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 static 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 static 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 static 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 static 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
1546 && !ka->boot_vcpu_runs_old_kvmclock;
1548 if (ka->use_master_clock)
1549 atomic_set(&kvm_guest_has_master_clock, 1);
1551 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1552 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1557 static void kvm_gen_update_masterclock(struct kvm *kvm)
1559 #ifdef CONFIG_X86_64
1561 struct kvm_vcpu *vcpu;
1562 struct kvm_arch *ka = &kvm->arch;
1564 spin_lock(&ka->pvclock_gtod_sync_lock);
1565 kvm_make_mclock_inprogress_request(kvm);
1566 /* no guest entries from this point */
1567 pvclock_update_vm_gtod_copy(kvm);
1569 kvm_for_each_vcpu(i, vcpu, kvm)
1570 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1572 /* guest entries allowed */
1573 kvm_for_each_vcpu(i, vcpu, kvm)
1574 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1576 spin_unlock(&ka->pvclock_gtod_sync_lock);
1580 static int kvm_guest_time_update(struct kvm_vcpu *v)
1582 unsigned long flags, this_tsc_khz;
1583 struct kvm_vcpu_arch *vcpu = &v->arch;
1584 struct kvm_arch *ka = &v->kvm->arch;
1586 u64 tsc_timestamp, host_tsc;
1587 struct pvclock_vcpu_time_info guest_hv_clock;
1589 bool use_master_clock;
1595 * If the host uses TSC clock, then passthrough TSC as stable
1598 spin_lock(&ka->pvclock_gtod_sync_lock);
1599 use_master_clock = ka->use_master_clock;
1600 if (use_master_clock) {
1601 host_tsc = ka->master_cycle_now;
1602 kernel_ns = ka->master_kernel_ns;
1604 spin_unlock(&ka->pvclock_gtod_sync_lock);
1606 /* Keep irq disabled to prevent changes to the clock */
1607 local_irq_save(flags);
1608 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1609 if (unlikely(this_tsc_khz == 0)) {
1610 local_irq_restore(flags);
1611 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1614 if (!use_master_clock) {
1615 host_tsc = native_read_tsc();
1616 kernel_ns = get_kernel_ns();
1619 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1622 * We may have to catch up the TSC to match elapsed wall clock
1623 * time for two reasons, even if kvmclock is used.
1624 * 1) CPU could have been running below the maximum TSC rate
1625 * 2) Broken TSC compensation resets the base at each VCPU
1626 * entry to avoid unknown leaps of TSC even when running
1627 * again on the same CPU. This may cause apparent elapsed
1628 * time to disappear, and the guest to stand still or run
1631 if (vcpu->tsc_catchup) {
1632 u64 tsc = compute_guest_tsc(v, kernel_ns);
1633 if (tsc > tsc_timestamp) {
1634 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1635 tsc_timestamp = tsc;
1639 local_irq_restore(flags);
1641 if (!vcpu->pv_time_enabled)
1644 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1645 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1646 &vcpu->hv_clock.tsc_shift,
1647 &vcpu->hv_clock.tsc_to_system_mul);
1648 vcpu->hw_tsc_khz = this_tsc_khz;
1651 /* With all the info we got, fill in the values */
1652 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1653 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1654 vcpu->last_guest_tsc = tsc_timestamp;
1656 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1657 &guest_hv_clock, sizeof(guest_hv_clock))))
1661 * The interface expects us to write an even number signaling that the
1662 * update is finished. Since the guest won't see the intermediate
1663 * state, we just increase by 2 at the end.
1665 vcpu->hv_clock.version = guest_hv_clock.version + 2;
1667 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1668 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1670 if (vcpu->pvclock_set_guest_stopped_request) {
1671 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1672 vcpu->pvclock_set_guest_stopped_request = false;
1675 /* If the host uses TSC clocksource, then it is stable */
1676 if (use_master_clock)
1677 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1679 vcpu->hv_clock.flags = pvclock_flags;
1681 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1683 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1685 sizeof(vcpu->hv_clock));
1690 * kvmclock updates which are isolated to a given vcpu, such as
1691 * vcpu->cpu migration, should not allow system_timestamp from
1692 * the rest of the vcpus to remain static. Otherwise ntp frequency
1693 * correction applies to one vcpu's system_timestamp but not
1696 * So in those cases, request a kvmclock update for all vcpus.
1697 * We need to rate-limit these requests though, as they can
1698 * considerably slow guests that have a large number of vcpus.
1699 * The time for a remote vcpu to update its kvmclock is bound
1700 * by the delay we use to rate-limit the updates.
1703 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1705 static void kvmclock_update_fn(struct work_struct *work)
1708 struct delayed_work *dwork = to_delayed_work(work);
1709 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1710 kvmclock_update_work);
1711 struct kvm *kvm = container_of(ka, struct kvm, arch);
1712 struct kvm_vcpu *vcpu;
1714 kvm_for_each_vcpu(i, vcpu, kvm) {
1715 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1716 kvm_vcpu_kick(vcpu);
1720 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1722 struct kvm *kvm = v->kvm;
1724 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1725 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1726 KVMCLOCK_UPDATE_DELAY);
1729 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1731 static void kvmclock_sync_fn(struct work_struct *work)
1733 struct delayed_work *dwork = to_delayed_work(work);
1734 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1735 kvmclock_sync_work);
1736 struct kvm *kvm = container_of(ka, struct kvm, arch);
1738 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1739 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1740 KVMCLOCK_SYNC_PERIOD);
1743 static bool msr_mtrr_valid(unsigned msr)
1746 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1747 case MSR_MTRRfix64K_00000:
1748 case MSR_MTRRfix16K_80000:
1749 case MSR_MTRRfix16K_A0000:
1750 case MSR_MTRRfix4K_C0000:
1751 case MSR_MTRRfix4K_C8000:
1752 case MSR_MTRRfix4K_D0000:
1753 case MSR_MTRRfix4K_D8000:
1754 case MSR_MTRRfix4K_E0000:
1755 case MSR_MTRRfix4K_E8000:
1756 case MSR_MTRRfix4K_F0000:
1757 case MSR_MTRRfix4K_F8000:
1758 case MSR_MTRRdefType:
1759 case MSR_IA32_CR_PAT:
1767 static bool valid_pat_type(unsigned t)
1769 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1772 static bool valid_mtrr_type(unsigned t)
1774 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1777 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1782 if (!msr_mtrr_valid(msr))
1785 if (msr == MSR_IA32_CR_PAT) {
1786 for (i = 0; i < 8; i++)
1787 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1790 } else if (msr == MSR_MTRRdefType) {
1793 return valid_mtrr_type(data & 0xff);
1794 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1795 for (i = 0; i < 8 ; i++)
1796 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1801 /* variable MTRRs */
1802 WARN_ON(!(msr >= 0x200 && msr < 0x200 + 2 * KVM_NR_VAR_MTRR));
1804 mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
1805 if ((msr & 1) == 0) {
1807 if (!valid_mtrr_type(data & 0xff))
1814 kvm_inject_gp(vcpu, 0);
1820 EXPORT_SYMBOL_GPL(kvm_mtrr_valid);
1822 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1824 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1826 if (!kvm_mtrr_valid(vcpu, msr, data))
1829 if (msr == MSR_MTRRdefType) {
1830 vcpu->arch.mtrr_state.def_type = data;
1831 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1832 } else if (msr == MSR_MTRRfix64K_00000)
1834 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1835 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1836 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1837 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1838 else if (msr == MSR_IA32_CR_PAT)
1839 vcpu->arch.pat = data;
1840 else { /* Variable MTRRs */
1841 int idx, is_mtrr_mask;
1844 idx = (msr - 0x200) / 2;
1845 is_mtrr_mask = msr - 0x200 - 2 * idx;
1848 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1851 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1855 kvm_mmu_reset_context(vcpu);
1859 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1861 u64 mcg_cap = vcpu->arch.mcg_cap;
1862 unsigned bank_num = mcg_cap & 0xff;
1865 case MSR_IA32_MCG_STATUS:
1866 vcpu->arch.mcg_status = data;
1868 case MSR_IA32_MCG_CTL:
1869 if (!(mcg_cap & MCG_CTL_P))
1871 if (data != 0 && data != ~(u64)0)
1873 vcpu->arch.mcg_ctl = data;
1876 if (msr >= MSR_IA32_MC0_CTL &&
1877 msr < MSR_IA32_MCx_CTL(bank_num)) {
1878 u32 offset = msr - MSR_IA32_MC0_CTL;
1879 /* only 0 or all 1s can be written to IA32_MCi_CTL
1880 * some Linux kernels though clear bit 10 in bank 4 to
1881 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1882 * this to avoid an uncatched #GP in the guest
1884 if ((offset & 0x3) == 0 &&
1885 data != 0 && (data | (1 << 10)) != ~(u64)0)
1887 vcpu->arch.mce_banks[offset] = data;
1895 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1897 struct kvm *kvm = vcpu->kvm;
1898 int lm = is_long_mode(vcpu);
1899 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1900 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1901 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1902 : kvm->arch.xen_hvm_config.blob_size_32;
1903 u32 page_num = data & ~PAGE_MASK;
1904 u64 page_addr = data & PAGE_MASK;
1909 if (page_num >= blob_size)
1912 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1917 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1926 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1928 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1931 static bool kvm_hv_msr_partition_wide(u32 msr)
1935 case HV_X64_MSR_GUEST_OS_ID:
1936 case HV_X64_MSR_HYPERCALL:
1937 case HV_X64_MSR_REFERENCE_TSC:
1938 case HV_X64_MSR_TIME_REF_COUNT:
1946 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1948 struct kvm *kvm = vcpu->kvm;
1951 case HV_X64_MSR_GUEST_OS_ID:
1952 kvm->arch.hv_guest_os_id = data;
1953 /* setting guest os id to zero disables hypercall page */
1954 if (!kvm->arch.hv_guest_os_id)
1955 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1957 case HV_X64_MSR_HYPERCALL: {
1962 /* if guest os id is not set hypercall should remain disabled */
1963 if (!kvm->arch.hv_guest_os_id)
1965 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1966 kvm->arch.hv_hypercall = data;
1969 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1970 addr = gfn_to_hva(kvm, gfn);
1971 if (kvm_is_error_hva(addr))
1973 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1974 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1975 if (__copy_to_user((void __user *)addr, instructions, 4))
1977 kvm->arch.hv_hypercall = data;
1978 mark_page_dirty(kvm, gfn);
1981 case HV_X64_MSR_REFERENCE_TSC: {
1983 HV_REFERENCE_TSC_PAGE tsc_ref;
1984 memset(&tsc_ref, 0, sizeof(tsc_ref));
1985 kvm->arch.hv_tsc_page = data;
1986 if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1988 gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1989 if (kvm_write_guest(kvm, gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
1990 &tsc_ref, sizeof(tsc_ref)))
1992 mark_page_dirty(kvm, gfn);
1996 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1997 "data 0x%llx\n", msr, data);
2003 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2006 case HV_X64_MSR_APIC_ASSIST_PAGE: {
2010 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
2011 vcpu->arch.hv_vapic = data;
2012 if (kvm_lapic_enable_pv_eoi(vcpu, 0))
2016 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
2017 addr = gfn_to_hva(vcpu->kvm, gfn);
2018 if (kvm_is_error_hva(addr))
2020 if (__clear_user((void __user *)addr, PAGE_SIZE))
2022 vcpu->arch.hv_vapic = data;
2023 mark_page_dirty(vcpu->kvm, gfn);
2024 if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
2028 case HV_X64_MSR_EOI:
2029 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
2030 case HV_X64_MSR_ICR:
2031 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
2032 case HV_X64_MSR_TPR:
2033 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
2035 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2036 "data 0x%llx\n", msr, data);
2043 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2045 gpa_t gpa = data & ~0x3f;
2047 /* Bits 2:5 are reserved, Should be zero */
2051 vcpu->arch.apf.msr_val = data;
2053 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2054 kvm_clear_async_pf_completion_queue(vcpu);
2055 kvm_async_pf_hash_reset(vcpu);
2059 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2063 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2064 kvm_async_pf_wakeup_all(vcpu);
2068 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2070 vcpu->arch.pv_time_enabled = false;
2073 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2077 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2080 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2081 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2082 vcpu->arch.st.accum_steal = delta;
2085 static void record_steal_time(struct kvm_vcpu *vcpu)
2087 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2090 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2091 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2094 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2095 vcpu->arch.st.steal.version += 2;
2096 vcpu->arch.st.accum_steal = 0;
2098 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2099 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2102 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2105 u32 msr = msr_info->index;
2106 u64 data = msr_info->data;
2109 case MSR_AMD64_NB_CFG:
2110 case MSR_IA32_UCODE_REV:
2111 case MSR_IA32_UCODE_WRITE:
2112 case MSR_VM_HSAVE_PA:
2113 case MSR_AMD64_PATCH_LOADER:
2114 case MSR_AMD64_BU_CFG2:
2118 return set_efer(vcpu, data);
2120 data &= ~(u64)0x40; /* ignore flush filter disable */
2121 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2122 data &= ~(u64)0x8; /* ignore TLB cache disable */
2123 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2125 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2130 case MSR_FAM10H_MMIO_CONF_BASE:
2132 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2137 case MSR_IA32_DEBUGCTLMSR:
2139 /* We support the non-activated case already */
2141 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2142 /* Values other than LBR and BTF are vendor-specific,
2143 thus reserved and should throw a #GP */
2146 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2149 case 0x200 ... 0x2ff:
2150 return set_msr_mtrr(vcpu, msr, data);
2151 case MSR_IA32_APICBASE:
2152 return kvm_set_apic_base(vcpu, msr_info);
2153 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2154 return kvm_x2apic_msr_write(vcpu, msr, data);
2155 case MSR_IA32_TSCDEADLINE:
2156 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2158 case MSR_IA32_TSC_ADJUST:
2159 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2160 if (!msr_info->host_initiated) {
2161 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2162 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2164 vcpu->arch.ia32_tsc_adjust_msr = data;
2167 case MSR_IA32_MISC_ENABLE:
2168 vcpu->arch.ia32_misc_enable_msr = data;
2170 case MSR_KVM_WALL_CLOCK_NEW:
2171 case MSR_KVM_WALL_CLOCK:
2172 vcpu->kvm->arch.wall_clock = data;
2173 kvm_write_wall_clock(vcpu->kvm, data);
2175 case MSR_KVM_SYSTEM_TIME_NEW:
2176 case MSR_KVM_SYSTEM_TIME: {
2178 struct kvm_arch *ka = &vcpu->kvm->arch;
2180 kvmclock_reset(vcpu);
2182 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2183 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2185 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2186 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2189 ka->boot_vcpu_runs_old_kvmclock = tmp;
2192 vcpu->arch.time = data;
2193 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2195 /* we verify if the enable bit is set... */
2199 gpa_offset = data & ~(PAGE_MASK | 1);
2201 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2202 &vcpu->arch.pv_time, data & ~1ULL,
2203 sizeof(struct pvclock_vcpu_time_info)))
2204 vcpu->arch.pv_time_enabled = false;
2206 vcpu->arch.pv_time_enabled = true;
2210 case MSR_KVM_ASYNC_PF_EN:
2211 if (kvm_pv_enable_async_pf(vcpu, data))
2214 case MSR_KVM_STEAL_TIME:
2216 if (unlikely(!sched_info_on()))
2219 if (data & KVM_STEAL_RESERVED_MASK)
2222 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2223 data & KVM_STEAL_VALID_BITS,
2224 sizeof(struct kvm_steal_time)))
2227 vcpu->arch.st.msr_val = data;
2229 if (!(data & KVM_MSR_ENABLED))
2232 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2235 accumulate_steal_time(vcpu);
2238 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2241 case MSR_KVM_PV_EOI_EN:
2242 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2246 case MSR_IA32_MCG_CTL:
2247 case MSR_IA32_MCG_STATUS:
2248 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2249 return set_msr_mce(vcpu, msr, data);
2251 /* Performance counters are not protected by a CPUID bit,
2252 * so we should check all of them in the generic path for the sake of
2253 * cross vendor migration.
