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"
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.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 <linux/sched/stat.h>
57 #include <linux/mem_encrypt.h>
59 #include <trace/events/kvm.h>
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 CREATE_TRACE_POINTS
74 #define MAX_IO_MSRS 256
75 #define KVM_MAX_MCE_BANKS 32
76 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
77 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
79 #define emul_to_vcpu(ctxt) \
80 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
83 * - enable syscall per default because its emulated by KVM
84 * - enable LME and LMA per default on 64 bit KVM
88 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
90 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
93 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
94 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
96 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
97 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
99 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
100 static void process_nmi(struct kvm_vcpu *vcpu);
101 static void enter_smm(struct kvm_vcpu *vcpu);
102 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
104 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
105 EXPORT_SYMBOL_GPL(kvm_x86_ops);
107 static bool __read_mostly ignore_msrs = 0;
108 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
110 static bool __read_mostly report_ignored_msrs = true;
111 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
113 unsigned int min_timer_period_us = 500;
114 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
116 static bool __read_mostly kvmclock_periodic_sync = true;
117 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
119 bool __read_mostly kvm_has_tsc_control;
120 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
121 u32 __read_mostly kvm_max_guest_tsc_khz;
122 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
123 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
124 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
125 u64 __read_mostly kvm_max_tsc_scaling_ratio;
126 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
127 u64 __read_mostly kvm_default_tsc_scaling_ratio;
128 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
130 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
131 static u32 __read_mostly tsc_tolerance_ppm = 250;
132 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
134 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
135 unsigned int __read_mostly lapic_timer_advance_ns = 0;
136 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
138 static bool __read_mostly vector_hashing = true;
139 module_param(vector_hashing, bool, S_IRUGO);
141 #define KVM_NR_SHARED_MSRS 16
143 struct kvm_shared_msrs_global {
145 u32 msrs[KVM_NR_SHARED_MSRS];
148 struct kvm_shared_msrs {
149 struct user_return_notifier urn;
151 struct kvm_shared_msr_values {
154 } values[KVM_NR_SHARED_MSRS];
157 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
158 static struct kvm_shared_msrs __percpu *shared_msrs;
160 struct kvm_stats_debugfs_item debugfs_entries[] = {
161 { "pf_fixed", VCPU_STAT(pf_fixed) },
162 { "pf_guest", VCPU_STAT(pf_guest) },
163 { "tlb_flush", VCPU_STAT(tlb_flush) },
164 { "invlpg", VCPU_STAT(invlpg) },
165 { "exits", VCPU_STAT(exits) },
166 { "io_exits", VCPU_STAT(io_exits) },
167 { "mmio_exits", VCPU_STAT(mmio_exits) },
168 { "signal_exits", VCPU_STAT(signal_exits) },
169 { "irq_window", VCPU_STAT(irq_window_exits) },
170 { "nmi_window", VCPU_STAT(nmi_window_exits) },
171 { "halt_exits", VCPU_STAT(halt_exits) },
172 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
173 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
174 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
175 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
176 { "hypercalls", VCPU_STAT(hypercalls) },
177 { "request_irq", VCPU_STAT(request_irq_exits) },
178 { "irq_exits", VCPU_STAT(irq_exits) },
179 { "host_state_reload", VCPU_STAT(host_state_reload) },
180 { "efer_reload", VCPU_STAT(efer_reload) },
181 { "fpu_reload", VCPU_STAT(fpu_reload) },
182 { "insn_emulation", VCPU_STAT(insn_emulation) },
183 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
184 { "irq_injections", VCPU_STAT(irq_injections) },
185 { "nmi_injections", VCPU_STAT(nmi_injections) },
186 { "req_event", VCPU_STAT(req_event) },
187 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
188 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
189 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
190 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
191 { "mmu_flooded", VM_STAT(mmu_flooded) },
192 { "mmu_recycled", VM_STAT(mmu_recycled) },
193 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
194 { "mmu_unsync", VM_STAT(mmu_unsync) },
195 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
196 { "largepages", VM_STAT(lpages) },
197 { "max_mmu_page_hash_collisions",
198 VM_STAT(max_mmu_page_hash_collisions) },
202 u64 __read_mostly host_xcr0;
204 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
206 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
209 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
210 vcpu->arch.apf.gfns[i] = ~0;
