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 <linux/pci.h>
48 #include <trace/events/kvm.h>
50 #define CREATE_TRACE_POINTS
53 #include <asm/debugreg.h>
60 #include <asm/pvclock.h>
61 #include <asm/div64.h>
63 #define MAX_IO_MSRS 256
64 #define KVM_MAX_MCE_BANKS 32
65 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
67 #define emul_to_vcpu(ctxt) \
68 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
71 * - enable syscall per default because its emulated by KVM
72 * - enable LME and LMA per default on 64 bit KVM
76 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
78 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
81 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
82 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
84 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
85 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
86 struct kvm_cpuid_entry2 __user *entries);
87 static void process_nmi(struct kvm_vcpu *vcpu);
89 struct kvm_x86_ops *kvm_x86_ops;
90 EXPORT_SYMBOL_GPL(kvm_x86_ops);
93 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
95 unsigned int min_timer_period_us = 500;
96 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
98 bool kvm_has_tsc_control;
99 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
100 u32 kvm_max_guest_tsc_khz;
101 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
103 #define KVM_NR_SHARED_MSRS 16
105 struct kvm_shared_msrs_global {
107 u32 msrs[KVM_NR_SHARED_MSRS];
110 struct kvm_shared_msrs {
111 struct user_return_notifier urn;
113 struct kvm_shared_msr_values {
116 } values[KVM_NR_SHARED_MSRS];
119 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
120 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
122 struct kvm_stats_debugfs_item debugfs_entries[] = {
123 { "pf_fixed", VCPU_STAT(pf_fixed) },
124 { "pf_guest", VCPU_STAT(pf_guest) },
125 { "tlb_flush", VCPU_STAT(tlb_flush) },
126 { "invlpg", VCPU_STAT(invlpg) },
127 { "exits", VCPU_STAT(exits) },
128 { "io_exits", VCPU_STAT(io_exits) },
129 { "mmio_exits", VCPU_STAT(mmio_exits) },
130 { "signal_exits", VCPU_STAT(signal_exits) },
131 { "irq_window", VCPU_STAT(irq_window_exits) },
132 { "nmi_window", VCPU_STAT(nmi_window_exits) },
133 { "halt_exits", VCPU_STAT(halt_exits) },
134 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
135 { "hypercalls", VCPU_STAT(hypercalls) },
136 { "request_irq", VCPU_STAT(request_irq_exits) },
137 { "irq_exits", VCPU_STAT(irq_exits) },
138 { "host_state_reload", VCPU_STAT(host_state_reload) },
139 { "efer_reload", VCPU_STAT(efer_reload) },
140 { "fpu_reload", VCPU_STAT(fpu_reload) },
141 { "insn_emulation", VCPU_STAT(insn_emulation) },
142 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
143 { "irq_injections", VCPU_STAT(irq_injections) },
144 { "nmi_injections", VCPU_STAT(nmi_injections) },
145 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
146 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
147 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
148 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
149 { "mmu_flooded", VM_STAT(mmu_flooded) },
150 { "mmu_recycled", VM_STAT(mmu_recycled) },
151 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
152 { "mmu_unsync", VM_STAT(mmu_unsync) },
153 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
154 { "largepages", VM_STAT(lpages) },
158 u64 __read_mostly host_xcr0;
160 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
162 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
165 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
166 vcpu->arch.apf.gfns[i] = ~0;
169 static void kvm_on_user_return(struct user_return_notifier *urn)
172 struct kvm_shared_msrs *locals
173 = container_of(urn, struct kvm_shared_msrs, urn);
174 struct kvm_shared_msr_values *values;
176 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
177 values = &locals->values[slot];
178 if (values->host != values->curr) {
179 wrmsrl(shared_msrs_global.msrs[slot], values->host);
180 values->curr = values->host;
183 locals->registered = false;
184 user_return_notifier_unregister(urn);
187 static void shared_msr_update(unsigned slot, u32 msr)
189 struct kvm_shared_msrs *smsr;
192 smsr = &__get_cpu_var(shared_msrs);
193 /* only read, and nobody should modify it at this time,
194 * so don't need lock */
195 if (slot >= shared_msrs_global.nr) {
196 printk(KERN_ERR "kvm: invalid MSR slot!");
199 rdmsrl_safe(msr, &value);
200 smsr->values[slot].host = value;
201 smsr->values[slot].curr = value;
204 void kvm_define_shared_msr(unsigned slot, u32 msr)
206 if (slot >= shared_msrs_global.nr)
207 shared_msrs_global.nr = slot + 1;
208 shared_msrs_global.msrs[slot] = msr;
209 /* we need ensured the shared_msr_global have been updated */
212 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
214 static void kvm_shared_msr_cpu_online(void)
218 for (i = 0; i < shared_msrs_global.nr; ++i)
219 shared_msr_update(i, shared_msrs_global.msrs[i]);
222 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
224 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
226 if (((value ^ smsr->values[slot].curr) & mask) == 0)
228 smsr->values[slot].curr = value;
229 wrmsrl(shared_msrs_global.msrs[slot], value);
230 if (!smsr->registered) {
231 smsr->urn.on_user_return = kvm_on_user_return;
232 user_return_notifier_register(&smsr->urn);
233 smsr->registered = true;
236 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
238 static void drop_user_return_notifiers(void *ignore)
240 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
242 if (smsr->registered)
243 kvm_on_user_return(&smsr->urn);
246 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
248 if (irqchip_in_kernel(vcpu->kvm))
249 return vcpu->arch.apic_base;
251 return vcpu->arch.apic_base;
253 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
255 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
257 /* TODO: reserve bits check */
258 if (irqchip_in_kernel(vcpu->kvm))
259 kvm_lapic_set_base(vcpu, data);
261 vcpu->arch.apic_base = data;
263 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
265 #define EXCPT_BENIGN 0
266 #define EXCPT_CONTRIBUTORY 1
269 static int exception_class(int vector)
279 return EXCPT_CONTRIBUTORY;
286 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
287 unsigned nr, bool has_error, u32 error_code,
293 kvm_make_request(KVM_REQ_EVENT, vcpu);
295 if (!vcpu->arch.exception.pending) {
297 vcpu->arch.exception.pending = true;
298 vcpu->arch.exception.has_error_code = has_error;
299 vcpu->arch.exception.nr = nr;
300 vcpu->arch.exception.error_code = error_code;
301 vcpu->arch.exception.reinject = reinject;
305 /* to check exception */
306 prev_nr = vcpu->arch.exception.nr;
307 if (prev_nr == DF_VECTOR) {
308 /* triple fault -> shutdown */
309 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
312 class1 = exception_class(prev_nr);
313 class2 = exception_class(nr);
314 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
315 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
316 /* generate double fault per SDM Table 5-5 */
317 vcpu->arch.exception.pending = true;
318 vcpu->arch.exception.has_error_code = true;
319 vcpu->arch.exception.nr = DF_VECTOR;
320 vcpu->arch.exception.error_code = 0;
322 /* replace previous exception with a new one in a hope
323 that instruction re-execution will regenerate lost
328 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
330 kvm_multiple_exception(vcpu, nr, false, 0, false);
332 EXPORT_SYMBOL_GPL(kvm_queue_exception);
334 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
336 kvm_multiple_exception(vcpu, nr, false, 0, true);
338 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
340 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
343 kvm_inject_gp(vcpu, 0);
345 kvm_x86_ops->skip_emulated_instruction(vcpu);
347 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
349 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
351 ++vcpu->stat.pf_guest;
352 vcpu->arch.cr2 = fault->address;
353 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
355 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
357 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
359 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
360 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
362 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
365 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
367 atomic_inc(&vcpu->arch.nmi_queued);
368 kvm_make_request(KVM_REQ_NMI, vcpu);
370 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
372 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
374 kvm_multiple_exception(vcpu, nr, true, error_code, false);
376 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
378 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
380 kvm_multiple_exception(vcpu, nr, true, error_code, true);
382 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
385 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
386 * a #GP and return false.
388 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
390 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
392 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
395 EXPORT_SYMBOL_GPL(kvm_require_cpl);
398 * This function will be used to read from the physical memory of the currently
399 * running guest. The difference to kvm_read_guest_page is that this function
400 * can read from guest physical or from the guest's guest physical memory.
402 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
403 gfn_t ngfn, void *data, int offset, int len,
409 ngpa = gfn_to_gpa(ngfn);
410 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
411 if (real_gfn == UNMAPPED_GVA)
414 real_gfn = gpa_to_gfn(real_gfn);
416 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
418 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
420 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
421 void *data, int offset, int len, u32 access)
423 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
424 data, offset, len, access);
428 * Load the pae pdptrs. Return true is they are all valid.
430 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
432 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
433 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
436 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
438 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
439 offset * sizeof(u64), sizeof(pdpte),
440 PFERR_USER_MASK|PFERR_WRITE_MASK);
445 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
446 if (is_present_gpte(pdpte[i]) &&
447 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
454 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
455 __set_bit(VCPU_EXREG_PDPTR,
456 (unsigned long *)&vcpu->arch.regs_avail);
457 __set_bit(VCPU_EXREG_PDPTR,
458 (unsigned long *)&vcpu->arch.regs_dirty);
463 EXPORT_SYMBOL_GPL(load_pdptrs);
465 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
467 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
473 if (is_long_mode(vcpu) || !is_pae(vcpu))
476 if (!test_bit(VCPU_EXREG_PDPTR,
477 (unsigned long *)&vcpu->arch.regs_avail))
480 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
481 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
482 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
483 PFERR_USER_MASK | PFERR_WRITE_MASK);
486 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
492 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
494 unsigned long old_cr0 = kvm_read_cr0(vcpu);
495 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
496 X86_CR0_CD | X86_CR0_NW;
501 if (cr0 & 0xffffffff00000000UL)
505 cr0 &= ~CR0_RESERVED_BITS;
507 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
510 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
513 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
515 if ((vcpu->arch.efer & EFER_LME)) {
520 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
525 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
530 kvm_x86_ops->set_cr0(vcpu, cr0);
532 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
533 kvm_clear_async_pf_completion_queue(vcpu);
534 kvm_async_pf_hash_reset(vcpu);
537 if ((cr0 ^ old_cr0) & update_bits)
538 kvm_mmu_reset_context(vcpu);
541 EXPORT_SYMBOL_GPL(kvm_set_cr0);
543 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
545 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
547 EXPORT_SYMBOL_GPL(kvm_lmsw);
549 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
553 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
554 if (index != XCR_XFEATURE_ENABLED_MASK)
557 if (!(xcr0 & XSTATE_FP))
559 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
561 if (xcr0 & ~host_xcr0)
563 vcpu->arch.xcr0 = xcr0;
564 vcpu->guest_xcr0_loaded = 0;
568 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
570 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
571 __kvm_set_xcr(vcpu, index, xcr)) {
572 kvm_inject_gp(vcpu, 0);
577 EXPORT_SYMBOL_GPL(kvm_set_xcr);
579 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
581 struct kvm_cpuid_entry2 *best;
583 if (!static_cpu_has(X86_FEATURE_XSAVE))
586 best = kvm_find_cpuid_entry(vcpu, 1, 0);
587 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
590 static bool guest_cpuid_has_smep(struct kvm_vcpu *vcpu)
592 struct kvm_cpuid_entry2 *best;
594 best = kvm_find_cpuid_entry(vcpu, 7, 0);
595 return best && (best->ebx & bit(X86_FEATURE_SMEP));
598 static bool guest_cpuid_has_fsgsbase(struct kvm_vcpu *vcpu)
600 struct kvm_cpuid_entry2 *best;
602 best = kvm_find_cpuid_entry(vcpu, 7, 0);
603 return best && (best->ebx & bit(X86_FEATURE_FSGSBASE));
606 static void update_cpuid(struct kvm_vcpu *vcpu)
608 struct kvm_cpuid_entry2 *best;
609 struct kvm_lapic *apic = vcpu->arch.apic;
611 best = kvm_find_cpuid_entry(vcpu, 1, 0);
615 /* Update OSXSAVE bit */
616 if (cpu_has_xsave && best->function == 0x1) {
617 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
618 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
619 best->ecx |= bit(X86_FEATURE_OSXSAVE);
623 if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER))
624 apic->lapic_timer.timer_mode_mask = 3 << 17;
626 apic->lapic_timer.timer_mode_mask = 1 << 17;
630 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
632 unsigned long old_cr4 = kvm_read_cr4(vcpu);
633 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
634 X86_CR4_PAE | X86_CR4_SMEP;
635 if (cr4 & CR4_RESERVED_BITS)
638 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
641 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
644 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
647 if (is_long_mode(vcpu)) {
648 if (!(cr4 & X86_CR4_PAE))
650 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
651 && ((cr4 ^ old_cr4) & pdptr_bits)
652 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
656 if (kvm_x86_ops->set_cr4(vcpu, cr4))
659 if ((cr4 ^ old_cr4) & pdptr_bits)
660 kvm_mmu_reset_context(vcpu);
662 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
667 EXPORT_SYMBOL_GPL(kvm_set_cr4);
669 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
671 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
672 kvm_mmu_sync_roots(vcpu);
673 kvm_mmu_flush_tlb(vcpu);
677 if (is_long_mode(vcpu)) {
678 if (cr3 & CR3_L_MODE_RESERVED_BITS)
682 if (cr3 & CR3_PAE_RESERVED_BITS)
684 if (is_paging(vcpu) &&
685 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
689 * We don't check reserved bits in nonpae mode, because
690 * this isn't enforced, and VMware depends on this.
695 * Does the new cr3 value map to physical memory? (Note, we
696 * catch an invalid cr3 even in real-mode, because it would
697 * cause trouble later on when we turn on paging anyway.)
699 * A real CPU would silently accept an invalid cr3 and would
700 * attempt to use it - with largely undefined (and often hard
701 * to debug) behavior on the guest side.
703 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
705 vcpu->arch.cr3 = cr3;
706 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
707 vcpu->arch.mmu.new_cr3(vcpu);
710 EXPORT_SYMBOL_GPL(kvm_set_cr3);
712 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
714 if (cr8 & CR8_RESERVED_BITS)
716 if (irqchip_in_kernel(vcpu->kvm))
717 kvm_lapic_set_tpr(vcpu, cr8);
719 vcpu->arch.cr8 = cr8;
722 EXPORT_SYMBOL_GPL(kvm_set_cr8);
724 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
726 if (irqchip_in_kernel(vcpu->kvm))
727 return kvm_lapic_get_cr8(vcpu);
729 return vcpu->arch.cr8;
731 EXPORT_SYMBOL_GPL(kvm_get_cr8);
733 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
737 vcpu->arch.db[dr] = val;
738 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
739 vcpu->arch.eff_db[dr] = val;
742 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
746 if (val & 0xffffffff00000000ULL)
748 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
751 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
755 if (val & 0xffffffff00000000ULL)
757 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
758 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
759 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
760 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
768 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
772 res = __kvm_set_dr(vcpu, dr, val);
774 kvm_queue_exception(vcpu, UD_VECTOR);
776 kvm_inject_gp(vcpu, 0);
780 EXPORT_SYMBOL_GPL(kvm_set_dr);
782 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
786 *val = vcpu->arch.db[dr];
789 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
793 *val = vcpu->arch.dr6;
796 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
800 *val = vcpu->arch.dr7;
807 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
809 if (_kvm_get_dr(vcpu, dr, val)) {
810 kvm_queue_exception(vcpu, UD_VECTOR);
815 EXPORT_SYMBOL_GPL(kvm_get_dr);
818 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
819 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
821 * This list is modified at module load time to reflect the
822 * capabilities of the host cpu. This capabilities test skips MSRs that are
823 * kvm-specific. Those are put in the beginning of the list.
826 #define KVM_SAVE_MSRS_BEGIN 9
827 static u32 msrs_to_save[] = {
828 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
829 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
830 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
831 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
832 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
835 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
837 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
840 static unsigned num_msrs_to_save;
842 static u32 emulated_msrs[] = {
843 MSR_IA32_TSCDEADLINE,
844 MSR_IA32_MISC_ENABLE,
849 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
851 u64 old_efer = vcpu->arch.efer;
853 if (efer & efer_reserved_bits)
857 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
860 if (efer & EFER_FFXSR) {
861 struct kvm_cpuid_entry2 *feat;
863 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
864 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
868 if (efer & EFER_SVME) {
869 struct kvm_cpuid_entry2 *feat;
871 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
872 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
877 efer |= vcpu->arch.efer & EFER_LMA;
879 kvm_x86_ops->set_efer(vcpu, efer);
881 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
883 /* Update reserved bits */
884 if ((efer ^ old_efer) & EFER_NX)
885 kvm_mmu_reset_context(vcpu);
890 void kvm_enable_efer_bits(u64 mask)
892 efer_reserved_bits &= ~mask;
894 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
897 * Writes msr value into into the appropriate "register".
