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"
31 #include <linux/clocksource.h>
32 #include <linux/interrupt.h>
33 #include <linux/kvm.h>
35 #include <linux/vmalloc.h>
36 #include <linux/module.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <trace/events/kvm.h>
53 #define CREATE_TRACE_POINTS
56 #include <asm/debugreg.h>
62 #include <asm/fpu-internal.h> /* Ugh! */
64 #include <asm/pvclock.h>
65 #include <asm/div64.h>
67 #define MAX_IO_MSRS 256
68 #define KVM_MAX_MCE_BANKS 32
69 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
71 #define emul_to_vcpu(ctxt) \
72 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
75 * - enable syscall per default because its emulated by KVM
76 * - enable LME and LMA per default on 64 bit KVM
80 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
82 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
85 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
86 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
88 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
89 static void process_nmi(struct kvm_vcpu *vcpu);
91 struct kvm_x86_ops *kvm_x86_ops;
92 EXPORT_SYMBOL_GPL(kvm_x86_ops);
94 static bool ignore_msrs = 0;
95 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
97 bool kvm_has_tsc_control;
98 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
99 u32 kvm_max_guest_tsc_khz;
100 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
102 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
103 static u32 tsc_tolerance_ppm = 250;
104 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
106 #define KVM_NR_SHARED_MSRS 16
108 struct kvm_shared_msrs_global {
110 u32 msrs[KVM_NR_SHARED_MSRS];
113 struct kvm_shared_msrs {
114 struct user_return_notifier urn;
116 struct kvm_shared_msr_values {
119 } values[KVM_NR_SHARED_MSRS];
122 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
123 static struct kvm_shared_msrs __percpu *shared_msrs;
125 struct kvm_stats_debugfs_item debugfs_entries[] = {
126 { "pf_fixed", VCPU_STAT(pf_fixed) },
127 { "pf_guest", VCPU_STAT(pf_guest) },
128 { "tlb_flush", VCPU_STAT(tlb_flush) },
129 { "invlpg", VCPU_STAT(invlpg) },
130 { "exits", VCPU_STAT(exits) },
131 { "io_exits", VCPU_STAT(io_exits) },
132 { "mmio_exits", VCPU_STAT(mmio_exits) },
133 { "signal_exits", VCPU_STAT(signal_exits) },
134 { "irq_window", VCPU_STAT(irq_window_exits) },
135 { "nmi_window", VCPU_STAT(nmi_window_exits) },
136 { "halt_exits", VCPU_STAT(halt_exits) },
137 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
138 { "hypercalls", VCPU_STAT(hypercalls) },
139 { "request_irq", VCPU_STAT(request_irq_exits) },
140 { "irq_exits", VCPU_STAT(irq_exits) },
141 { "host_state_reload", VCPU_STAT(host_state_reload) },
142 { "efer_reload", VCPU_STAT(efer_reload) },
143 { "fpu_reload", VCPU_STAT(fpu_reload) },
144 { "insn_emulation", VCPU_STAT(insn_emulation) },
145 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
146 { "irq_injections", VCPU_STAT(irq_injections) },
147 { "nmi_injections", VCPU_STAT(nmi_injections) },
148 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
149 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
150 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
151 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
152 { "mmu_flooded", VM_STAT(mmu_flooded) },
153 { "mmu_recycled", VM_STAT(mmu_recycled) },
154 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
155 { "mmu_unsync", VM_STAT(mmu_unsync) },
156 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
157 { "largepages", VM_STAT(lpages) },
161 u64 __read_mostly host_xcr0;
163 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
165 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
168 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
169 vcpu->arch.apf.gfns[i] = ~0;
172 static void kvm_on_user_return(struct user_return_notifier *urn)
175 struct kvm_shared_msrs *locals
176 = container_of(urn, struct kvm_shared_msrs, urn);
177 struct kvm_shared_msr_values *values;
179 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
180 values = &locals->values[slot];
181 if (values->host != values->curr) {
182 wrmsrl(shared_msrs_global.msrs[slot], values->host);
183 values->curr = values->host;
186 locals->registered = false;
187 user_return_notifier_unregister(urn);
190 static void shared_msr_update(unsigned slot, u32 msr)
193 unsigned int cpu = smp_processor_id();
194 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
196 /* only read, and nobody should modify it at this time,
197 * so don't need lock */
198 if (slot >= shared_msrs_global.nr) {
199 printk(KERN_ERR "kvm: invalid MSR slot!");
202 rdmsrl_safe(msr, &value);
203 smsr->values[slot].host = value;
204 smsr->values[slot].curr = value;
207 void kvm_define_shared_msr(unsigned slot, u32 msr)
209 if (slot >= shared_msrs_global.nr)
210 shared_msrs_global.nr = slot + 1;
211 shared_msrs_global.msrs[slot] = msr;
212 /* we need ensured the shared_msr_global have been updated */
215 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
217 static void kvm_shared_msr_cpu_online(void)
221 for (i = 0; i < shared_msrs_global.nr; ++i)
222 shared_msr_update(i, shared_msrs_global.msrs[i]);
225 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
227 unsigned int cpu = smp_processor_id();
228 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
230 if (((value ^ smsr->values[slot].curr) & mask) == 0)
232 smsr->values[slot].curr = value;
233 wrmsrl(shared_msrs_global.msrs[slot], value);
234 if (!smsr->registered) {
235 smsr->urn.on_user_return = kvm_on_user_return;
236 user_return_notifier_register(&smsr->urn);
237 smsr->registered = true;
240 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
242 static void drop_user_return_notifiers(void *ignore)
244 unsigned int cpu = smp_processor_id();
245 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
247 if (smsr->registered)
248 kvm_on_user_return(&smsr->urn);
251 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
253 return vcpu->arch.apic_base;
255 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
257 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
259 /* TODO: reserve bits check */
260 kvm_lapic_set_base(vcpu, data);
262 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
264 asmlinkage void kvm_spurious_fault(void)
266 /* Fault while not rebooting. We want the trace. */
269 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
271 #define EXCPT_BENIGN 0
272 #define EXCPT_CONTRIBUTORY 1
275 static int exception_class(int vector)
285 return EXCPT_CONTRIBUTORY;
292 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
293 unsigned nr, bool has_error, u32 error_code,
299 kvm_make_request(KVM_REQ_EVENT, vcpu);
301 if (!vcpu->arch.exception.pending) {
303 vcpu->arch.exception.pending = true;
304 vcpu->arch.exception.has_error_code = has_error;
305 vcpu->arch.exception.nr = nr;
306 vcpu->arch.exception.error_code = error_code;
307 vcpu->arch.exception.reinject = reinject;
311 /* to check exception */
312 prev_nr = vcpu->arch.exception.nr;
313 if (prev_nr == DF_VECTOR) {
314 /* triple fault -> shutdown */
315 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
318 class1 = exception_class(prev_nr);
319 class2 = exception_class(nr);
320 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
321 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
322 /* generate double fault per SDM Table 5-5 */
323 vcpu->arch.exception.pending = true;
324 vcpu->arch.exception.has_error_code = true;
325 vcpu->arch.exception.nr = DF_VECTOR;
326 vcpu->arch.exception.error_code = 0;
328 /* replace previous exception with a new one in a hope
329 that instruction re-execution will regenerate lost
334 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
336 kvm_multiple_exception(vcpu, nr, false, 0, false);
338 EXPORT_SYMBOL_GPL(kvm_queue_exception);
340 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
342 kvm_multiple_exception(vcpu, nr, false, 0, true);
344 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
346 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
349 kvm_inject_gp(vcpu, 0);
351 kvm_x86_ops->skip_emulated_instruction(vcpu);
353 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
355 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
357 ++vcpu->stat.pf_guest;
358 vcpu->arch.cr2 = fault->address;
359 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
361 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
363 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
365 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
366 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
368 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
371 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
373 atomic_inc(&vcpu->arch.nmi_queued);
374 kvm_make_request(KVM_REQ_NMI, vcpu);
376 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
378 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
380 kvm_multiple_exception(vcpu, nr, true, error_code, false);
382 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
384 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
386 kvm_multiple_exception(vcpu, nr, true, error_code, true);
388 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
391 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
392 * a #GP and return false.
394 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
396 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
398 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
401 EXPORT_SYMBOL_GPL(kvm_require_cpl);
404 * This function will be used to read from the physical memory of the currently
405 * running guest. The difference to kvm_read_guest_page is that this function
406 * can read from guest physical or from the guest's guest physical memory.
408 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
409 gfn_t ngfn, void *data, int offset, int len,
415 ngpa = gfn_to_gpa(ngfn);
416 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
417 if (real_gfn == UNMAPPED_GVA)
420 real_gfn = gpa_to_gfn(real_gfn);
422 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
424 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
426 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
427 void *data, int offset, int len, u32 access)
429 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
430 data, offset, len, access);
434 * Load the pae pdptrs. Return true is they are all valid.
436 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
438 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
439 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
442 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
444 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
445 offset * sizeof(u64), sizeof(pdpte),
446 PFERR_USER_MASK|PFERR_WRITE_MASK);
451 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
452 if (is_present_gpte(pdpte[i]) &&
453 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
460 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
461 __set_bit(VCPU_EXREG_PDPTR,
462 (unsigned long *)&vcpu->arch.regs_avail);
463 __set_bit(VCPU_EXREG_PDPTR,
464 (unsigned long *)&vcpu->arch.regs_dirty);
469 EXPORT_SYMBOL_GPL(load_pdptrs);
471 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
473 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
479 if (is_long_mode(vcpu) || !is_pae(vcpu))
482 if (!test_bit(VCPU_EXREG_PDPTR,
483 (unsigned long *)&vcpu->arch.regs_avail))
486 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
487 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
488 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
489 PFERR_USER_MASK | PFERR_WRITE_MASK);
492 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
498 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
500 unsigned long old_cr0 = kvm_read_cr0(vcpu);
501 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
502 X86_CR0_CD | X86_CR0_NW;
507 if (cr0 & 0xffffffff00000000UL)
511 cr0 &= ~CR0_RESERVED_BITS;
513 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
516 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
519 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
521 if ((vcpu->arch.efer & EFER_LME)) {
526 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
531 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
536 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
539 kvm_x86_ops->set_cr0(vcpu, cr0);
541 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
542 kvm_clear_async_pf_completion_queue(vcpu);
543 kvm_async_pf_hash_reset(vcpu);
546 if ((cr0 ^ old_cr0) & update_bits)
547 kvm_mmu_reset_context(vcpu);
550 EXPORT_SYMBOL_GPL(kvm_set_cr0);
552 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
554 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
556 EXPORT_SYMBOL_GPL(kvm_lmsw);
558 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
560 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
561 !vcpu->guest_xcr0_loaded) {
562 /* kvm_set_xcr() also depends on this */
563 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
564 vcpu->guest_xcr0_loaded = 1;
568 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
570 if (vcpu->guest_xcr0_loaded) {
571 if (vcpu->arch.xcr0 != host_xcr0)
572 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
573 vcpu->guest_xcr0_loaded = 0;
577 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
581 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
582 if (index != XCR_XFEATURE_ENABLED_MASK)
585 if (!(xcr0 & XSTATE_FP))
587 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
589 if (xcr0 & ~host_xcr0)
591 kvm_put_guest_xcr0(vcpu);
592 vcpu->arch.xcr0 = xcr0;
596 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
598 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
599 __kvm_set_xcr(vcpu, index, xcr)) {
600 kvm_inject_gp(vcpu, 0);
605 EXPORT_SYMBOL_GPL(kvm_set_xcr);
607 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
609 unsigned long old_cr4 = kvm_read_cr4(vcpu);
610 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
611 X86_CR4_PAE | X86_CR4_SMEP;
612 if (cr4 & CR4_RESERVED_BITS)
615 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
618 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
621 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
624 if (is_long_mode(vcpu)) {
625 if (!(cr4 & X86_CR4_PAE))
627 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
628 && ((cr4 ^ old_cr4) & pdptr_bits)
629 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
633 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
634 if (!guest_cpuid_has_pcid(vcpu))
637 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
638 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
642 if (kvm_x86_ops->set_cr4(vcpu, cr4))
645 if (((cr4 ^ old_cr4) & pdptr_bits) ||
646 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
647 kvm_mmu_reset_context(vcpu);
649 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
650 kvm_update_cpuid(vcpu);
654 EXPORT_SYMBOL_GPL(kvm_set_cr4);
656 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
658 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
659 kvm_mmu_sync_roots(vcpu);
660 kvm_mmu_flush_tlb(vcpu);
664 if (is_long_mode(vcpu)) {
665 if (kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) {
666 if (cr3 & CR3_PCID_ENABLED_RESERVED_BITS)
669 if (cr3 & CR3_L_MODE_RESERVED_BITS)
673 if (cr3 & CR3_PAE_RESERVED_BITS)
675 if (is_paging(vcpu) &&
676 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
680 * We don't check reserved bits in nonpae mode, because
681 * this isn't enforced, and VMware depends on this.
685 vcpu->arch.cr3 = cr3;
686 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
687 vcpu->arch.mmu.new_cr3(vcpu);
690 EXPORT_SYMBOL_GPL(kvm_set_cr3);
692 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
694 if (cr8 & CR8_RESERVED_BITS)
696 if (irqchip_in_kernel(vcpu->kvm))
697 kvm_lapic_set_tpr(vcpu, cr8);
699 vcpu->arch.cr8 = cr8;
702 EXPORT_SYMBOL_GPL(kvm_set_cr8);
704 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
706 if (irqchip_in_kernel(vcpu->kvm))
707 return kvm_lapic_get_cr8(vcpu);
709 return vcpu->arch.cr8;
711 EXPORT_SYMBOL_GPL(kvm_get_cr8);
713 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
717 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
718 dr7 = vcpu->arch.guest_debug_dr7;
720 dr7 = vcpu->arch.dr7;
721 kvm_x86_ops->set_dr7(vcpu, dr7);
722 vcpu->arch.switch_db_regs = (dr7 & DR7_BP_EN_MASK);
725 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
729 vcpu->arch.db[dr] = val;
730 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
731 vcpu->arch.eff_db[dr] = val;
734 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
738 if (val & 0xffffffff00000000ULL)
740 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
743 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
747 if (val & 0xffffffff00000000ULL)
749 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
750 kvm_update_dr7(vcpu);
757 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
761 res = __kvm_set_dr(vcpu, dr, val);
763 kvm_queue_exception(vcpu, UD_VECTOR);
765 kvm_inject_gp(vcpu, 0);
769 EXPORT_SYMBOL_GPL(kvm_set_dr);
771 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
775 *val = vcpu->arch.db[dr];
778 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
782 *val = vcpu->arch.dr6;
785 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
789 *val = vcpu->arch.dr7;
796 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
798 if (_kvm_get_dr(vcpu, dr, val)) {
799 kvm_queue_exception(vcpu, UD_VECTOR);
804 EXPORT_SYMBOL_GPL(kvm_get_dr);
806 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
808 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
812 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
815 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
816 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
819 EXPORT_SYMBOL_GPL(kvm_rdpmc);
822 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
823 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
825 * This list is modified at module load time to reflect the
826 * capabilities of the host cpu. This capabilities test skips MSRs that are
827 * kvm-specific. Those are put in the beginning of the list.
830 #define KVM_SAVE_MSRS_BEGIN 10
831 static u32 msrs_to_save[] = {
832 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
833 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
834 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
835 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
837 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
840 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
842 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
843 MSR_IA32_FEATURE_CONTROL
846 static unsigned num_msrs_to_save;
848 static const u32 emulated_msrs[] = {
850 MSR_IA32_TSCDEADLINE,
851 MSR_IA32_MISC_ENABLE,
856 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
858 if (efer & efer_reserved_bits)
861 if (efer & EFER_FFXSR) {
862 struct kvm_cpuid_entry2 *feat;
864 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
865 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
869 if (efer & EFER_SVME) {
870 struct kvm_cpuid_entry2 *feat;
872 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
873 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
879 EXPORT_SYMBOL_GPL(kvm_valid_efer);
881 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
883 u64 old_efer = vcpu->arch.efer;
885 if (!kvm_valid_efer(vcpu, efer))
889 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
893 efer |= vcpu->arch.efer & EFER_LMA;
895 kvm_x86_ops->set_efer(vcpu, efer);
897 /* Update reserved bits */
898 if ((efer ^ old_efer) & EFER_NX)
899 kvm_mmu_reset_context(vcpu);
904 void kvm_enable_efer_bits(u64 mask)
906 efer_reserved_bits &= ~mask;
908 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
912 * Writes msr value into into the appropriate "register".
913 * Returns 0 on success, non-0 otherwise.
914 * Assumes vcpu_load() was already called.
916 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
918 return kvm_x86_ops->set_msr(vcpu, msr);
922 * Adapt set_msr() to msr_io()'s calling convention
924 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
930 msr.host_initiated = true;
931 return kvm_set_msr(vcpu, &msr);
935 struct pvclock_gtod_data {
938 struct { /* extract of a clocksource struct */
946 /* open coded 'struct timespec' */
947 u64 monotonic_time_snsec;
948 time_t monotonic_time_sec;
951 static struct pvclock_gtod_data pvclock_gtod_data;
953 static void update_pvclock_gtod(struct timekeeper *tk)
955 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
957 write_seqcount_begin(&vdata->seq);
959 /* copy pvclock gtod data */
960 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
961 vdata->clock.cycle_last = tk->clock->cycle_last;
962 vdata->clock.mask = tk->clock->mask;
963 vdata->clock.mult = tk->mult;
964 vdata->clock.shift = tk->shift;
966 vdata->monotonic_time_sec = tk->xtime_sec
967 + tk->wall_to_monotonic.tv_sec;
968 vdata->monotonic_time_snsec = tk->xtime_nsec
969 + (tk->wall_to_monotonic.tv_nsec
971 while (vdata->monotonic_time_snsec >=
972 (((u64)NSEC_PER_SEC) << tk->shift)) {
973 vdata->monotonic_time_snsec -=
974 ((u64)NSEC_PER_SEC) << tk->shift;
975 vdata->monotonic_time_sec++;
978 write_seqcount_end(&vdata->seq);
983 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
987 struct pvclock_wall_clock wc;
988 struct timespec boot;
993 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
998 ++version; /* first time write, random junk */
1002 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1005 * The guest calculates current wall clock time by adding
1006 * system time (updated by kvm_guest_time_update below) to the
1007 * wall clock specified here. guest system time equals host
1008 * system time for us, thus we must fill in host boot time here.
