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"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/module.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <trace/events/kvm.h>
58 #define CREATE_TRACE_POINTS
61 #include <asm/debugreg.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
71 #define MAX_IO_MSRS 256
72 #define KVM_MAX_MCE_BANKS 32
73 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
75 #define emul_to_vcpu(ctxt) \
76 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
79 * - enable syscall per default because its emulated by KVM
80 * - enable LME and LMA per default on 64 bit KVM
84 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
86 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
89 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
90 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
92 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
93 static void process_nmi(struct kvm_vcpu *vcpu);
94 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
96 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
97 EXPORT_SYMBOL_GPL(kvm_x86_ops);
99 static bool __read_mostly ignore_msrs = 0;
100 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
102 unsigned int min_timer_period_us = 500;
103 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
105 static bool __read_mostly kvmclock_periodic_sync = true;
106 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
108 bool __read_mostly kvm_has_tsc_control;
109 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
110 u32 __read_mostly kvm_max_guest_tsc_khz;
111 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
112 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
113 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
114 u64 __read_mostly kvm_max_tsc_scaling_ratio;
115 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
116 static u64 __read_mostly kvm_default_tsc_scaling_ratio;
118 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
119 static u32 __read_mostly tsc_tolerance_ppm = 250;
120 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
122 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
123 unsigned int __read_mostly lapic_timer_advance_ns = 0;
124 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
126 static bool __read_mostly backwards_tsc_observed = false;
128 #define KVM_NR_SHARED_MSRS 16
130 struct kvm_shared_msrs_global {
132 u32 msrs[KVM_NR_SHARED_MSRS];
135 struct kvm_shared_msrs {
136 struct user_return_notifier urn;
138 struct kvm_shared_msr_values {
141 } values[KVM_NR_SHARED_MSRS];
144 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
145 static struct kvm_shared_msrs __percpu *shared_msrs;
147 struct kvm_stats_debugfs_item debugfs_entries[] = {
148 { "pf_fixed", VCPU_STAT(pf_fixed) },
149 { "pf_guest", VCPU_STAT(pf_guest) },
150 { "tlb_flush", VCPU_STAT(tlb_flush) },
151 { "invlpg", VCPU_STAT(invlpg) },
152 { "exits", VCPU_STAT(exits) },
153 { "io_exits", VCPU_STAT(io_exits) },
154 { "mmio_exits", VCPU_STAT(mmio_exits) },
155 { "signal_exits", VCPU_STAT(signal_exits) },
156 { "irq_window", VCPU_STAT(irq_window_exits) },
157 { "nmi_window", VCPU_STAT(nmi_window_exits) },
158 { "halt_exits", VCPU_STAT(halt_exits) },
159 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
160 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
161 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
162 { "hypercalls", VCPU_STAT(hypercalls) },
163 { "request_irq", VCPU_STAT(request_irq_exits) },
164 { "irq_exits", VCPU_STAT(irq_exits) },
165 { "host_state_reload", VCPU_STAT(host_state_reload) },
166 { "efer_reload", VCPU_STAT(efer_reload) },
167 { "fpu_reload", VCPU_STAT(fpu_reload) },
168 { "insn_emulation", VCPU_STAT(insn_emulation) },
169 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
170 { "irq_injections", VCPU_STAT(irq_injections) },
171 { "nmi_injections", VCPU_STAT(nmi_injections) },
172 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
173 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
174 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
175 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
176 { "mmu_flooded", VM_STAT(mmu_flooded) },
177 { "mmu_recycled", VM_STAT(mmu_recycled) },
178 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
179 { "mmu_unsync", VM_STAT(mmu_unsync) },
180 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
181 { "largepages", VM_STAT(lpages) },
185 u64 __read_mostly host_xcr0;
187 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
189 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
192 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
193 vcpu->arch.apf.gfns[i] = ~0;
196 static void kvm_on_user_return(struct user_return_notifier *urn)
199 struct kvm_shared_msrs *locals
200 = container_of(urn, struct kvm_shared_msrs, urn);
201 struct kvm_shared_msr_values *values;
203 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
204 values = &locals->values[slot];
205 if (values->host != values->curr) {
206 wrmsrl(shared_msrs_global.msrs[slot], values->host);
207 values->curr = values->host;
210 locals->registered = false;
211 user_return_notifier_unregister(urn);
214 static void shared_msr_update(unsigned slot, u32 msr)
217 unsigned int cpu = smp_processor_id();
218 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
220 /* only read, and nobody should modify it at this time,
221 * so don't need lock */
222 if (slot >= shared_msrs_global.nr) {
223 printk(KERN_ERR "kvm: invalid MSR slot!");
226 rdmsrl_safe(msr, &value);
227 smsr->values[slot].host = value;
228 smsr->values[slot].curr = value;
231 void kvm_define_shared_msr(unsigned slot, u32 msr)
233 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
234 shared_msrs_global.msrs[slot] = msr;
235 if (slot >= shared_msrs_global.nr)
236 shared_msrs_global.nr = slot + 1;
238 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
240 static void kvm_shared_msr_cpu_online(void)
244 for (i = 0; i < shared_msrs_global.nr; ++i)
245 shared_msr_update(i, shared_msrs_global.msrs[i]);
248 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
250 unsigned int cpu = smp_processor_id();
251 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
254 if (((value ^ smsr->values[slot].curr) & mask) == 0)
256 smsr->values[slot].curr = value;
257 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
261 if (!smsr->registered) {
262 smsr->urn.on_user_return = kvm_on_user_return;
263 user_return_notifier_register(&smsr->urn);
264 smsr->registered = true;
268 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
270 static void drop_user_return_notifiers(void)
272 unsigned int cpu = smp_processor_id();
273 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
275 if (smsr->registered)
276 kvm_on_user_return(&smsr->urn);
279 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
281 return vcpu->arch.apic_base;
283 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
285 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
287 u64 old_state = vcpu->arch.apic_base &
288 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
289 u64 new_state = msr_info->data &
290 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
291 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
292 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
294 if (!msr_info->host_initiated &&
295 ((msr_info->data & reserved_bits) != 0 ||
296 new_state == X2APIC_ENABLE ||
297 (new_state == MSR_IA32_APICBASE_ENABLE &&
298 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
299 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
303 kvm_lapic_set_base(vcpu, msr_info->data);
306 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
308 asmlinkage __visible void kvm_spurious_fault(void)
310 /* Fault while not rebooting. We want the trace. */
313 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
315 #define EXCPT_BENIGN 0
316 #define EXCPT_CONTRIBUTORY 1
319 static int exception_class(int vector)
329 return EXCPT_CONTRIBUTORY;
336 #define EXCPT_FAULT 0
338 #define EXCPT_ABORT 2
339 #define EXCPT_INTERRUPT 3
341 static int exception_type(int vector)
345 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
346 return EXCPT_INTERRUPT;
350 /* #DB is trap, as instruction watchpoints are handled elsewhere */
351 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
354 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
357 /* Reserved exceptions will result in fault */
361 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
362 unsigned nr, bool has_error, u32 error_code,
368 kvm_make_request(KVM_REQ_EVENT, vcpu);
370 if (!vcpu->arch.exception.pending) {
372 if (has_error && !is_protmode(vcpu))
374 vcpu->arch.exception.pending = true;
375 vcpu->arch.exception.has_error_code = has_error;
376 vcpu->arch.exception.nr = nr;
377 vcpu->arch.exception.error_code = error_code;
378 vcpu->arch.exception.reinject = reinject;
382 /* to check exception */
383 prev_nr = vcpu->arch.exception.nr;
384 if (prev_nr == DF_VECTOR) {
385 /* triple fault -> shutdown */
386 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
389 class1 = exception_class(prev_nr);
390 class2 = exception_class(nr);
391 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
392 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
393 /* generate double fault per SDM Table 5-5 */
394 vcpu->arch.exception.pending = true;
395 vcpu->arch.exception.has_error_code = true;
396 vcpu->arch.exception.nr = DF_VECTOR;
397 vcpu->arch.exception.error_code = 0;
399 /* replace previous exception with a new one in a hope
400 that instruction re-execution will regenerate lost
405 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
407 kvm_multiple_exception(vcpu, nr, false, 0, false);
409 EXPORT_SYMBOL_GPL(kvm_queue_exception);
411 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
413 kvm_multiple_exception(vcpu, nr, false, 0, true);
415 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
417 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
420 kvm_inject_gp(vcpu, 0);
422 kvm_x86_ops->skip_emulated_instruction(vcpu);
424 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
426 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
428 ++vcpu->stat.pf_guest;
429 vcpu->arch.cr2 = fault->address;
430 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
432 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
434 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
436 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
437 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
439 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
441 return fault->nested_page_fault;
444 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
446 atomic_inc(&vcpu->arch.nmi_queued);
447 kvm_make_request(KVM_REQ_NMI, vcpu);
449 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
451 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
453 kvm_multiple_exception(vcpu, nr, true, error_code, false);
455 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
457 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
459 kvm_multiple_exception(vcpu, nr, true, error_code, true);
461 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
464 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
465 * a #GP and return false.
467 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
469 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
471 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
474 EXPORT_SYMBOL_GPL(kvm_require_cpl);
476 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
478 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
481 kvm_queue_exception(vcpu, UD_VECTOR);
484 EXPORT_SYMBOL_GPL(kvm_require_dr);
487 * This function will be used to read from the physical memory of the currently
488 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
489 * can read from guest physical or from the guest's guest physical memory.
491 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
492 gfn_t ngfn, void *data, int offset, int len,
495 struct x86_exception exception;
499 ngpa = gfn_to_gpa(ngfn);
500 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
501 if (real_gfn == UNMAPPED_GVA)
504 real_gfn = gpa_to_gfn(real_gfn);
506 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
508 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
510 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
511 void *data, int offset, int len, u32 access)
513 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
514 data, offset, len, access);
518 * Load the pae pdptrs. Return true is they are all valid.
520 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
522 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
523 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
526 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
528 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
529 offset * sizeof(u64), sizeof(pdpte),
530 PFERR_USER_MASK|PFERR_WRITE_MASK);
535 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
536 if (is_present_gpte(pdpte[i]) &&
538 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
545 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
546 __set_bit(VCPU_EXREG_PDPTR,
547 (unsigned long *)&vcpu->arch.regs_avail);
548 __set_bit(VCPU_EXREG_PDPTR,
549 (unsigned long *)&vcpu->arch.regs_dirty);
554 EXPORT_SYMBOL_GPL(load_pdptrs);
556 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
558 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
564 if (is_long_mode(vcpu) || !is_pae(vcpu))
567 if (!test_bit(VCPU_EXREG_PDPTR,
568 (unsigned long *)&vcpu->arch.regs_avail))
571 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
572 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
573 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
574 PFERR_USER_MASK | PFERR_WRITE_MASK);
577 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
583 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
585 unsigned long old_cr0 = kvm_read_cr0(vcpu);
586 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
591 if (cr0 & 0xffffffff00000000UL)
595 cr0 &= ~CR0_RESERVED_BITS;
597 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
600 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
603 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
605 if ((vcpu->arch.efer & EFER_LME)) {
610 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
615 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
620 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
623 kvm_x86_ops->set_cr0(vcpu, cr0);
625 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
626 kvm_clear_async_pf_completion_queue(vcpu);
627 kvm_async_pf_hash_reset(vcpu);
630 if ((cr0 ^ old_cr0) & update_bits)
631 kvm_mmu_reset_context(vcpu);
633 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
634 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
635 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
636 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
640 EXPORT_SYMBOL_GPL(kvm_set_cr0);
642 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
644 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
646 EXPORT_SYMBOL_GPL(kvm_lmsw);
648 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
650 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
651 !vcpu->guest_xcr0_loaded) {
652 /* kvm_set_xcr() also depends on this */
653 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
654 vcpu->guest_xcr0_loaded = 1;
658 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
660 if (vcpu->guest_xcr0_loaded) {
661 if (vcpu->arch.xcr0 != host_xcr0)
662 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
663 vcpu->guest_xcr0_loaded = 0;
667 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
670 u64 old_xcr0 = vcpu->arch.xcr0;
673 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
674 if (index != XCR_XFEATURE_ENABLED_MASK)
676 if (!(xcr0 & XFEATURE_MASK_FP))
678 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
682 * Do not allow the guest to set bits that we do not support
683 * saving. However, xcr0 bit 0 is always set, even if the
684 * emulated CPU does not support XSAVE (see fx_init).
686 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
687 if (xcr0 & ~valid_bits)
690 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
691 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
694 if (xcr0 & XFEATURE_MASK_AVX512) {
695 if (!(xcr0 & XFEATURE_MASK_YMM))
697 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
700 kvm_put_guest_xcr0(vcpu);
701 vcpu->arch.xcr0 = xcr0;
703 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
704 kvm_update_cpuid(vcpu);
708 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
710 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
711 __kvm_set_xcr(vcpu, index, xcr)) {
712 kvm_inject_gp(vcpu, 0);
717 EXPORT_SYMBOL_GPL(kvm_set_xcr);
719 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
721 unsigned long old_cr4 = kvm_read_cr4(vcpu);
722 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
723 X86_CR4_SMEP | X86_CR4_SMAP;
725 if (cr4 & CR4_RESERVED_BITS)
728 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
731 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
734 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
737 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
740 if (is_long_mode(vcpu)) {
741 if (!(cr4 & X86_CR4_PAE))
743 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
744 && ((cr4 ^ old_cr4) & pdptr_bits)
745 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
749 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
750 if (!guest_cpuid_has_pcid(vcpu))
753 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
754 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
758 if (kvm_x86_ops->set_cr4(vcpu, cr4))
761 if (((cr4 ^ old_cr4) & pdptr_bits) ||
762 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
763 kvm_mmu_reset_context(vcpu);
765 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
766 kvm_update_cpuid(vcpu);
770 EXPORT_SYMBOL_GPL(kvm_set_cr4);
772 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
775 cr3 &= ~CR3_PCID_INVD;
778 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
779 kvm_mmu_sync_roots(vcpu);
780 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
784 if (is_long_mode(vcpu)) {
785 if (cr3 & CR3_L_MODE_RESERVED_BITS)
787 } else if (is_pae(vcpu) && is_paging(vcpu) &&
788 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
791 vcpu->arch.cr3 = cr3;
792 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
793 kvm_mmu_new_cr3(vcpu);
796 EXPORT_SYMBOL_GPL(kvm_set_cr3);
798 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
800 if (cr8 & CR8_RESERVED_BITS)
802 if (lapic_in_kernel(vcpu))
803 kvm_lapic_set_tpr(vcpu, cr8);
805 vcpu->arch.cr8 = cr8;
808 EXPORT_SYMBOL_GPL(kvm_set_cr8);
810 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
812 if (lapic_in_kernel(vcpu))
813 return kvm_lapic_get_cr8(vcpu);
815 return vcpu->arch.cr8;
817 EXPORT_SYMBOL_GPL(kvm_get_cr8);
819 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
823 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
824 for (i = 0; i < KVM_NR_DB_REGS; i++)
825 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
826 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
830 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
832 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
833 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
836 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
840 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
841 dr7 = vcpu->arch.guest_debug_dr7;
843 dr7 = vcpu->arch.dr7;
844 kvm_x86_ops->set_dr7(vcpu, dr7);
845 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
846 if (dr7 & DR7_BP_EN_MASK)
847 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
850 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
852 u64 fixed = DR6_FIXED_1;
854 if (!guest_cpuid_has_rtm(vcpu))
859 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
863 vcpu->arch.db[dr] = val;
864 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
865 vcpu->arch.eff_db[dr] = val;
870 if (val & 0xffffffff00000000ULL)
872 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
873 kvm_update_dr6(vcpu);
878 if (val & 0xffffffff00000000ULL)
880 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
881 kvm_update_dr7(vcpu);
888 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
890 if (__kvm_set_dr(vcpu, dr, val)) {
891 kvm_inject_gp(vcpu, 0);
896 EXPORT_SYMBOL_GPL(kvm_set_dr);
898 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
902 *val = vcpu->arch.db[dr];
907 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
908 *val = vcpu->arch.dr6;
910 *val = kvm_x86_ops->get_dr6(vcpu);
915 *val = vcpu->arch.dr7;
920 EXPORT_SYMBOL_GPL(kvm_get_dr);
922 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
924 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
928 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
931 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
932 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
935 EXPORT_SYMBOL_GPL(kvm_rdpmc);
938 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
939 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
941 * This list is modified at module load time to reflect the
942 * capabilities of the host cpu. This capabilities test skips MSRs that are
943 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
944 * may depend on host virtualization features rather than host cpu features.
947 static u32 msrs_to_save[] = {
948 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
951 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
953 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
954 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
957 static unsigned num_msrs_to_save;
959 static u32 emulated_msrs[] = {
960 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
961 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
962 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
963 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
964 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
965 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
968 HV_X64_MSR_VP_RUNTIME,
970 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
974 MSR_IA32_TSCDEADLINE,
975 MSR_IA32_MISC_ENABLE,
981 static unsigned num_emulated_msrs;
983 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
985 if (efer & efer_reserved_bits)
988 if (efer & EFER_FFXSR) {
989 struct kvm_cpuid_entry2 *feat;
991 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
992 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
996 if (efer & EFER_SVME) {
997 struct kvm_cpuid_entry2 *feat;
999 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1000 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1006 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1008 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1010 u64 old_efer = vcpu->arch.efer;
1012 if (!kvm_valid_efer(vcpu, efer))
1016 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1020 efer |= vcpu->arch.efer & EFER_LMA;
1022 kvm_x86_ops->set_efer(vcpu, efer);
1024 /* Update reserved bits */
1025 if ((efer ^ old_efer) & EFER_NX)
1026 kvm_mmu_reset_context(vcpu);
1031 void kvm_enable_efer_bits(u64 mask)
1033 efer_reserved_bits &= ~mask;
1035 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1038 * Writes msr value into into the appropriate "register".
1039 * Returns 0 on success, non-0 otherwise.
1040 * Assumes vcpu_load() was already called.
1042 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1044 switch (msr->index) {
1047 case MSR_KERNEL_GS_BASE:
1050 if (is_noncanonical_address(msr->data))
1053 case MSR_IA32_SYSENTER_EIP:
1054 case MSR_IA32_SYSENTER_ESP:
1056 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1057 * non-canonical address is written on Intel but not on
1058 * AMD (which ignores the top 32-bits, because it does
1059 * not implement 64-bit SYSENTER).
1061 * 64-bit code should hence be able to write a non-canonical
1062 * value on AMD. Making the address canonical ensures that
1063 * vmentry does not fail on Intel after writing a non-canonical
1064 * value, and that something deterministic happens if the guest
1065 * invokes 64-bit SYSENTER.
1067 msr->data = get_canonical(msr->data);
1069 return kvm_x86_ops->set_msr(vcpu, msr);
1071 EXPORT_SYMBOL_GPL(kvm_set_msr);
1074 * Adapt set_msr() to msr_io()'s calling convention
1076 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1078 struct msr_data msr;
1082 msr.host_initiated = true;
1083 r = kvm_get_msr(vcpu, &msr);
1091 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1093 struct msr_data msr;
1097 msr.host_initiated = true;
1098 return kvm_set_msr(vcpu, &msr);
1101 #ifdef CONFIG_X86_64
1102 struct pvclock_gtod_data {
1105 struct { /* extract of a clocksource struct */
1117 static struct pvclock_gtod_data pvclock_gtod_data;
1119 static void update_pvclock_gtod(struct timekeeper *tk)
1121 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1124 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1126 write_seqcount_begin(&vdata->seq);
1128 /* copy pvclock gtod data */
1129 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1130 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1131 vdata->clock.mask = tk->tkr_mono.mask;
1132 vdata->clock.mult = tk->tkr_mono.mult;
1133 vdata->clock.shift = tk->tkr_mono.shift;
1135 vdata->boot_ns = boot_ns;
1136 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1138 write_seqcount_end(&vdata->seq);
1142 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1145 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1146 * vcpu_enter_guest. This function is only called from
1147 * the physical CPU that is running vcpu.
1149 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1152 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1156 struct pvclock_wall_clock wc;
1157 struct timespec boot;
1162 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1167 ++version; /* first time write, random junk */
1171 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1174 * The guest calculates current wall clock time by adding
1175 * system time (updated by kvm_guest_time_update below) to the
1176 * wall clock specified here. guest system time equals host
1177 * system time for us, thus we must fill in host boot time here.
