 vcpu->arch.mp_state =
5754 KVM_MP_STATE_RUNNABLE;
5755 case KVM_MP_STATE_RUNNABLE:
5756 vcpu->arch.apf.halted = false;
5758 case KVM_MP_STATE_SIPI_RECEIVED:
5769 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5770 if (kvm_cpu_has_pending_timer(vcpu))
5771 kvm_inject_pending_timer_irqs(vcpu);
5773 if (dm_request_for_irq_injection(vcpu)) {
5775 vcpu->run->exit_reason = KVM_EXIT_INTR;
5776 ++vcpu->stat.request_irq_exits;
5779 kvm_check_async_pf_completion(vcpu);
5781 if (signal_pending(current)) {
5783 vcpu->run->exit_reason = KVM_EXIT_INTR;
5784 ++vcpu->stat.signal_exits;
5786 if (need_resched()) {
5787 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5789 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5793 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5800 static int complete_mmio(struct kvm_vcpu *vcpu)
5802 struct kvm_run *run = vcpu->run;
5805 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5808 if (vcpu->mmio_needed) {
5809 vcpu->mmio_needed = 0;
5810 if (!vcpu->mmio_is_write)
5811 memcpy(vcpu->mmio_data + vcpu->mmio_index,
5813 vcpu->mmio_index += 8;
5814 if (vcpu->mmio_index < vcpu->mmio_size) {
5815 run->exit_reason = KVM_EXIT_MMIO;
5816 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5817 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5818 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5819 run->mmio.is_write = vcpu->mmio_is_write;
5820 vcpu->mmio_needed = 1;
5823 if (vcpu->mmio_is_write)
5825 vcpu->mmio_read_completed = 1;
5827 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5828 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5829 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5830 if (r != EMULATE_DONE)
5835 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5840 if (!tsk_used_math(current) && init_fpu(current))
5843 if (vcpu->sigset_active)
5844 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5846 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5847 kvm_vcpu_block(vcpu);
5848 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5853 /* re-sync apic's tpr */
5854 if (!irqchip_in_kernel(vcpu->kvm)) {
5855 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5861 r = complete_mmio(vcpu);
5865 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5866 kvm_register_write(vcpu, VCPU_REGS_RAX,
5867 kvm_run->hypercall.ret);
5869 r = __vcpu_run(vcpu);
5872 post_kvm_run_save(vcpu);
5873 if (vcpu->sigset_active)
5874 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5879 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5881 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5883 * We are here if userspace calls get_regs() in the middle of
5884 * instruction emulation. Registers state needs to be copied
5885 * back from emulation context to vcpu. Usrapace shouldn't do
5886 * that usually, but some bad designed PV devices (vmware
5887 * backdoor interface) need this to work
5889 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5890 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5891 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5893 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5894 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5895 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5896 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5897 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5898 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5899 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5900 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5901 #ifdef CONFIG_X86_64
5902 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5903 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5904 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5905 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5906 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5907 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5908 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5909 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5912 regs->rip = kvm_rip_read(vcpu);
5913 regs->rflags = kvm_get_rflags(vcpu);
5918 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5920 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5921 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5923 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5924 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5925 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5926 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5927 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5928 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5929 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5930 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5931 #ifdef CONFIG_X86_64
5932 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5933 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5934 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5935 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5936 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5937 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5938 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5939 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5942 kvm_rip_write(vcpu, regs->rip);
5943 kvm_set_rflags(vcpu, regs->rflags);
5945 vcpu->arch.exception.pending = false;
5947 kvm_make_request(KVM_REQ_EVENT, vcpu);
5952 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5954 struct kvm_segment cs;
5956 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5960 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5962 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5963 struct kvm_sregs *sregs)