1012 if (kvm->arch.kvmclock_offset) {
1013 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1014 boot = timespec_sub(boot, ts);
1016 wc.sec = boot.tv_sec;
1017 wc.nsec = boot.tv_nsec;
1018 wc.version = version;
1020 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1023 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1026 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1028 uint32_t quotient, remainder;
1030 /* Don't try to replace with do_div(), this one calculates
1031 * "(dividend << 32) / divisor" */
1033 : "=a" (quotient), "=d" (remainder)
1034 : "0" (0), "1" (dividend), "r" (divisor) );
1038 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1039 s8 *pshift, u32 *pmultiplier)
1046 tps64 = base_khz * 1000LL;
1047 scaled64 = scaled_khz * 1000LL;
1048 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1053 tps32 = (uint32_t)tps64;
1054 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1055 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1063 *pmultiplier = div_frac(scaled64, tps32);
1065 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1066 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1069 static inline u64 get_kernel_ns(void)
1073 WARN_ON(preemptible());
1075 monotonic_to_bootbased(&ts);
1076 return timespec_to_ns(&ts);
1079 #ifdef CONFIG_X86_64
1080 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1083 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1084 unsigned long max_tsc_khz;
1086 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1088 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1089 vcpu->arch.virtual_tsc_shift);
1092 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1094 u64 v = (u64)khz * (1000000 + ppm);
1099 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1101 u32 thresh_lo, thresh_hi;
1102 int use_scaling = 0;
1104 /* tsc_khz can be zero if TSC calibration fails */
1105 if (this_tsc_khz == 0)
1108 /* Compute a scale to convert nanoseconds in TSC cycles */
1109 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1110 &vcpu->arch.virtual_tsc_shift,
1111 &vcpu->arch.virtual_tsc_mult);
1112 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1115 * Compute the variation in TSC rate which is acceptable
1116 * within the range of tolerance and decide if the
1117 * rate being applied is within that bounds of the hardware
1118 * rate. If so, no scaling or compensation need be done.
1120 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1121 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1122 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1123 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1126 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1129 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1131 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1132 vcpu->arch.virtual_tsc_mult,
1133 vcpu->arch.virtual_tsc_shift);
1134 tsc += vcpu->arch.this_tsc_write;
1138 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1140 #ifdef CONFIG_X86_64
1142 bool do_request = false;
1143 struct kvm_arch *ka = &vcpu->kvm->arch;
1144 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1146 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1147 atomic_read(&vcpu->kvm->online_vcpus));
1149 if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1150 if (!ka->use_master_clock)
1153 if (!vcpus_matched && ka->use_master_clock)
1157 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1159 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1160 atomic_read(&vcpu->kvm->online_vcpus),
1161 ka->use_master_clock, gtod->clock.vclock_mode);
1165 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1167 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1168 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1171 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1173 struct kvm *kvm = vcpu->kvm;
1174 u64 offset, ns, elapsed;
1175 unsigned long flags;
1178 u64 data = msr->data;
1180 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1181 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1182 ns = get_kernel_ns();
1183 elapsed = ns - kvm->arch.last_tsc_nsec;
1185 if (vcpu->arch.virtual_tsc_khz) {
1188 /* n.b - signed multiplication and division required */
1189 usdiff = data - kvm->arch.last_tsc_write;
1190 #ifdef CONFIG_X86_64
1191 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1193 /* do_div() only does unsigned */
1194 asm("1: idivl %[divisor]\n"
1195 "2: xor %%edx, %%edx\n"
1196 " movl $0, %[faulted]\n"
1198 ".section .fixup,\"ax\"\n"
1199 "4: movl $1, %[faulted]\n"
1203 _ASM_EXTABLE(1b, 4b)
1205 : "=A"(usdiff), [faulted] "=r" (faulted)
1206 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1209 do_div(elapsed, 1000);
1214 /* idivl overflow => difference is larger than USEC_PER_SEC */
1216 usdiff = USEC_PER_SEC;
1218 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1221 * Special case: TSC write with a small delta (1 second) of virtual
1222 * cycle time against real time is interpreted as an attempt to
1223 * synchronize the CPU.
1225 * For a reliable TSC, we can match TSC offsets, and for an unstable
1226 * TSC, we add elapsed time in this computation. We could let the
1227 * compensation code attempt to catch up if we fall behind, but
1228 * it's better to try to match offsets from the beginning.
1230 if (usdiff < USEC_PER_SEC &&
1231 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1232 if (!check_tsc_unstable()) {
1233 offset = kvm->arch.cur_tsc_offset;
1234 pr_debug("kvm: matched tsc offset for %llu\n", data);
1236 u64 delta = nsec_to_cycles(vcpu, elapsed);
1238 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1239 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1244 * We split periods of matched TSC writes into generations.
1245 * For each generation, we track the original measured
1246 * nanosecond time, offset, and write, so if TSCs are in
1247 * sync, we can match exact offset, and if not, we can match
1248 * exact software computation in compute_guest_tsc()
1250 * These values are tracked in kvm->arch.cur_xxx variables.
1252 kvm->arch.cur_tsc_generation++;
1253 kvm->arch.cur_tsc_nsec = ns;
1254 kvm->arch.cur_tsc_write = data;
1255 kvm->arch.cur_tsc_offset = offset;
1257 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1258 kvm->arch.cur_tsc_generation, data);
1262 * We also track th most recent recorded KHZ, write and time to
1263 * allow the matching interval to be extended at each write.
1265 kvm->arch.last_tsc_nsec = ns;
1266 kvm->arch.last_tsc_write = data;
1267 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1269 /* Reset of TSC must disable overshoot protection below */
1270 vcpu->arch.hv_clock.tsc_timestamp = 0;
1271 vcpu->arch.last_guest_tsc = data;
1273 /* Keep track of which generation this VCPU has synchronized to */
1274 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1275 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1276 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1278 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1279 update_ia32_tsc_adjust_msr(vcpu, offset);
1280 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1281 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1283 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1285 kvm->arch.nr_vcpus_matched_tsc++;
1287 kvm->arch.nr_vcpus_matched_tsc = 0;
1289 kvm_track_tsc_matching(vcpu);
1290 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1293 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1295 #ifdef CONFIG_X86_64
1297 static cycle_t read_tsc(void)
1303 * Empirically, a fence (of type that depends on the CPU)
1304 * before rdtsc is enough to ensure that rdtsc is ordered
1305 * with respect to loads. The various CPU manuals are unclear
1306 * as to whether rdtsc can be reordered with later loads,
1307 * but no one has ever seen it happen.
1310 ret = (cycle_t)vget_cycles();
1312 last = pvclock_gtod_data.clock.cycle_last;
1314 if (likely(ret >= last))
1318 * GCC likes to generate cmov here, but this branch is extremely
1319 * predictable (it's just a funciton of time and the likely is
1320 * very likely) and there's a data dependence, so force GCC
1321 * to generate a branch instead. I don't barrier() because
1322 * we don't actually need a barrier, and if this function
1323 * ever gets inlined it will generate worse code.
1329 static inline u64 vgettsc(cycle_t *cycle_now)
1332 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1334 *cycle_now = read_tsc();
1336 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1337 return v * gtod->clock.mult;
1340 static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1345 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1349 seq = read_seqcount_begin(>od->seq);
1350 mode = gtod->clock.vclock_mode;
1351 ts->tv_sec = gtod->monotonic_time_sec;
1352 ns = gtod->monotonic_time_snsec;
1353 ns += vgettsc(cycle_now);
1354 ns >>= gtod->clock.shift;
1355 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1356 timespec_add_ns(ts, ns);
1361 /* returns true if host is using tsc clocksource */
1362 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1366 /* checked again under seqlock below */
1367 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1370 if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1373 monotonic_to_bootbased(&ts);
1374 *kernel_ns = timespec_to_ns(&ts);
1382 * Assuming a stable TSC across physical CPUS, and a stable TSC
1383 * across virtual CPUs, the following condition is possible.
1384 * Each numbered line represents an event visible to both
1385 * CPUs at the next numbered event.
1387 * "timespecX" represents host monotonic time. "tscX" represents
1390 * VCPU0 on CPU0 | VCPU1 on CPU1
1392 * 1. read timespec0,tsc0
1393 * 2. | timespec1 = timespec0 + N
1395 * 3. transition to guest | transition to guest
1396 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1397 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1398 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1400 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1403 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1405 * - 0 < N - M => M < N
1407 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1408 * always the case (the difference between two distinct xtime instances
1409 * might be smaller then the difference between corresponding TSC reads,
1410 * when updating guest vcpus pvclock areas).
1412 * To avoid that problem, do not allow visibility of distinct
1413 * system_timestamp/tsc_timestamp values simultaneously: use a master
1414 * copy of host monotonic time values. Update that master copy
1417 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1421 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1423 #ifdef CONFIG_X86_64
1424 struct kvm_arch *ka = &kvm->arch;
1426 bool host_tsc_clocksource, vcpus_matched;
1428 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1429 atomic_read(&kvm->online_vcpus));
1432 * If the host uses TSC clock, then passthrough TSC as stable
1435 host_tsc_clocksource = kvm_get_time_and_clockread(
1436 &ka->master_kernel_ns,
1437 &ka->master_cycle_now);
1439 ka->use_master_clock = host_tsc_clocksource & vcpus_matched;
1441 if (ka->use_master_clock)
1442 atomic_set(&kvm_guest_has_master_clock, 1);
1444 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1445 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1450 static int kvm_guest_time_update(struct kvm_vcpu *v)
1452 unsigned long flags, this_tsc_khz;
1453 struct kvm_vcpu_arch *vcpu = &v->arch;
1454 struct kvm_arch *ka = &v->kvm->arch;
1455 s64 kernel_ns, max_kernel_ns;
1456 u64 tsc_timestamp, host_tsc;
1457 struct pvclock_vcpu_time_info guest_hv_clock;
1459 bool use_master_clock;
1465 * If the host uses TSC clock, then passthrough TSC as stable
1468 spin_lock(&ka->pvclock_gtod_sync_lock);
1469 use_master_clock = ka->use_master_clock;
1470 if (use_master_clock) {
1471 host_tsc = ka->master_cycle_now;
1472 kernel_ns = ka->master_kernel_ns;
1474 spin_unlock(&ka->pvclock_gtod_sync_lock);
1476 /* Keep irq disabled to prevent changes to the clock */
1477 local_irq_save(flags);
1478 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1479 if (unlikely(this_tsc_khz == 0)) {
1480 local_irq_restore(flags);
1481 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1484 if (!use_master_clock) {
1485 host_tsc = native_read_tsc();
1486 kernel_ns = get_kernel_ns();
1489 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1492 * We may have to catch up the TSC to match elapsed wall clock
1493 * time for two reasons, even if kvmclock is used.
1494 * 1) CPU could have been running below the maximum TSC rate
1495 * 2) Broken TSC compensation resets the base at each VCPU
1496 * entry to avoid unknown leaps of TSC even when running
1497 * again on the same CPU. This may cause apparent elapsed
1498 * time to disappear, and the guest to stand still or run
1501 if (vcpu->tsc_catchup) {
1502 u64 tsc = compute_guest_tsc(v, kernel_ns);
1503 if (tsc > tsc_timestamp) {
1504 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1505 tsc_timestamp = tsc;
1509 local_irq_restore(flags);
1511 if (!vcpu->pv_time_enabled)
1515 * Time as measured by the TSC may go backwards when resetting the base
1516 * tsc_timestamp. The reason for this is that the TSC resolution is
1517 * higher than the resolution of the other clock scales. Thus, many
1518 * possible measurments of the TSC correspond to one measurement of any
1519 * other clock, and so a spread of values is possible. This is not a
1520 * problem for the computation of the nanosecond clock; with TSC rates
1521 * around 1GHZ, there can only be a few cycles which correspond to one
1522 * nanosecond value, and any path through this code will inevitably
1523 * take longer than that. However, with the kernel_ns value itself,
1524 * the precision may be much lower, down to HZ granularity. If the
1525 * first sampling of TSC against kernel_ns ends in the low part of the
1526 * range, and the second in the high end of the range, we can get:
1528 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1530 * As the sampling errors potentially range in the thousands of cycles,
1531 * it is possible such a time value has already been observed by the
1532 * guest. To protect against this, we must compute the system time as
1533 * observed by the guest and ensure the new system time is greater.
1536 if (vcpu->hv_clock.tsc_timestamp) {
1537 max_kernel_ns = vcpu->last_guest_tsc -
1538 vcpu->hv_clock.tsc_timestamp;
1539 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1540 vcpu->hv_clock.tsc_to_system_mul,
1541 vcpu->hv_clock.tsc_shift);
1542 max_kernel_ns += vcpu->last_kernel_ns;
1545 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1546 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1547 &vcpu->hv_clock.tsc_shift,
1548 &vcpu->hv_clock.tsc_to_system_mul);
1549 vcpu->hw_tsc_khz = this_tsc_khz;
1552 /* with a master <monotonic time, tsc value> tuple,
1553 * pvclock clock reads always increase at the (scaled) rate
1554 * of guest TSC - no need to deal with sampling errors.
1556 if (!use_master_clock) {
1557 if (max_kernel_ns > kernel_ns)
1558 kernel_ns = max_kernel_ns;
1560 /* With all the info we got, fill in the values */
1561 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1562 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1563 vcpu->last_kernel_ns = kernel_ns;
1564 vcpu->last_guest_tsc = tsc_timestamp;
1567 * The interface expects us to write an even number signaling that the
1568 * update is finished. Since the guest won't see the intermediate
1569 * state, we just increase by 2 at the end.
1571 vcpu->hv_clock.version += 2;
1573 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1574 &guest_hv_clock, sizeof(guest_hv_clock))))
1577 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1578 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1580 if (vcpu->pvclock_set_guest_stopped_request) {
1581 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1582 vcpu->pvclock_set_guest_stopped_request = false;
1585 /* If the host uses TSC clocksource, then it is stable */
1586 if (use_master_clock)
1587 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1589 vcpu->hv_clock.flags = pvclock_flags;
1591 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1593 sizeof(vcpu->hv_clock));
1598 * kvmclock updates which are isolated to a given vcpu, such as
1599 * vcpu->cpu migration, should not allow system_timestamp from
1600 * the rest of the vcpus to remain static. Otherwise ntp frequency
1601 * correction applies to one vcpu's system_timestamp but not
1604 * So in those cases, request a kvmclock update for all vcpus.
1605 * The worst case for a remote vcpu to update its kvmclock
1606 * is then bounded by maximum nohz sleep latency.