1181 if (kvm->arch.kvmclock_offset) {
1182 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1183 boot = timespec_sub(boot, ts);
1185 wc.sec = boot.tv_sec;
1186 wc.nsec = boot.tv_nsec;
1187 wc.version = version;
1189 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1192 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1195 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1197 uint32_t quotient, remainder;
1199 /* Don't try to replace with do_div(), this one calculates
1200 * "(dividend << 32) / divisor" */
1202 : "=a" (quotient), "=d" (remainder)
1203 : "0" (0), "1" (dividend), "r" (divisor) );
1207 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1208 s8 *pshift, u32 *pmultiplier)
1215 tps64 = base_khz * 1000LL;
1216 scaled64 = scaled_khz * 1000LL;
1217 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1222 tps32 = (uint32_t)tps64;
1223 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1224 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1232 *pmultiplier = div_frac(scaled64, tps32);
1234 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1235 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1238 #ifdef CONFIG_X86_64
1239 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1242 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1243 static unsigned long max_tsc_khz;
1245 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1247 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1248 vcpu->arch.virtual_tsc_shift);
1251 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1253 u64 v = (u64)khz * (1000000 + ppm);
1258 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1262 /* Guest TSC same frequency as host TSC? */
1264 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1268 /* TSC scaling supported? */
1269 if (!kvm_has_tsc_control) {
1270 if (user_tsc_khz > tsc_khz) {
1271 vcpu->arch.tsc_catchup = 1;
1272 vcpu->arch.tsc_always_catchup = 1;
1275 WARN(1, "user requested TSC rate below hardware speed\n");
1280 /* TSC scaling required - calculate ratio */
1281 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1282 user_tsc_khz, tsc_khz);
1284 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1285 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1290 vcpu->arch.tsc_scaling_ratio = ratio;
1294 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1296 u32 thresh_lo, thresh_hi;
1297 int use_scaling = 0;
1299 /* tsc_khz can be zero if TSC calibration fails */
1300 if (this_tsc_khz == 0) {
1301 /* set tsc_scaling_ratio to a safe value */
1302 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1306 /* Compute a scale to convert nanoseconds in TSC cycles */
1307 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1308 &vcpu->arch.virtual_tsc_shift,
1309 &vcpu->arch.virtual_tsc_mult);
1310 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1313 * Compute the variation in TSC rate which is acceptable
1314 * within the range of tolerance and decide if the
1315 * rate being applied is within that bounds of the hardware
1316 * rate. If so, no scaling or compensation need be done.
1318 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1319 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1320 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1321 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1324 return set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1327 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1329 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1330 vcpu->arch.virtual_tsc_mult,
1331 vcpu->arch.virtual_tsc_shift);
1332 tsc += vcpu->arch.this_tsc_write;
1336 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1338 #ifdef CONFIG_X86_64
1340 struct kvm_arch *ka = &vcpu->kvm->arch;
1341 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1343 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1344 atomic_read(&vcpu->kvm->online_vcpus));
1347 * Once the masterclock is enabled, always perform request in
1348 * order to update it.
1350 * In order to enable masterclock, the host clocksource must be TSC
1351 * and the vcpus need to have matched TSCs. When that happens,
1352 * perform request to enable masterclock.
1354 if (ka->use_master_clock ||
1355 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1356 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1358 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1359 atomic_read(&vcpu->kvm->online_vcpus),
1360 ka->use_master_clock, gtod->clock.vclock_mode);
1364 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1366 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1367 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1371 * Multiply tsc by a fixed point number represented by ratio.
1373 * The most significant 64-N bits (mult) of ratio represent the
1374 * integral part of the fixed point number; the remaining N bits
1375 * (frac) represent the fractional part, ie. ratio represents a fixed
1376 * point number (mult + frac * 2^(-N)).
1378 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1380 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1382 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1385 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1388 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1390 if (ratio != kvm_default_tsc_scaling_ratio)
1391 _tsc = __scale_tsc(ratio, tsc);
1395 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1397 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1401 tsc = kvm_scale_tsc(vcpu, rdtsc());
1403 return target_tsc - tsc;
1406 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1408 return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
1410 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1412 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1414 struct kvm *kvm = vcpu->kvm;
1415 u64 offset, ns, elapsed;
1416 unsigned long flags;
1419 bool already_matched;
1420 u64 data = msr->data;
1422 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1423 offset = kvm_compute_tsc_offset(vcpu, data);
1424 ns = get_kernel_ns();
1425 elapsed = ns - kvm->arch.last_tsc_nsec;
1427 if (vcpu->arch.virtual_tsc_khz) {
1430 /* n.b - signed multiplication and division required */
1431 usdiff = data - kvm->arch.last_tsc_write;
1432 #ifdef CONFIG_X86_64
1433 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1435 /* do_div() only does unsigned */
1436 asm("1: idivl %[divisor]\n"
1437 "2: xor %%edx, %%edx\n"
1438 " movl $0, %[faulted]\n"
1440 ".section .fixup,\"ax\"\n"
1441 "4: movl $1, %[faulted]\n"
1445 _ASM_EXTABLE(1b, 4b)
1447 : "=A"(usdiff), [faulted] "=r" (faulted)
1448 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1451 do_div(elapsed, 1000);
1456 /* idivl overflow => difference is larger than USEC_PER_SEC */
1458 usdiff = USEC_PER_SEC;
1460 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1463 * Special case: TSC write with a small delta (1 second) of virtual
1464 * cycle time against real time is interpreted as an attempt to
1465 * synchronize the CPU.
1467 * For a reliable TSC, we can match TSC offsets, and for an unstable
1468 * TSC, we add elapsed time in this computation. We could let the
1469 * compensation code attempt to catch up if we fall behind, but
1470 * it's better to try to match offsets from the beginning.
1472 if (usdiff < USEC_PER_SEC &&
1473 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1474 if (!check_tsc_unstable()) {
1475 offset = kvm->arch.cur_tsc_offset;
1476 pr_debug("kvm: matched tsc offset for %llu\n", data);
1478 u64 delta = nsec_to_cycles(vcpu, elapsed);
1480 offset = kvm_compute_tsc_offset(vcpu, data);
1481 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1484 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1487 * We split periods of matched TSC writes into generations.
1488 * For each generation, we track the original measured
1489 * nanosecond time, offset, and write, so if TSCs are in
1490 * sync, we can match exact offset, and if not, we can match
1491 * exact software computation in compute_guest_tsc()
1493 * These values are tracked in kvm->arch.cur_xxx variables.
1495 kvm->arch.cur_tsc_generation++;
1496 kvm->arch.cur_tsc_nsec = ns;
1497 kvm->arch.cur_tsc_write = data;
1498 kvm->arch.cur_tsc_offset = offset;
1500 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1501 kvm->arch.cur_tsc_generation, data);
1505 * We also track th most recent recorded KHZ, write and time to
1506 * allow the matching interval to be extended at each write.
1508 kvm->arch.last_tsc_nsec = ns;
1509 kvm->arch.last_tsc_write = data;
1510 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1512 vcpu->arch.last_guest_tsc = data;
1514 /* Keep track of which generation this VCPU has synchronized to */
1515 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1516 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1517 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1519 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1520 update_ia32_tsc_adjust_msr(vcpu, offset);
1521 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1522 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1524 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1526 kvm->arch.nr_vcpus_matched_tsc = 0;
1527 } else if (!already_matched) {
1528 kvm->arch.nr_vcpus_matched_tsc++;
1531 kvm_track_tsc_matching(vcpu);
1532 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1535 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1537 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1540 kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1543 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1545 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1546 WARN_ON(adjustment < 0);
1547 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1548 kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1551 #ifdef CONFIG_X86_64
1553 static cycle_t read_tsc(void)
1555 cycle_t ret = (cycle_t)rdtsc_ordered();
1556 u64 last = pvclock_gtod_data.clock.cycle_last;
1558 if (likely(ret >= last))
1562 * GCC likes to generate cmov here, but this branch is extremely
1563 * predictable (it's just a funciton of time and the likely is
1564 * very likely) and there's a data dependence, so force GCC
1565 * to generate a branch instead. I don't barrier() because
1566 * we don't actually need a barrier, and if this function
1567 * ever gets inlined it will generate worse code.
1573 static inline u64 vgettsc(cycle_t *cycle_now)
1576 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1578 *cycle_now = read_tsc();
1580 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1581 return v * gtod->clock.mult;
1584 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1586 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1592 seq = read_seqcount_begin(>od->seq);
1593 mode = gtod->clock.vclock_mode;
1594 ns = gtod->nsec_base;
1595 ns += vgettsc(cycle_now);
1596 ns >>= gtod->clock.shift;
1597 ns += gtod->boot_ns;
1598 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1604 /* returns true if host is using tsc clocksource */
1605 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1607 /* checked again under seqlock below */
1608 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1611 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1617 * Assuming a stable TSC across physical CPUS, and a stable TSC
1618 * across virtual CPUs, the following condition is possible.
1619 * Each numbered line represents an event visible to both
1620 * CPUs at the next numbered event.
1622 * "timespecX" represents host monotonic time. "tscX" represents
1625 * VCPU0 on CPU0 | VCPU1 on CPU1
1627 * 1. read timespec0,tsc0
1628 * 2. | timespec1 = timespec0 + N
1630 * 3. transition to guest | transition to guest
1631 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1632 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1633 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1635 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1638 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1640 * - 0 < N - M => M < N
1642 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1643 * always the case (the difference between two distinct xtime instances
1644 * might be smaller then the difference between corresponding TSC reads,
1645 * when updating guest vcpus pvclock areas).
1647 * To avoid that problem, do not allow visibility of distinct
1648 * system_timestamp/tsc_timestamp values simultaneously: use a master
1649 * copy of host monotonic time values. Update that master copy
1652 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1656 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1658 #ifdef CONFIG_X86_64
1659 struct kvm_arch *ka = &kvm->arch;
1661 bool host_tsc_clocksource, vcpus_matched;
1663 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1664 atomic_read(&kvm->online_vcpus));
1667 * If the host uses TSC clock, then passthrough TSC as stable
1670 host_tsc_clocksource = kvm_get_time_and_clockread(
1671 &ka->master_kernel_ns,
1672 &ka->master_cycle_now);
1674 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1675 && !backwards_tsc_observed
1676 && !ka->boot_vcpu_runs_old_kvmclock;
1678 if (ka->use_master_clock)
1679 atomic_set(&kvm_guest_has_master_clock, 1);
1681 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1682 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1687 static void kvm_gen_update_masterclock(struct kvm *kvm)
1689 #ifdef CONFIG_X86_64
1691 struct kvm_vcpu *vcpu;
1692 struct kvm_arch *ka = &kvm->arch;
1694 spin_lock(&ka->pvclock_gtod_sync_lock);
1695 kvm_make_mclock_inprogress_request(kvm);
1696 /* no guest entries from this point */
1697 pvclock_update_vm_gtod_copy(kvm);
1699 kvm_for_each_vcpu(i, vcpu, kvm)
1700 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1702 /* guest entries allowed */
1703 kvm_for_each_vcpu(i, vcpu, kvm)
1704 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1706 spin_unlock(&ka->pvclock_gtod_sync_lock);
1710 static int kvm_guest_time_update(struct kvm_vcpu *v)
1712 unsigned long flags, this_tsc_khz, tgt_tsc_khz;
1713 struct kvm_vcpu_arch *vcpu = &v->arch;
1714 struct kvm_arch *ka = &v->kvm->arch;
1716 u64 tsc_timestamp, host_tsc;
1717 struct pvclock_vcpu_time_info guest_hv_clock;
1719 bool use_master_clock;
1725 * If the host uses TSC clock, then passthrough TSC as stable
1728 spin_lock(&ka->pvclock_gtod_sync_lock);
1729 use_master_clock = ka->use_master_clock;
1730 if (use_master_clock) {
1731 host_tsc = ka->master_cycle_now;
1732 kernel_ns = ka->master_kernel_ns;
1734 spin_unlock(&ka->pvclock_gtod_sync_lock);
1736 /* Keep irq disabled to prevent changes to the clock */
1737 local_irq_save(flags);
1738 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1739 if (unlikely(this_tsc_khz == 0)) {
1740 local_irq_restore(flags);
1741 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1744 if (!use_master_clock) {
1746 kernel_ns = get_kernel_ns();
1749 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1752 * We may have to catch up the TSC to match elapsed wall clock
1753 * time for two reasons, even if kvmclock is used.
1754 * 1) CPU could have been running below the maximum TSC rate
1755 * 2) Broken TSC compensation resets the base at each VCPU
1756 * entry to avoid unknown leaps of TSC even when running
1757 * again on the same CPU. This may cause apparent elapsed
1758 * time to disappear, and the guest to stand still or run
1761 if (vcpu->tsc_catchup) {
1762 u64 tsc = compute_guest_tsc(v, kernel_ns);
1763 if (tsc > tsc_timestamp) {
1764 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1765 tsc_timestamp = tsc;
1769 local_irq_restore(flags);
1771 if (!vcpu->pv_time_enabled)
1774 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1775 tgt_tsc_khz = kvm_has_tsc_control ?
1776 vcpu->virtual_tsc_khz : this_tsc_khz;
1777 kvm_get_time_scale(NSEC_PER_SEC / 1000, tgt_tsc_khz,
1778 &vcpu->hv_clock.tsc_shift,
1779 &vcpu->hv_clock.tsc_to_system_mul);
1780 vcpu->hw_tsc_khz = this_tsc_khz;
1783 /* With all the info we got, fill in the values */
1784 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1785 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1786 vcpu->last_guest_tsc = tsc_timestamp;
1788 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1789 &guest_hv_clock, sizeof(guest_hv_clock))))
1792 /* This VCPU is paused, but it's legal for a guest to read another
1793 * VCPU's kvmclock, so we really have to follow the specification where
1794 * it says that version is odd if data is being modified, and even after
1797 * Version field updates must be kept separate. This is because
1798 * kvm_write_guest_cached might use a "rep movs" instruction, and
1799 * writes within a string instruction are weakly ordered. So there
1800 * are three writes overall.
1802 * As a small optimization, only write the version field in the first
1803 * and third write. The vcpu->pv_time cache is still valid, because the
1804 * version field is the first in the struct.
1806 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1808 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1809 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1811 sizeof(vcpu->hv_clock.version));
1815 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1816 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1818 if (vcpu->pvclock_set_guest_stopped_request) {
1819 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1820 vcpu->pvclock_set_guest_stopped_request = false;
1823 /* If the host uses TSC clocksource, then it is stable */
1824 if (use_master_clock)
1825 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1827 vcpu->hv_clock.flags = pvclock_flags;
1829 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1831 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1833 sizeof(vcpu->hv_clock));
1837 vcpu->hv_clock.version++;
1838 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1840 sizeof(vcpu->hv_clock.version));
1845 * kvmclock updates which are isolated to a given vcpu, such as
1846 * vcpu->cpu migration, should not allow system_timestamp from
1847 * the rest of the vcpus to remain static. Otherwise ntp frequency
1848 * correction applies to one vcpu's system_timestamp but not
1851 * So in those cases, request a kvmclock update for all vcpus.
1852 * We need to rate-limit these requests though, as they can
1853 * considerably slow guests that have a large number of vcpus.
1854 * The time for a remote vcpu to update its kvmclock is bound
1855 * by the delay we use to rate-limit the updates.
1858 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1860 static void kvmclock_update_fn(struct work_struct *work)
1863 struct delayed_work *dwork = to_delayed_work(work);
1864 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1865 kvmclock_update_work);
1866 struct kvm *kvm = container_of(ka, struct kvm, arch);
1867 struct kvm_vcpu *vcpu;
1869 kvm_for_each_vcpu(i, vcpu, kvm) {
1870 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1871 kvm_vcpu_kick(vcpu);
1875 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1877 struct kvm *kvm = v->kvm;
1879 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1880 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1881 KVMCLOCK_UPDATE_DELAY);
1884 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1886 static void kvmclock_sync_fn(struct work_struct *work)
1888 struct delayed_work *dwork = to_delayed_work(work);
1889 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1890 kvmclock_sync_work);
1891 struct kvm *kvm = container_of(ka, struct kvm, arch);
1893 if (!kvmclock_periodic_sync)
1896 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1897 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1898 KVMCLOCK_SYNC_PERIOD);
1901 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1903 u64 mcg_cap = vcpu->arch.mcg_cap;
1904 unsigned bank_num = mcg_cap & 0xff;
1907 case MSR_IA32_MCG_STATUS:
1908 vcpu->arch.mcg_status = data;
1910 case MSR_IA32_MCG_CTL:
1911 if (!(mcg_cap & MCG_CTL_P))
1913 if (data != 0 && data != ~(u64)0)
1915 vcpu->arch.mcg_ctl = data;
1918 if (msr >= MSR_IA32_MC0_CTL &&
1919 msr < MSR_IA32_MCx_CTL(bank_num)) {
1920 u32 offset = msr - MSR_IA32_MC0_CTL;
1921 /* only 0 or all 1s can be written to IA32_MCi_CTL
1922 * some Linux kernels though clear bit 10 in bank 4 to
1923 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1924 * this to avoid an uncatched #GP in the guest
1926 if ((offset & 0x3) == 0 &&
1927 data != 0 && (data | (1 << 10)) != ~(u64)0)
1929 vcpu->arch.mce_banks[offset] = data;
1937 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1939 struct kvm *kvm = vcpu->kvm;
1940 int lm = is_long_mode(vcpu);
1941 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1942 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1943 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1944 : kvm->arch.xen_hvm_config.blob_size_32;
1945 u32 page_num = data & ~PAGE_MASK;
1946 u64 page_addr = data & PAGE_MASK;
1951 if (page_num >= blob_size)
1954 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1959 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1968 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1970 gpa_t gpa = data & ~0x3f;
1972 /* Bits 2:5 are reserved, Should be zero */
1976 vcpu->arch.apf.msr_val = data;
1978 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1979 kvm_clear_async_pf_completion_queue(vcpu);
1980 kvm_async_pf_hash_reset(vcpu);
1984 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1988 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1989 kvm_async_pf_wakeup_all(vcpu);
1993 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1995 vcpu->arch.pv_time_enabled = false;
1998 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2002 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2005 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2006 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2007 vcpu->arch.st.accum_steal = delta;
2010 static void record_steal_time(struct kvm_vcpu *vcpu)
2012 accumulate_steal_time(vcpu);
2014 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2017 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2018 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2021 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2022 vcpu->arch.st.steal.version += 2;
2023 vcpu->arch.st.accum_steal = 0;
2025 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2026 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2029 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2032 u32 msr = msr_info->index;
2033 u64 data = msr_info->data;
2036 case MSR_AMD64_NB_CFG:
2037 case MSR_IA32_UCODE_REV:
2038 case MSR_IA32_UCODE_WRITE:
2039 case MSR_VM_HSAVE_PA:
2040 case MSR_AMD64_PATCH_LOADER:
2041 case MSR_AMD64_BU_CFG2:
2045 return set_efer(vcpu, data);
2047 data &= ~(u64)0x40; /* ignore flush filter disable */
2048 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2049 data &= ~(u64)0x8; /* ignore TLB cache disable */
2050 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2052 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2057 case MSR_FAM10H_MMIO_CONF_BASE:
2059 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2064 case MSR_IA32_DEBUGCTLMSR:
2066 /* We support the non-activated case already */
2068 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2069 /* Values other than LBR and BTF are vendor-specific,
2070 thus reserved and should throw a #GP */
2073 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2076 case 0x200 ... 0x2ff:
2077 return kvm_mtrr_set_msr(vcpu, msr, data);
2078 case MSR_IA32_APICBASE:
2079 return kvm_set_apic_base(vcpu, msr_info);
2080 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2081 return kvm_x2apic_msr_write(vcpu, msr, data);
2082 case MSR_IA32_TSCDEADLINE:
2083 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2085 case MSR_IA32_TSC_ADJUST:
2086 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2087 if (!msr_info->host_initiated) {
2088 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2089 adjust_tsc_offset_guest(vcpu, adj);
2091 vcpu->arch.ia32_tsc_adjust_msr = data;
2094 case MSR_IA32_MISC_ENABLE:
2095 vcpu->arch.ia32_misc_enable_msr = data;
2097 case MSR_IA32_SMBASE:
2098 if (!msr_info->host_initiated)
2100 vcpu->arch.smbase = data;
2102 case MSR_KVM_WALL_CLOCK_NEW:
2103 case MSR_KVM_WALL_CLOCK:
2104 vcpu->kvm->arch.wall_clock = data;
2105 kvm_write_wall_clock(vcpu->kvm, data);
2107 case MSR_KVM_SYSTEM_TIME_NEW:
2108 case MSR_KVM_SYSTEM_TIME: {
2110 struct kvm_arch *ka = &vcpu->kvm->arch;
2112 kvmclock_reset(vcpu);
2114 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2115 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2117 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2118 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2121 ka->boot_vcpu_runs_old_kvmclock = tmp;
2124 vcpu->arch.time = data;
2125 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2127 /* we verify if the enable bit is set... */
2131 gpa_offset = data & ~(PAGE_MASK | 1);
2133 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2134 &vcpu->arch.pv_time, data & ~1ULL,
2135 sizeof(struct pvclock_vcpu_time_info)))
2136 vcpu->arch.pv_time_enabled = false;
2138 vcpu->arch.pv_time_enabled = true;
2142 case MSR_KVM_ASYNC_PF_EN:
2143 if (kvm_pv_enable_async_pf(vcpu, data))
2146 case MSR_KVM_STEAL_TIME:
2148 if (unlikely(!sched_info_on()))
2151 if (data & KVM_STEAL_RESERVED_MASK)
2154 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2155 data & KVM_STEAL_VALID_BITS,
2156 sizeof(struct kvm_steal_time)))
2159 vcpu->arch.st.msr_val = data;
2161 if (!(data & KVM_MSR_ENABLED))
2164 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2167 case MSR_KVM_PV_EOI_EN:
2168 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2172 case MSR_IA32_MCG_CTL:
2173 case MSR_IA32_MCG_STATUS:
2174 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2175 return set_msr_mce(vcpu, msr, data);
2177 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2178 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2179 pr = true; /* fall through */
2180 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2181 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2182 if (kvm_pmu_is_valid_msr(vcpu, msr))
2183 return kvm_pmu_set_msr(vcpu, msr_info);
2185 if (pr || data != 0)
2186 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2187 "0x%x data 0x%llx\n", msr, data);
2189 case MSR_K7_CLK_CTL:
2191 * Ignore all writes to this no longer documented MSR.
