2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
30 #include "assigned-dev.h"
32 #include <linux/clocksource.h>
33 #include <linux/interrupt.h>
34 #include <linux/kvm.h>
36 #include <linux/vmalloc.h>
37 #include <linux/module.h>
38 #include <linux/mman.h>
39 #include <linux/highmem.h>
40 #include <linux/iommu.h>
41 #include <linux/intel-iommu.h>
42 #include <linux/cpufreq.h>
43 #include <linux/user-return-notifier.h>
44 #include <linux/srcu.h>
45 #include <linux/slab.h>
46 #include <linux/perf_event.h>
47 #include <linux/uaccess.h>
48 #include <linux/hash.h>
49 #include <linux/pci.h>
50 #include <linux/timekeeper_internal.h>
51 #include <linux/pvclock_gtod.h>
52 #include <trace/events/kvm.h>
54 #define CREATE_TRACE_POINTS
57 #include <asm/debugreg.h>
63 #include <asm/fpu-internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
68 #define MAX_IO_MSRS 256
69 #define KVM_MAX_MCE_BANKS 32
70 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
72 #define emul_to_vcpu(ctxt) \
73 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
76 * - enable syscall per default because its emulated by KVM
77 * - enable LME and LMA per default on 64 bit KVM
81 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
83 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
86 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
87 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
89 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
90 static void process_nmi(struct kvm_vcpu *vcpu);
91 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
93 struct kvm_x86_ops *kvm_x86_ops;
94 EXPORT_SYMBOL_GPL(kvm_x86_ops);
96 static bool ignore_msrs = 0;
97 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
99 unsigned int min_timer_period_us = 500;
100 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
102 bool kvm_has_tsc_control;
103 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
104 u32 kvm_max_guest_tsc_khz;
105 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
107 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
108 static u32 tsc_tolerance_ppm = 250;
109 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
111 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
112 unsigned int lapic_timer_advance_ns = 0;
113 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
115 static bool backwards_tsc_observed = false;
117 #define KVM_NR_SHARED_MSRS 16
119 struct kvm_shared_msrs_global {
121 u32 msrs[KVM_NR_SHARED_MSRS];
124 struct kvm_shared_msrs {
125 struct user_return_notifier urn;
127 struct kvm_shared_msr_values {
130 } values[KVM_NR_SHARED_MSRS];
133 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
134 static struct kvm_shared_msrs __percpu *shared_msrs;
136 struct kvm_stats_debugfs_item debugfs_entries[] = {
137 { "pf_fixed", VCPU_STAT(pf_fixed) },
138 { "pf_guest", VCPU_STAT(pf_guest) },
139 { "tlb_flush", VCPU_STAT(tlb_flush) },
140 { "invlpg", VCPU_STAT(invlpg) },
141 { "exits", VCPU_STAT(exits) },
142 { "io_exits", VCPU_STAT(io_exits) },
143 { "mmio_exits", VCPU_STAT(mmio_exits) },
144 { "signal_exits", VCPU_STAT(signal_exits) },
145 { "irq_window", VCPU_STAT(irq_window_exits) },
146 { "nmi_window", VCPU_STAT(nmi_window_exits) },
147 { "halt_exits", VCPU_STAT(halt_exits) },
148 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
149 { "hypercalls", VCPU_STAT(hypercalls) },
150 { "request_irq", VCPU_STAT(request_irq_exits) },
151 { "irq_exits", VCPU_STAT(irq_exits) },
152 { "host_state_reload", VCPU_STAT(host_state_reload) },
153 { "efer_reload", VCPU_STAT(efer_reload) },
154 { "fpu_reload", VCPU_STAT(fpu_reload) },
155 { "insn_emulation", VCPU_STAT(insn_emulation) },
156 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
157 { "irq_injections", VCPU_STAT(irq_injections) },
158 { "nmi_injections", VCPU_STAT(nmi_injections) },
159 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
160 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
161 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
162 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
163 { "mmu_flooded", VM_STAT(mmu_flooded) },
164 { "mmu_recycled", VM_STAT(mmu_recycled) },
165 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
166 { "mmu_unsync", VM_STAT(mmu_unsync) },
167 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
168 { "largepages", VM_STAT(lpages) },
172 u64 __read_mostly host_xcr0;
174 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
176 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
179 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
180 vcpu->arch.apf.gfns[i] = ~0;
183 static void kvm_on_user_return(struct user_return_notifier *urn)
186 struct kvm_shared_msrs *locals
187 = container_of(urn, struct kvm_shared_msrs, urn);
188 struct kvm_shared_msr_values *values;
190 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
191 values = &locals->values[slot];
192 if (values->host != values->curr) {
193 wrmsrl(shared_msrs_global.msrs[slot], values->host);
194 values->curr = values->host;
197 locals->registered = false;
198 user_return_notifier_unregister(urn);
201 static void shared_msr_update(unsigned slot, u32 msr)
204 unsigned int cpu = smp_processor_id();
205 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
207 /* only read, and nobody should modify it at this time,
208 * so don't need lock */
209 if (slot >= shared_msrs_global.nr) {
210 printk(KERN_ERR "kvm: invalid MSR slot!");
213 rdmsrl_safe(msr, &value);
214 smsr->values[slot].host = value;
215 smsr->values[slot].curr = value;
218 void kvm_define_shared_msr(unsigned slot, u32 msr)
220 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
221 if (slot >= shared_msrs_global.nr)
222 shared_msrs_global.nr = slot + 1;
223 shared_msrs_global.msrs[slot] = msr;
224 /* we need ensured the shared_msr_global have been updated */
227 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
229 static void kvm_shared_msr_cpu_online(void)
233 for (i = 0; i < shared_msrs_global.nr; ++i)
234 shared_msr_update(i, shared_msrs_global.msrs[i]);
237 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
239 unsigned int cpu = smp_processor_id();
240 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
243 if (((value ^ smsr->values[slot].curr) & mask) == 0)
245 smsr->values[slot].curr = value;
246 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
250 if (!smsr->registered) {
251 smsr->urn.on_user_return = kvm_on_user_return;
252 user_return_notifier_register(&smsr->urn);
253 smsr->registered = true;
257 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
259 static void drop_user_return_notifiers(void)
261 unsigned int cpu = smp_processor_id();
262 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
264 if (smsr->registered)
265 kvm_on_user_return(&smsr->urn);
268 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
270 return vcpu->arch.apic_base;
272 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
274 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
276 u64 old_state = vcpu->arch.apic_base &
277 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
278 u64 new_state = msr_info->data &
279 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
280 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
281 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
283 if (!msr_info->host_initiated &&
284 ((msr_info->data & reserved_bits) != 0 ||
285 new_state == X2APIC_ENABLE ||
286 (new_state == MSR_IA32_APICBASE_ENABLE &&
287 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
288 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
292 kvm_lapic_set_base(vcpu, msr_info->data);
295 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
297 asmlinkage __visible void kvm_spurious_fault(void)
299 /* Fault while not rebooting. We want the trace. */
302 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
304 #define EXCPT_BENIGN 0
305 #define EXCPT_CONTRIBUTORY 1
308 static int exception_class(int vector)
318 return EXCPT_CONTRIBUTORY;
325 #define EXCPT_FAULT 0
327 #define EXCPT_ABORT 2
328 #define EXCPT_INTERRUPT 3
330 static int exception_type(int vector)
334 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
335 return EXCPT_INTERRUPT;
339 /* #DB is trap, as instruction watchpoints are handled elsewhere */
340 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
343 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
346 /* Reserved exceptions will result in fault */
350 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
351 unsigned nr, bool has_error, u32 error_code,
357 kvm_make_request(KVM_REQ_EVENT, vcpu);
359 if (!vcpu->arch.exception.pending) {
361 if (has_error && !is_protmode(vcpu))
363 vcpu->arch.exception.pending = true;
364 vcpu->arch.exception.has_error_code = has_error;
365 vcpu->arch.exception.nr = nr;
366 vcpu->arch.exception.error_code = error_code;
367 vcpu->arch.exception.reinject = reinject;
371 /* to check exception */
372 prev_nr = vcpu->arch.exception.nr;
373 if (prev_nr == DF_VECTOR) {
374 /* triple fault -> shutdown */
375 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
378 class1 = exception_class(prev_nr);
379 class2 = exception_class(nr);
380 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
381 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
382 /* generate double fault per SDM Table 5-5 */
383 vcpu->arch.exception.pending = true;
384 vcpu->arch.exception.has_error_code = true;
385 vcpu->arch.exception.nr = DF_VECTOR;
386 vcpu->arch.exception.error_code = 0;
388 /* replace previous exception with a new one in a hope
389 that instruction re-execution will regenerate lost
394 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
396 kvm_multiple_exception(vcpu, nr, false, 0, false);
398 EXPORT_SYMBOL_GPL(kvm_queue_exception);
400 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
402 kvm_multiple_exception(vcpu, nr, false, 0, true);
404 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
406 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
409 kvm_inject_gp(vcpu, 0);
411 kvm_x86_ops->skip_emulated_instruction(vcpu);
413 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
415 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
417 ++vcpu->stat.pf_guest;
418 vcpu->arch.cr2 = fault->address;
419 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
421 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
423 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
425 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
426 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
428 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
430 return fault->nested_page_fault;
433 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
435 atomic_inc(&vcpu->arch.nmi_queued);
436 kvm_make_request(KVM_REQ_NMI, vcpu);
438 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
440 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
442 kvm_multiple_exception(vcpu, nr, true, error_code, false);
444 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
446 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
448 kvm_multiple_exception(vcpu, nr, true, error_code, true);
450 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
453 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
454 * a #GP and return false.
456 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
458 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
460 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
463 EXPORT_SYMBOL_GPL(kvm_require_cpl);
465 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
467 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
470 kvm_queue_exception(vcpu, UD_VECTOR);
473 EXPORT_SYMBOL_GPL(kvm_require_dr);
476 * This function will be used to read from the physical memory of the currently
477 * running guest. The difference to kvm_read_guest_page is that this function
478 * can read from guest physical or from the guest's guest physical memory.
480 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
481 gfn_t ngfn, void *data, int offset, int len,
484 struct x86_exception exception;
488 ngpa = gfn_to_gpa(ngfn);
489 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
490 if (real_gfn == UNMAPPED_GVA)
493 real_gfn = gpa_to_gfn(real_gfn);
495 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
497 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
499 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
500 void *data, int offset, int len, u32 access)
502 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
503 data, offset, len, access);
507 * Load the pae pdptrs. Return true is they are all valid.
509 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
511 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
512 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
515 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
517 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
518 offset * sizeof(u64), sizeof(pdpte),
519 PFERR_USER_MASK|PFERR_WRITE_MASK);
524 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
525 if (is_present_gpte(pdpte[i]) &&
526 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
533 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
534 __set_bit(VCPU_EXREG_PDPTR,
535 (unsigned long *)&vcpu->arch.regs_avail);
536 __set_bit(VCPU_EXREG_PDPTR,
537 (unsigned long *)&vcpu->arch.regs_dirty);
542 EXPORT_SYMBOL_GPL(load_pdptrs);
544 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
546 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
552 if (is_long_mode(vcpu) || !is_pae(vcpu))
555 if (!test_bit(VCPU_EXREG_PDPTR,
556 (unsigned long *)&vcpu->arch.regs_avail))
559 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
560 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
561 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
562 PFERR_USER_MASK | PFERR_WRITE_MASK);
565 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
571 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
573 unsigned long old_cr0 = kvm_read_cr0(vcpu);
574 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
575 X86_CR0_CD | X86_CR0_NW;
580 if (cr0 & 0xffffffff00000000UL)
584 cr0 &= ~CR0_RESERVED_BITS;
586 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
589 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
592 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
594 if ((vcpu->arch.efer & EFER_LME)) {
599 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
604 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
609 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
612 kvm_x86_ops->set_cr0(vcpu, cr0);
614 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
615 kvm_clear_async_pf_completion_queue(vcpu);
616 kvm_async_pf_hash_reset(vcpu);
619 if ((cr0 ^ old_cr0) & update_bits)
620 kvm_mmu_reset_context(vcpu);
623 EXPORT_SYMBOL_GPL(kvm_set_cr0);
625 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
627 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
629 EXPORT_SYMBOL_GPL(kvm_lmsw);
631 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
633 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
634 !vcpu->guest_xcr0_loaded) {
635 /* kvm_set_xcr() also depends on this */
636 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
637 vcpu->guest_xcr0_loaded = 1;
641 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
643 if (vcpu->guest_xcr0_loaded) {
644 if (vcpu->arch.xcr0 != host_xcr0)
645 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
646 vcpu->guest_xcr0_loaded = 0;
650 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
653 u64 old_xcr0 = vcpu->arch.xcr0;
656 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
657 if (index != XCR_XFEATURE_ENABLED_MASK)
659 if (!(xcr0 & XSTATE_FP))
661 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
665 * Do not allow the guest to set bits that we do not support
666 * saving. However, xcr0 bit 0 is always set, even if the
667 * emulated CPU does not support XSAVE (see fx_init).
669 valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
670 if (xcr0 & ~valid_bits)
673 if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
676 if (xcr0 & XSTATE_AVX512) {
677 if (!(xcr0 & XSTATE_YMM))
679 if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512)
682 kvm_put_guest_xcr0(vcpu);
683 vcpu->arch.xcr0 = xcr0;
685 if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
686 kvm_update_cpuid(vcpu);
690 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
692 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
693 __kvm_set_xcr(vcpu, index, xcr)) {
694 kvm_inject_gp(vcpu, 0);
699 EXPORT_SYMBOL_GPL(kvm_set_xcr);
701 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
703 unsigned long old_cr4 = kvm_read_cr4(vcpu);
704 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
705 X86_CR4_PAE | X86_CR4_SMEP;
706 if (cr4 & CR4_RESERVED_BITS)
709 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
712 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
715 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
718 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
721 if (is_long_mode(vcpu)) {
722 if (!(cr4 & X86_CR4_PAE))
724 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
725 && ((cr4 ^ old_cr4) & pdptr_bits)
726 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
730 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
731 if (!guest_cpuid_has_pcid(vcpu))
734 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
735 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
739 if (kvm_x86_ops->set_cr4(vcpu, cr4))
742 if (((cr4 ^ old_cr4) & pdptr_bits) ||
743 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
744 kvm_mmu_reset_context(vcpu);
746 if ((cr4 ^ old_cr4) & X86_CR4_SMAP)
747 update_permission_bitmask(vcpu, vcpu->arch.walk_mmu, false);
749 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
750 kvm_update_cpuid(vcpu);
754 EXPORT_SYMBOL_GPL(kvm_set_cr4);
756 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
759 cr3 &= ~CR3_PCID_INVD;
762 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
763 kvm_mmu_sync_roots(vcpu);
764 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
768 if (is_long_mode(vcpu)) {
769 if (cr3 & CR3_L_MODE_RESERVED_BITS)
771 } else if (is_pae(vcpu) && is_paging(vcpu) &&
772 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
775 vcpu->arch.cr3 = cr3;
776 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
777 kvm_mmu_new_cr3(vcpu);
780 EXPORT_SYMBOL_GPL(kvm_set_cr3);
782 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
784 if (cr8 & CR8_RESERVED_BITS)
786 if (irqchip_in_kernel(vcpu->kvm))
787 kvm_lapic_set_tpr(vcpu, cr8);
789 vcpu->arch.cr8 = cr8;
792 EXPORT_SYMBOL_GPL(kvm_set_cr8);
794 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
796 if (irqchip_in_kernel(vcpu->kvm))
797 return kvm_lapic_get_cr8(vcpu);
799 return vcpu->arch.cr8;
801 EXPORT_SYMBOL_GPL(kvm_get_cr8);
803 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
805 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
806 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
809 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
813 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
814 dr7 = vcpu->arch.guest_debug_dr7;
816 dr7 = vcpu->arch.dr7;
817 kvm_x86_ops->set_dr7(vcpu, dr7);
818 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
819 if (dr7 & DR7_BP_EN_MASK)
820 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
823 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
825 u64 fixed = DR6_FIXED_1;
827 if (!guest_cpuid_has_rtm(vcpu))
832 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
836 vcpu->arch.db[dr] = val;
837 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
838 vcpu->arch.eff_db[dr] = val;
843 if (val & 0xffffffff00000000ULL)
845 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
846 kvm_update_dr6(vcpu);
851 if (val & 0xffffffff00000000ULL)
853 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
854 kvm_update_dr7(vcpu);
861 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
863 if (__kvm_set_dr(vcpu, dr, val)) {
864 kvm_inject_gp(vcpu, 0);
869 EXPORT_SYMBOL_GPL(kvm_set_dr);
871 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
875 *val = vcpu->arch.db[dr];
880 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
881 *val = vcpu->arch.dr6;
883 *val = kvm_x86_ops->get_dr6(vcpu);
888 *val = vcpu->arch.dr7;
893 EXPORT_SYMBOL_GPL(kvm_get_dr);
895 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
897 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
901 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
904 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
905 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
908 EXPORT_SYMBOL_GPL(kvm_rdpmc);
911 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
912 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
914 * This list is modified at module load time to reflect the
915 * capabilities of the host cpu. This capabilities test skips MSRs that are
916 * kvm-specific. Those are put in the beginning of the list.
919 #define KVM_SAVE_MSRS_BEGIN 12
920 static u32 msrs_to_save[] = {
921 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
922 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
923 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
924 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
925 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
927 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
930 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
932 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
933 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
936 static unsigned num_msrs_to_save;
938 static const u32 emulated_msrs[] = {
940 MSR_IA32_TSCDEADLINE,
941 MSR_IA32_MISC_ENABLE,
946 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
948 if (efer & efer_reserved_bits)
951 if (efer & EFER_FFXSR) {
952 struct kvm_cpuid_entry2 *feat;
954 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
955 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
959 if (efer & EFER_SVME) {
960 struct kvm_cpuid_entry2 *feat;
962 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
963 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
969 EXPORT_SYMBOL_GPL(kvm_valid_efer);
971 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
973 u64 old_efer = vcpu->arch.efer;
975 if (!kvm_valid_efer(vcpu, efer))
979 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
983 efer |= vcpu->arch.efer & EFER_LMA;
985 kvm_x86_ops->set_efer(vcpu, efer);
987 /* Update reserved bits */
988 if ((efer ^ old_efer) & EFER_NX)
989 kvm_mmu_reset_context(vcpu);
994 void kvm_enable_efer_bits(u64 mask)
996 efer_reserved_bits &= ~mask;
998 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1001 * Writes msr value into into the appropriate "register".
1002 * Returns 0 on success, non-0 otherwise.
1003 * Assumes vcpu_load() was already called.
1005 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1007 switch (msr->index) {
1010 case MSR_KERNEL_GS_BASE:
1013 if (is_noncanonical_address(msr->data))
1016 case MSR_IA32_SYSENTER_EIP:
1017 case MSR_IA32_SYSENTER_ESP:
1019 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1020 * non-canonical address is written on Intel but not on
1021 * AMD (which ignores the top 32-bits, because it does
1022 * not implement 64-bit SYSENTER).
1024 * 64-bit code should hence be able to write a non-canonical
1025 * value on AMD. Making the address canonical ensures that
1026 * vmentry does not fail on Intel after writing a non-canonical
1027 * value, and that something deterministic happens if the guest
1028 * invokes 64-bit SYSENTER.