1609 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1612 struct kvm *kvm = v->kvm;
1613 struct kvm_vcpu *vcpu;
1615 kvm_for_each_vcpu(i, vcpu, kvm) {
1616 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
1617 kvm_vcpu_kick(vcpu);
1621 static bool msr_mtrr_valid(unsigned msr)
1624 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1625 case MSR_MTRRfix64K_00000:
1626 case MSR_MTRRfix16K_80000:
1627 case MSR_MTRRfix16K_A0000:
1628 case MSR_MTRRfix4K_C0000:
1629 case MSR_MTRRfix4K_C8000:
1630 case MSR_MTRRfix4K_D0000:
1631 case MSR_MTRRfix4K_D8000:
1632 case MSR_MTRRfix4K_E0000:
1633 case MSR_MTRRfix4K_E8000:
1634 case MSR_MTRRfix4K_F0000:
1635 case MSR_MTRRfix4K_F8000:
1636 case MSR_MTRRdefType:
1637 case MSR_IA32_CR_PAT:
1645 static bool valid_pat_type(unsigned t)
1647 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1650 static bool valid_mtrr_type(unsigned t)
1652 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1655 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1659 if (!msr_mtrr_valid(msr))
1662 if (msr == MSR_IA32_CR_PAT) {
1663 for (i = 0; i < 8; i++)
1664 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1667 } else if (msr == MSR_MTRRdefType) {
1670 return valid_mtrr_type(data & 0xff);
1671 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1672 for (i = 0; i < 8 ; i++)
1673 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1678 /* variable MTRRs */
1679 return valid_mtrr_type(data & 0xff);
1682 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1684 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1686 if (!mtrr_valid(vcpu, msr, data))
1689 if (msr == MSR_MTRRdefType) {
1690 vcpu->arch.mtrr_state.def_type = data;
1691 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1692 } else if (msr == MSR_MTRRfix64K_00000)
1694 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1695 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1696 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1697 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1698 else if (msr == MSR_IA32_CR_PAT)
1699 vcpu->arch.pat = data;
1700 else { /* Variable MTRRs */
1701 int idx, is_mtrr_mask;
1704 idx = (msr - 0x200) / 2;
1705 is_mtrr_mask = msr - 0x200 - 2 * idx;
1708 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1711 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1715 kvm_mmu_reset_context(vcpu);
1719 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1721 u64 mcg_cap = vcpu->arch.mcg_cap;
1722 unsigned bank_num = mcg_cap & 0xff;
1725 case MSR_IA32_MCG_STATUS:
1726 vcpu->arch.mcg_status = data;
1728 case MSR_IA32_MCG_CTL:
1729 if (!(mcg_cap & MCG_CTL_P))
1731 if (data != 0 && data != ~(u64)0)
1733 vcpu->arch.mcg_ctl = data;
1736 if (msr >= MSR_IA32_MC0_CTL &&
1737 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1738 u32 offset = msr - MSR_IA32_MC0_CTL;
1739 /* only 0 or all 1s can be written to IA32_MCi_CTL
1740 * some Linux kernels though clear bit 10 in bank 4 to
1741 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1742 * this to avoid an uncatched #GP in the guest
1744 if ((offset & 0x3) == 0 &&
1745 data != 0 && (data | (1 << 10)) != ~(u64)0)
1747 vcpu->arch.mce_banks[offset] = data;
1755 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1757 struct kvm *kvm = vcpu->kvm;
1758 int lm = is_long_mode(vcpu);
1759 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1760 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1761 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1762 : kvm->arch.xen_hvm_config.blob_size_32;
1763 u32 page_num = data & ~PAGE_MASK;
1764 u64 page_addr = data & PAGE_MASK;
1769 if (page_num >= blob_size)
1772 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1777 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1786 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1788 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1791 static bool kvm_hv_msr_partition_wide(u32 msr)
1795 case HV_X64_MSR_GUEST_OS_ID:
1796 case HV_X64_MSR_HYPERCALL:
1804 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1806 struct kvm *kvm = vcpu->kvm;
1809 case HV_X64_MSR_GUEST_OS_ID:
1810 kvm->arch.hv_guest_os_id = data;
1811 /* setting guest os id to zero disables hypercall page */
1812 if (!kvm->arch.hv_guest_os_id)
1813 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1815 case HV_X64_MSR_HYPERCALL: {
1820 /* if guest os id is not set hypercall should remain disabled */
1821 if (!kvm->arch.hv_guest_os_id)
1823 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1824 kvm->arch.hv_hypercall = data;
1827 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1828 addr = gfn_to_hva(kvm, gfn);
1829 if (kvm_is_error_hva(addr))
1831 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1832 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1833 if (__copy_to_user((void __user *)addr, instructions, 4))
1835 kvm->arch.hv_hypercall = data;
1839 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1840 "data 0x%llx\n", msr, data);
1846 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1849 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1852 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1853 vcpu->arch.hv_vapic = data;
1856 addr = gfn_to_hva(vcpu->kvm, data >>
1857 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1858 if (kvm_is_error_hva(addr))
1860 if (__clear_user((void __user *)addr, PAGE_SIZE))
1862 vcpu->arch.hv_vapic = data;
1865 case HV_X64_MSR_EOI:
1866 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1867 case HV_X64_MSR_ICR:
1868 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1869 case HV_X64_MSR_TPR:
1870 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1872 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1873 "data 0x%llx\n", msr, data);
1880 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1882 gpa_t gpa = data & ~0x3f;
1884 /* Bits 2:5 are reserved, Should be zero */
1888 vcpu->arch.apf.msr_val = data;
1890 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1891 kvm_clear_async_pf_completion_queue(vcpu);
1892 kvm_async_pf_hash_reset(vcpu);
1896 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1900 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1901 kvm_async_pf_wakeup_all(vcpu);
1905 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1907 vcpu->arch.pv_time_enabled = false;
1910 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1914 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1917 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1918 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1919 vcpu->arch.st.accum_steal = delta;
1922 static void record_steal_time(struct kvm_vcpu *vcpu)
1924 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1927 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1928 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1931 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1932 vcpu->arch.st.steal.version += 2;
1933 vcpu->arch.st.accum_steal = 0;
1935 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1936 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1939 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1942 u32 msr = msr_info->index;
1943 u64 data = msr_info->data;
1946 case MSR_AMD64_NB_CFG:
1947 case MSR_IA32_UCODE_REV:
1948 case MSR_IA32_UCODE_WRITE:
1949 case MSR_VM_HSAVE_PA:
1950 case MSR_AMD64_PATCH_LOADER:
1951 case MSR_AMD64_BU_CFG2:
1955 return set_efer(vcpu, data);
1957 data &= ~(u64)0x40; /* ignore flush filter disable */
1958 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1959 data &= ~(u64)0x8; /* ignore TLB cache disable */
1961 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1966 case MSR_FAM10H_MMIO_CONF_BASE:
1968 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1973 case MSR_IA32_DEBUGCTLMSR:
1975 /* We support the non-activated case already */
1977 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1978 /* Values other than LBR and BTF are vendor-specific,
1979 thus reserved and should throw a #GP */
1982 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1985 case 0x200 ... 0x2ff:
1986 return set_msr_mtrr(vcpu, msr, data);
1987 case MSR_IA32_APICBASE:
1988 kvm_set_apic_base(vcpu, data);
1990 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1991 return kvm_x2apic_msr_write(vcpu, msr, data);
1992 case MSR_IA32_TSCDEADLINE:
1993 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1995 case MSR_IA32_TSC_ADJUST:
1996 if (guest_cpuid_has_tsc_adjust(vcpu)) {
1997 if (!msr_info->host_initiated) {
1998 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
1999 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2001 vcpu->arch.ia32_tsc_adjust_msr = data;
2004 case MSR_IA32_MISC_ENABLE:
2005 vcpu->arch.ia32_misc_enable_msr = data;
2007 case MSR_KVM_WALL_CLOCK_NEW:
2008 case MSR_KVM_WALL_CLOCK:
2009 vcpu->kvm->arch.wall_clock = data;
2010 kvm_write_wall_clock(vcpu->kvm, data);
2012 case MSR_KVM_SYSTEM_TIME_NEW:
2013 case MSR_KVM_SYSTEM_TIME: {
2015 kvmclock_reset(vcpu);
2017 vcpu->arch.time = data;
2018 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2020 /* we verify if the enable bit is set... */
2024 gpa_offset = data & ~(PAGE_MASK | 1);
2026 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2027 &vcpu->arch.pv_time, data & ~1ULL,
2028 sizeof(struct pvclock_vcpu_time_info)))
2029 vcpu->arch.pv_time_enabled = false;
2031 vcpu->arch.pv_time_enabled = true;
2035 case MSR_KVM_ASYNC_PF_EN:
2036 if (kvm_pv_enable_async_pf(vcpu, data))
2039 case MSR_KVM_STEAL_TIME:
2041 if (unlikely(!sched_info_on()))
2044 if (data & KVM_STEAL_RESERVED_MASK)
2047 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2048 data & KVM_STEAL_VALID_BITS,
2049 sizeof(struct kvm_steal_time)))
2052 vcpu->arch.st.msr_val = data;
2054 if (!(data & KVM_MSR_ENABLED))
2057 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2060 accumulate_steal_time(vcpu);
2063 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2066 case MSR_KVM_PV_EOI_EN:
2067 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2071 case MSR_IA32_MCG_CTL:
2072 case MSR_IA32_MCG_STATUS:
2073 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2074 return set_msr_mce(vcpu, msr, data);
2076 /* Performance counters are not protected by a CPUID bit,
2077 * so we should check all of them in the generic path for the sake of
2078 * cross vendor migration.
2079 * Writing a zero into the event select MSRs disables them,
2080 * which we perfectly emulate ;-). Any other value should be at least
2081 * reported, some guests depend on them.
2083 case MSR_K7_EVNTSEL0:
2084 case MSR_K7_EVNTSEL1:
2085 case MSR_K7_EVNTSEL2:
2086 case MSR_K7_EVNTSEL3:
2088 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2089 "0x%x data 0x%llx\n", msr, data);
2091 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2092 * so we ignore writes to make it happy.
2094 case MSR_K7_PERFCTR0:
2095 case MSR_K7_PERFCTR1:
2096 case MSR_K7_PERFCTR2:
2097 case MSR_K7_PERFCTR3:
2098 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2099 "0x%x data 0x%llx\n", msr, data);
2101 case MSR_P6_PERFCTR0:
2102 case MSR_P6_PERFCTR1:
2104 case MSR_P6_EVNTSEL0:
2105 case MSR_P6_EVNTSEL1:
2106 if (kvm_pmu_msr(vcpu, msr))
2107 return kvm_pmu_set_msr(vcpu, msr_info);
2109 if (pr || data != 0)
2110 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2111 "0x%x data 0x%llx\n", msr, data);
2113 case MSR_K7_CLK_CTL:
2115 * Ignore all writes to this no longer documented MSR.
2116 * Writes are only relevant for old K7 processors,
2117 * all pre-dating SVM, but a recommended workaround from
2118 * AMD for these chips. It is possible to specify the
2119 * affected processor models on the command line, hence
2120 * the need to ignore the workaround.
2123 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2124 if (kvm_hv_msr_partition_wide(msr)) {
2126 mutex_lock(&vcpu->kvm->lock);
2127 r = set_msr_hyperv_pw(vcpu, msr, data);
2128 mutex_unlock(&vcpu->kvm->lock);
2131 return set_msr_hyperv(vcpu, msr, data);
2133 case MSR_IA32_BBL_CR_CTL3:
2134 /* Drop writes to this legacy MSR -- see rdmsr
2135 * counterpart for further detail.
2137 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2139 case MSR_AMD64_OSVW_ID_LENGTH:
2140 if (!guest_cpuid_has_osvw(vcpu))
2142 vcpu->arch.osvw.length = data;
2144 case MSR_AMD64_OSVW_STATUS:
2145 if (!guest_cpuid_has_osvw(vcpu))
2147 vcpu->arch.osvw.status = data;
2150 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2151 return xen_hvm_config(vcpu, data);
2152 if (kvm_pmu_msr(vcpu, msr))
2153 return kvm_pmu_set_msr(vcpu, msr_info);
2155 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2159 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2166 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2170 * Reads an msr value (of 'msr_index') into 'pdata'.
2171 * Returns 0 on success, non-0 otherwise.
2172 * Assumes vcpu_load() was already called.
2174 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2176 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2179 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2181 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2183 if (!msr_mtrr_valid(msr))
2186 if (msr == MSR_MTRRdefType)
2187 *pdata = vcpu->arch.mtrr_state.def_type +
2188 (vcpu->arch.mtrr_state.enabled << 10);
2189 else if (msr == MSR_MTRRfix64K_00000)
2191 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2192 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2193 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2194 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2195 else if (msr == MSR_IA32_CR_PAT)
2196 *pdata = vcpu->arch.pat;
2197 else { /* Variable MTRRs */
2198 int idx, is_mtrr_mask;
2201 idx = (msr - 0x200) / 2;
2202 is_mtrr_mask = msr - 0x200 - 2 * idx;
2205 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2208 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2215 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2218 u64 mcg_cap = vcpu->arch.mcg_cap;
2219 unsigned bank_num = mcg_cap & 0xff;
2222 case MSR_IA32_P5_MC_ADDR:
2223 case MSR_IA32_P5_MC_TYPE:
2226 case MSR_IA32_MCG_CAP:
2227 data = vcpu->arch.mcg_cap;
2229 case MSR_IA32_MCG_CTL:
2230 if (!(mcg_cap & MCG_CTL_P))
2232 data = vcpu->arch.mcg_ctl;
2234 case MSR_IA32_MCG_STATUS:
2235 data = vcpu->arch.mcg_status;
2238 if (msr >= MSR_IA32_MC0_CTL &&
2239 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2240 u32 offset = msr - MSR_IA32_MC0_CTL;
2241 data = vcpu->arch.mce_banks[offset];
2250 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2253 struct kvm *kvm = vcpu->kvm;
2256 case HV_X64_MSR_GUEST_OS_ID:
2257 data = kvm->arch.hv_guest_os_id;
2259 case HV_X64_MSR_HYPERCALL:
2260 data = kvm->arch.hv_hypercall;
2263 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2271 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2276 case HV_X64_MSR_VP_INDEX: {
2279 kvm_for_each_vcpu(r, v, vcpu->kvm)
2284 case HV_X64_MSR_EOI:
2285 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2286 case HV_X64_MSR_ICR:
2287 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2288 case HV_X64_MSR_TPR:
2289 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2290 case HV_X64_MSR_APIC_ASSIST_PAGE:
2291 data = vcpu->arch.hv_vapic;
2294 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2301 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2306 case MSR_IA32_PLATFORM_ID:
2307 case MSR_IA32_EBL_CR_POWERON:
2308 case MSR_IA32_DEBUGCTLMSR:
2309 case MSR_IA32_LASTBRANCHFROMIP:
2310 case MSR_IA32_LASTBRANCHTOIP:
2311 case MSR_IA32_LASTINTFROMIP:
2312 case MSR_IA32_LASTINTTOIP:
2315 case MSR_VM_HSAVE_PA:
2316 case MSR_K7_EVNTSEL0:
2317 case MSR_K7_PERFCTR0:
2318 case MSR_K8_INT_PENDING_MSG:
2319 case MSR_AMD64_NB_CFG:
2320 case MSR_FAM10H_MMIO_CONF_BASE:
2321 case MSR_AMD64_BU_CFG2:
2324 case MSR_P6_PERFCTR0:
2325 case MSR_P6_PERFCTR1:
2326 case MSR_P6_EVNTSEL0:
2327 case MSR_P6_EVNTSEL1:
2328 if (kvm_pmu_msr(vcpu, msr))
2329 return kvm_pmu_get_msr(vcpu, msr, pdata);
2332 case MSR_IA32_UCODE_REV:
2333 data = 0x100000000ULL;
2336 data = 0x500 | KVM_NR_VAR_MTRR;
2338 case 0x200 ... 0x2ff:
2339 return get_msr_mtrr(vcpu, msr, pdata);
2340 case 0xcd: /* fsb frequency */
2344 * MSR_EBC_FREQUENCY_ID
2345 * Conservative value valid for even the basic CPU models.
2346 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2347 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2348 * and 266MHz for model 3, or 4. Set Core Clock
2349 * Frequency to System Bus Frequency Ratio to 1 (bits
2350 * 31:24) even though these are only valid for CPU
2351 * models > 2, however guests may end up dividing or
2352 * multiplying by zero otherwise.
2354 case MSR_EBC_FREQUENCY_ID:
2357 case MSR_IA32_APICBASE:
2358 data = kvm_get_apic_base(vcpu);
2360 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2361 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2363 case MSR_IA32_TSCDEADLINE:
2364 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2366 case MSR_IA32_TSC_ADJUST:
2367 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2369 case MSR_IA32_MISC_ENABLE:
2370 data = vcpu->arch.ia32_misc_enable_msr;
2372 case MSR_IA32_PERF_STATUS:
2373 /* TSC increment by tick */
2375 /* CPU multiplier */
2376 data |= (((uint64_t)4ULL) << 40);
2379 data = vcpu->arch.efer;
2381 case MSR_KVM_WALL_CLOCK:
2382 case MSR_KVM_WALL_CLOCK_NEW:
2383 data = vcpu->kvm->arch.wall_clock;
2385 case MSR_KVM_SYSTEM_TIME:
2386 case MSR_KVM_SYSTEM_TIME_NEW:
2387 data = vcpu->arch.time;
2389 case MSR_KVM_ASYNC_PF_EN:
2390 data = vcpu->arch.apf.msr_val;
2392 case MSR_KVM_STEAL_TIME:
2393 data = vcpu->arch.st.msr_val;
2395 case MSR_KVM_PV_EOI_EN:
2396 data = vcpu->arch.pv_eoi.msr_val;
2398 case MSR_IA32_P5_MC_ADDR:
2399 case MSR_IA32_P5_MC_TYPE:
2400 case MSR_IA32_MCG_CAP:
2401 case MSR_IA32_MCG_CTL:
2402 case MSR_IA32_MCG_STATUS:
2403 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2404 return get_msr_mce(vcpu, msr, pdata);
2405 case MSR_K7_CLK_CTL:
2407 * Provide expected ramp-up count for K7. All other
2408 * are set to zero, indicating minimum divisors for
2411 * This prevents guest kernels on AMD host with CPU
2412 * type 6, model 8 and higher from exploding due to
2413 * the rdmsr failing.
2417 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2418 if (kvm_hv_msr_partition_wide(msr)) {
2420 mutex_lock(&vcpu->kvm->lock);
2421 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2422 mutex_unlock(&vcpu->kvm->lock);
2425 return get_msr_hyperv(vcpu, msr, pdata);
2427 case MSR_IA32_BBL_CR_CTL3:
2428 /* This legacy MSR exists but isn't fully documented in current
2429 * silicon. It is however accessed by winxp in very narrow
2430 * scenarios where it sets bit #19, itself documented as
2431 * a "reserved" bit. Best effort attempt to source coherent
2432 * read data here should the balance of the register be
2433 * interpreted by the guest:
2435 * L2 cache control register 3: 64GB range, 256KB size,
2436 * enabled, latency 0x1, configured
2440 case MSR_AMD64_OSVW_ID_LENGTH:
2441 if (!guest_cpuid_has_osvw(vcpu))
2443 data = vcpu->arch.osvw.length;
2445 case MSR_AMD64_OSVW_STATUS:
2446 if (!guest_cpuid_has_osvw(vcpu))
2448 data = vcpu->arch.osvw.status;
2451 if (kvm_pmu_msr(vcpu, msr))
2452 return kvm_pmu_get_msr(vcpu, msr, pdata);
2454 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2457 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2465 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2468 * Read or write a bunch of msrs. All parameters are kernel addresses.
2470 * @return number of msrs set successfully.
2472 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2473 struct kvm_msr_entry *entries,
2474 int (*do_msr)(struct kvm_vcpu *vcpu,
2475 unsigned index, u64 *data))
2479 idx = srcu_read_lock(&vcpu->kvm->srcu);
2480 for (i = 0; i < msrs->nmsrs; ++i)
2481 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2483 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2489 * Read or write a bunch of msrs. Parameters are user addresses.
2491 * @return number of msrs set successfully.