213 static void kvm_on_user_return(struct user_return_notifier *urn)
216 struct kvm_shared_msrs *locals
217 = container_of(urn, struct kvm_shared_msrs, urn);
218 struct kvm_shared_msr_values *values;
222 * Disabling irqs at this point since the following code could be
223 * interrupted and executed through kvm_arch_hardware_disable()
225 local_irq_save(flags);
226 if (locals->registered) {
227 locals->registered = false;
228 user_return_notifier_unregister(urn);
230 local_irq_restore(flags);
231 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
232 values = &locals->values[slot];
233 if (values->host != values->curr) {
234 wrmsrl(shared_msrs_global.msrs[slot], values->host);
235 values->curr = values->host;
240 static void shared_msr_update(unsigned slot, u32 msr)
243 unsigned int cpu = smp_processor_id();
244 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
246 /* only read, and nobody should modify it at this time,
247 * so don't need lock */
248 if (slot >= shared_msrs_global.nr) {
249 printk(KERN_ERR "kvm: invalid MSR slot!");
252 rdmsrl_safe(msr, &value);
253 smsr->values[slot].host = value;
254 smsr->values[slot].curr = value;
257 void kvm_define_shared_msr(unsigned slot, u32 msr)
259 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
260 shared_msrs_global.msrs[slot] = msr;
261 if (slot >= shared_msrs_global.nr)
262 shared_msrs_global.nr = slot + 1;
264 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
266 static void kvm_shared_msr_cpu_online(void)
270 for (i = 0; i < shared_msrs_global.nr; ++i)
271 shared_msr_update(i, shared_msrs_global.msrs[i]);
274 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
276 unsigned int cpu = smp_processor_id();
277 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
280 if (((value ^ smsr->values[slot].curr) & mask) == 0)
282 smsr->values[slot].curr = value;
283 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
287 if (!smsr->registered) {
288 smsr->urn.on_user_return = kvm_on_user_return;
289 user_return_notifier_register(&smsr->urn);
290 smsr->registered = true;
294 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
296 static void drop_user_return_notifiers(void)
298 unsigned int cpu = smp_processor_id();
299 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
301 if (smsr->registered)
302 kvm_on_user_return(&smsr->urn);
305 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
307 return vcpu->arch.apic_base;
309 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
311 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
313 u64 old_state = vcpu->arch.apic_base &
314 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
315 u64 new_state = msr_info->data &
316 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
317 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
318 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
320 if ((msr_info->data & reserved_bits) || new_state == X2APIC_ENABLE)
322 if (!msr_info->host_initiated &&
323 ((new_state == MSR_IA32_APICBASE_ENABLE &&
324 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
325 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
329 kvm_lapic_set_base(vcpu, msr_info->data);
332 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
334 asmlinkage __visible void kvm_spurious_fault(void)
336 /* Fault while not rebooting. We want the trace. */
339 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
341 #define EXCPT_BENIGN 0
342 #define EXCPT_CONTRIBUTORY 1
345 static int exception_class(int vector)
355 return EXCPT_CONTRIBUTORY;
362 #define EXCPT_FAULT 0
364 #define EXCPT_ABORT 2
365 #define EXCPT_INTERRUPT 3
367 static int exception_type(int vector)
371 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
372 return EXCPT_INTERRUPT;
376 /* #DB is trap, as instruction watchpoints are handled elsewhere */
377 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
380 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
383 /* Reserved exceptions will result in fault */
387 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
388 unsigned nr, bool has_error, u32 error_code,
394 kvm_make_request(KVM_REQ_EVENT, vcpu);
396 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
398 if (has_error && !is_protmode(vcpu))
402 * On vmentry, vcpu->arch.exception.pending is only
403 * true if an event injection was blocked by
404 * nested_run_pending. In that case, however,
405 * vcpu_enter_guest requests an immediate exit,
406 * and the guest shouldn't proceed far enough to
409 WARN_ON_ONCE(vcpu->arch.exception.pending);
410 vcpu->arch.exception.injected = true;
412 vcpu->arch.exception.pending = true;
413 vcpu->arch.exception.injected = false;
415 vcpu->arch.exception.has_error_code = has_error;
416 vcpu->arch.exception.nr = nr;
417 vcpu->arch.exception.error_code = error_code;
421 /* to check exception */
422 prev_nr = vcpu->arch.exception.nr;
423 if (prev_nr == DF_VECTOR) {
424 /* triple fault -> shutdown */
425 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
428 class1 = exception_class(prev_nr);
429 class2 = exception_class(nr);
430 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
431 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
433 * Generate double fault per SDM Table 5-5. Set
434 * exception.pending = true so that the double fault
435 * can trigger a nested vmexit.