2254 * Writing a zero into the event select MSRs disables them,
2255 * which we perfectly emulate ;-). Any other value should be at least
2256 * reported, some guests depend on them.
2258 case MSR_K7_EVNTSEL0:
2259 case MSR_K7_EVNTSEL1:
2260 case MSR_K7_EVNTSEL2:
2261 case MSR_K7_EVNTSEL3:
2263 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2264 "0x%x data 0x%llx\n", msr, data);
2266 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2267 * so we ignore writes to make it happy.
2269 case MSR_K7_PERFCTR0:
2270 case MSR_K7_PERFCTR1:
2271 case MSR_K7_PERFCTR2:
2272 case MSR_K7_PERFCTR3:
2273 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2274 "0x%x data 0x%llx\n", msr, data);
2276 case MSR_P6_PERFCTR0:
2277 case MSR_P6_PERFCTR1:
2279 case MSR_P6_EVNTSEL0:
2280 case MSR_P6_EVNTSEL1:
2281 if (kvm_pmu_msr(vcpu, msr))
2282 return kvm_pmu_set_msr(vcpu, msr_info);
2284 if (pr || data != 0)
2285 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2286 "0x%x data 0x%llx\n", msr, data);
2288 case MSR_K7_CLK_CTL:
2290 * Ignore all writes to this no longer documented MSR.
2291 * Writes are only relevant for old K7 processors,
2292 * all pre-dating SVM, but a recommended workaround from
2293 * AMD for these chips. It is possible to specify the
2294 * affected processor models on the command line, hence
2295 * the need to ignore the workaround.
2298 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2299 if (kvm_hv_msr_partition_wide(msr)) {
2301 mutex_lock(&vcpu->kvm->lock);
2302 r = set_msr_hyperv_pw(vcpu, msr, data);
2303 mutex_unlock(&vcpu->kvm->lock);
2306 return set_msr_hyperv(vcpu, msr, data);
2308 case MSR_IA32_BBL_CR_CTL3:
2309 /* Drop writes to this legacy MSR -- see rdmsr
2310 * counterpart for further detail.
2312 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2314 case MSR_AMD64_OSVW_ID_LENGTH:
2315 if (!guest_cpuid_has_osvw(vcpu))
2317 vcpu->arch.osvw.length = data;
2319 case MSR_AMD64_OSVW_STATUS:
2320 if (!guest_cpuid_has_osvw(vcpu))
2322 vcpu->arch.osvw.status = data;
2325 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2326 return xen_hvm_config(vcpu, data);
2327 if (kvm_pmu_msr(vcpu, msr))
2328 return kvm_pmu_set_msr(vcpu, msr_info);
2330 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2334 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2341 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2345 * Reads an msr value (of 'msr_index') into 'pdata'.
2346 * Returns 0 on success, non-0 otherwise.
2347 * Assumes vcpu_load() was already called.
2349 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2351 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2353 EXPORT_SYMBOL_GPL(kvm_get_msr);
2355 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2357 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2359 if (!msr_mtrr_valid(msr))
2362 if (msr == MSR_MTRRdefType)
2363 *pdata = vcpu->arch.mtrr_state.def_type +
2364 (vcpu->arch.mtrr_state.enabled << 10);
2365 else if (msr == MSR_MTRRfix64K_00000)
2367 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2368 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2369 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2370 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2371 else if (msr == MSR_IA32_CR_PAT)
2372 *pdata = vcpu->arch.pat;
2373 else { /* Variable MTRRs */
2374 int idx, is_mtrr_mask;
2377 idx = (msr - 0x200) / 2;
2378 is_mtrr_mask = msr - 0x200 - 2 * idx;
2381 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2384 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2391 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2394 u64 mcg_cap = vcpu->arch.mcg_cap;
2395 unsigned bank_num = mcg_cap & 0xff;
2398 case MSR_IA32_P5_MC_ADDR:
2399 case MSR_IA32_P5_MC_TYPE:
2402 case MSR_IA32_MCG_CAP:
2403 data = vcpu->arch.mcg_cap;
2405 case MSR_IA32_MCG_CTL:
2406 if (!(mcg_cap & MCG_CTL_P))
2408 data = vcpu->arch.mcg_ctl;
2410 case MSR_IA32_MCG_STATUS:
2411 data = vcpu->arch.mcg_status;
2414 if (msr >= MSR_IA32_MC0_CTL &&
2415 msr < MSR_IA32_MCx_CTL(bank_num)) {
2416 u32 offset = msr - MSR_IA32_MC0_CTL;
2417 data = vcpu->arch.mce_banks[offset];
2426 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2429 struct kvm *kvm = vcpu->kvm;
2432 case HV_X64_MSR_GUEST_OS_ID:
2433 data = kvm->arch.hv_guest_os_id;
2435 case HV_X64_MSR_HYPERCALL:
2436 data = kvm->arch.hv_hypercall;
2438 case HV_X64_MSR_TIME_REF_COUNT: {
2440 div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2443 case HV_X64_MSR_REFERENCE_TSC:
2444 data = kvm->arch.hv_tsc_page;
2447 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2455 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2460 case HV_X64_MSR_VP_INDEX: {
2463 kvm_for_each_vcpu(r, v, vcpu->kvm) {
2471 case HV_X64_MSR_EOI:
2472 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2473 case HV_X64_MSR_ICR:
2474 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2475 case HV_X64_MSR_TPR:
2476 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2477 case HV_X64_MSR_APIC_ASSIST_PAGE:
2478 data = vcpu->arch.hv_vapic;
2481 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2488 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2493 case MSR_IA32_PLATFORM_ID:
2494 case MSR_IA32_EBL_CR_POWERON:
2495 case MSR_IA32_DEBUGCTLMSR:
2496 case MSR_IA32_LASTBRANCHFROMIP:
2497 case MSR_IA32_LASTBRANCHTOIP:
2498 case MSR_IA32_LASTINTFROMIP:
2499 case MSR_IA32_LASTINTTOIP:
2502 case MSR_VM_HSAVE_PA:
2503 case MSR_K7_EVNTSEL0:
2504 case MSR_K7_EVNTSEL1:
2505 case MSR_K7_EVNTSEL2:
2506 case MSR_K7_EVNTSEL3:
2507 case MSR_K7_PERFCTR0:
2508 case MSR_K7_PERFCTR1:
2509 case MSR_K7_PERFCTR2:
2510 case MSR_K7_PERFCTR3:
2511 case MSR_K8_INT_PENDING_MSG:
2512 case MSR_AMD64_NB_CFG:
2513 case MSR_FAM10H_MMIO_CONF_BASE:
2514 case MSR_AMD64_BU_CFG2:
2517 case MSR_P6_PERFCTR0:
2518 case MSR_P6_PERFCTR1:
2519 case MSR_P6_EVNTSEL0:
2520 case MSR_P6_EVNTSEL1:
2521 if (kvm_pmu_msr(vcpu, msr))
2522 return kvm_pmu_get_msr(vcpu, msr, pdata);
2525 case MSR_IA32_UCODE_REV:
2526 data = 0x100000000ULL;
2529 data = 0x500 | KVM_NR_VAR_MTRR;
2531 case 0x200 ... 0x2ff:
2532 return get_msr_mtrr(vcpu, msr, pdata);
2533 case 0xcd: /* fsb frequency */
2537 * MSR_EBC_FREQUENCY_ID
2538 * Conservative value valid for even the basic CPU models.
2539 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2540 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2541 * and 266MHz for model 3, or 4. Set Core Clock
2542 * Frequency to System Bus Frequency Ratio to 1 (bits
2543 * 31:24) even though these are only valid for CPU
2544 * models > 2, however guests may end up dividing or
2545 * multiplying by zero otherwise.
2547 case MSR_EBC_FREQUENCY_ID:
2550 case MSR_IA32_APICBASE:
2551 data = kvm_get_apic_base(vcpu);
2553 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2554 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2556 case MSR_IA32_TSCDEADLINE:
2557 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2559 case MSR_IA32_TSC_ADJUST:
2560 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2562 case MSR_IA32_MISC_ENABLE:
2563 data = vcpu->arch.ia32_misc_enable_msr;
2565 case MSR_IA32_PERF_STATUS:
2566 /* TSC increment by tick */
2568 /* CPU multiplier */
2569 data |= (((uint64_t)4ULL) << 40);
2572 data = vcpu->arch.efer;
2574 case MSR_KVM_WALL_CLOCK:
2575 case MSR_KVM_WALL_CLOCK_NEW:
2576 data = vcpu->kvm->arch.wall_clock;
2578 case MSR_KVM_SYSTEM_TIME:
2579 case MSR_KVM_SYSTEM_TIME_NEW:
2580 data = vcpu->arch.time;
2582 case MSR_KVM_ASYNC_PF_EN:
2583 data = vcpu->arch.apf.msr_val;
2585 case MSR_KVM_STEAL_TIME:
2586 data = vcpu->arch.st.msr_val;
2588 case MSR_KVM_PV_EOI_EN:
2589 data = vcpu->arch.pv_eoi.msr_val;
2591 case MSR_IA32_P5_MC_ADDR:
2592 case MSR_IA32_P5_MC_TYPE:
2593 case MSR_IA32_MCG_CAP:
2594 case MSR_IA32_MCG_CTL:
2595 case MSR_IA32_MCG_STATUS:
2596 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2597 return get_msr_mce(vcpu, msr, pdata);
2598 case MSR_K7_CLK_CTL:
2600 * Provide expected ramp-up count for K7. All other
2601 * are set to zero, indicating minimum divisors for
2604 * This prevents guest kernels on AMD host with CPU
2605 * type 6, model 8 and higher from exploding due to
2606 * the rdmsr failing.
2610 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2611 if (kvm_hv_msr_partition_wide(msr)) {
2613 mutex_lock(&vcpu->kvm->lock);
2614 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2615 mutex_unlock(&vcpu->kvm->lock);
2618 return get_msr_hyperv(vcpu, msr, pdata);
2620 case MSR_IA32_BBL_CR_CTL3:
2621 /* This legacy MSR exists but isn't fully documented in current
2622 * silicon. It is however accessed by winxp in very narrow
2623 * scenarios where it sets bit #19, itself documented as
2624 * a "reserved" bit. Best effort attempt to source coherent
2625 * read data here should the balance of the register be
2626 * interpreted by the guest:
2628 * L2 cache control register 3: 64GB range, 256KB size,
2629 * enabled, latency 0x1, configured
2633 case MSR_AMD64_OSVW_ID_LENGTH:
2634 if (!guest_cpuid_has_osvw(vcpu))
2636 data = vcpu->arch.osvw.length;
2638 case MSR_AMD64_OSVW_STATUS:
2639 if (!guest_cpuid_has_osvw(vcpu))
2641 data = vcpu->arch.osvw.status;
2644 if (kvm_pmu_msr(vcpu, msr))
2645 return kvm_pmu_get_msr(vcpu, msr, pdata);
2647 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2650 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2658 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2661 * Read or write a bunch of msrs. All parameters are kernel addresses.
2663 * @return number of msrs set successfully.
2665 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2666 struct kvm_msr_entry *entries,
2667 int (*do_msr)(struct kvm_vcpu *vcpu,
2668 unsigned index, u64 *data))
2672 idx = srcu_read_lock(&vcpu->kvm->srcu);
2673 for (i = 0; i < msrs->nmsrs; ++i)
2674 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2676 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2682 * Read or write a bunch of msrs. Parameters are user addresses.
2684 * @return number of msrs set successfully.