2493 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2494 int (*do_msr)(struct kvm_vcpu *vcpu,
2495 unsigned index, u64 *data),
2498 struct kvm_msrs msrs;
2499 struct kvm_msr_entry *entries;
2504 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2508 if (msrs.nmsrs >= MAX_IO_MSRS)
2511 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2512 entries = memdup_user(user_msrs->entries, size);
2513 if (IS_ERR(entries)) {
2514 r = PTR_ERR(entries);
2518 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2523 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2534 int kvm_dev_ioctl_check_extension(long ext)
2539 case KVM_CAP_IRQCHIP:
2541 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2542 case KVM_CAP_SET_TSS_ADDR:
2543 case KVM_CAP_EXT_CPUID:
2544 case KVM_CAP_CLOCKSOURCE:
2546 case KVM_CAP_NOP_IO_DELAY:
2547 case KVM_CAP_MP_STATE:
2548 case KVM_CAP_SYNC_MMU:
2549 case KVM_CAP_USER_NMI:
2550 case KVM_CAP_REINJECT_CONTROL:
2551 case KVM_CAP_IRQ_INJECT_STATUS:
2553 case KVM_CAP_IOEVENTFD:
2555 case KVM_CAP_PIT_STATE2:
2556 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2557 case KVM_CAP_XEN_HVM:
2558 case KVM_CAP_ADJUST_CLOCK:
2559 case KVM_CAP_VCPU_EVENTS:
2560 case KVM_CAP_HYPERV:
2561 case KVM_CAP_HYPERV_VAPIC:
2562 case KVM_CAP_HYPERV_SPIN:
2563 case KVM_CAP_PCI_SEGMENT:
2564 case KVM_CAP_DEBUGREGS:
2565 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2567 case KVM_CAP_ASYNC_PF:
2568 case KVM_CAP_GET_TSC_KHZ:
2569 case KVM_CAP_KVMCLOCK_CTRL:
2570 case KVM_CAP_READONLY_MEM:
2571 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2572 case KVM_CAP_ASSIGN_DEV_IRQ:
2573 case KVM_CAP_PCI_2_3:
2577 case KVM_CAP_COALESCED_MMIO:
2578 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2581 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2583 case KVM_CAP_NR_VCPUS:
2584 r = KVM_SOFT_MAX_VCPUS;
2586 case KVM_CAP_MAX_VCPUS:
2589 case KVM_CAP_NR_MEMSLOTS:
2590 r = KVM_USER_MEM_SLOTS;
2592 case KVM_CAP_PV_MMU: /* obsolete */
2595 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2597 r = iommu_present(&pci_bus_type);
2601 r = KVM_MAX_MCE_BANKS;
2606 case KVM_CAP_TSC_CONTROL:
2607 r = kvm_has_tsc_control;
2609 case KVM_CAP_TSC_DEADLINE_TIMER:
2610 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2620 long kvm_arch_dev_ioctl(struct file *filp,
2621 unsigned int ioctl, unsigned long arg)
2623 void __user *argp = (void __user *)arg;
2627 case KVM_GET_MSR_INDEX_LIST: {
2628 struct kvm_msr_list __user *user_msr_list = argp;
2629 struct kvm_msr_list msr_list;
2633 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2636 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2637 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2640 if (n < msr_list.nmsrs)
2643 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2644 num_msrs_to_save * sizeof(u32)))
2646 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2648 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2653 case KVM_GET_SUPPORTED_CPUID: {
2654 struct kvm_cpuid2 __user *cpuid_arg = argp;
2655 struct kvm_cpuid2 cpuid;
2658 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2660 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2661 cpuid_arg->entries);
2666 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2671 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2674 mce_cap = KVM_MCE_CAP_SUPPORTED;
2676 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2688 static void wbinvd_ipi(void *garbage)
2693 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2695 return vcpu->kvm->arch.iommu_domain &&
2696 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2699 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2701 /* Address WBINVD may be executed by guest */
2702 if (need_emulate_wbinvd(vcpu)) {
2703 if (kvm_x86_ops->has_wbinvd_exit())
2704 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2705 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2706 smp_call_function_single(vcpu->cpu,
2707 wbinvd_ipi, NULL, 1);
2710 kvm_x86_ops->vcpu_load(vcpu, cpu);
2712 /* Apply any externally detected TSC adjustments (due to suspend) */
2713 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2714 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2715 vcpu->arch.tsc_offset_adjustment = 0;
2716 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2719 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2720 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2721 native_read_tsc() - vcpu->arch.last_host_tsc;
2723 mark_tsc_unstable("KVM discovered backwards TSC");
2724 if (check_tsc_unstable()) {
2725 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2726 vcpu->arch.last_guest_tsc);
2727 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2728 vcpu->arch.tsc_catchup = 1;
2731 * On a host with synchronized TSC, there is no need to update
2732 * kvmclock on vcpu->cpu migration
2734 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2735 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2736 if (vcpu->cpu != cpu)
2737 kvm_migrate_timers(vcpu);
2741 accumulate_steal_time(vcpu);
2742 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2745 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2747 kvm_x86_ops->vcpu_put(vcpu);
2748 kvm_put_guest_fpu(vcpu);
2749 vcpu->arch.last_host_tsc = native_read_tsc();
2752 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2753 struct kvm_lapic_state *s)
2755 kvm_x86_ops->sync_pir_to_irr(vcpu);
2756 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2761 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2762 struct kvm_lapic_state *s)
2764 kvm_apic_post_state_restore(vcpu, s);
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 >= KVM_NR_INTERRUPTS)
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 = 0; /* never valid when reporting to user space */
2898 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2899 | KVM_VCPUEVENT_VALID_SHADOW);
2900 memset(&events->reserved, 0, sizeof(events->reserved));
2903 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2904 struct kvm_vcpu_events *events)
2906 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2907 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2908 | KVM_VCPUEVENT_VALID_SHADOW))
2912 vcpu->arch.exception.pending = events->exception.injected;
2913 vcpu->arch.exception.nr = events->exception.nr;
2914 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2915 vcpu->arch.exception.error_code = events->exception.error_code;
2917 vcpu->arch.interrupt.pending = events->interrupt.injected;
2918 vcpu->arch.interrupt.nr = events->interrupt.nr;
2919 vcpu->arch.interrupt.soft = events->interrupt.soft;
2920 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2921 kvm_x86_ops->set_interrupt_shadow(vcpu,
2922 events->interrupt.shadow);
2924 vcpu->arch.nmi_injected = events->nmi.injected;
2925 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2926 vcpu->arch.nmi_pending = events->nmi.pending;
2927 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2929 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2930 kvm_vcpu_has_lapic(vcpu))
2931 vcpu->arch.apic->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);
3033 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3034 * stopped by the hypervisor. This function will be called from the host only.
3035 * EINVAL is returned when the host attempts to set the flag for a guest that
3036 * does not support pv clocks.
3038 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3040 if (!vcpu->arch.pv_time_enabled)
3042 vcpu->arch.pvclock_set_guest_stopped_request = true;
3043 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3047 long kvm_arch_vcpu_ioctl(struct file *filp,
3048 unsigned int ioctl, unsigned long arg)
3050 struct kvm_vcpu *vcpu = filp->private_data;
3051 void __user *argp = (void __user *)arg;
3054 struct kvm_lapic_state *lapic;
3055 struct kvm_xsave *xsave;
3056 struct kvm_xcrs *xcrs;
3062 case KVM_GET_LAPIC: {
3064 if (!vcpu->arch.apic)
3066 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3071 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3075 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3080 case KVM_SET_LAPIC: {
3082 if (!vcpu->arch.apic)
3084 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3085 if (IS_ERR(u.lapic))
3086 return PTR_ERR(u.lapic);
3088 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3091 case KVM_INTERRUPT: {
3092 struct kvm_interrupt irq;
3095 if (copy_from_user(&irq, argp, sizeof irq))
3097 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3101 r = kvm_vcpu_ioctl_nmi(vcpu);
3104 case KVM_SET_CPUID: {
3105 struct kvm_cpuid __user *cpuid_arg = argp;
3106 struct kvm_cpuid cpuid;
3109 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3111 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3114 case KVM_SET_CPUID2: {
3115 struct kvm_cpuid2 __user *cpuid_arg = argp;
3116 struct kvm_cpuid2 cpuid;
3119 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3121 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3122 cpuid_arg->entries);
3125 case KVM_GET_CPUID2: {
3126 struct kvm_cpuid2 __user *cpuid_arg = argp;
3127 struct kvm_cpuid2 cpuid;
3130 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3132 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3133 cpuid_arg->entries);
3137 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3143 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3146 r = msr_io(vcpu, argp, do_set_msr, 0);
3148 case KVM_TPR_ACCESS_REPORTING: {
3149 struct kvm_tpr_access_ctl tac;
3152 if (copy_from_user(&tac, argp, sizeof tac))
3154 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3158 if (copy_to_user(argp, &tac, sizeof tac))
3163 case KVM_SET_VAPIC_ADDR: {
3164 struct kvm_vapic_addr va;
3167 if (!irqchip_in_kernel(vcpu->kvm))
3170 if (copy_from_user(&va, argp, sizeof va))
3173 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 = memdup_user(argp, sizeof(*u.xsave));
3254 if (IS_ERR(u.xsave))
3255 return PTR_ERR(u.xsave);
3257 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3260 case KVM_GET_XCRS: {
3261 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3266 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3269 if (copy_to_user(argp, u.xcrs,
3270 sizeof(struct kvm_xcrs)))
3275 case KVM_SET_XCRS: {
3276 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3278 return PTR_ERR(u.xcrs);
3280 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3283 case KVM_SET_TSC_KHZ: {
3287 user_tsc_khz = (u32)arg;
3289 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3292 if (user_tsc_khz == 0)
3293 user_tsc_khz = tsc_khz;
3295 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3300 case KVM_GET_TSC_KHZ: {
3301 r = vcpu->arch.virtual_tsc_khz;
3304 case KVM_KVMCLOCK_CTRL: {
3305 r = kvm_set_guest_paused(vcpu);
3316 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3318 return VM_FAULT_SIGBUS;
3321 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3325 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3327 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3331 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3334 kvm->arch.ept_identity_map_addr = ident_addr;
3338 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3339 u32 kvm_nr_mmu_pages)
3341 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3344 mutex_lock(&kvm->slots_lock);
3346 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3347 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3349 mutex_unlock(&kvm->slots_lock);
3353 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3355 return kvm->arch.n_max_mmu_pages;
3358 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3363 switch (chip->chip_id) {
3364 case KVM_IRQCHIP_PIC_MASTER:
3365 memcpy(&chip->chip.pic,
3366 &pic_irqchip(kvm)->pics[0],
3367 sizeof(struct kvm_pic_state));
3369 case KVM_IRQCHIP_PIC_SLAVE:
3370 memcpy(&chip->chip.pic,
3371 &pic_irqchip(kvm)->pics[1],
3372 sizeof(struct kvm_pic_state));
3374 case KVM_IRQCHIP_IOAPIC:
3375 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3384 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3389 switch (chip->chip_id) {
3390 case KVM_IRQCHIP_PIC_MASTER:
3391 spin_lock(&pic_irqchip(kvm)->lock);
3392 memcpy(&pic_irqchip(kvm)->pics[0],
3394 sizeof(struct kvm_pic_state));
3395 spin_unlock(&pic_irqchip(kvm)->lock);
3397 case KVM_IRQCHIP_PIC_SLAVE:
3398 spin_lock(&pic_irqchip(kvm)->lock);
3399 memcpy(&pic_irqchip(kvm)->pics[1],
3401 sizeof(struct kvm_pic_state));
3402 spin_unlock(&pic_irqchip(kvm)->lock);
3404 case KVM_IRQCHIP_IOAPIC:
3405 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3411 kvm_pic_update_irq(pic_irqchip(kvm));
3415 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3419 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3420 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3421 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3425 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3429 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3430 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3431 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3432 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3436 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3440 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3441 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3442 sizeof(ps->channels));
3443 ps->flags = kvm->arch.vpit->pit_state.flags;
3444 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3445 memset(&ps->reserved, 0, sizeof(ps->reserved));
3449 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3451 int r = 0, start = 0;
3452 u32 prev_legacy, cur_legacy;
3453 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3454 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3455 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3456 if (!prev_legacy && cur_legacy)
3458 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3459 sizeof(kvm->arch.vpit->pit_state.channels));
3460 kvm->arch.vpit->pit_state.flags = ps->flags;
3461 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3462 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3466 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3467 struct kvm_reinject_control *control)
3469 if (!kvm->arch.vpit)
3471 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3472 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3473 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3478 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3479 * @kvm: kvm instance
3480 * @log: slot id and address to which we copy the log
3482 * We need to keep it in mind that VCPU threads can write to the bitmap
3483 * concurrently. So, to avoid losing data, we keep the following order for
3486 * 1. Take a snapshot of the bit and clear it if needed.
3487 * 2. Write protect the corresponding page.
3488 * 3. Flush TLB's if needed.
3489 * 4. Copy the snapshot to the userspace.
3491 * Between 2 and 3, the guest may write to the page using the remaining TLB
3492 * entry. This is not a problem because the page will be reported dirty at
3493 * step 4 using the snapshot taken before and step 3 ensures that successive
3494 * writes will be logged for the next call.
3496 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3499 struct kvm_memory_slot *memslot;
3501 unsigned long *dirty_bitmap;
3502 unsigned long *dirty_bitmap_buffer;
3503 bool is_dirty = false;
3505 mutex_lock(&kvm->slots_lock);
3508 if (log->slot >= KVM_USER_MEM_SLOTS)
3511 memslot = id_to_memslot(kvm->memslots, log->slot);
3513 dirty_bitmap = memslot->dirty_bitmap;
3518 n = kvm_dirty_bitmap_bytes(memslot);
3520 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3521 memset(dirty_bitmap_buffer, 0, n);
3523 spin_lock(&kvm->mmu_lock);
3525 for (i = 0; i < n / sizeof(long); i++) {
3529 if (!dirty_bitmap[i])
3534 mask = xchg(&dirty_bitmap[i], 0);
3535 dirty_bitmap_buffer[i] = mask;
3537 offset = i * BITS_PER_LONG;
3538 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3541 kvm_flush_remote_tlbs(kvm);
3543 spin_unlock(&kvm->mmu_lock);
3546 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3551 mutex_unlock(&kvm->slots_lock);
3555 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3558 if (!irqchip_in_kernel(kvm))
3561 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3562 irq_event->irq, irq_event->level,
3567 long kvm_arch_vm_ioctl(struct file *filp,
3568 unsigned int ioctl, unsigned long arg)
3570 struct kvm *kvm = filp->private_data;
3571 void __user *argp = (void __user *)arg;
3574 * This union makes it completely explicit to gcc-3.x
3575 * that these two variables' stack usage should be
3576 * combined, not added together.