898 * Returns 0 on success, non-0 otherwise.
899 * Assumes vcpu_load() was already called.
901 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
906 case MSR_KERNEL_GS_BASE:
909 if (is_noncanonical_address(data))
912 case MSR_IA32_SYSENTER_EIP:
913 case MSR_IA32_SYSENTER_ESP:
915 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
916 * non-canonical address is written on Intel but not on
917 * AMD (which ignores the top 32-bits, because it does
918 * not implement 64-bit SYSENTER).
920 * 64-bit code should hence be able to write a non-canonical
921 * value on AMD. Making the address canonical ensures that
922 * vmentry does not fail on Intel after writing a non-canonical
923 * value, and that something deterministic happens if the guest
924 * invokes 64-bit SYSENTER.
926 data = get_canonical(data);
928 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
930 EXPORT_SYMBOL_GPL(kvm_set_msr);
933 * Adapt set_msr() to msr_io()'s calling convention
935 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
937 return kvm_set_msr(vcpu, index, *data);
940 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
944 struct pvclock_wall_clock wc;
945 struct timespec boot;
950 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
955 ++version; /* first time write, random junk */
959 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
962 * The guest calculates current wall clock time by adding
963 * system time (updated by kvm_guest_time_update below) to the
964 * wall clock specified here. guest system time equals host
965 * system time for us, thus we must fill in host boot time here.
969 wc.sec = boot.tv_sec;
970 wc.nsec = boot.tv_nsec;
971 wc.version = version;
973 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
976 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
979 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
981 uint32_t quotient, remainder;
983 /* Don't try to replace with do_div(), this one calculates
984 * "(dividend << 32) / divisor" */
986 : "=a" (quotient), "=d" (remainder)
987 : "0" (0), "1" (dividend), "r" (divisor) );
991 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
992 s8 *pshift, u32 *pmultiplier)
999 tps64 = base_khz * 1000LL;
1000 scaled64 = scaled_khz * 1000LL;
1001 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1006 tps32 = (uint32_t)tps64;
1007 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1008 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1016 *pmultiplier = div_frac(scaled64, tps32);
1018 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1019 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1022 static inline u64 get_kernel_ns(void)
1026 WARN_ON(preemptible());
1028 monotonic_to_bootbased(&ts);
1029 return timespec_to_ns(&ts);
1032 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1033 unsigned long max_tsc_khz;
1035 static inline int kvm_tsc_changes_freq(void)
1037 int cpu = get_cpu();
1038 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
1039 cpufreq_quick_get(cpu) != 0;
1044 u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
1046 if (vcpu->arch.virtual_tsc_khz)
1047 return vcpu->arch.virtual_tsc_khz;
1049 return __this_cpu_read(cpu_tsc_khz);
1052 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1056 WARN_ON(preemptible());
1057 if (kvm_tsc_changes_freq())
1058 printk_once(KERN_WARNING
1059 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
1060 ret = nsec * vcpu_tsc_khz(vcpu);
1061 do_div(ret, USEC_PER_SEC);
1065 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1067 /* Compute a scale to convert nanoseconds in TSC cycles */
1068 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1069 &vcpu->arch.tsc_catchup_shift,
1070 &vcpu->arch.tsc_catchup_mult);
1073 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1075 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1076 vcpu->arch.tsc_catchup_mult,
1077 vcpu->arch.tsc_catchup_shift);
1078 tsc += vcpu->arch.last_tsc_write;
1082 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1084 struct kvm *kvm = vcpu->kvm;
1085 u64 offset, ns, elapsed;
1086 unsigned long flags;
1089 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1090 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1091 ns = get_kernel_ns();
1092 elapsed = ns - kvm->arch.last_tsc_nsec;
1093 sdiff = data - kvm->arch.last_tsc_write;
1098 * Special case: close write to TSC within 5 seconds of
1099 * another CPU is interpreted as an attempt to synchronize
1100 * The 5 seconds is to accommodate host load / swapping as
1101 * well as any reset of TSC during the boot process.
1103 * In that case, for a reliable TSC, we can match TSC offsets,
1104 * or make a best guest using elapsed value.
1106 if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1107 elapsed < 5ULL * NSEC_PER_SEC) {
1108 if (!check_tsc_unstable()) {
1109 offset = kvm->arch.last_tsc_offset;
1110 pr_debug("kvm: matched tsc offset for %llu\n", data);
1112 u64 delta = nsec_to_cycles(vcpu, elapsed);
1114 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1116 ns = kvm->arch.last_tsc_nsec;
1118 kvm->arch.last_tsc_nsec = ns;
1119 kvm->arch.last_tsc_write = data;
1120 kvm->arch.last_tsc_offset = offset;
1121 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1122 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1124 /* Reset of TSC must disable overshoot protection below */
1125 vcpu->arch.hv_clock.tsc_timestamp = 0;
1126 vcpu->arch.last_tsc_write = data;
1127 vcpu->arch.last_tsc_nsec = ns;
1129 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1131 static int kvm_guest_time_update(struct kvm_vcpu *v)
1133 unsigned long flags;
1134 struct kvm_vcpu_arch *vcpu = &v->arch;
1135 unsigned long this_tsc_khz;
1136 s64 kernel_ns, max_kernel_ns;
1139 /* Keep irq disabled to prevent changes to the clock */
1140 local_irq_save(flags);
1141 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v);
1142 kernel_ns = get_kernel_ns();
1143 this_tsc_khz = vcpu_tsc_khz(v);
1144 if (unlikely(this_tsc_khz == 0)) {
1145 local_irq_restore(flags);
1146 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1151 * We may have to catch up the TSC to match elapsed wall clock
1152 * time for two reasons, even if kvmclock is used.
1153 * 1) CPU could have been running below the maximum TSC rate
1154 * 2) Broken TSC compensation resets the base at each VCPU
1155 * entry to avoid unknown leaps of TSC even when running
1156 * again on the same CPU. This may cause apparent elapsed
1157 * time to disappear, and the guest to stand still or run
1160 if (vcpu->tsc_catchup) {
1161 u64 tsc = compute_guest_tsc(v, kernel_ns);
1162 if (tsc > tsc_timestamp) {
1163 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1164 tsc_timestamp = tsc;
1168 local_irq_restore(flags);
1170 if (!vcpu->pv_time_enabled)
1174 * Time as measured by the TSC may go backwards when resetting the base
1175 * tsc_timestamp. The reason for this is that the TSC resolution is
1176 * higher than the resolution of the other clock scales. Thus, many
1177 * possible measurments of the TSC correspond to one measurement of any
1178 * other clock, and so a spread of values is possible. This is not a
1179 * problem for the computation of the nanosecond clock; with TSC rates
1180 * around 1GHZ, there can only be a few cycles which correspond to one
1181 * nanosecond value, and any path through this code will inevitably
1182 * take longer than that. However, with the kernel_ns value itself,
1183 * the precision may be much lower, down to HZ granularity. If the
1184 * first sampling of TSC against kernel_ns ends in the low part of the
1185 * range, and the second in the high end of the range, we can get:
1187 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1189 * As the sampling errors potentially range in the thousands of cycles,
1190 * it is possible such a time value has already been observed by the
1191 * guest. To protect against this, we must compute the system time as
1192 * observed by the guest and ensure the new system time is greater.
1195 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1196 max_kernel_ns = vcpu->last_guest_tsc -
1197 vcpu->hv_clock.tsc_timestamp;
1198 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1199 vcpu->hv_clock.tsc_to_system_mul,
1200 vcpu->hv_clock.tsc_shift);
1201 max_kernel_ns += vcpu->last_kernel_ns;
1204 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1205 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1206 &vcpu->hv_clock.tsc_shift,
1207 &vcpu->hv_clock.tsc_to_system_mul);
1208 vcpu->hw_tsc_khz = this_tsc_khz;
1211 if (max_kernel_ns > kernel_ns)
1212 kernel_ns = max_kernel_ns;
1214 /* With all the info we got, fill in the values */
1215 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1216 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1217 vcpu->last_kernel_ns = kernel_ns;
1218 vcpu->last_guest_tsc = tsc_timestamp;
1219 vcpu->hv_clock.flags = 0;
1222 * The interface expects us to write an even number signaling that the
1223 * update is finished. Since the guest won't see the intermediate
1224 * state, we just increase by 2 at the end.
1226 vcpu->hv_clock.version += 2;
1228 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1230 sizeof(vcpu->hv_clock));
1234 static bool msr_mtrr_valid(unsigned msr)
1237 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1238 case MSR_MTRRfix64K_00000:
1239 case MSR_MTRRfix16K_80000:
1240 case MSR_MTRRfix16K_A0000:
1241 case MSR_MTRRfix4K_C0000:
1242 case MSR_MTRRfix4K_C8000:
1243 case MSR_MTRRfix4K_D0000:
1244 case MSR_MTRRfix4K_D8000:
1245 case MSR_MTRRfix4K_E0000:
1246 case MSR_MTRRfix4K_E8000:
1247 case MSR_MTRRfix4K_F0000:
1248 case MSR_MTRRfix4K_F8000:
1249 case MSR_MTRRdefType:
1250 case MSR_IA32_CR_PAT:
1258 static bool valid_pat_type(unsigned t)
1260 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1263 static bool valid_mtrr_type(unsigned t)
1265 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1268 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1272 if (!msr_mtrr_valid(msr))
1275 if (msr == MSR_IA32_CR_PAT) {
1276 for (i = 0; i < 8; i++)
1277 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1280 } else if (msr == MSR_MTRRdefType) {
1283 return valid_mtrr_type(data & 0xff);
1284 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1285 for (i = 0; i < 8 ; i++)
1286 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1291 /* variable MTRRs */
1292 return valid_mtrr_type(data & 0xff);
1295 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1297 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1299 if (!mtrr_valid(vcpu, msr, data))
1302 if (msr == MSR_MTRRdefType) {
1303 vcpu->arch.mtrr_state.def_type = data;
1304 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1305 } else if (msr == MSR_MTRRfix64K_00000)
1307 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1308 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1309 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1310 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1311 else if (msr == MSR_IA32_CR_PAT)
1312 vcpu->arch.pat = data;
1313 else { /* Variable MTRRs */
1314 int idx, is_mtrr_mask;
1317 idx = (msr - 0x200) / 2;
1318 is_mtrr_mask = msr - 0x200 - 2 * idx;
1321 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1324 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1328 kvm_mmu_reset_context(vcpu);
1332 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1334 u64 mcg_cap = vcpu->arch.mcg_cap;
1335 unsigned bank_num = mcg_cap & 0xff;
1338 case MSR_IA32_MCG_STATUS:
1339 vcpu->arch.mcg_status = data;
1341 case MSR_IA32_MCG_CTL:
1342 if (!(mcg_cap & MCG_CTL_P))
1344 if (data != 0 && data != ~(u64)0)
1346 vcpu->arch.mcg_ctl = data;
1349 if (msr >= MSR_IA32_MC0_CTL &&
1350 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1351 u32 offset = msr - MSR_IA32_MC0_CTL;
1352 /* only 0 or all 1s can be written to IA32_MCi_CTL
1353 * some Linux kernels though clear bit 10 in bank 4 to
1354 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1355 * this to avoid an uncatched #GP in the guest
1357 if ((offset & 0x3) == 0 &&
1358 data != 0 && (data | (1 << 10)) != ~(u64)0)
1360 vcpu->arch.mce_banks[offset] = data;
1368 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1370 struct kvm *kvm = vcpu->kvm;
1371 int lm = is_long_mode(vcpu);
1372 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1373 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1374 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1375 : kvm->arch.xen_hvm_config.blob_size_32;
1376 u32 page_num = data & ~PAGE_MASK;
1377 u64 page_addr = data & PAGE_MASK;
1382 if (page_num >= blob_size)
1385 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1389 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1391 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1400 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1402 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1405 static bool kvm_hv_msr_partition_wide(u32 msr)
1409 case HV_X64_MSR_GUEST_OS_ID:
1410 case HV_X64_MSR_HYPERCALL:
1418 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1420 struct kvm *kvm = vcpu->kvm;
1423 case HV_X64_MSR_GUEST_OS_ID:
1424 kvm->arch.hv_guest_os_id = data;
1425 /* setting guest os id to zero disables hypercall page */
1426 if (!kvm->arch.hv_guest_os_id)
1427 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1429 case HV_X64_MSR_HYPERCALL: {
1434 /* if guest os id is not set hypercall should remain disabled */
1435 if (!kvm->arch.hv_guest_os_id)
1437 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1438 kvm->arch.hv_hypercall = data;
1441 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1442 addr = gfn_to_hva(kvm, gfn);
1443 if (kvm_is_error_hva(addr))
1445 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1446 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1447 if (__copy_to_user((void __user *)addr, instructions, 4))
1449 kvm->arch.hv_hypercall = data;
1453 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1454 "data 0x%llx\n", msr, data);
1460 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1463 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1466 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1467 vcpu->arch.hv_vapic = data;
1470 addr = gfn_to_hva(vcpu->kvm, data >>
1471 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1472 if (kvm_is_error_hva(addr))
1474 if (__clear_user((void __user *)addr, PAGE_SIZE))
1476 vcpu->arch.hv_vapic = data;
1479 case HV_X64_MSR_EOI:
1480 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1481 case HV_X64_MSR_ICR:
1482 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1483 case HV_X64_MSR_TPR:
1484 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1486 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1487 "data 0x%llx\n", msr, data);
1494 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1496 gpa_t gpa = data & ~0x3f;
1498 /* Bits 2:5 are resrved, Should be zero */
1502 vcpu->arch.apf.msr_val = data;
1504 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1505 kvm_clear_async_pf_completion_queue(vcpu);
1506 kvm_async_pf_hash_reset(vcpu);
1510 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1514 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1515 kvm_async_pf_wakeup_all(vcpu);
1519 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1521 vcpu->arch.pv_time_enabled = false;
1524 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1528 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1531 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1532 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1533 vcpu->arch.st.accum_steal = delta;
1536 static void record_steal_time(struct kvm_vcpu *vcpu)
1538 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1541 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1542 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1545 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1546 vcpu->arch.st.steal.version += 2;
1547 vcpu->arch.st.accum_steal = 0;
1549 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1550 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1553 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1557 return set_efer(vcpu, data);
1559 data &= ~(u64)0x40; /* ignore flush filter disable */
1560 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1562 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1567 case MSR_FAM10H_MMIO_CONF_BASE:
1569 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1574 case MSR_AMD64_NB_CFG:
1576 case MSR_IA32_DEBUGCTLMSR:
1578 /* We support the non-activated case already */
1580 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1581 /* Values other than LBR and BTF are vendor-specific,
1582 thus reserved and should throw a #GP */
1585 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1588 case MSR_IA32_UCODE_REV:
1589 case MSR_IA32_UCODE_WRITE:
1590 case MSR_VM_HSAVE_PA:
1591 case MSR_AMD64_PATCH_LOADER:
1593 case 0x200 ... 0x2ff:
1594 return set_msr_mtrr(vcpu, msr, data);
1595 case MSR_IA32_APICBASE:
1596 kvm_set_apic_base(vcpu, data);
1598 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1599 return kvm_x2apic_msr_write(vcpu, msr, data);
1600 case MSR_IA32_TSCDEADLINE:
1601 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1603 case MSR_IA32_MISC_ENABLE:
1604 vcpu->arch.ia32_misc_enable_msr = data;
1606 case MSR_KVM_WALL_CLOCK_NEW:
1607 case MSR_KVM_WALL_CLOCK:
1608 vcpu->kvm->arch.wall_clock = data;
1609 kvm_write_wall_clock(vcpu->kvm, data);
1611 case MSR_KVM_SYSTEM_TIME_NEW:
1612 case MSR_KVM_SYSTEM_TIME: {
1614 kvmclock_reset(vcpu);
1616 vcpu->arch.time = data;
1617 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1619 /* we verify if the enable bit is set... */
1623 gpa_offset = data & ~(PAGE_MASK | 1);
1625 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1626 &vcpu->arch.pv_time, data & ~1ULL,
1627 sizeof(struct pvclock_vcpu_time_info)))
1628 vcpu->arch.pv_time_enabled = false;
1630 vcpu->arch.pv_time_enabled = true;
1633 case MSR_KVM_ASYNC_PF_EN:
1634 if (kvm_pv_enable_async_pf(vcpu, data))
1637 case MSR_KVM_STEAL_TIME:
1639 if (unlikely(!sched_info_on()))
1642 if (data & KVM_STEAL_RESERVED_MASK)
1645 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1646 data & KVM_STEAL_VALID_BITS,
1647 sizeof(struct kvm_steal_time)))
1650 vcpu->arch.st.msr_val = data;
1652 if (!(data & KVM_MSR_ENABLED))
1655 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1658 accumulate_steal_time(vcpu);
1661 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1665 case MSR_IA32_MCG_CTL:
1666 case MSR_IA32_MCG_STATUS:
1667 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1668 return set_msr_mce(vcpu, msr, data);
1670 /* Performance counters are not protected by a CPUID bit,
1671 * so we should check all of them in the generic path for the sake of
1672 * cross vendor migration.