5967 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5968 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5969 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5970 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5971 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5972 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5974 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5975 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5977 kvm_x86_ops->get_idt(vcpu, &dt);
5978 sregs->idt.limit = dt.size;
5979 sregs->idt.base = dt.address;
5980 kvm_x86_ops->get_gdt(vcpu, &dt);
5981 sregs->gdt.limit = dt.size;
5982 sregs->gdt.base = dt.address;
5984 sregs->cr0 = kvm_read_cr0(vcpu);
5985 sregs->cr2 = vcpu->arch.cr2;
5986 sregs->cr3 = kvm_read_cr3(vcpu);
5987 sregs->cr4 = kvm_read_cr4(vcpu);
5988 sregs->cr8 = kvm_get_cr8(vcpu);
5989 sregs->efer = vcpu->arch.efer;
5990 sregs->apic_base = kvm_get_apic_base(vcpu);
5992 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5994 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5995 set_bit(vcpu->arch.interrupt.nr,
5996 (unsigned long *)sregs->interrupt_bitmap);
6001 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6002 struct kvm_mp_state *mp_state)
6004 mp_state->mp_state = vcpu->arch.mp_state;
6008 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6009 struct kvm_mp_state *mp_state)
6011 vcpu->arch.mp_state = mp_state->mp_state;
6012 kvm_make_request(KVM_REQ_EVENT, vcpu);
6016 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
6017 bool has_error_code, u32 error_code)
6019 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6022 init_emulate_ctxt(vcpu);
6024 ret = emulator_task_switch(ctxt, tss_selector, reason,
6025 has_error_code, error_code);
6028 return EMULATE_FAIL;
6030 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
6031 kvm_rip_write(vcpu, ctxt->eip);
6032 kvm_set_rflags(vcpu, ctxt->eflags);
6033 kvm_make_request(KVM_REQ_EVENT, vcpu);
6034 return EMULATE_DONE;
6036 EXPORT_SYMBOL_GPL(kvm_task_switch);
6038 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6039 struct kvm_sregs *sregs)
6041 int mmu_reset_needed = 0;
6042 int pending_vec, max_bits, idx;
6045 dt.size = sregs->idt.limit;
6046 dt.address = sregs->idt.base;
6047 kvm_x86_ops->set_idt(vcpu, &dt);
6048 dt.size = sregs->gdt.limit;
6049 dt.address = sregs->gdt.base;
6050 kvm_x86_ops->set_gdt(vcpu, &dt);
6052 vcpu->arch.cr2 = sregs->cr2;
6053 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6054 vcpu->arch.cr3 = sregs->cr3;
6055 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6057 kvm_set_cr8(vcpu, sregs->cr8);
6059 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6060 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6061 kvm_set_apic_base(vcpu, sregs->apic_base);
6063 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6064 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6065 vcpu->arch.cr0 = sregs->cr0;
6067 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6068 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6069 if (sregs->cr4 & X86_CR4_OSXSAVE)
6072 idx = srcu_read_lock(&vcpu->kvm->srcu);
6073 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6074 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6075 mmu_reset_needed = 1;
6077 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6079 if (mmu_reset_needed)
6080 kvm_mmu_reset_context(vcpu);
6082 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
6083 pending_vec = find_first_bit(
6084 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6085 if (pending_vec < max_bits) {
6086 kvm_queue_interrupt(vcpu, pending_vec, false);
6087 pr_debug("Set back pending irq %d\n", pending_vec);
6090 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6091 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6092 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6093 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6094 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6095 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6097 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6098 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6100 update_cr8_intercept(vcpu);
6102 /* Older userspace won't unhalt the vcpu on reset. */
6103 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6104 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6106 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6108 kvm_make_request(KVM_REQ_EVENT, vcpu);
6113 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6114 struct kvm_guest_debug *dbg)
6116 unsigned long rflags;
6119 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6121 if (vcpu->arch.exception.pending)
6123 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6124 kvm_queue_exception(vcpu, DB_VECTOR);
6126 kvm_queue_exception(vcpu, BP_VECTOR);
6130 * Read rflags as long as potentially injected trace flags are still
6133 rflags = kvm_get_rflags(vcpu);
6135 vcpu->guest_debug = dbg->control;
6136 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6137 vcpu->guest_debug = 0;
6139 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6140 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6141 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6142 vcpu->arch.switch_db_regs =
6143 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
6145 for (i = 0; i < KVM_NR_DB_REGS; i++)
6146 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6147 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
6150 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6151 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6152 get_segment_base(vcpu, VCPU_SREG_CS);
6155 * Trigger an rflags update that will inject or remove the trace
6158 kvm_set_rflags(vcpu, rflags);
6160 kvm_x86_ops->set_guest_debug(vcpu, dbg);
6170 * Translate a guest virtual address to a guest physical address.