3579 struct kvm_pit_state ps;
3580 struct kvm_pit_state2 ps2;
3581 struct kvm_pit_config pit_config;
3585 case KVM_SET_TSS_ADDR:
3586 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3588 case KVM_SET_IDENTITY_MAP_ADDR: {
3592 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3594 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3597 case KVM_SET_NR_MMU_PAGES:
3598 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3600 case KVM_GET_NR_MMU_PAGES:
3601 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3603 case KVM_CREATE_IRQCHIP: {
3604 struct kvm_pic *vpic;
3606 mutex_lock(&kvm->lock);
3609 goto create_irqchip_unlock;
3611 if (atomic_read(&kvm->online_vcpus))
3612 goto create_irqchip_unlock;
3614 vpic = kvm_create_pic(kvm);
3616 r = kvm_ioapic_init(kvm);
3618 mutex_lock(&kvm->slots_lock);
3619 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3621 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3623 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3625 mutex_unlock(&kvm->slots_lock);
3627 goto create_irqchip_unlock;
3630 goto create_irqchip_unlock;
3632 kvm->arch.vpic = vpic;
3634 r = kvm_setup_default_irq_routing(kvm);
3636 mutex_lock(&kvm->slots_lock);
3637 mutex_lock(&kvm->irq_lock);
3638 kvm_ioapic_destroy(kvm);
3639 kvm_destroy_pic(kvm);
3640 mutex_unlock(&kvm->irq_lock);
3641 mutex_unlock(&kvm->slots_lock);
3643 create_irqchip_unlock:
3644 mutex_unlock(&kvm->lock);
3647 case KVM_CREATE_PIT:
3648 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3650 case KVM_CREATE_PIT2:
3652 if (copy_from_user(&u.pit_config, argp,
3653 sizeof(struct kvm_pit_config)))
3656 mutex_lock(&kvm->slots_lock);
3659 goto create_pit_unlock;
3661 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3665 mutex_unlock(&kvm->slots_lock);
3667 case KVM_GET_IRQCHIP: {
3668 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3669 struct kvm_irqchip *chip;
3671 chip = memdup_user(argp, sizeof(*chip));
3678 if (!irqchip_in_kernel(kvm))
3679 goto get_irqchip_out;
3680 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3682 goto get_irqchip_out;
3684 if (copy_to_user(argp, chip, sizeof *chip))
3685 goto get_irqchip_out;
3691 case KVM_SET_IRQCHIP: {
3692 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3693 struct kvm_irqchip *chip;
3695 chip = memdup_user(argp, sizeof(*chip));
3702 if (!irqchip_in_kernel(kvm))
3703 goto set_irqchip_out;
3704 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3706 goto set_irqchip_out;
3714 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3717 if (!kvm->arch.vpit)
3719 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3723 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3730 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3733 if (!kvm->arch.vpit)
3735 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3738 case KVM_GET_PIT2: {
3740 if (!kvm->arch.vpit)
3742 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3746 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3751 case KVM_SET_PIT2: {
3753 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3756 if (!kvm->arch.vpit)
3758 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3761 case KVM_REINJECT_CONTROL: {
3762 struct kvm_reinject_control control;
3764 if (copy_from_user(&control, argp, sizeof(control)))
3766 r = kvm_vm_ioctl_reinject(kvm, &control);
3769 case KVM_XEN_HVM_CONFIG: {
3771 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3772 sizeof(struct kvm_xen_hvm_config)))
3775 if (kvm->arch.xen_hvm_config.flags)
3780 case KVM_SET_CLOCK: {
3781 struct kvm_clock_data user_ns;
3786 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3794 local_irq_disable();
3795 now_ns = get_kernel_ns();
3796 delta = user_ns.clock - now_ns;
3798 kvm->arch.kvmclock_offset = delta;
3801 case KVM_GET_CLOCK: {
3802 struct kvm_clock_data user_ns;
3805 local_irq_disable();
3806 now_ns = get_kernel_ns();
3807 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3810 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3813 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3826 static void kvm_init_msr_list(void)
3831 /* skip the first msrs in the list. KVM-specific */
3832 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3833 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3836 msrs_to_save[j] = msrs_to_save[i];
3839 num_msrs_to_save = j;
3842 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3850 if (!(vcpu->arch.apic &&
3851 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3852 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3863 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3870 if (!(vcpu->arch.apic &&
3871 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3872 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3874 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3884 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3885 struct kvm_segment *var, int seg)
3887 kvm_x86_ops->set_segment(vcpu, var, seg);
3890 void kvm_get_segment(struct kvm_vcpu *vcpu,
3891 struct kvm_segment *var, int seg)
3893 kvm_x86_ops->get_segment(vcpu, var, seg);
3896 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3899 struct x86_exception exception;
3901 BUG_ON(!mmu_is_nested(vcpu));
3903 /* NPT walks are always user-walks */
3904 access |= PFERR_USER_MASK;
3905 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3910 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3911 struct x86_exception *exception)
3913 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3914 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3917 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3918 struct x86_exception *exception)
3920 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3921 access |= PFERR_FETCH_MASK;
3922 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3925 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3926 struct x86_exception *exception)
3928 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3929 access |= PFERR_WRITE_MASK;
3930 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3933 /* uses this to access any guest's mapped memory without checking CPL */
3934 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3935 struct x86_exception *exception)
3937 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3940 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3941 struct kvm_vcpu *vcpu, u32 access,
3942 struct x86_exception *exception)
3945 int r = X86EMUL_CONTINUE;
3948 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3950 unsigned offset = addr & (PAGE_SIZE-1);
3951 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3954 if (gpa == UNMAPPED_GVA)
3955 return X86EMUL_PROPAGATE_FAULT;
3956 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3958 r = X86EMUL_IO_NEEDED;
3970 /* used for instruction fetching */
3971 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3972 gva_t addr, void *val, unsigned int bytes,
3973 struct x86_exception *exception)
3975 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3976 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3978 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3979 access | PFERR_FETCH_MASK,
3983 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3984 gva_t addr, void *val, unsigned int bytes,
3985 struct x86_exception *exception)
3987 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3988 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3990 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3993 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3995 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3996 gva_t addr, void *val, unsigned int bytes,
3997 struct x86_exception *exception)
3999 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4000 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4003 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4004 gva_t addr, void *val,
4006 struct x86_exception *exception)
4008 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4010 int r = X86EMUL_CONTINUE;
4013 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4016 unsigned offset = addr & (PAGE_SIZE-1);
4017 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4020 if (gpa == UNMAPPED_GVA)
4021 return X86EMUL_PROPAGATE_FAULT;
4022 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4024 r = X86EMUL_IO_NEEDED;
4035 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4037 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4038 gpa_t *gpa, struct x86_exception *exception,
4041 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4042 | (write ? PFERR_WRITE_MASK : 0);
4044 if (vcpu_match_mmio_gva(vcpu, gva)
4045 && !permission_fault(vcpu->arch.walk_mmu, vcpu->arch.access, access)) {
4046 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4047 (gva & (PAGE_SIZE - 1));
4048 trace_vcpu_match_mmio(gva, *gpa, write, false);
4052 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4054 if (*gpa == UNMAPPED_GVA)
4057 /* For APIC access vmexit */
4058 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4061 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4062 trace_vcpu_match_mmio(gva, *gpa, write, true);
4069 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4070 const void *val, int bytes)
4074 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4077 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4081 struct read_write_emulator_ops {
4082 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4084 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4085 void *val, int bytes);
4086 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4087 int bytes, void *val);
4088 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4089 void *val, int bytes);
4093 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4095 if (vcpu->mmio_read_completed) {
4096 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4097 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4098 vcpu->mmio_read_completed = 0;
4105 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4106 void *val, int bytes)
4108 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4111 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4112 void *val, int bytes)
4114 return emulator_write_phys(vcpu, gpa, val, bytes);
4117 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4119 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4120 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4123 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4124 void *val, int bytes)
4126 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4127 return X86EMUL_IO_NEEDED;
4130 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4131 void *val, int bytes)
4133 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4135 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4136 return X86EMUL_CONTINUE;
4139 static const struct read_write_emulator_ops read_emultor = {
4140 .read_write_prepare = read_prepare,
4141 .read_write_emulate = read_emulate,
4142 .read_write_mmio = vcpu_mmio_read,
4143 .read_write_exit_mmio = read_exit_mmio,
4146 static const struct read_write_emulator_ops write_emultor = {
4147 .read_write_emulate = write_emulate,
4148 .read_write_mmio = write_mmio,
4149 .read_write_exit_mmio = write_exit_mmio,
4153 static int emulator_read_write_onepage(unsigned long addr, void *val,
4155 struct x86_exception *exception,
4156 struct kvm_vcpu *vcpu,
4157 const struct read_write_emulator_ops *ops)
4161 bool write = ops->write;
4162 struct kvm_mmio_fragment *frag;
4164 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4167 return X86EMUL_PROPAGATE_FAULT;
4169 /* For APIC access vmexit */
4173 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4174 return X86EMUL_CONTINUE;
4178 * Is this MMIO handled locally?
4180 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4181 if (handled == bytes)
4182 return X86EMUL_CONTINUE;
4188 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4189 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4193 return X86EMUL_CONTINUE;
4196 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4197 void *val, unsigned int bytes,
4198 struct x86_exception *exception,
4199 const struct read_write_emulator_ops *ops)
4201 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4205 if (ops->read_write_prepare &&
4206 ops->read_write_prepare(vcpu, val, bytes))
4207 return X86EMUL_CONTINUE;
4209 vcpu->mmio_nr_fragments = 0;
4211 /* Crossing a page boundary? */
4212 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4215 now = -addr & ~PAGE_MASK;
4216 rc = emulator_read_write_onepage(addr, val, now, exception,
4219 if (rc != X86EMUL_CONTINUE)
4226 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4228 if (rc != X86EMUL_CONTINUE)
4231 if (!vcpu->mmio_nr_fragments)
4234 gpa = vcpu->mmio_fragments[0].gpa;
4236 vcpu->mmio_needed = 1;
4237 vcpu->mmio_cur_fragment = 0;
4239 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4240 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4241 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4242 vcpu->run->mmio.phys_addr = gpa;
4244 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4247 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4251 struct x86_exception *exception)
4253 return emulator_read_write(ctxt, addr, val, bytes,
4254 exception, &read_emultor);
4257 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4261 struct x86_exception *exception)
4263 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4264 exception, &write_emultor);
4267 #define CMPXCHG_TYPE(t, ptr, old, new) \
4268 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4270 #ifdef CONFIG_X86_64
4271 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4273 # define CMPXCHG64(ptr, old, new) \
4274 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4277 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4282 struct x86_exception *exception)
4284 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4290 /* guests cmpxchg8b have to be emulated atomically */
4291 if (bytes > 8 || (bytes & (bytes - 1)))
4294 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4296 if (gpa == UNMAPPED_GVA ||
4297 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4300 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4303 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4304 if (is_error_page(page))
4307 kaddr = kmap_atomic(page);
4308 kaddr += offset_in_page(gpa);
4311 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4314 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4317 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4320 exchanged = CMPXCHG64(kaddr, old, new);
4325 kunmap_atomic(kaddr);
4326 kvm_release_page_dirty(page);
4329 return X86EMUL_CMPXCHG_FAILED;
4331 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4333 return X86EMUL_CONTINUE;
4336 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4338 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4341 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4343 /* TODO: String I/O for in kernel device */
4346 if (vcpu->arch.pio.in)
4347 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4348 vcpu->arch.pio.size, pd);
4350 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4351 vcpu->arch.pio.port, vcpu->arch.pio.size,
4356 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4357 unsigned short port, void *val,
4358 unsigned int count, bool in)
4360 trace_kvm_pio(!in, port, size, count);
4362 vcpu->arch.pio.port = port;
4363 vcpu->arch.pio.in = in;
4364 vcpu->arch.pio.count = count;
4365 vcpu->arch.pio.size = size;
4367 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4368 vcpu->arch.pio.count = 0;
4372 vcpu->run->exit_reason = KVM_EXIT_IO;
4373 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4374 vcpu->run->io.size = size;
4375 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4376 vcpu->run->io.count = count;
4377 vcpu->run->io.port = port;
4382 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4383 int size, unsigned short port, void *val,
4386 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4389 if (vcpu->arch.pio.count)
4392 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4395 memcpy(val, vcpu->arch.pio_data, size * count);
4396 vcpu->arch.pio.count = 0;
4403 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4404 int size, unsigned short port,
4405 const void *val, unsigned int count)
4407 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4409 memcpy(vcpu->arch.pio_data, val, size * count);
4410 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4413 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4415 return kvm_x86_ops->get_segment_base(vcpu, seg);
4418 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4420 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4423 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4425 if (!need_emulate_wbinvd(vcpu))
4426 return X86EMUL_CONTINUE;
4428 if (kvm_x86_ops->has_wbinvd_exit()) {
4429 int cpu = get_cpu();
4431 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4432 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4433 wbinvd_ipi, NULL, 1);
4435 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4438 return X86EMUL_CONTINUE;
4440 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4442 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4444 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4447 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4449 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4452 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4455 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4458 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4460 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4463 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4465 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4466 unsigned long value;
4470 value = kvm_read_cr0(vcpu);
4473 value = vcpu->arch.cr2;
4476 value = kvm_read_cr3(vcpu);
4479 value = kvm_read_cr4(vcpu);
4482 value = kvm_get_cr8(vcpu);
4485 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4492 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4494 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4499 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4502 vcpu->arch.cr2 = val;
4505 res = kvm_set_cr3(vcpu, val);
4508 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4511 res = kvm_set_cr8(vcpu, val);
4514 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4521 static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4523 kvm_set_rflags(emul_to_vcpu(ctxt), val);
4526 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4528 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4531 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4533 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4536 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4538 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4541 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4543 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4546 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4548 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4551 static unsigned long emulator_get_cached_segment_base(
4552 struct x86_emulate_ctxt *ctxt, int seg)
4554 return get_segment_base(emul_to_vcpu(ctxt), seg);
4557 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4558 struct desc_struct *desc, u32 *base3,
4561 struct kvm_segment var;
4563 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4564 *selector = var.selector;
4567 memset(desc, 0, sizeof(*desc));
4573 set_desc_limit(desc, var.limit);
4574 set_desc_base(desc, (unsigned long)var.base);
4575 #ifdef CONFIG_X86_64
4577 *base3 = var.base >> 32;
4579 desc->type = var.type;
4581 desc->dpl = var.dpl;
4582 desc->p = var.present;
4583 desc->avl = var.avl;
4591 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4592 struct desc_struct *desc, u32 base3,
4595 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4596 struct kvm_segment var;
4598 var.selector = selector;
4599 var.base = get_desc_base(desc);
4600 #ifdef CONFIG_X86_64
4601 var.base |= ((u64)base3) << 32;
4603 var.limit = get_desc_limit(desc);
4605 var.limit = (var.limit << 12) | 0xfff;
4606 var.type = desc->type;
4607 var.present = desc->p;
4608 var.dpl = desc->dpl;
4613 var.avl = desc->avl;
4614 var.present = desc->p;
4615 var.unusable = !var.present;
4618 kvm_set_segment(vcpu, &var, seg);
4622 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4623 u32 msr_index, u64 *pdata)
4625 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4628 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4629 u32 msr_index, u64 data)
4631 struct msr_data msr;
4634 msr.index = msr_index;
4635 msr.host_initiated = false;
4636 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4639 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4640 u32 pmc, u64 *pdata)
4642 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4645 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4647 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4650 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4653 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4655 * CR0.TS may reference the host fpu state, not the guest fpu state,
4656 * so it may be clear at this point.
4661 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4666 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4667 struct x86_instruction_info *info,
4668 enum x86_intercept_stage stage)
4670 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4673 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4674 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4676 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4679 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4681 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4684 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4686 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4689 static const struct x86_emulate_ops emulate_ops = {
4690 .read_gpr = emulator_read_gpr,
4691 .write_gpr = emulator_write_gpr,
4692 .read_std = kvm_read_guest_virt_system,
4693 .write_std = kvm_write_guest_virt_system,
4694 .fetch = kvm_fetch_guest_virt,
4695 .read_emulated = emulator_read_emulated,
4696 .write_emulated = emulator_write_emulated,
4697 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4698 .invlpg = emulator_invlpg,
4699 .pio_in_emulated = emulator_pio_in_emulated,
4700 .pio_out_emulated = emulator_pio_out_emulated,
4701 .get_segment = emulator_get_segment,
4702 .set_segment = emulator_set_segment,
4703 .get_cached_segment_base = emulator_get_cached_segment_base,
4704 .get_gdt = emulator_get_gdt,
4705 .get_idt = emulator_get_idt,
4706 .set_gdt = emulator_set_gdt,
4707 .set_idt = emulator_set_idt,
4708 .get_cr = emulator_get_cr,
4709 .set_cr = emulator_set_cr,
4710 .set_rflags = emulator_set_rflags,
4711 .cpl = emulator_get_cpl,
4712 .get_dr = emulator_get_dr,
4713 .set_dr = emulator_set_dr,
4714 .set_msr = emulator_set_msr,
4715 .get_msr = emulator_get_msr,
4716 .read_pmc = emulator_read_pmc,
4717 .halt = emulator_halt,
4718 .wbinvd = emulator_wbinvd,
4719 .fix_hypercall = emulator_fix_hypercall,
4720 .get_fpu = emulator_get_fpu,
4721 .put_fpu = emulator_put_fpu,
4722 .intercept = emulator_intercept,
4723 .get_cpuid = emulator_get_cpuid,
4726 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4728 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4730 * an sti; sti; sequence only disable interrupts for the first
4731 * instruction. So, if the last instruction, be it emulated or
4732 * not, left the system with the INT_STI flag enabled, it
4733 * means that the last instruction is an sti. We should not
4734 * leave the flag on in this case. The same goes for mov ss
4736 if (!(int_shadow & mask))
4737 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4740 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4742 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4743 if (ctxt->exception.vector == PF_VECTOR)
4744 kvm_propagate_fault(vcpu, &ctxt->exception);
4745 else if (ctxt->exception.error_code_valid)
4746 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4747 ctxt->exception.error_code);
4749 kvm_queue_exception(vcpu, ctxt->exception.vector);
4752 static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
4754 memset(&ctxt->twobyte, 0,
4755 (void *)&ctxt->_regs - (void *)&ctxt->twobyte);
4757 ctxt->fetch.start = 0;
4758 ctxt->fetch.end = 0;
4759 ctxt->io_read.pos = 0;
4760 ctxt->io_read.end = 0;
4761 ctxt->mem_read.pos = 0;
4762 ctxt->mem_read.end = 0;
4765 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4767 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4770 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4772 ctxt->eflags = kvm_get_rflags(vcpu);
4773 ctxt->eip = kvm_rip_read(vcpu);
4774 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4775 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4776 cs_l ? X86EMUL_MODE_PROT64 :
4777 cs_db ? X86EMUL_MODE_PROT32 :
4778 X86EMUL_MODE_PROT16;
4779 ctxt->guest_mode = is_guest_mode(vcpu);
4781 init_decode_cache(ctxt);
4782 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4785 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4787 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4790 init_emulate_ctxt(vcpu);
4794 ctxt->_eip = ctxt->eip + inc_eip;
4795 ret = emulate_int_real(ctxt, irq);
4797 if (ret != X86EMUL_CONTINUE)
4798 return EMULATE_FAIL;
4800 ctxt->eip = ctxt->_eip;
4801 kvm_rip_write(vcpu, ctxt->eip);
4802 kvm_set_rflags(vcpu, ctxt->eflags);
4804 if (irq == NMI_VECTOR)
4805 vcpu->arch.nmi_pending = 0;
4807 vcpu->arch.interrupt.pending = false;
4809 return EMULATE_DONE;
4811 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4813 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4815 int r = EMULATE_DONE;
4817 ++vcpu->stat.insn_emulation_fail;
4818 trace_kvm_emulate_insn_failed(vcpu);
4819 if (!is_guest_mode(vcpu)) {
4820 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4821 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4822 vcpu->run->internal.ndata = 0;
4825 kvm_queue_exception(vcpu, UD_VECTOR);
4830 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4831 bool write_fault_to_shadow_pgtable,
4837 if (emulation_type & EMULTYPE_NO_REEXECUTE)
4840 if (!vcpu->arch.mmu.direct_map) {
4842 * Write permission should be allowed since only
4843 * write access need to be emulated.
4845 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4848 * If the mapping is invalid in guest, let cpu retry
4849 * it to generate fault.
4851 if (gpa == UNMAPPED_GVA)
4856 * Do not retry the unhandleable instruction if it faults on the
4857 * readonly host memory, otherwise it will goto a infinite loop:
4858 * retry instruction -> write #PF -> emulation fail -> retry
4859 * instruction -> ...
4861 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4864 * If the instruction failed on the error pfn, it can not be fixed,
4865 * report the error to userspace.
4867 if (is_error_noslot_pfn(pfn))
4870 kvm_release_pfn_clean(pfn);
4872 /* The instructions are well-emulated on direct mmu. */
4873 if (vcpu->arch.mmu.direct_map) {
4874 unsigned int indirect_shadow_pages;
4876 spin_lock(&vcpu->kvm->mmu_lock);
4877 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
4878 spin_unlock(&vcpu->kvm->mmu_lock);
4880 if (indirect_shadow_pages)
4881 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4887 * if emulation was due to access to shadowed page table
4888 * and it failed try to unshadow page and re-enter the
4889 * guest to let CPU execute the instruction.
4891 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4894 * If the access faults on its page table, it can not
4895 * be fixed by unprotecting shadow page and it should
4896 * be reported to userspace.