2686 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2687 int (*do_msr)(struct kvm_vcpu *vcpu,
2688 unsigned index, u64 *data),
2691 struct kvm_msrs msrs;
2692 struct kvm_msr_entry *entries;
2697 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2701 if (msrs.nmsrs >= MAX_IO_MSRS)
2704 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2705 entries = memdup_user(user_msrs->entries, size);
2706 if (IS_ERR(entries)) {
2707 r = PTR_ERR(entries);
2711 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2716 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2727 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2732 case KVM_CAP_IRQCHIP:
2734 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2735 case KVM_CAP_SET_TSS_ADDR:
2736 case KVM_CAP_EXT_CPUID:
2737 case KVM_CAP_EXT_EMUL_CPUID:
2738 case KVM_CAP_CLOCKSOURCE:
2740 case KVM_CAP_NOP_IO_DELAY:
2741 case KVM_CAP_MP_STATE:
2742 case KVM_CAP_SYNC_MMU:
2743 case KVM_CAP_USER_NMI:
2744 case KVM_CAP_REINJECT_CONTROL:
2745 case KVM_CAP_IRQ_INJECT_STATUS:
2747 case KVM_CAP_IOEVENTFD:
2748 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2750 case KVM_CAP_PIT_STATE2:
2751 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2752 case KVM_CAP_XEN_HVM:
2753 case KVM_CAP_ADJUST_CLOCK:
2754 case KVM_CAP_VCPU_EVENTS:
2755 case KVM_CAP_HYPERV:
2756 case KVM_CAP_HYPERV_VAPIC:
2757 case KVM_CAP_HYPERV_SPIN:
2758 case KVM_CAP_PCI_SEGMENT:
2759 case KVM_CAP_DEBUGREGS:
2760 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2762 case KVM_CAP_ASYNC_PF:
2763 case KVM_CAP_GET_TSC_KHZ:
2764 case KVM_CAP_KVMCLOCK_CTRL:
2765 case KVM_CAP_READONLY_MEM:
2766 case KVM_CAP_HYPERV_TIME:
2767 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2768 case KVM_CAP_TSC_DEADLINE_TIMER:
2769 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2770 case KVM_CAP_ASSIGN_DEV_IRQ:
2771 case KVM_CAP_PCI_2_3:
2775 case KVM_CAP_COALESCED_MMIO:
2776 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2779 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2781 case KVM_CAP_NR_VCPUS:
2782 r = KVM_SOFT_MAX_VCPUS;
2784 case KVM_CAP_MAX_VCPUS:
2787 case KVM_CAP_NR_MEMSLOTS:
2788 r = KVM_USER_MEM_SLOTS;
2790 case KVM_CAP_PV_MMU: /* obsolete */
2793 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2795 r = iommu_present(&pci_bus_type);
2799 r = KVM_MAX_MCE_BANKS;
2804 case KVM_CAP_TSC_CONTROL:
2805 r = kvm_has_tsc_control;
2815 long kvm_arch_dev_ioctl(struct file *filp,
2816 unsigned int ioctl, unsigned long arg)
2818 void __user *argp = (void __user *)arg;
2822 case KVM_GET_MSR_INDEX_LIST: {
2823 struct kvm_msr_list __user *user_msr_list = argp;
2824 struct kvm_msr_list msr_list;
2828 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2831 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2832 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2835 if (n < msr_list.nmsrs)
2838 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2839 num_msrs_to_save * sizeof(u32)))
2841 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2843 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2848 case KVM_GET_SUPPORTED_CPUID:
2849 case KVM_GET_EMULATED_CPUID: {
2850 struct kvm_cpuid2 __user *cpuid_arg = argp;
2851 struct kvm_cpuid2 cpuid;
2854 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2857 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2863 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2868 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2871 mce_cap = KVM_MCE_CAP_SUPPORTED;
2873 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2885 static void wbinvd_ipi(void *garbage)
2890 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2892 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2895 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2897 /* Address WBINVD may be executed by guest */
2898 if (need_emulate_wbinvd(vcpu)) {
2899 if (kvm_x86_ops->has_wbinvd_exit())
2900 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2901 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2902 smp_call_function_single(vcpu->cpu,
2903 wbinvd_ipi, NULL, 1);
2906 kvm_x86_ops->vcpu_load(vcpu, cpu);
2908 /* Apply any externally detected TSC adjustments (due to suspend) */
2909 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2910 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2911 vcpu->arch.tsc_offset_adjustment = 0;
2912 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2915 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2916 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2917 native_read_tsc() - vcpu->arch.last_host_tsc;
2919 mark_tsc_unstable("KVM discovered backwards TSC");
2920 if (check_tsc_unstable()) {
2921 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2922 vcpu->arch.last_guest_tsc);
2923 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2924 vcpu->arch.tsc_catchup = 1;
2927 * On a host with synchronized TSC, there is no need to update
2928 * kvmclock on vcpu->cpu migration
2930 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2931 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2932 if (vcpu->cpu != cpu)
2933 kvm_migrate_timers(vcpu);
2937 accumulate_steal_time(vcpu);
2938 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2941 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2943 kvm_x86_ops->vcpu_put(vcpu);
2944 kvm_put_guest_fpu(vcpu);
2945 vcpu->arch.last_host_tsc = native_read_tsc();
2948 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2949 struct kvm_lapic_state *s)
2951 kvm_x86_ops->sync_pir_to_irr(vcpu);
2952 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2957 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2958 struct kvm_lapic_state *s)
2960 kvm_apic_post_state_restore(vcpu, s);
2961 update_cr8_intercept(vcpu);
2966 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2967 struct kvm_interrupt *irq)
2969 if (irq->irq >= KVM_NR_INTERRUPTS)
2971 if (irqchip_in_kernel(vcpu->kvm))
2974 kvm_queue_interrupt(vcpu, irq->irq, false);
2975 kvm_make_request(KVM_REQ_EVENT, vcpu);
2980 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2982 kvm_inject_nmi(vcpu);
2987 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2988 struct kvm_tpr_access_ctl *tac)
2992 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2996 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3000 unsigned bank_num = mcg_cap & 0xff, bank;
3003 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3005 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
3008 vcpu->arch.mcg_cap = mcg_cap;
3009 /* Init IA32_MCG_CTL to all 1s */
3010 if (mcg_cap & MCG_CTL_P)
3011 vcpu->arch.mcg_ctl = ~(u64)0;
3012 /* Init IA32_MCi_CTL to all 1s */
3013 for (bank = 0; bank < bank_num; bank++)
3014 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3019 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3020 struct kvm_x86_mce *mce)
3022 u64 mcg_cap = vcpu->arch.mcg_cap;
3023 unsigned bank_num = mcg_cap & 0xff;
3024 u64 *banks = vcpu->arch.mce_banks;
3026 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3029 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3030 * reporting is disabled
3032 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3033 vcpu->arch.mcg_ctl != ~(u64)0)
3035 banks += 4 * mce->bank;
3037 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3038 * reporting is disabled for the bank
3040 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3042 if (mce->status & MCI_STATUS_UC) {
3043 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3044 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3045 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3048 if (banks[1] & MCI_STATUS_VAL)
3049 mce->status |= MCI_STATUS_OVER;
3050 banks[2] = mce->addr;
3051 banks[3] = mce->misc;
3052 vcpu->arch.mcg_status = mce->mcg_status;
3053 banks[1] = mce->status;
3054 kvm_queue_exception(vcpu, MC_VECTOR);
3055 } else if (!(banks[1] & MCI_STATUS_VAL)
3056 || !(banks[1] & MCI_STATUS_UC)) {
3057 if (banks[1] & MCI_STATUS_VAL)
3058 mce->status |= MCI_STATUS_OVER;
3059 banks[2] = mce->addr;
3060 banks[3] = mce->misc;
3061 banks[1] = mce->status;
3063 banks[1] |= MCI_STATUS_OVER;
3067 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3068 struct kvm_vcpu_events *events)
3071 events->exception.injected =
3072 vcpu->arch.exception.pending &&
3073 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3074 events->exception.nr = vcpu->arch.exception.nr;
3075 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3076 events->exception.pad = 0;
3077 events->exception.error_code = vcpu->arch.exception.error_code;
3079 events->interrupt.injected =
3080 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3081 events->interrupt.nr = vcpu->arch.interrupt.nr;
3082 events->interrupt.soft = 0;
3083 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3085 events->nmi.injected = vcpu->arch.nmi_injected;
3086 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3087 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3088 events->nmi.pad = 0;
3090 events->sipi_vector = 0; /* never valid when reporting to user space */
3092 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3093 | KVM_VCPUEVENT_VALID_SHADOW);
3094 memset(&events->reserved, 0, sizeof(events->reserved));
3097 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3098 struct kvm_vcpu_events *events)
3100 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3101 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3102 | KVM_VCPUEVENT_VALID_SHADOW))
3106 vcpu->arch.exception.pending = events->exception.injected;
3107 vcpu->arch.exception.nr = events->exception.nr;
3108 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3109 vcpu->arch.exception.error_code = events->exception.error_code;
3111 vcpu->arch.interrupt.pending = events->interrupt.injected;
3112 vcpu->arch.interrupt.nr = events->interrupt.nr;
3113 vcpu->arch.interrupt.soft = events->interrupt.soft;
3114 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3115 kvm_x86_ops->set_interrupt_shadow(vcpu,
3116 events->interrupt.shadow);
3118 vcpu->arch.nmi_injected = events->nmi.injected;
3119 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3120 vcpu->arch.nmi_pending = events->nmi.pending;
3121 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3123 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3124 kvm_vcpu_has_lapic(vcpu))
3125 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3127 kvm_make_request(KVM_REQ_EVENT, vcpu);
3132 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3133 struct kvm_debugregs *dbgregs)
3137 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3138 kvm_get_dr(vcpu, 6, &val);
3140 dbgregs->dr7 = vcpu->arch.dr7;
3142 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3145 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3146 struct kvm_debugregs *dbgregs)
3151 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3152 vcpu->arch.dr6 = dbgregs->dr6;
3153 kvm_update_dr6(vcpu);
3154 vcpu->arch.dr7 = dbgregs->dr7;
3155 kvm_update_dr7(vcpu);
3160 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3162 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3164 struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3165 u64 xstate_bv = xsave->xsave_hdr.xstate_bv;
3169 * Copy legacy XSAVE area, to avoid complications with CPUID
3170 * leaves 0 and 1 in the loop below.
3172 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3175 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3178 * Copy each region from the possibly compacted offset to the
3179 * non-compacted offset.
3181 valid = xstate_bv & ~XSTATE_FPSSE;
3183 u64 feature = valid & -valid;
3184 int index = fls64(feature) - 1;
3185 void *src = get_xsave_addr(xsave, feature);
3188 u32 size, offset, ecx, edx;
3189 cpuid_count(XSTATE_CPUID, index,
3190 &size, &offset, &ecx, &edx);
3191 memcpy(dest + offset, src, size);
3198 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3200 struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3201 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3205 * Copy legacy XSAVE area, to avoid complications with CPUID
3206 * leaves 0 and 1 in the loop below.
3208 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3210 /* Set XSTATE_BV and possibly XCOMP_BV. */
3211 xsave->xsave_hdr.xstate_bv = xstate_bv;
3213 xsave->xsave_hdr.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3216 * Copy each region from the non-compacted offset to the
3217 * possibly compacted offset.
3219 valid = xstate_bv & ~XSTATE_FPSSE;
3221 u64 feature = valid & -valid;
3222 int index = fls64(feature) - 1;
3223 void *dest = get_xsave_addr(xsave, feature);
3226 u32 size, offset, ecx, edx;
3227 cpuid_count(XSTATE_CPUID, index,
3228 &size, &offset, &ecx, &edx);
3229 memcpy(dest, src + offset, size);
3237 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3238 struct kvm_xsave *guest_xsave)
3240 if (cpu_has_xsave) {
3241 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3242 fill_xsave((u8 *) guest_xsave->region, vcpu);
3244 memcpy(guest_xsave->region,
3245 &vcpu->arch.guest_fpu.state->fxsave,
3246 sizeof(struct i387_fxsave_struct));
3247 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3252 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3253 struct kvm_xsave *guest_xsave)
3256 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3258 if (cpu_has_xsave) {
3260 * Here we allow setting states that are not present in
3261 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3262 * with old userspace.
3264 if (xstate_bv & ~kvm_supported_xcr0())
3266 load_xsave(vcpu, (u8 *)guest_xsave->region);
3268 if (xstate_bv & ~XSTATE_FPSSE)
3270 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3271 guest_xsave->region, sizeof(struct i387_fxsave_struct));
3276 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3277 struct kvm_xcrs *guest_xcrs)
3279 if (!cpu_has_xsave) {
3280 guest_xcrs->nr_xcrs = 0;
3284 guest_xcrs->nr_xcrs = 1;
3285 guest_xcrs->flags = 0;
3286 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3287 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3290 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3291 struct kvm_xcrs *guest_xcrs)
3298 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3301 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3302 /* Only support XCR0 currently */
3303 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3304 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3305 guest_xcrs->xcrs[i].value);
3314 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3315 * stopped by the hypervisor. This function will be called from the host only.
3316 * EINVAL is returned when the host attempts to set the flag for a guest that
3317 * does not support pv clocks.
3319 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3321 if (!vcpu->arch.pv_time_enabled)
3323 vcpu->arch.pvclock_set_guest_stopped_request = true;
3324 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3328 long kvm_arch_vcpu_ioctl(struct file *filp,
3329 unsigned int ioctl, unsigned long arg)
3331 struct kvm_vcpu *vcpu = filp->private_data;
3332 void __user *argp = (void __user *)arg;
3335 struct kvm_lapic_state *lapic;
3336 struct kvm_xsave *xsave;
3337 struct kvm_xcrs *xcrs;
3343 case KVM_GET_LAPIC: {
3345 if (!vcpu->arch.apic)
3347 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3352 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3356 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3361 case KVM_SET_LAPIC: {
3363 if (!vcpu->arch.apic)
3365 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3366 if (IS_ERR(u.lapic))
3367 return PTR_ERR(u.lapic);
3369 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3372 case KVM_INTERRUPT: {
3373 struct kvm_interrupt irq;
3376 if (copy_from_user(&irq, argp, sizeof irq))
3378 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3382 r = kvm_vcpu_ioctl_nmi(vcpu);
3385 case KVM_SET_CPUID: {
3386 struct kvm_cpuid __user *cpuid_arg = argp;
3387 struct kvm_cpuid cpuid;
3390 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3392 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3395 case KVM_SET_CPUID2: {
3396 struct kvm_cpuid2 __user *cpuid_arg = argp;
3397 struct kvm_cpuid2 cpuid;
3400 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3402 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3403 cpuid_arg->entries);
3406 case KVM_GET_CPUID2: {
3407 struct kvm_cpuid2 __user *cpuid_arg = argp;
3408 struct kvm_cpuid2 cpuid;
3411 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3413 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3414 cpuid_arg->entries);
3418 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3424 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3427 r = msr_io(vcpu, argp, do_set_msr, 0);
3429 case KVM_TPR_ACCESS_REPORTING: {
3430 struct kvm_tpr_access_ctl tac;
3433 if (copy_from_user(&tac, argp, sizeof tac))
3435 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3439 if (copy_to_user(argp, &tac, sizeof tac))
3444 case KVM_SET_VAPIC_ADDR: {
3445 struct kvm_vapic_addr va;
3448 if (!irqchip_in_kernel(vcpu->kvm))
3451 if (copy_from_user(&va, argp, sizeof va))
3453 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3456 case KVM_X86_SETUP_MCE: {
3460 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3462 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3465 case KVM_X86_SET_MCE: {
3466 struct kvm_x86_mce mce;
3469 if (copy_from_user(&mce, argp, sizeof mce))
3471 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3474 case KVM_GET_VCPU_EVENTS: {
3475 struct kvm_vcpu_events events;
3477 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3480 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3485 case KVM_SET_VCPU_EVENTS: {
3486 struct kvm_vcpu_events events;
3489 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3492 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3495 case KVM_GET_DEBUGREGS: {
3496 struct kvm_debugregs dbgregs;
3498 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3501 if (copy_to_user(argp, &dbgregs,
3502 sizeof(struct kvm_debugregs)))
3507 case KVM_SET_DEBUGREGS: {
3508 struct kvm_debugregs dbgregs;
3511 if (copy_from_user(&dbgregs, argp,
3512 sizeof(struct kvm_debugregs)))
3515 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3518 case KVM_GET_XSAVE: {
3519 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3524 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3527 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3532 case KVM_SET_XSAVE: {
3533 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3534 if (IS_ERR(u.xsave))
3535 return PTR_ERR(u.xsave);
3537 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3540 case KVM_GET_XCRS: {
3541 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3546 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3549 if (copy_to_user(argp, u.xcrs,
3550 sizeof(struct kvm_xcrs)))
3555 case KVM_SET_XCRS: {
3556 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3558 return PTR_ERR(u.xcrs);
3560 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3563 case KVM_SET_TSC_KHZ: {
3567 user_tsc_khz = (u32)arg;
3569 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3572 if (user_tsc_khz == 0)
3573 user_tsc_khz = tsc_khz;
3575 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3580 case KVM_GET_TSC_KHZ: {
3581 r = vcpu->arch.virtual_tsc_khz;
3584 case KVM_KVMCLOCK_CTRL: {
3585 r = kvm_set_guest_paused(vcpu);
3596 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3598 return VM_FAULT_SIGBUS;
3601 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3605 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3607 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3611 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3614 kvm->arch.ept_identity_map_addr = ident_addr;
3618 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3619 u32 kvm_nr_mmu_pages)
3621 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3624 mutex_lock(&kvm->slots_lock);
3626 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3627 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3629 mutex_unlock(&kvm->slots_lock);
3633 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3635 return kvm->arch.n_max_mmu_pages;
3638 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3643 switch (chip->chip_id) {
3644 case KVM_IRQCHIP_PIC_MASTER:
3645 memcpy(&chip->chip.pic,
3646 &pic_irqchip(kvm)->pics[0],
3647 sizeof(struct kvm_pic_state));
3649 case KVM_IRQCHIP_PIC_SLAVE:
3650 memcpy(&chip->chip.pic,
3651 &pic_irqchip(kvm)->pics[1],
3652 sizeof(struct kvm_pic_state));
3654 case KVM_IRQCHIP_IOAPIC:
3655 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3664 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3669 switch (chip->chip_id) {
3670 case KVM_IRQCHIP_PIC_MASTER:
3671 spin_lock(&pic_irqchip(kvm)->lock);
3672 memcpy(&pic_irqchip(kvm)->pics[0],
3674 sizeof(struct kvm_pic_state));
3675 spin_unlock(&pic_irqchip(kvm)->lock);
3677 case KVM_IRQCHIP_PIC_SLAVE:
3678 spin_lock(&pic_irqchip(kvm)->lock);
3679 memcpy(&pic_irqchip(kvm)->pics[1],
3681 sizeof(struct kvm_pic_state));
3682 spin_unlock(&pic_irqchip(kvm)->lock);
3684 case KVM_IRQCHIP_IOAPIC:
3685 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3691 kvm_pic_update_irq(pic_irqchip(kvm));
3695 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3699 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3700 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3701 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3705 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3709 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3710 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3711 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3712 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3716 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3720 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3721 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3722 sizeof(ps->channels));
3723 ps->flags = kvm->arch.vpit->pit_state.flags;
3724 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3725 memset(&ps->reserved, 0, sizeof(ps->reserved));
3729 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3731 int r = 0, start = 0;
3732 u32 prev_legacy, cur_legacy;
3733 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3734 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3735 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3736 if (!prev_legacy && cur_legacy)
3738 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3739 sizeof(kvm->arch.vpit->pit_state.channels));
3740 kvm->arch.vpit->pit_state.flags = ps->flags;
3741 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3742 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3746 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3747 struct kvm_reinject_control *control)
3749 if (!kvm->arch.vpit)
3751 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3752 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3753 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3758 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3759 * @kvm: kvm instance
3760 * @log: slot id and address to which we copy the log
3762 * Steps 1-4 below provide general overview of dirty page logging. See
3763 * kvm_get_dirty_log_protect() function description for additional details.