6172 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6173 struct kvm_translation *tr)
6175 unsigned long vaddr = tr->linear_address;
6179 idx = srcu_read_lock(&vcpu->kvm->srcu);
6180 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6181 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6182 tr->physical_address = gpa;
6183 tr->valid = gpa != UNMAPPED_GVA;
6190 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6192 struct i387_fxsave_struct *fxsave =
6193 &vcpu->arch.guest_fpu.state->fxsave;
6195 memcpy(fpu->fpr, fxsave->st_space, 128);
6196 fpu->fcw = fxsave->cwd;
6197 fpu->fsw = fxsave->swd;
6198 fpu->ftwx = fxsave->twd;
6199 fpu->last_opcode = fxsave->fop;
6200 fpu->last_ip = fxsave->rip;
6201 fpu->last_dp = fxsave->rdp;
6202 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6207 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6209 struct i387_fxsave_struct *fxsave =
6210 &vcpu->arch.guest_fpu.state->fxsave;
6212 memcpy(fxsave->st_space, fpu->fpr, 128);
6213 fxsave->cwd = fpu->fcw;
6214 fxsave->swd = fpu->fsw;
6215 fxsave->twd = fpu->ftwx;
6216 fxsave->fop = fpu->last_opcode;
6217 fxsave->rip = fpu->last_ip;
6218 fxsave->rdp = fpu->last_dp;
6219 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6224 int fx_init(struct kvm_vcpu *vcpu)
6228 err = fpu_alloc(&vcpu->arch.guest_fpu);
6232 fpu_finit(&vcpu->arch.guest_fpu);
6235 * Ensure guest xcr0 is valid for loading
6237 vcpu->arch.xcr0 = XSTATE_FP;
6239 vcpu->arch.cr0 |= X86_CR0_ET;
6243 EXPORT_SYMBOL_GPL(fx_init);
6245 static void fx_free(struct kvm_vcpu *vcpu)
6247 fpu_free(&vcpu->arch.guest_fpu);
6250 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6252 if (vcpu->guest_fpu_loaded)
6256 * Restore all possible states in the guest,
6257 * and assume host would use all available bits.
6258 * Guest xcr0 would be loaded later.
6260 kvm_put_guest_xcr0(vcpu);
6261 vcpu->guest_fpu_loaded = 1;
6262 unlazy_fpu(current);
6263 fpu_restore_checking(&vcpu->arch.guest_fpu);
6267 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6269 kvm_put_guest_xcr0(vcpu);
6271 if (!vcpu->guest_fpu_loaded)
6274 vcpu->guest_fpu_loaded = 0;
6275 fpu_save_init(&vcpu->arch.guest_fpu);
6276 ++vcpu->stat.fpu_reload;
6277 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6281 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6283 kvmclock_reset(vcpu);
6285 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6287 kvm_x86_ops->vcpu_free(vcpu);
6290 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6293 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6294 printk_once(KERN_WARNING
6295 "kvm: SMP vm created on host with unstable TSC; "
6296 "guest TSC will not be reliable\n");
6297 return kvm_x86_ops->vcpu_create(kvm, id);
6300 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6304 vcpu->arch.mtrr_state.have_fixed = 1;
6306 r = kvm_arch_vcpu_reset(vcpu);
6308 r = kvm_mmu_setup(vcpu);
6314 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6316 vcpu->arch.apf.msr_val = 0;
6319 kvm_mmu_unload(vcpu);
6323 kvm_x86_ops->vcpu_free(vcpu);
6326 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6328 vcpu->arch.nmi_pending = false;
6329 vcpu->arch.nmi_injected = false;
6331 vcpu->arch.switch_db_regs = 0;
6332 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6333 vcpu->arch.dr6 = DR6_FIXED_1;
6334 vcpu->arch.dr7 = DR7_FIXED_1;
6336 kvm_make_request(KVM_REQ_EVENT, vcpu);
6337 vcpu->arch.apf.msr_val = 0;
6338 vcpu->arch.st.msr_val = 0;
6340 kvmclock_reset(vcpu);
6342 kvm_clear_async_pf_completion_queue(vcpu);
6343 kvm_async_pf_hash_reset(vcpu);
6344 vcpu->arch.apf.halted = false;
6346 return kvm_x86_ops->vcpu_reset(vcpu);
6349 int kvm_arch_hardware_enable(void *garbage)
6352 struct kvm_vcpu *vcpu;
6355 kvm_shared_msr_cpu_online();
6356 list_for_each_entry(kvm, &vm_list, vm_list)
6357 kvm_for_each_vcpu(i, vcpu, kvm)
6358 if (vcpu->cpu == smp_processor_id())
6359 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6360 return kvm_x86_ops->hardware_enable(garbage);