2192 * Writes are only relevant for old K7 processors,
2193 * all pre-dating SVM, but a recommended workaround from
2194 * AMD for these chips. It is possible to specify the
2195 * affected processor models on the command line, hence
2196 * the need to ignore the workaround.
2199 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2200 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2201 case HV_X64_MSR_CRASH_CTL:
2202 return kvm_hv_set_msr_common(vcpu, msr, data,
2203 msr_info->host_initiated);
2204 case MSR_IA32_BBL_CR_CTL3:
2205 /* Drop writes to this legacy MSR -- see rdmsr
2206 * counterpart for further detail.
2208 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2210 case MSR_AMD64_OSVW_ID_LENGTH:
2211 if (!guest_cpuid_has_osvw(vcpu))
2213 vcpu->arch.osvw.length = data;
2215 case MSR_AMD64_OSVW_STATUS:
2216 if (!guest_cpuid_has_osvw(vcpu))
2218 vcpu->arch.osvw.status = data;
2221 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2222 return xen_hvm_config(vcpu, data);
2223 if (kvm_pmu_is_valid_msr(vcpu, msr))
2224 return kvm_pmu_set_msr(vcpu, msr_info);
2226 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2230 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2237 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2241 * Reads an msr value (of 'msr_index') into 'pdata'.
2242 * Returns 0 on success, non-0 otherwise.
2243 * Assumes vcpu_load() was already called.
2245 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2247 return kvm_x86_ops->get_msr(vcpu, msr);
2249 EXPORT_SYMBOL_GPL(kvm_get_msr);
2251 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2254 u64 mcg_cap = vcpu->arch.mcg_cap;
2255 unsigned bank_num = mcg_cap & 0xff;
2258 case MSR_IA32_P5_MC_ADDR:
2259 case MSR_IA32_P5_MC_TYPE:
2262 case MSR_IA32_MCG_CAP:
2263 data = vcpu->arch.mcg_cap;
2265 case MSR_IA32_MCG_CTL:
2266 if (!(mcg_cap & MCG_CTL_P))
2268 data = vcpu->arch.mcg_ctl;
2270 case MSR_IA32_MCG_STATUS:
2271 data = vcpu->arch.mcg_status;
2274 if (msr >= MSR_IA32_MC0_CTL &&
2275 msr < MSR_IA32_MCx_CTL(bank_num)) {
2276 u32 offset = msr - MSR_IA32_MC0_CTL;
2277 data = vcpu->arch.mce_banks[offset];
2286 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2288 switch (msr_info->index) {
2289 case MSR_IA32_PLATFORM_ID:
2290 case MSR_IA32_EBL_CR_POWERON:
2291 case MSR_IA32_DEBUGCTLMSR:
2292 case MSR_IA32_LASTBRANCHFROMIP:
2293 case MSR_IA32_LASTBRANCHTOIP:
2294 case MSR_IA32_LASTINTFROMIP:
2295 case MSR_IA32_LASTINTTOIP:
2297 case MSR_K8_TSEG_ADDR:
2298 case MSR_K8_TSEG_MASK:
2300 case MSR_VM_HSAVE_PA:
2301 case MSR_K8_INT_PENDING_MSG:
2302 case MSR_AMD64_NB_CFG:
2303 case MSR_FAM10H_MMIO_CONF_BASE:
2304 case MSR_AMD64_BU_CFG2:
2307 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2308 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2309 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2310 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2311 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2312 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2315 case MSR_IA32_UCODE_REV:
2316 msr_info->data = 0x100000000ULL;
2319 case 0x200 ... 0x2ff:
2320 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2321 case 0xcd: /* fsb frequency */
2325 * MSR_EBC_FREQUENCY_ID
2326 * Conservative value valid for even the basic CPU models.
2327 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2328 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2329 * and 266MHz for model 3, or 4. Set Core Clock
2330 * Frequency to System Bus Frequency Ratio to 1 (bits
2331 * 31:24) even though these are only valid for CPU
2332 * models > 2, however guests may end up dividing or
2333 * multiplying by zero otherwise.
2335 case MSR_EBC_FREQUENCY_ID:
2336 msr_info->data = 1 << 24;
2338 case MSR_IA32_APICBASE:
2339 msr_info->data = kvm_get_apic_base(vcpu);
2341 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2342 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2344 case MSR_IA32_TSCDEADLINE:
2345 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2347 case MSR_IA32_TSC_ADJUST:
2348 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2350 case MSR_IA32_MISC_ENABLE:
2351 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2353 case MSR_IA32_SMBASE:
2354 if (!msr_info->host_initiated)
2356 msr_info->data = vcpu->arch.smbase;
2358 case MSR_IA32_PERF_STATUS:
2359 /* TSC increment by tick */
2360 msr_info->data = 1000ULL;
2361 /* CPU multiplier */
2362 msr_info->data |= (((uint64_t)4ULL) << 40);
2365 msr_info->data = vcpu->arch.efer;
2367 case MSR_KVM_WALL_CLOCK:
2368 case MSR_KVM_WALL_CLOCK_NEW:
2369 msr_info->data = vcpu->kvm->arch.wall_clock;
2371 case MSR_KVM_SYSTEM_TIME:
2372 case MSR_KVM_SYSTEM_TIME_NEW:
2373 msr_info->data = vcpu->arch.time;
2375 case MSR_KVM_ASYNC_PF_EN:
2376 msr_info->data = vcpu->arch.apf.msr_val;
2378 case MSR_KVM_STEAL_TIME:
2379 msr_info->data = vcpu->arch.st.msr_val;
2381 case MSR_KVM_PV_EOI_EN:
2382 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2384 case MSR_IA32_P5_MC_ADDR:
2385 case MSR_IA32_P5_MC_TYPE:
2386 case MSR_IA32_MCG_CAP:
2387 case MSR_IA32_MCG_CTL:
2388 case MSR_IA32_MCG_STATUS:
2389 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2390 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2391 case MSR_K7_CLK_CTL:
2393 * Provide expected ramp-up count for K7. All other
2394 * are set to zero, indicating minimum divisors for
2397 * This prevents guest kernels on AMD host with CPU
2398 * type 6, model 8 and higher from exploding due to
2399 * the rdmsr failing.
2401 msr_info->data = 0x20000000;
2403 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2404 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2405 case HV_X64_MSR_CRASH_CTL:
2406 return kvm_hv_get_msr_common(vcpu,
2407 msr_info->index, &msr_info->data);
2409 case MSR_IA32_BBL_CR_CTL3:
2410 /* This legacy MSR exists but isn't fully documented in current
2411 * silicon. It is however accessed by winxp in very narrow
2412 * scenarios where it sets bit #19, itself documented as
2413 * a "reserved" bit. Best effort attempt to source coherent
2414 * read data here should the balance of the register be
2415 * interpreted by the guest:
2417 * L2 cache control register 3: 64GB range, 256KB size,
2418 * enabled, latency 0x1, configured
2420 msr_info->data = 0xbe702111;
2422 case MSR_AMD64_OSVW_ID_LENGTH:
2423 if (!guest_cpuid_has_osvw(vcpu))
2425 msr_info->data = vcpu->arch.osvw.length;
2427 case MSR_AMD64_OSVW_STATUS:
2428 if (!guest_cpuid_has_osvw(vcpu))
2430 msr_info->data = vcpu->arch.osvw.status;
2433 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2434 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2436 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2439 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2446 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2449 * Read or write a bunch of msrs. All parameters are kernel addresses.
2451 * @return number of msrs set successfully.
2453 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2454 struct kvm_msr_entry *entries,
2455 int (*do_msr)(struct kvm_vcpu *vcpu,
2456 unsigned index, u64 *data))
2460 idx = srcu_read_lock(&vcpu->kvm->srcu);
2461 for (i = 0; i < msrs->nmsrs; ++i)
2462 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2464 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2470 * Read or write a bunch of msrs. Parameters are user addresses.
2472 * @return number of msrs set successfully.
2474 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2475 int (*do_msr)(struct kvm_vcpu *vcpu,
2476 unsigned index, u64 *data),
2479 struct kvm_msrs msrs;
2480 struct kvm_msr_entry *entries;
2485 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2489 if (msrs.nmsrs >= MAX_IO_MSRS)
2492 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2493 entries = memdup_user(user_msrs->entries, size);
2494 if (IS_ERR(entries)) {
2495 r = PTR_ERR(entries);
2499 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2504 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2515 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2520 case KVM_CAP_IRQCHIP:
2522 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2523 case KVM_CAP_SET_TSS_ADDR:
2524 case KVM_CAP_EXT_CPUID:
2525 case KVM_CAP_EXT_EMUL_CPUID:
2526 case KVM_CAP_CLOCKSOURCE:
2528 case KVM_CAP_NOP_IO_DELAY:
2529 case KVM_CAP_MP_STATE:
2530 case KVM_CAP_SYNC_MMU:
2531 case KVM_CAP_USER_NMI:
2532 case KVM_CAP_REINJECT_CONTROL:
2533 case KVM_CAP_IRQ_INJECT_STATUS:
2534 case KVM_CAP_IOEVENTFD:
2535 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2537 case KVM_CAP_PIT_STATE2:
2538 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2539 case KVM_CAP_XEN_HVM:
2540 case KVM_CAP_ADJUST_CLOCK:
2541 case KVM_CAP_VCPU_EVENTS:
2542 case KVM_CAP_HYPERV:
2543 case KVM_CAP_HYPERV_VAPIC:
2544 case KVM_CAP_HYPERV_SPIN:
2545 case KVM_CAP_HYPERV_SYNIC:
2546 case KVM_CAP_PCI_SEGMENT:
2547 case KVM_CAP_DEBUGREGS:
2548 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2550 case KVM_CAP_ASYNC_PF:
2551 case KVM_CAP_GET_TSC_KHZ:
2552 case KVM_CAP_KVMCLOCK_CTRL:
2553 case KVM_CAP_READONLY_MEM:
2554 case KVM_CAP_HYPERV_TIME:
2555 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2556 case KVM_CAP_TSC_DEADLINE_TIMER:
2557 case KVM_CAP_ENABLE_CAP_VM:
2558 case KVM_CAP_DISABLE_QUIRKS:
2559 case KVM_CAP_SET_BOOT_CPU_ID:
2560 case KVM_CAP_SPLIT_IRQCHIP:
2561 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2562 case KVM_CAP_ASSIGN_DEV_IRQ:
2563 case KVM_CAP_PCI_2_3:
2567 case KVM_CAP_X86_SMM:
2568 /* SMBASE is usually relocated above 1M on modern chipsets,
2569 * and SMM handlers might indeed rely on 4G segment limits,
2570 * so do not report SMM to be available if real mode is
2571 * emulated via vm86 mode. Still, do not go to great lengths
2572 * to avoid userspace's usage of the feature, because it is a
2573 * fringe case that is not enabled except via specific settings
2574 * of the module parameters.
2576 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2578 case KVM_CAP_COALESCED_MMIO:
2579 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2582 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2584 case KVM_CAP_NR_VCPUS:
2585 r = KVM_SOFT_MAX_VCPUS;
2587 case KVM_CAP_MAX_VCPUS:
2590 case KVM_CAP_NR_MEMSLOTS:
2591 r = KVM_USER_MEM_SLOTS;
2593 case KVM_CAP_PV_MMU: /* obsolete */
2596 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2598 r = iommu_present(&pci_bus_type);
2602 r = KVM_MAX_MCE_BANKS;
2607 case KVM_CAP_TSC_CONTROL:
2608 r = kvm_has_tsc_control;
2618 long kvm_arch_dev_ioctl(struct file *filp,
2619 unsigned int ioctl, unsigned long arg)
2621 void __user *argp = (void __user *)arg;
2625 case KVM_GET_MSR_INDEX_LIST: {
2626 struct kvm_msr_list __user *user_msr_list = argp;
2627 struct kvm_msr_list msr_list;
2631 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2634 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2635 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2638 if (n < msr_list.nmsrs)
2641 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2642 num_msrs_to_save * sizeof(u32)))
2644 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2646 num_emulated_msrs * sizeof(u32)))
2651 case KVM_GET_SUPPORTED_CPUID:
2652 case KVM_GET_EMULATED_CPUID: {
2653 struct kvm_cpuid2 __user *cpuid_arg = argp;
2654 struct kvm_cpuid2 cpuid;
2657 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2660 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2666 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2671 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2674 mce_cap = KVM_MCE_CAP_SUPPORTED;
2676 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2688 static void wbinvd_ipi(void *garbage)
2693 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2695 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2698 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2700 /* Address WBINVD may be executed by guest */
2701 if (need_emulate_wbinvd(vcpu)) {
2702 if (kvm_x86_ops->has_wbinvd_exit())
2703 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2704 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2705 smp_call_function_single(vcpu->cpu,
2706 wbinvd_ipi, NULL, 1);
2709 kvm_x86_ops->vcpu_load(vcpu, cpu);
2711 /* Apply any externally detected TSC adjustments (due to suspend) */
2712 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2713 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2714 vcpu->arch.tsc_offset_adjustment = 0;
2715 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2718 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2719 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2720 rdtsc() - vcpu->arch.last_host_tsc;
2722 mark_tsc_unstable("KVM discovered backwards TSC");
2723 if (check_tsc_unstable()) {
2724 u64 offset = kvm_compute_tsc_offset(vcpu,
2725 vcpu->arch.last_guest_tsc);
2726 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2727 vcpu->arch.tsc_catchup = 1;
2730 * On a host with synchronized TSC, there is no need to update
2731 * kvmclock on vcpu->cpu migration
2733 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2734 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2735 if (vcpu->cpu != cpu)
2736 kvm_migrate_timers(vcpu);
2740 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2743 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2745 kvm_x86_ops->vcpu_put(vcpu);
2746 kvm_put_guest_fpu(vcpu);
2747 vcpu->arch.last_host_tsc = rdtsc();
2750 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2751 struct kvm_lapic_state *s)
2753 if (vcpu->arch.apicv_active)
2754 kvm_x86_ops->sync_pir_to_irr(vcpu);
2756 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2761 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2762 struct kvm_lapic_state *s)
2764 kvm_apic_post_state_restore(vcpu, s);
2765 update_cr8_intercept(vcpu);
2770 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2772 return (!lapic_in_kernel(vcpu) ||
2773 kvm_apic_accept_pic_intr(vcpu));
2777 * if userspace requested an interrupt window, check that the
2778 * interrupt window is open.
2780 * No need to exit to userspace if we already have an interrupt queued.
2782 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2784 return kvm_arch_interrupt_allowed(vcpu) &&
2785 !kvm_cpu_has_interrupt(vcpu) &&
2786 !kvm_event_needs_reinjection(vcpu) &&
2787 kvm_cpu_accept_dm_intr(vcpu);
2790 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2791 struct kvm_interrupt *irq)
2793 if (irq->irq >= KVM_NR_INTERRUPTS)
2796 if (!irqchip_in_kernel(vcpu->kvm)) {
2797 kvm_queue_interrupt(vcpu, irq->irq, false);
2798 kvm_make_request(KVM_REQ_EVENT, vcpu);
2803 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2804 * fail for in-kernel 8259.
2806 if (pic_in_kernel(vcpu->kvm))
2809 if (vcpu->arch.pending_external_vector != -1)
2812 vcpu->arch.pending_external_vector = irq->irq;
2813 kvm_make_request(KVM_REQ_EVENT, vcpu);
2817 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2819 kvm_inject_nmi(vcpu);
2824 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2826 kvm_make_request(KVM_REQ_SMI, vcpu);
2831 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2832 struct kvm_tpr_access_ctl *tac)
2836 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2840 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2844 unsigned bank_num = mcg_cap & 0xff, bank;
2847 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2849 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2852 vcpu->arch.mcg_cap = mcg_cap;
2853 /* Init IA32_MCG_CTL to all 1s */
2854 if (mcg_cap & MCG_CTL_P)
2855 vcpu->arch.mcg_ctl = ~(u64)0;
2856 /* Init IA32_MCi_CTL to all 1s */
2857 for (bank = 0; bank < bank_num; bank++)
2858 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2863 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2864 struct kvm_x86_mce *mce)
2866 u64 mcg_cap = vcpu->arch.mcg_cap;
2867 unsigned bank_num = mcg_cap & 0xff;
2868 u64 *banks = vcpu->arch.mce_banks;
2870 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2873 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2874 * reporting is disabled
2876 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2877 vcpu->arch.mcg_ctl != ~(u64)0)
2879 banks += 4 * mce->bank;
2881 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2882 * reporting is disabled for the bank
2884 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2886 if (mce->status & MCI_STATUS_UC) {
2887 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2888 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2889 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2892 if (banks[1] & MCI_STATUS_VAL)
2893 mce->status |= MCI_STATUS_OVER;
2894 banks[2] = mce->addr;
2895 banks[3] = mce->misc;
2896 vcpu->arch.mcg_status = mce->mcg_status;
2897 banks[1] = mce->status;
2898 kvm_queue_exception(vcpu, MC_VECTOR);
2899 } else if (!(banks[1] & MCI_STATUS_VAL)
2900 || !(banks[1] & MCI_STATUS_UC)) {
2901 if (banks[1] & MCI_STATUS_VAL)
2902 mce->status |= MCI_STATUS_OVER;
2903 banks[2] = mce->addr;
2904 banks[3] = mce->misc;
2905 banks[1] = mce->status;
2907 banks[1] |= MCI_STATUS_OVER;
2911 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2912 struct kvm_vcpu_events *events)
2915 events->exception.injected =
2916 vcpu->arch.exception.pending &&
2917 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2918 events->exception.nr = vcpu->arch.exception.nr;
2919 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2920 events->exception.pad = 0;
2921 events->exception.error_code = vcpu->arch.exception.error_code;
2923 events->interrupt.injected =
2924 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2925 events->interrupt.nr = vcpu->arch.interrupt.nr;
2926 events->interrupt.soft = 0;
2927 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2929 events->nmi.injected = vcpu->arch.nmi_injected;
2930 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2931 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2932 events->nmi.pad = 0;
2934 events->sipi_vector = 0; /* never valid when reporting to user space */
2936 events->smi.smm = is_smm(vcpu);
2937 events->smi.pending = vcpu->arch.smi_pending;
2938 events->smi.smm_inside_nmi =
2939 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2940 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2942 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2943 | KVM_VCPUEVENT_VALID_SHADOW
2944 | KVM_VCPUEVENT_VALID_SMM);
2945 memset(&events->reserved, 0, sizeof(events->reserved));
2948 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2949 struct kvm_vcpu_events *events)
2951 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2952 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2953 | KVM_VCPUEVENT_VALID_SHADOW
2954 | KVM_VCPUEVENT_VALID_SMM))
2958 vcpu->arch.exception.pending = events->exception.injected;
2959 vcpu->arch.exception.nr = events->exception.nr;
2960 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2961 vcpu->arch.exception.error_code = events->exception.error_code;
2963 vcpu->arch.interrupt.pending = events->interrupt.injected;
2964 vcpu->arch.interrupt.nr = events->interrupt.nr;
2965 vcpu->arch.interrupt.soft = events->interrupt.soft;
2966 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2967 kvm_x86_ops->set_interrupt_shadow(vcpu,
2968 events->interrupt.shadow);
2970 vcpu->arch.nmi_injected = events->nmi.injected;
2971 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2972 vcpu->arch.nmi_pending = events->nmi.pending;
2973 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2975 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2976 kvm_vcpu_has_lapic(vcpu))
2977 vcpu->arch.apic->sipi_vector = events->sipi_vector;
2979 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
2980 if (events->smi.smm)
2981 vcpu->arch.hflags |= HF_SMM_MASK;
2983 vcpu->arch.hflags &= ~HF_SMM_MASK;
2984 vcpu->arch.smi_pending = events->smi.pending;
2985 if (events->smi.smm_inside_nmi)
2986 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
2988 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
2989 if (kvm_vcpu_has_lapic(vcpu)) {
2990 if (events->smi.latched_init)
2991 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2993 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2997 kvm_make_request(KVM_REQ_EVENT, vcpu);
3002 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3003 struct kvm_debugregs *dbgregs)
3007 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3008 kvm_get_dr(vcpu, 6, &val);
3010 dbgregs->dr7 = vcpu->arch.dr7;
3012 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3015 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3016 struct kvm_debugregs *dbgregs)
3021 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3022 kvm_update_dr0123(vcpu);
3023 vcpu->arch.dr6 = dbgregs->dr6;
3024 kvm_update_dr6(vcpu);
3025 vcpu->arch.dr7 = dbgregs->dr7;
3026 kvm_update_dr7(vcpu);
3031 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3033 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3035 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3036 u64 xstate_bv = xsave->header.xfeatures;
3040 * Copy legacy XSAVE area, to avoid complications with CPUID
3041 * leaves 0 and 1 in the loop below.