3765 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3766 * always flush the TLB (step 4) even if previous step failed and the dirty
3767 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3768 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3769 * writes will be marked dirty for next log read.
3771 * 1. Take a snapshot of the bit and clear it if needed.
3772 * 2. Write protect the corresponding page.
3773 * 3. Copy the snapshot to the userspace.
3774 * 4. Flush TLB's if needed.
3776 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3778 bool is_dirty = false;
3781 mutex_lock(&kvm->slots_lock);
3783 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3786 * All the TLBs can be flushed out of mmu lock, see the comments in
3787 * kvm_mmu_slot_remove_write_access().
3789 lockdep_assert_held(&kvm->slots_lock);
3791 kvm_flush_remote_tlbs(kvm);
3793 mutex_unlock(&kvm->slots_lock);
3797 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3800 if (!irqchip_in_kernel(kvm))
3803 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3804 irq_event->irq, irq_event->level,
3809 long kvm_arch_vm_ioctl(struct file *filp,
3810 unsigned int ioctl, unsigned long arg)
3812 struct kvm *kvm = filp->private_data;
3813 void __user *argp = (void __user *)arg;
3816 * This union makes it completely explicit to gcc-3.x
3817 * that these two variables' stack usage should be
3818 * combined, not added together.
3821 struct kvm_pit_state ps;
3822 struct kvm_pit_state2 ps2;
3823 struct kvm_pit_config pit_config;
3827 case KVM_SET_TSS_ADDR:
3828 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3830 case KVM_SET_IDENTITY_MAP_ADDR: {
3834 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3836 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3839 case KVM_SET_NR_MMU_PAGES:
3840 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3842 case KVM_GET_NR_MMU_PAGES:
3843 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3845 case KVM_CREATE_IRQCHIP: {
3846 struct kvm_pic *vpic;
3848 mutex_lock(&kvm->lock);
3851 goto create_irqchip_unlock;
3853 if (atomic_read(&kvm->online_vcpus))
3854 goto create_irqchip_unlock;
3856 vpic = kvm_create_pic(kvm);
3858 r = kvm_ioapic_init(kvm);
3860 mutex_lock(&kvm->slots_lock);
3861 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3863 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3865 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3867 mutex_unlock(&kvm->slots_lock);
3869 goto create_irqchip_unlock;
3872 goto create_irqchip_unlock;
3874 kvm->arch.vpic = vpic;
3876 r = kvm_setup_default_irq_routing(kvm);
3878 mutex_lock(&kvm->slots_lock);
3879 mutex_lock(&kvm->irq_lock);
3880 kvm_ioapic_destroy(kvm);
3881 kvm_destroy_pic(kvm);
3882 mutex_unlock(&kvm->irq_lock);
3883 mutex_unlock(&kvm->slots_lock);
3885 create_irqchip_unlock:
3886 mutex_unlock(&kvm->lock);
3889 case KVM_CREATE_PIT:
3890 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3892 case KVM_CREATE_PIT2:
3894 if (copy_from_user(&u.pit_config, argp,
3895 sizeof(struct kvm_pit_config)))
3898 mutex_lock(&kvm->slots_lock);
3901 goto create_pit_unlock;
3903 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3907 mutex_unlock(&kvm->slots_lock);
3909 case KVM_GET_IRQCHIP: {
3910 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3911 struct kvm_irqchip *chip;
3913 chip = memdup_user(argp, sizeof(*chip));
3920 if (!irqchip_in_kernel(kvm))
3921 goto get_irqchip_out;
3922 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3924 goto get_irqchip_out;
3926 if (copy_to_user(argp, chip, sizeof *chip))
3927 goto get_irqchip_out;
3933 case KVM_SET_IRQCHIP: {
3934 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3935 struct kvm_irqchip *chip;
3937 chip = memdup_user(argp, sizeof(*chip));
3944 if (!irqchip_in_kernel(kvm))
3945 goto set_irqchip_out;
3946 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3948 goto set_irqchip_out;
3956 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3959 if (!kvm->arch.vpit)
3961 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3965 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3972 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3975 if (!kvm->arch.vpit)
3977 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3980 case KVM_GET_PIT2: {
3982 if (!kvm->arch.vpit)
3984 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3988 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3993 case KVM_SET_PIT2: {
3995 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3998 if (!kvm->arch.vpit)
4000 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4003 case KVM_REINJECT_CONTROL: {
4004 struct kvm_reinject_control control;
4006 if (copy_from_user(&control, argp, sizeof(control)))
4008 r = kvm_vm_ioctl_reinject(kvm, &control);
4011 case KVM_XEN_HVM_CONFIG: {
4013 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4014 sizeof(struct kvm_xen_hvm_config)))
4017 if (kvm->arch.xen_hvm_config.flags)
4022 case KVM_SET_CLOCK: {
4023 struct kvm_clock_data user_ns;
4028 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4036 local_irq_disable();
4037 now_ns = get_kernel_ns();
4038 delta = user_ns.clock - now_ns;
4040 kvm->arch.kvmclock_offset = delta;
4041 kvm_gen_update_masterclock(kvm);
4044 case KVM_GET_CLOCK: {
4045 struct kvm_clock_data user_ns;
4048 local_irq_disable();
4049 now_ns = get_kernel_ns();
4050 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4053 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4056 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4063 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4069 static void kvm_init_msr_list(void)
4074 /* skip the first msrs in the list. KVM-specific */
4075 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
4076 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4080 * Even MSRs that are valid in the host may not be exposed
4081 * to the guests in some cases. We could work around this
4082 * in VMX with the generic MSR save/load machinery, but it
4083 * is not really worthwhile since it will really only
4084 * happen with nested virtualization.
4086 switch (msrs_to_save[i]) {
4087 case MSR_IA32_BNDCFGS:
4088 if (!kvm_x86_ops->mpx_supported())
4096 msrs_to_save[j] = msrs_to_save[i];
4099 num_msrs_to_save = j;
4102 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4110 if (!(vcpu->arch.apic &&
4111 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
4112 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
4123 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4130 if (!(vcpu->arch.apic &&
4131 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
4132 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
4134 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4144 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4145 struct kvm_segment *var, int seg)
4147 kvm_x86_ops->set_segment(vcpu, var, seg);
4150 void kvm_get_segment(struct kvm_vcpu *vcpu,
4151 struct kvm_segment *var, int seg)
4153 kvm_x86_ops->get_segment(vcpu, var, seg);
4156 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4157 struct x86_exception *exception)
4161 BUG_ON(!mmu_is_nested(vcpu));
4163 /* NPT walks are always user-walks */
4164 access |= PFERR_USER_MASK;
4165 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4170 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4171 struct x86_exception *exception)
4173 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4174 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4177 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4178 struct x86_exception *exception)
4180 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4181 access |= PFERR_FETCH_MASK;
4182 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4185 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4186 struct x86_exception *exception)
4188 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4189 access |= PFERR_WRITE_MASK;
4190 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4193 /* uses this to access any guest's mapped memory without checking CPL */
4194 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4195 struct x86_exception *exception)
4197 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4200 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4201 struct kvm_vcpu *vcpu, u32 access,
4202 struct x86_exception *exception)
4205 int r = X86EMUL_CONTINUE;
4208 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4210 unsigned offset = addr & (PAGE_SIZE-1);
4211 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4214 if (gpa == UNMAPPED_GVA)
4215 return X86EMUL_PROPAGATE_FAULT;
4216 ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, data,
4219 r = X86EMUL_IO_NEEDED;
4231 /* used for instruction fetching */
4232 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4233 gva_t addr, void *val, unsigned int bytes,
4234 struct x86_exception *exception)
4236 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4237 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4241 /* Inline kvm_read_guest_virt_helper for speed. */
4242 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4244 if (unlikely(gpa == UNMAPPED_GVA))
4245 return X86EMUL_PROPAGATE_FAULT;
4247 offset = addr & (PAGE_SIZE-1);
4248 if (WARN_ON(offset + bytes > PAGE_SIZE))
4249 bytes = (unsigned)PAGE_SIZE - offset;
4250 ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, val,
4252 if (unlikely(ret < 0))
4253 return X86EMUL_IO_NEEDED;
4255 return X86EMUL_CONTINUE;
4258 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4259 gva_t addr, void *val, unsigned int bytes,
4260 struct x86_exception *exception)
4262 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4263 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4265 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4268 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4270 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4271 gva_t addr, void *val, unsigned int bytes,
4272 struct x86_exception *exception)
4274 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4275 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4278 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4279 gva_t addr, void *val,
4281 struct x86_exception *exception)
4283 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4285 int r = X86EMUL_CONTINUE;
4288 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4291 unsigned offset = addr & (PAGE_SIZE-1);
4292 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4295 if (gpa == UNMAPPED_GVA)
4296 return X86EMUL_PROPAGATE_FAULT;
4297 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4299 r = X86EMUL_IO_NEEDED;
4310 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4312 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4313 gpa_t *gpa, struct x86_exception *exception,
4316 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4317 | (write ? PFERR_WRITE_MASK : 0);
4319 if (vcpu_match_mmio_gva(vcpu, gva)
4320 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4321 vcpu->arch.access, access)) {
4322 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4323 (gva & (PAGE_SIZE - 1));
4324 trace_vcpu_match_mmio(gva, *gpa, write, false);
4328 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4330 if (*gpa == UNMAPPED_GVA)
4333 /* For APIC access vmexit */
4334 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4337 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4338 trace_vcpu_match_mmio(gva, *gpa, write, true);
4345 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4346 const void *val, int bytes)
4350 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4353 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4357 struct read_write_emulator_ops {
4358 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4360 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4361 void *val, int bytes);
4362 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4363 int bytes, void *val);
4364 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4365 void *val, int bytes);
4369 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4371 if (vcpu->mmio_read_completed) {
4372 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4373 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4374 vcpu->mmio_read_completed = 0;
4381 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4382 void *val, int bytes)
4384 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4387 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4388 void *val, int bytes)
4390 return emulator_write_phys(vcpu, gpa, val, bytes);
4393 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4395 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4396 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4399 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4400 void *val, int bytes)
4402 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4403 return X86EMUL_IO_NEEDED;
4406 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4407 void *val, int bytes)
4409 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4411 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4412 return X86EMUL_CONTINUE;
4415 static const struct read_write_emulator_ops read_emultor = {
4416 .read_write_prepare = read_prepare,
4417 .read_write_emulate = read_emulate,
4418 .read_write_mmio = vcpu_mmio_read,
4419 .read_write_exit_mmio = read_exit_mmio,
4422 static const struct read_write_emulator_ops write_emultor = {
4423 .read_write_emulate = write_emulate,
4424 .read_write_mmio = write_mmio,
4425 .read_write_exit_mmio = write_exit_mmio,
4429 static int emulator_read_write_onepage(unsigned long addr, void *val,
4431 struct x86_exception *exception,
4432 struct kvm_vcpu *vcpu,
4433 const struct read_write_emulator_ops *ops)
4437 bool write = ops->write;
4438 struct kvm_mmio_fragment *frag;
4440 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4443 return X86EMUL_PROPAGATE_FAULT;
4445 /* For APIC access vmexit */
4449 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4450 return X86EMUL_CONTINUE;
4454 * Is this MMIO handled locally?
4456 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4457 if (handled == bytes)
4458 return X86EMUL_CONTINUE;
4464 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4465 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4469 return X86EMUL_CONTINUE;
4472 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4473 void *val, unsigned int bytes,
4474 struct x86_exception *exception,
4475 const struct read_write_emulator_ops *ops)
4477 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4481 if (ops->read_write_prepare &&
4482 ops->read_write_prepare(vcpu, val, bytes))
4483 return X86EMUL_CONTINUE;
4485 vcpu->mmio_nr_fragments = 0;
4487 /* Crossing a page boundary? */
4488 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4491 now = -addr & ~PAGE_MASK;
4492 rc = emulator_read_write_onepage(addr, val, now, exception,
4495 if (rc != X86EMUL_CONTINUE)
4502 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4504 if (rc != X86EMUL_CONTINUE)
4507 if (!vcpu->mmio_nr_fragments)
4510 gpa = vcpu->mmio_fragments[0].gpa;
4512 vcpu->mmio_needed = 1;
4513 vcpu->mmio_cur_fragment = 0;
4515 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4516 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4517 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4518 vcpu->run->mmio.phys_addr = gpa;
4520 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4523 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4527 struct x86_exception *exception)
4529 return emulator_read_write(ctxt, addr, val, bytes,
4530 exception, &read_emultor);
4533 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4537 struct x86_exception *exception)
4539 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4540 exception, &write_emultor);
4543 #define CMPXCHG_TYPE(t, ptr, old, new) \
4544 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4546 #ifdef CONFIG_X86_64
4547 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4549 # define CMPXCHG64(ptr, old, new) \
4550 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4553 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4558 struct x86_exception *exception)
4560 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4566 /* guests cmpxchg8b have to be emulated atomically */
4567 if (bytes > 8 || (bytes & (bytes - 1)))
4570 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4572 if (gpa == UNMAPPED_GVA ||
4573 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4576 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4579 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4580 if (is_error_page(page))
4583 kaddr = kmap_atomic(page);
4584 kaddr += offset_in_page(gpa);
4587 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4590 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4593 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4596 exchanged = CMPXCHG64(kaddr, old, new);
4601 kunmap_atomic(kaddr);
4602 kvm_release_page_dirty(page);
4605 return X86EMUL_CMPXCHG_FAILED;
4607 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
4608 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4610 return X86EMUL_CONTINUE;
4613 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4615 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4618 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4620 /* TODO: String I/O for in kernel device */
4623 if (vcpu->arch.pio.in)
4624 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4625 vcpu->arch.pio.size, pd);
4627 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4628 vcpu->arch.pio.port, vcpu->arch.pio.size,
4633 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4634 unsigned short port, void *val,
4635 unsigned int count, bool in)
4637 vcpu->arch.pio.port = port;
4638 vcpu->arch.pio.in = in;
4639 vcpu->arch.pio.count = count;
4640 vcpu->arch.pio.size = size;
4642 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4643 vcpu->arch.pio.count = 0;
4647 vcpu->run->exit_reason = KVM_EXIT_IO;
4648 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4649 vcpu->run->io.size = size;
4650 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4651 vcpu->run->io.count = count;
4652 vcpu->run->io.port = port;
4657 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4658 int size, unsigned short port, void *val,
4661 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4664 if (vcpu->arch.pio.count)
4667 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4670 memcpy(val, vcpu->arch.pio_data, size * count);
4671 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4672 vcpu->arch.pio.count = 0;
4679 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4680 int size, unsigned short port,
4681 const void *val, unsigned int count)
4683 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4685 memcpy(vcpu->arch.pio_data, val, size * count);
4686 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4687 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4690 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4692 return kvm_x86_ops->get_segment_base(vcpu, seg);
4695 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4697 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4700 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4702 if (!need_emulate_wbinvd(vcpu))
4703 return X86EMUL_CONTINUE;
4705 if (kvm_x86_ops->has_wbinvd_exit()) {
4706 int cpu = get_cpu();
4708 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4709 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4710 wbinvd_ipi, NULL, 1);
4712 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4715 return X86EMUL_CONTINUE;
4717 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4719 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4721 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4724 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4726 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4729 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4732 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4735 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4737 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4740 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4742 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4743 unsigned long value;
4747 value = kvm_read_cr0(vcpu);
4750 value = vcpu->arch.cr2;
4753 value = kvm_read_cr3(vcpu);
4756 value = kvm_read_cr4(vcpu);
4759 value = kvm_get_cr8(vcpu);
4762 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4769 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4771 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4776 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4779 vcpu->arch.cr2 = val;
4782 res = kvm_set_cr3(vcpu, val);
4785 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4788 res = kvm_set_cr8(vcpu, val);
4791 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4798 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4800 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4803 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4805 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4808 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4810 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4813 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4815 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4818 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4820 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4823 static unsigned long emulator_get_cached_segment_base(
4824 struct x86_emulate_ctxt *ctxt, int seg)
4826 return get_segment_base(emul_to_vcpu(ctxt), seg);
4829 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4830 struct desc_struct *desc, u32 *base3,
4833 struct kvm_segment var;
4835 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4836 *selector = var.selector;
4839 memset(desc, 0, sizeof(*desc));
4845 set_desc_limit(desc, var.limit);
4846 set_desc_base(desc, (unsigned long)var.base);
4847 #ifdef CONFIG_X86_64
4849 *base3 = var.base >> 32;
4851 desc->type = var.type;
4853 desc->dpl = var.dpl;
4854 desc->p = var.present;
4855 desc->avl = var.avl;
4863 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4864 struct desc_struct *desc, u32 base3,
4867 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4868 struct kvm_segment var;
4870 var.selector = selector;
4871 var.base = get_desc_base(desc);
4872 #ifdef CONFIG_X86_64
4873 var.base |= ((u64)base3) << 32;
4875 var.limit = get_desc_limit(desc);
4877 var.limit = (var.limit << 12) | 0xfff;
4878 var.type = desc->type;
4879 var.dpl = desc->dpl;
4884 var.avl = desc->avl;
4885 var.present = desc->p;
4886 var.unusable = !var.present;
4889 kvm_set_segment(vcpu, &var, seg);
4893 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4894 u32 msr_index, u64 *pdata)
4896 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4899 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4900 u32 msr_index, u64 data)
4902 struct msr_data msr;
4905 msr.index = msr_index;
4906 msr.host_initiated = false;
4907 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4910 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4913 return kvm_pmu_check_pmc(emul_to_vcpu(ctxt), pmc);
4916 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4917 u32 pmc, u64 *pdata)
4919 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4922 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4924 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4927 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4930 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4932 * CR0.TS may reference the host fpu state, not the guest fpu state,
4933 * so it may be clear at this point.