6363 void kvm_arch_hardware_disable(void *garbage)
6365 kvm_x86_ops->hardware_disable(garbage);
6366 drop_user_return_notifiers(garbage);
6369 int kvm_arch_hardware_setup(void)
6371 return kvm_x86_ops->hardware_setup();
6374 void kvm_arch_hardware_unsetup(void)
6376 kvm_x86_ops->hardware_unsetup();
6379 void kvm_arch_check_processor_compat(void *rtn)
6381 kvm_x86_ops->check_processor_compatibility(rtn);
6384 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6390 BUG_ON(vcpu->kvm == NULL);
6393 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6394 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6395 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6396 vcpu->arch.mmu.translate_gpa = translate_gpa;
6397 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6398 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6399 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6401 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6403 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6408 vcpu->arch.pio_data = page_address(page);
6410 kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6412 r = kvm_mmu_create(vcpu);
6414 goto fail_free_pio_data;
6416 if (irqchip_in_kernel(kvm)) {
6417 r = kvm_create_lapic(vcpu);
6419 goto fail_mmu_destroy;
6422 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6424 if (!vcpu->arch.mce_banks) {
6426 goto fail_free_lapic;
6428 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6430 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6431 goto fail_free_mce_banks;
6433 kvm_async_pf_hash_reset(vcpu);
6436 fail_free_mce_banks:
6437 kfree(vcpu->arch.mce_banks);
6439 kvm_free_lapic(vcpu);
6441 kvm_mmu_destroy(vcpu);
6443 free_page((unsigned long)vcpu->arch.pio_data);
6448 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6452 kfree(vcpu->arch.mce_banks);
6453 kvm_free_lapic(vcpu);
6454 idx = srcu_read_lock(&vcpu->kvm->srcu);
6455 kvm_mmu_destroy(vcpu);
6456 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6457 free_page((unsigned long)vcpu->arch.pio_data);
6460 int kvm_arch_init_vm(struct kvm *kvm)
6462 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6463 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6465 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6466 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6468 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6473 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6476 kvm_mmu_unload(vcpu);
6480 static void kvm_free_vcpus(struct kvm *kvm)
6483 struct kvm_vcpu *vcpu;
6486 * Unpin any mmu pages first.
6488 kvm_for_each_vcpu(i, vcpu, kvm) {
6489 kvm_clear_async_pf_completion_queue(vcpu);
6490 kvm_unload_vcpu_mmu(vcpu);
6492 kvm_for_each_vcpu(i, vcpu, kvm)
6493 kvm_arch_vcpu_free(vcpu);
6495 mutex_lock(&kvm->lock);
6496 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6497 kvm->vcpus[i] = NULL;
6499 atomic_set(&kvm->online_vcpus, 0);
6500 mutex_unlock(&kvm->lock);
6503 void kvm_arch_sync_events(struct kvm *kvm)
6505 kvm_free_all_assigned_devices(kvm);
6509 void kvm_arch_destroy_vm(struct kvm *kvm)
6511 kvm_iommu_unmap_guest(kvm);
6512 kfree(kvm->arch.vpic);
6513 kfree(kvm->arch.vioapic);
6514 kvm_free_vcpus(kvm);
6515 if (kvm->arch.apic_access_page)
6516 put_page(kvm->arch.apic_access_page);
6517 if (kvm->arch.ept_identity_pagetable)
6518 put_page(kvm->arch.ept_identity_pagetable);
6521 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6522 struct kvm_memory_slot *memslot,
6523 struct kvm_memory_slot old,
6524 struct kvm_userspace_memory_region *mem,
6527 int npages = memslot->npages;
6528 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6530 /* Prevent internal slot pages from being moved by fork()/COW. */
6531 if (memslot->id >= KVM_MEMORY_SLOTS)
6532 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6534 /*To keep backward compatibility with older userspace,
6535 *x86 needs to hanlde !user_alloc case.