4898 return !write_fault_to_shadow_pgtable;
4901 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4902 unsigned long cr2, int emulation_type)
4904 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4905 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4907 last_retry_eip = vcpu->arch.last_retry_eip;
4908 last_retry_addr = vcpu->arch.last_retry_addr;
4911 * If the emulation is caused by #PF and it is non-page_table
4912 * writing instruction, it means the VM-EXIT is caused by shadow
4913 * page protected, we can zap the shadow page and retry this
4914 * instruction directly.
4916 * Note: if the guest uses a non-page-table modifying instruction
4917 * on the PDE that points to the instruction, then we will unmap
4918 * the instruction and go to an infinite loop. So, we cache the
4919 * last retried eip and the last fault address, if we meet the eip
4920 * and the address again, we can break out of the potential infinite
4923 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4925 if (!(emulation_type & EMULTYPE_RETRY))
4928 if (x86_page_table_writing_insn(ctxt))
4931 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4934 vcpu->arch.last_retry_eip = ctxt->eip;
4935 vcpu->arch.last_retry_addr = cr2;
4937 if (!vcpu->arch.mmu.direct_map)
4938 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4940 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4945 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
4946 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
4948 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
4957 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
4958 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
4963 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, int *r)
4965 struct kvm_run *kvm_run = vcpu->run;
4968 * Use the "raw" value to see if TF was passed to the processor.
4969 * Note that the new value of the flags has not been saved yet.
4971 * This is correct even for TF set by the guest, because "the
4972 * processor will not generate this exception after the instruction
4973 * that sets the TF flag".
4975 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
4977 if (unlikely(rflags & X86_EFLAGS_TF)) {
4978 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
4979 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1;
4980 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
4981 kvm_run->debug.arch.exception = DB_VECTOR;
4982 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4983 *r = EMULATE_USER_EXIT;
4985 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
4987 * "Certain debug exceptions may clear bit 0-3. The
4988 * remaining contents of the DR6 register are never
4989 * cleared by the processor".
4991 vcpu->arch.dr6 &= ~15;
4992 vcpu->arch.dr6 |= DR6_BS;
4993 kvm_queue_exception(vcpu, DB_VECTOR);
4998 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5000 struct kvm_run *kvm_run = vcpu->run;
5001 unsigned long eip = vcpu->arch.emulate_ctxt.eip;
5004 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5005 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5006 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5007 vcpu->arch.guest_debug_dr7,
5011 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5012 kvm_run->debug.arch.pc = kvm_rip_read(vcpu) +
5013 get_segment_base(vcpu, VCPU_SREG_CS);
5015 kvm_run->debug.arch.exception = DB_VECTOR;
5016 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5017 *r = EMULATE_USER_EXIT;
5022 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK)) {
5023 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5028 vcpu->arch.dr6 &= ~15;
5029 vcpu->arch.dr6 |= dr6;
5030 kvm_queue_exception(vcpu, DB_VECTOR);
5039 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5046 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5047 bool writeback = true;
5048 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5051 * Clear write_fault_to_shadow_pgtable here to ensure it is
5054 vcpu->arch.write_fault_to_shadow_pgtable = false;
5055 kvm_clear_exception_queue(vcpu);
5057 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5058 init_emulate_ctxt(vcpu);
5061 * We will reenter on the same instruction since
5062 * we do not set complete_userspace_io. This does not
5063 * handle watchpoints yet, those would be handled in
5066 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5069 ctxt->interruptibility = 0;
5070 ctxt->have_exception = false;
5071 ctxt->perm_ok = false;
5073 ctxt->only_vendor_specific_insn
5074 = emulation_type & EMULTYPE_TRAP_UD;
5076 r = x86_decode_insn(ctxt, insn, insn_len);
5078 trace_kvm_emulate_insn_start(vcpu);
5079 ++vcpu->stat.insn_emulation;
5080 if (r != EMULATION_OK) {
5081 if (emulation_type & EMULTYPE_TRAP_UD)
5082 return EMULATE_FAIL;
5083 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5085 return EMULATE_DONE;
5086 if (emulation_type & EMULTYPE_SKIP)
5087 return EMULATE_FAIL;
5088 return handle_emulation_failure(vcpu);
5092 if (emulation_type & EMULTYPE_SKIP) {
5093 kvm_rip_write(vcpu, ctxt->_eip);
5094 return EMULATE_DONE;
5097 if (retry_instruction(ctxt, cr2, emulation_type))
5098 return EMULATE_DONE;
5100 /* this is needed for vmware backdoor interface to work since it
5101 changes registers values during IO operation */
5102 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5103 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5104 emulator_invalidate_register_cache(ctxt);
5108 r = x86_emulate_insn(ctxt);
5110 if (r == EMULATION_INTERCEPTED)
5111 return EMULATE_DONE;
5113 if (r == EMULATION_FAILED) {
5114 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5116 return EMULATE_DONE;
5118 return handle_emulation_failure(vcpu);
5121 if (ctxt->have_exception) {
5122 inject_emulated_exception(vcpu);
5124 } else if (vcpu->arch.pio.count) {
5125 if (!vcpu->arch.pio.in)
5126 vcpu->arch.pio.count = 0;
5129 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5131 r = EMULATE_USER_EXIT;
5132 } else if (vcpu->mmio_needed) {
5133 if (!vcpu->mmio_is_write)
5135 r = EMULATE_USER_EXIT;
5136 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5137 } else if (r == EMULATION_RESTART)
5143 toggle_interruptibility(vcpu, ctxt->interruptibility);
5144 kvm_make_request(KVM_REQ_EVENT, vcpu);
5145 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5146 kvm_rip_write(vcpu, ctxt->eip);
5147 if (r == EMULATE_DONE)
5148 kvm_vcpu_check_singlestep(vcpu, &r);
5149 kvm_set_rflags(vcpu, ctxt->eflags);
5151 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5155 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5157 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5159 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5160 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5161 size, port, &val, 1);
5162 /* do not return to emulator after return from userspace */
5163 vcpu->arch.pio.count = 0;
5166 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5168 static void tsc_bad(void *info)
5170 __this_cpu_write(cpu_tsc_khz, 0);
5173 static void tsc_khz_changed(void *data)
5175 struct cpufreq_freqs *freq = data;
5176 unsigned long khz = 0;
5180 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5181 khz = cpufreq_quick_get(raw_smp_processor_id());
5184 __this_cpu_write(cpu_tsc_khz, khz);
5187 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5190 struct cpufreq_freqs *freq = data;
5192 struct kvm_vcpu *vcpu;
5193 int i, send_ipi = 0;
5196 * We allow guests to temporarily run on slowing clocks,
5197 * provided we notify them after, or to run on accelerating
5198 * clocks, provided we notify them before. Thus time never
5201 * However, we have a problem. We can't atomically update
5202 * the frequency of a given CPU from this function; it is
5203 * merely a notifier, which can be called from any CPU.
5204 * Changing the TSC frequency at arbitrary points in time
5205 * requires a recomputation of local variables related to
5206 * the TSC for each VCPU. We must flag these local variables
5207 * to be updated and be sure the update takes place with the
5208 * new frequency before any guests proceed.
5210 * Unfortunately, the combination of hotplug CPU and frequency
5211 * change creates an intractable locking scenario; the order
5212 * of when these callouts happen is undefined with respect to
5213 * CPU hotplug, and they can race with each other. As such,
5214 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5215 * undefined; you can actually have a CPU frequency change take
5216 * place in between the computation of X and the setting of the
5217 * variable. To protect against this problem, all updates of
5218 * the per_cpu tsc_khz variable are done in an interrupt
5219 * protected IPI, and all callers wishing to update the value
5220 * must wait for a synchronous IPI to complete (which is trivial
5221 * if the caller is on the CPU already). This establishes the
5222 * necessary total order on variable updates.
5224 * Note that because a guest time update may take place
5225 * anytime after the setting of the VCPU's request bit, the
5226 * correct TSC value must be set before the request. However,
5227 * to ensure the update actually makes it to any guest which
5228 * starts running in hardware virtualization between the set
5229 * and the acquisition of the spinlock, we must also ping the
5230 * CPU after setting the request bit.
5234 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5236 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5239 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5241 raw_spin_lock(&kvm_lock);
5242 list_for_each_entry(kvm, &vm_list, vm_list) {
5243 kvm_for_each_vcpu(i, vcpu, kvm) {
5244 if (vcpu->cpu != freq->cpu)
5246 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5247 if (vcpu->cpu != smp_processor_id())
5251 raw_spin_unlock(&kvm_lock);
5253 if (freq->old < freq->new && send_ipi) {
5255 * We upscale the frequency. Must make the guest
5256 * doesn't see old kvmclock values while running with
5257 * the new frequency, otherwise we risk the guest sees
5258 * time go backwards.
5260 * In case we update the frequency for another cpu
5261 * (which might be in guest context) send an interrupt
5262 * to kick the cpu out of guest context. Next time
5263 * guest context is entered kvmclock will be updated,
5264 * so the guest will not see stale values.
5266 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5271 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5272 .notifier_call = kvmclock_cpufreq_notifier
5275 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5276 unsigned long action, void *hcpu)
5278 unsigned int cpu = (unsigned long)hcpu;
5282 case CPU_DOWN_FAILED:
5283 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5285 case CPU_DOWN_PREPARE:
5286 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5292 static struct notifier_block kvmclock_cpu_notifier_block = {
5293 .notifier_call = kvmclock_cpu_notifier,
5294 .priority = -INT_MAX
5297 static void kvm_timer_init(void)
5301 max_tsc_khz = tsc_khz;
5302 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5303 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5304 #ifdef CONFIG_CPU_FREQ
5305 struct cpufreq_policy policy;
5306 memset(&policy, 0, sizeof(policy));
5308 cpufreq_get_policy(&policy, cpu);
5309 if (policy.cpuinfo.max_freq)
5310 max_tsc_khz = policy.cpuinfo.max_freq;
5313 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5314 CPUFREQ_TRANSITION_NOTIFIER);
5316 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5317 for_each_online_cpu(cpu)
5318 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5321 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5323 int kvm_is_in_guest(void)
5325 return __this_cpu_read(current_vcpu) != NULL;
5328 static int kvm_is_user_mode(void)
5332 if (__this_cpu_read(current_vcpu))
5333 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5335 return user_mode != 0;
5338 static unsigned long kvm_get_guest_ip(void)
5340 unsigned long ip = 0;
5342 if (__this_cpu_read(current_vcpu))
5343 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5348 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5349 .is_in_guest = kvm_is_in_guest,
5350 .is_user_mode = kvm_is_user_mode,
5351 .get_guest_ip = kvm_get_guest_ip,
5354 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5356 __this_cpu_write(current_vcpu, vcpu);
5358 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5360 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5362 __this_cpu_write(current_vcpu, NULL);
5364 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5366 static void kvm_set_mmio_spte_mask(void)
5369 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5372 * Set the reserved bits and the present bit of an paging-structure
5373 * entry to generate page fault with PFER.RSV = 1.
5375 /* Mask the reserved physical address bits. */
5376 mask = ((1ull << (51 - maxphyaddr + 1)) - 1) << maxphyaddr;
5378 /* Bit 62 is always reserved for 32bit host. */
5379 mask |= 0x3ull << 62;
5381 /* Set the present bit. */
5384 #ifdef CONFIG_X86_64
5386 * If reserved bit is not supported, clear the present bit to disable
5389 if (maxphyaddr == 52)
5393 kvm_mmu_set_mmio_spte_mask(mask);
5396 #ifdef CONFIG_X86_64
5397 static void pvclock_gtod_update_fn(struct work_struct *work)
5401 struct kvm_vcpu *vcpu;
5404 raw_spin_lock(&kvm_lock);
5405 list_for_each_entry(kvm, &vm_list, vm_list)
5406 kvm_for_each_vcpu(i, vcpu, kvm)
5407 set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5408 atomic_set(&kvm_guest_has_master_clock, 0);
5409 raw_spin_unlock(&kvm_lock);
5412 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5415 * Notification about pvclock gtod data update.
5417 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5420 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5421 struct timekeeper *tk = priv;
5423 update_pvclock_gtod(tk);
5425 /* disable master clock if host does not trust, or does not
5426 * use, TSC clocksource
5428 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5429 atomic_read(&kvm_guest_has_master_clock) != 0)
5430 queue_work(system_long_wq, &pvclock_gtod_work);
5435 static struct notifier_block pvclock_gtod_notifier = {
5436 .notifier_call = pvclock_gtod_notify,
5440 int kvm_arch_init(void *opaque)
5443 struct kvm_x86_ops *ops = opaque;
5446 printk(KERN_ERR "kvm: already loaded the other module\n");
5451 if (!ops->cpu_has_kvm_support()) {
5452 printk(KERN_ERR "kvm: no hardware support\n");
5456 if (ops->disabled_by_bios()) {
5457 printk(KERN_ERR "kvm: disabled by bios\n");
5463 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5465 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5469 r = kvm_mmu_module_init();
5471 goto out_free_percpu;
5473 kvm_set_mmio_spte_mask();
5474 kvm_init_msr_list();
5477 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5478 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5482 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5485 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5488 #ifdef CONFIG_X86_64
5489 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5495 free_percpu(shared_msrs);
5500 void kvm_arch_exit(void)
5502 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5504 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5505 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5506 CPUFREQ_TRANSITION_NOTIFIER);
5507 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5508 #ifdef CONFIG_X86_64
5509 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5512 kvm_mmu_module_exit();
5513 free_percpu(shared_msrs);
5516 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5518 ++vcpu->stat.halt_exits;
5519 if (irqchip_in_kernel(vcpu->kvm)) {
5520 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5523 vcpu->run->exit_reason = KVM_EXIT_HLT;
5527 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5529 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5531 u64 param, ingpa, outgpa, ret;
5532 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5533 bool fast, longmode;
5537 * hypercall generates UD from non zero cpl and real mode
5540 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5541 kvm_queue_exception(vcpu, UD_VECTOR);
5545 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5546 longmode = is_long_mode(vcpu) && cs_l == 1;
5549 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5550 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5551 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5552 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5553 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5554 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5556 #ifdef CONFIG_X86_64
5558 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5559 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5560 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5564 code = param & 0xffff;
5565 fast = (param >> 16) & 0x1;
5566 rep_cnt = (param >> 32) & 0xfff;
5567 rep_idx = (param >> 48) & 0xfff;
5569 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5572 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5573 kvm_vcpu_on_spin(vcpu);
5576 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5580 ret = res | (((u64)rep_done & 0xfff) << 32);
5582 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5584 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5585 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5592 * kvm_pv_kick_cpu_op: Kick a vcpu.
5594 * @apicid - apicid of vcpu to be kicked.
5596 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5598 struct kvm_lapic_irq lapic_irq;
5600 lapic_irq.shorthand = 0;
5601 lapic_irq.dest_mode = 0;
5602 lapic_irq.dest_id = apicid;
5604 lapic_irq.delivery_mode = APIC_DM_REMRD;
5605 kvm_irq_delivery_to_apic(kvm, 0, &lapic_irq, NULL);
5608 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5610 unsigned long nr, a0, a1, a2, a3, ret;
5613 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5614 return kvm_hv_hypercall(vcpu);
5616 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5617 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5618 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5619 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5620 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5622 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5624 if (!is_long_mode(vcpu)) {
5632 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5638 case KVM_HC_VAPIC_POLL_IRQ:
5641 case KVM_HC_KICK_CPU:
5642 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5650 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5651 ++vcpu->stat.hypercalls;
5654 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5656 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5658 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5659 char instruction[3];
5660 unsigned long rip = kvm_rip_read(vcpu);
5662 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5664 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5668 * Check if userspace requested an interrupt window, and that the
5669 * interrupt window is open.
5671 * No need to exit to userspace if we already have an interrupt queued.
5673 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5675 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5676 vcpu->run->request_interrupt_window &&
5677 kvm_arch_interrupt_allowed(vcpu));
5680 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5682 struct kvm_run *kvm_run = vcpu->run;
5684 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5685 kvm_run->cr8 = kvm_get_cr8(vcpu);
5686 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5687 if (irqchip_in_kernel(vcpu->kvm))
5688 kvm_run->ready_for_interrupt_injection = 1;
5690 kvm_run->ready_for_interrupt_injection =
5691 kvm_arch_interrupt_allowed(vcpu) &&
5692 !kvm_cpu_has_interrupt(vcpu) &&
5693 !kvm_event_needs_reinjection(vcpu);
5696 static int vapic_enter(struct kvm_vcpu *vcpu)
5698 struct kvm_lapic *apic = vcpu->arch.apic;
5701 if (!apic || !apic->vapic_addr)
5704 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5705 if (is_error_page(page))
5708 vcpu->arch.apic->vapic_page = page;
5712 static void vapic_exit(struct kvm_vcpu *vcpu)
5714 struct kvm_lapic *apic = vcpu->arch.apic;
5717 if (!apic || !apic->vapic_addr)
5720 idx = srcu_read_lock(&vcpu->kvm->srcu);
5721 kvm_release_page_dirty(apic->vapic_page);
5722 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5723 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5726 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5730 if (!kvm_x86_ops->update_cr8_intercept)
5733 if (!vcpu->arch.apic)
5736 if (!vcpu->arch.apic->vapic_addr)
5737 max_irr = kvm_lapic_find_highest_irr(vcpu);
5744 tpr = kvm_lapic_get_cr8(vcpu);
5746 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5749 static void inject_pending_event(struct kvm_vcpu *vcpu)
5751 /* try to reinject previous events if any */
5752 if (vcpu->arch.exception.pending) {
5753 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5754 vcpu->arch.exception.has_error_code,
5755 vcpu->arch.exception.error_code);
5756 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5757 vcpu->arch.exception.has_error_code,
5758 vcpu->arch.exception.error_code,
5759 vcpu->arch.exception.reinject);
5763 if (vcpu->arch.nmi_injected) {
5764 kvm_x86_ops->set_nmi(vcpu);
5768 if (vcpu->arch.interrupt.pending) {
5769 kvm_x86_ops->set_irq(vcpu);
5773 /* try to inject new event if pending */
5774 if (vcpu->arch.nmi_pending) {
5775 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5776 --vcpu->arch.nmi_pending;
5777 vcpu->arch.nmi_injected = true;
5778 kvm_x86_ops->set_nmi(vcpu);
5780 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
5781 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5782 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5784 kvm_x86_ops->set_irq(vcpu);
5789 static void process_nmi(struct kvm_vcpu *vcpu)
5794 * x86 is limited to one NMI running, and one NMI pending after it.