1030 msr->data = get_canonical(msr->data);
1032 return kvm_x86_ops->set_msr(vcpu, msr);
1034 EXPORT_SYMBOL_GPL(kvm_set_msr);
1037 * Adapt set_msr() to msr_io()'s calling convention
1039 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1041 struct msr_data msr;
1045 msr.host_initiated = true;
1046 return kvm_set_msr(vcpu, &msr);
1049 #ifdef CONFIG_X86_64
1050 struct pvclock_gtod_data {
1053 struct { /* extract of a clocksource struct */
1065 static struct pvclock_gtod_data pvclock_gtod_data;
1067 static void update_pvclock_gtod(struct timekeeper *tk)
1069 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1072 boot_ns = ktime_to_ns(ktime_add(tk->tkr.base_mono, tk->offs_boot));
1074 write_seqcount_begin(&vdata->seq);
1076 /* copy pvclock gtod data */
1077 vdata->clock.vclock_mode = tk->tkr.clock->archdata.vclock_mode;
1078 vdata->clock.cycle_last = tk->tkr.cycle_last;
1079 vdata->clock.mask = tk->tkr.mask;
1080 vdata->clock.mult = tk->tkr.mult;
1081 vdata->clock.shift = tk->tkr.shift;
1083 vdata->boot_ns = boot_ns;
1084 vdata->nsec_base = tk->tkr.xtime_nsec;
1086 write_seqcount_end(&vdata->seq);
1090 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1093 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1094 * vcpu_enter_guest. This function is only called from
1095 * the physical CPU that is running vcpu.
1097 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1100 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1104 struct pvclock_wall_clock wc;
1105 struct timespec boot;
1110 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1115 ++version; /* first time write, random junk */
1119 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1122 * The guest calculates current wall clock time by adding
1123 * system time (updated by kvm_guest_time_update below) to the
1124 * wall clock specified here. guest system time equals host
1125 * system time for us, thus we must fill in host boot time here.
1129 if (kvm->arch.kvmclock_offset) {
1130 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1131 boot = timespec_sub(boot, ts);
1133 wc.sec = boot.tv_sec;
1134 wc.nsec = boot.tv_nsec;
1135 wc.version = version;
1137 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1140 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1143 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1145 uint32_t quotient, remainder;
1147 /* Don't try to replace with do_div(), this one calculates
1148 * "(dividend << 32) / divisor" */
1150 : "=a" (quotient), "=d" (remainder)
1151 : "0" (0), "1" (dividend), "r" (divisor) );
1155 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1156 s8 *pshift, u32 *pmultiplier)
1163 tps64 = base_khz * 1000LL;
1164 scaled64 = scaled_khz * 1000LL;
1165 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1170 tps32 = (uint32_t)tps64;
1171 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1172 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1180 *pmultiplier = div_frac(scaled64, tps32);
1182 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1183 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1186 static inline u64 get_kernel_ns(void)
1188 return ktime_get_boot_ns();
1191 #ifdef CONFIG_X86_64
1192 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1195 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1196 unsigned long max_tsc_khz;
1198 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1200 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1201 vcpu->arch.virtual_tsc_shift);
1204 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1206 u64 v = (u64)khz * (1000000 + ppm);
1211 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1213 u32 thresh_lo, thresh_hi;
1214 int use_scaling = 0;
1216 /* tsc_khz can be zero if TSC calibration fails */
1217 if (this_tsc_khz == 0)
1220 /* Compute a scale to convert nanoseconds in TSC cycles */
1221 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1222 &vcpu->arch.virtual_tsc_shift,
1223 &vcpu->arch.virtual_tsc_mult);
1224 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1227 * Compute the variation in TSC rate which is acceptable
1228 * within the range of tolerance and decide if the
1229 * rate being applied is within that bounds of the hardware
1230 * rate. If so, no scaling or compensation need be done.
1232 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1233 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1234 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1235 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1238 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1241 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1243 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1244 vcpu->arch.virtual_tsc_mult,
1245 vcpu->arch.virtual_tsc_shift);
1246 tsc += vcpu->arch.this_tsc_write;
1250 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1252 #ifdef CONFIG_X86_64
1254 struct kvm_arch *ka = &vcpu->kvm->arch;
1255 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1257 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1258 atomic_read(&vcpu->kvm->online_vcpus));
1261 * Once the masterclock is enabled, always perform request in
1262 * order to update it.
1264 * In order to enable masterclock, the host clocksource must be TSC
1265 * and the vcpus need to have matched TSCs. When that happens,
1266 * perform request to enable masterclock.
1268 if (ka->use_master_clock ||
1269 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1270 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1272 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1273 atomic_read(&vcpu->kvm->online_vcpus),
1274 ka->use_master_clock, gtod->clock.vclock_mode);
1278 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1280 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1281 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1284 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1286 struct kvm *kvm = vcpu->kvm;
1287 u64 offset, ns, elapsed;
1288 unsigned long flags;
1291 bool already_matched;
1292 u64 data = msr->data;
1294 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1295 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1296 ns = get_kernel_ns();
1297 elapsed = ns - kvm->arch.last_tsc_nsec;
1299 if (vcpu->arch.virtual_tsc_khz) {
1302 /* n.b - signed multiplication and division required */
1303 usdiff = data - kvm->arch.last_tsc_write;
1304 #ifdef CONFIG_X86_64
1305 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1307 /* do_div() only does unsigned */
1308 asm("1: idivl %[divisor]\n"
1309 "2: xor %%edx, %%edx\n"
1310 " movl $0, %[faulted]\n"
1312 ".section .fixup,\"ax\"\n"
1313 "4: movl $1, %[faulted]\n"
1317 _ASM_EXTABLE(1b, 4b)
1319 : "=A"(usdiff), [faulted] "=r" (faulted)
1320 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1323 do_div(elapsed, 1000);
1328 /* idivl overflow => difference is larger than USEC_PER_SEC */
1330 usdiff = USEC_PER_SEC;
1332 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1335 * Special case: TSC write with a small delta (1 second) of virtual
1336 * cycle time against real time is interpreted as an attempt to
1337 * synchronize the CPU.
1339 * For a reliable TSC, we can match TSC offsets, and for an unstable
1340 * TSC, we add elapsed time in this computation. We could let the
1341 * compensation code attempt to catch up if we fall behind, but
1342 * it's better to try to match offsets from the beginning.
1344 if (usdiff < USEC_PER_SEC &&
1345 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1346 if (!check_tsc_unstable()) {
1347 offset = kvm->arch.cur_tsc_offset;
1348 pr_debug("kvm: matched tsc offset for %llu\n", data);
1350 u64 delta = nsec_to_cycles(vcpu, elapsed);
1352 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1353 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1356 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1359 * We split periods of matched TSC writes into generations.
1360 * For each generation, we track the original measured
1361 * nanosecond time, offset, and write, so if TSCs are in
1362 * sync, we can match exact offset, and if not, we can match
1363 * exact software computation in compute_guest_tsc()
1365 * These values are tracked in kvm->arch.cur_xxx variables.
1367 kvm->arch.cur_tsc_generation++;
1368 kvm->arch.cur_tsc_nsec = ns;
1369 kvm->arch.cur_tsc_write = data;
1370 kvm->arch.cur_tsc_offset = offset;
1372 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1373 kvm->arch.cur_tsc_generation, data);
1377 * We also track th most recent recorded KHZ, write and time to
1378 * allow the matching interval to be extended at each write.
1380 kvm->arch.last_tsc_nsec = ns;
1381 kvm->arch.last_tsc_write = data;
1382 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1384 vcpu->arch.last_guest_tsc = data;
1386 /* Keep track of which generation this VCPU has synchronized to */
1387 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1388 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1389 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1391 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1392 update_ia32_tsc_adjust_msr(vcpu, offset);
1393 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1394 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1396 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1398 kvm->arch.nr_vcpus_matched_tsc = 0;
1399 } else if (!already_matched) {
1400 kvm->arch.nr_vcpus_matched_tsc++;
1403 kvm_track_tsc_matching(vcpu);
1404 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1407 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1409 #ifdef CONFIG_X86_64
1411 static cycle_t read_tsc(void)
1417 * Empirically, a fence (of type that depends on the CPU)
1418 * before rdtsc is enough to ensure that rdtsc is ordered
1419 * with respect to loads. The various CPU manuals are unclear
1420 * as to whether rdtsc can be reordered with later loads,
1421 * but no one has ever seen it happen.
1424 ret = (cycle_t)vget_cycles();
1426 last = pvclock_gtod_data.clock.cycle_last;
1428 if (likely(ret >= last))
1432 * GCC likes to generate cmov here, but this branch is extremely
1433 * predictable (it's just a funciton of time and the likely is
1434 * very likely) and there's a data dependence, so force GCC
1435 * to generate a branch instead. I don't barrier() because
1436 * we don't actually need a barrier, and if this function
1437 * ever gets inlined it will generate worse code.
1443 static inline u64 vgettsc(cycle_t *cycle_now)
1446 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1448 *cycle_now = read_tsc();
1450 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1451 return v * gtod->clock.mult;
1454 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1456 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1462 seq = read_seqcount_begin(>od->seq);
1463 mode = gtod->clock.vclock_mode;
1464 ns = gtod->nsec_base;
1465 ns += vgettsc(cycle_now);
1466 ns >>= gtod->clock.shift;
1467 ns += gtod->boot_ns;
1468 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1474 /* returns true if host is using tsc clocksource */
1475 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1477 /* checked again under seqlock below */
1478 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1481 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1487 * Assuming a stable TSC across physical CPUS, and a stable TSC
1488 * across virtual CPUs, the following condition is possible.
1489 * Each numbered line represents an event visible to both
1490 * CPUs at the next numbered event.
1492 * "timespecX" represents host monotonic time. "tscX" represents
1495 * VCPU0 on CPU0 | VCPU1 on CPU1
1497 * 1. read timespec0,tsc0
1498 * 2. | timespec1 = timespec0 + N
1500 * 3. transition to guest | transition to guest
1501 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1502 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1503 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1505 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1508 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1510 * - 0 < N - M => M < N
1512 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1513 * always the case (the difference between two distinct xtime instances
1514 * might be smaller then the difference between corresponding TSC reads,
1515 * when updating guest vcpus pvclock areas).
1517 * To avoid that problem, do not allow visibility of distinct
1518 * system_timestamp/tsc_timestamp values simultaneously: use a master
1519 * copy of host monotonic time values. Update that master copy
1522 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1526 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1528 #ifdef CONFIG_X86_64
1529 struct kvm_arch *ka = &kvm->arch;
1531 bool host_tsc_clocksource, vcpus_matched;
1533 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1534 atomic_read(&kvm->online_vcpus));
1537 * If the host uses TSC clock, then passthrough TSC as stable
1540 host_tsc_clocksource = kvm_get_time_and_clockread(
1541 &ka->master_kernel_ns,
1542 &ka->master_cycle_now);
1544 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1545 && !backwards_tsc_observed;
1547 if (ka->use_master_clock)
1548 atomic_set(&kvm_guest_has_master_clock, 1);
1550 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1551 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1556 static void kvm_gen_update_masterclock(struct kvm *kvm)
1558 #ifdef CONFIG_X86_64
1560 struct kvm_vcpu *vcpu;
1561 struct kvm_arch *ka = &kvm->arch;
1563 spin_lock(&ka->pvclock_gtod_sync_lock);
1564 kvm_make_mclock_inprogress_request(kvm);
1565 /* no guest entries from this point */
1566 pvclock_update_vm_gtod_copy(kvm);
1568 kvm_for_each_vcpu(i, vcpu, kvm)
1569 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1571 /* guest entries allowed */
1572 kvm_for_each_vcpu(i, vcpu, kvm)
1573 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1575 spin_unlock(&ka->pvclock_gtod_sync_lock);
1579 static int kvm_guest_time_update(struct kvm_vcpu *v)
1581 unsigned long flags, this_tsc_khz;
1582 struct kvm_vcpu_arch *vcpu = &v->arch;
1583 struct kvm_arch *ka = &v->kvm->arch;
1585 u64 tsc_timestamp, host_tsc;
1586 struct pvclock_vcpu_time_info guest_hv_clock;
1588 bool use_master_clock;
1594 * If the host uses TSC clock, then passthrough TSC as stable
1597 spin_lock(&ka->pvclock_gtod_sync_lock);
1598 use_master_clock = ka->use_master_clock;
1599 if (use_master_clock) {
1600 host_tsc = ka->master_cycle_now;
1601 kernel_ns = ka->master_kernel_ns;
1603 spin_unlock(&ka->pvclock_gtod_sync_lock);
1605 /* Keep irq disabled to prevent changes to the clock */
1606 local_irq_save(flags);
1607 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1608 if (unlikely(this_tsc_khz == 0)) {
1609 local_irq_restore(flags);
1610 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1613 if (!use_master_clock) {
1614 host_tsc = native_read_tsc();
1615 kernel_ns = get_kernel_ns();
1618 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1621 * We may have to catch up the TSC to match elapsed wall clock
1622 * time for two reasons, even if kvmclock is used.
1623 * 1) CPU could have been running below the maximum TSC rate
1624 * 2) Broken TSC compensation resets the base at each VCPU
1625 * entry to avoid unknown leaps of TSC even when running
1626 * again on the same CPU. This may cause apparent elapsed
1627 * time to disappear, and the guest to stand still or run
1630 if (vcpu->tsc_catchup) {
1631 u64 tsc = compute_guest_tsc(v, kernel_ns);
1632 if (tsc > tsc_timestamp) {
1633 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1634 tsc_timestamp = tsc;
1638 local_irq_restore(flags);
1640 if (!vcpu->pv_time_enabled)
1643 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1644 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1645 &vcpu->hv_clock.tsc_shift,
1646 &vcpu->hv_clock.tsc_to_system_mul);
1647 vcpu->hw_tsc_khz = this_tsc_khz;
1650 /* With all the info we got, fill in the values */
1651 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1652 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1653 vcpu->last_guest_tsc = tsc_timestamp;
1655 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1656 &guest_hv_clock, sizeof(guest_hv_clock))))
1660 * The interface expects us to write an even number signaling that the
1661 * update is finished. Since the guest won't see the intermediate
1662 * state, we just increase by 2 at the end.
1664 vcpu->hv_clock.version = guest_hv_clock.version + 2;
1666 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1667 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1669 if (vcpu->pvclock_set_guest_stopped_request) {
1670 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1671 vcpu->pvclock_set_guest_stopped_request = false;
1674 /* If the host uses TSC clocksource, then it is stable */
1675 if (use_master_clock)
1676 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1678 vcpu->hv_clock.flags = pvclock_flags;
1680 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1682 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1684 sizeof(vcpu->hv_clock));
1689 * kvmclock updates which are isolated to a given vcpu, such as
1690 * vcpu->cpu migration, should not allow system_timestamp from
1691 * the rest of the vcpus to remain static. Otherwise ntp frequency
1692 * correction applies to one vcpu's system_timestamp but not
1695 * So in those cases, request a kvmclock update for all vcpus.
1696 * We need to rate-limit these requests though, as they can
1697 * considerably slow guests that have a large number of vcpus.
1698 * The time for a remote vcpu to update its kvmclock is bound
1699 * by the delay we use to rate-limit the updates.