6538 if (npages && !old.rmap) {
6539 unsigned long userspace_addr;
6541 down_write(¤t->mm->mmap_sem);
6542 userspace_addr = do_mmap(NULL, 0,
6544 PROT_READ | PROT_WRITE,
6547 up_write(¤t->mm->mmap_sem);
6549 if (IS_ERR((void *)userspace_addr))
6550 return PTR_ERR((void *)userspace_addr);
6552 memslot->userspace_addr = userspace_addr;
6560 void kvm_arch_commit_memory_region(struct kvm *kvm,
6561 struct kvm_userspace_memory_region *mem,
6562 struct kvm_memory_slot old,
6566 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6568 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6571 down_write(¤t->mm->mmap_sem);
6572 ret = do_munmap(current->mm, old.userspace_addr,
6573 old.npages * PAGE_SIZE);
6574 up_write(¤t->mm->mmap_sem);
6577 "kvm_vm_ioctl_set_memory_region: "
6578 "failed to munmap memory\n");
6581 if (!kvm->arch.n_requested_mmu_pages)
6582 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6584 spin_lock(&kvm->mmu_lock);
6586 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6587 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6588 spin_unlock(&kvm->mmu_lock);
6591 void kvm_arch_flush_shadow(struct kvm *kvm)
6593 kvm_mmu_zap_all(kvm);
6594 kvm_reload_remote_mmus(kvm);
6597 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6599 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6600 !vcpu->arch.apf.halted)
6601 || !list_empty_careful(&vcpu->async_pf.done)
6602 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6603 || vcpu->arch.nmi_pending ||
6604 (kvm_arch_interrupt_allowed(vcpu) &&
6605 kvm_cpu_has_interrupt(vcpu));
6608 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6611 int cpu = vcpu->cpu;
6613 if (waitqueue_active(&vcpu->wq)) {
6614 wake_up_interruptible(&vcpu->wq);
6615 ++vcpu->stat.halt_wakeup;
6619 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6620 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6621 smp_send_reschedule(cpu);
6625 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6627 return kvm_x86_ops->interrupt_allowed(vcpu);
6630 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6632 unsigned long current_rip = kvm_rip_read(vcpu) +
6633 get_segment_base(vcpu, VCPU_SREG_CS);
6635 return current_rip == linear_rip;
6637 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6639 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6641 unsigned long rflags;
6643 rflags = kvm_x86_ops->get_rflags(vcpu);
6644 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6645 rflags &= ~X86_EFLAGS_TF;
6648 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6650 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6652 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6653 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6654 rflags |= X86_EFLAGS_TF;
6655 kvm_x86_ops->set_rflags(vcpu, rflags);
6656 kvm_make_request(KVM_REQ_EVENT, vcpu);
6658 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6660 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6664 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6665 is_error_page(work->page))
6668 r = kvm_mmu_reload(vcpu);
6672 if (!vcpu->arch.mmu.direct_map &&
6673 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6676 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6679 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6681 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6684 static inline u32 kvm_async_pf_next_probe(u32 key)
6686 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6689 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6691 u32 key = kvm_async_pf_hash_fn(gfn);
6693 while (vcpu->arch.apf.gfns[key] != ~0)
6694 key = kvm_async_pf_next_probe(key);
6696 vcpu->arch.apf.gfns[key] = gfn;
6699 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6702 u32 key = kvm_async_pf_hash_fn(gfn);
6704 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6705 (vcpu->arch.apf.gfns[key] != gfn &&
6706 vcpu->arch.apf.gfns[key] != ~0); i++)
6707 key = kvm_async_pf_next_probe(key);
6712 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6714 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6717 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6721 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6723 vcpu->arch.apf.gfns[i] = ~0;
6725 j = kvm_async_pf_next_probe(j);
6726 if (vcpu->arch.apf.gfns[j] == ~0)
6728 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6730 * k lies cyclically in ]i,j]
6732 * |....j i.k.| or |.k..j i...|
6734 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6735 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6740 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6743 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6747 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6748 struct kvm_async_pf *work)
6750 struct x86_exception fault;
6752 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6753 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6755 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6756 (vcpu->arch.apf.send_user_only &&
6757 kvm_x86_ops->get_cpl(vcpu) == 0))
6758 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6759 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6760 fault.vector = PF_VECTOR;
6761 fault.error_code_valid = true;
6762 fault.error_code = 0;
6763 fault.nested_page_fault = false;
6764 fault.address = work->arch.token;
6765 kvm_inject_page_fault(vcpu, &fault);
6769 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6770 struct kvm_async_pf *work)
6772 struct x86_exception fault;
6774 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6775 if (is_error_page(work->page))
6776 work->arch.token = ~0; /* broadcast wakeup */
6778 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6780 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6781 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6782 fault.vector = PF_VECTOR;
6783 fault.error_code_valid = true;
6784 fault.error_code = 0;
6785 fault.nested_page_fault = false;
6786 fault.address = work->arch.token;
6787 kvm_inject_page_fault(vcpu, &fault);
6789 vcpu->arch.apf.halted = false;
6792 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6794 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6797 return !kvm_event_needs_reinjection(vcpu) &&
6798 kvm_x86_ops->interrupt_allowed(vcpu);
6801 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6802 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6803 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6804 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6805 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6806 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6807 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6808 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6809 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6810 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6811 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6812 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);