4938 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4943 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4944 struct x86_instruction_info *info,
4945 enum x86_intercept_stage stage)
4947 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4950 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4951 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4953 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4956 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4958 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4961 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4963 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4966 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
4968 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
4971 static const struct x86_emulate_ops emulate_ops = {
4972 .read_gpr = emulator_read_gpr,
4973 .write_gpr = emulator_write_gpr,
4974 .read_std = kvm_read_guest_virt_system,
4975 .write_std = kvm_write_guest_virt_system,
4976 .fetch = kvm_fetch_guest_virt,
4977 .read_emulated = emulator_read_emulated,
4978 .write_emulated = emulator_write_emulated,
4979 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4980 .invlpg = emulator_invlpg,
4981 .pio_in_emulated = emulator_pio_in_emulated,
4982 .pio_out_emulated = emulator_pio_out_emulated,
4983 .get_segment = emulator_get_segment,
4984 .set_segment = emulator_set_segment,
4985 .get_cached_segment_base = emulator_get_cached_segment_base,
4986 .get_gdt = emulator_get_gdt,
4987 .get_idt = emulator_get_idt,
4988 .set_gdt = emulator_set_gdt,
4989 .set_idt = emulator_set_idt,
4990 .get_cr = emulator_get_cr,
4991 .set_cr = emulator_set_cr,
4992 .cpl = emulator_get_cpl,
4993 .get_dr = emulator_get_dr,
4994 .set_dr = emulator_set_dr,
4995 .set_msr = emulator_set_msr,
4996 .get_msr = emulator_get_msr,
4997 .check_pmc = emulator_check_pmc,
4998 .read_pmc = emulator_read_pmc,
4999 .halt = emulator_halt,
5000 .wbinvd = emulator_wbinvd,
5001 .fix_hypercall = emulator_fix_hypercall,
5002 .get_fpu = emulator_get_fpu,
5003 .put_fpu = emulator_put_fpu,
5004 .intercept = emulator_intercept,
5005 .get_cpuid = emulator_get_cpuid,
5006 .set_nmi_mask = emulator_set_nmi_mask,
5009 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5011 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5013 * an sti; sti; sequence only disable interrupts for the first
5014 * instruction. So, if the last instruction, be it emulated or
5015 * not, left the system with the INT_STI flag enabled, it
5016 * means that the last instruction is an sti. We should not
5017 * leave the flag on in this case. The same goes for mov ss
5019 if (int_shadow & mask)
5021 if (unlikely(int_shadow || mask)) {
5022 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5024 kvm_make_request(KVM_REQ_EVENT, vcpu);
5028 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5030 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5031 if (ctxt->exception.vector == PF_VECTOR)
5032 return kvm_propagate_fault(vcpu, &ctxt->exception);
5034 if (ctxt->exception.error_code_valid)
5035 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5036 ctxt->exception.error_code);
5038 kvm_queue_exception(vcpu, ctxt->exception.vector);
5042 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5044 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5047 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5049 ctxt->eflags = kvm_get_rflags(vcpu);
5050 ctxt->eip = kvm_rip_read(vcpu);
5051 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5052 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5053 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5054 cs_db ? X86EMUL_MODE_PROT32 :
5055 X86EMUL_MODE_PROT16;
5056 ctxt->guest_mode = is_guest_mode(vcpu);
5058 init_decode_cache(ctxt);
5059 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5062 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5064 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5067 init_emulate_ctxt(vcpu);
5071 ctxt->_eip = ctxt->eip + inc_eip;
5072 ret = emulate_int_real(ctxt, irq);
5074 if (ret != X86EMUL_CONTINUE)
5075 return EMULATE_FAIL;
5077 ctxt->eip = ctxt->_eip;
5078 kvm_rip_write(vcpu, ctxt->eip);
5079 kvm_set_rflags(vcpu, ctxt->eflags);
5081 if (irq == NMI_VECTOR)
5082 vcpu->arch.nmi_pending = 0;
5084 vcpu->arch.interrupt.pending = false;
5086 return EMULATE_DONE;
5088 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5090 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5092 int r = EMULATE_DONE;
5094 ++vcpu->stat.insn_emulation_fail;
5095 trace_kvm_emulate_insn_failed(vcpu);
5096 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5097 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5098 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5099 vcpu->run->internal.ndata = 0;
5102 kvm_queue_exception(vcpu, UD_VECTOR);
5107 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5108 bool write_fault_to_shadow_pgtable,
5114 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5117 if (!vcpu->arch.mmu.direct_map) {
5119 * Write permission should be allowed since only
5120 * write access need to be emulated.
5122 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5125 * If the mapping is invalid in guest, let cpu retry
5126 * it to generate fault.
5128 if (gpa == UNMAPPED_GVA)
5133 * Do not retry the unhandleable instruction if it faults on the
5134 * readonly host memory, otherwise it will goto a infinite loop:
5135 * retry instruction -> write #PF -> emulation fail -> retry
5136 * instruction -> ...
5138 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5141 * If the instruction failed on the error pfn, it can not be fixed,
5142 * report the error to userspace.
5144 if (is_error_noslot_pfn(pfn))
5147 kvm_release_pfn_clean(pfn);
5149 /* The instructions are well-emulated on direct mmu. */
5150 if (vcpu->arch.mmu.direct_map) {
5151 unsigned int indirect_shadow_pages;
5153 spin_lock(&vcpu->kvm->mmu_lock);
5154 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5155 spin_unlock(&vcpu->kvm->mmu_lock);
5157 if (indirect_shadow_pages)
5158 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5164 * if emulation was due to access to shadowed page table
5165 * and it failed try to unshadow page and re-enter the
5166 * guest to let CPU execute the instruction.
5168 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5171 * If the access faults on its page table, it can not
5172 * be fixed by unprotecting shadow page and it should
5173 * be reported to userspace.
5175 return !write_fault_to_shadow_pgtable;
5178 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5179 unsigned long cr2, int emulation_type)
5181 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5182 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5184 last_retry_eip = vcpu->arch.last_retry_eip;
5185 last_retry_addr = vcpu->arch.last_retry_addr;
5188 * If the emulation is caused by #PF and it is non-page_table
5189 * writing instruction, it means the VM-EXIT is caused by shadow
5190 * page protected, we can zap the shadow page and retry this
5191 * instruction directly.
5193 * Note: if the guest uses a non-page-table modifying instruction
5194 * on the PDE that points to the instruction, then we will unmap
5195 * the instruction and go to an infinite loop. So, we cache the
5196 * last retried eip and the last fault address, if we meet the eip
5197 * and the address again, we can break out of the potential infinite
5200 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5202 if (!(emulation_type & EMULTYPE_RETRY))
5205 if (x86_page_table_writing_insn(ctxt))
5208 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5211 vcpu->arch.last_retry_eip = ctxt->eip;
5212 vcpu->arch.last_retry_addr = cr2;
5214 if (!vcpu->arch.mmu.direct_map)
5215 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5217 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5222 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5223 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5225 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5234 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5235 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5240 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5242 struct kvm_run *kvm_run = vcpu->run;
5245 * rflags is the old, "raw" value of the flags. The new value has
5246 * not been saved yet.
5248 * This is correct even for TF set by the guest, because "the
5249 * processor will not generate this exception after the instruction
5250 * that sets the TF flag".
5252 if (unlikely(rflags & X86_EFLAGS_TF)) {
5253 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5254 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5256 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5257 kvm_run->debug.arch.exception = DB_VECTOR;
5258 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5259 *r = EMULATE_USER_EXIT;
5261 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5263 * "Certain debug exceptions may clear bit 0-3. The
5264 * remaining contents of the DR6 register are never
5265 * cleared by the processor".
5267 vcpu->arch.dr6 &= ~15;
5268 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5269 kvm_queue_exception(vcpu, DB_VECTOR);
5274 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5276 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5277 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5278 struct kvm_run *kvm_run = vcpu->run;
5279 unsigned long eip = kvm_get_linear_rip(vcpu);
5280 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5281 vcpu->arch.guest_debug_dr7,
5285 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5286 kvm_run->debug.arch.pc = eip;
5287 kvm_run->debug.arch.exception = DB_VECTOR;
5288 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5289 *r = EMULATE_USER_EXIT;
5294 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5295 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5296 unsigned long eip = kvm_get_linear_rip(vcpu);
5297 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5302 vcpu->arch.dr6 &= ~15;
5303 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5304 kvm_queue_exception(vcpu, DB_VECTOR);
5313 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5320 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5321 bool writeback = true;
5322 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5325 * Clear write_fault_to_shadow_pgtable here to ensure it is
5328 vcpu->arch.write_fault_to_shadow_pgtable = false;
5329 kvm_clear_exception_queue(vcpu);
5331 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5332 init_emulate_ctxt(vcpu);
5335 * We will reenter on the same instruction since
5336 * we do not set complete_userspace_io. This does not
5337 * handle watchpoints yet, those would be handled in
5340 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5343 ctxt->interruptibility = 0;
5344 ctxt->have_exception = false;
5345 ctxt->exception.vector = -1;
5346 ctxt->perm_ok = false;
5348 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5350 r = x86_decode_insn(ctxt, insn, insn_len);
5352 trace_kvm_emulate_insn_start(vcpu);
5353 ++vcpu->stat.insn_emulation;
5354 if (r != EMULATION_OK) {
5355 if (emulation_type & EMULTYPE_TRAP_UD)
5356 return EMULATE_FAIL;
5357 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5359 return EMULATE_DONE;
5360 if (emulation_type & EMULTYPE_SKIP)
5361 return EMULATE_FAIL;
5362 return handle_emulation_failure(vcpu);
5366 if (emulation_type & EMULTYPE_SKIP) {
5367 kvm_rip_write(vcpu, ctxt->_eip);
5368 if (ctxt->eflags & X86_EFLAGS_RF)
5369 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5370 return EMULATE_DONE;
5373 if (retry_instruction(ctxt, cr2, emulation_type))
5374 return EMULATE_DONE;
5376 /* this is needed for vmware backdoor interface to work since it
5377 changes registers values during IO operation */
5378 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5379 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5380 emulator_invalidate_register_cache(ctxt);
5384 r = x86_emulate_insn(ctxt);
5386 if (r == EMULATION_INTERCEPTED)
5387 return EMULATE_DONE;
5389 if (r == EMULATION_FAILED) {
5390 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5392 return EMULATE_DONE;
5394 return handle_emulation_failure(vcpu);
5397 if (ctxt->have_exception) {
5399 if (inject_emulated_exception(vcpu))
5401 } else if (vcpu->arch.pio.count) {
5402 if (!vcpu->arch.pio.in) {
5403 /* FIXME: return into emulator if single-stepping. */
5404 vcpu->arch.pio.count = 0;
5407 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5409 r = EMULATE_USER_EXIT;
5410 } else if (vcpu->mmio_needed) {
5411 if (!vcpu->mmio_is_write)
5413 r = EMULATE_USER_EXIT;
5414 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5415 } else if (r == EMULATION_RESTART)
5421 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5422 toggle_interruptibility(vcpu, ctxt->interruptibility);
5423 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5424 kvm_rip_write(vcpu, ctxt->eip);
5425 if (r == EMULATE_DONE)
5426 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5427 if (!ctxt->have_exception ||
5428 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5429 __kvm_set_rflags(vcpu, ctxt->eflags);
5432 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5433 * do nothing, and it will be requested again as soon as
5434 * the shadow expires. But we still need to check here,
5435 * because POPF has no interrupt shadow.
5437 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5438 kvm_make_request(KVM_REQ_EVENT, vcpu);
5440 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5444 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5446 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5448 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5449 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5450 size, port, &val, 1);
5451 /* do not return to emulator after return from userspace */
5452 vcpu->arch.pio.count = 0;
5455 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5457 static void tsc_bad(void *info)
5459 __this_cpu_write(cpu_tsc_khz, 0);
5462 static void tsc_khz_changed(void *data)
5464 struct cpufreq_freqs *freq = data;
5465 unsigned long khz = 0;
5469 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5470 khz = cpufreq_quick_get(raw_smp_processor_id());
5473 __this_cpu_write(cpu_tsc_khz, khz);
5476 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5479 struct cpufreq_freqs *freq = data;
5481 struct kvm_vcpu *vcpu;
5482 int i, send_ipi = 0;
5485 * We allow guests to temporarily run on slowing clocks,
5486 * provided we notify them after, or to run on accelerating
5487 * clocks, provided we notify them before. Thus time never
5490 * However, we have a problem. We can't atomically update
5491 * the frequency of a given CPU from this function; it is
5492 * merely a notifier, which can be called from any CPU.
5493 * Changing the TSC frequency at arbitrary points in time
5494 * requires a recomputation of local variables related to
5495 * the TSC for each VCPU. We must flag these local variables
5496 * to be updated and be sure the update takes place with the
5497 * new frequency before any guests proceed.
5499 * Unfortunately, the combination of hotplug CPU and frequency
5500 * change creates an intractable locking scenario; the order
5501 * of when these callouts happen is undefined with respect to
5502 * CPU hotplug, and they can race with each other. As such,
5503 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5504 * undefined; you can actually have a CPU frequency change take
5505 * place in between the computation of X and the setting of the
5506 * variable. To protect against this problem, all updates of
5507 * the per_cpu tsc_khz variable are done in an interrupt
5508 * protected IPI, and all callers wishing to update the value
5509 * must wait for a synchronous IPI to complete (which is trivial
5510 * if the caller is on the CPU already). This establishes the
5511 * necessary total order on variable updates.
5513 * Note that because a guest time update may take place
5514 * anytime after the setting of the VCPU's request bit, the
5515 * correct TSC value must be set before the request. However,
5516 * to ensure the update actually makes it to any guest which
5517 * starts running in hardware virtualization between the set
5518 * and the acquisition of the spinlock, we must also ping the
5519 * CPU after setting the request bit.