1702 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1704 static void kvmclock_update_fn(struct work_struct *work)
1707 struct delayed_work *dwork = to_delayed_work(work);
1708 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1709 kvmclock_update_work);
1710 struct kvm *kvm = container_of(ka, struct kvm, arch);
1711 struct kvm_vcpu *vcpu;
1713 kvm_for_each_vcpu(i, vcpu, kvm) {
1714 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1715 kvm_vcpu_kick(vcpu);
1719 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1721 struct kvm *kvm = v->kvm;
1723 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1724 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1725 KVMCLOCK_UPDATE_DELAY);
1728 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1730 static void kvmclock_sync_fn(struct work_struct *work)
1732 struct delayed_work *dwork = to_delayed_work(work);
1733 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1734 kvmclock_sync_work);
1735 struct kvm *kvm = container_of(ka, struct kvm, arch);
1737 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1738 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1739 KVMCLOCK_SYNC_PERIOD);
1742 static bool msr_mtrr_valid(unsigned msr)
1745 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1746 case MSR_MTRRfix64K_00000:
1747 case MSR_MTRRfix16K_80000:
1748 case MSR_MTRRfix16K_A0000:
1749 case MSR_MTRRfix4K_C0000:
1750 case MSR_MTRRfix4K_C8000:
1751 case MSR_MTRRfix4K_D0000:
1752 case MSR_MTRRfix4K_D8000:
1753 case MSR_MTRRfix4K_E0000:
1754 case MSR_MTRRfix4K_E8000:
1755 case MSR_MTRRfix4K_F0000:
1756 case MSR_MTRRfix4K_F8000:
1757 case MSR_MTRRdefType:
1758 case MSR_IA32_CR_PAT:
1766 static bool valid_pat_type(unsigned t)
1768 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1771 static bool valid_mtrr_type(unsigned t)
1773 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1776 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1781 if (!msr_mtrr_valid(msr))
1784 if (msr == MSR_IA32_CR_PAT) {
1785 for (i = 0; i < 8; i++)
1786 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1789 } else if (msr == MSR_MTRRdefType) {
1792 return valid_mtrr_type(data & 0xff);
1793 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1794 for (i = 0; i < 8 ; i++)
1795 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1800 /* variable MTRRs */
1801 WARN_ON(!(msr >= 0x200 && msr < 0x200 + 2 * KVM_NR_VAR_MTRR));
1803 mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
1804 if ((msr & 1) == 0) {
1806 if (!valid_mtrr_type(data & 0xff))
1813 kvm_inject_gp(vcpu, 0);
1819 EXPORT_SYMBOL_GPL(kvm_mtrr_valid);
1821 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1823 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1825 if (!kvm_mtrr_valid(vcpu, msr, data))
1828 if (msr == MSR_MTRRdefType) {
1829 vcpu->arch.mtrr_state.def_type = data;
1830 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1831 } else if (msr == MSR_MTRRfix64K_00000)
1833 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1834 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1835 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1836 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1837 else if (msr == MSR_IA32_CR_PAT)
1838 vcpu->arch.pat = data;
1839 else { /* Variable MTRRs */
1840 int idx, is_mtrr_mask;
1843 idx = (msr - 0x200) / 2;
1844 is_mtrr_mask = msr - 0x200 - 2 * idx;
1847 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1850 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1854 kvm_mmu_reset_context(vcpu);
1858 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1860 u64 mcg_cap = vcpu->arch.mcg_cap;
1861 unsigned bank_num = mcg_cap & 0xff;
1864 case MSR_IA32_MCG_STATUS:
1865 vcpu->arch.mcg_status = data;
1867 case MSR_IA32_MCG_CTL:
1868 if (!(mcg_cap & MCG_CTL_P))
1870 if (data != 0 && data != ~(u64)0)
1872 vcpu->arch.mcg_ctl = data;
1875 if (msr >= MSR_IA32_MC0_CTL &&
1876 msr < MSR_IA32_MCx_CTL(bank_num)) {
1877 u32 offset = msr - MSR_IA32_MC0_CTL;
1878 /* only 0 or all 1s can be written to IA32_MCi_CTL
1879 * some Linux kernels though clear bit 10 in bank 4 to
1880 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1881 * this to avoid an uncatched #GP in the guest
1883 if ((offset & 0x3) == 0 &&
1884 data != 0 && (data | (1 << 10)) != ~(u64)0)
1886 vcpu->arch.mce_banks[offset] = data;
1894 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1896 struct kvm *kvm = vcpu->kvm;
1897 int lm = is_long_mode(vcpu);
1898 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1899 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1900 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1901 : kvm->arch.xen_hvm_config.blob_size_32;
1902 u32 page_num = data & ~PAGE_MASK;
1903 u64 page_addr = data & PAGE_MASK;
1908 if (page_num >= blob_size)
1911 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1916 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1925 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1927 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1930 static bool kvm_hv_msr_partition_wide(u32 msr)
1934 case HV_X64_MSR_GUEST_OS_ID:
1935 case HV_X64_MSR_HYPERCALL:
1936 case HV_X64_MSR_REFERENCE_TSC:
1937 case HV_X64_MSR_TIME_REF_COUNT:
1945 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1947 struct kvm *kvm = vcpu->kvm;
1950 case HV_X64_MSR_GUEST_OS_ID:
1951 kvm->arch.hv_guest_os_id = data;
1952 /* setting guest os id to zero disables hypercall page */
1953 if (!kvm->arch.hv_guest_os_id)
1954 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1956 case HV_X64_MSR_HYPERCALL: {
1961 /* if guest os id is not set hypercall should remain disabled */
1962 if (!kvm->arch.hv_guest_os_id)
1964 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1965 kvm->arch.hv_hypercall = data;
1968 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1969 addr = gfn_to_hva(kvm, gfn);
1970 if (kvm_is_error_hva(addr))
1972 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1973 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1974 if (__copy_to_user((void __user *)addr, instructions, 4))
1976 kvm->arch.hv_hypercall = data;
1977 mark_page_dirty(kvm, gfn);
1980 case HV_X64_MSR_REFERENCE_TSC: {
1982 HV_REFERENCE_TSC_PAGE tsc_ref;
1983 memset(&tsc_ref, 0, sizeof(tsc_ref));
1984 kvm->arch.hv_tsc_page = data;
1985 if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1987 gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1988 if (kvm_write_guest(kvm, gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
1989 &tsc_ref, sizeof(tsc_ref)))
1991 mark_page_dirty(kvm, gfn);
1995 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1996 "data 0x%llx\n", msr, data);
2002 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2005 case HV_X64_MSR_APIC_ASSIST_PAGE: {
2009 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
2010 vcpu->arch.hv_vapic = data;
2011 if (kvm_lapic_enable_pv_eoi(vcpu, 0))
2015 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
2016 addr = gfn_to_hva(vcpu->kvm, gfn);
2017 if (kvm_is_error_hva(addr))
2019 if (__clear_user((void __user *)addr, PAGE_SIZE))
2021 vcpu->arch.hv_vapic = data;
2022 mark_page_dirty(vcpu->kvm, gfn);
2023 if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
2027 case HV_X64_MSR_EOI:
2028 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
2029 case HV_X64_MSR_ICR:
2030 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
2031 case HV_X64_MSR_TPR:
2032 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
2034 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2035 "data 0x%llx\n", msr, data);
2042 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2044 gpa_t gpa = data & ~0x3f;
2046 /* Bits 2:5 are reserved, Should be zero */
2050 vcpu->arch.apf.msr_val = data;
2052 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2053 kvm_clear_async_pf_completion_queue(vcpu);
2054 kvm_async_pf_hash_reset(vcpu);
2058 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2062 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2063 kvm_async_pf_wakeup_all(vcpu);
2067 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2069 vcpu->arch.pv_time_enabled = false;
2072 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2076 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2079 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2080 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2081 vcpu->arch.st.accum_steal = delta;
2084 static void record_steal_time(struct kvm_vcpu *vcpu)
2086 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2089 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2090 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2093 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2094 vcpu->arch.st.steal.version += 2;
2095 vcpu->arch.st.accum_steal = 0;
2097 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2098 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2101 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2104 u32 msr = msr_info->index;
2105 u64 data = msr_info->data;
2108 case MSR_AMD64_NB_CFG:
2109 case MSR_IA32_UCODE_REV:
2110 case MSR_IA32_UCODE_WRITE:
2111 case MSR_VM_HSAVE_PA:
2112 case MSR_AMD64_PATCH_LOADER:
2113 case MSR_AMD64_BU_CFG2:
2117 return set_efer(vcpu, data);
2119 data &= ~(u64)0x40; /* ignore flush filter disable */
2120 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2121 data &= ~(u64)0x8; /* ignore TLB cache disable */
2122 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2124 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2129 case MSR_FAM10H_MMIO_CONF_BASE:
2131 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2136 case MSR_IA32_DEBUGCTLMSR:
2138 /* We support the non-activated case already */
2140 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2141 /* Values other than LBR and BTF are vendor-specific,
2142 thus reserved and should throw a #GP */
2145 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2148 case 0x200 ... 0x2ff:
2149 return set_msr_mtrr(vcpu, msr, data);
2150 case MSR_IA32_APICBASE:
2151 return kvm_set_apic_base(vcpu, msr_info);
2152 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2153 return kvm_x2apic_msr_write(vcpu, msr, data);
2154 case MSR_IA32_TSCDEADLINE:
2155 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2157 case MSR_IA32_TSC_ADJUST:
2158 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2159 if (!msr_info->host_initiated) {
2160 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2161 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2163 vcpu->arch.ia32_tsc_adjust_msr = data;
2166 case MSR_IA32_MISC_ENABLE:
2167 vcpu->arch.ia32_misc_enable_msr = data;
2169 case MSR_KVM_WALL_CLOCK_NEW:
2170 case MSR_KVM_WALL_CLOCK:
2171 vcpu->kvm->arch.wall_clock = data;
2172 kvm_write_wall_clock(vcpu->kvm, data);
2174 case MSR_KVM_SYSTEM_TIME_NEW:
2175 case MSR_KVM_SYSTEM_TIME: {
2177 kvmclock_reset(vcpu);
2179 vcpu->arch.time = data;
2180 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2182 /* we verify if the enable bit is set... */
2186 gpa_offset = data & ~(PAGE_MASK | 1);
2188 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2189 &vcpu->arch.pv_time, data & ~1ULL,
2190 sizeof(struct pvclock_vcpu_time_info)))
2191 vcpu->arch.pv_time_enabled = false;
2193 vcpu->arch.pv_time_enabled = true;
2197 case MSR_KVM_ASYNC_PF_EN:
2198 if (kvm_pv_enable_async_pf(vcpu, data))
2201 case MSR_KVM_STEAL_TIME:
2203 if (unlikely(!sched_info_on()))
2206 if (data & KVM_STEAL_RESERVED_MASK)
2209 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2210 data & KVM_STEAL_VALID_BITS,
2211 sizeof(struct kvm_steal_time)))
2214 vcpu->arch.st.msr_val = data;
2216 if (!(data & KVM_MSR_ENABLED))
2219 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2222 accumulate_steal_time(vcpu);
2225 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2228 case MSR_KVM_PV_EOI_EN:
2229 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2233 case MSR_IA32_MCG_CTL:
2234 case MSR_IA32_MCG_STATUS:
2235 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2236 return set_msr_mce(vcpu, msr, data);
2238 /* Performance counters are not protected by a CPUID bit,
2239 * so we should check all of them in the generic path for the sake of
2240 * cross vendor migration.
2241 * Writing a zero into the event select MSRs disables them,
2242 * which we perfectly emulate ;-). Any other value should be at least
2243 * reported, some guests depend on them.
2245 case MSR_K7_EVNTSEL0:
2246 case MSR_K7_EVNTSEL1:
2247 case MSR_K7_EVNTSEL2:
2248 case MSR_K7_EVNTSEL3:
2250 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2251 "0x%x data 0x%llx\n", msr, data);
2253 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2254 * so we ignore writes to make it happy.
2256 case MSR_K7_PERFCTR0:
2257 case MSR_K7_PERFCTR1:
2258 case MSR_K7_PERFCTR2:
2259 case MSR_K7_PERFCTR3:
2260 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2261 "0x%x data 0x%llx\n", msr, data);
2263 case MSR_P6_PERFCTR0:
2264 case MSR_P6_PERFCTR1:
2266 case MSR_P6_EVNTSEL0:
2267 case MSR_P6_EVNTSEL1:
2268 if (kvm_pmu_msr(vcpu, msr))
2269 return kvm_pmu_set_msr(vcpu, msr_info);
2271 if (pr || data != 0)
2272 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2273 "0x%x data 0x%llx\n", msr, data);
2275 case MSR_K7_CLK_CTL:
2277 * Ignore all writes to this no longer documented MSR.
2278 * Writes are only relevant for old K7 processors,
2279 * all pre-dating SVM, but a recommended workaround from
2280 * AMD for these chips. It is possible to specify the
2281 * affected processor models on the command line, hence
2282 * the need to ignore the workaround.
2285 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2286 if (kvm_hv_msr_partition_wide(msr)) {
2288 mutex_lock(&vcpu->kvm->lock);
2289 r = set_msr_hyperv_pw(vcpu, msr, data);
2290 mutex_unlock(&vcpu->kvm->lock);
2293 return set_msr_hyperv(vcpu, msr, data);
2295 case MSR_IA32_BBL_CR_CTL3:
2296 /* Drop writes to this legacy MSR -- see rdmsr
2297 * counterpart for further detail.
2299 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2301 case MSR_AMD64_OSVW_ID_LENGTH:
2302 if (!guest_cpuid_has_osvw(vcpu))
2304 vcpu->arch.osvw.length = data;
2306 case MSR_AMD64_OSVW_STATUS:
2307 if (!guest_cpuid_has_osvw(vcpu))
2309 vcpu->arch.osvw.status = data;
2312 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2313 return xen_hvm_config(vcpu, data);
2314 if (kvm_pmu_msr(vcpu, msr))
2315 return kvm_pmu_set_msr(vcpu, msr_info);
2317 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2321 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2328 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2332 * Reads an msr value (of 'msr_index') into 'pdata'.
2333 * Returns 0 on success, non-0 otherwise.
2334 * Assumes vcpu_load() was already called.
2336 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2338 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2340 EXPORT_SYMBOL_GPL(kvm_get_msr);
2342 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2344 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2346 if (!msr_mtrr_valid(msr))
2349 if (msr == MSR_MTRRdefType)
2350 *pdata = vcpu->arch.mtrr_state.def_type +
2351 (vcpu->arch.mtrr_state.enabled << 10);
2352 else if (msr == MSR_MTRRfix64K_00000)
2354 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2355 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2356 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2357 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2358 else if (msr == MSR_IA32_CR_PAT)
2359 *pdata = vcpu->arch.pat;
2360 else { /* Variable MTRRs */
2361 int idx, is_mtrr_mask;
2364 idx = (msr - 0x200) / 2;
2365 is_mtrr_mask = msr - 0x200 - 2 * idx;
2368 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2371 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2378 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2381 u64 mcg_cap = vcpu->arch.mcg_cap;
2382 unsigned bank_num = mcg_cap & 0xff;
2385 case MSR_IA32_P5_MC_ADDR:
2386 case MSR_IA32_P5_MC_TYPE:
2389 case MSR_IA32_MCG_CAP:
2390 data = vcpu->arch.mcg_cap;
2392 case MSR_IA32_MCG_CTL:
2393 if (!(mcg_cap & MCG_CTL_P))
2395 data = vcpu->arch.mcg_ctl;
2397 case MSR_IA32_MCG_STATUS:
2398 data = vcpu->arch.mcg_status;
2401 if (msr >= MSR_IA32_MC0_CTL &&
2402 msr < MSR_IA32_MCx_CTL(bank_num)) {
2403 u32 offset = msr - MSR_IA32_MC0_CTL;
2404 data = vcpu->arch.mce_banks[offset];
2413 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2416 struct kvm *kvm = vcpu->kvm;
2419 case HV_X64_MSR_GUEST_OS_ID:
2420 data = kvm->arch.hv_guest_os_id;
2422 case HV_X64_MSR_HYPERCALL:
2423 data = kvm->arch.hv_hypercall;
2425 case HV_X64_MSR_TIME_REF_COUNT: {
2427 div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2430 case HV_X64_MSR_REFERENCE_TSC:
2431 data = kvm->arch.hv_tsc_page;
2434 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2442 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2447 case HV_X64_MSR_VP_INDEX: {
2450 kvm_for_each_vcpu(r, v, vcpu->kvm) {
2458 case HV_X64_MSR_EOI:
2459 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2460 case HV_X64_MSR_ICR:
2461 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2462 case HV_X64_MSR_TPR:
2463 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2464 case HV_X64_MSR_APIC_ASSIST_PAGE:
2465 data = vcpu->arch.hv_vapic;
2468 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2475 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2480 case MSR_IA32_PLATFORM_ID:
2481 case MSR_IA32_EBL_CR_POWERON:
2482 case MSR_IA32_DEBUGCTLMSR:
2483 case MSR_IA32_LASTBRANCHFROMIP:
2484 case MSR_IA32_LASTBRANCHTOIP:
2485 case MSR_IA32_LASTINTFROMIP:
2486 case MSR_IA32_LASTINTTOIP:
2489 case MSR_VM_HSAVE_PA:
2490 case MSR_K7_EVNTSEL0:
2491 case MSR_K7_EVNTSEL1:
2492 case MSR_K7_EVNTSEL2:
2493 case MSR_K7_EVNTSEL3:
2494 case MSR_K7_PERFCTR0:
2495 case MSR_K7_PERFCTR1:
2496 case MSR_K7_PERFCTR2:
2497 case MSR_K7_PERFCTR3:
2498 case MSR_K8_INT_PENDING_MSG:
2499 case MSR_AMD64_NB_CFG:
2500 case MSR_FAM10H_MMIO_CONF_BASE:
2501 case MSR_AMD64_BU_CFG2:
2504 case MSR_P6_PERFCTR0:
2505 case MSR_P6_PERFCTR1:
2506 case MSR_P6_EVNTSEL0:
2507 case MSR_P6_EVNTSEL1:
2508 if (kvm_pmu_msr(vcpu, msr))
2509 return kvm_pmu_get_msr(vcpu, msr, pdata);
2512 case MSR_IA32_UCODE_REV:
2513 data = 0x100000000ULL;
2516 data = 0x500 | KVM_NR_VAR_MTRR;
2518 case 0x200 ... 0x2ff:
2519 return get_msr_mtrr(vcpu, msr, pdata);
2520 case 0xcd: /* fsb frequency */
2524 * MSR_EBC_FREQUENCY_ID
2525 * Conservative value valid for even the basic CPU models.
2526 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2527 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2528 * and 266MHz for model 3, or 4. Set Core Clock
2529 * Frequency to System Bus Frequency Ratio to 1 (bits
2530 * 31:24) even though these are only valid for CPU
2531 * models > 2, however guests may end up dividing or
2532 * multiplying by zero otherwise.
2534 case MSR_EBC_FREQUENCY_ID:
2537 case MSR_IA32_APICBASE:
2538 data = kvm_get_apic_base(vcpu);
2540 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2541 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2543 case MSR_IA32_TSCDEADLINE:
2544 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2546 case MSR_IA32_TSC_ADJUST:
2547 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2549 case MSR_IA32_MISC_ENABLE:
2550 data = vcpu->arch.ia32_misc_enable_msr;
2552 case MSR_IA32_PERF_STATUS:
2553 /* TSC increment by tick */
2555 /* CPU multiplier */
2556 data |= (((uint64_t)4ULL) << 40);
2559 data = vcpu->arch.efer;
2561 case MSR_KVM_WALL_CLOCK:
2562 case MSR_KVM_WALL_CLOCK_NEW:
2563 data = vcpu->kvm->arch.wall_clock;
2565 case MSR_KVM_SYSTEM_TIME:
2566 case MSR_KVM_SYSTEM_TIME_NEW:
2567 data = vcpu->arch.time;
2569 case MSR_KVM_ASYNC_PF_EN:
2570 data = vcpu->arch.apf.msr_val;
2572 case MSR_KVM_STEAL_TIME:
2573 data = vcpu->arch.st.msr_val;
2575 case MSR_KVM_PV_EOI_EN:
2576 data = vcpu->arch.pv_eoi.msr_val;
2578 case MSR_IA32_P5_MC_ADDR:
2579 case MSR_IA32_P5_MC_TYPE:
2580 case MSR_IA32_MCG_CAP:
2581 case MSR_IA32_MCG_CTL:
2582 case MSR_IA32_MCG_STATUS:
2583 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2584 return get_msr_mce(vcpu, msr, pdata);
2585 case MSR_K7_CLK_CTL:
2587 * Provide expected ramp-up count for K7. All other
2588 * are set to zero, indicating minimum divisors for
2591 * This prevents guest kernels on AMD host with CPU
2592 * type 6, model 8 and higher from exploding due to
2593 * the rdmsr failing.
2597 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2598 if (kvm_hv_msr_partition_wide(msr)) {
2600 mutex_lock(&vcpu->kvm->lock);
2601 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2602 mutex_unlock(&vcpu->kvm->lock);
2605 return get_msr_hyperv(vcpu, msr, pdata);
2607 case MSR_IA32_BBL_CR_CTL3:
2608 /* This legacy MSR exists but isn't fully documented in current
2609 * silicon. It is however accessed by winxp in very narrow
2610 * scenarios where it sets bit #19, itself documented as
2611 * a "reserved" bit. Best effort attempt to source coherent
2612 * read data here should the balance of the register be
2613 * interpreted by the guest:
2615 * L2 cache control register 3: 64GB range, 256KB size,
2616 * enabled, latency 0x1, configured
2620 case MSR_AMD64_OSVW_ID_LENGTH:
2621 if (!guest_cpuid_has_osvw(vcpu))
2623 data = vcpu->arch.osvw.length;
2625 case MSR_AMD64_OSVW_STATUS:
2626 if (!guest_cpuid_has_osvw(vcpu))
2628 data = vcpu->arch.osvw.status;
2631 if (kvm_pmu_msr(vcpu, msr))
2632 return kvm_pmu_get_msr(vcpu, msr, pdata);
2634 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2637 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2645 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2648 * Read or write a bunch of msrs. All parameters are kernel addresses.
2650 * @return number of msrs set successfully.
2652 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2653 struct kvm_msr_entry *entries,
2654 int (*do_msr)(struct kvm_vcpu *vcpu,
2655 unsigned index, u64 *data))
2659 idx = srcu_read_lock(&vcpu->kvm->srcu);
2660 for (i = 0; i < msrs->nmsrs; ++i)
2661 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2663 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2669 * Read or write a bunch of msrs. Parameters are user addresses.
2671 * @return number of msrs set successfully.