1673 * Writing a zero into the event select MSRs disables them,
1674 * which we perfectly emulate ;-). Any other value should be at least
1675 * reported, some guests depend on them.
1677 case MSR_P6_EVNTSEL0:
1678 case MSR_P6_EVNTSEL1:
1679 case MSR_K7_EVNTSEL0:
1680 case MSR_K7_EVNTSEL1:
1681 case MSR_K7_EVNTSEL2:
1682 case MSR_K7_EVNTSEL3:
1684 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1685 "0x%x data 0x%llx\n", msr, data);
1687 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1688 * so we ignore writes to make it happy.
1690 case MSR_P6_PERFCTR0:
1691 case MSR_P6_PERFCTR1:
1692 case MSR_K7_PERFCTR0:
1693 case MSR_K7_PERFCTR1:
1694 case MSR_K7_PERFCTR2:
1695 case MSR_K7_PERFCTR3:
1696 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1697 "0x%x data 0x%llx\n", msr, data);
1699 case MSR_K7_CLK_CTL:
1701 * Ignore all writes to this no longer documented MSR.
1702 * Writes are only relevant for old K7 processors,
1703 * all pre-dating SVM, but a recommended workaround from
1704 * AMD for these chips. It is possible to speicify the
1705 * affected processor models on the command line, hence
1706 * the need to ignore the workaround.
1709 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1710 if (kvm_hv_msr_partition_wide(msr)) {
1712 mutex_lock(&vcpu->kvm->lock);
1713 r = set_msr_hyperv_pw(vcpu, msr, data);
1714 mutex_unlock(&vcpu->kvm->lock);
1717 return set_msr_hyperv(vcpu, msr, data);
1719 case MSR_IA32_BBL_CR_CTL3:
1720 /* Drop writes to this legacy MSR -- see rdmsr
1721 * counterpart for further detail.
1723 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1726 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1727 return xen_hvm_config(vcpu, data);
1729 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1733 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1740 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1744 * Reads an msr value (of 'msr_index') into 'pdata'.
1745 * Returns 0 on success, non-0 otherwise.
1746 * Assumes vcpu_load() was already called.
1748 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1750 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1753 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1755 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1757 if (!msr_mtrr_valid(msr))
1760 if (msr == MSR_MTRRdefType)
1761 *pdata = vcpu->arch.mtrr_state.def_type +
1762 (vcpu->arch.mtrr_state.enabled << 10);
1763 else if (msr == MSR_MTRRfix64K_00000)
1765 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1766 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1767 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1768 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1769 else if (msr == MSR_IA32_CR_PAT)
1770 *pdata = vcpu->arch.pat;
1771 else { /* Variable MTRRs */
1772 int idx, is_mtrr_mask;
1775 idx = (msr - 0x200) / 2;
1776 is_mtrr_mask = msr - 0x200 - 2 * idx;
1779 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1782 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1789 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1792 u64 mcg_cap = vcpu->arch.mcg_cap;
1793 unsigned bank_num = mcg_cap & 0xff;
1796 case MSR_IA32_P5_MC_ADDR:
1797 case MSR_IA32_P5_MC_TYPE:
1800 case MSR_IA32_MCG_CAP:
1801 data = vcpu->arch.mcg_cap;
1803 case MSR_IA32_MCG_CTL:
1804 if (!(mcg_cap & MCG_CTL_P))
1806 data = vcpu->arch.mcg_ctl;
1808 case MSR_IA32_MCG_STATUS:
1809 data = vcpu->arch.mcg_status;
1812 if (msr >= MSR_IA32_MC0_CTL &&
1813 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1814 u32 offset = msr - MSR_IA32_MC0_CTL;
1815 data = vcpu->arch.mce_banks[offset];
1824 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1827 struct kvm *kvm = vcpu->kvm;
1830 case HV_X64_MSR_GUEST_OS_ID:
1831 data = kvm->arch.hv_guest_os_id;
1833 case HV_X64_MSR_HYPERCALL:
1834 data = kvm->arch.hv_hypercall;
1837 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1845 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1850 case HV_X64_MSR_VP_INDEX: {
1853 kvm_for_each_vcpu(r, v, vcpu->kvm)
1858 case HV_X64_MSR_EOI:
1859 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1860 case HV_X64_MSR_ICR:
1861 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1862 case HV_X64_MSR_TPR:
1863 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1864 case HV_X64_MSR_APIC_ASSIST_PAGE:
1865 data = vcpu->arch.hv_vapic;
1868 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1875 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1880 case MSR_IA32_PLATFORM_ID:
1881 case MSR_IA32_EBL_CR_POWERON:
1882 case MSR_IA32_DEBUGCTLMSR:
1883 case MSR_IA32_LASTBRANCHFROMIP:
1884 case MSR_IA32_LASTBRANCHTOIP:
1885 case MSR_IA32_LASTINTFROMIP:
1886 case MSR_IA32_LASTINTTOIP:
1888 case MSR_K8_TSEG_ADDR:
1889 case MSR_K8_TSEG_MASK:
1891 case MSR_VM_HSAVE_PA:
1892 case MSR_P6_PERFCTR0:
1893 case MSR_P6_PERFCTR1:
1894 case MSR_P6_EVNTSEL0:
1895 case MSR_P6_EVNTSEL1:
1896 case MSR_K7_EVNTSEL0:
1897 case MSR_K7_PERFCTR0:
1898 case MSR_K8_INT_PENDING_MSG:
1899 case MSR_AMD64_NB_CFG:
1900 case MSR_FAM10H_MMIO_CONF_BASE:
1903 case MSR_IA32_UCODE_REV:
1904 data = 0x100000000ULL;
1907 data = 0x500 | KVM_NR_VAR_MTRR;
1909 case 0x200 ... 0x2ff:
1910 return get_msr_mtrr(vcpu, msr, pdata);
1911 case 0xcd: /* fsb frequency */
1915 * MSR_EBC_FREQUENCY_ID
1916 * Conservative value valid for even the basic CPU models.
1917 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1918 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1919 * and 266MHz for model 3, or 4. Set Core Clock
1920 * Frequency to System Bus Frequency Ratio to 1 (bits
1921 * 31:24) even though these are only valid for CPU
1922 * models > 2, however guests may end up dividing or
1923 * multiplying by zero otherwise.
1925 case MSR_EBC_FREQUENCY_ID:
1928 case MSR_IA32_APICBASE:
1929 data = kvm_get_apic_base(vcpu);
1931 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1932 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1934 case MSR_IA32_TSCDEADLINE:
1935 data = kvm_get_lapic_tscdeadline_msr(vcpu);
1937 case MSR_IA32_MISC_ENABLE:
1938 data = vcpu->arch.ia32_misc_enable_msr;
1940 case MSR_IA32_PERF_STATUS:
1941 /* TSC increment by tick */
1943 /* CPU multiplier */
1944 data |= (((uint64_t)4ULL) << 40);
1947 data = vcpu->arch.efer;
1949 case MSR_KVM_WALL_CLOCK:
1950 case MSR_KVM_WALL_CLOCK_NEW:
1951 data = vcpu->kvm->arch.wall_clock;
1953 case MSR_KVM_SYSTEM_TIME:
1954 case MSR_KVM_SYSTEM_TIME_NEW:
1955 data = vcpu->arch.time;
1957 case MSR_KVM_ASYNC_PF_EN:
1958 data = vcpu->arch.apf.msr_val;
1960 case MSR_KVM_STEAL_TIME:
1961 data = vcpu->arch.st.msr_val;
1963 case MSR_IA32_P5_MC_ADDR:
1964 case MSR_IA32_P5_MC_TYPE:
1965 case MSR_IA32_MCG_CAP:
1966 case MSR_IA32_MCG_CTL:
1967 case MSR_IA32_MCG_STATUS:
1968 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1969 return get_msr_mce(vcpu, msr, pdata);
1970 case MSR_K7_CLK_CTL:
1972 * Provide expected ramp-up count for K7. All other
1973 * are set to zero, indicating minimum divisors for
1976 * This prevents guest kernels on AMD host with CPU
1977 * type 6, model 8 and higher from exploding due to
1978 * the rdmsr failing.
1982 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1983 if (kvm_hv_msr_partition_wide(msr)) {
1985 mutex_lock(&vcpu->kvm->lock);
1986 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1987 mutex_unlock(&vcpu->kvm->lock);
1990 return get_msr_hyperv(vcpu, msr, pdata);
1992 case MSR_IA32_BBL_CR_CTL3:
1993 /* This legacy MSR exists but isn't fully documented in current
1994 * silicon. It is however accessed by winxp in very narrow
1995 * scenarios where it sets bit #19, itself documented as
1996 * a "reserved" bit. Best effort attempt to source coherent
1997 * read data here should the balance of the register be
1998 * interpreted by the guest:
2000 * L2 cache control register 3: 64GB range, 256KB size,
2001 * enabled, latency 0x1, configured
2007 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2010 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2018 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2021 * Read or write a bunch of msrs. All parameters are kernel addresses.
2023 * @return number of msrs set successfully.
2025 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2026 struct kvm_msr_entry *entries,
2027 int (*do_msr)(struct kvm_vcpu *vcpu,
2028 unsigned index, u64 *data))
2032 idx = srcu_read_lock(&vcpu->kvm->srcu);
2033 for (i = 0; i < msrs->nmsrs; ++i)
2034 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2036 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2042 * Read or write a bunch of msrs. Parameters are user addresses.
2044 * @return number of msrs set successfully.
2046 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2047 int (*do_msr)(struct kvm_vcpu *vcpu,
2048 unsigned index, u64 *data),
2051 struct kvm_msrs msrs;
2052 struct kvm_msr_entry *entries;
2057 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2061 if (msrs.nmsrs >= MAX_IO_MSRS)
2065 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2066 entries = kmalloc(size, GFP_KERNEL);
2071 if (copy_from_user(entries, user_msrs->entries, size))
2074 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2079 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2090 int kvm_dev_ioctl_check_extension(long ext)
2095 case KVM_CAP_IRQCHIP:
2097 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2098 case KVM_CAP_SET_TSS_ADDR:
2099 case KVM_CAP_EXT_CPUID:
2100 case KVM_CAP_CLOCKSOURCE:
2102 case KVM_CAP_NOP_IO_DELAY:
2103 case KVM_CAP_MP_STATE:
2104 case KVM_CAP_SYNC_MMU:
2105 case KVM_CAP_USER_NMI:
2106 case KVM_CAP_REINJECT_CONTROL:
2107 case KVM_CAP_IRQ_INJECT_STATUS:
2108 case KVM_CAP_ASSIGN_DEV_IRQ:
2110 case KVM_CAP_IOEVENTFD:
2112 case KVM_CAP_PIT_STATE2:
2113 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2114 case KVM_CAP_XEN_HVM:
2115 case KVM_CAP_ADJUST_CLOCK:
2116 case KVM_CAP_VCPU_EVENTS:
2117 case KVM_CAP_HYPERV:
2118 case KVM_CAP_HYPERV_VAPIC:
2119 case KVM_CAP_HYPERV_SPIN:
2120 case KVM_CAP_PCI_SEGMENT:
2121 case KVM_CAP_DEBUGREGS:
2122 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2124 case KVM_CAP_ASYNC_PF:
2125 case KVM_CAP_GET_TSC_KHZ:
2128 case KVM_CAP_COALESCED_MMIO:
2129 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2132 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2134 case KVM_CAP_NR_VCPUS:
2135 r = KVM_SOFT_MAX_VCPUS;
2137 case KVM_CAP_MAX_VCPUS:
2140 case KVM_CAP_NR_MEMSLOTS:
2141 r = KVM_MEMORY_SLOTS;
2143 case KVM_CAP_PV_MMU: /* obsolete */
2147 r = iommu_present(&pci_bus_type);
2150 r = KVM_MAX_MCE_BANKS;
2155 case KVM_CAP_TSC_CONTROL:
2156 r = kvm_has_tsc_control;
2158 case KVM_CAP_TSC_DEADLINE_TIMER:
2159 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2169 long kvm_arch_dev_ioctl(struct file *filp,
2170 unsigned int ioctl, unsigned long arg)
2172 void __user *argp = (void __user *)arg;
2176 case KVM_GET_MSR_INDEX_LIST: {
2177 struct kvm_msr_list __user *user_msr_list = argp;
2178 struct kvm_msr_list msr_list;
2182 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2185 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2186 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2189 if (n < msr_list.nmsrs)
2192 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2193 num_msrs_to_save * sizeof(u32)))
2195 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2197 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2202 case KVM_GET_SUPPORTED_CPUID: {
2203 struct kvm_cpuid2 __user *cpuid_arg = argp;
2204 struct kvm_cpuid2 cpuid;
2207 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2209 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2210 cpuid_arg->entries);
2215 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2220 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2223 mce_cap = KVM_MCE_CAP_SUPPORTED;
2225 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2237 static void wbinvd_ipi(void *garbage)
2242 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2244 return vcpu->kvm->arch.iommu_domain &&
2245 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2248 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2250 /* Address WBINVD may be executed by guest */
2251 if (need_emulate_wbinvd(vcpu)) {
2252 if (kvm_x86_ops->has_wbinvd_exit())
2253 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2254 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2255 smp_call_function_single(vcpu->cpu,
2256 wbinvd_ipi, NULL, 1);
2259 kvm_x86_ops->vcpu_load(vcpu, cpu);
2260 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2261 /* Make sure TSC doesn't go backwards */
2265 tsc = kvm_x86_ops->read_l1_tsc(vcpu);
2266 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2267 tsc - vcpu->arch.last_guest_tsc;
2270 mark_tsc_unstable("KVM discovered backwards TSC");
2271 if (check_tsc_unstable()) {
2272 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2273 vcpu->arch.tsc_catchup = 1;
2275 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2276 if (vcpu->cpu != cpu)
2277 kvm_migrate_timers(vcpu);
2281 accumulate_steal_time(vcpu);
2282 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2285 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2287 kvm_x86_ops->vcpu_put(vcpu);
2288 kvm_put_guest_fpu(vcpu);
2289 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
2292 static int is_efer_nx(void)
2294 unsigned long long efer = 0;
2296 rdmsrl_safe(MSR_EFER, &efer);
2297 return efer & EFER_NX;
2300 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2303 struct kvm_cpuid_entry2 *e, *entry;
2306 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2307 e = &vcpu->arch.cpuid_entries[i];
2308 if (e->function == 0x80000001) {
2313 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2314 entry->edx &= ~(1 << 20);
2315 printk(KERN_INFO "kvm: guest NX capability removed\n");
2319 /* when an old userspace process fills a new kernel module */
2320 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2321 struct kvm_cpuid *cpuid,
2322 struct kvm_cpuid_entry __user *entries)
2325 struct kvm_cpuid_entry *cpuid_entries;
2328 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2331 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2335 if (copy_from_user(cpuid_entries, entries,
2336 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2338 for (i = 0; i < cpuid->nent; i++) {
2339 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2340 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2341 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2342 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2343 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2344 vcpu->arch.cpuid_entries[i].index = 0;
2345 vcpu->arch.cpuid_entries[i].flags = 0;
2346 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2347 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2348 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2350 vcpu->arch.cpuid_nent = cpuid->nent;
2351 cpuid_fix_nx_cap(vcpu);
2353 kvm_apic_set_version(vcpu);
2354 kvm_x86_ops->cpuid_update(vcpu);
2358 vfree(cpuid_entries);
2363 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2364 struct kvm_cpuid2 *cpuid,
2365 struct kvm_cpuid_entry2 __user *entries)
2370 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2373 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2374 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2376 vcpu->arch.cpuid_nent = cpuid->nent;
2377 kvm_apic_set_version(vcpu);
2378 kvm_x86_ops->cpuid_update(vcpu);
2386 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2387 struct kvm_cpuid2 *cpuid,
2388 struct kvm_cpuid_entry2 __user *entries)
2393 if (cpuid->nent < vcpu->arch.cpuid_nent)
2396 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2397 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2402 cpuid->nent = vcpu->arch.cpuid_nent;
2406 static void cpuid_mask(u32 *word, int wordnum)
2408 *word &= boot_cpu_data.x86_capability[wordnum];
2411 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2414 entry->function = function;
2415 entry->index = index;
2416 cpuid_count(entry->function, entry->index,
2417 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2421 static bool supported_xcr0_bit(unsigned bit)
2423 u64 mask = ((u64)1 << bit);
2425 return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
2428 #define F(x) bit(X86_FEATURE_##x)
2430 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2431 u32 index, int *nent, int maxnent)
2433 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2434 #ifdef CONFIG_X86_64
2435 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2437 unsigned f_lm = F(LM);
2439 unsigned f_gbpages = 0;
2442 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2445 const u32 kvm_supported_word0_x86_features =
2446 F(FPU) | F(VME) | F(DE) | F(PSE) |
2447 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2448 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2449 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2450 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2451 0 /* Reserved, DS, ACPI */ | F(MMX) |
2452 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2453 0 /* HTT, TM, Reserved, PBE */;
2454 /* cpuid 0x80000001.edx */
2455 const u32 kvm_supported_word1_x86_features =
2456 F(FPU) | F(VME) | F(DE) | F(PSE) |
2457 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2458 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2459 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2460 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2461 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2462 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2463 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2465 const u32 kvm_supported_word4_x86_features =
2466 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2467 0 /* DS-CPL, VMX, SMX, EST */ |
2468 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2469 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2470 0 /* Reserved, DCA */ | F(XMM4_1) |
2471 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2472 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2473 F(F16C) | F(RDRAND);
2474 /* cpuid 0x80000001.ecx */
2475 const u32 kvm_supported_word6_x86_features =
2476 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2477 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2478 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2479 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2481 /* cpuid 0xC0000001.edx */
2482 const u32 kvm_supported_word5_x86_features =
2483 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2484 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2488 const u32 kvm_supported_word9_x86_features =
2489 F(SMEP) | F(FSGSBASE) | F(ERMS);
2491 /* all calls to cpuid_count() should be made on the same cpu */
2493 do_cpuid_1_ent(entry, function, index);
2498 entry->eax = min(entry->eax, (u32)0xd);
2501 entry->edx &= kvm_supported_word0_x86_features;
2502 cpuid_mask(&entry->edx, 0);
2503 entry->ecx &= kvm_supported_word4_x86_features;
2504 cpuid_mask(&entry->ecx, 4);
2505 /* we support x2apic emulation even if host does not support
2506 * it since we emulate x2apic in software */
2507 entry->ecx |= F(X2APIC);
2509 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2510 * may return different values. This forces us to get_cpu() before
2511 * issuing the first command, and also to emulate this annoying behavior
2512 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2514 int t, times = entry->eax & 0xff;
2516 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2517 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2518 for (t = 1; t < times && *nent < maxnent; ++t) {
2519 do_cpuid_1_ent(&entry[t], function, 0);
2520 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2525 /* function 4 has additional index. */