5795 * If an NMI is already in progress, limit further NMIs to just one.
5796 * Otherwise, allow two (and we'll inject the first one immediately).
5798 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5801 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5802 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5803 kvm_make_request(KVM_REQ_EVENT, vcpu);
5806 static void kvm_gen_update_masterclock(struct kvm *kvm)
5808 #ifdef CONFIG_X86_64
5810 struct kvm_vcpu *vcpu;
5811 struct kvm_arch *ka = &kvm->arch;
5813 spin_lock(&ka->pvclock_gtod_sync_lock);
5814 kvm_make_mclock_inprogress_request(kvm);
5815 /* no guest entries from this point */
5816 pvclock_update_vm_gtod_copy(kvm);
5818 kvm_for_each_vcpu(i, vcpu, kvm)
5819 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
5821 /* guest entries allowed */
5822 kvm_for_each_vcpu(i, vcpu, kvm)
5823 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
5825 spin_unlock(&ka->pvclock_gtod_sync_lock);
5829 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
5831 u64 eoi_exit_bitmap[4];
5834 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
5837 memset(eoi_exit_bitmap, 0, 32);
5840 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
5841 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
5842 kvm_apic_update_tmr(vcpu, tmr);
5845 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5848 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5849 vcpu->run->request_interrupt_window;
5850 bool req_immediate_exit = false;
5852 if (vcpu->requests) {
5853 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5854 kvm_mmu_unload(vcpu);
5855 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5856 __kvm_migrate_timers(vcpu);
5857 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5858 kvm_gen_update_masterclock(vcpu->kvm);
5859 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
5860 kvm_gen_kvmclock_update(vcpu);
5861 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5862 r = kvm_guest_time_update(vcpu);
5866 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5867 kvm_mmu_sync_roots(vcpu);
5868 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5869 kvm_x86_ops->tlb_flush(vcpu);
5870 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5871 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5875 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5876 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5880 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5881 vcpu->fpu_active = 0;
5882 kvm_x86_ops->fpu_deactivate(vcpu);
5884 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5885 /* Page is swapped out. Do synthetic halt */
5886 vcpu->arch.apf.halted = true;
5890 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5891 record_steal_time(vcpu);
5892 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5894 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5895 kvm_handle_pmu_event(vcpu);
5896 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5897 kvm_deliver_pmi(vcpu);
5898 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
5899 vcpu_scan_ioapic(vcpu);
5902 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5903 kvm_apic_accept_events(vcpu);
5904 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
5909 inject_pending_event(vcpu);
5911 /* enable NMI/IRQ window open exits if needed */
5912 if (vcpu->arch.nmi_pending)
5913 req_immediate_exit =
5914 kvm_x86_ops->enable_nmi_window(vcpu) != 0;
5915 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
5916 req_immediate_exit =
5917 kvm_x86_ops->enable_irq_window(vcpu) != 0;
5919 if (kvm_lapic_enabled(vcpu)) {
5921 * Update architecture specific hints for APIC
5922 * virtual interrupt delivery.
5924 if (kvm_x86_ops->hwapic_irr_update)
5925 kvm_x86_ops->hwapic_irr_update(vcpu,
5926 kvm_lapic_find_highest_irr(vcpu));
5927 update_cr8_intercept(vcpu);
5928 kvm_lapic_sync_to_vapic(vcpu);
5932 r = kvm_mmu_reload(vcpu);
5934 goto cancel_injection;
5939 kvm_x86_ops->prepare_guest_switch(vcpu);
5940 if (vcpu->fpu_active)
5941 kvm_load_guest_fpu(vcpu);
5942 kvm_load_guest_xcr0(vcpu);
5944 vcpu->mode = IN_GUEST_MODE;
5946 /* We should set ->mode before check ->requests,
5947 * see the comment in make_all_cpus_request.
5951 local_irq_disable();
5953 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5954 || need_resched() || signal_pending(current)) {
5955 vcpu->mode = OUTSIDE_GUEST_MODE;
5960 goto cancel_injection;
5963 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5965 if (req_immediate_exit)
5966 smp_send_reschedule(vcpu->cpu);
5970 if (unlikely(vcpu->arch.switch_db_regs)) {
5972 set_debugreg(vcpu->arch.eff_db[0], 0);
5973 set_debugreg(vcpu->arch.eff_db[1], 1);
5974 set_debugreg(vcpu->arch.eff_db[2], 2);
5975 set_debugreg(vcpu->arch.eff_db[3], 3);
5978 trace_kvm_entry(vcpu->vcpu_id);
5979 kvm_x86_ops->run(vcpu);
5982 * If the guest has used debug registers, at least dr7
5983 * will be disabled while returning to the host.
5984 * If we don't have active breakpoints in the host, we don't
5985 * care about the messed up debug address registers. But if
5986 * we have some of them active, restore the old state.
5988 if (hw_breakpoint_active())
5989 hw_breakpoint_restore();
5991 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
5994 vcpu->mode = OUTSIDE_GUEST_MODE;
5997 /* Interrupt is enabled by handle_external_intr() */
5998 kvm_x86_ops->handle_external_intr(vcpu);
6003 * We must have an instruction between local_irq_enable() and
6004 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6005 * the interrupt shadow. The stat.exits increment will do nicely.
6006 * But we need to prevent reordering, hence this barrier():
6014 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6017 * Profile KVM exit RIPs:
6019 if (unlikely(prof_on == KVM_PROFILING)) {
6020 unsigned long rip = kvm_rip_read(vcpu);
6021 profile_hit(KVM_PROFILING, (void *)rip);
6024 if (unlikely(vcpu->arch.tsc_always_catchup))
6025 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6027 if (vcpu->arch.apic_attention)
6028 kvm_lapic_sync_from_vapic(vcpu);
6030 r = kvm_x86_ops->handle_exit(vcpu);
6034 kvm_x86_ops->cancel_injection(vcpu);
6035 if (unlikely(vcpu->arch.apic_attention))
6036 kvm_lapic_sync_from_vapic(vcpu);
6042 static int __vcpu_run(struct kvm_vcpu *vcpu)
6045 struct kvm *kvm = vcpu->kvm;
6047 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6048 r = vapic_enter(vcpu);
6050 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6056 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6057 !vcpu->arch.apf.halted)
6058 r = vcpu_enter_guest(vcpu);
6060 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6061 kvm_vcpu_block(vcpu);
6062 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6063 if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
6064 kvm_apic_accept_events(vcpu);
6065 switch(vcpu->arch.mp_state) {
6066 case KVM_MP_STATE_HALTED:
6067 vcpu->arch.pv.pv_unhalted = false;
6068 vcpu->arch.mp_state =
6069 KVM_MP_STATE_RUNNABLE;
6070 case KVM_MP_STATE_RUNNABLE:
6071 vcpu->arch.apf.halted = false;
6073 case KVM_MP_STATE_INIT_RECEIVED:
6085 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6086 if (kvm_cpu_has_pending_timer(vcpu))
6087 kvm_inject_pending_timer_irqs(vcpu);
6089 if (dm_request_for_irq_injection(vcpu)) {
6091 vcpu->run->exit_reason = KVM_EXIT_INTR;
6092 ++vcpu->stat.request_irq_exits;
6095 kvm_check_async_pf_completion(vcpu);
6097 if (signal_pending(current)) {
6099 vcpu->run->exit_reason = KVM_EXIT_INTR;
6100 ++vcpu->stat.signal_exits;
6102 if (need_resched()) {
6103 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6105 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6109 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6116 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6119 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6120 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6121 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6122 if (r != EMULATE_DONE)
6127 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6129 BUG_ON(!vcpu->arch.pio.count);
6131 return complete_emulated_io(vcpu);
6135 * Implements the following, as a state machine:
6139 * for each mmio piece in the fragment
6147 * for each mmio piece in the fragment
6152 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6154 struct kvm_run *run = vcpu->run;
6155 struct kvm_mmio_fragment *frag;
6158 BUG_ON(!vcpu->mmio_needed);
6160 /* Complete previous fragment */
6161 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6162 len = min(8u, frag->len);
6163 if (!vcpu->mmio_is_write)
6164 memcpy(frag->data, run->mmio.data, len);
6166 if (frag->len <= 8) {
6167 /* Switch to the next fragment. */
6169 vcpu->mmio_cur_fragment++;
6171 /* Go forward to the next mmio piece. */
6177 if (vcpu->mmio_cur_fragment == vcpu->mmio_nr_fragments) {
6178 vcpu->mmio_needed = 0;
6179 if (vcpu->mmio_is_write)
6181 vcpu->mmio_read_completed = 1;
6182 return complete_emulated_io(vcpu);
6185 run->exit_reason = KVM_EXIT_MMIO;
6186 run->mmio.phys_addr = frag->gpa;
6187 if (vcpu->mmio_is_write)
6188 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6189 run->mmio.len = min(8u, frag->len);
6190 run->mmio.is_write = vcpu->mmio_is_write;
6191 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6196 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6201 if (!tsk_used_math(current) && init_fpu(current))
6204 if (vcpu->sigset_active)
6205 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6207 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6208 kvm_vcpu_block(vcpu);
6209 kvm_apic_accept_events(vcpu);
6210 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6215 /* re-sync apic's tpr */
6216 if (!irqchip_in_kernel(vcpu->kvm)) {
6217 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6223 if (unlikely(vcpu->arch.complete_userspace_io)) {
6224 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6225 vcpu->arch.complete_userspace_io = NULL;
6230 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6232 r = __vcpu_run(vcpu);
6235 post_kvm_run_save(vcpu);
6236 if (vcpu->sigset_active)
6237 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6242 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6244 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6246 * We are here if userspace calls get_regs() in the middle of
6247 * instruction emulation. Registers state needs to be copied
6248 * back from emulation context to vcpu. Userspace shouldn't do
6249 * that usually, but some bad designed PV devices (vmware
6250 * backdoor interface) need this to work
6252 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6253 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6255 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6256 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6257 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6258 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6259 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6260 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6261 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6262 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6263 #ifdef CONFIG_X86_64
6264 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6265 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6266 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6267 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6268 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6269 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6270 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6271 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6274 regs->rip = kvm_rip_read(vcpu);
6275 regs->rflags = kvm_get_rflags(vcpu);
6280 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6282 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6283 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6285 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6286 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6287 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6288 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6289 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6290 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6291 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6292 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6293 #ifdef CONFIG_X86_64
6294 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6295 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6296 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6297 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6298 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6299 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6300 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6301 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6304 kvm_rip_write(vcpu, regs->rip);
6305 kvm_set_rflags(vcpu, regs->rflags);
6307 vcpu->arch.exception.pending = false;
6309 kvm_make_request(KVM_REQ_EVENT, vcpu);
6314 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6316 struct kvm_segment cs;
6318 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6322 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6324 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6325 struct kvm_sregs *sregs)
6329 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6330 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6331 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6332 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6333 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6334 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6336 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6337 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6339 kvm_x86_ops->get_idt(vcpu, &dt);
6340 sregs->idt.limit = dt.size;
6341 sregs->idt.base = dt.address;
6342 kvm_x86_ops->get_gdt(vcpu, &dt);
6343 sregs->gdt.limit = dt.size;
6344 sregs->gdt.base = dt.address;
6346 sregs->cr0 = kvm_read_cr0(vcpu);
6347 sregs->cr2 = vcpu->arch.cr2;
6348 sregs->cr3 = kvm_read_cr3(vcpu);
6349 sregs->cr4 = kvm_read_cr4(vcpu);
6350 sregs->cr8 = kvm_get_cr8(vcpu);
6351 sregs->efer = vcpu->arch.efer;
6352 sregs->apic_base = kvm_get_apic_base(vcpu);
6354 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6356 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6357 set_bit(vcpu->arch.interrupt.nr,
6358 (unsigned long *)sregs->interrupt_bitmap);
6363 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6364 struct kvm_mp_state *mp_state)
6366 kvm_apic_accept_events(vcpu);
6367 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6368 vcpu->arch.pv.pv_unhalted)
6369 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6371 mp_state->mp_state = vcpu->arch.mp_state;
6376 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6377 struct kvm_mp_state *mp_state)
6379 if (!kvm_vcpu_has_lapic(vcpu) &&
6380 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6383 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6384 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6385 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6387 vcpu->arch.mp_state = mp_state->mp_state;
6388 kvm_make_request(KVM_REQ_EVENT, vcpu);
6392 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6393 int reason, bool has_error_code, u32 error_code)
6395 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6398 init_emulate_ctxt(vcpu);
6400 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6401 has_error_code, error_code);
6404 return EMULATE_FAIL;
6406 kvm_rip_write(vcpu, ctxt->eip);
6407 kvm_set_rflags(vcpu, ctxt->eflags);
6408 kvm_make_request(KVM_REQ_EVENT, vcpu);
6409 return EMULATE_DONE;
6411 EXPORT_SYMBOL_GPL(kvm_task_switch);
6413 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6414 struct kvm_sregs *sregs)
6416 int mmu_reset_needed = 0;
6417 int pending_vec, max_bits, idx;
6420 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6423 dt.size = sregs->idt.limit;
6424 dt.address = sregs->idt.base;
6425 kvm_x86_ops->set_idt(vcpu, &dt);
6426 dt.size = sregs->gdt.limit;
6427 dt.address = sregs->gdt.base;
6428 kvm_x86_ops->set_gdt(vcpu, &dt);
6430 vcpu->arch.cr2 = sregs->cr2;
6431 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6432 vcpu->arch.cr3 = sregs->cr3;
6433 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6435 kvm_set_cr8(vcpu, sregs->cr8);
6437 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6438 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6439 kvm_set_apic_base(vcpu, sregs->apic_base);
6441 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6442 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6443 vcpu->arch.cr0 = sregs->cr0;
6445 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6446 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6447 if (sregs->cr4 & X86_CR4_OSXSAVE)
6448 kvm_update_cpuid(vcpu);
6450 idx = srcu_read_lock(&vcpu->kvm->srcu);
6451 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6452 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6453 mmu_reset_needed = 1;
6455 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6457 if (mmu_reset_needed)
6458 kvm_mmu_reset_context(vcpu);
6460 max_bits = KVM_NR_INTERRUPTS;
6461 pending_vec = find_first_bit(
6462 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6463 if (pending_vec < max_bits) {
6464 kvm_queue_interrupt(vcpu, pending_vec, false);
6465 pr_debug("Set back pending irq %d\n", pending_vec);
6468 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6469 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6470 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6471 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6472 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6473 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6475 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6476 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6478 update_cr8_intercept(vcpu);
6480 /* Older userspace won't unhalt the vcpu on reset. */
6481 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6482 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6484 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6486 kvm_make_request(KVM_REQ_EVENT, vcpu);
6491 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6492 struct kvm_guest_debug *dbg)
6494 unsigned long rflags;
6497 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6499 if (vcpu->arch.exception.pending)
6501 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6502 kvm_queue_exception(vcpu, DB_VECTOR);
6504 kvm_queue_exception(vcpu, BP_VECTOR);
6508 * Read rflags as long as potentially injected trace flags are still
6511 rflags = kvm_get_rflags(vcpu);
6513 vcpu->guest_debug = dbg->control;
6514 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6515 vcpu->guest_debug = 0;
6517 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6518 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6519 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6520 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6522 for (i = 0; i < KVM_NR_DB_REGS; i++)
6523 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6525 kvm_update_dr7(vcpu);
6527 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6528 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6529 get_segment_base(vcpu, VCPU_SREG_CS);
6532 * Trigger an rflags update that will inject or remove the trace
6535 kvm_set_rflags(vcpu, rflags);
6537 kvm_x86_ops->update_db_bp_intercept(vcpu);
6547 * Translate a guest virtual address to a guest physical address.
6549 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6550 struct kvm_translation *tr)
6552 unsigned long vaddr = tr->linear_address;
6556 idx = srcu_read_lock(&vcpu->kvm->srcu);
6557 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6558 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6559 tr->physical_address = gpa;
6560 tr->valid = gpa != UNMAPPED_GVA;
6567 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6569 struct i387_fxsave_struct *fxsave =
6570 &vcpu->arch.guest_fpu.state->fxsave;
6572 memcpy(fpu->fpr, fxsave->st_space, 128);
6573 fpu->fcw = fxsave->cwd;
6574 fpu->fsw = fxsave->swd;
6575 fpu->ftwx = fxsave->twd;
6576 fpu->last_opcode = fxsave->fop;
6577 fpu->last_ip = fxsave->rip;
6578 fpu->last_dp = fxsave->rdp;
6579 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6584 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6586 struct i387_fxsave_struct *fxsave =
6587 &vcpu->arch.guest_fpu.state->fxsave;
6589 memcpy(fxsave->st_space, fpu->fpr, 128);
6590 fxsave->cwd = fpu->fcw;
6591 fxsave->swd = fpu->fsw;
6592 fxsave->twd = fpu->ftwx;
6593 fxsave->fop = fpu->last_opcode;
6594 fxsave->rip = fpu->last_ip;
6595 fxsave->rdp = fpu->last_dp;
6596 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6601 int fx_init(struct kvm_vcpu *vcpu)
6605 err = fpu_alloc(&vcpu->arch.guest_fpu);
6609 fpu_finit(&vcpu->arch.guest_fpu);
6612 * Ensure guest xcr0 is valid for loading
6614 vcpu->arch.xcr0 = XSTATE_FP;
6616 vcpu->arch.cr0 |= X86_CR0_ET;
6620 EXPORT_SYMBOL_GPL(fx_init);
6622 static void fx_free(struct kvm_vcpu *vcpu)
6624 fpu_free(&vcpu->arch.guest_fpu);
6627 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6629 if (vcpu->guest_fpu_loaded)
6633 * Restore all possible states in the guest,
6634 * and assume host would use all available bits.