2673 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2674 int (*do_msr)(struct kvm_vcpu *vcpu,
2675 unsigned index, u64 *data),
2678 struct kvm_msrs msrs;
2679 struct kvm_msr_entry *entries;
2684 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2688 if (msrs.nmsrs >= MAX_IO_MSRS)
2691 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2692 entries = memdup_user(user_msrs->entries, size);
2693 if (IS_ERR(entries)) {
2694 r = PTR_ERR(entries);
2698 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2703 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2714 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2719 case KVM_CAP_IRQCHIP:
2721 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2722 case KVM_CAP_SET_TSS_ADDR:
2723 case KVM_CAP_EXT_CPUID:
2724 case KVM_CAP_EXT_EMUL_CPUID:
2725 case KVM_CAP_CLOCKSOURCE:
2727 case KVM_CAP_NOP_IO_DELAY:
2728 case KVM_CAP_MP_STATE:
2729 case KVM_CAP_SYNC_MMU:
2730 case KVM_CAP_USER_NMI:
2731 case KVM_CAP_REINJECT_CONTROL:
2732 case KVM_CAP_IRQ_INJECT_STATUS:
2734 case KVM_CAP_IOEVENTFD:
2735 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2737 case KVM_CAP_PIT_STATE2:
2738 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2739 case KVM_CAP_XEN_HVM:
2740 case KVM_CAP_ADJUST_CLOCK:
2741 case KVM_CAP_VCPU_EVENTS:
2742 case KVM_CAP_HYPERV:
2743 case KVM_CAP_HYPERV_VAPIC:
2744 case KVM_CAP_HYPERV_SPIN:
2745 case KVM_CAP_PCI_SEGMENT:
2746 case KVM_CAP_DEBUGREGS:
2747 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2749 case KVM_CAP_ASYNC_PF:
2750 case KVM_CAP_GET_TSC_KHZ:
2751 case KVM_CAP_KVMCLOCK_CTRL:
2752 case KVM_CAP_READONLY_MEM:
2753 case KVM_CAP_HYPERV_TIME:
2754 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2755 case KVM_CAP_TSC_DEADLINE_TIMER:
2756 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2757 case KVM_CAP_ASSIGN_DEV_IRQ:
2758 case KVM_CAP_PCI_2_3:
2762 case KVM_CAP_COALESCED_MMIO:
2763 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2766 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2768 case KVM_CAP_NR_VCPUS:
2769 r = KVM_SOFT_MAX_VCPUS;
2771 case KVM_CAP_MAX_VCPUS:
2774 case KVM_CAP_NR_MEMSLOTS:
2775 r = KVM_USER_MEM_SLOTS;
2777 case KVM_CAP_PV_MMU: /* obsolete */
2780 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2782 r = iommu_present(&pci_bus_type);
2786 r = KVM_MAX_MCE_BANKS;
2791 case KVM_CAP_TSC_CONTROL:
2792 r = kvm_has_tsc_control;
2802 long kvm_arch_dev_ioctl(struct file *filp,
2803 unsigned int ioctl, unsigned long arg)
2805 void __user *argp = (void __user *)arg;
2809 case KVM_GET_MSR_INDEX_LIST: {
2810 struct kvm_msr_list __user *user_msr_list = argp;
2811 struct kvm_msr_list msr_list;
2815 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2818 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2819 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2822 if (n < msr_list.nmsrs)
2825 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2826 num_msrs_to_save * sizeof(u32)))
2828 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2830 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2835 case KVM_GET_SUPPORTED_CPUID:
2836 case KVM_GET_EMULATED_CPUID: {
2837 struct kvm_cpuid2 __user *cpuid_arg = argp;
2838 struct kvm_cpuid2 cpuid;
2841 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2844 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2850 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2855 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2858 mce_cap = KVM_MCE_CAP_SUPPORTED;
2860 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2872 static void wbinvd_ipi(void *garbage)
2877 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2879 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2882 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2884 /* Address WBINVD may be executed by guest */
2885 if (need_emulate_wbinvd(vcpu)) {
2886 if (kvm_x86_ops->has_wbinvd_exit())
2887 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2888 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2889 smp_call_function_single(vcpu->cpu,
2890 wbinvd_ipi, NULL, 1);
2893 kvm_x86_ops->vcpu_load(vcpu, cpu);
2895 /* Apply any externally detected TSC adjustments (due to suspend) */
2896 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2897 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2898 vcpu->arch.tsc_offset_adjustment = 0;
2899 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2902 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2903 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2904 native_read_tsc() - vcpu->arch.last_host_tsc;
2906 mark_tsc_unstable("KVM discovered backwards TSC");
2907 if (check_tsc_unstable()) {
2908 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2909 vcpu->arch.last_guest_tsc);
2910 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2911 vcpu->arch.tsc_catchup = 1;
2914 * On a host with synchronized TSC, there is no need to update
2915 * kvmclock on vcpu->cpu migration
2917 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2918 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2919 if (vcpu->cpu != cpu)
2920 kvm_migrate_timers(vcpu);
2924 accumulate_steal_time(vcpu);
2925 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2928 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2930 kvm_x86_ops->vcpu_put(vcpu);
2931 kvm_put_guest_fpu(vcpu);
2932 vcpu->arch.last_host_tsc = native_read_tsc();
2935 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2936 struct kvm_lapic_state *s)
2938 kvm_x86_ops->sync_pir_to_irr(vcpu);
2939 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2944 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2945 struct kvm_lapic_state *s)
2947 kvm_apic_post_state_restore(vcpu, s);
2948 update_cr8_intercept(vcpu);
2953 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2954 struct kvm_interrupt *irq)
2956 if (irq->irq >= KVM_NR_INTERRUPTS)
2958 if (irqchip_in_kernel(vcpu->kvm))
2961 kvm_queue_interrupt(vcpu, irq->irq, false);
2962 kvm_make_request(KVM_REQ_EVENT, vcpu);
2967 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2969 kvm_inject_nmi(vcpu);
2974 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2975 struct kvm_tpr_access_ctl *tac)
2979 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2983 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2987 unsigned bank_num = mcg_cap & 0xff, bank;
2990 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2992 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2995 vcpu->arch.mcg_cap = mcg_cap;
2996 /* Init IA32_MCG_CTL to all 1s */
2997 if (mcg_cap & MCG_CTL_P)
2998 vcpu->arch.mcg_ctl = ~(u64)0;
2999 /* Init IA32_MCi_CTL to all 1s */
3000 for (bank = 0; bank < bank_num; bank++)
3001 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3006 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3007 struct kvm_x86_mce *mce)
3009 u64 mcg_cap = vcpu->arch.mcg_cap;
3010 unsigned bank_num = mcg_cap & 0xff;
3011 u64 *banks = vcpu->arch.mce_banks;
3013 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3016 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3017 * reporting is disabled
3019 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3020 vcpu->arch.mcg_ctl != ~(u64)0)
3022 banks += 4 * mce->bank;
3024 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3025 * reporting is disabled for the bank
3027 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3029 if (mce->status & MCI_STATUS_UC) {
3030 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3031 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3032 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3035 if (banks[1] & MCI_STATUS_VAL)
3036 mce->status |= MCI_STATUS_OVER;
3037 banks[2] = mce->addr;
3038 banks[3] = mce->misc;
3039 vcpu->arch.mcg_status = mce->mcg_status;
3040 banks[1] = mce->status;
3041 kvm_queue_exception(vcpu, MC_VECTOR);
3042 } else if (!(banks[1] & MCI_STATUS_VAL)
3043 || !(banks[1] & MCI_STATUS_UC)) {
3044 if (banks[1] & MCI_STATUS_VAL)
3045 mce->status |= MCI_STATUS_OVER;
3046 banks[2] = mce->addr;
3047 banks[3] = mce->misc;
3048 banks[1] = mce->status;
3050 banks[1] |= MCI_STATUS_OVER;
3054 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3055 struct kvm_vcpu_events *events)
3058 events->exception.injected =
3059 vcpu->arch.exception.pending &&
3060 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3061 events->exception.nr = vcpu->arch.exception.nr;
3062 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3063 events->exception.pad = 0;
3064 events->exception.error_code = vcpu->arch.exception.error_code;
3066 events->interrupt.injected =
3067 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3068 events->interrupt.nr = vcpu->arch.interrupt.nr;
3069 events->interrupt.soft = 0;
3070 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3072 events->nmi.injected = vcpu->arch.nmi_injected;
3073 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3074 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3075 events->nmi.pad = 0;
3077 events->sipi_vector = 0; /* never valid when reporting to user space */
3079 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3080 | KVM_VCPUEVENT_VALID_SHADOW);
3081 memset(&events->reserved, 0, sizeof(events->reserved));
3084 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3085 struct kvm_vcpu_events *events)
3087 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3088 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3089 | KVM_VCPUEVENT_VALID_SHADOW))
3093 vcpu->arch.exception.pending = events->exception.injected;
3094 vcpu->arch.exception.nr = events->exception.nr;
3095 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3096 vcpu->arch.exception.error_code = events->exception.error_code;
3098 vcpu->arch.interrupt.pending = events->interrupt.injected;
3099 vcpu->arch.interrupt.nr = events->interrupt.nr;
3100 vcpu->arch.interrupt.soft = events->interrupt.soft;
3101 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3102 kvm_x86_ops->set_interrupt_shadow(vcpu,
3103 events->interrupt.shadow);
3105 vcpu->arch.nmi_injected = events->nmi.injected;
3106 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3107 vcpu->arch.nmi_pending = events->nmi.pending;
3108 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3110 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3111 kvm_vcpu_has_lapic(vcpu))
3112 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3114 kvm_make_request(KVM_REQ_EVENT, vcpu);
3119 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3120 struct kvm_debugregs *dbgregs)
3124 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3125 kvm_get_dr(vcpu, 6, &val);
3127 dbgregs->dr7 = vcpu->arch.dr7;
3129 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3132 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3133 struct kvm_debugregs *dbgregs)
3138 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3139 vcpu->arch.dr6 = dbgregs->dr6;
3140 kvm_update_dr6(vcpu);
3141 vcpu->arch.dr7 = dbgregs->dr7;
3142 kvm_update_dr7(vcpu);
3147 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3149 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3151 struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3152 u64 xstate_bv = xsave->xsave_hdr.xstate_bv;
3156 * Copy legacy XSAVE area, to avoid complications with CPUID
3157 * leaves 0 and 1 in the loop below.
3159 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3162 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3165 * Copy each region from the possibly compacted offset to the
3166 * non-compacted offset.
3168 valid = xstate_bv & ~XSTATE_FPSSE;
3170 u64 feature = valid & -valid;
3171 int index = fls64(feature) - 1;
3172 void *src = get_xsave_addr(xsave, feature);
3175 u32 size, offset, ecx, edx;
3176 cpuid_count(XSTATE_CPUID, index,
3177 &size, &offset, &ecx, &edx);
3178 memcpy(dest + offset, src, size);
3185 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3187 struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3188 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3192 * Copy legacy XSAVE area, to avoid complications with CPUID
3193 * leaves 0 and 1 in the loop below.
3195 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3197 /* Set XSTATE_BV and possibly XCOMP_BV. */
3198 xsave->xsave_hdr.xstate_bv = xstate_bv;
3200 xsave->xsave_hdr.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3203 * Copy each region from the non-compacted offset to the
3204 * possibly compacted offset.
3206 valid = xstate_bv & ~XSTATE_FPSSE;
3208 u64 feature = valid & -valid;
3209 int index = fls64(feature) - 1;
3210 void *dest = get_xsave_addr(xsave, feature);
3213 u32 size, offset, ecx, edx;
3214 cpuid_count(XSTATE_CPUID, index,
3215 &size, &offset, &ecx, &edx);
3216 memcpy(dest, src + offset, size);
3224 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3225 struct kvm_xsave *guest_xsave)
3227 if (cpu_has_xsave) {
3228 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3229 fill_xsave((u8 *) guest_xsave->region, vcpu);
3231 memcpy(guest_xsave->region,
3232 &vcpu->arch.guest_fpu.state->fxsave,
3233 sizeof(struct i387_fxsave_struct));
3234 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3239 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3240 struct kvm_xsave *guest_xsave)
3243 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3245 if (cpu_has_xsave) {
3247 * Here we allow setting states that are not present in
3248 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3249 * with old userspace.
3251 if (xstate_bv & ~kvm_supported_xcr0())
3253 load_xsave(vcpu, (u8 *)guest_xsave->region);
3255 if (xstate_bv & ~XSTATE_FPSSE)
3257 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3258 guest_xsave->region, sizeof(struct i387_fxsave_struct));
3263 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3264 struct kvm_xcrs *guest_xcrs)
3266 if (!cpu_has_xsave) {
3267 guest_xcrs->nr_xcrs = 0;
3271 guest_xcrs->nr_xcrs = 1;
3272 guest_xcrs->flags = 0;
3273 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3274 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3277 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3278 struct kvm_xcrs *guest_xcrs)
3285 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3288 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3289 /* Only support XCR0 currently */
3290 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3291 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3292 guest_xcrs->xcrs[i].value);
3301 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3302 * stopped by the hypervisor. This function will be called from the host only.
3303 * EINVAL is returned when the host attempts to set the flag for a guest that
3304 * does not support pv clocks.
3306 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3308 if (!vcpu->arch.pv_time_enabled)
3310 vcpu->arch.pvclock_set_guest_stopped_request = true;
3311 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3315 long kvm_arch_vcpu_ioctl(struct file *filp,
3316 unsigned int ioctl, unsigned long arg)
3318 struct kvm_vcpu *vcpu = filp->private_data;
3319 void __user *argp = (void __user *)arg;
3322 struct kvm_lapic_state *lapic;
3323 struct kvm_xsave *xsave;
3324 struct kvm_xcrs *xcrs;
3330 case KVM_GET_LAPIC: {
3332 if (!vcpu->arch.apic)
3334 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3339 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3343 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3348 case KVM_SET_LAPIC: {
3350 if (!vcpu->arch.apic)
3352 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3353 if (IS_ERR(u.lapic))
3354 return PTR_ERR(u.lapic);
3356 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3359 case KVM_INTERRUPT: {
3360 struct kvm_interrupt irq;
3363 if (copy_from_user(&irq, argp, sizeof irq))
3365 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3369 r = kvm_vcpu_ioctl_nmi(vcpu);
3372 case KVM_SET_CPUID: {
3373 struct kvm_cpuid __user *cpuid_arg = argp;
3374 struct kvm_cpuid cpuid;
3377 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3379 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3382 case KVM_SET_CPUID2: {
3383 struct kvm_cpuid2 __user *cpuid_arg = argp;
3384 struct kvm_cpuid2 cpuid;
3387 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3389 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3390 cpuid_arg->entries);
3393 case KVM_GET_CPUID2: {
3394 struct kvm_cpuid2 __user *cpuid_arg = argp;
3395 struct kvm_cpuid2 cpuid;
3398 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3400 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3401 cpuid_arg->entries);
3405 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3411 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3414 r = msr_io(vcpu, argp, do_set_msr, 0);
3416 case KVM_TPR_ACCESS_REPORTING: {
3417 struct kvm_tpr_access_ctl tac;
3420 if (copy_from_user(&tac, argp, sizeof tac))
3422 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3426 if (copy_to_user(argp, &tac, sizeof tac))
3431 case KVM_SET_VAPIC_ADDR: {
3432 struct kvm_vapic_addr va;
3435 if (!irqchip_in_kernel(vcpu->kvm))
3438 if (copy_from_user(&va, argp, sizeof va))
3440 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3443 case KVM_X86_SETUP_MCE: {
3447 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3449 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3452 case KVM_X86_SET_MCE: {
3453 struct kvm_x86_mce mce;
3456 if (copy_from_user(&mce, argp, sizeof mce))
3458 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3461 case KVM_GET_VCPU_EVENTS: {
3462 struct kvm_vcpu_events events;
3464 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3467 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3472 case KVM_SET_VCPU_EVENTS: {
3473 struct kvm_vcpu_events events;
3476 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3479 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3482 case KVM_GET_DEBUGREGS: {
3483 struct kvm_debugregs dbgregs;
3485 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3488 if (copy_to_user(argp, &dbgregs,
3489 sizeof(struct kvm_debugregs)))
3494 case KVM_SET_DEBUGREGS: {
3495 struct kvm_debugregs dbgregs;
3498 if (copy_from_user(&dbgregs, argp,
3499 sizeof(struct kvm_debugregs)))
3502 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3505 case KVM_GET_XSAVE: {
3506 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3511 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3514 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3519 case KVM_SET_XSAVE: {
3520 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3521 if (IS_ERR(u.xsave))
3522 return PTR_ERR(u.xsave);
3524 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3527 case KVM_GET_XCRS: {
3528 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3533 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3536 if (copy_to_user(argp, u.xcrs,
3537 sizeof(struct kvm_xcrs)))
3542 case KVM_SET_XCRS: {
3543 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3545 return PTR_ERR(u.xcrs);
3547 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3550 case KVM_SET_TSC_KHZ: {
3554 user_tsc_khz = (u32)arg;
3556 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3559 if (user_tsc_khz == 0)
3560 user_tsc_khz = tsc_khz;
3562 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3567 case KVM_GET_TSC_KHZ: {
3568 r = vcpu->arch.virtual_tsc_khz;
3571 case KVM_KVMCLOCK_CTRL: {
3572 r = kvm_set_guest_paused(vcpu);
3583 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3585 return VM_FAULT_SIGBUS;
3588 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3592 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3594 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3598 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3601 kvm->arch.ept_identity_map_addr = ident_addr;
3605 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3606 u32 kvm_nr_mmu_pages)
3608 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3611 mutex_lock(&kvm->slots_lock);
3613 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3614 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3616 mutex_unlock(&kvm->slots_lock);
3620 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3622 return kvm->arch.n_max_mmu_pages;
3625 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3630 switch (chip->chip_id) {
3631 case KVM_IRQCHIP_PIC_MASTER:
3632 memcpy(&chip->chip.pic,
3633 &pic_irqchip(kvm)->pics[0],
3634 sizeof(struct kvm_pic_state));
3636 case KVM_IRQCHIP_PIC_SLAVE:
3637 memcpy(&chip->chip.pic,
3638 &pic_irqchip(kvm)->pics[1],
3639 sizeof(struct kvm_pic_state));
3641 case KVM_IRQCHIP_IOAPIC:
3642 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3651 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3656 switch (chip->chip_id) {
3657 case KVM_IRQCHIP_PIC_MASTER:
3658 spin_lock(&pic_irqchip(kvm)->lock);
3659 memcpy(&pic_irqchip(kvm)->pics[0],
3661 sizeof(struct kvm_pic_state));
3662 spin_unlock(&pic_irqchip(kvm)->lock);
3664 case KVM_IRQCHIP_PIC_SLAVE:
3665 spin_lock(&pic_irqchip(kvm)->lock);
3666 memcpy(&pic_irqchip(kvm)->pics[1],
3668 sizeof(struct kvm_pic_state));
3669 spin_unlock(&pic_irqchip(kvm)->lock);
3671 case KVM_IRQCHIP_IOAPIC:
3672 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3678 kvm_pic_update_irq(pic_irqchip(kvm));
3682 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3686 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3687 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3688 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3692 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3696 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3697 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3698 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3699 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3703 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3707 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3708 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3709 sizeof(ps->channels));
3710 ps->flags = kvm->arch.vpit->pit_state.flags;
3711 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3712 memset(&ps->reserved, 0, sizeof(ps->reserved));
3716 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3718 int r = 0, start = 0;
3719 u32 prev_legacy, cur_legacy;
3720 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3721 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3722 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3723 if (!prev_legacy && cur_legacy)
3725 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3726 sizeof(kvm->arch.vpit->pit_state.channels));
3727 kvm->arch.vpit->pit_state.flags = ps->flags;
3728 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3729 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3733 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3734 struct kvm_reinject_control *control)
3736 if (!kvm->arch.vpit)
3738 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3739 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3740 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3745 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3746 * @kvm: kvm instance
3747 * @log: slot id and address to which we copy the log
3749 * We need to keep it in mind that VCPU threads can write to the bitmap
3750 * concurrently. So, to avoid losing data, we keep the following order for
3753 * 1. Take a snapshot of the bit and clear it if needed.
3754 * 2. Write protect the corresponding page.
3755 * 3. Flush TLB's if needed.
3756 * 4. Copy the snapshot to the userspace.
3758 * Between 2 and 3, the guest may write to the page using the remaining TLB
3759 * entry. This is not a problem because the page will be reported dirty at
3760 * step 4 using the snapshot taken before and step 3 ensures that successive
3761 * writes will be logged for the next call.
3763 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3766 struct kvm_memory_slot *memslot;
3768 unsigned long *dirty_bitmap;
3769 unsigned long *dirty_bitmap_buffer;
3770 bool is_dirty = false;
3772 mutex_lock(&kvm->slots_lock);
3775 if (log->slot >= KVM_USER_MEM_SLOTS)
3778 memslot = id_to_memslot(kvm->memslots, log->slot);
3780 dirty_bitmap = memslot->dirty_bitmap;
3785 n = kvm_dirty_bitmap_bytes(memslot);
3787 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3788 memset(dirty_bitmap_buffer, 0, n);
3790 spin_lock(&kvm->mmu_lock);
3792 for (i = 0; i < n / sizeof(long); i++) {
3796 if (!dirty_bitmap[i])
3801 mask = xchg(&dirty_bitmap[i], 0);
3802 dirty_bitmap_buffer[i] = mask;
3804 offset = i * BITS_PER_LONG;
3805 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3808 spin_unlock(&kvm->mmu_lock);
3810 /* See the comments in kvm_mmu_slot_remove_write_access(). */
3811 lockdep_assert_held(&kvm->slots_lock);
3814 * All the TLBs can be flushed out of mmu lock, see the comments in
3815 * kvm_mmu_slot_remove_write_access().