3043 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3046 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3049 * Copy each region from the possibly compacted offset to the
3050 * non-compacted offset.
3052 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3054 u64 feature = valid & -valid;
3055 int index = fls64(feature) - 1;
3056 void *src = get_xsave_addr(xsave, feature);
3059 u32 size, offset, ecx, edx;
3060 cpuid_count(XSTATE_CPUID, index,
3061 &size, &offset, &ecx, &edx);
3062 memcpy(dest + offset, src, size);
3069 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3071 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3072 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3076 * Copy legacy XSAVE area, to avoid complications with CPUID
3077 * leaves 0 and 1 in the loop below.
3079 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3081 /* Set XSTATE_BV and possibly XCOMP_BV. */
3082 xsave->header.xfeatures = xstate_bv;
3084 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3087 * Copy each region from the non-compacted offset to the
3088 * possibly compacted offset.
3090 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3092 u64 feature = valid & -valid;
3093 int index = fls64(feature) - 1;
3094 void *dest = get_xsave_addr(xsave, feature);
3097 u32 size, offset, ecx, edx;
3098 cpuid_count(XSTATE_CPUID, index,
3099 &size, &offset, &ecx, &edx);
3100 memcpy(dest, src + offset, size);
3107 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3108 struct kvm_xsave *guest_xsave)
3110 if (cpu_has_xsave) {
3111 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3112 fill_xsave((u8 *) guest_xsave->region, vcpu);
3114 memcpy(guest_xsave->region,
3115 &vcpu->arch.guest_fpu.state.fxsave,
3116 sizeof(struct fxregs_state));
3117 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3118 XFEATURE_MASK_FPSSE;
3122 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3123 struct kvm_xsave *guest_xsave)
3126 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3128 if (cpu_has_xsave) {
3130 * Here we allow setting states that are not present in
3131 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3132 * with old userspace.
3134 if (xstate_bv & ~kvm_supported_xcr0())
3136 load_xsave(vcpu, (u8 *)guest_xsave->region);
3138 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3140 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3141 guest_xsave->region, sizeof(struct fxregs_state));
3146 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3147 struct kvm_xcrs *guest_xcrs)
3149 if (!cpu_has_xsave) {
3150 guest_xcrs->nr_xcrs = 0;
3154 guest_xcrs->nr_xcrs = 1;
3155 guest_xcrs->flags = 0;
3156 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3157 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3160 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3161 struct kvm_xcrs *guest_xcrs)
3168 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3171 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3172 /* Only support XCR0 currently */
3173 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3174 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3175 guest_xcrs->xcrs[i].value);
3184 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3185 * stopped by the hypervisor. This function will be called from the host only.
3186 * EINVAL is returned when the host attempts to set the flag for a guest that
3187 * does not support pv clocks.
3189 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3191 if (!vcpu->arch.pv_time_enabled)
3193 vcpu->arch.pvclock_set_guest_stopped_request = true;
3194 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3198 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3199 struct kvm_enable_cap *cap)
3205 case KVM_CAP_HYPERV_SYNIC:
3206 return kvm_hv_activate_synic(vcpu);
3212 long kvm_arch_vcpu_ioctl(struct file *filp,
3213 unsigned int ioctl, unsigned long arg)
3215 struct kvm_vcpu *vcpu = filp->private_data;
3216 void __user *argp = (void __user *)arg;
3219 struct kvm_lapic_state *lapic;
3220 struct kvm_xsave *xsave;
3221 struct kvm_xcrs *xcrs;
3227 case KVM_GET_LAPIC: {
3229 if (!vcpu->arch.apic)
3231 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3236 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3240 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3245 case KVM_SET_LAPIC: {
3247 if (!vcpu->arch.apic)
3249 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3250 if (IS_ERR(u.lapic))
3251 return PTR_ERR(u.lapic);
3253 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3256 case KVM_INTERRUPT: {
3257 struct kvm_interrupt irq;
3260 if (copy_from_user(&irq, argp, sizeof irq))
3262 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3266 r = kvm_vcpu_ioctl_nmi(vcpu);
3270 r = kvm_vcpu_ioctl_smi(vcpu);
3273 case KVM_SET_CPUID: {
3274 struct kvm_cpuid __user *cpuid_arg = argp;
3275 struct kvm_cpuid cpuid;
3278 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3280 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3283 case KVM_SET_CPUID2: {
3284 struct kvm_cpuid2 __user *cpuid_arg = argp;
3285 struct kvm_cpuid2 cpuid;
3288 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3290 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3291 cpuid_arg->entries);
3294 case KVM_GET_CPUID2: {
3295 struct kvm_cpuid2 __user *cpuid_arg = argp;
3296 struct kvm_cpuid2 cpuid;
3299 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3301 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3302 cpuid_arg->entries);
3306 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3312 r = msr_io(vcpu, argp, do_get_msr, 1);
3315 r = msr_io(vcpu, argp, do_set_msr, 0);
3317 case KVM_TPR_ACCESS_REPORTING: {
3318 struct kvm_tpr_access_ctl tac;
3321 if (copy_from_user(&tac, argp, sizeof tac))
3323 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3327 if (copy_to_user(argp, &tac, sizeof tac))
3332 case KVM_SET_VAPIC_ADDR: {
3333 struct kvm_vapic_addr va;
3336 if (!lapic_in_kernel(vcpu))
3339 if (copy_from_user(&va, argp, sizeof va))
3341 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3344 case KVM_X86_SETUP_MCE: {
3348 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3350 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3353 case KVM_X86_SET_MCE: {
3354 struct kvm_x86_mce mce;
3357 if (copy_from_user(&mce, argp, sizeof mce))
3359 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3362 case KVM_GET_VCPU_EVENTS: {
3363 struct kvm_vcpu_events events;
3365 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3368 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3373 case KVM_SET_VCPU_EVENTS: {
3374 struct kvm_vcpu_events events;
3377 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3380 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3383 case KVM_GET_DEBUGREGS: {
3384 struct kvm_debugregs dbgregs;
3386 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3389 if (copy_to_user(argp, &dbgregs,
3390 sizeof(struct kvm_debugregs)))
3395 case KVM_SET_DEBUGREGS: {
3396 struct kvm_debugregs dbgregs;
3399 if (copy_from_user(&dbgregs, argp,
3400 sizeof(struct kvm_debugregs)))
3403 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3406 case KVM_GET_XSAVE: {
3407 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3412 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3415 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3420 case KVM_SET_XSAVE: {
3421 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3422 if (IS_ERR(u.xsave))
3423 return PTR_ERR(u.xsave);
3425 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3428 case KVM_GET_XCRS: {
3429 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3434 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3437 if (copy_to_user(argp, u.xcrs,
3438 sizeof(struct kvm_xcrs)))
3443 case KVM_SET_XCRS: {
3444 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3446 return PTR_ERR(u.xcrs);
3448 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3451 case KVM_SET_TSC_KHZ: {
3455 user_tsc_khz = (u32)arg;
3457 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3460 if (user_tsc_khz == 0)
3461 user_tsc_khz = tsc_khz;
3463 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3468 case KVM_GET_TSC_KHZ: {
3469 r = vcpu->arch.virtual_tsc_khz;
3472 case KVM_KVMCLOCK_CTRL: {
3473 r = kvm_set_guest_paused(vcpu);
3476 case KVM_ENABLE_CAP: {
3477 struct kvm_enable_cap cap;
3480 if (copy_from_user(&cap, argp, sizeof(cap)))
3482 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3493 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3495 return VM_FAULT_SIGBUS;
3498 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3502 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3504 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3508 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3511 kvm->arch.ept_identity_map_addr = ident_addr;
3515 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3516 u32 kvm_nr_mmu_pages)
3518 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3521 mutex_lock(&kvm->slots_lock);
3523 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3524 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3526 mutex_unlock(&kvm->slots_lock);
3530 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3532 return kvm->arch.n_max_mmu_pages;
3535 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3540 switch (chip->chip_id) {
3541 case KVM_IRQCHIP_PIC_MASTER:
3542 memcpy(&chip->chip.pic,
3543 &pic_irqchip(kvm)->pics[0],
3544 sizeof(struct kvm_pic_state));
3546 case KVM_IRQCHIP_PIC_SLAVE:
3547 memcpy(&chip->chip.pic,
3548 &pic_irqchip(kvm)->pics[1],
3549 sizeof(struct kvm_pic_state));
3551 case KVM_IRQCHIP_IOAPIC:
3552 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3561 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3566 switch (chip->chip_id) {
3567 case KVM_IRQCHIP_PIC_MASTER:
3568 spin_lock(&pic_irqchip(kvm)->lock);
3569 memcpy(&pic_irqchip(kvm)->pics[0],
3571 sizeof(struct kvm_pic_state));
3572 spin_unlock(&pic_irqchip(kvm)->lock);
3574 case KVM_IRQCHIP_PIC_SLAVE:
3575 spin_lock(&pic_irqchip(kvm)->lock);
3576 memcpy(&pic_irqchip(kvm)->pics[1],
3578 sizeof(struct kvm_pic_state));
3579 spin_unlock(&pic_irqchip(kvm)->lock);
3581 case KVM_IRQCHIP_IOAPIC:
3582 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3588 kvm_pic_update_irq(pic_irqchip(kvm));
3592 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3594 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3595 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3596 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3600 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3602 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3603 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3604 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3605 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3609 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3611 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3612 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3613 sizeof(ps->channels));
3614 ps->flags = kvm->arch.vpit->pit_state.flags;
3615 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3616 memset(&ps->reserved, 0, sizeof(ps->reserved));
3620 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3623 u32 prev_legacy, cur_legacy;
3624 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3625 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3626 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3627 if (!prev_legacy && cur_legacy)
3629 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3630 sizeof(kvm->arch.vpit->pit_state.channels));
3631 kvm->arch.vpit->pit_state.flags = ps->flags;
3632 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3633 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3637 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3638 struct kvm_reinject_control *control)
3640 if (!kvm->arch.vpit)
3642 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3643 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3644 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3649 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3650 * @kvm: kvm instance
3651 * @log: slot id and address to which we copy the log
3653 * Steps 1-4 below provide general overview of dirty page logging. See
3654 * kvm_get_dirty_log_protect() function description for additional details.
3656 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3657 * always flush the TLB (step 4) even if previous step failed and the dirty
3658 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3659 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3660 * writes will be marked dirty for next log read.
3662 * 1. Take a snapshot of the bit and clear it if needed.
3663 * 2. Write protect the corresponding page.
3664 * 3. Copy the snapshot to the userspace.
3665 * 4. Flush TLB's if needed.
3667 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3669 bool is_dirty = false;
3672 mutex_lock(&kvm->slots_lock);
3675 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3677 if (kvm_x86_ops->flush_log_dirty)
3678 kvm_x86_ops->flush_log_dirty(kvm);
3680 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3683 * All the TLBs can be flushed out of mmu lock, see the comments in
3684 * kvm_mmu_slot_remove_write_access().
3686 lockdep_assert_held(&kvm->slots_lock);
3688 kvm_flush_remote_tlbs(kvm);
3690 mutex_unlock(&kvm->slots_lock);
3694 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3697 if (!irqchip_in_kernel(kvm))
3700 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3701 irq_event->irq, irq_event->level,
3706 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3707 struct kvm_enable_cap *cap)
3715 case KVM_CAP_DISABLE_QUIRKS:
3716 kvm->arch.disabled_quirks = cap->args[0];
3719 case KVM_CAP_SPLIT_IRQCHIP: {
3720 mutex_lock(&kvm->lock);
3722 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3723 goto split_irqchip_unlock;
3725 if (irqchip_in_kernel(kvm))
3726 goto split_irqchip_unlock;
3727 if (atomic_read(&kvm->online_vcpus))
3728 goto split_irqchip_unlock;
3729 r = kvm_setup_empty_irq_routing(kvm);
3731 goto split_irqchip_unlock;
3732 /* Pairs with irqchip_in_kernel. */
3734 kvm->arch.irqchip_split = true;
3735 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3737 split_irqchip_unlock:
3738 mutex_unlock(&kvm->lock);
3748 long kvm_arch_vm_ioctl(struct file *filp,
3749 unsigned int ioctl, unsigned long arg)
3751 struct kvm *kvm = filp->private_data;
3752 void __user *argp = (void __user *)arg;
3755 * This union makes it completely explicit to gcc-3.x
3756 * that these two variables' stack usage should be
3757 * combined, not added together.
3760 struct kvm_pit_state ps;
3761 struct kvm_pit_state2 ps2;
3762 struct kvm_pit_config pit_config;
3766 case KVM_SET_TSS_ADDR:
3767 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3769 case KVM_SET_IDENTITY_MAP_ADDR: {
3773 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3775 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3778 case KVM_SET_NR_MMU_PAGES:
3779 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3781 case KVM_GET_NR_MMU_PAGES:
3782 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3784 case KVM_CREATE_IRQCHIP: {
3785 struct kvm_pic *vpic;
3787 mutex_lock(&kvm->lock);
3790 goto create_irqchip_unlock;
3792 if (atomic_read(&kvm->online_vcpus))
3793 goto create_irqchip_unlock;
3795 vpic = kvm_create_pic(kvm);
3797 r = kvm_ioapic_init(kvm);
3799 mutex_lock(&kvm->slots_lock);
3800 kvm_destroy_pic(vpic);
3801 mutex_unlock(&kvm->slots_lock);
3802 goto create_irqchip_unlock;
3805 goto create_irqchip_unlock;
3806 r = kvm_setup_default_irq_routing(kvm);
3808 mutex_lock(&kvm->slots_lock);
3809 mutex_lock(&kvm->irq_lock);
3810 kvm_ioapic_destroy(kvm);
3811 kvm_destroy_pic(vpic);
3812 mutex_unlock(&kvm->irq_lock);
3813 mutex_unlock(&kvm->slots_lock);
3814 goto create_irqchip_unlock;
3816 /* Write kvm->irq_routing before kvm->arch.vpic. */
3818 kvm->arch.vpic = vpic;
3819 create_irqchip_unlock:
3820 mutex_unlock(&kvm->lock);
3823 case KVM_CREATE_PIT:
3824 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3826 case KVM_CREATE_PIT2:
3828 if (copy_from_user(&u.pit_config, argp,
3829 sizeof(struct kvm_pit_config)))
3832 mutex_lock(&kvm->slots_lock);
3835 goto create_pit_unlock;
3837 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3841 mutex_unlock(&kvm->slots_lock);
3843 case KVM_GET_IRQCHIP: {
3844 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3845 struct kvm_irqchip *chip;
3847 chip = memdup_user(argp, sizeof(*chip));
3854 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3855 goto get_irqchip_out;
3856 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3858 goto get_irqchip_out;
3860 if (copy_to_user(argp, chip, sizeof *chip))
3861 goto get_irqchip_out;
3867 case KVM_SET_IRQCHIP: {
3868 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3869 struct kvm_irqchip *chip;
3871 chip = memdup_user(argp, sizeof(*chip));
3878 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3879 goto set_irqchip_out;
3880 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3882 goto set_irqchip_out;
3890 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3893 if (!kvm->arch.vpit)
3895 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3899 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3906 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3909 if (!kvm->arch.vpit)
3911 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3914 case KVM_GET_PIT2: {
3916 if (!kvm->arch.vpit)
3918 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3922 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3927 case KVM_SET_PIT2: {
3929 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3932 if (!kvm->arch.vpit)
3934 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3937 case KVM_REINJECT_CONTROL: {
3938 struct kvm_reinject_control control;
3940 if (copy_from_user(&control, argp, sizeof(control)))
3942 r = kvm_vm_ioctl_reinject(kvm, &control);
3945 case KVM_SET_BOOT_CPU_ID:
3947 mutex_lock(&kvm->lock);
3948 if (atomic_read(&kvm->online_vcpus) != 0)
3951 kvm->arch.bsp_vcpu_id = arg;
3952 mutex_unlock(&kvm->lock);
3954 case KVM_XEN_HVM_CONFIG: {
3956 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3957 sizeof(struct kvm_xen_hvm_config)))
3960 if (kvm->arch.xen_hvm_config.flags)
3965 case KVM_SET_CLOCK: {
3966 struct kvm_clock_data user_ns;
3971 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3979 local_irq_disable();
3980 now_ns = get_kernel_ns();
3981 delta = user_ns.clock - now_ns;
3983 kvm->arch.kvmclock_offset = delta;
3984 kvm_gen_update_masterclock(kvm);
3987 case KVM_GET_CLOCK: {
3988 struct kvm_clock_data user_ns;
3991 local_irq_disable();
3992 now_ns = get_kernel_ns();
3993 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3996 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3999 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4004 case KVM_ENABLE_CAP: {
4005 struct kvm_enable_cap cap;
4008 if (copy_from_user(&cap, argp, sizeof(cap)))
4010 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4014 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4020 static void kvm_init_msr_list(void)
4025 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4026 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4030 * Even MSRs that are valid in the host may not be exposed
4031 * to the guests in some cases. We could work around this
4032 * in VMX with the generic MSR save/load machinery, but it
4033 * is not really worthwhile since it will really only
4034 * happen with nested virtualization.