2529 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2530 /* read more entries until cache_type is zero */
2531 for (i = 1; *nent < maxnent; ++i) {
2532 cache_type = entry[i - 1].eax & 0x1f;
2535 do_cpuid_1_ent(&entry[i], function, i);
2537 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2543 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2544 /* Mask ebx against host capbability word 9 */
2546 entry->ebx &= kvm_supported_word9_x86_features;
2547 cpuid_mask(&entry->ebx, 9);
2557 /* function 0xb has additional index. */
2561 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2562 /* read more entries until level_type is zero */
2563 for (i = 1; *nent < maxnent; ++i) {
2564 level_type = entry[i - 1].ecx & 0xff00;
2567 do_cpuid_1_ent(&entry[i], function, i);
2569 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2577 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2578 for (idx = 1, i = 1; *nent < maxnent && idx < 64; ++idx) {
2579 do_cpuid_1_ent(&entry[i], function, idx);
2580 if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
2583 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2589 case KVM_CPUID_SIGNATURE: {
2590 char signature[12] = "KVMKVMKVM\0\0";
2591 u32 *sigptr = (u32 *)signature;
2593 entry->ebx = sigptr[0];
2594 entry->ecx = sigptr[1];
2595 entry->edx = sigptr[2];
2598 case KVM_CPUID_FEATURES:
2599 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2600 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2601 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2602 (1 << KVM_FEATURE_ASYNC_PF) |
2603 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2605 if (sched_info_on())
2606 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
2613 entry->eax = min(entry->eax, 0x8000001a);
2616 entry->edx &= kvm_supported_word1_x86_features;
2617 cpuid_mask(&entry->edx, 1);
2618 entry->ecx &= kvm_supported_word6_x86_features;
2619 cpuid_mask(&entry->ecx, 6);
2622 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
2623 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
2624 unsigned phys_as = entry->eax & 0xff;
2627 g_phys_as = phys_as;
2628 entry->eax = g_phys_as | (virt_as << 8);
2629 entry->ebx = entry->edx = 0;
2633 entry->ecx = entry->edx = 0;
2639 /*Add support for Centaur's CPUID instruction*/
2641 /*Just support up to 0xC0000004 now*/
2642 entry->eax = min(entry->eax, 0xC0000004);
2645 entry->edx &= kvm_supported_word5_x86_features;
2646 cpuid_mask(&entry->edx, 5);
2648 case 3: /* Processor serial number */
2649 case 5: /* MONITOR/MWAIT */
2650 case 6: /* Thermal management */
2651 case 0xA: /* Architectural Performance Monitoring */
2652 case 0x80000007: /* Advanced power management */
2657 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
2661 kvm_x86_ops->set_supported_cpuid(function, entry);
2668 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2669 struct kvm_cpuid_entry2 __user *entries)
2671 struct kvm_cpuid_entry2 *cpuid_entries;
2672 int limit, nent = 0, r = -E2BIG;
2675 if (cpuid->nent < 1)
2677 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2678 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2680 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2684 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2685 limit = cpuid_entries[0].eax;
2686 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2687 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2688 &nent, cpuid->nent);
2690 if (nent >= cpuid->nent)
2693 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2694 limit = cpuid_entries[nent - 1].eax;
2695 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2696 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2697 &nent, cpuid->nent);
2702 if (nent >= cpuid->nent)
2705 /* Add support for Centaur's CPUID instruction. */
2706 if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2707 do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2708 &nent, cpuid->nent);
2711 if (nent >= cpuid->nent)
2714 limit = cpuid_entries[nent - 1].eax;
2715 for (func = 0xC0000001;
2716 func <= limit && nent < cpuid->nent; ++func)
2717 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2718 &nent, cpuid->nent);
2721 if (nent >= cpuid->nent)
2725 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2729 if (nent >= cpuid->nent)
2732 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2736 if (nent >= cpuid->nent)
2740 if (copy_to_user(entries, cpuid_entries,
2741 nent * sizeof(struct kvm_cpuid_entry2)))
2747 vfree(cpuid_entries);
2752 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2753 struct kvm_lapic_state *s)
2755 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2760 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2761 struct kvm_lapic_state *s)
2763 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2764 kvm_apic_post_state_restore(vcpu);
2765 update_cr8_intercept(vcpu);
2770 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2771 struct kvm_interrupt *irq)
2773 if (irq->irq < 0 || irq->irq >= 256)
2775 if (irqchip_in_kernel(vcpu->kvm))
2778 kvm_queue_interrupt(vcpu, irq->irq, false);
2779 kvm_make_request(KVM_REQ_EVENT, vcpu);
2784 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2786 kvm_inject_nmi(vcpu);
2791 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2792 struct kvm_tpr_access_ctl *tac)
2796 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2800 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2804 unsigned bank_num = mcg_cap & 0xff, bank;
2807 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2809 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2812 vcpu->arch.mcg_cap = mcg_cap;
2813 /* Init IA32_MCG_CTL to all 1s */
2814 if (mcg_cap & MCG_CTL_P)
2815 vcpu->arch.mcg_ctl = ~(u64)0;
2816 /* Init IA32_MCi_CTL to all 1s */
2817 for (bank = 0; bank < bank_num; bank++)
2818 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2823 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2824 struct kvm_x86_mce *mce)
2826 u64 mcg_cap = vcpu->arch.mcg_cap;
2827 unsigned bank_num = mcg_cap & 0xff;
2828 u64 *banks = vcpu->arch.mce_banks;
2830 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2833 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2834 * reporting is disabled
2836 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2837 vcpu->arch.mcg_ctl != ~(u64)0)
2839 banks += 4 * mce->bank;
2841 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2842 * reporting is disabled for the bank
2844 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2846 if (mce->status & MCI_STATUS_UC) {
2847 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2848 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2849 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2852 if (banks[1] & MCI_STATUS_VAL)
2853 mce->status |= MCI_STATUS_OVER;
2854 banks[2] = mce->addr;
2855 banks[3] = mce->misc;
2856 vcpu->arch.mcg_status = mce->mcg_status;
2857 banks[1] = mce->status;
2858 kvm_queue_exception(vcpu, MC_VECTOR);
2859 } else if (!(banks[1] & MCI_STATUS_VAL)
2860 || !(banks[1] & MCI_STATUS_UC)) {
2861 if (banks[1] & MCI_STATUS_VAL)
2862 mce->status |= MCI_STATUS_OVER;
2863 banks[2] = mce->addr;
2864 banks[3] = mce->misc;
2865 banks[1] = mce->status;
2867 banks[1] |= MCI_STATUS_OVER;
2871 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2872 struct kvm_vcpu_events *events)
2875 events->exception.injected =
2876 vcpu->arch.exception.pending &&
2877 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2878 events->exception.nr = vcpu->arch.exception.nr;
2879 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2880 events->exception.pad = 0;
2881 events->exception.error_code = vcpu->arch.exception.error_code;
2883 events->interrupt.injected =
2884 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2885 events->interrupt.nr = vcpu->arch.interrupt.nr;
2886 events->interrupt.soft = 0;
2887 events->interrupt.shadow =
2888 kvm_x86_ops->get_interrupt_shadow(vcpu,
2889 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2891 events->nmi.injected = vcpu->arch.nmi_injected;
2892 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2893 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2894 events->nmi.pad = 0;
2896 events->sipi_vector = vcpu->arch.sipi_vector;
2898 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2899 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2900 | KVM_VCPUEVENT_VALID_SHADOW);
2901 memset(&events->reserved, 0, sizeof(events->reserved));
2904 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2905 struct kvm_vcpu_events *events)
2907 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2908 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2909 | KVM_VCPUEVENT_VALID_SHADOW))
2913 vcpu->arch.exception.pending = events->exception.injected;
2914 vcpu->arch.exception.nr = events->exception.nr;
2915 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2916 vcpu->arch.exception.error_code = events->exception.error_code;
2918 vcpu->arch.interrupt.pending = events->interrupt.injected;
2919 vcpu->arch.interrupt.nr = events->interrupt.nr;
2920 vcpu->arch.interrupt.soft = events->interrupt.soft;
2921 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2922 kvm_x86_ops->set_interrupt_shadow(vcpu,
2923 events->interrupt.shadow);
2925 vcpu->arch.nmi_injected = events->nmi.injected;
2926 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2927 vcpu->arch.nmi_pending = events->nmi.pending;
2928 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2930 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2931 vcpu->arch.sipi_vector = events->sipi_vector;
2933 kvm_make_request(KVM_REQ_EVENT, vcpu);
2938 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2939 struct kvm_debugregs *dbgregs)
2941 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2942 dbgregs->dr6 = vcpu->arch.dr6;
2943 dbgregs->dr7 = vcpu->arch.dr7;
2945 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2948 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2949 struct kvm_debugregs *dbgregs)
2954 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2955 vcpu->arch.dr6 = dbgregs->dr6;
2956 vcpu->arch.dr7 = dbgregs->dr7;
2961 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2962 struct kvm_xsave *guest_xsave)
2965 memcpy(guest_xsave->region,
2966 &vcpu->arch.guest_fpu.state->xsave,
2969 memcpy(guest_xsave->region,
2970 &vcpu->arch.guest_fpu.state->fxsave,
2971 sizeof(struct i387_fxsave_struct));
2972 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2977 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2978 struct kvm_xsave *guest_xsave)
2981 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2984 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2985 guest_xsave->region, xstate_size);
2987 if (xstate_bv & ~XSTATE_FPSSE)
2989 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2990 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2995 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2996 struct kvm_xcrs *guest_xcrs)
2998 if (!cpu_has_xsave) {
2999 guest_xcrs->nr_xcrs = 0;
3003 guest_xcrs->nr_xcrs = 1;
3004 guest_xcrs->flags = 0;
3005 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3006 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3009 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3010 struct kvm_xcrs *guest_xcrs)
3017 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3020 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3021 /* Only support XCR0 currently */
3022 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
3023 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3024 guest_xcrs->xcrs[0].value);
3032 long kvm_arch_vcpu_ioctl(struct file *filp,
3033 unsigned int ioctl, unsigned long arg)
3035 struct kvm_vcpu *vcpu = filp->private_data;
3036 void __user *argp = (void __user *)arg;
3039 struct kvm_lapic_state *lapic;
3040 struct kvm_xsave *xsave;
3041 struct kvm_xcrs *xcrs;
3047 case KVM_GET_LAPIC: {
3049 if (!vcpu->arch.apic)
3051 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3056 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3060 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3065 case KVM_SET_LAPIC: {
3067 if (!vcpu->arch.apic)
3069 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3074 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
3076 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3082 case KVM_INTERRUPT: {
3083 struct kvm_interrupt irq;
3086 if (copy_from_user(&irq, argp, sizeof irq))
3088 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3095 r = kvm_vcpu_ioctl_nmi(vcpu);
3101 case KVM_SET_CPUID: {
3102 struct kvm_cpuid __user *cpuid_arg = argp;
3103 struct kvm_cpuid cpuid;
3106 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3108 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3113 case KVM_SET_CPUID2: {
3114 struct kvm_cpuid2 __user *cpuid_arg = argp;
3115 struct kvm_cpuid2 cpuid;
3118 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3120 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3121 cpuid_arg->entries);
3126 case KVM_GET_CPUID2: {
3127 struct kvm_cpuid2 __user *cpuid_arg = argp;
3128 struct kvm_cpuid2 cpuid;
3131 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3133 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3134 cpuid_arg->entries);
3138 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3144 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3147 r = msr_io(vcpu, argp, do_set_msr, 0);
3149 case KVM_TPR_ACCESS_REPORTING: {
3150 struct kvm_tpr_access_ctl tac;
3153 if (copy_from_user(&tac, argp, sizeof tac))
3155 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3159 if (copy_to_user(argp, &tac, sizeof tac))
3164 case KVM_SET_VAPIC_ADDR: {
3165 struct kvm_vapic_addr va;
3168 if (!irqchip_in_kernel(vcpu->kvm))
3171 if (copy_from_user(&va, argp, sizeof va))
3173 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3176 case KVM_X86_SETUP_MCE: {
3180 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3182 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3185 case KVM_X86_SET_MCE: {
3186 struct kvm_x86_mce mce;
3189 if (copy_from_user(&mce, argp, sizeof mce))
3191 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3194 case KVM_GET_VCPU_EVENTS: {
3195 struct kvm_vcpu_events events;
3197 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3200 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3205 case KVM_SET_VCPU_EVENTS: {
3206 struct kvm_vcpu_events events;
3209 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3212 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3215 case KVM_GET_DEBUGREGS: {
3216 struct kvm_debugregs dbgregs;
3218 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3221 if (copy_to_user(argp, &dbgregs,
3222 sizeof(struct kvm_debugregs)))
3227 case KVM_SET_DEBUGREGS: {
3228 struct kvm_debugregs dbgregs;
3231 if (copy_from_user(&dbgregs, argp,
3232 sizeof(struct kvm_debugregs)))
3235 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3238 case KVM_GET_XSAVE: {
3239 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3244 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3247 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3252 case KVM_SET_XSAVE: {
3253 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3259 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3262 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3265 case KVM_GET_XCRS: {
3266 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3271 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3274 if (copy_to_user(argp, u.xcrs,
3275 sizeof(struct kvm_xcrs)))
3280 case KVM_SET_XCRS: {
3281 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3287 if (copy_from_user(u.xcrs, argp,
3288 sizeof(struct kvm_xcrs)))
3291 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3294 case KVM_SET_TSC_KHZ: {
3298 if (!kvm_has_tsc_control)
3301 user_tsc_khz = (u32)arg;
3303 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3306 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3311 case KVM_GET_TSC_KHZ: {
3313 if (check_tsc_unstable())
3316 r = vcpu_tsc_khz(vcpu);
3328 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3332 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3334 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3338 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3341 kvm->arch.ept_identity_map_addr = ident_addr;
3345 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3346 u32 kvm_nr_mmu_pages)
3348 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3351 mutex_lock(&kvm->slots_lock);
3352 spin_lock(&kvm->mmu_lock);
3354 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3355 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3357 spin_unlock(&kvm->mmu_lock);
3358 mutex_unlock(&kvm->slots_lock);
3362 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3364 return kvm->arch.n_max_mmu_pages;
3367 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3372 switch (chip->chip_id) {
3373 case KVM_IRQCHIP_PIC_MASTER:
3374 memcpy(&chip->chip.pic,
3375 &pic_irqchip(kvm)->pics[0],
3376 sizeof(struct kvm_pic_state));
3378 case KVM_IRQCHIP_PIC_SLAVE:
3379 memcpy(&chip->chip.pic,
3380 &pic_irqchip(kvm)->pics[1],
3381 sizeof(struct kvm_pic_state));
3383 case KVM_IRQCHIP_IOAPIC:
3384 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3393 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3398 switch (chip->chip_id) {
3399 case KVM_IRQCHIP_PIC_MASTER:
3400 spin_lock(&pic_irqchip(kvm)->lock);
3401 memcpy(&pic_irqchip(kvm)->pics[0],
3403 sizeof(struct kvm_pic_state));
3404 spin_unlock(&pic_irqchip(kvm)->lock);
3406 case KVM_IRQCHIP_PIC_SLAVE:
3407 spin_lock(&pic_irqchip(kvm)->lock);
3408 memcpy(&pic_irqchip(kvm)->pics[1],
3410 sizeof(struct kvm_pic_state));
3411 spin_unlock(&pic_irqchip(kvm)->lock);
3413 case KVM_IRQCHIP_IOAPIC:
3414 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3420 kvm_pic_update_irq(pic_irqchip(kvm));
3424 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3428 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3429 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3430 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3434 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3438 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3439 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3440 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3441 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3445 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3449 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3450 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3451 sizeof(ps->channels));
3452 ps->flags = kvm->arch.vpit->pit_state.flags;
3453 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3454 memset(&ps->reserved, 0, sizeof(ps->reserved));
3458 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3460 int r = 0, start = 0;
3461 u32 prev_legacy, cur_legacy;
3462 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3463 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3464 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3465 if (!prev_legacy && cur_legacy)
3467 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3468 sizeof(kvm->arch.vpit->pit_state.channels));
3469 kvm->arch.vpit->pit_state.flags = ps->flags;
3470 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3471 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3475 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3476 struct kvm_reinject_control *control)
3478 if (!kvm->arch.vpit)
3480 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3481 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3482 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3487 * Get (and clear) the dirty memory log for a memory slot.