5523 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5525 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5528 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5530 spin_lock(&kvm_lock);
5531 list_for_each_entry(kvm, &vm_list, vm_list) {
5532 kvm_for_each_vcpu(i, vcpu, kvm) {
5533 if (vcpu->cpu != freq->cpu)
5535 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5536 if (vcpu->cpu != smp_processor_id())
5540 spin_unlock(&kvm_lock);
5542 if (freq->old < freq->new && send_ipi) {
5544 * We upscale the frequency. Must make the guest
5545 * doesn't see old kvmclock values while running with
5546 * the new frequency, otherwise we risk the guest sees
5547 * time go backwards.
5549 * In case we update the frequency for another cpu
5550 * (which might be in guest context) send an interrupt
5551 * to kick the cpu out of guest context. Next time
5552 * guest context is entered kvmclock will be updated,
5553 * so the guest will not see stale values.
5555 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5560 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5561 .notifier_call = kvmclock_cpufreq_notifier
5564 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5565 unsigned long action, void *hcpu)
5567 unsigned int cpu = (unsigned long)hcpu;
5571 case CPU_DOWN_FAILED:
5572 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5574 case CPU_DOWN_PREPARE:
5575 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5581 static struct notifier_block kvmclock_cpu_notifier_block = {
5582 .notifier_call = kvmclock_cpu_notifier,
5583 .priority = -INT_MAX
5586 static void kvm_timer_init(void)
5590 max_tsc_khz = tsc_khz;
5592 cpu_notifier_register_begin();
5593 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5594 #ifdef CONFIG_CPU_FREQ
5595 struct cpufreq_policy policy;
5596 memset(&policy, 0, sizeof(policy));
5598 cpufreq_get_policy(&policy, cpu);
5599 if (policy.cpuinfo.max_freq)
5600 max_tsc_khz = policy.cpuinfo.max_freq;
5603 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5604 CPUFREQ_TRANSITION_NOTIFIER);
5606 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5607 for_each_online_cpu(cpu)
5608 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5610 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5611 cpu_notifier_register_done();
5615 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5617 int kvm_is_in_guest(void)
5619 return __this_cpu_read(current_vcpu) != NULL;
5622 static int kvm_is_user_mode(void)
5626 if (__this_cpu_read(current_vcpu))
5627 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5629 return user_mode != 0;
5632 static unsigned long kvm_get_guest_ip(void)
5634 unsigned long ip = 0;
5636 if (__this_cpu_read(current_vcpu))
5637 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5642 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5643 .is_in_guest = kvm_is_in_guest,
5644 .is_user_mode = kvm_is_user_mode,
5645 .get_guest_ip = kvm_get_guest_ip,
5648 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5650 __this_cpu_write(current_vcpu, vcpu);
5652 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5654 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5656 __this_cpu_write(current_vcpu, NULL);
5658 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5660 static void kvm_set_mmio_spte_mask(void)
5663 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5666 * Set the reserved bits and the present bit of an paging-structure
5667 * entry to generate page fault with PFER.RSV = 1.
5669 /* Mask the reserved physical address bits. */
5670 mask = rsvd_bits(maxphyaddr, 51);
5672 /* Bit 62 is always reserved for 32bit host. */
5673 mask |= 0x3ull << 62;
5675 /* Set the present bit. */
5678 #ifdef CONFIG_X86_64
5680 * If reserved bit is not supported, clear the present bit to disable
5683 if (maxphyaddr == 52)
5687 kvm_mmu_set_mmio_spte_mask(mask);
5690 #ifdef CONFIG_X86_64
5691 static void pvclock_gtod_update_fn(struct work_struct *work)
5695 struct kvm_vcpu *vcpu;
5698 spin_lock(&kvm_lock);
5699 list_for_each_entry(kvm, &vm_list, vm_list)
5700 kvm_for_each_vcpu(i, vcpu, kvm)
5701 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5702 atomic_set(&kvm_guest_has_master_clock, 0);
5703 spin_unlock(&kvm_lock);
5706 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5709 * Notification about pvclock gtod data update.
5711 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5714 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5715 struct timekeeper *tk = priv;
5717 update_pvclock_gtod(tk);
5719 /* disable master clock if host does not trust, or does not
5720 * use, TSC clocksource
5722 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5723 atomic_read(&kvm_guest_has_master_clock) != 0)
5724 queue_work(system_long_wq, &pvclock_gtod_work);
5729 static struct notifier_block pvclock_gtod_notifier = {
5730 .notifier_call = pvclock_gtod_notify,
5734 int kvm_arch_init(void *opaque)
5737 struct kvm_x86_ops *ops = opaque;
5740 printk(KERN_ERR "kvm: already loaded the other module\n");
5745 if (!ops->cpu_has_kvm_support()) {
5746 printk(KERN_ERR "kvm: no hardware support\n");
5750 if (ops->disabled_by_bios()) {
5751 printk(KERN_ERR "kvm: disabled by bios\n");
5757 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5759 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5763 r = kvm_mmu_module_init();
5765 goto out_free_percpu;
5767 kvm_set_mmio_spte_mask();
5770 kvm_init_msr_list();
5772 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5773 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5777 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5780 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5783 #ifdef CONFIG_X86_64
5784 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5790 free_percpu(shared_msrs);
5795 void kvm_arch_exit(void)
5797 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5799 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5800 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5801 CPUFREQ_TRANSITION_NOTIFIER);
5802 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5803 #ifdef CONFIG_X86_64
5804 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5807 kvm_mmu_module_exit();
5808 free_percpu(shared_msrs);
5811 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5813 ++vcpu->stat.halt_exits;
5814 if (irqchip_in_kernel(vcpu->kvm)) {
5815 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5818 vcpu->run->exit_reason = KVM_EXIT_HLT;
5822 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5824 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5826 u64 param, ingpa, outgpa, ret;
5827 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5828 bool fast, longmode;
5831 * hypercall generates UD from non zero cpl and real mode
5834 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5835 kvm_queue_exception(vcpu, UD_VECTOR);
5839 longmode = is_64_bit_mode(vcpu);
5842 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5843 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5844 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5845 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5846 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5847 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5849 #ifdef CONFIG_X86_64
5851 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5852 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5853 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5857 code = param & 0xffff;
5858 fast = (param >> 16) & 0x1;
5859 rep_cnt = (param >> 32) & 0xfff;
5860 rep_idx = (param >> 48) & 0xfff;
5862 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5865 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5866 kvm_vcpu_on_spin(vcpu);
5869 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5873 ret = res | (((u64)rep_done & 0xfff) << 32);
5875 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5877 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5878 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5885 * kvm_pv_kick_cpu_op: Kick a vcpu.
5887 * @apicid - apicid of vcpu to be kicked.
5889 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5891 struct kvm_lapic_irq lapic_irq;
5893 lapic_irq.shorthand = 0;
5894 lapic_irq.dest_mode = 0;
5895 lapic_irq.dest_id = apicid;
5897 lapic_irq.delivery_mode = APIC_DM_REMRD;
5898 kvm_irq_delivery_to_apic(kvm, 0, &lapic_irq, NULL);
5901 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5903 unsigned long nr, a0, a1, a2, a3, ret;
5904 int op_64_bit, r = 1;
5906 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5907 return kvm_hv_hypercall(vcpu);
5909 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5910 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5911 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5912 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5913 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5915 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5917 op_64_bit = is_64_bit_mode(vcpu);
5926 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5932 case KVM_HC_VAPIC_POLL_IRQ:
5935 case KVM_HC_KICK_CPU:
5936 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5946 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5947 ++vcpu->stat.hypercalls;
5950 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5952 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5954 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5955 char instruction[3];
5956 unsigned long rip = kvm_rip_read(vcpu);
5958 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5960 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5964 * Check if userspace requested an interrupt window, and that the
5965 * interrupt window is open.
5967 * No need to exit to userspace if we already have an interrupt queued.
5969 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5971 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5972 vcpu->run->request_interrupt_window &&
5973 kvm_arch_interrupt_allowed(vcpu));
5976 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5978 struct kvm_run *kvm_run = vcpu->run;
5980 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5981 kvm_run->cr8 = kvm_get_cr8(vcpu);
5982 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5983 if (irqchip_in_kernel(vcpu->kvm))
5984 kvm_run->ready_for_interrupt_injection = 1;
5986 kvm_run->ready_for_interrupt_injection =
5987 kvm_arch_interrupt_allowed(vcpu) &&
5988 !kvm_cpu_has_interrupt(vcpu) &&
5989 !kvm_event_needs_reinjection(vcpu);
5992 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5996 if (!kvm_x86_ops->update_cr8_intercept)
5999 if (!vcpu->arch.apic)
6002 if (!vcpu->arch.apic->vapic_addr)
6003 max_irr = kvm_lapic_find_highest_irr(vcpu);
6010 tpr = kvm_lapic_get_cr8(vcpu);
6012 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6015 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6019 /* try to reinject previous events if any */
6020 if (vcpu->arch.exception.pending) {
6021 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6022 vcpu->arch.exception.has_error_code,
6023 vcpu->arch.exception.error_code);
6025 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6026 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6029 if (vcpu->arch.exception.nr == DB_VECTOR &&
6030 (vcpu->arch.dr7 & DR7_GD)) {
6031 vcpu->arch.dr7 &= ~DR7_GD;
6032 kvm_update_dr7(vcpu);
6035 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6036 vcpu->arch.exception.has_error_code,
6037 vcpu->arch.exception.error_code,
6038 vcpu->arch.exception.reinject);
6042 if (vcpu->arch.nmi_injected) {
6043 kvm_x86_ops->set_nmi(vcpu);
6047 if (vcpu->arch.interrupt.pending) {
6048 kvm_x86_ops->set_irq(vcpu);
6052 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6053 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6058 /* try to inject new event if pending */
6059 if (vcpu->arch.nmi_pending) {
6060 if (kvm_x86_ops->nmi_allowed(vcpu)) {
6061 --vcpu->arch.nmi_pending;
6062 vcpu->arch.nmi_injected = true;
6063 kvm_x86_ops->set_nmi(vcpu);
6065 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6067 * Because interrupts can be injected asynchronously, we are
6068 * calling check_nested_events again here to avoid a race condition.
6069 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6070 * proposal and current concerns. Perhaps we should be setting
6071 * KVM_REQ_EVENT only on certain events and not unconditionally?
6073 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6074 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6078 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6079 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6081 kvm_x86_ops->set_irq(vcpu);
6087 static void process_nmi(struct kvm_vcpu *vcpu)
6092 * x86 is limited to one NMI running, and one NMI pending after it.
6093 * If an NMI is already in progress, limit further NMIs to just one.
6094 * Otherwise, allow two (and we'll inject the first one immediately).
6096 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6099 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6100 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6101 kvm_make_request(KVM_REQ_EVENT, vcpu);
6104 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6106 u64 eoi_exit_bitmap[4];
6109 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6112 memset(eoi_exit_bitmap, 0, 32);
6115 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6116 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6117 kvm_apic_update_tmr(vcpu, tmr);
6120 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6122 ++vcpu->stat.tlb_flush;
6123 kvm_x86_ops->tlb_flush(vcpu);
6126 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6128 struct page *page = NULL;
6130 if (!irqchip_in_kernel(vcpu->kvm))
6133 if (!kvm_x86_ops->set_apic_access_page_addr)
6136 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6137 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6140 * Do not pin apic access page in memory, the MMU notifier
6141 * will call us again if it is migrated or swapped out.
6145 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6147 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6148 unsigned long address)
6151 * The physical address of apic access page is stored in the VMCS.
6152 * Update it when it becomes invalid.
6154 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6155 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6159 * Returns 1 to let __vcpu_run() continue the guest execution loop without
6160 * exiting to the userspace. Otherwise, the value will be returned to the
6163 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6166 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6167 vcpu->run->request_interrupt_window;
6168 bool req_immediate_exit = false;
6170 if (vcpu->requests) {
6171 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6172 kvm_mmu_unload(vcpu);
6173 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6174 __kvm_migrate_timers(vcpu);
6175 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6176 kvm_gen_update_masterclock(vcpu->kvm);
6177 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6178 kvm_gen_kvmclock_update(vcpu);
6179 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6180 r = kvm_guest_time_update(vcpu);
6184 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6185 kvm_mmu_sync_roots(vcpu);
6186 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6187 kvm_vcpu_flush_tlb(vcpu);
6188 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6189 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6193 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6194 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6198 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6199 vcpu->fpu_active = 0;
6200 kvm_x86_ops->fpu_deactivate(vcpu);
6202 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6203 /* Page is swapped out. Do synthetic halt */
6204 vcpu->arch.apf.halted = true;
6208 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6209 record_steal_time(vcpu);
6210 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6212 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6213 kvm_handle_pmu_event(vcpu);
6214 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6215 kvm_deliver_pmi(vcpu);
6216 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6217 vcpu_scan_ioapic(vcpu);
6218 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6219 kvm_vcpu_reload_apic_access_page(vcpu);
6222 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6223 kvm_apic_accept_events(vcpu);
6224 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6229 if (inject_pending_event(vcpu, req_int_win) != 0)
6230 req_immediate_exit = true;
6231 /* enable NMI/IRQ window open exits if needed */
6232 else if (vcpu->arch.nmi_pending)
6233 kvm_x86_ops->enable_nmi_window(vcpu);
6234 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6235 kvm_x86_ops->enable_irq_window(vcpu);
6237 if (kvm_lapic_enabled(vcpu)) {
6239 * Update architecture specific hints for APIC
6240 * virtual interrupt delivery.
6242 if (kvm_x86_ops->hwapic_irr_update)
6243 kvm_x86_ops->hwapic_irr_update(vcpu,
6244 kvm_lapic_find_highest_irr(vcpu));
6245 update_cr8_intercept(vcpu);
6246 kvm_lapic_sync_to_vapic(vcpu);
6250 r = kvm_mmu_reload(vcpu);
6252 goto cancel_injection;
6257 kvm_x86_ops->prepare_guest_switch(vcpu);
6258 if (vcpu->fpu_active)
6259 kvm_load_guest_fpu(vcpu);
6260 kvm_load_guest_xcr0(vcpu);
6262 vcpu->mode = IN_GUEST_MODE;
6264 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6266 /* We should set ->mode before check ->requests,
6267 * see the comment in make_all_cpus_request.
6269 smp_mb__after_srcu_read_unlock();
6271 local_irq_disable();
6273 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6274 || need_resched() || signal_pending(current)) {
6275 vcpu->mode = OUTSIDE_GUEST_MODE;
6279 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6281 goto cancel_injection;
6284 if (req_immediate_exit)
6285 smp_send_reschedule(vcpu->cpu);
6289 if (unlikely(vcpu->arch.switch_db_regs)) {
6291 set_debugreg(vcpu->arch.eff_db[0], 0);
6292 set_debugreg(vcpu->arch.eff_db[1], 1);
6293 set_debugreg(vcpu->arch.eff_db[2], 2);
6294 set_debugreg(vcpu->arch.eff_db[3], 3);
6295 set_debugreg(vcpu->arch.dr6, 6);
6298 trace_kvm_entry(vcpu->vcpu_id);
6299 wait_lapic_expire(vcpu);
6300 kvm_x86_ops->run(vcpu);
6303 * Do this here before restoring debug registers on the host. And
6304 * since we do this before handling the vmexit, a DR access vmexit
6305 * can (a) read the correct value of the debug registers, (b) set
6306 * KVM_DEBUGREG_WONT_EXIT again.
6308 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6311 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6312 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6313 for (i = 0; i < KVM_NR_DB_REGS; i++)
6314 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6318 * If the guest has used debug registers, at least dr7
6319 * will be disabled while returning to the host.
6320 * If we don't have active breakpoints in the host, we don't
6321 * care about the messed up debug address registers. But if
6322 * we have some of them active, restore the old state.