4036 switch (msrs_to_save[i]) {
4037 case MSR_IA32_BNDCFGS:
4038 if (!kvm_x86_ops->mpx_supported())
4046 msrs_to_save[j] = msrs_to_save[i];
4049 num_msrs_to_save = j;
4051 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4052 switch (emulated_msrs[i]) {
4053 case MSR_IA32_SMBASE:
4054 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4062 emulated_msrs[j] = emulated_msrs[i];
4065 num_emulated_msrs = j;
4068 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4076 if (!(vcpu->arch.apic &&
4077 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4078 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4089 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4096 if (!(vcpu->arch.apic &&
4097 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4099 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4101 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4111 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4112 struct kvm_segment *var, int seg)
4114 kvm_x86_ops->set_segment(vcpu, var, seg);
4117 void kvm_get_segment(struct kvm_vcpu *vcpu,
4118 struct kvm_segment *var, int seg)
4120 kvm_x86_ops->get_segment(vcpu, var, seg);
4123 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4124 struct x86_exception *exception)
4128 BUG_ON(!mmu_is_nested(vcpu));
4130 /* NPT walks are always user-walks */
4131 access |= PFERR_USER_MASK;
4132 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4137 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4138 struct x86_exception *exception)
4140 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4141 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4144 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4145 struct x86_exception *exception)
4147 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4148 access |= PFERR_FETCH_MASK;
4149 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4152 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4153 struct x86_exception *exception)
4155 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4156 access |= PFERR_WRITE_MASK;
4157 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4160 /* uses this to access any guest's mapped memory without checking CPL */
4161 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4162 struct x86_exception *exception)
4164 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4167 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4168 struct kvm_vcpu *vcpu, u32 access,
4169 struct x86_exception *exception)
4172 int r = X86EMUL_CONTINUE;
4175 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4177 unsigned offset = addr & (PAGE_SIZE-1);
4178 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4181 if (gpa == UNMAPPED_GVA)
4182 return X86EMUL_PROPAGATE_FAULT;
4183 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4186 r = X86EMUL_IO_NEEDED;
4198 /* used for instruction fetching */
4199 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4200 gva_t addr, void *val, unsigned int bytes,
4201 struct x86_exception *exception)
4203 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4204 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4208 /* Inline kvm_read_guest_virt_helper for speed. */
4209 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4211 if (unlikely(gpa == UNMAPPED_GVA))
4212 return X86EMUL_PROPAGATE_FAULT;
4214 offset = addr & (PAGE_SIZE-1);
4215 if (WARN_ON(offset + bytes > PAGE_SIZE))
4216 bytes = (unsigned)PAGE_SIZE - offset;
4217 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4219 if (unlikely(ret < 0))
4220 return X86EMUL_IO_NEEDED;
4222 return X86EMUL_CONTINUE;
4225 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4226 gva_t addr, void *val, unsigned int bytes,
4227 struct x86_exception *exception)
4229 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4230 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4232 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4235 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4237 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4238 gva_t addr, void *val, unsigned int bytes,
4239 struct x86_exception *exception)
4241 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4242 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4245 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4246 unsigned long addr, void *val, unsigned int bytes)
4248 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4249 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4251 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4254 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4255 gva_t addr, void *val,
4257 struct x86_exception *exception)
4259 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4261 int r = X86EMUL_CONTINUE;
4264 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4267 unsigned offset = addr & (PAGE_SIZE-1);
4268 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4271 if (gpa == UNMAPPED_GVA)
4272 return X86EMUL_PROPAGATE_FAULT;
4273 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4275 r = X86EMUL_IO_NEEDED;
4286 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4288 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4289 gpa_t *gpa, struct x86_exception *exception,
4292 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4293 | (write ? PFERR_WRITE_MASK : 0);
4295 if (vcpu_match_mmio_gva(vcpu, gva)
4296 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4297 vcpu->arch.access, access)) {
4298 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4299 (gva & (PAGE_SIZE - 1));
4300 trace_vcpu_match_mmio(gva, *gpa, write, false);
4304 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4306 if (*gpa == UNMAPPED_GVA)
4309 /* For APIC access vmexit */
4310 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4313 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4314 trace_vcpu_match_mmio(gva, *gpa, write, true);
4321 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4322 const void *val, int bytes)
4326 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4329 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4333 struct read_write_emulator_ops {
4334 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4336 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4337 void *val, int bytes);
4338 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4339 int bytes, void *val);
4340 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4341 void *val, int bytes);
4345 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4347 if (vcpu->mmio_read_completed) {
4348 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4349 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4350 vcpu->mmio_read_completed = 0;
4357 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4358 void *val, int bytes)
4360 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4363 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4364 void *val, int bytes)
4366 return emulator_write_phys(vcpu, gpa, val, bytes);
4369 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4371 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4372 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4375 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4376 void *val, int bytes)
4378 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4379 return X86EMUL_IO_NEEDED;
4382 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4383 void *val, int bytes)
4385 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4387 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4388 return X86EMUL_CONTINUE;
4391 static const struct read_write_emulator_ops read_emultor = {
4392 .read_write_prepare = read_prepare,
4393 .read_write_emulate = read_emulate,
4394 .read_write_mmio = vcpu_mmio_read,
4395 .read_write_exit_mmio = read_exit_mmio,
4398 static const struct read_write_emulator_ops write_emultor = {
4399 .read_write_emulate = write_emulate,
4400 .read_write_mmio = write_mmio,
4401 .read_write_exit_mmio = write_exit_mmio,
4405 static int emulator_read_write_onepage(unsigned long addr, void *val,
4407 struct x86_exception *exception,
4408 struct kvm_vcpu *vcpu,
4409 const struct read_write_emulator_ops *ops)
4413 bool write = ops->write;
4414 struct kvm_mmio_fragment *frag;
4416 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4419 return X86EMUL_PROPAGATE_FAULT;
4421 /* For APIC access vmexit */
4425 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4426 return X86EMUL_CONTINUE;
4430 * Is this MMIO handled locally?
4432 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4433 if (handled == bytes)
4434 return X86EMUL_CONTINUE;
4440 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4441 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4445 return X86EMUL_CONTINUE;
4448 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4450 void *val, unsigned int bytes,
4451 struct x86_exception *exception,
4452 const struct read_write_emulator_ops *ops)
4454 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4458 if (ops->read_write_prepare &&
4459 ops->read_write_prepare(vcpu, val, bytes))
4460 return X86EMUL_CONTINUE;
4462 vcpu->mmio_nr_fragments = 0;
4464 /* Crossing a page boundary? */
4465 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4468 now = -addr & ~PAGE_MASK;
4469 rc = emulator_read_write_onepage(addr, val, now, exception,
4472 if (rc != X86EMUL_CONTINUE)
4475 if (ctxt->mode != X86EMUL_MODE_PROT64)
4481 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4483 if (rc != X86EMUL_CONTINUE)
4486 if (!vcpu->mmio_nr_fragments)
4489 gpa = vcpu->mmio_fragments[0].gpa;
4491 vcpu->mmio_needed = 1;
4492 vcpu->mmio_cur_fragment = 0;
4494 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4495 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4496 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4497 vcpu->run->mmio.phys_addr = gpa;
4499 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4502 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4506 struct x86_exception *exception)
4508 return emulator_read_write(ctxt, addr, val, bytes,
4509 exception, &read_emultor);
4512 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4516 struct x86_exception *exception)
4518 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4519 exception, &write_emultor);
4522 #define CMPXCHG_TYPE(t, ptr, old, new) \
4523 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4525 #ifdef CONFIG_X86_64
4526 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4528 # define CMPXCHG64(ptr, old, new) \
4529 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4532 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4537 struct x86_exception *exception)
4539 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4545 /* guests cmpxchg8b have to be emulated atomically */
4546 if (bytes > 8 || (bytes & (bytes - 1)))
4549 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4551 if (gpa == UNMAPPED_GVA ||
4552 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4555 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4558 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4559 if (is_error_page(page))
4562 kaddr = kmap_atomic(page);
4563 kaddr += offset_in_page(gpa);
4566 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4569 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4572 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4575 exchanged = CMPXCHG64(kaddr, old, new);
4580 kunmap_atomic(kaddr);
4581 kvm_release_page_dirty(page);
4584 return X86EMUL_CMPXCHG_FAILED;
4586 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4587 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4589 return X86EMUL_CONTINUE;
4592 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4594 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4597 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4599 /* TODO: String I/O for in kernel device */
4602 if (vcpu->arch.pio.in)
4603 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4604 vcpu->arch.pio.size, pd);
4606 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4607 vcpu->arch.pio.port, vcpu->arch.pio.size,
4612 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4613 unsigned short port, void *val,
4614 unsigned int count, bool in)
4616 vcpu->arch.pio.port = port;
4617 vcpu->arch.pio.in = in;
4618 vcpu->arch.pio.count = count;
4619 vcpu->arch.pio.size = size;
4621 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4622 vcpu->arch.pio.count = 0;
4626 vcpu->run->exit_reason = KVM_EXIT_IO;
4627 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4628 vcpu->run->io.size = size;
4629 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4630 vcpu->run->io.count = count;
4631 vcpu->run->io.port = port;
4636 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4637 int size, unsigned short port, void *val,
4640 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4643 if (vcpu->arch.pio.count)
4646 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4649 memcpy(val, vcpu->arch.pio_data, size * count);
4650 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4651 vcpu->arch.pio.count = 0;
4658 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4659 int size, unsigned short port,
4660 const void *val, unsigned int count)
4662 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4664 memcpy(vcpu->arch.pio_data, val, size * count);
4665 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4666 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4669 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4671 return kvm_x86_ops->get_segment_base(vcpu, seg);
4674 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4676 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4679 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4681 if (!need_emulate_wbinvd(vcpu))
4682 return X86EMUL_CONTINUE;
4684 if (kvm_x86_ops->has_wbinvd_exit()) {
4685 int cpu = get_cpu();
4687 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4688 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4689 wbinvd_ipi, NULL, 1);
4691 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4694 return X86EMUL_CONTINUE;
4697 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4699 kvm_x86_ops->skip_emulated_instruction(vcpu);
4700 return kvm_emulate_wbinvd_noskip(vcpu);
4702 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4706 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4708 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4711 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4712 unsigned long *dest)
4714 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4717 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4718 unsigned long value)
4721 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4724 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4726 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4729 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4731 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4732 unsigned long value;
4736 value = kvm_read_cr0(vcpu);
4739 value = vcpu->arch.cr2;
4742 value = kvm_read_cr3(vcpu);
4745 value = kvm_read_cr4(vcpu);
4748 value = kvm_get_cr8(vcpu);
4751 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4758 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4760 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4765 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4768 vcpu->arch.cr2 = val;
4771 res = kvm_set_cr3(vcpu, val);
4774 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4777 res = kvm_set_cr8(vcpu, val);
4780 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4787 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4789 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4792 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4794 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4797 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4799 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4802 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4804 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4807 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4809 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4812 static unsigned long emulator_get_cached_segment_base(
4813 struct x86_emulate_ctxt *ctxt, int seg)
4815 return get_segment_base(emul_to_vcpu(ctxt), seg);
4818 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4819 struct desc_struct *desc, u32 *base3,
4822 struct kvm_segment var;
4824 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4825 *selector = var.selector;
4828 memset(desc, 0, sizeof(*desc));
4834 set_desc_limit(desc, var.limit);
4835 set_desc_base(desc, (unsigned long)var.base);
4836 #ifdef CONFIG_X86_64
4838 *base3 = var.base >> 32;
4840 desc->type = var.type;
4842 desc->dpl = var.dpl;
4843 desc->p = var.present;
4844 desc->avl = var.avl;
4852 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4853 struct desc_struct *desc, u32 base3,
4856 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4857 struct kvm_segment var;
4859 var.selector = selector;
4860 var.base = get_desc_base(desc);
4861 #ifdef CONFIG_X86_64
4862 var.base |= ((u64)base3) << 32;
4864 var.limit = get_desc_limit(desc);
4866 var.limit = (var.limit << 12) | 0xfff;
4867 var.type = desc->type;
4868 var.dpl = desc->dpl;
4873 var.avl = desc->avl;
4874 var.present = desc->p;
4875 var.unusable = !var.present;
4878 kvm_set_segment(vcpu, &var, seg);
4882 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4883 u32 msr_index, u64 *pdata)
4885 struct msr_data msr;
4888 msr.index = msr_index;
4889 msr.host_initiated = false;
4890 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4898 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4899 u32 msr_index, u64 data)
4901 struct msr_data msr;
4904 msr.index = msr_index;
4905 msr.host_initiated = false;
4906 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4909 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4911 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4913 return vcpu->arch.smbase;
4916 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4918 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4920 vcpu->arch.smbase = smbase;
4923 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4926 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4929 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4930 u32 pmc, u64 *pdata)
4932 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4935 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4937 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4940 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4943 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4945 * CR0.TS may reference the host fpu state, not the guest fpu state,
4946 * so it may be clear at this point.
4951 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4956 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4957 struct x86_instruction_info *info,
4958 enum x86_intercept_stage stage)
4960 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4963 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4964 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4966 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4969 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4971 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4974 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4976 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4979 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
4981 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
4984 static const struct x86_emulate_ops emulate_ops = {
4985 .read_gpr = emulator_read_gpr,
4986 .write_gpr = emulator_write_gpr,
4987 .read_std = kvm_read_guest_virt_system,
4988 .write_std = kvm_write_guest_virt_system,
4989 .read_phys = kvm_read_guest_phys_system,
4990 .fetch = kvm_fetch_guest_virt,
4991 .read_emulated = emulator_read_emulated,
4992 .write_emulated = emulator_write_emulated,
4993 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4994 .invlpg = emulator_invlpg,
4995 .pio_in_emulated = emulator_pio_in_emulated,
4996 .pio_out_emulated = emulator_pio_out_emulated,
4997 .get_segment = emulator_get_segment,
4998 .set_segment = emulator_set_segment,
4999 .get_cached_segment_base = emulator_get_cached_segment_base,
5000 .get_gdt = emulator_get_gdt,
5001 .get_idt = emulator_get_idt,
5002 .set_gdt = emulator_set_gdt,
5003 .set_idt = emulator_set_idt,
5004 .get_cr = emulator_get_cr,
5005 .set_cr = emulator_set_cr,
5006 .cpl = emulator_get_cpl,
5007 .get_dr = emulator_get_dr,
5008 .set_dr = emulator_set_dr,
5009 .get_smbase = emulator_get_smbase,
5010 .set_smbase = emulator_set_smbase,
5011 .set_msr = emulator_set_msr,
5012 .get_msr = emulator_get_msr,
5013 .check_pmc = emulator_check_pmc,
5014 .read_pmc = emulator_read_pmc,
5015 .halt = emulator_halt,
5016 .wbinvd = emulator_wbinvd,
5017 .fix_hypercall = emulator_fix_hypercall,
5018 .get_fpu = emulator_get_fpu,
5019 .put_fpu = emulator_put_fpu,
5020 .intercept = emulator_intercept,
5021 .get_cpuid = emulator_get_cpuid,
5022 .set_nmi_mask = emulator_set_nmi_mask,
5025 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5027 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5029 * an sti; sti; sequence only disable interrupts for the first
5030 * instruction. So, if the last instruction, be it emulated or
5031 * not, left the system with the INT_STI flag enabled, it
5032 * means that the last instruction is an sti. We should not
5033 * leave the flag on in this case. The same goes for mov ss
5035 if (int_shadow & mask)
5037 if (unlikely(int_shadow || mask)) {
5038 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5040 kvm_make_request(KVM_REQ_EVENT, vcpu);
5044 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5046 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5047 if (ctxt->exception.vector == PF_VECTOR)
5048 return kvm_propagate_fault(vcpu, &ctxt->exception);
5050 if (ctxt->exception.error_code_valid)
5051 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5052 ctxt->exception.error_code);
5054 kvm_queue_exception(vcpu, ctxt->exception.vector);
5058 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5060 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5063 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5065 ctxt->eflags = kvm_get_rflags(vcpu);
5066 ctxt->eip = kvm_rip_read(vcpu);
5067 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5068 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5069 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5070 cs_db ? X86EMUL_MODE_PROT32 :
5071 X86EMUL_MODE_PROT16;
5072 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5073 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5074 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5075 ctxt->emul_flags = vcpu->arch.hflags;
5077 init_decode_cache(ctxt);
5078 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5081 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5083 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5086 init_emulate_ctxt(vcpu);
5090 ctxt->_eip = ctxt->eip + inc_eip;
5091 ret = emulate_int_real(ctxt, irq);
5093 if (ret != X86EMUL_CONTINUE)
5094 return EMULATE_FAIL;
5096 ctxt->eip = ctxt->_eip;
5097 kvm_rip_write(vcpu, ctxt->eip);
5098 kvm_set_rflags(vcpu, ctxt->eflags);
5100 if (irq == NMI_VECTOR)
5101 vcpu->arch.nmi_pending = 0;
5103 vcpu->arch.interrupt.pending = false;
5105 return EMULATE_DONE;
5107 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5109 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5111 int r = EMULATE_DONE;
5113 ++vcpu->stat.insn_emulation_fail;
5114 trace_kvm_emulate_insn_failed(vcpu);
5115 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5116 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5117 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5118 vcpu->run->internal.ndata = 0;
5121 kvm_queue_exception(vcpu, UD_VECTOR);
5126 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5127 bool write_fault_to_shadow_pgtable,
5133 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5136 if (!vcpu->arch.mmu.direct_map) {
5138 * Write permission should be allowed since only
5139 * write access need to be emulated.
5141 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5144 * If the mapping is invalid in guest, let cpu retry
5145 * it to generate fault.
5147 if (gpa == UNMAPPED_GVA)
5152 * Do not retry the unhandleable instruction if it faults on the
5153 * readonly host memory, otherwise it will goto a infinite loop:
5154 * retry instruction -> write #PF -> emulation fail -> retry
5155 * instruction -> ...
5157 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5160 * If the instruction failed on the error pfn, it can not be fixed,
5161 * report the error to userspace.
5163 if (is_error_noslot_pfn(pfn))
5166 kvm_release_pfn_clean(pfn);
5168 /* The instructions are well-emulated on direct mmu. */
5169 if (vcpu->arch.mmu.direct_map) {
5170 unsigned int indirect_shadow_pages;
5172 spin_lock(&vcpu->kvm->mmu_lock);
5173 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5174 spin_unlock(&vcpu->kvm->mmu_lock);
5176 if (indirect_shadow_pages)
5177 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5183 * if emulation was due to access to shadowed page table
5184 * and it failed try to unshadow page and re-enter the
5185 * guest to let CPU execute the instruction.
5187 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5190 * If the access faults on its page table, it can not
5191 * be fixed by unprotecting shadow page and it should
5192 * be reported to userspace.
5194 return !write_fault_to_shadow_pgtable;
5197 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5198 unsigned long cr2, int emulation_type)
5200 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5201 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5203 last_retry_eip = vcpu->arch.last_retry_eip;
5204 last_retry_addr = vcpu->arch.last_retry_addr;
5207 * If the emulation is caused by #PF and it is non-page_table
5208 * writing instruction, it means the VM-EXIT is caused by shadow
5209 * page protected, we can zap the shadow page and retry this
5210 * instruction directly.
5212 * Note: if the guest uses a non-page-table modifying instruction
5213 * on the PDE that points to the instruction, then we will unmap
5214 * the instruction and go to an infinite loop. So, we cache the
5215 * last retried eip and the last fault address, if we meet the eip
5216 * and the address again, we can break out of the potential infinite
5219 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5221 if (!(emulation_type & EMULTYPE_RETRY))
5224 if (x86_page_table_writing_insn(ctxt))
5227 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5230 vcpu->arch.last_retry_eip = ctxt->eip;
5231 vcpu->arch.last_retry_addr = cr2;
5233 if (!vcpu->arch.mmu.direct_map)
5234 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5236 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5241 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5242 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5244 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5246 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5247 /* This is a good place to trace that we are exiting SMM. */
5248 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5250 if (unlikely(vcpu->arch.smi_pending)) {
5251 kvm_make_request(KVM_REQ_SMI, vcpu);
5252 vcpu->arch.smi_pending = 0;
5254 /* Process a latched INIT, if any. */
5255 kvm_make_request(KVM_REQ_EVENT, vcpu);
5259 kvm_mmu_reset_context(vcpu);
5262 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5264 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5266 vcpu->arch.hflags = emul_flags;
5268 if (changed & HF_SMM_MASK)
5269 kvm_smm_changed(vcpu);
5272 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5281 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5282 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5287 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5289 struct kvm_run *kvm_run = vcpu->run;
5292 * rflags is the old, "raw" value of the flags. The new value has
5293 * not been saved yet.
5295 * This is correct even for TF set by the guest, because "the
5296 * processor will not generate this exception after the instruction
5297 * that sets the TF flag".
5299 if (unlikely(rflags & X86_EFLAGS_TF)) {
5300 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5301 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5303 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5304 kvm_run->debug.arch.exception = DB_VECTOR;
5305 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5306 *r = EMULATE_USER_EXIT;
5308 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5310 * "Certain debug exceptions may clear bit 0-3. The
5311 * remaining contents of the DR6 register are never
5312 * cleared by the processor".