3818 kvm_flush_remote_tlbs(kvm);
3821 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3826 mutex_unlock(&kvm->slots_lock);
3830 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3833 if (!irqchip_in_kernel(kvm))
3836 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3837 irq_event->irq, irq_event->level,
3842 long kvm_arch_vm_ioctl(struct file *filp,
3843 unsigned int ioctl, unsigned long arg)
3845 struct kvm *kvm = filp->private_data;
3846 void __user *argp = (void __user *)arg;
3849 * This union makes it completely explicit to gcc-3.x
3850 * that these two variables' stack usage should be
3851 * combined, not added together.
3854 struct kvm_pit_state ps;
3855 struct kvm_pit_state2 ps2;
3856 struct kvm_pit_config pit_config;
3860 case KVM_SET_TSS_ADDR:
3861 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3863 case KVM_SET_IDENTITY_MAP_ADDR: {
3867 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3869 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3872 case KVM_SET_NR_MMU_PAGES:
3873 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3875 case KVM_GET_NR_MMU_PAGES:
3876 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3878 case KVM_CREATE_IRQCHIP: {
3879 struct kvm_pic *vpic;
3881 mutex_lock(&kvm->lock);
3884 goto create_irqchip_unlock;
3886 if (atomic_read(&kvm->online_vcpus))
3887 goto create_irqchip_unlock;
3889 vpic = kvm_create_pic(kvm);
3891 r = kvm_ioapic_init(kvm);
3893 mutex_lock(&kvm->slots_lock);
3894 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3896 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3898 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3900 mutex_unlock(&kvm->slots_lock);
3902 goto create_irqchip_unlock;
3905 goto create_irqchip_unlock;
3907 kvm->arch.vpic = vpic;
3909 r = kvm_setup_default_irq_routing(kvm);
3911 mutex_lock(&kvm->slots_lock);
3912 mutex_lock(&kvm->irq_lock);
3913 kvm_ioapic_destroy(kvm);
3914 kvm_destroy_pic(kvm);
3915 mutex_unlock(&kvm->irq_lock);
3916 mutex_unlock(&kvm->slots_lock);
3918 create_irqchip_unlock:
3919 mutex_unlock(&kvm->lock);
3922 case KVM_CREATE_PIT:
3923 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3925 case KVM_CREATE_PIT2:
3927 if (copy_from_user(&u.pit_config, argp,
3928 sizeof(struct kvm_pit_config)))
3931 mutex_lock(&kvm->slots_lock);
3934 goto create_pit_unlock;
3936 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3940 mutex_unlock(&kvm->slots_lock);
3942 case KVM_GET_IRQCHIP: {
3943 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3944 struct kvm_irqchip *chip;
3946 chip = memdup_user(argp, sizeof(*chip));
3953 if (!irqchip_in_kernel(kvm))
3954 goto get_irqchip_out;
3955 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3957 goto get_irqchip_out;
3959 if (copy_to_user(argp, chip, sizeof *chip))
3960 goto get_irqchip_out;
3966 case KVM_SET_IRQCHIP: {
3967 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3968 struct kvm_irqchip *chip;
3970 chip = memdup_user(argp, sizeof(*chip));
3977 if (!irqchip_in_kernel(kvm))
3978 goto set_irqchip_out;
3979 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3981 goto set_irqchip_out;
3989 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3992 if (!kvm->arch.vpit)
3994 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3998 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4005 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4008 if (!kvm->arch.vpit)
4010 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4013 case KVM_GET_PIT2: {
4015 if (!kvm->arch.vpit)
4017 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4021 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4026 case KVM_SET_PIT2: {
4028 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4031 if (!kvm->arch.vpit)
4033 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4036 case KVM_REINJECT_CONTROL: {
4037 struct kvm_reinject_control control;
4039 if (copy_from_user(&control, argp, sizeof(control)))
4041 r = kvm_vm_ioctl_reinject(kvm, &control);
4044 case KVM_XEN_HVM_CONFIG: {
4046 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4047 sizeof(struct kvm_xen_hvm_config)))
4050 if (kvm->arch.xen_hvm_config.flags)
4055 case KVM_SET_CLOCK: {
4056 struct kvm_clock_data user_ns;
4061 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4069 local_irq_disable();
4070 now_ns = get_kernel_ns();
4071 delta = user_ns.clock - now_ns;
4073 kvm->arch.kvmclock_offset = delta;
4074 kvm_gen_update_masterclock(kvm);
4077 case KVM_GET_CLOCK: {
4078 struct kvm_clock_data user_ns;
4081 local_irq_disable();
4082 now_ns = get_kernel_ns();
4083 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4086 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4089 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4096 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4102 static void kvm_init_msr_list(void)
4107 /* skip the first msrs in the list. KVM-specific */
4108 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
4109 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4113 * Even MSRs that are valid in the host may not be exposed
4114 * to the guests in some cases. We could work around this
4115 * in VMX with the generic MSR save/load machinery, but it
4116 * is not really worthwhile since it will really only
4117 * happen with nested virtualization.
4119 switch (msrs_to_save[i]) {
4120 case MSR_IA32_BNDCFGS:
4121 if (!kvm_x86_ops->mpx_supported())
4129 msrs_to_save[j] = msrs_to_save[i];
4132 num_msrs_to_save = j;
4135 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4143 if (!(vcpu->arch.apic &&
4144 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
4145 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
4156 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4163 if (!(vcpu->arch.apic &&
4164 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
4165 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
4167 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4177 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4178 struct kvm_segment *var, int seg)
4180 kvm_x86_ops->set_segment(vcpu, var, seg);
4183 void kvm_get_segment(struct kvm_vcpu *vcpu,
4184 struct kvm_segment *var, int seg)
4186 kvm_x86_ops->get_segment(vcpu, var, seg);
4189 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4190 struct x86_exception *exception)
4194 BUG_ON(!mmu_is_nested(vcpu));
4196 /* NPT walks are always user-walks */
4197 access |= PFERR_USER_MASK;
4198 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4203 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4204 struct x86_exception *exception)
4206 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4207 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4210 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4211 struct x86_exception *exception)
4213 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4214 access |= PFERR_FETCH_MASK;
4215 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4218 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4219 struct x86_exception *exception)
4221 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4222 access |= PFERR_WRITE_MASK;
4223 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4226 /* uses this to access any guest's mapped memory without checking CPL */
4227 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4228 struct x86_exception *exception)
4230 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4233 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4234 struct kvm_vcpu *vcpu, u32 access,
4235 struct x86_exception *exception)
4238 int r = X86EMUL_CONTINUE;
4241 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4243 unsigned offset = addr & (PAGE_SIZE-1);
4244 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4247 if (gpa == UNMAPPED_GVA)
4248 return X86EMUL_PROPAGATE_FAULT;
4249 ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, data,
4252 r = X86EMUL_IO_NEEDED;
4264 /* used for instruction fetching */
4265 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4266 gva_t addr, void *val, unsigned int bytes,
4267 struct x86_exception *exception)
4269 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4270 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4274 /* Inline kvm_read_guest_virt_helper for speed. */
4275 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4277 if (unlikely(gpa == UNMAPPED_GVA))
4278 return X86EMUL_PROPAGATE_FAULT;
4280 offset = addr & (PAGE_SIZE-1);
4281 if (WARN_ON(offset + bytes > PAGE_SIZE))
4282 bytes = (unsigned)PAGE_SIZE - offset;
4283 ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, val,
4285 if (unlikely(ret < 0))
4286 return X86EMUL_IO_NEEDED;
4288 return X86EMUL_CONTINUE;
4291 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4292 gva_t addr, void *val, unsigned int bytes,
4293 struct x86_exception *exception)
4295 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4296 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4298 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4301 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4303 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4304 gva_t addr, void *val, unsigned int bytes,
4305 struct x86_exception *exception)
4307 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4308 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4311 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4312 gva_t addr, void *val,
4314 struct x86_exception *exception)
4316 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4318 int r = X86EMUL_CONTINUE;
4321 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4324 unsigned offset = addr & (PAGE_SIZE-1);
4325 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4328 if (gpa == UNMAPPED_GVA)
4329 return X86EMUL_PROPAGATE_FAULT;
4330 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4332 r = X86EMUL_IO_NEEDED;
4343 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4345 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4346 gpa_t *gpa, struct x86_exception *exception,
4349 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4350 | (write ? PFERR_WRITE_MASK : 0);
4352 if (vcpu_match_mmio_gva(vcpu, gva)
4353 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4354 vcpu->arch.access, access)) {
4355 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4356 (gva & (PAGE_SIZE - 1));
4357 trace_vcpu_match_mmio(gva, *gpa, write, false);
4361 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4363 if (*gpa == UNMAPPED_GVA)
4366 /* For APIC access vmexit */
4367 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4370 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4371 trace_vcpu_match_mmio(gva, *gpa, write, true);
4378 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4379 const void *val, int bytes)
4383 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4386 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4390 struct read_write_emulator_ops {
4391 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4393 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4394 void *val, int bytes);
4395 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4396 int bytes, void *val);
4397 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4398 void *val, int bytes);
4402 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4404 if (vcpu->mmio_read_completed) {
4405 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4406 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4407 vcpu->mmio_read_completed = 0;
4414 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4415 void *val, int bytes)
4417 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4420 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4421 void *val, int bytes)
4423 return emulator_write_phys(vcpu, gpa, val, bytes);
4426 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4428 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4429 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4432 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4433 void *val, int bytes)
4435 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4436 return X86EMUL_IO_NEEDED;
4439 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4440 void *val, int bytes)
4442 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4444 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4445 return X86EMUL_CONTINUE;
4448 static const struct read_write_emulator_ops read_emultor = {
4449 .read_write_prepare = read_prepare,
4450 .read_write_emulate = read_emulate,
4451 .read_write_mmio = vcpu_mmio_read,
4452 .read_write_exit_mmio = read_exit_mmio,
4455 static const struct read_write_emulator_ops write_emultor = {
4456 .read_write_emulate = write_emulate,
4457 .read_write_mmio = write_mmio,
4458 .read_write_exit_mmio = write_exit_mmio,
4462 static int emulator_read_write_onepage(unsigned long addr, void *val,
4464 struct x86_exception *exception,
4465 struct kvm_vcpu *vcpu,
4466 const struct read_write_emulator_ops *ops)
4470 bool write = ops->write;
4471 struct kvm_mmio_fragment *frag;
4473 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4476 return X86EMUL_PROPAGATE_FAULT;
4478 /* For APIC access vmexit */
4482 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4483 return X86EMUL_CONTINUE;
4487 * Is this MMIO handled locally?
4489 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4490 if (handled == bytes)
4491 return X86EMUL_CONTINUE;
4497 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4498 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4502 return X86EMUL_CONTINUE;
4505 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4506 void *val, unsigned int bytes,
4507 struct x86_exception *exception,
4508 const struct read_write_emulator_ops *ops)
4510 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4514 if (ops->read_write_prepare &&
4515 ops->read_write_prepare(vcpu, val, bytes))
4516 return X86EMUL_CONTINUE;
4518 vcpu->mmio_nr_fragments = 0;
4520 /* Crossing a page boundary? */
4521 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4524 now = -addr & ~PAGE_MASK;
4525 rc = emulator_read_write_onepage(addr, val, now, exception,
4528 if (rc != X86EMUL_CONTINUE)
4535 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4537 if (rc != X86EMUL_CONTINUE)
4540 if (!vcpu->mmio_nr_fragments)
4543 gpa = vcpu->mmio_fragments[0].gpa;
4545 vcpu->mmio_needed = 1;
4546 vcpu->mmio_cur_fragment = 0;
4548 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4549 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4550 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4551 vcpu->run->mmio.phys_addr = gpa;
4553 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4556 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4560 struct x86_exception *exception)
4562 return emulator_read_write(ctxt, addr, val, bytes,
4563 exception, &read_emultor);
4566 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4570 struct x86_exception *exception)
4572 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4573 exception, &write_emultor);
4576 #define CMPXCHG_TYPE(t, ptr, old, new) \
4577 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4579 #ifdef CONFIG_X86_64
4580 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4582 # define CMPXCHG64(ptr, old, new) \
4583 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4586 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4591 struct x86_exception *exception)
4593 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4599 /* guests cmpxchg8b have to be emulated atomically */
4600 if (bytes > 8 || (bytes & (bytes - 1)))
4603 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4605 if (gpa == UNMAPPED_GVA ||
4606 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4609 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4612 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4613 if (is_error_page(page))
4616 kaddr = kmap_atomic(page);
4617 kaddr += offset_in_page(gpa);
4620 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4623 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4626 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4629 exchanged = CMPXCHG64(kaddr, old, new);
4634 kunmap_atomic(kaddr);
4635 kvm_release_page_dirty(page);
4638 return X86EMUL_CMPXCHG_FAILED;
4640 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
4641 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4643 return X86EMUL_CONTINUE;
4646 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4648 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4651 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4653 /* TODO: String I/O for in kernel device */
4656 if (vcpu->arch.pio.in)
4657 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4658 vcpu->arch.pio.size, pd);
4660 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4661 vcpu->arch.pio.port, vcpu->arch.pio.size,
4666 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4667 unsigned short port, void *val,
4668 unsigned int count, bool in)
4670 vcpu->arch.pio.port = port;
4671 vcpu->arch.pio.in = in;
4672 vcpu->arch.pio.count = count;
4673 vcpu->arch.pio.size = size;
4675 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4676 vcpu->arch.pio.count = 0;
4680 vcpu->run->exit_reason = KVM_EXIT_IO;
4681 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4682 vcpu->run->io.size = size;
4683 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4684 vcpu->run->io.count = count;
4685 vcpu->run->io.port = port;
4690 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4691 int size, unsigned short port, void *val,
4694 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4697 if (vcpu->arch.pio.count)
4700 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4703 memcpy(val, vcpu->arch.pio_data, size * count);
4704 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4705 vcpu->arch.pio.count = 0;
4712 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4713 int size, unsigned short port,
4714 const void *val, unsigned int count)
4716 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4718 memcpy(vcpu->arch.pio_data, val, size * count);
4719 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4720 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4723 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4725 return kvm_x86_ops->get_segment_base(vcpu, seg);
4728 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4730 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4733 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4735 if (!need_emulate_wbinvd(vcpu))
4736 return X86EMUL_CONTINUE;
4738 if (kvm_x86_ops->has_wbinvd_exit()) {
4739 int cpu = get_cpu();
4741 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4742 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4743 wbinvd_ipi, NULL, 1);
4745 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4748 return X86EMUL_CONTINUE;
4750 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4752 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4754 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4757 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4759 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4762 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4765 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4768 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4770 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4773 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4775 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4776 unsigned long value;
4780 value = kvm_read_cr0(vcpu);
4783 value = vcpu->arch.cr2;
4786 value = kvm_read_cr3(vcpu);
4789 value = kvm_read_cr4(vcpu);
4792 value = kvm_get_cr8(vcpu);
4795 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4802 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4804 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4809 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4812 vcpu->arch.cr2 = val;
4815 res = kvm_set_cr3(vcpu, val);
4818 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4821 res = kvm_set_cr8(vcpu, val);
4824 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4831 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4833 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4836 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4838 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4841 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4843 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4846 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4848 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4851 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4853 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4856 static unsigned long emulator_get_cached_segment_base(
4857 struct x86_emulate_ctxt *ctxt, int seg)
4859 return get_segment_base(emul_to_vcpu(ctxt), seg);
4862 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4863 struct desc_struct *desc, u32 *base3,
4866 struct kvm_segment var;
4868 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4869 *selector = var.selector;
4872 memset(desc, 0, sizeof(*desc));
4878 set_desc_limit(desc, var.limit);
4879 set_desc_base(desc, (unsigned long)var.base);
4880 #ifdef CONFIG_X86_64
4882 *base3 = var.base >> 32;
4884 desc->type = var.type;
4886 desc->dpl = var.dpl;
4887 desc->p = var.present;
4888 desc->avl = var.avl;
4896 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4897 struct desc_struct *desc, u32 base3,
4900 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4901 struct kvm_segment var;
4903 var.selector = selector;
4904 var.base = get_desc_base(desc);
4905 #ifdef CONFIG_X86_64
4906 var.base |= ((u64)base3) << 32;
4908 var.limit = get_desc_limit(desc);
4910 var.limit = (var.limit << 12) | 0xfff;
4911 var.type = desc->type;
4912 var.dpl = desc->dpl;
4917 var.avl = desc->avl;
4918 var.present = desc->p;
4919 var.unusable = !var.present;
4922 kvm_set_segment(vcpu, &var, seg);
4926 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4927 u32 msr_index, u64 *pdata)
4929 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4932 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4933 u32 msr_index, u64 data)
4935 struct msr_data msr;
4938 msr.index = msr_index;
4939 msr.host_initiated = false;
4940 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4943 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4946 return kvm_pmu_check_pmc(emul_to_vcpu(ctxt), pmc);
4949 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4950 u32 pmc, u64 *pdata)
4952 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4955 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4957 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4960 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4963 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4965 * CR0.TS may reference the host fpu state, not the guest fpu state,
4966 * so it may be clear at this point.