6324 if (hw_breakpoint_active())
6325 hw_breakpoint_restore();
6327 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6330 vcpu->mode = OUTSIDE_GUEST_MODE;
6333 /* Interrupt is enabled by handle_external_intr() */
6334 kvm_x86_ops->handle_external_intr(vcpu);
6339 * We must have an instruction between local_irq_enable() and
6340 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6341 * the interrupt shadow. The stat.exits increment will do nicely.
6342 * But we need to prevent reordering, hence this barrier():
6350 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6353 * Profile KVM exit RIPs:
6355 if (unlikely(prof_on == KVM_PROFILING)) {
6356 unsigned long rip = kvm_rip_read(vcpu);
6357 profile_hit(KVM_PROFILING, (void *)rip);
6360 if (unlikely(vcpu->arch.tsc_always_catchup))
6361 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6363 if (vcpu->arch.apic_attention)
6364 kvm_lapic_sync_from_vapic(vcpu);
6366 r = kvm_x86_ops->handle_exit(vcpu);
6370 kvm_x86_ops->cancel_injection(vcpu);
6371 if (unlikely(vcpu->arch.apic_attention))
6372 kvm_lapic_sync_from_vapic(vcpu);
6378 static int __vcpu_run(struct kvm_vcpu *vcpu)
6381 struct kvm *kvm = vcpu->kvm;
6383 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6387 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6388 !vcpu->arch.apf.halted)
6389 r = vcpu_enter_guest(vcpu);
6391 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6392 kvm_vcpu_block(vcpu);
6393 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6394 if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
6395 kvm_apic_accept_events(vcpu);
6396 switch(vcpu->arch.mp_state) {
6397 case KVM_MP_STATE_HALTED:
6398 vcpu->arch.pv.pv_unhalted = false;
6399 vcpu->arch.mp_state =
6400 KVM_MP_STATE_RUNNABLE;
6401 case KVM_MP_STATE_RUNNABLE:
6402 vcpu->arch.apf.halted = false;
6404 case KVM_MP_STATE_INIT_RECEIVED:
6416 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6417 if (kvm_cpu_has_pending_timer(vcpu))
6418 kvm_inject_pending_timer_irqs(vcpu);
6420 if (dm_request_for_irq_injection(vcpu)) {
6422 vcpu->run->exit_reason = KVM_EXIT_INTR;
6423 ++vcpu->stat.request_irq_exits;
6426 kvm_check_async_pf_completion(vcpu);
6428 if (signal_pending(current)) {
6430 vcpu->run->exit_reason = KVM_EXIT_INTR;
6431 ++vcpu->stat.signal_exits;
6433 if (need_resched()) {
6434 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6436 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6440 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6445 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6448 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6449 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6450 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6451 if (r != EMULATE_DONE)
6456 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6458 BUG_ON(!vcpu->arch.pio.count);
6460 return complete_emulated_io(vcpu);
6464 * Implements the following, as a state machine:
6468 * for each mmio piece in the fragment
6476 * for each mmio piece in the fragment
6481 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6483 struct kvm_run *run = vcpu->run;
6484 struct kvm_mmio_fragment *frag;
6487 BUG_ON(!vcpu->mmio_needed);
6489 /* Complete previous fragment */
6490 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6491 len = min(8u, frag->len);
6492 if (!vcpu->mmio_is_write)
6493 memcpy(frag->data, run->mmio.data, len);
6495 if (frag->len <= 8) {
6496 /* Switch to the next fragment. */
6498 vcpu->mmio_cur_fragment++;
6500 /* Go forward to the next mmio piece. */
6506 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6507 vcpu->mmio_needed = 0;
6509 /* FIXME: return into emulator if single-stepping. */
6510 if (vcpu->mmio_is_write)
6512 vcpu->mmio_read_completed = 1;
6513 return complete_emulated_io(vcpu);
6516 run->exit_reason = KVM_EXIT_MMIO;
6517 run->mmio.phys_addr = frag->gpa;
6518 if (vcpu->mmio_is_write)
6519 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6520 run->mmio.len = min(8u, frag->len);
6521 run->mmio.is_write = vcpu->mmio_is_write;
6522 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6527 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6532 if (!tsk_used_math(current) && init_fpu(current))
6535 if (vcpu->sigset_active)
6536 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6538 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6539 kvm_vcpu_block(vcpu);
6540 kvm_apic_accept_events(vcpu);
6541 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6546 /* re-sync apic's tpr */
6547 if (!irqchip_in_kernel(vcpu->kvm)) {
6548 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6554 if (unlikely(vcpu->arch.complete_userspace_io)) {
6555 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6556 vcpu->arch.complete_userspace_io = NULL;
6561 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6563 r = __vcpu_run(vcpu);
6566 post_kvm_run_save(vcpu);
6567 if (vcpu->sigset_active)
6568 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6573 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6575 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6577 * We are here if userspace calls get_regs() in the middle of
6578 * instruction emulation. Registers state needs to be copied
6579 * back from emulation context to vcpu. Userspace shouldn't do
6580 * that usually, but some bad designed PV devices (vmware
6581 * backdoor interface) need this to work
6583 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6584 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6586 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6587 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6588 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6589 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6590 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6591 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6592 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6593 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6594 #ifdef CONFIG_X86_64
6595 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6596 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6597 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6598 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6599 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6600 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6601 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6602 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6605 regs->rip = kvm_rip_read(vcpu);
6606 regs->rflags = kvm_get_rflags(vcpu);
6611 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6613 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6614 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6616 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6617 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6618 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6619 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6620 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6621 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6622 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6623 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6624 #ifdef CONFIG_X86_64
6625 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6626 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6627 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6628 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6629 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6630 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6631 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6632 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6635 kvm_rip_write(vcpu, regs->rip);
6636 kvm_set_rflags(vcpu, regs->rflags);
6638 vcpu->arch.exception.pending = false;
6640 kvm_make_request(KVM_REQ_EVENT, vcpu);
6645 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6647 struct kvm_segment cs;
6649 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6653 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6655 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6656 struct kvm_sregs *sregs)
6660 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6661 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6662 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6663 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6664 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6665 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6667 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6668 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6670 kvm_x86_ops->get_idt(vcpu, &dt);
6671 sregs->idt.limit = dt.size;
6672 sregs->idt.base = dt.address;
6673 kvm_x86_ops->get_gdt(vcpu, &dt);
6674 sregs->gdt.limit = dt.size;
6675 sregs->gdt.base = dt.address;
6677 sregs->cr0 = kvm_read_cr0(vcpu);
6678 sregs->cr2 = vcpu->arch.cr2;
6679 sregs->cr3 = kvm_read_cr3(vcpu);
6680 sregs->cr4 = kvm_read_cr4(vcpu);
6681 sregs->cr8 = kvm_get_cr8(vcpu);
6682 sregs->efer = vcpu->arch.efer;
6683 sregs->apic_base = kvm_get_apic_base(vcpu);
6685 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6687 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6688 set_bit(vcpu->arch.interrupt.nr,
6689 (unsigned long *)sregs->interrupt_bitmap);
6694 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6695 struct kvm_mp_state *mp_state)
6697 kvm_apic_accept_events(vcpu);
6698 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6699 vcpu->arch.pv.pv_unhalted)
6700 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6702 mp_state->mp_state = vcpu->arch.mp_state;
6707 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6708 struct kvm_mp_state *mp_state)
6710 if (!kvm_vcpu_has_lapic(vcpu) &&
6711 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6714 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6715 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6716 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6718 vcpu->arch.mp_state = mp_state->mp_state;
6719 kvm_make_request(KVM_REQ_EVENT, vcpu);
6723 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6724 int reason, bool has_error_code, u32 error_code)
6726 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6729 init_emulate_ctxt(vcpu);
6731 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6732 has_error_code, error_code);
6735 return EMULATE_FAIL;
6737 kvm_rip_write(vcpu, ctxt->eip);
6738 kvm_set_rflags(vcpu, ctxt->eflags);
6739 kvm_make_request(KVM_REQ_EVENT, vcpu);
6740 return EMULATE_DONE;
6742 EXPORT_SYMBOL_GPL(kvm_task_switch);
6744 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6745 struct kvm_sregs *sregs)
6747 struct msr_data apic_base_msr;
6748 int mmu_reset_needed = 0;
6749 int pending_vec, max_bits, idx;
6752 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6755 dt.size = sregs->idt.limit;
6756 dt.address = sregs->idt.base;
6757 kvm_x86_ops->set_idt(vcpu, &dt);
6758 dt.size = sregs->gdt.limit;
6759 dt.address = sregs->gdt.base;
6760 kvm_x86_ops->set_gdt(vcpu, &dt);
6762 vcpu->arch.cr2 = sregs->cr2;
6763 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6764 vcpu->arch.cr3 = sregs->cr3;
6765 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6767 kvm_set_cr8(vcpu, sregs->cr8);
6769 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6770 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6771 apic_base_msr.data = sregs->apic_base;
6772 apic_base_msr.host_initiated = true;
6773 kvm_set_apic_base(vcpu, &apic_base_msr);
6775 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6776 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6777 vcpu->arch.cr0 = sregs->cr0;
6779 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6780 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6781 if (sregs->cr4 & X86_CR4_OSXSAVE)
6782 kvm_update_cpuid(vcpu);
6784 idx = srcu_read_lock(&vcpu->kvm->srcu);
6785 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6786 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6787 mmu_reset_needed = 1;
6789 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6791 if (mmu_reset_needed)
6792 kvm_mmu_reset_context(vcpu);
6794 max_bits = KVM_NR_INTERRUPTS;
6795 pending_vec = find_first_bit(
6796 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6797 if (pending_vec < max_bits) {
6798 kvm_queue_interrupt(vcpu, pending_vec, false);
6799 pr_debug("Set back pending irq %d\n", pending_vec);
6802 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6803 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6804 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6805 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6806 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6807 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6809 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6810 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6812 update_cr8_intercept(vcpu);
6814 /* Older userspace won't unhalt the vcpu on reset. */
6815 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6816 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6818 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6820 kvm_make_request(KVM_REQ_EVENT, vcpu);
6825 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6826 struct kvm_guest_debug *dbg)
6828 unsigned long rflags;
6831 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6833 if (vcpu->arch.exception.pending)
6835 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6836 kvm_queue_exception(vcpu, DB_VECTOR);
6838 kvm_queue_exception(vcpu, BP_VECTOR);
6842 * Read rflags as long as potentially injected trace flags are still
6845 rflags = kvm_get_rflags(vcpu);
6847 vcpu->guest_debug = dbg->control;
6848 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6849 vcpu->guest_debug = 0;
6851 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6852 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6853 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6854 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6856 for (i = 0; i < KVM_NR_DB_REGS; i++)
6857 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6859 kvm_update_dr7(vcpu);
6861 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6862 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6863 get_segment_base(vcpu, VCPU_SREG_CS);
6866 * Trigger an rflags update that will inject or remove the trace
6869 kvm_set_rflags(vcpu, rflags);
6871 kvm_x86_ops->update_db_bp_intercept(vcpu);
6881 * Translate a guest virtual address to a guest physical address.
6883 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6884 struct kvm_translation *tr)
6886 unsigned long vaddr = tr->linear_address;
6890 idx = srcu_read_lock(&vcpu->kvm->srcu);
6891 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6892 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6893 tr->physical_address = gpa;
6894 tr->valid = gpa != UNMAPPED_GVA;
6901 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6903 struct i387_fxsave_struct *fxsave =
6904 &vcpu->arch.guest_fpu.state->fxsave;
6906 memcpy(fpu->fpr, fxsave->st_space, 128);
6907 fpu->fcw = fxsave->cwd;
6908 fpu->fsw = fxsave->swd;
6909 fpu->ftwx = fxsave->twd;
6910 fpu->last_opcode = fxsave->fop;
6911 fpu->last_ip = fxsave->rip;
6912 fpu->last_dp = fxsave->rdp;
6913 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6918 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6920 struct i387_fxsave_struct *fxsave =
6921 &vcpu->arch.guest_fpu.state->fxsave;
6923 memcpy(fxsave->st_space, fpu->fpr, 128);
6924 fxsave->cwd = fpu->fcw;
6925 fxsave->swd = fpu->fsw;
6926 fxsave->twd = fpu->ftwx;
6927 fxsave->fop = fpu->last_opcode;
6928 fxsave->rip = fpu->last_ip;
6929 fxsave->rdp = fpu->last_dp;
6930 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6935 int fx_init(struct kvm_vcpu *vcpu)
6939 err = fpu_alloc(&vcpu->arch.guest_fpu);
6943 fpu_finit(&vcpu->arch.guest_fpu);
6945 vcpu->arch.guest_fpu.state->xsave.xsave_hdr.xcomp_bv =
6946 host_xcr0 | XSTATE_COMPACTION_ENABLED;
6949 * Ensure guest xcr0 is valid for loading
6951 vcpu->arch.xcr0 = XSTATE_FP;
6953 vcpu->arch.cr0 |= X86_CR0_ET;
6957 EXPORT_SYMBOL_GPL(fx_init);
6959 static void fx_free(struct kvm_vcpu *vcpu)
6961 fpu_free(&vcpu->arch.guest_fpu);
6964 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6966 if (vcpu->guest_fpu_loaded)
6970 * Restore all possible states in the guest,
6971 * and assume host would use all available bits.
6972 * Guest xcr0 would be loaded later.
6974 kvm_put_guest_xcr0(vcpu);
6975 vcpu->guest_fpu_loaded = 1;
6976 __kernel_fpu_begin();
6977 fpu_restore_checking(&vcpu->arch.guest_fpu);
6981 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6983 kvm_put_guest_xcr0(vcpu);
6985 if (!vcpu->guest_fpu_loaded)
6988 vcpu->guest_fpu_loaded = 0;
6989 fpu_save_init(&vcpu->arch.guest_fpu);
6991 ++vcpu->stat.fpu_reload;
6992 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6996 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6998 kvmclock_reset(vcpu);
7000 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7002 kvm_x86_ops->vcpu_free(vcpu);
7005 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7008 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7009 printk_once(KERN_WARNING
7010 "kvm: SMP vm created on host with unstable TSC; "
7011 "guest TSC will not be reliable\n");
7012 return kvm_x86_ops->vcpu_create(kvm, id);
7015 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7019 vcpu->arch.mtrr_state.have_fixed = 1;
7020 r = vcpu_load(vcpu);
7023 kvm_vcpu_reset(vcpu);
7024 kvm_mmu_setup(vcpu);
7030 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7032 struct msr_data msr;
7033 struct kvm *kvm = vcpu->kvm;
7035 if (vcpu_load(vcpu))
7038 msr.index = MSR_IA32_TSC;
7039 msr.host_initiated = true;
7040 kvm_write_tsc(vcpu, &msr);
7043 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7044 KVMCLOCK_SYNC_PERIOD);
7047 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7050 vcpu->arch.apf.msr_val = 0;
7052 r = vcpu_load(vcpu);
7054 kvm_mmu_unload(vcpu);
7058 kvm_x86_ops->vcpu_free(vcpu);
7061 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
7063 atomic_set(&vcpu->arch.nmi_queued, 0);
7064 vcpu->arch.nmi_pending = 0;
7065 vcpu->arch.nmi_injected = false;
7066 kvm_clear_interrupt_queue(vcpu);
7067 kvm_clear_exception_queue(vcpu);
7069 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7070 vcpu->arch.dr6 = DR6_INIT;
7071 kvm_update_dr6(vcpu);
7072 vcpu->arch.dr7 = DR7_FIXED_1;
7073 kvm_update_dr7(vcpu);
7075 kvm_make_request(KVM_REQ_EVENT, vcpu);
7076 vcpu->arch.apf.msr_val = 0;
7077 vcpu->arch.st.msr_val = 0;
7079 kvmclock_reset(vcpu);
7081 kvm_clear_async_pf_completion_queue(vcpu);
7082 kvm_async_pf_hash_reset(vcpu);
7083 vcpu->arch.apf.halted = false;
7085 kvm_pmu_reset(vcpu);
7087 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7088 vcpu->arch.regs_avail = ~0;
7089 vcpu->arch.regs_dirty = ~0;
7091 kvm_x86_ops->vcpu_reset(vcpu);
7094 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7096 struct kvm_segment cs;
7098 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7099 cs.selector = vector << 8;
7100 cs.base = vector << 12;
7101 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7102 kvm_rip_write(vcpu, 0);
7105 int kvm_arch_hardware_enable(void)
7108 struct kvm_vcpu *vcpu;
7113 bool stable, backwards_tsc = false;
7115 kvm_shared_msr_cpu_online();
7116 ret = kvm_x86_ops->hardware_enable();
7120 local_tsc = native_read_tsc();
7121 stable = !check_tsc_unstable();
7122 list_for_each_entry(kvm, &vm_list, vm_list) {
7123 kvm_for_each_vcpu(i, vcpu, kvm) {
7124 if (!stable && vcpu->cpu == smp_processor_id())
7125 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7126 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7127 backwards_tsc = true;
7128 if (vcpu->arch.last_host_tsc > max_tsc)
7129 max_tsc = vcpu->arch.last_host_tsc;
7135 * Sometimes, even reliable TSCs go backwards. This happens on
7136 * platforms that reset TSC during suspend or hibernate actions, but
7137 * maintain synchronization. We must compensate. Fortunately, we can
7138 * detect that condition here, which happens early in CPU bringup,
7139 * before any KVM threads can be running. Unfortunately, we can't
7140 * bring the TSCs fully up to date with real time, as we aren't yet far
7141 * enough into CPU bringup that we know how much real time has actually
7142 * elapsed; our helper function, get_kernel_ns() will be using boot
7143 * variables that haven't been updated yet.