3489 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3490 struct kvm_dirty_log *log)
3493 struct kvm_memory_slot *memslot;
3495 unsigned long is_dirty = 0;
3497 mutex_lock(&kvm->slots_lock);
3500 if (log->slot >= KVM_MEMORY_SLOTS)
3503 memslot = &kvm->memslots->memslots[log->slot];
3505 if (!memslot->dirty_bitmap)
3508 n = kvm_dirty_bitmap_bytes(memslot);
3510 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3511 is_dirty = memslot->dirty_bitmap[i];
3513 /* If nothing is dirty, don't bother messing with page tables. */
3515 struct kvm_memslots *slots, *old_slots;
3516 unsigned long *dirty_bitmap;
3518 dirty_bitmap = memslot->dirty_bitmap_head;
3519 if (memslot->dirty_bitmap == dirty_bitmap)
3520 dirty_bitmap += n / sizeof(long);
3521 memset(dirty_bitmap, 0, n);
3524 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3527 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3528 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3529 slots->generation++;
3531 old_slots = kvm->memslots;
3532 rcu_assign_pointer(kvm->memslots, slots);
3533 synchronize_srcu_expedited(&kvm->srcu);
3534 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3537 spin_lock(&kvm->mmu_lock);
3538 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3539 spin_unlock(&kvm->mmu_lock);
3542 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3546 if (clear_user(log->dirty_bitmap, n))
3552 mutex_unlock(&kvm->slots_lock);
3556 long kvm_arch_vm_ioctl(struct file *filp,
3557 unsigned int ioctl, unsigned long arg)
3559 struct kvm *kvm = filp->private_data;
3560 void __user *argp = (void __user *)arg;
3563 * This union makes it completely explicit to gcc-3.x
3564 * that these two variables' stack usage should be
3565 * combined, not added together.
3568 struct kvm_pit_state ps;
3569 struct kvm_pit_state2 ps2;
3570 struct kvm_pit_config pit_config;
3574 case KVM_SET_TSS_ADDR:
3575 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3579 case KVM_SET_IDENTITY_MAP_ADDR: {
3583 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3585 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3590 case KVM_SET_NR_MMU_PAGES:
3591 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3595 case KVM_GET_NR_MMU_PAGES:
3596 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3598 case KVM_CREATE_IRQCHIP: {
3599 struct kvm_pic *vpic;
3601 mutex_lock(&kvm->lock);
3604 goto create_irqchip_unlock;
3606 if (atomic_read(&kvm->online_vcpus))
3607 goto create_irqchip_unlock;
3609 vpic = kvm_create_pic(kvm);
3611 r = kvm_ioapic_init(kvm);
3613 mutex_lock(&kvm->slots_lock);
3614 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3616 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3618 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3620 mutex_unlock(&kvm->slots_lock);
3622 goto create_irqchip_unlock;
3625 goto create_irqchip_unlock;
3627 kvm->arch.vpic = vpic;
3629 r = kvm_setup_default_irq_routing(kvm);
3631 mutex_lock(&kvm->slots_lock);
3632 mutex_lock(&kvm->irq_lock);
3633 kvm_ioapic_destroy(kvm);
3634 kvm_destroy_pic(kvm);
3635 mutex_unlock(&kvm->irq_lock);
3636 mutex_unlock(&kvm->slots_lock);
3638 create_irqchip_unlock:
3639 mutex_unlock(&kvm->lock);
3642 case KVM_CREATE_PIT:
3643 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3645 case KVM_CREATE_PIT2:
3647 if (copy_from_user(&u.pit_config, argp,
3648 sizeof(struct kvm_pit_config)))
3651 mutex_lock(&kvm->slots_lock);
3654 goto create_pit_unlock;
3656 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3660 mutex_unlock(&kvm->slots_lock);
3662 case KVM_IRQ_LINE_STATUS:
3663 case KVM_IRQ_LINE: {
3664 struct kvm_irq_level irq_event;
3667 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3670 if (irqchip_in_kernel(kvm)) {
3672 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3673 irq_event.irq, irq_event.level);
3674 if (ioctl == KVM_IRQ_LINE_STATUS) {
3676 irq_event.status = status;
3677 if (copy_to_user(argp, &irq_event,
3685 case KVM_GET_IRQCHIP: {
3686 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3687 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3693 if (copy_from_user(chip, argp, sizeof *chip))
3694 goto get_irqchip_out;
3696 if (!irqchip_in_kernel(kvm))
3697 goto get_irqchip_out;
3698 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3700 goto get_irqchip_out;
3702 if (copy_to_user(argp, chip, sizeof *chip))
3703 goto get_irqchip_out;
3711 case KVM_SET_IRQCHIP: {
3712 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3713 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3719 if (copy_from_user(chip, argp, sizeof *chip))
3720 goto set_irqchip_out;
3722 if (!irqchip_in_kernel(kvm))
3723 goto set_irqchip_out;
3724 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3726 goto set_irqchip_out;
3736 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3739 if (!kvm->arch.vpit)
3741 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3745 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3752 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3755 if (!kvm->arch.vpit)
3757 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3763 case KVM_GET_PIT2: {
3765 if (!kvm->arch.vpit)
3767 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3771 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3776 case KVM_SET_PIT2: {
3778 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3781 if (!kvm->arch.vpit)
3783 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3789 case KVM_REINJECT_CONTROL: {
3790 struct kvm_reinject_control control;
3792 if (copy_from_user(&control, argp, sizeof(control)))
3794 r = kvm_vm_ioctl_reinject(kvm, &control);
3800 case KVM_XEN_HVM_CONFIG: {
3802 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3803 sizeof(struct kvm_xen_hvm_config)))
3806 if (kvm->arch.xen_hvm_config.flags)
3811 case KVM_SET_CLOCK: {
3812 struct kvm_clock_data user_ns;
3817 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3825 local_irq_disable();
3826 now_ns = get_kernel_ns();
3827 delta = user_ns.clock - now_ns;
3829 kvm->arch.kvmclock_offset = delta;
3832 case KVM_GET_CLOCK: {
3833 struct kvm_clock_data user_ns;
3836 local_irq_disable();
3837 now_ns = get_kernel_ns();
3838 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3841 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3844 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3857 static void kvm_init_msr_list(void)
3862 /* skip the first msrs in the list. KVM-specific */
3863 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3864 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3867 msrs_to_save[j] = msrs_to_save[i];
3870 num_msrs_to_save = j;
3873 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3881 if (!(vcpu->arch.apic &&
3882 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3883 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3894 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3901 if (!(vcpu->arch.apic &&
3902 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3903 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3905 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3915 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3916 struct kvm_segment *var, int seg)
3918 kvm_x86_ops->set_segment(vcpu, var, seg);
3921 void kvm_get_segment(struct kvm_vcpu *vcpu,
3922 struct kvm_segment *var, int seg)
3924 kvm_x86_ops->get_segment(vcpu, var, seg);
3927 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3932 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3935 struct x86_exception exception;
3937 BUG_ON(!mmu_is_nested(vcpu));
3939 /* NPT walks are always user-walks */
3940 access |= PFERR_USER_MASK;
3941 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3946 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3947 struct x86_exception *exception)
3949 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3950 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3953 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3954 struct x86_exception *exception)
3956 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3957 access |= PFERR_FETCH_MASK;
3958 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3961 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3962 struct x86_exception *exception)
3964 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3965 access |= PFERR_WRITE_MASK;
3966 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3969 /* uses this to access any guest's mapped memory without checking CPL */
3970 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3971 struct x86_exception *exception)
3973 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3976 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3977 struct kvm_vcpu *vcpu, u32 access,
3978 struct x86_exception *exception)
3981 int r = X86EMUL_CONTINUE;
3984 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3986 unsigned offset = addr & (PAGE_SIZE-1);
3987 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3990 if (gpa == UNMAPPED_GVA)
3991 return X86EMUL_PROPAGATE_FAULT;
3992 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3994 r = X86EMUL_IO_NEEDED;
4006 /* used for instruction fetching */
4007 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4008 gva_t addr, void *val, unsigned int bytes,
4009 struct x86_exception *exception)
4011 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4012 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4014 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
4015 access | PFERR_FETCH_MASK,
4019 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4020 gva_t addr, void *val, unsigned int bytes,
4021 struct x86_exception *exception)
4023 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4024 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4026 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4029 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4031 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4032 gva_t addr, void *val, unsigned int bytes,
4033 struct x86_exception *exception)
4035 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4036 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4039 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4040 gva_t addr, void *val,
4042 struct x86_exception *exception)
4044 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4046 int r = X86EMUL_CONTINUE;
4049 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4052 unsigned offset = addr & (PAGE_SIZE-1);
4053 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4056 if (gpa == UNMAPPED_GVA)
4057 return X86EMUL_PROPAGATE_FAULT;
4058 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4060 r = X86EMUL_IO_NEEDED;
4071 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4073 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4074 gpa_t *gpa, struct x86_exception *exception,
4077 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4079 if (vcpu_match_mmio_gva(vcpu, gva) &&
4080 check_write_user_access(vcpu, write, access,
4081 vcpu->arch.access)) {
4082 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4083 (gva & (PAGE_SIZE - 1));
4084 trace_vcpu_match_mmio(gva, *gpa, write, false);
4089 access |= PFERR_WRITE_MASK;
4091 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4093 if (*gpa == UNMAPPED_GVA)
4096 /* For APIC access vmexit */
4097 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4100 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4101 trace_vcpu_match_mmio(gva, *gpa, write, true);
4108 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4109 const void *val, int bytes)
4113 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4116 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
4120 struct read_write_emulator_ops {
4121 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4123 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4124 void *val, int bytes);
4125 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4126 int bytes, void *val);
4127 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4128 void *val, int bytes);
4132 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4134 if (vcpu->mmio_read_completed) {
4135 memcpy(val, vcpu->mmio_data, bytes);
4136 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4137 vcpu->mmio_phys_addr, *(u64 *)val);
4138 vcpu->mmio_read_completed = 0;
4145 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4146 void *val, int bytes)
4148 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4151 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4152 void *val, int bytes)
4154 return emulator_write_phys(vcpu, gpa, val, bytes);
4157 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4159 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4160 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4163 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4164 void *val, int bytes)
4166 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4167 return X86EMUL_IO_NEEDED;
4170 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4171 void *val, int bytes)
4173 memcpy(vcpu->mmio_data, val, bytes);
4174 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
4175 return X86EMUL_CONTINUE;
4178 static struct read_write_emulator_ops read_emultor = {
4179 .read_write_prepare = read_prepare,
4180 .read_write_emulate = read_emulate,
4181 .read_write_mmio = vcpu_mmio_read,
4182 .read_write_exit_mmio = read_exit_mmio,
4185 static struct read_write_emulator_ops write_emultor = {
4186 .read_write_emulate = write_emulate,
4187 .read_write_mmio = write_mmio,
4188 .read_write_exit_mmio = write_exit_mmio,
4192 static int emulator_read_write_onepage(unsigned long addr, void *val,
4194 struct x86_exception *exception,
4195 struct kvm_vcpu *vcpu,
4196 struct read_write_emulator_ops *ops)
4200 bool write = ops->write;
4202 if (ops->read_write_prepare &&
4203 ops->read_write_prepare(vcpu, val, bytes))
4204 return X86EMUL_CONTINUE;
4206 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4209 return X86EMUL_PROPAGATE_FAULT;
4211 /* For APIC access vmexit */
4215 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4216 return X86EMUL_CONTINUE;
4220 * Is this MMIO handled locally?
4222 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4223 if (handled == bytes)
4224 return X86EMUL_CONTINUE;
4230 vcpu->mmio_needed = 1;
4231 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4232 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
4233 vcpu->mmio_size = bytes;
4234 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
4235 vcpu->run->mmio.is_write = vcpu->mmio_is_write = write;
4236 vcpu->mmio_index = 0;
4238 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4241 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4242 void *val, unsigned int bytes,
4243 struct x86_exception *exception,
4244 struct read_write_emulator_ops *ops)
4246 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4248 /* Crossing a page boundary? */
4249 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4252 now = -addr & ~PAGE_MASK;
4253 rc = emulator_read_write_onepage(addr, val, now, exception,
4256 if (rc != X86EMUL_CONTINUE)
4263 return emulator_read_write_onepage(addr, val, bytes, exception,
4267 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4271 struct x86_exception *exception)
4273 return emulator_read_write(ctxt, addr, val, bytes,
4274 exception, &read_emultor);
4277 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4281 struct x86_exception *exception)
4283 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4284 exception, &write_emultor);
4287 #define CMPXCHG_TYPE(t, ptr, old, new) \
4288 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4290 #ifdef CONFIG_X86_64
4291 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4293 # define CMPXCHG64(ptr, old, new) \
4294 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4297 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4302 struct x86_exception *exception)
4304 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4310 /* guests cmpxchg8b have to be emulated atomically */
4311 if (bytes > 8 || (bytes & (bytes - 1)))
4314 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4316 if (gpa == UNMAPPED_GVA ||
4317 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4320 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4323 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4324 if (is_error_page(page)) {
4325 kvm_release_page_clean(page);
4329 kaddr = kmap_atomic(page, KM_USER0);
4330 kaddr += offset_in_page(gpa);
4333 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4336 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4339 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4342 exchanged = CMPXCHG64(kaddr, old, new);
4347 kunmap_atomic(kaddr, KM_USER0);
4348 kvm_release_page_dirty(page);
4351 return X86EMUL_CMPXCHG_FAILED;
4353 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4355 return X86EMUL_CONTINUE;
4358 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4360 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4363 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4365 /* TODO: String I/O for in kernel device */
4368 if (vcpu->arch.pio.in)
4369 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4370 vcpu->arch.pio.size, pd);
4372 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4373 vcpu->arch.pio.port, vcpu->arch.pio.size,
4379 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4380 int size, unsigned short port, void *val,
4383 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4385 if (vcpu->arch.pio.count)
4388 trace_kvm_pio(0, port, size, count);
4390 vcpu->arch.pio.port = port;
4391 vcpu->arch.pio.in = 1;
4392 vcpu->arch.pio.count = count;
4393 vcpu->arch.pio.size = size;
4395 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4397 memcpy(val, vcpu->arch.pio_data, size * count);
4398 vcpu->arch.pio.count = 0;
4402 vcpu->run->exit_reason = KVM_EXIT_IO;
4403 vcpu->run->io.direction = KVM_EXIT_IO_IN;
4404 vcpu->run->io.size = size;
4405 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4406 vcpu->run->io.count = count;
4407 vcpu->run->io.port = port;
4412 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4413 int size, unsigned short port,
4414 const void *val, unsigned int count)
4416 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4418 trace_kvm_pio(1, port, size, count);
4420 vcpu->arch.pio.port = port;
4421 vcpu->arch.pio.in = 0;
4422 vcpu->arch.pio.count = count;
4423 vcpu->arch.pio.size = size;
4425 memcpy(vcpu->arch.pio_data, val, size * count);
4427 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4428 vcpu->arch.pio.count = 0;
4432 vcpu->run->exit_reason = KVM_EXIT_IO;
4433 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4434 vcpu->run->io.size = size;
4435 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4436 vcpu->run->io.count = count;
4437 vcpu->run->io.port = port;
4442 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4444 return kvm_x86_ops->get_segment_base(vcpu, seg);
4447 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4449 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4452 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4454 if (!need_emulate_wbinvd(vcpu))
4455 return X86EMUL_CONTINUE;
4457 if (kvm_x86_ops->has_wbinvd_exit()) {
4458 int cpu = get_cpu();
4460 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4461 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4462 wbinvd_ipi, NULL, 1);
4464 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4467 return X86EMUL_CONTINUE;
4469 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4471 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4473 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4476 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4478 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4481 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4484 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4487 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4489 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4492 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4494 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4495 unsigned long value;
4499 value = kvm_read_cr0(vcpu);
4502 value = vcpu->arch.cr2;
4505 value = kvm_read_cr3(vcpu);
4508 value = kvm_read_cr4(vcpu);
4511 value = kvm_get_cr8(vcpu);
4514 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4521 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4523 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4528 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4531 vcpu->arch.cr2 = val;
4534 res = kvm_set_cr3(vcpu, val);
4537 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4540 res = kvm_set_cr8(vcpu, val);
4543 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4550 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4552 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4555 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4557 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4560 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4562 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4565 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4567 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4570 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4572 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4575 static unsigned long emulator_get_cached_segment_base(
4576 struct x86_emulate_ctxt *ctxt, int seg)
4578 return get_segment_base(emul_to_vcpu(ctxt), seg);
4581 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4582 struct desc_struct *desc, u32 *base3,
4585 struct kvm_segment var;
4587 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4588 *selector = var.selector;
4595 set_desc_limit(desc, var.limit);
4596 set_desc_base(desc, (unsigned long)var.base);
4597 #ifdef CONFIG_X86_64
4599 *base3 = var.base >> 32;
4601 desc->type = var.type;
4603 desc->dpl = var.dpl;
4604 desc->p = var.present;
4605 desc->avl = var.avl;
4613 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4614 struct desc_struct *desc, u32 base3,
4617 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4618 struct kvm_segment var;
4620 var.selector = selector;
4621 var.base = get_desc_base(desc);
4622 #ifdef CONFIG_X86_64
4623 var.base |= ((u64)base3) << 32;
4625 var.limit = get_desc_limit(desc);
4627 var.limit = (var.limit << 12) | 0xfff;
4628 var.type = desc->type;
4629 var.present = desc->p;
4630 var.dpl = desc->dpl;
4635 var.avl = desc->avl;
4636 var.present = desc->p;
4637 var.unusable = !var.present;
4640 kvm_set_segment(vcpu, &var, seg);
4644 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4645 u32 msr_index, u64 *pdata)
4647 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4650 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4651 u32 msr_index, u64 data)
4653 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4656 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4658 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4661 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4664 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4666 * CR0.TS may reference the host fpu state, not the guest fpu state,
4667 * so it may be clear at this point.