4971 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4976 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4977 struct x86_instruction_info *info,
4978 enum x86_intercept_stage stage)
4980 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4983 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4984 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4986 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4989 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4991 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4994 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4996 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4999 static const struct x86_emulate_ops emulate_ops = {
5000 .read_gpr = emulator_read_gpr,
5001 .write_gpr = emulator_write_gpr,
5002 .read_std = kvm_read_guest_virt_system,
5003 .write_std = kvm_write_guest_virt_system,
5004 .fetch = kvm_fetch_guest_virt,
5005 .read_emulated = emulator_read_emulated,
5006 .write_emulated = emulator_write_emulated,
5007 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5008 .invlpg = emulator_invlpg,
5009 .pio_in_emulated = emulator_pio_in_emulated,
5010 .pio_out_emulated = emulator_pio_out_emulated,
5011 .get_segment = emulator_get_segment,
5012 .set_segment = emulator_set_segment,
5013 .get_cached_segment_base = emulator_get_cached_segment_base,
5014 .get_gdt = emulator_get_gdt,
5015 .get_idt = emulator_get_idt,
5016 .set_gdt = emulator_set_gdt,
5017 .set_idt = emulator_set_idt,
5018 .get_cr = emulator_get_cr,
5019 .set_cr = emulator_set_cr,
5020 .cpl = emulator_get_cpl,
5021 .get_dr = emulator_get_dr,
5022 .set_dr = emulator_set_dr,
5023 .set_msr = emulator_set_msr,
5024 .get_msr = emulator_get_msr,
5025 .check_pmc = emulator_check_pmc,
5026 .read_pmc = emulator_read_pmc,
5027 .halt = emulator_halt,
5028 .wbinvd = emulator_wbinvd,
5029 .fix_hypercall = emulator_fix_hypercall,
5030 .get_fpu = emulator_get_fpu,
5031 .put_fpu = emulator_put_fpu,
5032 .intercept = emulator_intercept,
5033 .get_cpuid = emulator_get_cpuid,
5036 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5038 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5040 * an sti; sti; sequence only disable interrupts for the first
5041 * instruction. So, if the last instruction, be it emulated or
5042 * not, left the system with the INT_STI flag enabled, it
5043 * means that the last instruction is an sti. We should not
5044 * leave the flag on in this case. The same goes for mov ss
5046 if (int_shadow & mask)
5048 if (unlikely(int_shadow || mask)) {
5049 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5051 kvm_make_request(KVM_REQ_EVENT, vcpu);
5055 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5057 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5058 if (ctxt->exception.vector == PF_VECTOR)
5059 return kvm_propagate_fault(vcpu, &ctxt->exception);
5061 if (ctxt->exception.error_code_valid)
5062 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5063 ctxt->exception.error_code);
5065 kvm_queue_exception(vcpu, ctxt->exception.vector);
5069 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5071 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5074 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5076 ctxt->eflags = kvm_get_rflags(vcpu);
5077 ctxt->eip = kvm_rip_read(vcpu);
5078 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5079 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5080 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5081 cs_db ? X86EMUL_MODE_PROT32 :
5082 X86EMUL_MODE_PROT16;
5083 ctxt->guest_mode = is_guest_mode(vcpu);
5085 init_decode_cache(ctxt);
5086 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5089 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5091 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5094 init_emulate_ctxt(vcpu);
5098 ctxt->_eip = ctxt->eip + inc_eip;
5099 ret = emulate_int_real(ctxt, irq);
5101 if (ret != X86EMUL_CONTINUE)
5102 return EMULATE_FAIL;
5104 ctxt->eip = ctxt->_eip;
5105 kvm_rip_write(vcpu, ctxt->eip);
5106 kvm_set_rflags(vcpu, ctxt->eflags);
5108 if (irq == NMI_VECTOR)
5109 vcpu->arch.nmi_pending = 0;
5111 vcpu->arch.interrupt.pending = false;
5113 return EMULATE_DONE;
5115 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5117 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5119 int r = EMULATE_DONE;
5121 ++vcpu->stat.insn_emulation_fail;
5122 trace_kvm_emulate_insn_failed(vcpu);
5123 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5124 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5125 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5126 vcpu->run->internal.ndata = 0;
5129 kvm_queue_exception(vcpu, UD_VECTOR);
5134 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5135 bool write_fault_to_shadow_pgtable,
5141 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5144 if (!vcpu->arch.mmu.direct_map) {
5146 * Write permission should be allowed since only
5147 * write access need to be emulated.
5149 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5152 * If the mapping is invalid in guest, let cpu retry
5153 * it to generate fault.
5155 if (gpa == UNMAPPED_GVA)
5160 * Do not retry the unhandleable instruction if it faults on the
5161 * readonly host memory, otherwise it will goto a infinite loop:
5162 * retry instruction -> write #PF -> emulation fail -> retry
5163 * instruction -> ...
5165 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5168 * If the instruction failed on the error pfn, it can not be fixed,
5169 * report the error to userspace.
5171 if (is_error_noslot_pfn(pfn))
5174 kvm_release_pfn_clean(pfn);
5176 /* The instructions are well-emulated on direct mmu. */
5177 if (vcpu->arch.mmu.direct_map) {
5178 unsigned int indirect_shadow_pages;
5180 spin_lock(&vcpu->kvm->mmu_lock);
5181 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5182 spin_unlock(&vcpu->kvm->mmu_lock);
5184 if (indirect_shadow_pages)
5185 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5191 * if emulation was due to access to shadowed page table
5192 * and it failed try to unshadow page and re-enter the
5193 * guest to let CPU execute the instruction.
5195 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5198 * If the access faults on its page table, it can not
5199 * be fixed by unprotecting shadow page and it should
5200 * be reported to userspace.
5202 return !write_fault_to_shadow_pgtable;
5205 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5206 unsigned long cr2, int emulation_type)
5208 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5209 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5211 last_retry_eip = vcpu->arch.last_retry_eip;
5212 last_retry_addr = vcpu->arch.last_retry_addr;
5215 * If the emulation is caused by #PF and it is non-page_table
5216 * writing instruction, it means the VM-EXIT is caused by shadow
5217 * page protected, we can zap the shadow page and retry this
5218 * instruction directly.
5220 * Note: if the guest uses a non-page-table modifying instruction
5221 * on the PDE that points to the instruction, then we will unmap
5222 * the instruction and go to an infinite loop. So, we cache the
5223 * last retried eip and the last fault address, if we meet the eip
5224 * and the address again, we can break out of the potential infinite
5227 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5229 if (!(emulation_type & EMULTYPE_RETRY))
5232 if (x86_page_table_writing_insn(ctxt))
5235 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5238 vcpu->arch.last_retry_eip = ctxt->eip;
5239 vcpu->arch.last_retry_addr = cr2;
5241 if (!vcpu->arch.mmu.direct_map)
5242 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5244 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5249 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5250 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5252 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5261 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5262 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5267 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5269 struct kvm_run *kvm_run = vcpu->run;
5272 * rflags is the old, "raw" value of the flags. The new value has
5273 * not been saved yet.
5275 * This is correct even for TF set by the guest, because "the
5276 * processor will not generate this exception after the instruction
5277 * that sets the TF flag".
5279 if (unlikely(rflags & X86_EFLAGS_TF)) {
5280 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5281 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5283 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5284 kvm_run->debug.arch.exception = DB_VECTOR;
5285 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5286 *r = EMULATE_USER_EXIT;
5288 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5290 * "Certain debug exceptions may clear bit 0-3. The
5291 * remaining contents of the DR6 register are never
5292 * cleared by the processor".
5294 vcpu->arch.dr6 &= ~15;
5295 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5296 kvm_queue_exception(vcpu, DB_VECTOR);
5301 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5303 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5304 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5305 struct kvm_run *kvm_run = vcpu->run;
5306 unsigned long eip = kvm_get_linear_rip(vcpu);
5307 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5308 vcpu->arch.guest_debug_dr7,
5312 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5313 kvm_run->debug.arch.pc = eip;
5314 kvm_run->debug.arch.exception = DB_VECTOR;
5315 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5316 *r = EMULATE_USER_EXIT;
5321 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5322 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5323 unsigned long eip = kvm_get_linear_rip(vcpu);
5324 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5329 vcpu->arch.dr6 &= ~15;
5330 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5331 kvm_queue_exception(vcpu, DB_VECTOR);
5340 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5347 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5348 bool writeback = true;
5349 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5352 * Clear write_fault_to_shadow_pgtable here to ensure it is
5355 vcpu->arch.write_fault_to_shadow_pgtable = false;
5356 kvm_clear_exception_queue(vcpu);
5358 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5359 init_emulate_ctxt(vcpu);
5362 * We will reenter on the same instruction since
5363 * we do not set complete_userspace_io. This does not
5364 * handle watchpoints yet, those would be handled in
5367 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5370 ctxt->interruptibility = 0;
5371 ctxt->have_exception = false;
5372 ctxt->exception.vector = -1;
5373 ctxt->perm_ok = false;
5375 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5377 r = x86_decode_insn(ctxt, insn, insn_len);
5379 trace_kvm_emulate_insn_start(vcpu);
5380 ++vcpu->stat.insn_emulation;
5381 if (r != EMULATION_OK) {
5382 if (emulation_type & EMULTYPE_TRAP_UD)
5383 return EMULATE_FAIL;
5384 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5386 return EMULATE_DONE;
5387 if (emulation_type & EMULTYPE_SKIP)
5388 return EMULATE_FAIL;
5389 return handle_emulation_failure(vcpu);
5393 if (emulation_type & EMULTYPE_SKIP) {
5394 kvm_rip_write(vcpu, ctxt->_eip);
5395 if (ctxt->eflags & X86_EFLAGS_RF)
5396 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5397 return EMULATE_DONE;
5400 if (retry_instruction(ctxt, cr2, emulation_type))
5401 return EMULATE_DONE;
5403 /* this is needed for vmware backdoor interface to work since it
5404 changes registers values during IO operation */
5405 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5406 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5407 emulator_invalidate_register_cache(ctxt);
5411 r = x86_emulate_insn(ctxt);
5413 if (r == EMULATION_INTERCEPTED)
5414 return EMULATE_DONE;
5416 if (r == EMULATION_FAILED) {
5417 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5419 return EMULATE_DONE;
5421 return handle_emulation_failure(vcpu);
5424 if (ctxt->have_exception) {
5426 if (inject_emulated_exception(vcpu))
5428 } else if (vcpu->arch.pio.count) {
5429 if (!vcpu->arch.pio.in) {
5430 /* FIXME: return into emulator if single-stepping. */
5431 vcpu->arch.pio.count = 0;
5434 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5436 r = EMULATE_USER_EXIT;
5437 } else if (vcpu->mmio_needed) {
5438 if (!vcpu->mmio_is_write)
5440 r = EMULATE_USER_EXIT;
5441 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5442 } else if (r == EMULATION_RESTART)
5448 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5449 toggle_interruptibility(vcpu, ctxt->interruptibility);
5450 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5451 kvm_rip_write(vcpu, ctxt->eip);
5452 if (r == EMULATE_DONE)
5453 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5454 if (!ctxt->have_exception ||
5455 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5456 __kvm_set_rflags(vcpu, ctxt->eflags);
5459 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5460 * do nothing, and it will be requested again as soon as
5461 * the shadow expires. But we still need to check here,
5462 * because POPF has no interrupt shadow.
5464 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5465 kvm_make_request(KVM_REQ_EVENT, vcpu);
5467 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5471 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5473 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5475 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5476 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5477 size, port, &val, 1);
5478 /* do not return to emulator after return from userspace */
5479 vcpu->arch.pio.count = 0;
5482 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5484 static void tsc_bad(void *info)
5486 __this_cpu_write(cpu_tsc_khz, 0);
5489 static void tsc_khz_changed(void *data)
5491 struct cpufreq_freqs *freq = data;
5492 unsigned long khz = 0;
5496 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5497 khz = cpufreq_quick_get(raw_smp_processor_id());
5500 __this_cpu_write(cpu_tsc_khz, khz);
5503 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5506 struct cpufreq_freqs *freq = data;
5508 struct kvm_vcpu *vcpu;
5509 int i, send_ipi = 0;
5512 * We allow guests to temporarily run on slowing clocks,
5513 * provided we notify them after, or to run on accelerating
5514 * clocks, provided we notify them before. Thus time never
5517 * However, we have a problem. We can't atomically update
5518 * the frequency of a given CPU from this function; it is
5519 * merely a notifier, which can be called from any CPU.
5520 * Changing the TSC frequency at arbitrary points in time
5521 * requires a recomputation of local variables related to
5522 * the TSC for each VCPU. We must flag these local variables
5523 * to be updated and be sure the update takes place with the
5524 * new frequency before any guests proceed.
5526 * Unfortunately, the combination of hotplug CPU and frequency
5527 * change creates an intractable locking scenario; the order
5528 * of when these callouts happen is undefined with respect to
5529 * CPU hotplug, and they can race with each other. As such,
5530 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5531 * undefined; you can actually have a CPU frequency change take
5532 * place in between the computation of X and the setting of the
5533 * variable. To protect against this problem, all updates of
5534 * the per_cpu tsc_khz variable are done in an interrupt
5535 * protected IPI, and all callers wishing to update the value
5536 * must wait for a synchronous IPI to complete (which is trivial
5537 * if the caller is on the CPU already). This establishes the
5538 * necessary total order on variable updates.
5540 * Note that because a guest time update may take place
5541 * anytime after the setting of the VCPU's request bit, the
5542 * correct TSC value must be set before the request. However,
5543 * to ensure the update actually makes it to any guest which
5544 * starts running in hardware virtualization between the set
5545 * and the acquisition of the spinlock, we must also ping the
5546 * CPU after setting the request bit.
5550 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5552 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5555 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5557 spin_lock(&kvm_lock);
5558 list_for_each_entry(kvm, &vm_list, vm_list) {
5559 kvm_for_each_vcpu(i, vcpu, kvm) {
5560 if (vcpu->cpu != freq->cpu)
5562 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5563 if (vcpu->cpu != smp_processor_id())
5567 spin_unlock(&kvm_lock);
5569 if (freq->old < freq->new && send_ipi) {
5571 * We upscale the frequency. Must make the guest
5572 * doesn't see old kvmclock values while running with
5573 * the new frequency, otherwise we risk the guest sees
5574 * time go backwards.
5576 * In case we update the frequency for another cpu
5577 * (which might be in guest context) send an interrupt
5578 * to kick the cpu out of guest context. Next time
5579 * guest context is entered kvmclock will be updated,
5580 * so the guest will not see stale values.
5582 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5587 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5588 .notifier_call = kvmclock_cpufreq_notifier
5591 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5592 unsigned long action, void *hcpu)
5594 unsigned int cpu = (unsigned long)hcpu;
5598 case CPU_DOWN_FAILED:
5599 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5601 case CPU_DOWN_PREPARE:
5602 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5608 static struct notifier_block kvmclock_cpu_notifier_block = {
5609 .notifier_call = kvmclock_cpu_notifier,
5610 .priority = -INT_MAX
5613 static void kvm_timer_init(void)
5617 max_tsc_khz = tsc_khz;
5619 cpu_notifier_register_begin();
5620 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5621 #ifdef CONFIG_CPU_FREQ
5622 struct cpufreq_policy policy;
5623 memset(&policy, 0, sizeof(policy));
5625 cpufreq_get_policy(&policy, cpu);
5626 if (policy.cpuinfo.max_freq)
5627 max_tsc_khz = policy.cpuinfo.max_freq;
5630 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5631 CPUFREQ_TRANSITION_NOTIFIER);
5633 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5634 for_each_online_cpu(cpu)
5635 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5637 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5638 cpu_notifier_register_done();
5642 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5644 int kvm_is_in_guest(void)
5646 return __this_cpu_read(current_vcpu) != NULL;
5649 static int kvm_is_user_mode(void)
5653 if (__this_cpu_read(current_vcpu))
5654 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5656 return user_mode != 0;
5659 static unsigned long kvm_get_guest_ip(void)
5661 unsigned long ip = 0;
5663 if (__this_cpu_read(current_vcpu))
5664 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5669 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5670 .is_in_guest = kvm_is_in_guest,
5671 .is_user_mode = kvm_is_user_mode,
5672 .get_guest_ip = kvm_get_guest_ip,
5675 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5677 __this_cpu_write(current_vcpu, vcpu);
5679 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5681 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5683 __this_cpu_write(current_vcpu, NULL);
5685 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5687 static void kvm_set_mmio_spte_mask(void)
5690 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5693 * Set the reserved bits and the present bit of an paging-structure
5694 * entry to generate page fault with PFER.RSV = 1.
5696 /* Mask the reserved physical address bits. */
5697 mask = rsvd_bits(maxphyaddr, 51);
5699 /* Bit 62 is always reserved for 32bit host. */
5700 mask |= 0x3ull << 62;
5702 /* Set the present bit. */
5705 #ifdef CONFIG_X86_64
5707 * If reserved bit is not supported, clear the present bit to disable
5710 if (maxphyaddr == 52)
5714 kvm_mmu_set_mmio_spte_mask(mask);
5717 #ifdef CONFIG_X86_64
5718 static void pvclock_gtod_update_fn(struct work_struct *work)
5722 struct kvm_vcpu *vcpu;
5725 spin_lock(&kvm_lock);
5726 list_for_each_entry(kvm, &vm_list, vm_list)
5727 kvm_for_each_vcpu(i, vcpu, kvm)
5728 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5729 atomic_set(&kvm_guest_has_master_clock, 0);
5730 spin_unlock(&kvm_lock);
5733 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5736 * Notification about pvclock gtod data update.
5738 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5741 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5742 struct timekeeper *tk = priv;
5744 update_pvclock_gtod(tk);
5746 /* disable master clock if host does not trust, or does not
5747 * use, TSC clocksource
5749 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5750 atomic_read(&kvm_guest_has_master_clock) != 0)
5751 queue_work(system_long_wq, &pvclock_gtod_work);
5756 static struct notifier_block pvclock_gtod_notifier = {
5757 .notifier_call = pvclock_gtod_notify,
5761 int kvm_arch_init(void *opaque)
5764 struct kvm_x86_ops *ops = opaque;
5767 printk(KERN_ERR "kvm: already loaded the other module\n");
5772 if (!ops->cpu_has_kvm_support()) {
5773 printk(KERN_ERR "kvm: no hardware support\n");
5777 if (ops->disabled_by_bios()) {
5778 printk(KERN_ERR "kvm: disabled by bios\n");
5784 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5786 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5790 r = kvm_mmu_module_init();
5792 goto out_free_percpu;
5794 kvm_set_mmio_spte_mask();
5797 kvm_init_msr_list();
5799 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5800 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5804 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5807 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5810 #ifdef CONFIG_X86_64
5811 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5817 free_percpu(shared_msrs);
5822 void kvm_arch_exit(void)
5824 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5826 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5827 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5828 CPUFREQ_TRANSITION_NOTIFIER);
5829 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5830 #ifdef CONFIG_X86_64
5831 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5834 kvm_mmu_module_exit();
5835 free_percpu(shared_msrs);
5838 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5840 ++vcpu->stat.halt_exits;
5841 if (irqchip_in_kernel(vcpu->kvm)) {
5842 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5845 vcpu->run->exit_reason = KVM_EXIT_HLT;
5849 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5851 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5853 u64 param, ingpa, outgpa, ret;
5854 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5855 bool fast, longmode;
5858 * hypercall generates UD from non zero cpl and real mode
5861 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5862 kvm_queue_exception(vcpu, UD_VECTOR);
5866 longmode = is_64_bit_mode(vcpu);
5869 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5870 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5871 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5872 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5873 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5874 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5876 #ifdef CONFIG_X86_64
5878 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5879 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5880 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5884 code = param & 0xffff;
5885 fast = (param >> 16) & 0x1;
5886 rep_cnt = (param >> 32) & 0xfff;
5887 rep_idx = (param >> 48) & 0xfff;
5889 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5892 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5893 kvm_vcpu_on_spin(vcpu);
5896 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5900 ret = res | (((u64)rep_done & 0xfff) << 32);
5902 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5904 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5905 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5912 * kvm_pv_kick_cpu_op: Kick a vcpu.
5914 * @apicid - apicid of vcpu to be kicked.
5916 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5918 struct kvm_lapic_irq lapic_irq;
5920 lapic_irq.shorthand = 0;
5921 lapic_irq.dest_mode = 0;
5922 lapic_irq.dest_id = apicid;
5924 lapic_irq.delivery_mode = APIC_DM_REMRD;
5925 kvm_irq_delivery_to_apic(kvm, 0, &lapic_irq, NULL);
5928 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5930 unsigned long nr, a0, a1, a2, a3, ret;
5931 int op_64_bit, r = 1;
5933 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5934 return kvm_hv_hypercall(vcpu);
5936 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5937 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5938 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5939 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5940 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5942 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5944 op_64_bit = is_64_bit_mode(vcpu);
5953 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5959 case KVM_HC_VAPIC_POLL_IRQ:
5962 case KVM_HC_KICK_CPU:
5963 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5973 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5974 ++vcpu->stat.hypercalls;
5977 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5979 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5981 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5982 char instruction[3];
5983 unsigned long rip = kvm_rip_read(vcpu);
5985 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5987 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5991 * Check if userspace requested an interrupt window, and that the
5992 * interrupt window is open.
5994 * No need to exit to userspace if we already have an interrupt queued.