7145 * So we simply find the maximum observed TSC above, then record the
7146 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7147 * the adjustment will be applied. Note that we accumulate
7148 * adjustments, in case multiple suspend cycles happen before some VCPU
7149 * gets a chance to run again. In the event that no KVM threads get a
7150 * chance to run, we will miss the entire elapsed period, as we'll have
7151 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7152 * loose cycle time. This isn't too big a deal, since the loss will be
7153 * uniform across all VCPUs (not to mention the scenario is extremely
7154 * unlikely). It is possible that a second hibernate recovery happens
7155 * much faster than a first, causing the observed TSC here to be
7156 * smaller; this would require additional padding adjustment, which is
7157 * why we set last_host_tsc to the local tsc observed here.
7159 * N.B. - this code below runs only on platforms with reliable TSC,
7160 * as that is the only way backwards_tsc is set above. Also note
7161 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7162 * have the same delta_cyc adjustment applied if backwards_tsc
7163 * is detected. Note further, this adjustment is only done once,
7164 * as we reset last_host_tsc on all VCPUs to stop this from being
7165 * called multiple times (one for each physical CPU bringup).
7167 * Platforms with unreliable TSCs don't have to deal with this, they
7168 * will be compensated by the logic in vcpu_load, which sets the TSC to
7169 * catchup mode. This will catchup all VCPUs to real time, but cannot
7170 * guarantee that they stay in perfect synchronization.
7172 if (backwards_tsc) {
7173 u64 delta_cyc = max_tsc - local_tsc;
7174 backwards_tsc_observed = true;
7175 list_for_each_entry(kvm, &vm_list, vm_list) {
7176 kvm_for_each_vcpu(i, vcpu, kvm) {
7177 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7178 vcpu->arch.last_host_tsc = local_tsc;
7179 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7183 * We have to disable TSC offset matching.. if you were
7184 * booting a VM while issuing an S4 host suspend....
7185 * you may have some problem. Solving this issue is
7186 * left as an exercise to the reader.
7188 kvm->arch.last_tsc_nsec = 0;
7189 kvm->arch.last_tsc_write = 0;
7196 void kvm_arch_hardware_disable(void)
7198 kvm_x86_ops->hardware_disable();
7199 drop_user_return_notifiers();
7202 int kvm_arch_hardware_setup(void)
7204 return kvm_x86_ops->hardware_setup();
7207 void kvm_arch_hardware_unsetup(void)
7209 kvm_x86_ops->hardware_unsetup();
7212 void kvm_arch_check_processor_compat(void *rtn)
7214 kvm_x86_ops->check_processor_compatibility(rtn);
7217 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7219 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7222 struct static_key kvm_no_apic_vcpu __read_mostly;
7224 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7230 BUG_ON(vcpu->kvm == NULL);
7233 vcpu->arch.pv.pv_unhalted = false;
7234 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7235 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
7236 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7238 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7240 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7245 vcpu->arch.pio_data = page_address(page);
7247 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7249 r = kvm_mmu_create(vcpu);
7251 goto fail_free_pio_data;
7253 if (irqchip_in_kernel(kvm)) {
7254 r = kvm_create_lapic(vcpu);
7256 goto fail_mmu_destroy;
7258 static_key_slow_inc(&kvm_no_apic_vcpu);
7260 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7262 if (!vcpu->arch.mce_banks) {
7264 goto fail_free_lapic;
7266 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7268 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7270 goto fail_free_mce_banks;
7275 goto fail_free_wbinvd_dirty_mask;
7277 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7278 vcpu->arch.pv_time_enabled = false;
7280 vcpu->arch.guest_supported_xcr0 = 0;
7281 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7283 kvm_async_pf_hash_reset(vcpu);
7287 fail_free_wbinvd_dirty_mask:
7288 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7289 fail_free_mce_banks:
7290 kfree(vcpu->arch.mce_banks);
7292 kvm_free_lapic(vcpu);
7294 kvm_mmu_destroy(vcpu);
7296 free_page((unsigned long)vcpu->arch.pio_data);
7301 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7305 kvm_pmu_destroy(vcpu);
7306 kfree(vcpu->arch.mce_banks);
7307 kvm_free_lapic(vcpu);
7308 idx = srcu_read_lock(&vcpu->kvm->srcu);
7309 kvm_mmu_destroy(vcpu);
7310 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7311 free_page((unsigned long)vcpu->arch.pio_data);
7312 if (!irqchip_in_kernel(vcpu->kvm))
7313 static_key_slow_dec(&kvm_no_apic_vcpu);
7316 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7318 kvm_x86_ops->sched_in(vcpu, cpu);
7321 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7326 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7327 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7328 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7329 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7330 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7332 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7333 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7334 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7335 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7336 &kvm->arch.irq_sources_bitmap);
7338 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7339 mutex_init(&kvm->arch.apic_map_lock);
7340 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7342 pvclock_update_vm_gtod_copy(kvm);
7344 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7345 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7350 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7353 r = vcpu_load(vcpu);
7355 kvm_mmu_unload(vcpu);
7359 static void kvm_free_vcpus(struct kvm *kvm)
7362 struct kvm_vcpu *vcpu;
7365 * Unpin any mmu pages first.
7367 kvm_for_each_vcpu(i, vcpu, kvm) {
7368 kvm_clear_async_pf_completion_queue(vcpu);
7369 kvm_unload_vcpu_mmu(vcpu);
7371 kvm_for_each_vcpu(i, vcpu, kvm)
7372 kvm_arch_vcpu_free(vcpu);
7374 mutex_lock(&kvm->lock);
7375 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7376 kvm->vcpus[i] = NULL;
7378 atomic_set(&kvm->online_vcpus, 0);
7379 mutex_unlock(&kvm->lock);
7382 void kvm_arch_sync_events(struct kvm *kvm)
7384 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7385 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7386 kvm_free_all_assigned_devices(kvm);
7390 void kvm_arch_destroy_vm(struct kvm *kvm)
7392 if (current->mm == kvm->mm) {
7394 * Free memory regions allocated on behalf of userspace,
7395 * unless the the memory map has changed due to process exit
7398 struct kvm_userspace_memory_region mem;
7399 memset(&mem, 0, sizeof(mem));
7400 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7401 kvm_set_memory_region(kvm, &mem);
7403 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7404 kvm_set_memory_region(kvm, &mem);
7406 mem.slot = TSS_PRIVATE_MEMSLOT;
7407 kvm_set_memory_region(kvm, &mem);
7409 kvm_iommu_unmap_guest(kvm);
7410 kfree(kvm->arch.vpic);
7411 kfree(kvm->arch.vioapic);
7412 kvm_free_vcpus(kvm);
7413 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7416 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7417 struct kvm_memory_slot *dont)
7421 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7422 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7423 kvm_kvfree(free->arch.rmap[i]);
7424 free->arch.rmap[i] = NULL;
7429 if (!dont || free->arch.lpage_info[i - 1] !=
7430 dont->arch.lpage_info[i - 1]) {
7431 kvm_kvfree(free->arch.lpage_info[i - 1]);
7432 free->arch.lpage_info[i - 1] = NULL;
7437 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7438 unsigned long npages)
7442 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7447 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7448 slot->base_gfn, level) + 1;
7450 slot->arch.rmap[i] =
7451 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7452 if (!slot->arch.rmap[i])
7457 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7458 sizeof(*slot->arch.lpage_info[i - 1]));
7459 if (!slot->arch.lpage_info[i - 1])
7462 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7463 slot->arch.lpage_info[i - 1][0].write_count = 1;
7464 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7465 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7466 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7468 * If the gfn and userspace address are not aligned wrt each
7469 * other, or if explicitly asked to, disable large page
7470 * support for this slot
7472 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7473 !kvm_largepages_enabled()) {
7476 for (j = 0; j < lpages; ++j)
7477 slot->arch.lpage_info[i - 1][j].write_count = 1;
7484 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7485 kvm_kvfree(slot->arch.rmap[i]);
7486 slot->arch.rmap[i] = NULL;
7490 kvm_kvfree(slot->arch.lpage_info[i - 1]);
7491 slot->arch.lpage_info[i - 1] = NULL;
7496 void kvm_arch_memslots_updated(struct kvm *kvm)
7499 * memslots->generation has been incremented.
7500 * mmio generation may have reached its maximum value.
7502 kvm_mmu_invalidate_mmio_sptes(kvm);
7505 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7506 struct kvm_memory_slot *memslot,
7507 struct kvm_userspace_memory_region *mem,
7508 enum kvm_mr_change change)
7511 * Only private memory slots need to be mapped here since
7512 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7514 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7515 unsigned long userspace_addr;
7518 * MAP_SHARED to prevent internal slot pages from being moved
7521 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7522 PROT_READ | PROT_WRITE,
7523 MAP_SHARED | MAP_ANONYMOUS, 0);
7525 if (IS_ERR((void *)userspace_addr))
7526 return PTR_ERR((void *)userspace_addr);
7528 memslot->userspace_addr = userspace_addr;
7534 void kvm_arch_commit_memory_region(struct kvm *kvm,
7535 struct kvm_userspace_memory_region *mem,
7536 const struct kvm_memory_slot *old,
7537 enum kvm_mr_change change)
7540 int nr_mmu_pages = 0;
7542 if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7545 ret = vm_munmap(old->userspace_addr,
7546 old->npages * PAGE_SIZE);
7549 "kvm_vm_ioctl_set_memory_region: "
7550 "failed to munmap memory\n");
7553 if (!kvm->arch.n_requested_mmu_pages)
7554 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7557 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7559 * Write protect all pages for dirty logging.
7561 * All the sptes including the large sptes which point to this
7562 * slot are set to readonly. We can not create any new large
7563 * spte on this slot until the end of the logging.
7565 * See the comments in fast_page_fault().
7567 if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
7568 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7571 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7573 kvm_mmu_invalidate_zap_all_pages(kvm);
7576 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7577 struct kvm_memory_slot *slot)
7579 kvm_mmu_invalidate_zap_all_pages(kvm);
7582 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7584 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7585 kvm_x86_ops->check_nested_events(vcpu, false);
7587 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7588 !vcpu->arch.apf.halted)
7589 || !list_empty_careful(&vcpu->async_pf.done)
7590 || kvm_apic_has_events(vcpu)
7591 || vcpu->arch.pv.pv_unhalted
7592 || atomic_read(&vcpu->arch.nmi_queued) ||
7593 (kvm_arch_interrupt_allowed(vcpu) &&
7594 kvm_cpu_has_interrupt(vcpu));
7597 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7599 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7602 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7604 return kvm_x86_ops->interrupt_allowed(vcpu);
7607 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
7609 if (is_64_bit_mode(vcpu))
7610 return kvm_rip_read(vcpu);
7611 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
7612 kvm_rip_read(vcpu));
7614 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
7616 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7618 return kvm_get_linear_rip(vcpu) == linear_rip;
7620 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7622 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7624 unsigned long rflags;
7626 rflags = kvm_x86_ops->get_rflags(vcpu);
7627 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7628 rflags &= ~X86_EFLAGS_TF;
7631 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7633 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7635 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7636 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7637 rflags |= X86_EFLAGS_TF;
7638 kvm_x86_ops->set_rflags(vcpu, rflags);
7641 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7643 __kvm_set_rflags(vcpu, rflags);
7644 kvm_make_request(KVM_REQ_EVENT, vcpu);
7646 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7648 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7652 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7656 r = kvm_mmu_reload(vcpu);
7660 if (!vcpu->arch.mmu.direct_map &&
7661 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7664 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7667 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7669 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7672 static inline u32 kvm_async_pf_next_probe(u32 key)
7674 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7677 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7679 u32 key = kvm_async_pf_hash_fn(gfn);
7681 while (vcpu->arch.apf.gfns[key] != ~0)
7682 key = kvm_async_pf_next_probe(key);
7684 vcpu->arch.apf.gfns[key] = gfn;
7687 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7690 u32 key = kvm_async_pf_hash_fn(gfn);
7692 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7693 (vcpu->arch.apf.gfns[key] != gfn &&
7694 vcpu->arch.apf.gfns[key] != ~0); i++)
7695 key = kvm_async_pf_next_probe(key);
7700 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7702 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7705 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7709 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7711 vcpu->arch.apf.gfns[i] = ~0;
7713 j = kvm_async_pf_next_probe(j);
7714 if (vcpu->arch.apf.gfns[j] == ~0)
7716 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7718 * k lies cyclically in ]i,j]
7720 * |....j i.k.| or |.k..j i...|
7722 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7723 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7728 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7731 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7735 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7736 struct kvm_async_pf *work)
7738 struct x86_exception fault;
7740 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7741 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7743 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7744 (vcpu->arch.apf.send_user_only &&
7745 kvm_x86_ops->get_cpl(vcpu) == 0))
7746 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7747 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7748 fault.vector = PF_VECTOR;
7749 fault.error_code_valid = true;
7750 fault.error_code = 0;
7751 fault.nested_page_fault = false;
7752 fault.address = work->arch.token;
7753 kvm_inject_page_fault(vcpu, &fault);
7757 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7758 struct kvm_async_pf *work)
7760 struct x86_exception fault;
7762 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7763 if (work->wakeup_all)
7764 work->arch.token = ~0; /* broadcast wakeup */
7766 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7768 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7769 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7770 fault.vector = PF_VECTOR;
7771 fault.error_code_valid = true;
7772 fault.error_code = 0;
7773 fault.nested_page_fault = false;
7774 fault.address = work->arch.token;
7775 kvm_inject_page_fault(vcpu, &fault);
7777 vcpu->arch.apf.halted = false;
7778 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7781 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7783 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7786 return !kvm_event_needs_reinjection(vcpu) &&
7787 kvm_x86_ops->interrupt_allowed(vcpu);
7790 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
7792 atomic_inc(&kvm->arch.noncoherent_dma_count);
7794 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
7796 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
7798 atomic_dec(&kvm->arch.noncoherent_dma_count);
7800 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
7802 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
7804 return atomic_read(&kvm->arch.noncoherent_dma_count);
7806 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
7808 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7809 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7810 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7811 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7812 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7813 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7814 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7815 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7816 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7817 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7818 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7819 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7820 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7821 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);