4672 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4677 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4678 struct x86_instruction_info *info,
4679 enum x86_intercept_stage stage)
4681 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4684 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4685 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4687 struct kvm_cpuid_entry2 *cpuid = NULL;
4690 cpuid = kvm_find_cpuid_entry(emul_to_vcpu(ctxt),
4706 static struct x86_emulate_ops emulate_ops = {
4707 .read_std = kvm_read_guest_virt_system,
4708 .write_std = kvm_write_guest_virt_system,
4709 .fetch = kvm_fetch_guest_virt,
4710 .read_emulated = emulator_read_emulated,
4711 .write_emulated = emulator_write_emulated,
4712 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4713 .invlpg = emulator_invlpg,
4714 .pio_in_emulated = emulator_pio_in_emulated,
4715 .pio_out_emulated = emulator_pio_out_emulated,
4716 .get_segment = emulator_get_segment,
4717 .set_segment = emulator_set_segment,
4718 .get_cached_segment_base = emulator_get_cached_segment_base,
4719 .get_gdt = emulator_get_gdt,
4720 .get_idt = emulator_get_idt,
4721 .set_gdt = emulator_set_gdt,
4722 .set_idt = emulator_set_idt,
4723 .get_cr = emulator_get_cr,
4724 .set_cr = emulator_set_cr,
4725 .cpl = emulator_get_cpl,
4726 .get_dr = emulator_get_dr,
4727 .set_dr = emulator_set_dr,
4728 .set_msr = emulator_set_msr,
4729 .get_msr = emulator_get_msr,
4730 .halt = emulator_halt,
4731 .wbinvd = emulator_wbinvd,
4732 .fix_hypercall = emulator_fix_hypercall,
4733 .get_fpu = emulator_get_fpu,
4734 .put_fpu = emulator_put_fpu,
4735 .intercept = emulator_intercept,
4736 .get_cpuid = emulator_get_cpuid,
4739 static void cache_all_regs(struct kvm_vcpu *vcpu)
4741 kvm_register_read(vcpu, VCPU_REGS_RAX);
4742 kvm_register_read(vcpu, VCPU_REGS_RSP);
4743 kvm_register_read(vcpu, VCPU_REGS_RIP);
4744 vcpu->arch.regs_dirty = ~0;
4747 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4749 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4751 * an sti; sti; sequence only disable interrupts for the first
4752 * instruction. So, if the last instruction, be it emulated or
4753 * not, left the system with the INT_STI flag enabled, it
4754 * means that the last instruction is an sti. We should not
4755 * leave the flag on in this case. The same goes for mov ss
4757 if (!(int_shadow & mask))
4758 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4761 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4763 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4764 if (ctxt->exception.vector == PF_VECTOR)
4765 kvm_propagate_fault(vcpu, &ctxt->exception);
4766 else if (ctxt->exception.error_code_valid)
4767 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4768 ctxt->exception.error_code);
4770 kvm_queue_exception(vcpu, ctxt->exception.vector);
4773 static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
4774 const unsigned long *regs)
4776 memset(&ctxt->twobyte, 0,
4777 (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4778 memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
4780 ctxt->fetch.start = 0;
4781 ctxt->fetch.end = 0;
4782 ctxt->io_read.pos = 0;
4783 ctxt->io_read.end = 0;
4784 ctxt->mem_read.pos = 0;
4785 ctxt->mem_read.end = 0;
4788 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4790 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4794 * TODO: fix emulate.c to use guest_read/write_register
4795 * instead of direct ->regs accesses, can save hundred cycles
4796 * on Intel for instructions that don't read/change RSP, for
4799 cache_all_regs(vcpu);
4801 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4803 ctxt->eflags = kvm_get_rflags(vcpu);
4804 ctxt->eip = kvm_rip_read(vcpu);
4805 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4806 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4807 cs_l ? X86EMUL_MODE_PROT64 :
4808 cs_db ? X86EMUL_MODE_PROT32 :
4809 X86EMUL_MODE_PROT16;
4810 ctxt->guest_mode = is_guest_mode(vcpu);
4812 init_decode_cache(ctxt, vcpu->arch.regs);
4813 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4816 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4818 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4821 init_emulate_ctxt(vcpu);
4825 ctxt->_eip = ctxt->eip + inc_eip;
4826 ret = emulate_int_real(ctxt, irq);
4828 if (ret != X86EMUL_CONTINUE)
4829 return EMULATE_FAIL;
4831 ctxt->eip = ctxt->_eip;
4832 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4833 kvm_rip_write(vcpu, ctxt->eip);
4834 kvm_set_rflags(vcpu, ctxt->eflags);
4836 if (irq == NMI_VECTOR)
4837 vcpu->arch.nmi_pending = 0;
4839 vcpu->arch.interrupt.pending = false;
4841 return EMULATE_DONE;
4843 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4845 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4847 int r = EMULATE_DONE;
4849 ++vcpu->stat.insn_emulation_fail;
4850 trace_kvm_emulate_insn_failed(vcpu);
4851 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
4852 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4853 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4854 vcpu->run->internal.ndata = 0;
4857 kvm_queue_exception(vcpu, UD_VECTOR);
4862 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4870 * if emulation was due to access to shadowed page table
4871 * and it failed try to unshadow page and re-entetr the
4872 * guest to let CPU execute the instruction.
4874 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4877 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4879 if (gpa == UNMAPPED_GVA)
4880 return true; /* let cpu generate fault */
4882 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4888 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4895 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4896 bool writeback = true;
4898 kvm_clear_exception_queue(vcpu);
4900 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4901 init_emulate_ctxt(vcpu);
4902 ctxt->interruptibility = 0;
4903 ctxt->have_exception = false;
4904 ctxt->perm_ok = false;
4906 ctxt->only_vendor_specific_insn
4907 = emulation_type & EMULTYPE_TRAP_UD;
4909 r = x86_decode_insn(ctxt, insn, insn_len);
4911 trace_kvm_emulate_insn_start(vcpu);
4912 ++vcpu->stat.insn_emulation;
4913 if (r != EMULATION_OK) {
4914 if (emulation_type & EMULTYPE_TRAP_UD)
4915 return EMULATE_FAIL;
4916 if (reexecute_instruction(vcpu, cr2))
4917 return EMULATE_DONE;
4918 if (emulation_type & EMULTYPE_SKIP)
4919 return EMULATE_FAIL;
4920 return handle_emulation_failure(vcpu);
4924 if (emulation_type & EMULTYPE_SKIP) {
4925 kvm_rip_write(vcpu, ctxt->_eip);
4926 return EMULATE_DONE;
4929 /* this is needed for vmware backdoor interface to work since it
4930 changes registers values during IO operation */
4931 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4932 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4933 memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
4937 r = x86_emulate_insn(ctxt);
4939 if (r == EMULATION_INTERCEPTED)
4940 return EMULATE_DONE;
4942 if (r == EMULATION_FAILED) {
4943 if (reexecute_instruction(vcpu, cr2))
4944 return EMULATE_DONE;
4946 return handle_emulation_failure(vcpu);
4949 if (ctxt->have_exception) {
4950 inject_emulated_exception(vcpu);
4952 } else if (vcpu->arch.pio.count) {
4953 if (!vcpu->arch.pio.in)
4954 vcpu->arch.pio.count = 0;
4957 r = EMULATE_DO_MMIO;
4958 } else if (vcpu->mmio_needed) {
4959 if (!vcpu->mmio_is_write)
4961 r = EMULATE_DO_MMIO;
4962 } else if (r == EMULATION_RESTART)
4968 toggle_interruptibility(vcpu, ctxt->interruptibility);
4969 kvm_set_rflags(vcpu, ctxt->eflags);
4970 kvm_make_request(KVM_REQ_EVENT, vcpu);
4971 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4972 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4973 kvm_rip_write(vcpu, ctxt->eip);
4975 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4979 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4981 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4983 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4984 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4985 size, port, &val, 1);
4986 /* do not return to emulator after return from userspace */
4987 vcpu->arch.pio.count = 0;
4990 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4992 static void tsc_bad(void *info)
4994 __this_cpu_write(cpu_tsc_khz, 0);
4997 static void tsc_khz_changed(void *data)
4999 struct cpufreq_freqs *freq = data;
5000 unsigned long khz = 0;
5004 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5005 khz = cpufreq_quick_get(raw_smp_processor_id());
5008 __this_cpu_write(cpu_tsc_khz, khz);
5011 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5014 struct cpufreq_freqs *freq = data;
5016 struct kvm_vcpu *vcpu;
5017 int i, send_ipi = 0;
5020 * We allow guests to temporarily run on slowing clocks,
5021 * provided we notify them after, or to run on accelerating
5022 * clocks, provided we notify them before. Thus time never
5025 * However, we have a problem. We can't atomically update
5026 * the frequency of a given CPU from this function; it is
5027 * merely a notifier, which can be called from any CPU.
5028 * Changing the TSC frequency at arbitrary points in time
5029 * requires a recomputation of local variables related to
5030 * the TSC for each VCPU. We must flag these local variables
5031 * to be updated and be sure the update takes place with the
5032 * new frequency before any guests proceed.
5034 * Unfortunately, the combination of hotplug CPU and frequency
5035 * change creates an intractable locking scenario; the order
5036 * of when these callouts happen is undefined with respect to
5037 * CPU hotplug, and they can race with each other. As such,
5038 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5039 * undefined; you can actually have a CPU frequency change take
5040 * place in between the computation of X and the setting of the
5041 * variable. To protect against this problem, all updates of
5042 * the per_cpu tsc_khz variable are done in an interrupt
5043 * protected IPI, and all callers wishing to update the value
5044 * must wait for a synchronous IPI to complete (which is trivial
5045 * if the caller is on the CPU already). This establishes the
5046 * necessary total order on variable updates.
5048 * Note that because a guest time update may take place
5049 * anytime after the setting of the VCPU's request bit, the
5050 * correct TSC value must be set before the request. However,
5051 * to ensure the update actually makes it to any guest which
5052 * starts running in hardware virtualization between the set
5053 * and the acquisition of the spinlock, we must also ping the
5054 * CPU after setting the request bit.
5058 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5060 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5063 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5065 raw_spin_lock(&kvm_lock);
5066 list_for_each_entry(kvm, &vm_list, vm_list) {
5067 kvm_for_each_vcpu(i, vcpu, kvm) {
5068 if (vcpu->cpu != freq->cpu)
5070 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5071 if (vcpu->cpu != smp_processor_id())
5075 raw_spin_unlock(&kvm_lock);
5077 if (freq->old < freq->new && send_ipi) {
5079 * We upscale the frequency. Must make the guest
5080 * doesn't see old kvmclock values while running with
5081 * the new frequency, otherwise we risk the guest sees
5082 * time go backwards.
5084 * In case we update the frequency for another cpu
5085 * (which might be in guest context) send an interrupt
5086 * to kick the cpu out of guest context. Next time
5087 * guest context is entered kvmclock will be updated,
5088 * so the guest will not see stale values.
5090 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5095 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5096 .notifier_call = kvmclock_cpufreq_notifier
5099 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5100 unsigned long action, void *hcpu)
5102 unsigned int cpu = (unsigned long)hcpu;
5106 case CPU_DOWN_FAILED:
5107 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5109 case CPU_DOWN_PREPARE:
5110 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5116 static struct notifier_block kvmclock_cpu_notifier_block = {
5117 .notifier_call = kvmclock_cpu_notifier,
5118 .priority = -INT_MAX
5121 static void kvm_timer_init(void)
5125 max_tsc_khz = tsc_khz;
5126 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5127 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5128 #ifdef CONFIG_CPU_FREQ
5129 struct cpufreq_policy policy;
5130 memset(&policy, 0, sizeof(policy));
5132 cpufreq_get_policy(&policy, cpu);
5133 if (policy.cpuinfo.max_freq)
5134 max_tsc_khz = policy.cpuinfo.max_freq;
5137 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5138 CPUFREQ_TRANSITION_NOTIFIER);
5140 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5141 for_each_online_cpu(cpu)
5142 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5145 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5147 static int kvm_is_in_guest(void)
5149 return percpu_read(current_vcpu) != NULL;
5152 static int kvm_is_user_mode(void)
5156 if (percpu_read(current_vcpu))
5157 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
5159 return user_mode != 0;
5162 static unsigned long kvm_get_guest_ip(void)
5164 unsigned long ip = 0;
5166 if (percpu_read(current_vcpu))
5167 ip = kvm_rip_read(percpu_read(current_vcpu));
5172 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5173 .is_in_guest = kvm_is_in_guest,
5174 .is_user_mode = kvm_is_user_mode,
5175 .get_guest_ip = kvm_get_guest_ip,
5178 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5180 percpu_write(current_vcpu, vcpu);
5182 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5184 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5186 percpu_write(current_vcpu, NULL);
5188 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5190 static void kvm_set_mmio_spte_mask(void)
5193 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5196 * Set the reserved bits and the present bit of an paging-structure
5197 * entry to generate page fault with PFER.RSV = 1.