5996 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5998 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5999 vcpu->run->request_interrupt_window &&
6000 kvm_arch_interrupt_allowed(vcpu));
6003 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6005 struct kvm_run *kvm_run = vcpu->run;
6007 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6008 kvm_run->cr8 = kvm_get_cr8(vcpu);
6009 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6010 if (irqchip_in_kernel(vcpu->kvm))
6011 kvm_run->ready_for_interrupt_injection = 1;
6013 kvm_run->ready_for_interrupt_injection =
6014 kvm_arch_interrupt_allowed(vcpu) &&
6015 !kvm_cpu_has_interrupt(vcpu) &&
6016 !kvm_event_needs_reinjection(vcpu);
6019 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6023 if (!kvm_x86_ops->update_cr8_intercept)
6026 if (!vcpu->arch.apic)
6029 if (!vcpu->arch.apic->vapic_addr)
6030 max_irr = kvm_lapic_find_highest_irr(vcpu);
6037 tpr = kvm_lapic_get_cr8(vcpu);
6039 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6042 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6046 /* try to reinject previous events if any */
6047 if (vcpu->arch.exception.pending) {
6048 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6049 vcpu->arch.exception.has_error_code,
6050 vcpu->arch.exception.error_code);
6052 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6053 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6056 if (vcpu->arch.exception.nr == DB_VECTOR &&
6057 (vcpu->arch.dr7 & DR7_GD)) {
6058 vcpu->arch.dr7 &= ~DR7_GD;
6059 kvm_update_dr7(vcpu);
6062 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6063 vcpu->arch.exception.has_error_code,
6064 vcpu->arch.exception.error_code,
6065 vcpu->arch.exception.reinject);
6069 if (vcpu->arch.nmi_injected) {
6070 kvm_x86_ops->set_nmi(vcpu);
6074 if (vcpu->arch.interrupt.pending) {
6075 kvm_x86_ops->set_irq(vcpu);
6079 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6080 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6085 /* try to inject new event if pending */
6086 if (vcpu->arch.nmi_pending) {
6087 if (kvm_x86_ops->nmi_allowed(vcpu)) {
6088 --vcpu->arch.nmi_pending;
6089 vcpu->arch.nmi_injected = true;
6090 kvm_x86_ops->set_nmi(vcpu);
6092 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6094 * Because interrupts can be injected asynchronously, we are
6095 * calling check_nested_events again here to avoid a race condition.
6096 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6097 * proposal and current concerns. Perhaps we should be setting
6098 * KVM_REQ_EVENT only on certain events and not unconditionally?
6100 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6101 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6105 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6106 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6108 kvm_x86_ops->set_irq(vcpu);
6114 static void process_nmi(struct kvm_vcpu *vcpu)
6119 * x86 is limited to one NMI running, and one NMI pending after it.
6120 * If an NMI is already in progress, limit further NMIs to just one.
6121 * Otherwise, allow two (and we'll inject the first one immediately).
6123 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6126 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6127 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6128 kvm_make_request(KVM_REQ_EVENT, vcpu);
6131 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6133 u64 eoi_exit_bitmap[4];
6136 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6139 memset(eoi_exit_bitmap, 0, 32);
6142 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6143 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6144 kvm_apic_update_tmr(vcpu, tmr);
6147 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6149 ++vcpu->stat.tlb_flush;
6150 kvm_x86_ops->tlb_flush(vcpu);
6153 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6155 struct page *page = NULL;
6157 if (!irqchip_in_kernel(vcpu->kvm))
6160 if (!kvm_x86_ops->set_apic_access_page_addr)
6163 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6164 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6167 * Do not pin apic access page in memory, the MMU notifier
6168 * will call us again if it is migrated or swapped out.
6172 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6174 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6175 unsigned long address)
6178 * The physical address of apic access page is stored in the VMCS.
6179 * Update it when it becomes invalid.
6181 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6182 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6186 * Returns 1 to let __vcpu_run() continue the guest execution loop without
6187 * exiting to the userspace. Otherwise, the value will be returned to the
6190 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6193 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6194 vcpu->run->request_interrupt_window;
6195 bool req_immediate_exit = false;
6197 if (vcpu->requests) {
6198 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6199 kvm_mmu_unload(vcpu);
6200 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6201 __kvm_migrate_timers(vcpu);
6202 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6203 kvm_gen_update_masterclock(vcpu->kvm);
6204 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6205 kvm_gen_kvmclock_update(vcpu);
6206 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6207 r = kvm_guest_time_update(vcpu);
6211 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6212 kvm_mmu_sync_roots(vcpu);
6213 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6214 kvm_vcpu_flush_tlb(vcpu);
6215 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6216 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6220 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6221 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6225 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6226 vcpu->fpu_active = 0;
6227 kvm_x86_ops->fpu_deactivate(vcpu);
6229 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6230 /* Page is swapped out. Do synthetic halt */
6231 vcpu->arch.apf.halted = true;
6235 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6236 record_steal_time(vcpu);
6237 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6239 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6240 kvm_handle_pmu_event(vcpu);
6241 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6242 kvm_deliver_pmi(vcpu);
6243 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6244 vcpu_scan_ioapic(vcpu);
6245 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6246 kvm_vcpu_reload_apic_access_page(vcpu);
6249 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6250 kvm_apic_accept_events(vcpu);
6251 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6256 if (inject_pending_event(vcpu, req_int_win) != 0)
6257 req_immediate_exit = true;
6258 /* enable NMI/IRQ window open exits if needed */
6259 else if (vcpu->arch.nmi_pending)
6260 kvm_x86_ops->enable_nmi_window(vcpu);
6261 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6262 kvm_x86_ops->enable_irq_window(vcpu);
6264 if (kvm_lapic_enabled(vcpu)) {
6266 * Update architecture specific hints for APIC
6267 * virtual interrupt delivery.
6269 if (kvm_x86_ops->hwapic_irr_update)
6270 kvm_x86_ops->hwapic_irr_update(vcpu,
6271 kvm_lapic_find_highest_irr(vcpu));
6272 update_cr8_intercept(vcpu);
6273 kvm_lapic_sync_to_vapic(vcpu);
6277 r = kvm_mmu_reload(vcpu);
6279 goto cancel_injection;
6284 kvm_x86_ops->prepare_guest_switch(vcpu);
6285 if (vcpu->fpu_active)
6286 kvm_load_guest_fpu(vcpu);
6287 kvm_load_guest_xcr0(vcpu);
6289 vcpu->mode = IN_GUEST_MODE;
6291 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6293 /* We should set ->mode before check ->requests,
6294 * see the comment in make_all_cpus_request.
6296 smp_mb__after_srcu_read_unlock();
6298 local_irq_disable();
6300 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6301 || need_resched() || signal_pending(current)) {
6302 vcpu->mode = OUTSIDE_GUEST_MODE;
6306 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6308 goto cancel_injection;
6311 if (req_immediate_exit)
6312 smp_send_reschedule(vcpu->cpu);
6316 if (unlikely(vcpu->arch.switch_db_regs)) {
6318 set_debugreg(vcpu->arch.eff_db[0], 0);
6319 set_debugreg(vcpu->arch.eff_db[1], 1);
6320 set_debugreg(vcpu->arch.eff_db[2], 2);
6321 set_debugreg(vcpu->arch.eff_db[3], 3);
6322 set_debugreg(vcpu->arch.dr6, 6);
6325 trace_kvm_entry(vcpu->vcpu_id);
6326 wait_lapic_expire(vcpu);
6327 kvm_x86_ops->run(vcpu);
6330 * Do this here before restoring debug registers on the host. And
6331 * since we do this before handling the vmexit, a DR access vmexit
6332 * can (a) read the correct value of the debug registers, (b) set
6333 * KVM_DEBUGREG_WONT_EXIT again.
6335 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6338 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6339 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6340 for (i = 0; i < KVM_NR_DB_REGS; i++)
6341 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6345 * If the guest has used debug registers, at least dr7
6346 * will be disabled while returning to the host.
6347 * If we don't have active breakpoints in the host, we don't
6348 * care about the messed up debug address registers. But if
6349 * we have some of them active, restore the old state.
6351 if (hw_breakpoint_active())
6352 hw_breakpoint_restore();
6354 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6357 vcpu->mode = OUTSIDE_GUEST_MODE;
6360 /* Interrupt is enabled by handle_external_intr() */
6361 kvm_x86_ops->handle_external_intr(vcpu);
6366 * We must have an instruction between local_irq_enable() and
6367 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6368 * the interrupt shadow. The stat.exits increment will do nicely.
6369 * But we need to prevent reordering, hence this barrier():
6377 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6380 * Profile KVM exit RIPs:
6382 if (unlikely(prof_on == KVM_PROFILING)) {
6383 unsigned long rip = kvm_rip_read(vcpu);
6384 profile_hit(KVM_PROFILING, (void *)rip);
6387 if (unlikely(vcpu->arch.tsc_always_catchup))
6388 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6390 if (vcpu->arch.apic_attention)
6391 kvm_lapic_sync_from_vapic(vcpu);
6393 r = kvm_x86_ops->handle_exit(vcpu);
6397 kvm_x86_ops->cancel_injection(vcpu);
6398 if (unlikely(vcpu->arch.apic_attention))
6399 kvm_lapic_sync_from_vapic(vcpu);
6405 static int __vcpu_run(struct kvm_vcpu *vcpu)
6408 struct kvm *kvm = vcpu->kvm;
6410 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6414 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6415 !vcpu->arch.apf.halted)
6416 r = vcpu_enter_guest(vcpu);
6418 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6419 kvm_vcpu_block(vcpu);
6420 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6421 if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
6422 kvm_apic_accept_events(vcpu);
6423 switch(vcpu->arch.mp_state) {
6424 case KVM_MP_STATE_HALTED:
6425 vcpu->arch.pv.pv_unhalted = false;
6426 vcpu->arch.mp_state =
6427 KVM_MP_STATE_RUNNABLE;
6428 case KVM_MP_STATE_RUNNABLE:
6429 vcpu->arch.apf.halted = false;
6431 case KVM_MP_STATE_INIT_RECEIVED:
6443 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6444 if (kvm_cpu_has_pending_timer(vcpu))
6445 kvm_inject_pending_timer_irqs(vcpu);
6447 if (dm_request_for_irq_injection(vcpu)) {
6449 vcpu->run->exit_reason = KVM_EXIT_INTR;
6450 ++vcpu->stat.request_irq_exits;
6453 kvm_check_async_pf_completion(vcpu);
6455 if (signal_pending(current)) {
6457 vcpu->run->exit_reason = KVM_EXIT_INTR;
6458 ++vcpu->stat.signal_exits;
6460 if (need_resched()) {
6461 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6463 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6467 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6472 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6475 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6476 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6477 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6478 if (r != EMULATE_DONE)
6483 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6485 BUG_ON(!vcpu->arch.pio.count);
6487 return complete_emulated_io(vcpu);
6491 * Implements the following, as a state machine:
6495 * for each mmio piece in the fragment
6503 * for each mmio piece in the fragment
6508 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6510 struct kvm_run *run = vcpu->run;
6511 struct kvm_mmio_fragment *frag;
6514 BUG_ON(!vcpu->mmio_needed);
6516 /* Complete previous fragment */
6517 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6518 len = min(8u, frag->len);
6519 if (!vcpu->mmio_is_write)
6520 memcpy(frag->data, run->mmio.data, len);
6522 if (frag->len <= 8) {
6523 /* Switch to the next fragment. */
6525 vcpu->mmio_cur_fragment++;
6527 /* Go forward to the next mmio piece. */
6533 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6534 vcpu->mmio_needed = 0;
6536 /* FIXME: return into emulator if single-stepping. */
6537 if (vcpu->mmio_is_write)
6539 vcpu->mmio_read_completed = 1;
6540 return complete_emulated_io(vcpu);
6543 run->exit_reason = KVM_EXIT_MMIO;
6544 run->mmio.phys_addr = frag->gpa;
6545 if (vcpu->mmio_is_write)
6546 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6547 run->mmio.len = min(8u, frag->len);
6548 run->mmio.is_write = vcpu->mmio_is_write;
6549 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6554 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6559 if (!tsk_used_math(current) && init_fpu(current))
6562 if (vcpu->sigset_active)
6563 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6565 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6566 kvm_vcpu_block(vcpu);
6567 kvm_apic_accept_events(vcpu);
6568 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6573 /* re-sync apic's tpr */
6574 if (!irqchip_in_kernel(vcpu->kvm)) {
6575 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6581 if (unlikely(vcpu->arch.complete_userspace_io)) {
6582 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6583 vcpu->arch.complete_userspace_io = NULL;
6588 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6590 r = __vcpu_run(vcpu);
6593 post_kvm_run_save(vcpu);
6594 if (vcpu->sigset_active)
6595 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6600 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6602 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6604 * We are here if userspace calls get_regs() in the middle of
6605 * instruction emulation. Registers state needs to be copied
6606 * back from emulation context to vcpu. Userspace shouldn't do
6607 * that usually, but some bad designed PV devices (vmware
6608 * backdoor interface) need this to work
6610 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6611 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6613 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6614 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6615 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6616 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6617 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6618 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6619 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6620 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6621 #ifdef CONFIG_X86_64
6622 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6623 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6624 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6625 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6626 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6627 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6628 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6629 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6632 regs->rip = kvm_rip_read(vcpu);
6633 regs->rflags = kvm_get_rflags(vcpu);
6638 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6640 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6641 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6643 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6644 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6645 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6646 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6647 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6648 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6649 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6650 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6651 #ifdef CONFIG_X86_64
6652 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6653 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6654 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6655 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6656 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6657 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6658 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6659 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6662 kvm_rip_write(vcpu, regs->rip);
6663 kvm_set_rflags(vcpu, regs->rflags);
6665 vcpu->arch.exception.pending = false;
6667 kvm_make_request(KVM_REQ_EVENT, vcpu);
6672 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6674 struct kvm_segment cs;
6676 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6680 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6682 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6683 struct kvm_sregs *sregs)
6687 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6688 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6689 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6690 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6691 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6692 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6694 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6695 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6697 kvm_x86_ops->get_idt(vcpu, &dt);
6698 sregs->idt.limit = dt.size;
6699 sregs->idt.base = dt.address;
6700 kvm_x86_ops->get_gdt(vcpu, &dt);
6701 sregs->gdt.limit = dt.size;
6702 sregs->gdt.base = dt.address;
6704 sregs->cr0 = kvm_read_cr0(vcpu);
6705 sregs->cr2 = vcpu->arch.cr2;
6706 sregs->cr3 = kvm_read_cr3(vcpu);
6707 sregs->cr4 = kvm_read_cr4(vcpu);
6708 sregs->cr8 = kvm_get_cr8(vcpu);
6709 sregs->efer = vcpu->arch.efer;
6710 sregs->apic_base = kvm_get_apic_base(vcpu);
6712 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6714 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6715 set_bit(vcpu->arch.interrupt.nr,
6716 (unsigned long *)sregs->interrupt_bitmap);
6721 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6722 struct kvm_mp_state *mp_state)
6724 kvm_apic_accept_events(vcpu);
6725 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6726 vcpu->arch.pv.pv_unhalted)
6727 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6729 mp_state->mp_state = vcpu->arch.mp_state;
6734 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6735 struct kvm_mp_state *mp_state)
6737 if (!kvm_vcpu_has_lapic(vcpu) &&
6738 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6741 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6742 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6743 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6745 vcpu->arch.mp_state = mp_state->mp_state;
6746 kvm_make_request(KVM_REQ_EVENT, vcpu);
6750 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6751 int reason, bool has_error_code, u32 error_code)
6753 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6756 init_emulate_ctxt(vcpu);
6758 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6759 has_error_code, error_code);
6762 return EMULATE_FAIL;
6764 kvm_rip_write(vcpu, ctxt->eip);
6765 kvm_set_rflags(vcpu, ctxt->eflags);
6766 kvm_make_request(KVM_REQ_EVENT, vcpu);
6767 return EMULATE_DONE;
6769 EXPORT_SYMBOL_GPL(kvm_task_switch);
6771 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6772 struct kvm_sregs *sregs)
6774 struct msr_data apic_base_msr;
6775 int mmu_reset_needed = 0;
6776 int pending_vec, max_bits, idx;
6779 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6782 dt.size = sregs->idt.limit;
6783 dt.address = sregs->idt.base;
6784 kvm_x86_ops->set_idt(vcpu, &dt);
6785 dt.size = sregs->gdt.limit;
6786 dt.address = sregs->gdt.base;
6787 kvm_x86_ops->set_gdt(vcpu, &dt);
6789 vcpu->arch.cr2 = sregs->cr2;
6790 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6791 vcpu->arch.cr3 = sregs->cr3;
6792 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6794 kvm_set_cr8(vcpu, sregs->cr8);
6796 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6797 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6798 apic_base_msr.data = sregs->apic_base;
6799 apic_base_msr.host_initiated = true;
6800 kvm_set_apic_base(vcpu, &apic_base_msr);
6802 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6803 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6804 vcpu->arch.cr0 = sregs->cr0;
6806 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6807 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6808 if (sregs->cr4 & X86_CR4_OSXSAVE)
6809 kvm_update_cpuid(vcpu);
6811 idx = srcu_read_lock(&vcpu->kvm->srcu);
6812 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6813 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6814 mmu_reset_needed = 1;
6816 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6818 if (mmu_reset_needed)
6819 kvm_mmu_reset_context(vcpu);
6821 max_bits = KVM_NR_INTERRUPTS;
6822 pending_vec = find_first_bit(
6823 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6824 if (pending_vec < max_bits) {
6825 kvm_queue_interrupt(vcpu, pending_vec, false);
6826 pr_debug("Set back pending irq %d\n", pending_vec);
6829 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6830 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6831 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6832 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6833 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6834 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6836 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6837 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6839 update_cr8_intercept(vcpu);
6841 /* Older userspace won't unhalt the vcpu on reset. */
6842 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6843 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6845 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6847 kvm_make_request(KVM_REQ_EVENT, vcpu);
6852 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6853 struct kvm_guest_debug *dbg)
6855 unsigned long rflags;
6858 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6860 if (vcpu->arch.exception.pending)
6862 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6863 kvm_queue_exception(vcpu, DB_VECTOR);
6865 kvm_queue_exception(vcpu, BP_VECTOR);
6869 * Read rflags as long as potentially injected trace flags are still
6872 rflags = kvm_get_rflags(vcpu);
6874 vcpu->guest_debug = dbg->control;
6875 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6876 vcpu->guest_debug = 0;
6878 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6879 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6880 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6881 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6883 for (i = 0; i < KVM_NR_DB_REGS; i++)
6884 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6886 kvm_update_dr7(vcpu);
6888 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6889 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6890 get_segment_base(vcpu, VCPU_SREG_CS);
6893 * Trigger an rflags update that will inject or remove the trace
6896 kvm_set_rflags(vcpu, rflags);
6898 kvm_x86_ops->update_db_bp_intercept(vcpu);
6908 * Translate a guest virtual address to a guest physical address.