6635 * Guest xcr0 would be loaded later.
6637 kvm_put_guest_xcr0(vcpu);
6638 vcpu->guest_fpu_loaded = 1;
6639 __kernel_fpu_begin();
6640 fpu_restore_checking(&vcpu->arch.guest_fpu);
6644 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6646 kvm_put_guest_xcr0(vcpu);
6648 if (!vcpu->guest_fpu_loaded)
6651 vcpu->guest_fpu_loaded = 0;
6652 fpu_save_init(&vcpu->arch.guest_fpu);
6654 ++vcpu->stat.fpu_reload;
6655 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6659 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6661 kvmclock_reset(vcpu);
6663 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6665 kvm_x86_ops->vcpu_free(vcpu);
6668 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6671 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6672 printk_once(KERN_WARNING
6673 "kvm: SMP vm created on host with unstable TSC; "
6674 "guest TSC will not be reliable\n");
6675 return kvm_x86_ops->vcpu_create(kvm, id);
6678 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6682 vcpu->arch.mtrr_state.have_fixed = 1;
6683 r = vcpu_load(vcpu);
6686 kvm_vcpu_reset(vcpu);
6687 r = kvm_mmu_setup(vcpu);
6693 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6696 struct msr_data msr;
6698 r = vcpu_load(vcpu);
6702 msr.index = MSR_IA32_TSC;
6703 msr.host_initiated = true;
6704 kvm_write_tsc(vcpu, &msr);
6710 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6713 vcpu->arch.apf.msr_val = 0;
6715 r = vcpu_load(vcpu);
6717 kvm_mmu_unload(vcpu);
6721 kvm_x86_ops->vcpu_free(vcpu);
6724 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6726 atomic_set(&vcpu->arch.nmi_queued, 0);
6727 vcpu->arch.nmi_pending = 0;
6728 vcpu->arch.nmi_injected = false;
6730 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6731 vcpu->arch.dr6 = DR6_FIXED_1;
6732 vcpu->arch.dr7 = DR7_FIXED_1;
6733 kvm_update_dr7(vcpu);
6735 kvm_make_request(KVM_REQ_EVENT, vcpu);
6736 vcpu->arch.apf.msr_val = 0;
6737 vcpu->arch.st.msr_val = 0;
6739 kvmclock_reset(vcpu);
6741 kvm_clear_async_pf_completion_queue(vcpu);
6742 kvm_async_pf_hash_reset(vcpu);
6743 vcpu->arch.apf.halted = false;
6745 kvm_pmu_reset(vcpu);
6747 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6748 vcpu->arch.regs_avail = ~0;
6749 vcpu->arch.regs_dirty = ~0;
6751 kvm_x86_ops->vcpu_reset(vcpu);
6754 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, unsigned int vector)
6756 struct kvm_segment cs;
6758 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6759 cs.selector = vector << 8;
6760 cs.base = vector << 12;
6761 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6762 kvm_rip_write(vcpu, 0);
6765 int kvm_arch_hardware_enable(void *garbage)
6768 struct kvm_vcpu *vcpu;
6773 bool stable, backwards_tsc = false;
6775 kvm_shared_msr_cpu_online();
6776 ret = kvm_x86_ops->hardware_enable(garbage);
6780 local_tsc = native_read_tsc();
6781 stable = !check_tsc_unstable();
6782 list_for_each_entry(kvm, &vm_list, vm_list) {
6783 kvm_for_each_vcpu(i, vcpu, kvm) {
6784 if (!stable && vcpu->cpu == smp_processor_id())
6785 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6786 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6787 backwards_tsc = true;
6788 if (vcpu->arch.last_host_tsc > max_tsc)
6789 max_tsc = vcpu->arch.last_host_tsc;
6795 * Sometimes, even reliable TSCs go backwards. This happens on
6796 * platforms that reset TSC during suspend or hibernate actions, but
6797 * maintain synchronization. We must compensate. Fortunately, we can
6798 * detect that condition here, which happens early in CPU bringup,
6799 * before any KVM threads can be running. Unfortunately, we can't
6800 * bring the TSCs fully up to date with real time, as we aren't yet far
6801 * enough into CPU bringup that we know how much real time has actually
6802 * elapsed; our helper function, get_kernel_ns() will be using boot
6803 * variables that haven't been updated yet.
6805 * So we simply find the maximum observed TSC above, then record the
6806 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6807 * the adjustment will be applied. Note that we accumulate
6808 * adjustments, in case multiple suspend cycles happen before some VCPU
6809 * gets a chance to run again. In the event that no KVM threads get a
6810 * chance to run, we will miss the entire elapsed period, as we'll have
6811 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6812 * loose cycle time. This isn't too big a deal, since the loss will be
6813 * uniform across all VCPUs (not to mention the scenario is extremely
6814 * unlikely). It is possible that a second hibernate recovery happens
6815 * much faster than a first, causing the observed TSC here to be
6816 * smaller; this would require additional padding adjustment, which is
6817 * why we set last_host_tsc to the local tsc observed here.
6819 * N.B. - this code below runs only on platforms with reliable TSC,
6820 * as that is the only way backwards_tsc is set above. Also note
6821 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6822 * have the same delta_cyc adjustment applied if backwards_tsc
6823 * is detected. Note further, this adjustment is only done once,
6824 * as we reset last_host_tsc on all VCPUs to stop this from being
6825 * called multiple times (one for each physical CPU bringup).
6827 * Platforms with unreliable TSCs don't have to deal with this, they
6828 * will be compensated by the logic in vcpu_load, which sets the TSC to
6829 * catchup mode. This will catchup all VCPUs to real time, but cannot
6830 * guarantee that they stay in perfect synchronization.
6832 if (backwards_tsc) {
6833 u64 delta_cyc = max_tsc - local_tsc;
6834 list_for_each_entry(kvm, &vm_list, vm_list) {
6835 kvm_for_each_vcpu(i, vcpu, kvm) {
6836 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6837 vcpu->arch.last_host_tsc = local_tsc;
6838 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6843 * We have to disable TSC offset matching.. if you were
6844 * booting a VM while issuing an S4 host suspend....
6845 * you may have some problem. Solving this issue is
6846 * left as an exercise to the reader.
6848 kvm->arch.last_tsc_nsec = 0;
6849 kvm->arch.last_tsc_write = 0;
6856 void kvm_arch_hardware_disable(void *garbage)
6858 kvm_x86_ops->hardware_disable(garbage);
6859 drop_user_return_notifiers(garbage);
6862 int kvm_arch_hardware_setup(void)
6864 return kvm_x86_ops->hardware_setup();
6867 void kvm_arch_hardware_unsetup(void)
6869 kvm_x86_ops->hardware_unsetup();
6872 void kvm_arch_check_processor_compat(void *rtn)
6874 kvm_x86_ops->check_processor_compatibility(rtn);
6877 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6879 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6882 struct static_key kvm_no_apic_vcpu __read_mostly;
6884 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6890 BUG_ON(vcpu->kvm == NULL);
6893 vcpu->arch.pv.pv_unhalted = false;
6894 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6895 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6896 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6898 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6900 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6905 vcpu->arch.pio_data = page_address(page);
6907 kvm_set_tsc_khz(vcpu, max_tsc_khz);
6909 r = kvm_mmu_create(vcpu);
6911 goto fail_free_pio_data;
6913 if (irqchip_in_kernel(kvm)) {
6914 r = kvm_create_lapic(vcpu);
6916 goto fail_mmu_destroy;
6918 static_key_slow_inc(&kvm_no_apic_vcpu);
6920 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6922 if (!vcpu->arch.mce_banks) {
6924 goto fail_free_lapic;
6926 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6928 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
6930 goto fail_free_mce_banks;
6935 goto fail_free_wbinvd_dirty_mask;
6937 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
6938 vcpu->arch.pv_time_enabled = false;
6939 kvm_async_pf_hash_reset(vcpu);
6943 fail_free_wbinvd_dirty_mask:
6944 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6945 fail_free_mce_banks:
6946 kfree(vcpu->arch.mce_banks);
6948 kvm_free_lapic(vcpu);
6950 kvm_mmu_destroy(vcpu);
6952 free_page((unsigned long)vcpu->arch.pio_data);
6957 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6961 kvm_pmu_destroy(vcpu);
6962 kfree(vcpu->arch.mce_banks);
6963 kvm_free_lapic(vcpu);
6964 idx = srcu_read_lock(&vcpu->kvm->srcu);
6965 kvm_mmu_destroy(vcpu);
6966 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6967 free_page((unsigned long)vcpu->arch.pio_data);
6968 if (!irqchip_in_kernel(vcpu->kvm))
6969 static_key_slow_dec(&kvm_no_apic_vcpu);
6972 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
6977 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6978 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
6979 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6981 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6982 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6983 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6984 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
6985 &kvm->arch.irq_sources_bitmap);
6987 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6988 mutex_init(&kvm->arch.apic_map_lock);
6989 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
6991 pvclock_update_vm_gtod_copy(kvm);
6996 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6999 r = vcpu_load(vcpu);
7001 kvm_mmu_unload(vcpu);
7005 static void kvm_free_vcpus(struct kvm *kvm)
7008 struct kvm_vcpu *vcpu;
7011 * Unpin any mmu pages first.
7013 kvm_for_each_vcpu(i, vcpu, kvm) {
7014 kvm_clear_async_pf_completion_queue(vcpu);
7015 kvm_unload_vcpu_mmu(vcpu);
7017 kvm_for_each_vcpu(i, vcpu, kvm)
7018 kvm_arch_vcpu_free(vcpu);
7020 mutex_lock(&kvm->lock);
7021 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7022 kvm->vcpus[i] = NULL;
7024 atomic_set(&kvm->online_vcpus, 0);
7025 mutex_unlock(&kvm->lock);
7028 void kvm_arch_sync_events(struct kvm *kvm)
7030 kvm_free_all_assigned_devices(kvm);
7034 void kvm_arch_destroy_vm(struct kvm *kvm)
7036 if (current->mm == kvm->mm) {
7038 * Free memory regions allocated on behalf of userspace,
7039 * unless the the memory map has changed due to process exit
7042 struct kvm_userspace_memory_region mem;
7043 memset(&mem, 0, sizeof(mem));
7044 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7045 kvm_set_memory_region(kvm, &mem);
7047 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7048 kvm_set_memory_region(kvm, &mem);
7050 mem.slot = TSS_PRIVATE_MEMSLOT;
7051 kvm_set_memory_region(kvm, &mem);
7053 kvm_iommu_unmap_guest(kvm);
7054 kfree(kvm->arch.vpic);
7055 kfree(kvm->arch.vioapic);
7056 kvm_free_vcpus(kvm);
7057 if (kvm->arch.apic_access_page)
7058 put_page(kvm->arch.apic_access_page);
7059 if (kvm->arch.ept_identity_pagetable)
7060 put_page(kvm->arch.ept_identity_pagetable);
7061 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7064 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
7065 struct kvm_memory_slot *dont)
7069 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7070 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7071 kvm_kvfree(free->arch.rmap[i]);
7072 free->arch.rmap[i] = NULL;
7077 if (!dont || free->arch.lpage_info[i - 1] !=
7078 dont->arch.lpage_info[i - 1]) {
7079 kvm_kvfree(free->arch.lpage_info[i - 1]);
7080 free->arch.lpage_info[i - 1] = NULL;
7085 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
7089 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7094 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7095 slot->base_gfn, level) + 1;
7097 slot->arch.rmap[i] =
7098 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7099 if (!slot->arch.rmap[i])
7104 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7105 sizeof(*slot->arch.lpage_info[i - 1]));
7106 if (!slot->arch.lpage_info[i - 1])
7109 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7110 slot->arch.lpage_info[i - 1][0].write_count = 1;
7111 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7112 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7113 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7115 * If the gfn and userspace address are not aligned wrt each
7116 * other, or if explicitly asked to, disable large page
7117 * support for this slot
7119 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7120 !kvm_largepages_enabled()) {
7123 for (j = 0; j < lpages; ++j)
7124 slot->arch.lpage_info[i - 1][j].write_count = 1;
7131 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7132 kvm_kvfree(slot->arch.rmap[i]);
7133 slot->arch.rmap[i] = NULL;
7137 kvm_kvfree(slot->arch.lpage_info[i - 1]);
7138 slot->arch.lpage_info[i - 1] = NULL;
7143 void kvm_arch_memslots_updated(struct kvm *kvm)
7146 * memslots->generation has been incremented.
7147 * mmio generation may have reached its maximum value.
7149 kvm_mmu_invalidate_mmio_sptes(kvm);
7152 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7153 struct kvm_memory_slot *memslot,
7154 struct kvm_userspace_memory_region *mem,
7155 enum kvm_mr_change change)
7158 * Only private memory slots need to be mapped here since
7159 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7161 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7162 unsigned long userspace_addr;
7165 * MAP_SHARED to prevent internal slot pages from being moved
7168 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7169 PROT_READ | PROT_WRITE,
7170 MAP_SHARED | MAP_ANONYMOUS, 0);
7172 if (IS_ERR((void *)userspace_addr))
7173 return PTR_ERR((void *)userspace_addr);
7175 memslot->userspace_addr = userspace_addr;
7181 void kvm_arch_commit_memory_region(struct kvm *kvm,
7182 struct kvm_userspace_memory_region *mem,
7183 const struct kvm_memory_slot *old,
7184 enum kvm_mr_change change)
7187 int nr_mmu_pages = 0;
7189 if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7192 ret = vm_munmap(old->userspace_addr,
7193 old->npages * PAGE_SIZE);
7196 "kvm_vm_ioctl_set_memory_region: "
7197 "failed to munmap memory\n");
7200 if (!kvm->arch.n_requested_mmu_pages)
7201 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7204 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7206 * Write protect all pages for dirty logging.
7207 * Existing largepage mappings are destroyed here and new ones will
7208 * not be created until the end of the logging.
7210 if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
7211 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7214 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7216 kvm_mmu_invalidate_zap_all_pages(kvm);
7219 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7220 struct kvm_memory_slot *slot)
7222 kvm_mmu_invalidate_zap_all_pages(kvm);
7225 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7227 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7228 !vcpu->arch.apf.halted)
7229 || !list_empty_careful(&vcpu->async_pf.done)
7230 || kvm_apic_has_events(vcpu)
7231 || vcpu->arch.pv.pv_unhalted
7232 || atomic_read(&vcpu->arch.nmi_queued) ||
7233 (kvm_arch_interrupt_allowed(vcpu) &&
7234 kvm_cpu_has_interrupt(vcpu));
7237 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7239 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7242 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7244 return kvm_x86_ops->interrupt_allowed(vcpu);
7247 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7249 unsigned long current_rip = kvm_rip_read(vcpu) +
7250 get_segment_base(vcpu, VCPU_SREG_CS);
7252 return current_rip == linear_rip;
7254 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7256 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7258 unsigned long rflags;
7260 rflags = kvm_x86_ops->get_rflags(vcpu);
7261 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7262 rflags &= ~X86_EFLAGS_TF;
7265 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7267 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7269 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7270 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7271 rflags |= X86_EFLAGS_TF;
7272 kvm_x86_ops->set_rflags(vcpu, rflags);
7273 kvm_make_request(KVM_REQ_EVENT, vcpu);
7275 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7277 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7281 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7282 is_error_page(work->page))
7285 r = kvm_mmu_reload(vcpu);
7289 if (!vcpu->arch.mmu.direct_map &&
7290 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7293 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7296 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7298 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7301 static inline u32 kvm_async_pf_next_probe(u32 key)
7303 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7306 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7308 u32 key = kvm_async_pf_hash_fn(gfn);
7310 while (vcpu->arch.apf.gfns[key] != ~0)
7311 key = kvm_async_pf_next_probe(key);
7313 vcpu->arch.apf.gfns[key] = gfn;
7316 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7319 u32 key = kvm_async_pf_hash_fn(gfn);
7321 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7322 (vcpu->arch.apf.gfns[key] != gfn &&
7323 vcpu->arch.apf.gfns[key] != ~0); i++)
7324 key = kvm_async_pf_next_probe(key);
7329 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7331 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7334 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7338 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7340 vcpu->arch.apf.gfns[i] = ~0;
7342 j = kvm_async_pf_next_probe(j);
7343 if (vcpu->arch.apf.gfns[j] == ~0)
7345 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7347 * k lies cyclically in ]i,j]
7349 * |....j i.k.| or |.k..j i...|
7351 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7352 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7357 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7360 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7364 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7365 struct kvm_async_pf *work)
7367 struct x86_exception fault;
7369 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7370 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7372 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7373 (vcpu->arch.apf.send_user_only &&
7374 kvm_x86_ops->get_cpl(vcpu) == 0))
7375 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7376 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7377 fault.vector = PF_VECTOR;
7378 fault.error_code_valid = true;
7379 fault.error_code = 0;
7380 fault.nested_page_fault = false;
7381 fault.address = work->arch.token;
7382 kvm_inject_page_fault(vcpu, &fault);
7386 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7387 struct kvm_async_pf *work)
7389 struct x86_exception fault;
7391 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7392 if (is_error_page(work->page))
7393 work->arch.token = ~0; /* broadcast wakeup */
7395 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7397 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7398 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7399 fault.vector = PF_VECTOR;
7400 fault.error_code_valid = true;
7401 fault.error_code = 0;
7402 fault.nested_page_fault = false;
7403 fault.address = work->arch.token;
7404 kvm_inject_page_fault(vcpu, &fault);
7406 vcpu->arch.apf.halted = false;
7407 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7410 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7412 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7415 return !kvm_event_needs_reinjection(vcpu) &&
7416 kvm_x86_ops->interrupt_allowed(vcpu);
7419 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7420 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7421 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7422 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7423 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7424 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7425 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7426 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7427 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7428 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7429 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7430 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7431 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);