5199 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5202 #ifdef CONFIG_X86_64
5204 * If reserved bit is not supported, clear the present bit to disable
5207 if (maxphyaddr == 52)
5211 kvm_mmu_set_mmio_spte_mask(mask);
5214 int kvm_arch_init(void *opaque)
5217 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5220 printk(KERN_ERR "kvm: already loaded the other module\n");
5225 if (!ops->cpu_has_kvm_support()) {
5226 printk(KERN_ERR "kvm: no hardware support\n");
5230 if (ops->disabled_by_bios()) {
5231 printk(KERN_ERR "kvm: disabled by bios\n");
5236 r = kvm_mmu_module_init();
5240 kvm_set_mmio_spte_mask();
5241 kvm_init_msr_list();
5244 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5245 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5249 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5252 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5260 void kvm_arch_exit(void)
5262 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5264 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5265 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5266 CPUFREQ_TRANSITION_NOTIFIER);
5267 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5269 kvm_mmu_module_exit();
5272 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5274 ++vcpu->stat.halt_exits;
5275 if (irqchip_in_kernel(vcpu->kvm)) {
5276 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5279 vcpu->run->exit_reason = KVM_EXIT_HLT;
5283 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5285 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
5288 if (is_long_mode(vcpu))
5291 return a0 | ((gpa_t)a1 << 32);
5294 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5296 u64 param, ingpa, outgpa, ret;
5297 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5298 bool fast, longmode;
5302 * hypercall generates UD from non zero cpl and real mode
5305 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5306 kvm_queue_exception(vcpu, UD_VECTOR);
5310 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5311 longmode = is_long_mode(vcpu) && cs_l == 1;
5314 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5315 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5316 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5317 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5318 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5319 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5321 #ifdef CONFIG_X86_64
5323 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5324 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5325 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5329 code = param & 0xffff;
5330 fast = (param >> 16) & 0x1;
5331 rep_cnt = (param >> 32) & 0xfff;
5332 rep_idx = (param >> 48) & 0xfff;
5334 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5337 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5338 kvm_vcpu_on_spin(vcpu);
5341 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5345 ret = res | (((u64)rep_done & 0xfff) << 32);
5347 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5349 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5350 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5356 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5358 unsigned long nr, a0, a1, a2, a3, ret;
5361 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5362 return kvm_hv_hypercall(vcpu);
5364 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5365 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5366 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5367 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5368 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5370 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5372 if (!is_long_mode(vcpu)) {
5380 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5386 case KVM_HC_VAPIC_POLL_IRQ:
5390 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5397 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5398 ++vcpu->stat.hypercalls;
5401 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5403 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5405 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5406 char instruction[3];
5407 unsigned long rip = kvm_rip_read(vcpu);
5410 * Blow out the MMU to ensure that no other VCPU has an active mapping
5411 * to ensure that the updated hypercall appears atomically across all
5414 kvm_mmu_zap_all(vcpu->kvm);
5416 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5418 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5421 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5423 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5424 int j, nent = vcpu->arch.cpuid_nent;
5426 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5427 /* when no next entry is found, the current entry[i] is reselected */
5428 for (j = i + 1; ; j = (j + 1) % nent) {
5429 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5430 if (ej->function == e->function) {
5431 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5435 return 0; /* silence gcc, even though control never reaches here */
5438 /* find an entry with matching function, matching index (if needed), and that
5439 * should be read next (if it's stateful) */
5440 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5441 u32 function, u32 index)
5443 if (e->function != function)
5445 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5447 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5448 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5453 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5454 u32 function, u32 index)
5457 struct kvm_cpuid_entry2 *best = NULL;
5459 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5460 struct kvm_cpuid_entry2 *e;
5462 e = &vcpu->arch.cpuid_entries[i];
5463 if (is_matching_cpuid_entry(e, function, index)) {
5464 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5465 move_to_next_stateful_cpuid_entry(vcpu, i);
5472 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5474 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5476 struct kvm_cpuid_entry2 *best;
5478 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5479 if (!best || best->eax < 0x80000008)
5481 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5483 return best->eax & 0xff;
5489 * If no match is found, check whether we exceed the vCPU's limit
5490 * and return the content of the highest valid _standard_ leaf instead.
5491 * This is to satisfy the CPUID specification.
5493 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5494 u32 function, u32 index)
5496 struct kvm_cpuid_entry2 *maxlevel;
5498 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5499 if (!maxlevel || maxlevel->eax >= function)
5501 if (function & 0x80000000) {
5502 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5506 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5509 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5511 u32 function, index;
5512 struct kvm_cpuid_entry2 *best;
5514 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5515 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5516 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5517 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5518 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5519 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5520 best = kvm_find_cpuid_entry(vcpu, function, index);
5523 best = check_cpuid_limit(vcpu, function, index);
5526 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5527 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5528 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5529 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5531 kvm_x86_ops->skip_emulated_instruction(vcpu);
5532 trace_kvm_cpuid(function,
5533 kvm_register_read(vcpu, VCPU_REGS_RAX),
5534 kvm_register_read(vcpu, VCPU_REGS_RBX),
5535 kvm_register_read(vcpu, VCPU_REGS_RCX),
5536 kvm_register_read(vcpu, VCPU_REGS_RDX));
5538 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5541 * Check if userspace requested an interrupt window, and that the
5542 * interrupt window is open.
5544 * No need to exit to userspace if we already have an interrupt queued.
5546 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5548 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5549 vcpu->run->request_interrupt_window &&
5550 kvm_arch_interrupt_allowed(vcpu));
5553 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5555 struct kvm_run *kvm_run = vcpu->run;
5557 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5558 kvm_run->cr8 = kvm_get_cr8(vcpu);
5559 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5560 if (irqchip_in_kernel(vcpu->kvm))
5561 kvm_run->ready_for_interrupt_injection = 1;
5563 kvm_run->ready_for_interrupt_injection =
5564 kvm_arch_interrupt_allowed(vcpu) &&
5565 !kvm_cpu_has_interrupt(vcpu) &&
5566 !kvm_event_needs_reinjection(vcpu);
5569 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5573 if (!kvm_x86_ops->update_cr8_intercept)
5576 if (!vcpu->arch.apic)
5579 if (!vcpu->arch.apic->vapic_addr)
5580 max_irr = kvm_lapic_find_highest_irr(vcpu);
5587 tpr = kvm_lapic_get_cr8(vcpu);
5589 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5592 static void inject_pending_event(struct kvm_vcpu *vcpu)
5594 /* try to reinject previous events if any */
5595 if (vcpu->arch.exception.pending) {
5596 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5597 vcpu->arch.exception.has_error_code,
5598 vcpu->arch.exception.error_code);
5599 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5600 vcpu->arch.exception.has_error_code,
5601 vcpu->arch.exception.error_code,
5602 vcpu->arch.exception.reinject);
5606 if (vcpu->arch.nmi_injected) {
5607 kvm_x86_ops->set_nmi(vcpu);
5611 if (vcpu->arch.interrupt.pending) {
5612 kvm_x86_ops->set_irq(vcpu);
5616 /* try to inject new event if pending */
5617 if (vcpu->arch.nmi_pending) {
5618 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5619 --vcpu->arch.nmi_pending;
5620 vcpu->arch.nmi_injected = true;
5621 kvm_x86_ops->set_nmi(vcpu);
5623 } else if (kvm_cpu_has_interrupt(vcpu)) {
5624 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5625 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5627 kvm_x86_ops->set_irq(vcpu);
5632 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5634 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5635 !vcpu->guest_xcr0_loaded) {
5636 /* kvm_set_xcr() also depends on this */
5637 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5638 vcpu->guest_xcr0_loaded = 1;
5642 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5644 if (vcpu->guest_xcr0_loaded) {
5645 if (vcpu->arch.xcr0 != host_xcr0)
5646 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5647 vcpu->guest_xcr0_loaded = 0;
5651 static void process_nmi(struct kvm_vcpu *vcpu)
5656 * x86 is limited to one NMI running, and one NMI pending after it.
5657 * If an NMI is already in progress, limit further NMIs to just one.
5658 * Otherwise, allow two (and we'll inject the first one immediately).
5660 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5663 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5664 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5665 kvm_make_request(KVM_REQ_EVENT, vcpu);
5668 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5671 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5672 vcpu->run->request_interrupt_window;
5674 if (vcpu->requests) {
5675 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5676 kvm_mmu_unload(vcpu);
5677 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5678 __kvm_migrate_timers(vcpu);
5679 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5680 r = kvm_guest_time_update(vcpu);
5684 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5685 kvm_mmu_sync_roots(vcpu);
5686 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5687 kvm_x86_ops->tlb_flush(vcpu);
5688 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5689 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5693 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5694 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5698 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5699 vcpu->fpu_active = 0;
5700 kvm_x86_ops->fpu_deactivate(vcpu);
5702 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5703 /* Page is swapped out. Do synthetic halt */
5704 vcpu->arch.apf.halted = true;
5708 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5709 record_steal_time(vcpu);
5710 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5715 r = kvm_mmu_reload(vcpu);
5719 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5720 inject_pending_event(vcpu);
5722 /* enable NMI/IRQ window open exits if needed */
5723 if (vcpu->arch.nmi_pending)
5724 kvm_x86_ops->enable_nmi_window(vcpu);
5725 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5726 kvm_x86_ops->enable_irq_window(vcpu);
5728 if (kvm_lapic_enabled(vcpu)) {
5729 update_cr8_intercept(vcpu);
5730 kvm_lapic_sync_to_vapic(vcpu);
5736 kvm_x86_ops->prepare_guest_switch(vcpu);
5737 if (vcpu->fpu_active)
5738 kvm_load_guest_fpu(vcpu);
5739 kvm_load_guest_xcr0(vcpu);
5741 vcpu->mode = IN_GUEST_MODE;
5743 /* We should set ->mode before check ->requests,
5744 * see the comment in make_all_cpus_request.
5748 local_irq_disable();
5750 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5751 || need_resched() || signal_pending(current)) {
5752 vcpu->mode = OUTSIDE_GUEST_MODE;
5756 kvm_x86_ops->cancel_injection(vcpu);
5761 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5765 if (unlikely(vcpu->arch.switch_db_regs)) {
5767 set_debugreg(vcpu->arch.eff_db[0], 0);
5768 set_debugreg(vcpu->arch.eff_db[1], 1);
5769 set_debugreg(vcpu->arch.eff_db[2], 2);
5770 set_debugreg(vcpu->arch.eff_db[3], 3);
5773 trace_kvm_entry(vcpu->vcpu_id);
5774 kvm_x86_ops->run(vcpu);
5777 * If the guest has used debug registers, at least dr7
5778 * will be disabled while returning to the host.
5779 * If we don't have active breakpoints in the host, we don't
5780 * care about the messed up debug address registers. But if
5781 * we have some of them active, restore the old state.
5783 if (hw_breakpoint_active())
5784 hw_breakpoint_restore();
5786 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
5788 vcpu->mode = OUTSIDE_GUEST_MODE;
5795 * We must have an instruction between local_irq_enable() and
5796 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5797 * the interrupt shadow. The stat.exits increment will do nicely.
5798 * But we need to prevent reordering, hence this barrier():
5806 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5809 * Profile KVM exit RIPs:
5811 if (unlikely(prof_on == KVM_PROFILING)) {
5812 unsigned long rip = kvm_rip_read(vcpu);
5813 profile_hit(KVM_PROFILING, (void *)rip);
5817 kvm_lapic_sync_from_vapic(vcpu);
5819 r = kvm_x86_ops->handle_exit(vcpu);
5825 static int __vcpu_run(struct kvm_vcpu *vcpu)
5828 struct kvm *kvm = vcpu->kvm;
5830 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5831 pr_debug("vcpu %d received sipi with vector # %x\n",
5832 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5833 kvm_lapic_reset(vcpu);
5834 r = kvm_arch_vcpu_reset(vcpu);
5837 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5840 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5844 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5845 !vcpu->arch.apf.halted)
5846 r = vcpu_enter_guest(vcpu);
5848 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5849 kvm_vcpu_block(vcpu);
5850 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5851 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5853 switch(vcpu->arch.mp_state) {
5854 case KVM_MP_STATE_HALTED:
5855 vcpu->arch.mp_state =
5856 KVM_MP_STATE_RUNNABLE;
5857 case KVM_MP_STATE_RUNNABLE:
5858 vcpu->arch.apf.halted = false;
5860 case KVM_MP_STATE_SIPI_RECEIVED:
5871 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5872 if (kvm_cpu_has_pending_timer(vcpu))
5873 kvm_inject_pending_timer_irqs(vcpu);
5875 if (dm_request_for_irq_injection(vcpu)) {
5877 vcpu->run->exit_reason = KVM_EXIT_INTR;
5878 ++vcpu->stat.request_irq_exits;
5881 kvm_check_async_pf_completion(vcpu);
5883 if (signal_pending(current)) {
5885 vcpu->run->exit_reason = KVM_EXIT_INTR;
5886 ++vcpu->stat.signal_exits;
5888 if (need_resched()) {
5889 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5891 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5895 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5900 static int complete_mmio(struct kvm_vcpu *vcpu)
5902 struct kvm_run *run = vcpu->run;
5905 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5908 if (vcpu->mmio_needed) {
5909 vcpu->mmio_needed = 0;
5910 if (!vcpu->mmio_is_write)
5911 memcpy(vcpu->mmio_data + vcpu->mmio_index,
5913 vcpu->mmio_index += 8;
5914 if (vcpu->mmio_index < vcpu->mmio_size) {
5915 run->exit_reason = KVM_EXIT_MMIO;
5916 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5917 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5918 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5919 run->mmio.is_write = vcpu->mmio_is_write;
5920 vcpu->mmio_needed = 1;
5923 if (vcpu->mmio_is_write)
5925 vcpu->mmio_read_completed = 1;
5927 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5928 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5929 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5930 if (r != EMULATE_DONE)
5935 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5940 if (!tsk_used_math(current) && init_fpu(current))
5943 if (vcpu->sigset_active)
5944 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5946 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5947 kvm_vcpu_block(vcpu);
5948 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5953 /* re-sync apic's tpr */
5954 if (!irqchip_in_kernel(vcpu->kvm)) {
5955 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5961 r = complete_mmio(vcpu);
5965 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5966 kvm_register_write(vcpu, VCPU_REGS_RAX,
5967 kvm_run->hypercall.ret);
5969 r = __vcpu_run(vcpu);
5972 post_kvm_run_save(vcpu);
5973 if (vcpu->sigset_active)
5974 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5979 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5981 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5983 * We are here if userspace calls get_regs() in the middle of
5984 * instruction emulation. Registers state needs to be copied
5985 * back from emulation context to vcpu. Usrapace shouldn't do
5986 * that usually, but some bad designed PV devices (vmware
5987 * backdoor interface) need this to work
5989 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5990 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5991 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5993 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5994 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5995 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5996 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5997 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5998 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5999 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6000 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6001 #ifdef CONFIG_X86_64
6002 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6003 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6004 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6005 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6006 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6007 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6008 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6009 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6012 regs->rip = kvm_rip_read(vcpu);
6013 regs->rflags = kvm_get_rflags(vcpu);
6018 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6020 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6021 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6023 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6024 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6025 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6026 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6027 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6028 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6029 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6030 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6031 #ifdef CONFIG_X86_64
6032 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6033 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6034 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6035 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6036 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6037 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6038 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6039 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6042 kvm_rip_write(vcpu, regs->rip);
6043 kvm_set_rflags(vcpu, regs->rflags);
6045 vcpu->arch.exception.pending = false;
6047 kvm_make_request(KVM_REQ_EVENT, vcpu);
6052 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6054 struct kvm_segment cs;
6056 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6060 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6062 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6063 struct kvm_sregs *sregs)
6067 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6068 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6069 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6070 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6071 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6072 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6074 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6075 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6077 kvm_x86_ops->get_idt(vcpu, &dt);
6078 sregs->idt.limit = dt.size;
6079 sregs->idt.base = dt.address;
6080 kvm_x86_ops->get_gdt(vcpu, &dt);
6081 sregs->gdt.limit = dt.size;
6082 sregs->gdt.base = dt.address;
6084 sregs->cr0 = kvm_read_cr0(vcpu);
6085 sregs->cr2 = vcpu->arch.cr2;
6086 sregs->cr3 = kvm_read_cr3(vcpu);
6087 sregs->cr4 = kvm_read_cr4(vcpu);
6088 sregs->cr8 = kvm_get_cr8(vcpu);
6089 sregs->efer = vcpu->arch.efer;
6090 sregs->apic_base = kvm_get_apic_base(vcpu);
6092 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6094 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6095 set_bit(vcpu->arch.interrupt.nr,
6096 (unsigned long *)sregs->interrupt_bitmap);
6101 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6102 struct kvm_mp_state *mp_state)
6104 mp_state->mp_state = vcpu->arch.mp_state;
6108 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6109 struct kvm_mp_state *mp_state)
6111 vcpu->arch.mp_state = mp_state->mp_state;
6112 kvm_make_request(KVM_REQ_EVENT, vcpu);
6116 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
6117 bool has_error_code, u32 error_code)
6119 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6122 init_emulate_ctxt(vcpu);
6124 ret = emulator_task_switch(ctxt, tss_selector, reason,
6125 has_error_code, error_code);
6128 return EMULATE_FAIL;
6130 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
6131 kvm_rip_write(vcpu, ctxt->eip);
6132 kvm_set_rflags(vcpu, ctxt->eflags);
6133 kvm_make_request(KVM_REQ_EVENT, vcpu);
6134 return EMULATE_DONE;
6136 EXPORT_SYMBOL_GPL(kvm_task_switch);
6138 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6139 struct kvm_sregs *sregs)
6141 int mmu_reset_needed = 0;
6142 int pending_vec, max_bits, idx;
6145 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6148 dt.size = sregs->idt.limit;
6149 dt.address = sregs->idt.base;
6150 kvm_x86_ops->set_idt(vcpu, &dt);
6151 dt.size = sregs->gdt.limit;
6152 dt.address = sregs->gdt.base;
6153 kvm_x86_ops->set_gdt(vcpu, &dt);
6155 vcpu->arch.cr2 = sregs->cr2;
6156 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6157 vcpu->arch.cr3 = sregs->cr3;
6158 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6160 kvm_set_cr8(vcpu, sregs->cr8);
6162 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6163 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6164 kvm_set_apic_base(vcpu, sregs->apic_base);
6166 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6167 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6168 vcpu->arch.cr0 = sregs->cr0;
6170 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6171 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6172 if (sregs->cr4 & X86_CR4_OSXSAVE)
6175 idx = srcu_read_lock(&vcpu->kvm->srcu);
6176 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6177 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6178 mmu_reset_needed = 1;
6180 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6182 if (mmu_reset_needed)
6183 kvm_mmu_reset_context(vcpu);
6185 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
6186 pending_vec = find_first_bit(
6187 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6188 if (pending_vec < max_bits) {
6189 kvm_queue_interrupt(vcpu, pending_vec, false);
6190 pr_debug("Set back pending irq %d\n", pending_vec);
6193 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6194 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6195 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6196 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6197 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6198 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6200 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6201 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6203 update_cr8_intercept(vcpu);
6205 /* Older userspace won't unhalt the vcpu on reset. */
6206 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6207 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6209 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6211 kvm_make_request(KVM_REQ_EVENT, vcpu);
6216 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6217 struct kvm_guest_debug *dbg)
6219 unsigned long rflags;
6222 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6224 if (vcpu->arch.exception.pending)
6226 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6227 kvm_queue_exception(vcpu, DB_VECTOR);
6229 kvm_queue_exception(vcpu, BP_VECTOR);
6233 * Read rflags as long as potentially injected trace flags are still
6236 rflags = kvm_get_rflags(vcpu);
6238 vcpu->guest_debug = dbg->control;
6239 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6240 vcpu->guest_debug = 0;
6242 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6243 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6244 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6245 vcpu->arch.switch_db_regs =
6246 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
6248 for (i = 0; i < KVM_NR_DB_REGS; i++)
6249 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6250 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
6253 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6254 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6255 get_segment_base(vcpu, VCPU_SREG_CS);
6258 * Trigger an rflags update that will inject or remove the trace
6261 kvm_set_rflags(vcpu, rflags);
6263 kvm_x86_ops->set_guest_debug(vcpu, dbg);
6273 * Translate a guest virtual address to a guest physical address.