5314 vcpu->arch.dr6 &= ~15;
5315 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5316 kvm_queue_exception(vcpu, DB_VECTOR);
5321 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5323 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5324 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5325 struct kvm_run *kvm_run = vcpu->run;
5326 unsigned long eip = kvm_get_linear_rip(vcpu);
5327 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5328 vcpu->arch.guest_debug_dr7,
5332 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5333 kvm_run->debug.arch.pc = eip;
5334 kvm_run->debug.arch.exception = DB_VECTOR;
5335 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5336 *r = EMULATE_USER_EXIT;
5341 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5342 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5343 unsigned long eip = kvm_get_linear_rip(vcpu);
5344 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5349 vcpu->arch.dr6 &= ~15;
5350 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5351 kvm_queue_exception(vcpu, DB_VECTOR);
5360 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5367 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5368 bool writeback = true;
5369 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5372 * Clear write_fault_to_shadow_pgtable here to ensure it is
5375 vcpu->arch.write_fault_to_shadow_pgtable = false;
5376 kvm_clear_exception_queue(vcpu);
5378 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5379 init_emulate_ctxt(vcpu);
5382 * We will reenter on the same instruction since
5383 * we do not set complete_userspace_io. This does not
5384 * handle watchpoints yet, those would be handled in
5387 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5390 ctxt->interruptibility = 0;
5391 ctxt->have_exception = false;
5392 ctxt->exception.vector = -1;
5393 ctxt->perm_ok = false;
5395 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5397 r = x86_decode_insn(ctxt, insn, insn_len);
5399 trace_kvm_emulate_insn_start(vcpu);
5400 ++vcpu->stat.insn_emulation;
5401 if (r != EMULATION_OK) {
5402 if (emulation_type & EMULTYPE_TRAP_UD)
5403 return EMULATE_FAIL;
5404 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5406 return EMULATE_DONE;
5407 if (emulation_type & EMULTYPE_SKIP)
5408 return EMULATE_FAIL;
5409 return handle_emulation_failure(vcpu);
5413 if (emulation_type & EMULTYPE_SKIP) {
5414 kvm_rip_write(vcpu, ctxt->_eip);
5415 if (ctxt->eflags & X86_EFLAGS_RF)
5416 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5417 return EMULATE_DONE;
5420 if (retry_instruction(ctxt, cr2, emulation_type))
5421 return EMULATE_DONE;
5423 /* this is needed for vmware backdoor interface to work since it
5424 changes registers values during IO operation */
5425 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5426 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5427 emulator_invalidate_register_cache(ctxt);
5431 r = x86_emulate_insn(ctxt);
5433 if (r == EMULATION_INTERCEPTED)
5434 return EMULATE_DONE;
5436 if (r == EMULATION_FAILED) {
5437 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5439 return EMULATE_DONE;
5441 return handle_emulation_failure(vcpu);
5444 if (ctxt->have_exception) {
5446 if (inject_emulated_exception(vcpu))
5448 } else if (vcpu->arch.pio.count) {
5449 if (!vcpu->arch.pio.in) {
5450 /* FIXME: return into emulator if single-stepping. */
5451 vcpu->arch.pio.count = 0;
5454 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5456 r = EMULATE_USER_EXIT;
5457 } else if (vcpu->mmio_needed) {
5458 if (!vcpu->mmio_is_write)
5460 r = EMULATE_USER_EXIT;
5461 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5462 } else if (r == EMULATION_RESTART)
5468 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5469 toggle_interruptibility(vcpu, ctxt->interruptibility);
5470 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5471 if (vcpu->arch.hflags != ctxt->emul_flags)
5472 kvm_set_hflags(vcpu, ctxt->emul_flags);
5473 kvm_rip_write(vcpu, ctxt->eip);
5474 if (r == EMULATE_DONE)
5475 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5476 if (!ctxt->have_exception ||
5477 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5478 __kvm_set_rflags(vcpu, ctxt->eflags);
5481 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5482 * do nothing, and it will be requested again as soon as
5483 * the shadow expires. But we still need to check here,
5484 * because POPF has no interrupt shadow.
5486 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5487 kvm_make_request(KVM_REQ_EVENT, vcpu);
5489 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5493 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5495 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5497 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5498 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5499 size, port, &val, 1);
5500 /* do not return to emulator after return from userspace */
5501 vcpu->arch.pio.count = 0;
5504 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5506 static void tsc_bad(void *info)
5508 __this_cpu_write(cpu_tsc_khz, 0);
5511 static void tsc_khz_changed(void *data)
5513 struct cpufreq_freqs *freq = data;
5514 unsigned long khz = 0;
5518 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5519 khz = cpufreq_quick_get(raw_smp_processor_id());
5522 __this_cpu_write(cpu_tsc_khz, khz);
5525 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5528 struct cpufreq_freqs *freq = data;
5530 struct kvm_vcpu *vcpu;
5531 int i, send_ipi = 0;
5534 * We allow guests to temporarily run on slowing clocks,
5535 * provided we notify them after, or to run on accelerating
5536 * clocks, provided we notify them before. Thus time never
5539 * However, we have a problem. We can't atomically update
5540 * the frequency of a given CPU from this function; it is
5541 * merely a notifier, which can be called from any CPU.
5542 * Changing the TSC frequency at arbitrary points in time
5543 * requires a recomputation of local variables related to
5544 * the TSC for each VCPU. We must flag these local variables
5545 * to be updated and be sure the update takes place with the
5546 * new frequency before any guests proceed.
5548 * Unfortunately, the combination of hotplug CPU and frequency
5549 * change creates an intractable locking scenario; the order
5550 * of when these callouts happen is undefined with respect to
5551 * CPU hotplug, and they can race with each other. As such,
5552 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5553 * undefined; you can actually have a CPU frequency change take
5554 * place in between the computation of X and the setting of the
5555 * variable. To protect against this problem, all updates of
5556 * the per_cpu tsc_khz variable are done in an interrupt
5557 * protected IPI, and all callers wishing to update the value
5558 * must wait for a synchronous IPI to complete (which is trivial
5559 * if the caller is on the CPU already). This establishes the
5560 * necessary total order on variable updates.
5562 * Note that because a guest time update may take place
5563 * anytime after the setting of the VCPU's request bit, the
5564 * correct TSC value must be set before the request. However,
5565 * to ensure the update actually makes it to any guest which
5566 * starts running in hardware virtualization between the set
5567 * and the acquisition of the spinlock, we must also ping the
5568 * CPU after setting the request bit.
5572 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5574 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5577 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5579 spin_lock(&kvm_lock);
5580 list_for_each_entry(kvm, &vm_list, vm_list) {
5581 kvm_for_each_vcpu(i, vcpu, kvm) {
5582 if (vcpu->cpu != freq->cpu)
5584 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5585 if (vcpu->cpu != smp_processor_id())
5589 spin_unlock(&kvm_lock);
5591 if (freq->old < freq->new && send_ipi) {
5593 * We upscale the frequency. Must make the guest
5594 * doesn't see old kvmclock values while running with
5595 * the new frequency, otherwise we risk the guest sees
5596 * time go backwards.
5598 * In case we update the frequency for another cpu
5599 * (which might be in guest context) send an interrupt
5600 * to kick the cpu out of guest context. Next time
5601 * guest context is entered kvmclock will be updated,
5602 * so the guest will not see stale values.
5604 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5609 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5610 .notifier_call = kvmclock_cpufreq_notifier
5613 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5614 unsigned long action, void *hcpu)
5616 unsigned int cpu = (unsigned long)hcpu;
5620 case CPU_DOWN_FAILED:
5621 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5623 case CPU_DOWN_PREPARE:
5624 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5630 static struct notifier_block kvmclock_cpu_notifier_block = {
5631 .notifier_call = kvmclock_cpu_notifier,
5632 .priority = -INT_MAX
5635 static void kvm_timer_init(void)
5639 max_tsc_khz = tsc_khz;
5641 cpu_notifier_register_begin();
5642 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5643 #ifdef CONFIG_CPU_FREQ
5644 struct cpufreq_policy policy;
5645 memset(&policy, 0, sizeof(policy));
5647 cpufreq_get_policy(&policy, cpu);
5648 if (policy.cpuinfo.max_freq)
5649 max_tsc_khz = policy.cpuinfo.max_freq;
5652 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5653 CPUFREQ_TRANSITION_NOTIFIER);
5655 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5656 for_each_online_cpu(cpu)
5657 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5659 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5660 cpu_notifier_register_done();
5664 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5666 int kvm_is_in_guest(void)
5668 return __this_cpu_read(current_vcpu) != NULL;
5671 static int kvm_is_user_mode(void)
5675 if (__this_cpu_read(current_vcpu))
5676 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5678 return user_mode != 0;
5681 static unsigned long kvm_get_guest_ip(void)
5683 unsigned long ip = 0;
5685 if (__this_cpu_read(current_vcpu))
5686 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5691 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5692 .is_in_guest = kvm_is_in_guest,
5693 .is_user_mode = kvm_is_user_mode,
5694 .get_guest_ip = kvm_get_guest_ip,
5697 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5699 __this_cpu_write(current_vcpu, vcpu);
5701 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5703 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5705 __this_cpu_write(current_vcpu, NULL);
5707 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5709 static void kvm_set_mmio_spte_mask(void)
5712 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5715 * Set the reserved bits and the present bit of an paging-structure
5716 * entry to generate page fault with PFER.RSV = 1.
5718 /* Mask the reserved physical address bits. */
5719 mask = rsvd_bits(maxphyaddr, 51);
5721 /* Bit 62 is always reserved for 32bit host. */
5722 mask |= 0x3ull << 62;
5724 /* Set the present bit. */
5727 #ifdef CONFIG_X86_64
5729 * If reserved bit is not supported, clear the present bit to disable
5732 if (maxphyaddr == 52)
5736 kvm_mmu_set_mmio_spte_mask(mask);
5739 #ifdef CONFIG_X86_64
5740 static void pvclock_gtod_update_fn(struct work_struct *work)
5744 struct kvm_vcpu *vcpu;
5747 spin_lock(&kvm_lock);
5748 list_for_each_entry(kvm, &vm_list, vm_list)
5749 kvm_for_each_vcpu(i, vcpu, kvm)
5750 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5751 atomic_set(&kvm_guest_has_master_clock, 0);
5752 spin_unlock(&kvm_lock);
5755 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5758 * Notification about pvclock gtod data update.
5760 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5763 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5764 struct timekeeper *tk = priv;
5766 update_pvclock_gtod(tk);
5768 /* disable master clock if host does not trust, or does not
5769 * use, TSC clocksource
5771 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5772 atomic_read(&kvm_guest_has_master_clock) != 0)
5773 queue_work(system_long_wq, &pvclock_gtod_work);
5778 static struct notifier_block pvclock_gtod_notifier = {
5779 .notifier_call = pvclock_gtod_notify,
5783 int kvm_arch_init(void *opaque)
5786 struct kvm_x86_ops *ops = opaque;
5789 printk(KERN_ERR "kvm: already loaded the other module\n");
5794 if (!ops->cpu_has_kvm_support()) {
5795 printk(KERN_ERR "kvm: no hardware support\n");
5799 if (ops->disabled_by_bios()) {
5800 printk(KERN_ERR "kvm: disabled by bios\n");
5806 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5808 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5812 r = kvm_mmu_module_init();
5814 goto out_free_percpu;
5816 kvm_set_mmio_spte_mask();
5820 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5821 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5825 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5828 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5831 #ifdef CONFIG_X86_64
5832 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5838 free_percpu(shared_msrs);
5843 void kvm_arch_exit(void)
5845 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5847 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5848 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5849 CPUFREQ_TRANSITION_NOTIFIER);
5850 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5851 #ifdef CONFIG_X86_64
5852 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5855 kvm_mmu_module_exit();
5856 free_percpu(shared_msrs);
5859 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5861 ++vcpu->stat.halt_exits;
5862 if (lapic_in_kernel(vcpu)) {
5863 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5866 vcpu->run->exit_reason = KVM_EXIT_HLT;
5870 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5872 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5874 kvm_x86_ops->skip_emulated_instruction(vcpu);
5875 return kvm_vcpu_halt(vcpu);
5877 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5880 * kvm_pv_kick_cpu_op: Kick a vcpu.
5882 * @apicid - apicid of vcpu to be kicked.
5884 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5886 struct kvm_lapic_irq lapic_irq;
5888 lapic_irq.shorthand = 0;
5889 lapic_irq.dest_mode = 0;
5890 lapic_irq.dest_id = apicid;
5891 lapic_irq.msi_redir_hint = false;
5893 lapic_irq.delivery_mode = APIC_DM_REMRD;
5894 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5897 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
5899 vcpu->arch.apicv_active = false;
5900 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
5903 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5905 unsigned long nr, a0, a1, a2, a3, ret;
5906 int op_64_bit, r = 1;
5908 kvm_x86_ops->skip_emulated_instruction(vcpu);
5910 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5911 return kvm_hv_hypercall(vcpu);
5913 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5914 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5915 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5916 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5917 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5919 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5921 op_64_bit = is_64_bit_mode(vcpu);
5930 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5936 case KVM_HC_VAPIC_POLL_IRQ:
5939 case KVM_HC_KICK_CPU:
5940 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5950 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5951 ++vcpu->stat.hypercalls;
5954 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5956 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5958 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5959 char instruction[3];
5960 unsigned long rip = kvm_rip_read(vcpu);
5962 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5964 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5967 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5969 return vcpu->run->request_interrupt_window &&
5970 likely(!pic_in_kernel(vcpu->kvm));
5973 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5975 struct kvm_run *kvm_run = vcpu->run;
5977 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5978 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
5979 kvm_run->cr8 = kvm_get_cr8(vcpu);
5980 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5981 kvm_run->ready_for_interrupt_injection =
5982 pic_in_kernel(vcpu->kvm) ||
5983 kvm_vcpu_ready_for_interrupt_injection(vcpu);
5986 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5990 if (!kvm_x86_ops->update_cr8_intercept)
5993 if (!vcpu->arch.apic)
5996 if (vcpu->arch.apicv_active)
5999 if (!vcpu->arch.apic->vapic_addr)
6000 max_irr = kvm_lapic_find_highest_irr(vcpu);
6007 tpr = kvm_lapic_get_cr8(vcpu);
6009 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6012 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6016 /* try to reinject previous events if any */
6017 if (vcpu->arch.exception.pending) {
6018 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6019 vcpu->arch.exception.has_error_code,
6020 vcpu->arch.exception.error_code);
6022 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6023 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6026 if (vcpu->arch.exception.nr == DB_VECTOR &&
6027 (vcpu->arch.dr7 & DR7_GD)) {
6028 vcpu->arch.dr7 &= ~DR7_GD;
6029 kvm_update_dr7(vcpu);
6032 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6033 vcpu->arch.exception.has_error_code,
6034 vcpu->arch.exception.error_code,
6035 vcpu->arch.exception.reinject);
6039 if (vcpu->arch.nmi_injected) {
6040 kvm_x86_ops->set_nmi(vcpu);
6044 if (vcpu->arch.interrupt.pending) {
6045 kvm_x86_ops->set_irq(vcpu);
6049 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6050 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6055 /* try to inject new event if pending */
6056 if (vcpu->arch.nmi_pending) {
6057 if (kvm_x86_ops->nmi_allowed(vcpu)) {
6058 --vcpu->arch.nmi_pending;
6059 vcpu->arch.nmi_injected = true;
6060 kvm_x86_ops->set_nmi(vcpu);
6062 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6064 * Because interrupts can be injected asynchronously, we are
6065 * calling check_nested_events again here to avoid a race condition.
6066 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6067 * proposal and current concerns. Perhaps we should be setting
6068 * KVM_REQ_EVENT only on certain events and not unconditionally?
6070 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6071 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6075 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6076 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6078 kvm_x86_ops->set_irq(vcpu);
6084 static void process_nmi(struct kvm_vcpu *vcpu)
6089 * x86 is limited to one NMI running, and one NMI pending after it.
6090 * If an NMI is already in progress, limit further NMIs to just one.
6091 * Otherwise, allow two (and we'll inject the first one immediately).
6093 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6096 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6097 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6098 kvm_make_request(KVM_REQ_EVENT, vcpu);
6101 #define put_smstate(type, buf, offset, val) \
6102 *(type *)((buf) + (offset) - 0x7e00) = val
6104 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
6107 flags |= seg->g << 23;
6108 flags |= seg->db << 22;
6109 flags |= seg->l << 21;
6110 flags |= seg->avl << 20;
6111 flags |= seg->present << 15;
6112 flags |= seg->dpl << 13;
6113 flags |= seg->s << 12;
6114 flags |= seg->type << 8;
6118 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6120 struct kvm_segment seg;
6123 kvm_get_segment(vcpu, &seg, n);
6124 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6127 offset = 0x7f84 + n * 12;
6129 offset = 0x7f2c + (n - 3) * 12;
6131 put_smstate(u32, buf, offset + 8, seg.base);
6132 put_smstate(u32, buf, offset + 4, seg.limit);
6133 put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
6136 #ifdef CONFIG_X86_64
6137 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6139 struct kvm_segment seg;
6143 kvm_get_segment(vcpu, &seg, n);
6144 offset = 0x7e00 + n * 16;
6146 flags = process_smi_get_segment_flags(&seg) >> 8;
6147 put_smstate(u16, buf, offset, seg.selector);
6148 put_smstate(u16, buf, offset + 2, flags);
6149 put_smstate(u32, buf, offset + 4, seg.limit);
6150 put_smstate(u64, buf, offset + 8, seg.base);
6154 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6157 struct kvm_segment seg;
6161 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6162 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6163 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6164 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6166 for (i = 0; i < 8; i++)
6167 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6169 kvm_get_dr(vcpu, 6, &val);
6170 put_smstate(u32, buf, 0x7fcc, (u32)val);
6171 kvm_get_dr(vcpu, 7, &val);
6172 put_smstate(u32, buf, 0x7fc8, (u32)val);
6174 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6175 put_smstate(u32, buf, 0x7fc4, seg.selector);
6176 put_smstate(u32, buf, 0x7f64, seg.base);
6177 put_smstate(u32, buf, 0x7f60, seg.limit);
6178 put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
6180 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6181 put_smstate(u32, buf, 0x7fc0, seg.selector);
6182 put_smstate(u32, buf, 0x7f80, seg.base);
6183 put_smstate(u32, buf, 0x7f7c, seg.limit);
6184 put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6186 kvm_x86_ops->get_gdt(vcpu, &dt);
6187 put_smstate(u32, buf, 0x7f74, dt.address);
6188 put_smstate(u32, buf, 0x7f70, dt.size);
6190 kvm_x86_ops->get_idt(vcpu, &dt);
6191 put_smstate(u32, buf, 0x7f58, dt.address);
6192 put_smstate(u32, buf, 0x7f54, dt.size);
6194 for (i = 0; i < 6; i++)
6195 process_smi_save_seg_32(vcpu, buf, i);
6197 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6200 put_smstate(u32, buf, 0x7efc, 0x00020000);
6201 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6204 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6206 #ifdef CONFIG_X86_64
6208 struct kvm_segment seg;
6212 for (i = 0; i < 16; i++)
6213 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6215 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6216 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6218 kvm_get_dr(vcpu, 6, &val);
6219 put_smstate(u64, buf, 0x7f68, val);
6220 kvm_get_dr(vcpu, 7, &val);
6221 put_smstate(u64, buf, 0x7f60, val);
6223 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6224 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6225 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6227 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6230 put_smstate(u32, buf, 0x7efc, 0x00020064);
6232 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6234 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6235 put_smstate(u16, buf, 0x7e90, seg.selector);
6236 put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6237 put_smstate(u32, buf, 0x7e94, seg.limit);
6238 put_smstate(u64, buf, 0x7e98, seg.base);
6240 kvm_x86_ops->get_idt(vcpu, &dt);
6241 put_smstate(u32, buf, 0x7e84, dt.size);
6242 put_smstate(u64, buf, 0x7e88, dt.address);
6244 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6245 put_smstate(u16, buf, 0x7e70, seg.selector);
6246 put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6247 put_smstate(u32, buf, 0x7e74, seg.limit);
6248 put_smstate(u64, buf, 0x7e78, seg.base);
6250 kvm_x86_ops->get_gdt(vcpu, &dt);
6251 put_smstate(u32, buf, 0x7e64, dt.size);
6252 put_smstate(u64, buf, 0x7e68, dt.address);
6254 for (i = 0; i < 6; i++)
6255 process_smi_save_seg_64(vcpu, buf, i);
6261 static void process_smi(struct kvm_vcpu *vcpu)
6263 struct kvm_segment cs, ds;
6269 vcpu->arch.smi_pending = true;
6273 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6274 vcpu->arch.hflags |= HF_SMM_MASK;
6275 memset(buf, 0, 512);
6276 if (guest_cpuid_has_longmode(vcpu))
6277 process_smi_save_state_64(vcpu, buf);
6279 process_smi_save_state_32(vcpu, buf);
6281 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6283 if (kvm_x86_ops->get_nmi_mask(vcpu))
6284 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6286 kvm_x86_ops->set_nmi_mask(vcpu, true);
6288 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6289 kvm_rip_write(vcpu, 0x8000);
6291 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6292 kvm_x86_ops->set_cr0(vcpu, cr0);
6293 vcpu->arch.cr0 = cr0;
6295 kvm_x86_ops->set_cr4(vcpu, 0);
6297 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6298 dt.address = dt.size = 0;
6299 kvm_x86_ops->set_idt(vcpu, &dt);
6301 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6303 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6304 cs.base = vcpu->arch.smbase;
6309 cs.limit = ds.limit = 0xffffffff;
6310 cs.type = ds.type = 0x3;
6311 cs.dpl = ds.dpl = 0;
6316 cs.avl = ds.avl = 0;
6317 cs.present = ds.present = 1;
6318 cs.unusable = ds.unusable = 0;
6319 cs.padding = ds.padding = 0;
6321 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6322 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6323 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6324 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6325 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6326 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6328 if (guest_cpuid_has_longmode(vcpu))
6329 kvm_x86_ops->set_efer(vcpu, 0);
6331 kvm_update_cpuid(vcpu);
6332 kvm_mmu_reset_context(vcpu);
6335 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6337 u64 eoi_exit_bitmap[4];
6339 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6342 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6344 if (irqchip_split(vcpu->kvm))
6345 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6347 if (vcpu->arch.apicv_active)
6348 kvm_x86_ops->sync_pir_to_irr(vcpu);
6349 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6351 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6352 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6353 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6356 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6358 ++vcpu->stat.tlb_flush;
6359 kvm_x86_ops->tlb_flush(vcpu);
6362 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6364 struct page *page = NULL;
6366 if (!lapic_in_kernel(vcpu))
6369 if (!kvm_x86_ops->set_apic_access_page_addr)
6372 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6373 if (is_error_page(page))
6375 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6378 * Do not pin apic access page in memory, the MMU notifier
6379 * will call us again if it is migrated or swapped out.