6910 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6911 struct kvm_translation *tr)
6913 unsigned long vaddr = tr->linear_address;
6917 idx = srcu_read_lock(&vcpu->kvm->srcu);
6918 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6919 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6920 tr->physical_address = gpa;
6921 tr->valid = gpa != UNMAPPED_GVA;
6928 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6930 struct i387_fxsave_struct *fxsave =
6931 &vcpu->arch.guest_fpu.state->fxsave;
6933 memcpy(fpu->fpr, fxsave->st_space, 128);
6934 fpu->fcw = fxsave->cwd;
6935 fpu->fsw = fxsave->swd;
6936 fpu->ftwx = fxsave->twd;
6937 fpu->last_opcode = fxsave->fop;
6938 fpu->last_ip = fxsave->rip;
6939 fpu->last_dp = fxsave->rdp;
6940 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6945 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6947 struct i387_fxsave_struct *fxsave =
6948 &vcpu->arch.guest_fpu.state->fxsave;
6950 memcpy(fxsave->st_space, fpu->fpr, 128);
6951 fxsave->cwd = fpu->fcw;
6952 fxsave->swd = fpu->fsw;
6953 fxsave->twd = fpu->ftwx;
6954 fxsave->fop = fpu->last_opcode;
6955 fxsave->rip = fpu->last_ip;
6956 fxsave->rdp = fpu->last_dp;
6957 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6962 int fx_init(struct kvm_vcpu *vcpu)
6966 err = fpu_alloc(&vcpu->arch.guest_fpu);
6970 fpu_finit(&vcpu->arch.guest_fpu);
6972 vcpu->arch.guest_fpu.state->xsave.xsave_hdr.xcomp_bv =
6973 host_xcr0 | XSTATE_COMPACTION_ENABLED;
6976 * Ensure guest xcr0 is valid for loading
6978 vcpu->arch.xcr0 = XSTATE_FP;
6980 vcpu->arch.cr0 |= X86_CR0_ET;
6984 EXPORT_SYMBOL_GPL(fx_init);
6986 static void fx_free(struct kvm_vcpu *vcpu)
6988 fpu_free(&vcpu->arch.guest_fpu);
6991 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6993 if (vcpu->guest_fpu_loaded)
6997 * Restore all possible states in the guest,
6998 * and assume host would use all available bits.
6999 * Guest xcr0 would be loaded later.
7001 kvm_put_guest_xcr0(vcpu);
7002 vcpu->guest_fpu_loaded = 1;
7003 __kernel_fpu_begin();
7004 fpu_restore_checking(&vcpu->arch.guest_fpu);
7008 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7010 kvm_put_guest_xcr0(vcpu);
7012 if (!vcpu->guest_fpu_loaded)
7015 vcpu->guest_fpu_loaded = 0;
7016 fpu_save_init(&vcpu->arch.guest_fpu);
7018 ++vcpu->stat.fpu_reload;
7019 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7023 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7025 kvmclock_reset(vcpu);
7027 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7029 kvm_x86_ops->vcpu_free(vcpu);
7032 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7035 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7036 printk_once(KERN_WARNING
7037 "kvm: SMP vm created on host with unstable TSC; "
7038 "guest TSC will not be reliable\n");
7039 return kvm_x86_ops->vcpu_create(kvm, id);
7042 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7046 vcpu->arch.mtrr_state.have_fixed = 1;
7047 r = vcpu_load(vcpu);
7050 kvm_vcpu_reset(vcpu);
7051 kvm_mmu_setup(vcpu);
7057 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7060 struct msr_data msr;
7061 struct kvm *kvm = vcpu->kvm;
7063 r = vcpu_load(vcpu);
7067 msr.index = MSR_IA32_TSC;
7068 msr.host_initiated = true;
7069 kvm_write_tsc(vcpu, &msr);
7072 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7073 KVMCLOCK_SYNC_PERIOD);
7078 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7081 vcpu->arch.apf.msr_val = 0;
7083 r = vcpu_load(vcpu);
7085 kvm_mmu_unload(vcpu);
7089 kvm_x86_ops->vcpu_free(vcpu);
7092 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
7094 atomic_set(&vcpu->arch.nmi_queued, 0);
7095 vcpu->arch.nmi_pending = 0;
7096 vcpu->arch.nmi_injected = false;
7097 kvm_clear_interrupt_queue(vcpu);
7098 kvm_clear_exception_queue(vcpu);
7100 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7101 vcpu->arch.dr6 = DR6_INIT;
7102 kvm_update_dr6(vcpu);
7103 vcpu->arch.dr7 = DR7_FIXED_1;
7104 kvm_update_dr7(vcpu);
7106 kvm_make_request(KVM_REQ_EVENT, vcpu);
7107 vcpu->arch.apf.msr_val = 0;
7108 vcpu->arch.st.msr_val = 0;
7110 kvmclock_reset(vcpu);
7112 kvm_clear_async_pf_completion_queue(vcpu);
7113 kvm_async_pf_hash_reset(vcpu);
7114 vcpu->arch.apf.halted = false;
7116 kvm_pmu_reset(vcpu);
7118 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7119 vcpu->arch.regs_avail = ~0;
7120 vcpu->arch.regs_dirty = ~0;
7122 kvm_x86_ops->vcpu_reset(vcpu);
7125 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7127 struct kvm_segment cs;
7129 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7130 cs.selector = vector << 8;
7131 cs.base = vector << 12;
7132 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7133 kvm_rip_write(vcpu, 0);
7136 int kvm_arch_hardware_enable(void)
7139 struct kvm_vcpu *vcpu;
7144 bool stable, backwards_tsc = false;
7146 kvm_shared_msr_cpu_online();
7147 ret = kvm_x86_ops->hardware_enable();
7151 local_tsc = native_read_tsc();
7152 stable = !check_tsc_unstable();
7153 list_for_each_entry(kvm, &vm_list, vm_list) {
7154 kvm_for_each_vcpu(i, vcpu, kvm) {
7155 if (!stable && vcpu->cpu == smp_processor_id())
7156 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7157 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7158 backwards_tsc = true;
7159 if (vcpu->arch.last_host_tsc > max_tsc)
7160 max_tsc = vcpu->arch.last_host_tsc;
7166 * Sometimes, even reliable TSCs go backwards. This happens on
7167 * platforms that reset TSC during suspend or hibernate actions, but
7168 * maintain synchronization. We must compensate. Fortunately, we can
7169 * detect that condition here, which happens early in CPU bringup,
7170 * before any KVM threads can be running. Unfortunately, we can't
7171 * bring the TSCs fully up to date with real time, as we aren't yet far
7172 * enough into CPU bringup that we know how much real time has actually
7173 * elapsed; our helper function, get_kernel_ns() will be using boot
7174 * variables that haven't been updated yet.
7176 * So we simply find the maximum observed TSC above, then record the
7177 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7178 * the adjustment will be applied. Note that we accumulate
7179 * adjustments, in case multiple suspend cycles happen before some VCPU
7180 * gets a chance to run again. In the event that no KVM threads get a
7181 * chance to run, we will miss the entire elapsed period, as we'll have
7182 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7183 * loose cycle time. This isn't too big a deal, since the loss will be
7184 * uniform across all VCPUs (not to mention the scenario is extremely
7185 * unlikely). It is possible that a second hibernate recovery happens
7186 * much faster than a first, causing the observed TSC here to be
7187 * smaller; this would require additional padding adjustment, which is
7188 * why we set last_host_tsc to the local tsc observed here.
7190 * N.B. - this code below runs only on platforms with reliable TSC,
7191 * as that is the only way backwards_tsc is set above. Also note
7192 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7193 * have the same delta_cyc adjustment applied if backwards_tsc
7194 * is detected. Note further, this adjustment is only done once,
7195 * as we reset last_host_tsc on all VCPUs to stop this from being
7196 * called multiple times (one for each physical CPU bringup).
7198 * Platforms with unreliable TSCs don't have to deal with this, they
7199 * will be compensated by the logic in vcpu_load, which sets the TSC to
7200 * catchup mode. This will catchup all VCPUs to real time, but cannot
7201 * guarantee that they stay in perfect synchronization.
7203 if (backwards_tsc) {
7204 u64 delta_cyc = max_tsc - local_tsc;
7205 backwards_tsc_observed = true;
7206 list_for_each_entry(kvm, &vm_list, vm_list) {
7207 kvm_for_each_vcpu(i, vcpu, kvm) {
7208 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7209 vcpu->arch.last_host_tsc = local_tsc;
7210 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7214 * We have to disable TSC offset matching.. if you were
7215 * booting a VM while issuing an S4 host suspend....
7216 * you may have some problem. Solving this issue is
7217 * left as an exercise to the reader.
7219 kvm->arch.last_tsc_nsec = 0;
7220 kvm->arch.last_tsc_write = 0;
7227 void kvm_arch_hardware_disable(void)
7229 kvm_x86_ops->hardware_disable();
7230 drop_user_return_notifiers();
7233 int kvm_arch_hardware_setup(void)
7235 return kvm_x86_ops->hardware_setup();
7238 void kvm_arch_hardware_unsetup(void)
7240 kvm_x86_ops->hardware_unsetup();
7243 void kvm_arch_check_processor_compat(void *rtn)
7245 kvm_x86_ops->check_processor_compatibility(rtn);
7248 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7250 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7253 struct static_key kvm_no_apic_vcpu __read_mostly;
7255 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7261 BUG_ON(vcpu->kvm == NULL);
7264 vcpu->arch.pv.pv_unhalted = false;
7265 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7266 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
7267 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7269 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7271 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7276 vcpu->arch.pio_data = page_address(page);
7278 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7280 r = kvm_mmu_create(vcpu);
7282 goto fail_free_pio_data;
7284 if (irqchip_in_kernel(kvm)) {
7285 r = kvm_create_lapic(vcpu);
7287 goto fail_mmu_destroy;
7289 static_key_slow_inc(&kvm_no_apic_vcpu);
7291 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7293 if (!vcpu->arch.mce_banks) {
7295 goto fail_free_lapic;
7297 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7299 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7301 goto fail_free_mce_banks;
7306 goto fail_free_wbinvd_dirty_mask;
7308 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7309 vcpu->arch.pv_time_enabled = false;
7311 vcpu->arch.guest_supported_xcr0 = 0;
7312 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7314 kvm_async_pf_hash_reset(vcpu);
7318 fail_free_wbinvd_dirty_mask:
7319 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7320 fail_free_mce_banks:
7321 kfree(vcpu->arch.mce_banks);
7323 kvm_free_lapic(vcpu);
7325 kvm_mmu_destroy(vcpu);
7327 free_page((unsigned long)vcpu->arch.pio_data);
7332 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7336 kvm_pmu_destroy(vcpu);
7337 kfree(vcpu->arch.mce_banks);
7338 kvm_free_lapic(vcpu);
7339 idx = srcu_read_lock(&vcpu->kvm->srcu);
7340 kvm_mmu_destroy(vcpu);
7341 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7342 free_page((unsigned long)vcpu->arch.pio_data);
7343 if (!irqchip_in_kernel(vcpu->kvm))
7344 static_key_slow_dec(&kvm_no_apic_vcpu);
7347 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7349 kvm_x86_ops->sched_in(vcpu, cpu);
7352 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7357 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7358 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7359 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7360 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7361 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7363 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7364 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7365 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7366 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7367 &kvm->arch.irq_sources_bitmap);
7369 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7370 mutex_init(&kvm->arch.apic_map_lock);
7371 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7373 pvclock_update_vm_gtod_copy(kvm);
7375 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7376 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7381 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7384 r = vcpu_load(vcpu);
7386 kvm_mmu_unload(vcpu);
7390 static void kvm_free_vcpus(struct kvm *kvm)
7393 struct kvm_vcpu *vcpu;
7396 * Unpin any mmu pages first.
7398 kvm_for_each_vcpu(i, vcpu, kvm) {
7399 kvm_clear_async_pf_completion_queue(vcpu);
7400 kvm_unload_vcpu_mmu(vcpu);
7402 kvm_for_each_vcpu(i, vcpu, kvm)
7403 kvm_arch_vcpu_free(vcpu);
7405 mutex_lock(&kvm->lock);
7406 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7407 kvm->vcpus[i] = NULL;
7409 atomic_set(&kvm->online_vcpus, 0);
7410 mutex_unlock(&kvm->lock);
7413 void kvm_arch_sync_events(struct kvm *kvm)
7415 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7416 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7417 kvm_free_all_assigned_devices(kvm);
7421 void kvm_arch_destroy_vm(struct kvm *kvm)
7423 if (current->mm == kvm->mm) {
7425 * Free memory regions allocated on behalf of userspace,
7426 * unless the the memory map has changed due to process exit
7429 struct kvm_userspace_memory_region mem;
7430 memset(&mem, 0, sizeof(mem));
7431 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7432 kvm_set_memory_region(kvm, &mem);
7434 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7435 kvm_set_memory_region(kvm, &mem);
7437 mem.slot = TSS_PRIVATE_MEMSLOT;
7438 kvm_set_memory_region(kvm, &mem);
7440 kvm_iommu_unmap_guest(kvm);
7441 kfree(kvm->arch.vpic);
7442 kfree(kvm->arch.vioapic);
7443 kvm_free_vcpus(kvm);
7444 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7447 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7448 struct kvm_memory_slot *dont)
7452 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7453 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7454 kvm_kvfree(free->arch.rmap[i]);
7455 free->arch.rmap[i] = NULL;
7460 if (!dont || free->arch.lpage_info[i - 1] !=
7461 dont->arch.lpage_info[i - 1]) {
7462 kvm_kvfree(free->arch.lpage_info[i - 1]);
7463 free->arch.lpage_info[i - 1] = NULL;
7468 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7469 unsigned long npages)
7473 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7478 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7479 slot->base_gfn, level) + 1;
7481 slot->arch.rmap[i] =
7482 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7483 if (!slot->arch.rmap[i])
7488 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7489 sizeof(*slot->arch.lpage_info[i - 1]));
7490 if (!slot->arch.lpage_info[i - 1])
7493 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7494 slot->arch.lpage_info[i - 1][0].write_count = 1;
7495 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7496 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7497 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7499 * If the gfn and userspace address are not aligned wrt each
7500 * other, or if explicitly asked to, disable large page
7501 * support for this slot
7503 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7504 !kvm_largepages_enabled()) {
7507 for (j = 0; j < lpages; ++j)
7508 slot->arch.lpage_info[i - 1][j].write_count = 1;
7515 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7516 kvm_kvfree(slot->arch.rmap[i]);
7517 slot->arch.rmap[i] = NULL;
7521 kvm_kvfree(slot->arch.lpage_info[i - 1]);
7522 slot->arch.lpage_info[i - 1] = NULL;
7527 void kvm_arch_memslots_updated(struct kvm *kvm)
7530 * memslots->generation has been incremented.
7531 * mmio generation may have reached its maximum value.
7533 kvm_mmu_invalidate_mmio_sptes(kvm);
7536 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7537 struct kvm_memory_slot *memslot,
7538 struct kvm_userspace_memory_region *mem,
7539 enum kvm_mr_change change)
7542 * Only private memory slots need to be mapped here since
7543 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7545 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7546 unsigned long userspace_addr;
7549 * MAP_SHARED to prevent internal slot pages from being moved
7552 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7553 PROT_READ | PROT_WRITE,
7554 MAP_SHARED | MAP_ANONYMOUS, 0);
7556 if (IS_ERR((void *)userspace_addr))
7557 return PTR_ERR((void *)userspace_addr);
7559 memslot->userspace_addr = userspace_addr;
7565 void kvm_arch_commit_memory_region(struct kvm *kvm,
7566 struct kvm_userspace_memory_region *mem,
7567 const struct kvm_memory_slot *old,
7568 enum kvm_mr_change change)
7571 int nr_mmu_pages = 0;
7573 if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7576 ret = vm_munmap(old->userspace_addr,
7577 old->npages * PAGE_SIZE);
7580 "kvm_vm_ioctl_set_memory_region: "
7581 "failed to munmap memory\n");
7584 if (!kvm->arch.n_requested_mmu_pages)
7585 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7588 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7590 * Write protect all pages for dirty logging.
7592 * All the sptes including the large sptes which point to this
7593 * slot are set to readonly. We can not create any new large
7594 * spte on this slot until the end of the logging.
7596 * See the comments in fast_page_fault().
7598 if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
7599 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7602 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7604 kvm_mmu_invalidate_zap_all_pages(kvm);
7607 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7608 struct kvm_memory_slot *slot)
7610 kvm_mmu_invalidate_zap_all_pages(kvm);
7613 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7615 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7616 kvm_x86_ops->check_nested_events(vcpu, false);
7618 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7619 !vcpu->arch.apf.halted)
7620 || !list_empty_careful(&vcpu->async_pf.done)
7621 || kvm_apic_has_events(vcpu)
7622 || vcpu->arch.pv.pv_unhalted
7623 || atomic_read(&vcpu->arch.nmi_queued) ||
7624 (kvm_arch_interrupt_allowed(vcpu) &&
7625 kvm_cpu_has_interrupt(vcpu));
7628 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7630 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7633 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7635 return kvm_x86_ops->interrupt_allowed(vcpu);
7638 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
7640 if (is_64_bit_mode(vcpu))
7641 return kvm_rip_read(vcpu);
7642 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
7643 kvm_rip_read(vcpu));
7645 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
7647 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7649 return kvm_get_linear_rip(vcpu) == linear_rip;
7651 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7653 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7655 unsigned long rflags;
7657 rflags = kvm_x86_ops->get_rflags(vcpu);
7658 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7659 rflags &= ~X86_EFLAGS_TF;
7662 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7664 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7666 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7667 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7668 rflags |= X86_EFLAGS_TF;
7669 kvm_x86_ops->set_rflags(vcpu, rflags);
7672 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7674 __kvm_set_rflags(vcpu, rflags);
7675 kvm_make_request(KVM_REQ_EVENT, vcpu);
7677 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7679 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7683 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7687 r = kvm_mmu_reload(vcpu);
7691 if (!vcpu->arch.mmu.direct_map &&
7692 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7695 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7698 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7700 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7703 static inline u32 kvm_async_pf_next_probe(u32 key)
7705 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7708 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7710 u32 key = kvm_async_pf_hash_fn(gfn);
7712 while (vcpu->arch.apf.gfns[key] != ~0)
7713 key = kvm_async_pf_next_probe(key);
7715 vcpu->arch.apf.gfns[key] = gfn;
7718 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7721 u32 key = kvm_async_pf_hash_fn(gfn);
7723 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7724 (vcpu->arch.apf.gfns[key] != gfn &&
7725 vcpu->arch.apf.gfns[key] != ~0); i++)
7726 key = kvm_async_pf_next_probe(key);
7731 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7733 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7736 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7740 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7742 vcpu->arch.apf.gfns[i] = ~0;
7744 j = kvm_async_pf_next_probe(j);
7745 if (vcpu->arch.apf.gfns[j] == ~0)
7747 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7749 * k lies cyclically in ]i,j]
7751 * |....j i.k.| or |.k..j i...|
7753 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7754 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7759 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7762 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7766 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7767 struct kvm_async_pf *work)
7769 struct x86_exception fault;
7771 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7772 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7774 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7775 (vcpu->arch.apf.send_user_only &&
7776 kvm_x86_ops->get_cpl(vcpu) == 0))
7777 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7778 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7779 fault.vector = PF_VECTOR;
7780 fault.error_code_valid = true;
7781 fault.error_code = 0;
7782 fault.nested_page_fault = false;
7783 fault.address = work->arch.token;
7784 kvm_inject_page_fault(vcpu, &fault);
7788 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7789 struct kvm_async_pf *work)
7791 struct x86_exception fault;
7793 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7794 if (work->wakeup_all)
7795 work->arch.token = ~0; /* broadcast wakeup */
7797 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7799 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7800 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7801 fault.vector = PF_VECTOR;
7802 fault.error_code_valid = true;
7803 fault.error_code = 0;
7804 fault.nested_page_fault = false;
7805 fault.address = work->arch.token;
7806 kvm_inject_page_fault(vcpu, &fault);
7808 vcpu->arch.apf.halted = false;
7809 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7812 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7814 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7817 return !kvm_event_needs_reinjection(vcpu) &&
7818 kvm_x86_ops->interrupt_allowed(vcpu);
7821 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
7823 atomic_inc(&kvm->arch.noncoherent_dma_count);
7825 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
7827 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
7829 atomic_dec(&kvm->arch.noncoherent_dma_count);
7831 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
7833 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
7835 return atomic_read(&kvm->arch.noncoherent_dma_count);
7837 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
7839 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7840 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7841 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7842 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7843 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7844 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7845 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7846 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7847 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7848 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7849 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7850 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7851 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7852 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);