6275 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6276 struct kvm_translation *tr)
6278 unsigned long vaddr = tr->linear_address;
6282 idx = srcu_read_lock(&vcpu->kvm->srcu);
6283 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6284 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6285 tr->physical_address = gpa;
6286 tr->valid = gpa != UNMAPPED_GVA;
6293 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6295 struct i387_fxsave_struct *fxsave =
6296 &vcpu->arch.guest_fpu.state->fxsave;
6298 memcpy(fpu->fpr, fxsave->st_space, 128);
6299 fpu->fcw = fxsave->cwd;
6300 fpu->fsw = fxsave->swd;
6301 fpu->ftwx = fxsave->twd;
6302 fpu->last_opcode = fxsave->fop;
6303 fpu->last_ip = fxsave->rip;
6304 fpu->last_dp = fxsave->rdp;
6305 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6310 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6312 struct i387_fxsave_struct *fxsave =
6313 &vcpu->arch.guest_fpu.state->fxsave;
6315 memcpy(fxsave->st_space, fpu->fpr, 128);
6316 fxsave->cwd = fpu->fcw;
6317 fxsave->swd = fpu->fsw;
6318 fxsave->twd = fpu->ftwx;
6319 fxsave->fop = fpu->last_opcode;
6320 fxsave->rip = fpu->last_ip;
6321 fxsave->rdp = fpu->last_dp;
6322 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6327 int fx_init(struct kvm_vcpu *vcpu)
6331 err = fpu_alloc(&vcpu->arch.guest_fpu);
6335 fpu_finit(&vcpu->arch.guest_fpu);
6338 * Ensure guest xcr0 is valid for loading
6340 vcpu->arch.xcr0 = XSTATE_FP;
6342 vcpu->arch.cr0 |= X86_CR0_ET;
6346 EXPORT_SYMBOL_GPL(fx_init);
6348 static void fx_free(struct kvm_vcpu *vcpu)
6350 fpu_free(&vcpu->arch.guest_fpu);
6353 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6355 if (vcpu->guest_fpu_loaded)
6359 * Restore all possible states in the guest,
6360 * and assume host would use all available bits.
6361 * Guest xcr0 would be loaded later.
6363 kvm_put_guest_xcr0(vcpu);
6364 vcpu->guest_fpu_loaded = 1;
6365 unlazy_fpu(current);
6366 fpu_restore_checking(&vcpu->arch.guest_fpu);
6370 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6372 kvm_put_guest_xcr0(vcpu);
6374 if (!vcpu->guest_fpu_loaded)
6377 vcpu->guest_fpu_loaded = 0;
6378 fpu_save_init(&vcpu->arch.guest_fpu);
6379 ++vcpu->stat.fpu_reload;
6380 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6384 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6386 kvmclock_reset(vcpu);
6388 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6390 kvm_x86_ops->vcpu_free(vcpu);
6393 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6396 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6397 printk_once(KERN_WARNING
6398 "kvm: SMP vm created on host with unstable TSC; "
6399 "guest TSC will not be reliable\n");
6400 return kvm_x86_ops->vcpu_create(kvm, id);
6403 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6407 vcpu->arch.mtrr_state.have_fixed = 1;
6409 r = kvm_arch_vcpu_reset(vcpu);
6411 r = kvm_mmu_setup(vcpu);
6417 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6419 vcpu->arch.apf.msr_val = 0;
6422 kvm_mmu_unload(vcpu);
6426 kvm_x86_ops->vcpu_free(vcpu);
6429 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6431 atomic_set(&vcpu->arch.nmi_queued, 0);
6432 vcpu->arch.nmi_pending = 0;
6433 vcpu->arch.nmi_injected = false;
6435 vcpu->arch.switch_db_regs = 0;
6436 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6437 vcpu->arch.dr6 = DR6_FIXED_1;
6438 vcpu->arch.dr7 = DR7_FIXED_1;
6440 kvm_make_request(KVM_REQ_EVENT, vcpu);
6441 vcpu->arch.apf.msr_val = 0;
6442 vcpu->arch.st.msr_val = 0;
6444 kvmclock_reset(vcpu);
6446 kvm_clear_async_pf_completion_queue(vcpu);
6447 kvm_async_pf_hash_reset(vcpu);
6448 vcpu->arch.apf.halted = false;
6450 return kvm_x86_ops->vcpu_reset(vcpu);
6453 int kvm_arch_hardware_enable(void *garbage)
6456 struct kvm_vcpu *vcpu;
6459 kvm_shared_msr_cpu_online();
6460 list_for_each_entry(kvm, &vm_list, vm_list)
6461 kvm_for_each_vcpu(i, vcpu, kvm)
6462 if (vcpu->cpu == smp_processor_id())
6463 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6464 return kvm_x86_ops->hardware_enable(garbage);
6467 void kvm_arch_hardware_disable(void *garbage)
6469 kvm_x86_ops->hardware_disable(garbage);
6470 drop_user_return_notifiers(garbage);
6473 int kvm_arch_hardware_setup(void)
6475 return kvm_x86_ops->hardware_setup();
6478 void kvm_arch_hardware_unsetup(void)
6480 kvm_x86_ops->hardware_unsetup();
6483 void kvm_arch_check_processor_compat(void *rtn)
6485 kvm_x86_ops->check_processor_compatibility(rtn);
6488 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6490 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6493 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6499 BUG_ON(vcpu->kvm == NULL);
6502 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6503 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6504 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6505 vcpu->arch.mmu.translate_gpa = translate_gpa;
6506 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6507 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6508 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6510 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6512 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6517 vcpu->arch.pio_data = page_address(page);
6519 kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6521 r = kvm_mmu_create(vcpu);
6523 goto fail_free_pio_data;
6525 if (irqchip_in_kernel(kvm)) {
6526 r = kvm_create_lapic(vcpu);
6528 goto fail_mmu_destroy;
6531 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6533 if (!vcpu->arch.mce_banks) {
6535 goto fail_free_lapic;
6537 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6539 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6540 goto fail_free_mce_banks;
6542 vcpu->arch.pv_time_enabled = false;
6543 kvm_async_pf_hash_reset(vcpu);
6546 fail_free_mce_banks:
6547 kfree(vcpu->arch.mce_banks);
6549 kvm_free_lapic(vcpu);
6551 kvm_mmu_destroy(vcpu);
6553 free_page((unsigned long)vcpu->arch.pio_data);
6558 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6562 kfree(vcpu->arch.mce_banks);
6563 kvm_free_lapic(vcpu);
6564 idx = srcu_read_lock(&vcpu->kvm->srcu);
6565 kvm_mmu_destroy(vcpu);
6566 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6567 free_page((unsigned long)vcpu->arch.pio_data);
6570 int kvm_arch_init_vm(struct kvm *kvm)
6572 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6573 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6575 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6576 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6578 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6583 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6586 kvm_mmu_unload(vcpu);
6590 static void kvm_free_vcpus(struct kvm *kvm)
6593 struct kvm_vcpu *vcpu;
6596 * Unpin any mmu pages first.
6598 kvm_for_each_vcpu(i, vcpu, kvm) {
6599 kvm_clear_async_pf_completion_queue(vcpu);
6600 kvm_unload_vcpu_mmu(vcpu);
6602 kvm_for_each_vcpu(i, vcpu, kvm)
6603 kvm_arch_vcpu_free(vcpu);
6605 mutex_lock(&kvm->lock);
6606 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6607 kvm->vcpus[i] = NULL;
6609 atomic_set(&kvm->online_vcpus, 0);
6610 mutex_unlock(&kvm->lock);
6613 void kvm_arch_sync_events(struct kvm *kvm)
6615 kvm_free_all_assigned_devices(kvm);
6619 void kvm_arch_destroy_vm(struct kvm *kvm)
6621 kvm_iommu_unmap_guest(kvm);
6622 kfree(kvm->arch.vpic);
6623 kfree(kvm->arch.vioapic);
6624 kvm_free_vcpus(kvm);
6625 if (kvm->arch.apic_access_page)
6626 put_page(kvm->arch.apic_access_page);
6627 if (kvm->arch.ept_identity_pagetable)
6628 put_page(kvm->arch.ept_identity_pagetable);
6631 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6632 struct kvm_memory_slot *memslot,
6633 struct kvm_memory_slot old,
6634 struct kvm_userspace_memory_region *mem,
6637 int npages = memslot->npages;
6638 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6640 /* Prevent internal slot pages from being moved by fork()/COW. */
6641 if (memslot->id >= KVM_MEMORY_SLOTS)
6642 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6644 /*To keep backward compatibility with older userspace,
6645 *x86 needs to hanlde !user_alloc case.
6648 if (npages && !old.rmap) {
6649 unsigned long userspace_addr;
6651 down_write(¤t->mm->mmap_sem);
6652 userspace_addr = do_mmap(NULL, 0,
6654 PROT_READ | PROT_WRITE,
6657 up_write(¤t->mm->mmap_sem);
6659 if (IS_ERR((void *)userspace_addr))
6660 return PTR_ERR((void *)userspace_addr);
6662 memslot->userspace_addr = userspace_addr;
6670 void kvm_arch_commit_memory_region(struct kvm *kvm,
6671 struct kvm_userspace_memory_region *mem,
6672 struct kvm_memory_slot old,
6676 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6678 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6681 down_write(¤t->mm->mmap_sem);
6682 ret = do_munmap(current->mm, old.userspace_addr,
6683 old.npages * PAGE_SIZE);
6684 up_write(¤t->mm->mmap_sem);
6687 "kvm_vm_ioctl_set_memory_region: "
6688 "failed to munmap memory\n");
6691 if (!kvm->arch.n_requested_mmu_pages)
6692 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6694 spin_lock(&kvm->mmu_lock);
6696 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6697 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6698 spin_unlock(&kvm->mmu_lock);
6701 void kvm_arch_flush_shadow(struct kvm *kvm)
6703 kvm_mmu_zap_all(kvm);
6704 kvm_reload_remote_mmus(kvm);
6707 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6709 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6710 !vcpu->arch.apf.halted)
6711 || !list_empty_careful(&vcpu->async_pf.done)
6712 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6713 || atomic_read(&vcpu->arch.nmi_queued) ||
6714 (kvm_arch_interrupt_allowed(vcpu) &&
6715 kvm_cpu_has_interrupt(vcpu));
6718 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6721 int cpu = vcpu->cpu;
6723 if (waitqueue_active(&vcpu->wq)) {
6724 wake_up_interruptible(&vcpu->wq);
6725 ++vcpu->stat.halt_wakeup;
6729 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6730 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6731 smp_send_reschedule(cpu);
6735 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6737 return kvm_x86_ops->interrupt_allowed(vcpu);
6740 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6742 unsigned long current_rip = kvm_rip_read(vcpu) +
6743 get_segment_base(vcpu, VCPU_SREG_CS);
6745 return current_rip == linear_rip;
6747 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6749 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6751 unsigned long rflags;
6753 rflags = kvm_x86_ops->get_rflags(vcpu);
6754 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6755 rflags &= ~X86_EFLAGS_TF;
6758 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6760 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6762 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6763 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6764 rflags |= X86_EFLAGS_TF;
6765 kvm_x86_ops->set_rflags(vcpu, rflags);
6766 kvm_make_request(KVM_REQ_EVENT, vcpu);
6768 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6770 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6774 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6775 is_error_page(work->page))
6778 r = kvm_mmu_reload(vcpu);
6782 if (!vcpu->arch.mmu.direct_map &&
6783 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6786 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6789 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6791 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6794 static inline u32 kvm_async_pf_next_probe(u32 key)
6796 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6799 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6801 u32 key = kvm_async_pf_hash_fn(gfn);
6803 while (vcpu->arch.apf.gfns[key] != ~0)
6804 key = kvm_async_pf_next_probe(key);
6806 vcpu->arch.apf.gfns[key] = gfn;
6809 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6812 u32 key = kvm_async_pf_hash_fn(gfn);
6814 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6815 (vcpu->arch.apf.gfns[key] != gfn &&
6816 vcpu->arch.apf.gfns[key] != ~0); i++)
6817 key = kvm_async_pf_next_probe(key);
6822 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6824 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6827 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6831 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6833 vcpu->arch.apf.gfns[i] = ~0;
6835 j = kvm_async_pf_next_probe(j);
6836 if (vcpu->arch.apf.gfns[j] == ~0)
6838 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6840 * k lies cyclically in ]i,j]
6842 * |....j i.k.| or |.k..j i...|
6844 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6845 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6850 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6853 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6857 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6858 struct kvm_async_pf *work)
6860 struct x86_exception fault;
6862 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6863 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6865 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6866 (vcpu->arch.apf.send_user_only &&
6867 kvm_x86_ops->get_cpl(vcpu) == 0))
6868 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6869 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6870 fault.vector = PF_VECTOR;
6871 fault.error_code_valid = true;
6872 fault.error_code = 0;
6873 fault.nested_page_fault = false;
6874 fault.address = work->arch.token;
6875 kvm_inject_page_fault(vcpu, &fault);
6879 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6880 struct kvm_async_pf *work)
6882 struct x86_exception fault;
6884 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6885 if (is_error_page(work->page))
6886 work->arch.token = ~0; /* broadcast wakeup */
6888 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6890 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6891 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6892 fault.vector = PF_VECTOR;
6893 fault.error_code_valid = true;
6894 fault.error_code = 0;
6895 fault.nested_page_fault = false;
6896 fault.address = work->arch.token;
6897 kvm_inject_page_fault(vcpu, &fault);
6899 vcpu->arch.apf.halted = false;
6902 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6904 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6907 return !kvm_event_needs_reinjection(vcpu) &&
6908 kvm_x86_ops->interrupt_allowed(vcpu);
6911 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6912 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6913 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6914 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6915 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6916 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6917 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6918 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6919 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6920 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6921 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6922 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);