6383 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6385 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6386 unsigned long address)
6389 * The physical address of apic access page is stored in the VMCS.
6390 * Update it when it becomes invalid.
6392 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6393 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6397 * Returns 1 to let vcpu_run() continue the guest execution loop without
6398 * exiting to the userspace. Otherwise, the value will be returned to the
6401 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6405 dm_request_for_irq_injection(vcpu) &&
6406 kvm_cpu_accept_dm_intr(vcpu);
6408 bool req_immediate_exit = false;
6410 if (vcpu->requests) {
6411 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6412 kvm_mmu_unload(vcpu);
6413 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6414 __kvm_migrate_timers(vcpu);
6415 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6416 kvm_gen_update_masterclock(vcpu->kvm);
6417 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6418 kvm_gen_kvmclock_update(vcpu);
6419 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6420 r = kvm_guest_time_update(vcpu);
6424 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6425 kvm_mmu_sync_roots(vcpu);
6426 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6427 kvm_vcpu_flush_tlb(vcpu);
6428 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6429 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6433 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6434 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6438 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6439 vcpu->fpu_active = 0;
6440 kvm_x86_ops->fpu_deactivate(vcpu);
6442 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6443 /* Page is swapped out. Do synthetic halt */
6444 vcpu->arch.apf.halted = true;
6448 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6449 record_steal_time(vcpu);
6450 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6452 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6454 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6455 kvm_pmu_handle_event(vcpu);
6456 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6457 kvm_pmu_deliver_pmi(vcpu);
6458 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6459 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6460 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6461 vcpu->arch.ioapic_handled_vectors)) {
6462 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6463 vcpu->run->eoi.vector =
6464 vcpu->arch.pending_ioapic_eoi;
6469 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6470 vcpu_scan_ioapic(vcpu);
6471 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6472 kvm_vcpu_reload_apic_access_page(vcpu);
6473 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6474 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6475 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6479 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6480 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6481 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6488 * KVM_REQ_EVENT is not set when posted interrupts are set by
6489 * VT-d hardware, so we have to update RVI unconditionally.
6491 if (kvm_lapic_enabled(vcpu)) {
6493 * Update architecture specific hints for APIC
6494 * virtual interrupt delivery.
6496 if (vcpu->arch.apicv_active)
6497 kvm_x86_ops->hwapic_irr_update(vcpu,
6498 kvm_lapic_find_highest_irr(vcpu));
6501 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6502 kvm_apic_accept_events(vcpu);
6503 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6508 if (inject_pending_event(vcpu, req_int_win) != 0)
6509 req_immediate_exit = true;
6510 /* enable NMI/IRQ window open exits if needed */
6511 else if (vcpu->arch.nmi_pending)
6512 kvm_x86_ops->enable_nmi_window(vcpu);
6513 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6514 kvm_x86_ops->enable_irq_window(vcpu);
6516 if (kvm_lapic_enabled(vcpu)) {
6517 update_cr8_intercept(vcpu);
6518 kvm_lapic_sync_to_vapic(vcpu);
6522 r = kvm_mmu_reload(vcpu);
6524 goto cancel_injection;
6529 kvm_x86_ops->prepare_guest_switch(vcpu);
6530 if (vcpu->fpu_active)
6531 kvm_load_guest_fpu(vcpu);
6532 kvm_load_guest_xcr0(vcpu);
6534 vcpu->mode = IN_GUEST_MODE;
6536 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6538 /* We should set ->mode before check ->requests,
6539 * see the comment in make_all_cpus_request.
6541 smp_mb__after_srcu_read_unlock();
6543 local_irq_disable();
6545 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6546 || need_resched() || signal_pending(current)) {
6547 vcpu->mode = OUTSIDE_GUEST_MODE;
6551 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6553 goto cancel_injection;
6556 if (req_immediate_exit)
6557 smp_send_reschedule(vcpu->cpu);
6559 __kvm_guest_enter();
6561 if (unlikely(vcpu->arch.switch_db_regs)) {
6563 set_debugreg(vcpu->arch.eff_db[0], 0);
6564 set_debugreg(vcpu->arch.eff_db[1], 1);
6565 set_debugreg(vcpu->arch.eff_db[2], 2);
6566 set_debugreg(vcpu->arch.eff_db[3], 3);
6567 set_debugreg(vcpu->arch.dr6, 6);
6568 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6571 trace_kvm_entry(vcpu->vcpu_id);
6572 wait_lapic_expire(vcpu);
6573 kvm_x86_ops->run(vcpu);
6576 * Do this here before restoring debug registers on the host. And
6577 * since we do this before handling the vmexit, a DR access vmexit
6578 * can (a) read the correct value of the debug registers, (b) set
6579 * KVM_DEBUGREG_WONT_EXIT again.
6581 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6584 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6585 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6586 for (i = 0; i < KVM_NR_DB_REGS; i++)
6587 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6591 * If the guest has used debug registers, at least dr7
6592 * will be disabled while returning to the host.
6593 * If we don't have active breakpoints in the host, we don't
6594 * care about the messed up debug address registers. But if
6595 * we have some of them active, restore the old state.
6597 if (hw_breakpoint_active())
6598 hw_breakpoint_restore();
6600 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6602 vcpu->mode = OUTSIDE_GUEST_MODE;
6605 /* Interrupt is enabled by handle_external_intr() */
6606 kvm_x86_ops->handle_external_intr(vcpu);
6611 * We must have an instruction between local_irq_enable() and
6612 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6613 * the interrupt shadow. The stat.exits increment will do nicely.
6614 * But we need to prevent reordering, hence this barrier():
6622 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6625 * Profile KVM exit RIPs:
6627 if (unlikely(prof_on == KVM_PROFILING)) {
6628 unsigned long rip = kvm_rip_read(vcpu);
6629 profile_hit(KVM_PROFILING, (void *)rip);
6632 if (unlikely(vcpu->arch.tsc_always_catchup))
6633 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6635 if (vcpu->arch.apic_attention)
6636 kvm_lapic_sync_from_vapic(vcpu);
6638 r = kvm_x86_ops->handle_exit(vcpu);
6642 kvm_x86_ops->cancel_injection(vcpu);
6643 if (unlikely(vcpu->arch.apic_attention))
6644 kvm_lapic_sync_from_vapic(vcpu);
6649 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6651 if (!kvm_arch_vcpu_runnable(vcpu) &&
6652 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6653 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6654 kvm_vcpu_block(vcpu);
6655 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6657 if (kvm_x86_ops->post_block)
6658 kvm_x86_ops->post_block(vcpu);
6660 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6664 kvm_apic_accept_events(vcpu);
6665 switch(vcpu->arch.mp_state) {
6666 case KVM_MP_STATE_HALTED:
6667 vcpu->arch.pv.pv_unhalted = false;
6668 vcpu->arch.mp_state =
6669 KVM_MP_STATE_RUNNABLE;
6670 case KVM_MP_STATE_RUNNABLE:
6671 vcpu->arch.apf.halted = false;
6673 case KVM_MP_STATE_INIT_RECEIVED:
6682 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6684 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6685 !vcpu->arch.apf.halted);
6688 static int vcpu_run(struct kvm_vcpu *vcpu)
6691 struct kvm *kvm = vcpu->kvm;
6693 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6696 if (kvm_vcpu_running(vcpu)) {
6697 r = vcpu_enter_guest(vcpu);
6699 r = vcpu_block(kvm, vcpu);
6705 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6706 if (kvm_cpu_has_pending_timer(vcpu))
6707 kvm_inject_pending_timer_irqs(vcpu);
6709 if (dm_request_for_irq_injection(vcpu) &&
6710 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
6712 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6713 ++vcpu->stat.request_irq_exits;
6717 kvm_check_async_pf_completion(vcpu);
6719 if (signal_pending(current)) {
6721 vcpu->run->exit_reason = KVM_EXIT_INTR;
6722 ++vcpu->stat.signal_exits;
6725 if (need_resched()) {
6726 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6728 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6732 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6737 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6740 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6741 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6742 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6743 if (r != EMULATE_DONE)
6748 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6750 BUG_ON(!vcpu->arch.pio.count);
6752 return complete_emulated_io(vcpu);
6756 * Implements the following, as a state machine:
6760 * for each mmio piece in the fragment
6768 * for each mmio piece in the fragment
6773 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6775 struct kvm_run *run = vcpu->run;
6776 struct kvm_mmio_fragment *frag;
6779 BUG_ON(!vcpu->mmio_needed);
6781 /* Complete previous fragment */
6782 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6783 len = min(8u, frag->len);
6784 if (!vcpu->mmio_is_write)
6785 memcpy(frag->data, run->mmio.data, len);
6787 if (frag->len <= 8) {
6788 /* Switch to the next fragment. */
6790 vcpu->mmio_cur_fragment++;
6792 /* Go forward to the next mmio piece. */
6798 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6799 vcpu->mmio_needed = 0;
6801 /* FIXME: return into emulator if single-stepping. */
6802 if (vcpu->mmio_is_write)
6804 vcpu->mmio_read_completed = 1;
6805 return complete_emulated_io(vcpu);
6808 run->exit_reason = KVM_EXIT_MMIO;
6809 run->mmio.phys_addr = frag->gpa;
6810 if (vcpu->mmio_is_write)
6811 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6812 run->mmio.len = min(8u, frag->len);
6813 run->mmio.is_write = vcpu->mmio_is_write;
6814 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6819 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6821 struct fpu *fpu = ¤t->thread.fpu;
6825 fpu__activate_curr(fpu);
6827 if (vcpu->sigset_active)
6828 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6830 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6831 kvm_vcpu_block(vcpu);
6832 kvm_apic_accept_events(vcpu);
6833 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6838 /* re-sync apic's tpr */
6839 if (!lapic_in_kernel(vcpu)) {
6840 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6846 if (unlikely(vcpu->arch.complete_userspace_io)) {
6847 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6848 vcpu->arch.complete_userspace_io = NULL;
6853 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6858 post_kvm_run_save(vcpu);
6859 if (vcpu->sigset_active)
6860 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6865 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6867 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6869 * We are here if userspace calls get_regs() in the middle of
6870 * instruction emulation. Registers state needs to be copied
6871 * back from emulation context to vcpu. Userspace shouldn't do
6872 * that usually, but some bad designed PV devices (vmware
6873 * backdoor interface) need this to work
6875 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6876 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6878 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6879 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6880 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6881 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6882 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6883 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6884 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6885 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6886 #ifdef CONFIG_X86_64
6887 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6888 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6889 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6890 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6891 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6892 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6893 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6894 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6897 regs->rip = kvm_rip_read(vcpu);
6898 regs->rflags = kvm_get_rflags(vcpu);
6903 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6905 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6906 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6908 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6909 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6910 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6911 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6912 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6913 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6914 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6915 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6916 #ifdef CONFIG_X86_64
6917 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6918 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6919 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6920 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6921 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6922 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6923 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6924 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6927 kvm_rip_write(vcpu, regs->rip);
6928 kvm_set_rflags(vcpu, regs->rflags);
6930 vcpu->arch.exception.pending = false;
6932 kvm_make_request(KVM_REQ_EVENT, vcpu);
6937 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6939 struct kvm_segment cs;
6941 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6945 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6947 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6948 struct kvm_sregs *sregs)
6952 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6953 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6954 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6955 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6956 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6957 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6959 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6960 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6962 kvm_x86_ops->get_idt(vcpu, &dt);
6963 sregs->idt.limit = dt.size;
6964 sregs->idt.base = dt.address;
6965 kvm_x86_ops->get_gdt(vcpu, &dt);
6966 sregs->gdt.limit = dt.size;
6967 sregs->gdt.base = dt.address;
6969 sregs->cr0 = kvm_read_cr0(vcpu);
6970 sregs->cr2 = vcpu->arch.cr2;
6971 sregs->cr3 = kvm_read_cr3(vcpu);
6972 sregs->cr4 = kvm_read_cr4(vcpu);
6973 sregs->cr8 = kvm_get_cr8(vcpu);
6974 sregs->efer = vcpu->arch.efer;
6975 sregs->apic_base = kvm_get_apic_base(vcpu);
6977 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6979 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6980 set_bit(vcpu->arch.interrupt.nr,
6981 (unsigned long *)sregs->interrupt_bitmap);
6986 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6987 struct kvm_mp_state *mp_state)
6989 kvm_apic_accept_events(vcpu);
6990 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6991 vcpu->arch.pv.pv_unhalted)
6992 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6994 mp_state->mp_state = vcpu->arch.mp_state;
6999 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7000 struct kvm_mp_state *mp_state)
7002 if (!kvm_vcpu_has_lapic(vcpu) &&
7003 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7006 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7007 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7008 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7010 vcpu->arch.mp_state = mp_state->mp_state;
7011 kvm_make_request(KVM_REQ_EVENT, vcpu);
7015 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7016 int reason, bool has_error_code, u32 error_code)
7018 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7021 init_emulate_ctxt(vcpu);
7023 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7024 has_error_code, error_code);
7027 return EMULATE_FAIL;
7029 kvm_rip_write(vcpu, ctxt->eip);
7030 kvm_set_rflags(vcpu, ctxt->eflags);
7031 kvm_make_request(KVM_REQ_EVENT, vcpu);
7032 return EMULATE_DONE;
7034 EXPORT_SYMBOL_GPL(kvm_task_switch);
7036 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7037 struct kvm_sregs *sregs)
7039 struct msr_data apic_base_msr;
7040 int mmu_reset_needed = 0;
7041 int pending_vec, max_bits, idx;
7044 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7047 dt.size = sregs->idt.limit;
7048 dt.address = sregs->idt.base;
7049 kvm_x86_ops->set_idt(vcpu, &dt);
7050 dt.size = sregs->gdt.limit;
7051 dt.address = sregs->gdt.base;
7052 kvm_x86_ops->set_gdt(vcpu, &dt);
7054 vcpu->arch.cr2 = sregs->cr2;
7055 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7056 vcpu->arch.cr3 = sregs->cr3;
7057 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7059 kvm_set_cr8(vcpu, sregs->cr8);
7061 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7062 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7063 apic_base_msr.data = sregs->apic_base;
7064 apic_base_msr.host_initiated = true;
7065 kvm_set_apic_base(vcpu, &apic_base_msr);
7067 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7068 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7069 vcpu->arch.cr0 = sregs->cr0;
7071 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7072 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7073 if (sregs->cr4 & X86_CR4_OSXSAVE)
7074 kvm_update_cpuid(vcpu);
7076 idx = srcu_read_lock(&vcpu->kvm->srcu);
7077 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7078 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7079 mmu_reset_needed = 1;
7081 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7083 if (mmu_reset_needed)
7084 kvm_mmu_reset_context(vcpu);
7086 max_bits = KVM_NR_INTERRUPTS;
7087 pending_vec = find_first_bit(
7088 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7089 if (pending_vec < max_bits) {
7090 kvm_queue_interrupt(vcpu, pending_vec, false);
7091 pr_debug("Set back pending irq %d\n", pending_vec);
7094 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7095 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7096 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7097 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7098 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7099 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7101 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7102 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7104 update_cr8_intercept(vcpu);
7106 /* Older userspace won't unhalt the vcpu on reset. */
7107 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7108 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7110 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7112 kvm_make_request(KVM_REQ_EVENT, vcpu);
7117 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7118 struct kvm_guest_debug *dbg)
7120 unsigned long rflags;
7123 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7125 if (vcpu->arch.exception.pending)
7127 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7128 kvm_queue_exception(vcpu, DB_VECTOR);
7130 kvm_queue_exception(vcpu, BP_VECTOR);
7134 * Read rflags as long as potentially injected trace flags are still
7137 rflags = kvm_get_rflags(vcpu);
7139 vcpu->guest_debug = dbg->control;
7140 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7141 vcpu->guest_debug = 0;
7143 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7144 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7145 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7146 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7148 for (i = 0; i < KVM_NR_DB_REGS; i++)
7149 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7151 kvm_update_dr7(vcpu);
7153 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7154 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7155 get_segment_base(vcpu, VCPU_SREG_CS);
7158 * Trigger an rflags update that will inject or remove the trace
7161 kvm_set_rflags(vcpu, rflags);
7163 kvm_x86_ops->update_bp_intercept(vcpu);
7173 * Translate a guest virtual address to a guest physical address.
7175 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7176 struct kvm_translation *tr)
7178 unsigned long vaddr = tr->linear_address;
7182 idx = srcu_read_lock(&vcpu->kvm->srcu);
7183 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7184 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7185 tr->physical_address = gpa;
7186 tr->valid = gpa != UNMAPPED_GVA;
7193 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7195 struct fxregs_state *fxsave =
7196 &vcpu->arch.guest_fpu.state.fxsave;
7198 memcpy(fpu->fpr, fxsave->st_space, 128);
7199 fpu->fcw = fxsave->cwd;
7200 fpu->fsw = fxsave->swd;
7201 fpu->ftwx = fxsave->twd;
7202 fpu->last_opcode = fxsave->fop;
7203 fpu->last_ip = fxsave->rip;
7204 fpu->last_dp = fxsave->rdp;
7205 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7210 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7212 struct fxregs_state *fxsave =
7213 &vcpu->arch.guest_fpu.state.fxsave;
7215 memcpy(fxsave->st_space, fpu->fpr, 128);
7216 fxsave->cwd = fpu->fcw;
7217 fxsave->swd = fpu->fsw;
7218 fxsave->twd = fpu->ftwx;
7219 fxsave->fop = fpu->last_opcode;
7220 fxsave->rip = fpu->last_ip;
7221 fxsave->rdp = fpu->last_dp;
7222 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7227 static void fx_init(struct kvm_vcpu *vcpu)
7229 fpstate_init(&vcpu->arch.guest_fpu.state);
7231 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7232 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7235 * Ensure guest xcr0 is valid for loading
7237 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7239 vcpu->arch.cr0 |= X86_CR0_ET;
7242 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7244 if (vcpu->guest_fpu_loaded)
7248 * Restore all possible states in the guest,
7249 * and assume host would use all available bits.
7250 * Guest xcr0 would be loaded later.
7252 kvm_put_guest_xcr0(vcpu);
7253 vcpu->guest_fpu_loaded = 1;
7254 __kernel_fpu_begin();
7255 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7259 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7261 kvm_put_guest_xcr0(vcpu);
7263 if (!vcpu->guest_fpu_loaded) {
7264 vcpu->fpu_counter = 0;
7268 vcpu->guest_fpu_loaded = 0;
7269 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7271 ++vcpu->stat.fpu_reload;
7273 * If using eager FPU mode, or if the guest is a frequent user
7274 * of the FPU, just leave the FPU active for next time.
7275 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7276 * the FPU in bursts will revert to loading it on demand.
7278 if (!vcpu->arch.eager_fpu) {
7279 if (++vcpu->fpu_counter < 5)
7280 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7285 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7287 kvmclock_reset(vcpu);
7289 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7290 kvm_x86_ops->vcpu_free(vcpu);
7293 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7296 struct kvm_vcpu *vcpu;
7298 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7299 printk_once(KERN_WARNING
7300 "kvm: SMP vm created on host with unstable TSC; "
7301 "guest TSC will not be reliable\n");
7303 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7308 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7312 kvm_vcpu_mtrr_init(vcpu);
7313 r = vcpu_load(vcpu);
7316 kvm_vcpu_reset(vcpu, false);
7317 kvm_mmu_setup(vcpu);
7322 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7324 struct msr_data msr;
7325 struct kvm *kvm = vcpu->kvm;
7327 if (vcpu_load(vcpu))
7330 msr.index = MSR_IA32_TSC;
7331 msr.host_initiated = true;
7332 kvm_write_tsc(vcpu, &msr);
7335 if (!kvmclock_periodic_sync)
7338 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7339 KVMCLOCK_SYNC_PERIOD);
7342 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7345 vcpu->arch.apf.msr_val = 0;
7347 r = vcpu_load(vcpu);
7349 kvm_mmu_unload(vcpu);
7352 kvm_x86_ops->vcpu_free(vcpu);
7355 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7357 vcpu->arch.hflags = 0;
7359 atomic_set(&vcpu->arch.nmi_queued, 0);
7360 vcpu->arch.nmi_pending = 0;
7361 vcpu->arch.nmi_injected = false;
7362 kvm_clear_interrupt_queue(vcpu);
7363 kvm_clear_exception_queue(vcpu);
7365 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7366 kvm_update_dr0123(vcpu);
7367 vcpu->arch.dr6 = DR6_INIT;
7368 kvm_update_dr6(vcpu);
7369 vcpu->arch.dr7 = DR7_FIXED_1;
7370 kvm_update_dr7(vcpu);
7374 kvm_make_request(KVM_REQ_EVENT, vcpu);
7375 vcpu->arch.apf.msr_val = 0;
7376 vcpu->arch.st.msr_val = 0;
7378 kvmclock_reset(vcpu);
7380 kvm_clear_async_pf_completion_queue(vcpu);
7381 kvm_async_pf_hash_reset(vcpu);
7382 vcpu->arch.apf.halted = false;
7385 kvm_pmu_reset(vcpu);
7386 vcpu->arch.smbase = 0x30000;
7389 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7390 vcpu->arch.regs_avail = ~0;
7391 vcpu->arch.regs_dirty = ~0;
7393 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7396 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7398 struct kvm_segment cs;
7400 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7401 cs.selector = vector << 8;
7402 cs.base = vector << 12;
7403 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7404 kvm_rip_write(vcpu, 0);
7407 int kvm_arch_hardware_enable(void)
7410 struct kvm_vcpu *vcpu;
7415 bool stable, backwards_tsc = false;
7417 kvm_shared_msr_cpu_online();
7418 ret = kvm_x86_ops->hardware_enable();
7422 local_tsc = rdtsc();
7423 stable = !check_tsc_unstable();
7424 list_for_each_entry(kvm, &vm_list, vm_list) {
7425 kvm_for_each_vcpu(i, vcpu, kvm) {
7426 if (!stable && vcpu->cpu == smp_processor_id())
7427 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7428 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7429 backwards_tsc = true;
7430 if (vcpu->arch.last_host_tsc > max_tsc)
7431 max_tsc = vcpu->arch.last_host_tsc;
7437 * Sometimes, even reliable TSCs go backwards. This happens on
7438 * platforms that reset TSC during suspend or hibernate actions, but
7439 * maintain synchronization. We must compensate. Fortunately, we can
7440 * detect that condition here, which happens early in CPU bringup,
7441 * before any KVM threads can be running. Unfortunately, we can't
7442 * bring the TSCs fully up to date with real time, as we aren't yet far
7443 * enough into CPU bringup that we know how much real time has actually
7444 * elapsed; our helper function, get_kernel_ns() will be using boot
7445 * variables that haven't been updated yet.
7447 * So we simply find the maximum observed TSC above, then record the
7448 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7449 * the adjustment will be applied. Note that we accumulate
7450 * adjustments, in case multiple suspend cycles happen before some VCPU
7451 * gets a chance to run again. In the event that no KVM threads get a
7452 * chance to run, we will miss the entire elapsed period, as we'll have
7453 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7454 * loose cycle time. This isn't too big a deal, since the loss will be
7455 * uniform across all VCPUs (not to mention the scenario is extremely
7456 * unlikely). It is possible that a second hibernate recovery happens
7457 * much faster than a first, causing the observed TSC here to be
7458 * smaller; this would require additional padding adjustment, which is
7459 * why we set last_host_tsc to the local tsc observed here.
7461 * N.B. - this code below runs only on platforms with reliable TSC,
7462 * as that is the only way backwards_tsc is set above. Also note
7463 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7464 * have the same delta_cyc adjustment applied if backwards_tsc
7465 * is detected. Note further, this adjustment is only done once,
7466 * as we reset last_host_tsc on all VCPUs to stop this from being
7467 * called multiple times (one for each physical CPU bringup).
7469 * Platforms with unreliable TSCs don't have to deal with this, they
7470 * will be compensated by the logic in vcpu_load, which sets the TSC to
7471 * catchup mode. This will catchup all VCPUs to real time, but cannot
7472 * guarantee that they stay in perfect synchronization.
7474 if (backwards_tsc) {
7475 u64 delta_cyc = max_tsc - local_tsc;
7476 backwards_tsc_observed = true;
7477 list_for_each_entry(kvm, &vm_list, vm_list) {
7478 kvm_for_each_vcpu(i, vcpu, kvm) {
7479 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7480 vcpu->arch.last_host_tsc = local_tsc;
7481 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7485 * We have to disable TSC offset matching.. if you were
7486 * booting a VM while issuing an S4 host suspend....
7487 * you may have some problem. Solving this issue is
7488 * left as an exercise to the reader.
7490 kvm->arch.last_tsc_nsec = 0;
7491 kvm->arch.last_tsc_write = 0;
7498 void kvm_arch_hardware_disable(void)
7500 kvm_x86_ops->hardware_disable();
7501 drop_user_return_notifiers();
7504 int kvm_arch_hardware_setup(void)
7508 r = kvm_x86_ops->hardware_setup();
7512 if (kvm_has_tsc_control) {
7514 * Make sure the user can only configure tsc_khz values that
7515 * fit into a signed integer.
7516 * A min value is not calculated needed because it will always
7517 * be 1 on all machines.
7519 u64 max = min(0x7fffffffULL,
7520 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7521 kvm_max_guest_tsc_khz = max;
7523 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7526 kvm_init_msr_list();
7530 void kvm_arch_hardware_unsetup(void)
7532 kvm_x86_ops->hardware_unsetup();
7535 void kvm_arch_check_processor_compat(void *rtn)
7537 kvm_x86_ops->check_processor_compatibility(rtn);
7540 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7542 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7544 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7546 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7548 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7551 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7553 return irqchip_in_kernel(vcpu->kvm) == lapic_in_kernel(vcpu);
7556 struct static_key kvm_no_apic_vcpu __read_mostly;
7558 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7564 BUG_ON(vcpu->kvm == NULL);
7567 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7568 vcpu->arch.pv.pv_unhalted = false;
7569 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7570 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7571 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7573 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7575 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7580 vcpu->arch.pio_data = page_address(page);
7582 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7584 r = kvm_mmu_create(vcpu);
7586 goto fail_free_pio_data;
7588 if (irqchip_in_kernel(kvm)) {
7589 r = kvm_create_lapic(vcpu);
7591 goto fail_mmu_destroy;
7593 static_key_slow_inc(&kvm_no_apic_vcpu);
7595 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7597 if (!vcpu->arch.mce_banks) {
7599 goto fail_free_lapic;
7601 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7603 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7605 goto fail_free_mce_banks;
7610 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7611 vcpu->arch.pv_time_enabled = false;
7613 vcpu->arch.guest_supported_xcr0 = 0;
7614 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7616 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7618 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7620 kvm_async_pf_hash_reset(vcpu);
7623 vcpu->arch.pending_external_vector = -1;
7625 kvm_hv_vcpu_init(vcpu);
7629 fail_free_mce_banks:
7630 kfree(vcpu->arch.mce_banks);
7632 kvm_free_lapic(vcpu);
7634 kvm_mmu_destroy(vcpu);
7636 free_page((unsigned long)vcpu->arch.pio_data);
7641 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7645 kvm_pmu_destroy(vcpu);
7646 kfree(vcpu->arch.mce_banks);
7647 kvm_free_lapic(vcpu);
7648 idx = srcu_read_lock(&vcpu->kvm->srcu);
7649 kvm_mmu_destroy(vcpu);
7650 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7651 free_page((unsigned long)vcpu->arch.pio_data);
7652 if (!lapic_in_kernel(vcpu))
7653 static_key_slow_dec(&kvm_no_apic_vcpu);
7656 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7658 kvm_x86_ops->sched_in(vcpu, cpu);
7661 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7666 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7667 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7668 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7669 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7670 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7672 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7673 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7674 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7675 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7676 &kvm->arch.irq_sources_bitmap);
7678 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7679 mutex_init(&kvm->arch.apic_map_lock);
7680 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7682 pvclock_update_vm_gtod_copy(kvm);
7684 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7685 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7690 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7693 r = vcpu_load(vcpu);
7695 kvm_mmu_unload(vcpu);
7699 static void kvm_free_vcpus(struct kvm *kvm)
7702 struct kvm_vcpu *vcpu;
7705 * Unpin any mmu pages first.
7707 kvm_for_each_vcpu(i, vcpu, kvm) {
7708 kvm_clear_async_pf_completion_queue(vcpu);
7709 kvm_unload_vcpu_mmu(vcpu);
7711 kvm_for_each_vcpu(i, vcpu, kvm)
7712 kvm_arch_vcpu_free(vcpu);
7714 mutex_lock(&kvm->lock);
7715 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7716 kvm->vcpus[i] = NULL;
7718 atomic_set(&kvm->online_vcpus, 0);
7719 mutex_unlock(&kvm->lock);
7722 void kvm_arch_sync_events(struct kvm *kvm)
7724 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7725 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7726 kvm_free_all_assigned_devices(kvm);
7730 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7734 struct kvm_memslots *slots = kvm_memslots(kvm);
7735 struct kvm_memory_slot *slot, old;
7737 /* Called with kvm->slots_lock held. */
7738 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7741 slot = id_to_memslot(slots, id);
7743 if (WARN_ON(slot->npages))
7747 * MAP_SHARED to prevent internal slot pages from being moved
7750 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7751 MAP_SHARED | MAP_ANONYMOUS, 0);
7752 if (IS_ERR((void *)hva))
7753 return PTR_ERR((void *)hva);
7762 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7763 struct kvm_userspace_memory_region m;
7765 m.slot = id | (i << 16);
7767 m.guest_phys_addr = gpa;
7768 m.userspace_addr = hva;
7769 m.memory_size = size;
7770 r = __kvm_set_memory_region(kvm, &m);
7776 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7782 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7784 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7788 mutex_lock(&kvm->slots_lock);
7789 r = __x86_set_memory_region(kvm, id, gpa, size);
7790 mutex_unlock(&kvm->slots_lock);
7794 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7796 void kvm_arch_destroy_vm(struct kvm *kvm)
7798 if (current->mm == kvm->mm) {
7800 * Free memory regions allocated on behalf of userspace,
7801 * unless the the memory map has changed due to process exit
7804 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7805 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7806 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7808 kvm_iommu_unmap_guest(kvm);
7809 kfree(kvm->arch.vpic);
7810 kfree(kvm->arch.vioapic);
7811 kvm_free_vcpus(kvm);
7812 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7815 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7816 struct kvm_memory_slot *dont)
7820 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7821 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7822 kvfree(free->arch.rmap[i]);
7823 free->arch.rmap[i] = NULL;
7828 if (!dont || free->arch.lpage_info[i - 1] !=
7829 dont->arch.lpage_info[i - 1]) {
7830 kvfree(free->arch.lpage_info[i - 1]);
7831 free->arch.lpage_info[i - 1] = NULL;
7836 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7837 unsigned long npages)
7841 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7846 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7847 slot->base_gfn, level) + 1;
7849 slot->arch.rmap[i] =
7850 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7851 if (!slot->arch.rmap[i])
7856 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7857 sizeof(*slot->arch.lpage_info[i - 1]));
7858 if (!slot->arch.lpage_info[i - 1])
7861 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7862 slot->arch.lpage_info[i - 1][0].write_count = 1;
7863 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7864 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7865 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7867 * If the gfn and userspace address are not aligned wrt each
7868 * other, or if explicitly asked to, disable large page
7869 * support for this slot
7871 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7872 !kvm_largepages_enabled()) {
7875 for (j = 0; j < lpages; ++j)
7876 slot->arch.lpage_info[i - 1][j].write_count = 1;
7883 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7884 kvfree(slot->arch.rmap[i]);
7885 slot->arch.rmap[i] = NULL;
7889 kvfree(slot->arch.lpage_info[i - 1]);
7890 slot->arch.lpage_info[i - 1] = NULL;
7895 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7898 * memslots->generation has been incremented.
7899 * mmio generation may have reached its maximum value.
7901 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7904 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7905 struct kvm_memory_slot *memslot,
7906 const struct kvm_userspace_memory_region *mem,
7907 enum kvm_mr_change change)
7912 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7913 struct kvm_memory_slot *new)
7915 /* Still write protect RO slot */
7916 if (new->flags & KVM_MEM_READONLY) {
7917 kvm_mmu_slot_remove_write_access(kvm, new);
7922 * Call kvm_x86_ops dirty logging hooks when they are valid.
7924 * kvm_x86_ops->slot_disable_log_dirty is called when:
7926 * - KVM_MR_CREATE with dirty logging is disabled
7927 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7929 * The reason is, in case of PML, we need to set D-bit for any slots
7930 * with dirty logging disabled in order to eliminate unnecessary GPA
7931 * logging in PML buffer (and potential PML buffer full VMEXT). This
7932 * guarantees leaving PML enabled during guest's lifetime won't have
7933 * any additonal overhead from PML when guest is running with dirty
7934 * logging disabled for memory slots.
7936 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7937 * to dirty logging mode.
7939 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7941 * In case of write protect:
7943 * Write protect all pages for dirty logging.
7945 * All the sptes including the large sptes which point to this
7946 * slot are set to readonly. We can not create any new large
7947 * spte on this slot until the end of the logging.
7949 * See the comments in fast_page_fault().
7951 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7952 if (kvm_x86_ops->slot_enable_log_dirty)
7953 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7955 kvm_mmu_slot_remove_write_access(kvm, new);
7957 if (kvm_x86_ops->slot_disable_log_dirty)
7958 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7962 void kvm_arch_commit_memory_region(struct kvm *kvm,
7963 const struct kvm_userspace_memory_region *mem,
7964 const struct kvm_memory_slot *old,
7965 const struct kvm_memory_slot *new,
7966 enum kvm_mr_change change)
7968 int nr_mmu_pages = 0;
7970 if (!kvm->arch.n_requested_mmu_pages)
7971 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7974 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7977 * Dirty logging tracks sptes in 4k granularity, meaning that large
7978 * sptes have to be split. If live migration is successful, the guest
7979 * in the source machine will be destroyed and large sptes will be
7980 * created in the destination. However, if the guest continues to run
7981 * in the source machine (for example if live migration fails), small
7982 * sptes will remain around and cause bad performance.
7984 * Scan sptes if dirty logging has been stopped, dropping those
7985 * which can be collapsed into a single large-page spte. Later
7986 * page faults will create the large-page sptes.
7988 if ((change != KVM_MR_DELETE) &&
7989 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7990 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7991 kvm_mmu_zap_collapsible_sptes(kvm, new);
7994 * Set up write protection and/or dirty logging for the new slot.
7996 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7997 * been zapped so no dirty logging staff is needed for old slot. For
7998 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7999 * new and it's also covered when dealing with the new slot.
8001 * FIXME: const-ify all uses of struct kvm_memory_slot.
8003 if (change != KVM_MR_DELETE)
8004 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8007 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8009 kvm_mmu_invalidate_zap_all_pages(kvm);
8012 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8013 struct kvm_memory_slot *slot)
8015 kvm_mmu_invalidate_zap_all_pages(kvm);
8018 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8020 if (!list_empty_careful(&vcpu->async_pf.done))
8023 if (kvm_apic_has_events(vcpu))
8026 if (vcpu->arch.pv.pv_unhalted)
8029 if (atomic_read(&vcpu->arch.nmi_queued))
8032 if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8035 if (kvm_arch_interrupt_allowed(vcpu) &&
8036 kvm_cpu_has_interrupt(vcpu))
8042 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8044 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8045 kvm_x86_ops->check_nested_events(vcpu, false);
8047 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8050 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8052 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8055 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8057 return kvm_x86_ops->interrupt_allowed(vcpu);
8060 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8062 if (is_64_bit_mode(vcpu))
8063 return kvm_rip_read(vcpu);
8064 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8065 kvm_rip_read(vcpu));
8067 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8069 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8071 return kvm_get_linear_rip(vcpu) == linear_rip;
8073 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8075 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8077 unsigned long rflags;
8079 rflags = kvm_x86_ops->get_rflags(vcpu);
8080 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8081 rflags &= ~X86_EFLAGS_TF;
8084 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8086 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8088 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8089 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8090 rflags |= X86_EFLAGS_TF;
8091 kvm_x86_ops->set_rflags(vcpu, rflags);
8094 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8096 __kvm_set_rflags(vcpu, rflags);
8097 kvm_make_request(KVM_REQ_EVENT, vcpu);
8099 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8101 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8105 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8109 r = kvm_mmu_reload(vcpu);
8113 if (!vcpu->arch.mmu.direct_map &&
8114 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8117 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8120 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8122 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8125 static inline u32 kvm_async_pf_next_probe(u32 key)
8127 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8130 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8132 u32 key = kvm_async_pf_hash_fn(gfn);
8134 while (vcpu->arch.apf.gfns[key] != ~0)
8135 key = kvm_async_pf_next_probe(key);
8137 vcpu->arch.apf.gfns[key] = gfn;
8140 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8143 u32 key = kvm_async_pf_hash_fn(gfn);
8145 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8146 (vcpu->arch.apf.gfns[key] != gfn &&
8147 vcpu->arch.apf.gfns[key] != ~0); i++)
8148 key = kvm_async_pf_next_probe(key);
8153 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8155 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8158 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8162 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8164 vcpu->arch.apf.gfns[i] = ~0;
8166 j = kvm_async_pf_next_probe(j);
8167 if (vcpu->arch.apf.gfns[j] == ~0)
8169 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8171 * k lies cyclically in ]i,j]
8173 * |....j i.k.| or |.k..j i...|
8175 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8176 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8181 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8184 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8188 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8189 struct kvm_async_pf *work)
8191 struct x86_exception fault;
8193 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8194 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8196 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8197 (vcpu->arch.apf.send_user_only &&
8198 kvm_x86_ops->get_cpl(vcpu) == 0))
8199 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8200 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8201 fault.vector = PF_VECTOR;
8202 fault.error_code_valid = true;
8203 fault.error_code = 0;
8204 fault.nested_page_fault = false;
8205 fault.address = work->arch.token;
8206 kvm_inject_page_fault(vcpu, &fault);
8210 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8211 struct kvm_async_pf *work)
8213 struct x86_exception fault;
8215 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8216 if (work->wakeup_all)
8217 work->arch.token = ~0; /* broadcast wakeup */
8219 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8221 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8222 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8223 fault.vector = PF_VECTOR;
8224 fault.error_code_valid = true;
8225 fault.error_code = 0;
8226 fault.nested_page_fault = false;
8227 fault.address = work->arch.token;
8228 kvm_inject_page_fault(vcpu, &fault);
8230 vcpu->arch.apf.halted = false;
8231 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8234 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8236 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8239 return !kvm_event_needs_reinjection(vcpu) &&
8240 kvm_x86_ops->interrupt_allowed(vcpu);
8243 void kvm_arch_start_assignment(struct kvm *kvm)
8245 atomic_inc(&kvm->arch.assigned_device_count);
8247 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8249 void kvm_arch_end_assignment(struct kvm *kvm)
8251 atomic_dec(&kvm->arch.assigned_device_count);
8253 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8255 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8257 return atomic_read(&kvm->arch.assigned_device_count);
8259 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8261 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8263 atomic_inc(&kvm->arch.noncoherent_dma_count);
8265 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8267 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8269 atomic_dec(&kvm->arch.noncoherent_dma_count);
8271 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8273 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8275 return atomic_read(&kvm->arch.noncoherent_dma_count);
8277 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8279 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8280 struct irq_bypass_producer *prod)
8282 struct kvm_kernel_irqfd *irqfd =
8283 container_of(cons, struct kvm_kernel_irqfd, consumer);
8285 if (kvm_x86_ops->update_pi_irte) {
8286 irqfd->producer = prod;
8287 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8288 prod->irq, irqfd->gsi, 1);
8294 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8295 struct irq_bypass_producer *prod)
8298 struct kvm_kernel_irqfd *irqfd =
8299 container_of(cons, struct kvm_kernel_irqfd, consumer);
8301 if (!kvm_x86_ops->update_pi_irte) {
8302 WARN_ON(irqfd->producer != NULL);
8306 WARN_ON(irqfd->producer != prod);
8307 irqfd->producer = NULL;
8310 * When producer of consumer is unregistered, we change back to
8311 * remapped mode, so we can re-use the current implementation
8312 * when the irq is masked/disabed or the consumer side (KVM
8313 * int this case doesn't want to receive the interrupts.
8315 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8317 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8318 " fails: %d\n", irqfd->consumer.token, ret);
8321 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8322 uint32_t guest_irq, bool set)
8324 if (!kvm_x86_ops->update_pi_irte)
8327 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8330 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8331 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8332 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8333 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8334 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8335 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8336 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8337 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8338 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8339 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8340 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8341 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8342 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8343 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8344 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8345 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8346 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);