437 vcpu->arch.exception.pending = true;
438 vcpu->arch.exception.injected = false;
439 vcpu->arch.exception.has_error_code = true;
440 vcpu->arch.exception.nr = DF_VECTOR;
441 vcpu->arch.exception.error_code = 0;
443 /* replace previous exception with a new one in a hope
444 that instruction re-execution will regenerate lost
449 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
451 kvm_multiple_exception(vcpu, nr, false, 0, false);
453 EXPORT_SYMBOL_GPL(kvm_queue_exception);
455 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
457 kvm_multiple_exception(vcpu, nr, false, 0, true);
459 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
461 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
464 kvm_inject_gp(vcpu, 0);
466 return kvm_skip_emulated_instruction(vcpu);
470 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
472 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
474 ++vcpu->stat.pf_guest;
475 vcpu->arch.exception.nested_apf =
476 is_guest_mode(vcpu) && fault->async_page_fault;
477 if (vcpu->arch.exception.nested_apf)
478 vcpu->arch.apf.nested_apf_token = fault->address;
480 vcpu->arch.cr2 = fault->address;
481 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
483 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
485 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
487 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
488 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
490 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
492 return fault->nested_page_fault;
495 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
497 atomic_inc(&vcpu->arch.nmi_queued);
498 kvm_make_request(KVM_REQ_NMI, vcpu);
500 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
502 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
504 kvm_multiple_exception(vcpu, nr, true, error_code, false);
506 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
508 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
510 kvm_multiple_exception(vcpu, nr, true, error_code, true);
512 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
515 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
516 * a #GP and return false.
518 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
520 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
522 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
525 EXPORT_SYMBOL_GPL(kvm_require_cpl);
527 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
529 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
532 kvm_queue_exception(vcpu, UD_VECTOR);
535 EXPORT_SYMBOL_GPL(kvm_require_dr);
538 * This function will be used to read from the physical memory of the currently
539 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
540 * can read from guest physical or from the guest's guest physical memory.
542 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
543 gfn_t ngfn, void *data, int offset, int len,
546 struct x86_exception exception;
550 ngpa = gfn_to_gpa(ngfn);
551 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
552 if (real_gfn == UNMAPPED_GVA)
555 real_gfn = gpa_to_gfn(real_gfn);
557 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
559 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
561 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
562 void *data, int offset, int len, u32 access)
564 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
565 data, offset, len, access);
569 * Load the pae pdptrs. Return true is they are all valid.
571 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
573 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
574 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
577 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
579 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
580 offset * sizeof(u64), sizeof(pdpte),
581 PFERR_USER_MASK|PFERR_WRITE_MASK);
586 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
587 if ((pdpte[i] & PT_PRESENT_MASK) &&
589 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
596 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
597 __set_bit(VCPU_EXREG_PDPTR,
598 (unsigned long *)&vcpu->arch.regs_avail);
599 __set_bit(VCPU_EXREG_PDPTR,
600 (unsigned long *)&vcpu->arch.regs_dirty);
605 EXPORT_SYMBOL_GPL(load_pdptrs);
607 bool pdptrs_changed(struct kvm_vcpu *vcpu)
609 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
615 if (is_long_mode(vcpu) || !is_pae(vcpu))
618 if (!test_bit(VCPU_EXREG_PDPTR,
619 (unsigned long *)&vcpu->arch.regs_avail))
622 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
623 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
624 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
625 PFERR_USER_MASK | PFERR_WRITE_MASK);
628 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
633 EXPORT_SYMBOL_GPL(pdptrs_changed);
635 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
637 unsigned long old_cr0 = kvm_read_cr0(vcpu);
638 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
643 if (cr0 & 0xffffffff00000000UL)
647 cr0 &= ~CR0_RESERVED_BITS;
649 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
652 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
655 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
657 if ((vcpu->arch.efer & EFER_LME)) {
662 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
667 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
672 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
675 kvm_x86_ops->set_cr0(vcpu, cr0);
677 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
678 kvm_clear_async_pf_completion_queue(vcpu);
679 kvm_async_pf_hash_reset(vcpu);
682 if ((cr0 ^ old_cr0) & update_bits)
683 kvm_mmu_reset_context(vcpu);
685 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
686 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
687 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
688 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
692 EXPORT_SYMBOL_GPL(kvm_set_cr0);
694 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
696 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
698 EXPORT_SYMBOL_GPL(kvm_lmsw);
700 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
702 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
703 !vcpu->guest_xcr0_loaded) {
704 /* kvm_set_xcr() also depends on this */
705 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
706 vcpu->guest_xcr0_loaded = 1;
710 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
712 if (vcpu->guest_xcr0_loaded) {
713 if (vcpu->arch.xcr0 != host_xcr0)
714 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
715 vcpu->guest_xcr0_loaded = 0;
719 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
722 u64 old_xcr0 = vcpu->arch.xcr0;
725 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
726 if (index != XCR_XFEATURE_ENABLED_MASK)
728 if (!(xcr0 & XFEATURE_MASK_FP))
730 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
734 * Do not allow the guest to set bits that we do not support
735 * saving. However, xcr0 bit 0 is always set, even if the
736 * emulated CPU does not support XSAVE (see fx_init).
738 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
739 if (xcr0 & ~valid_bits)
742 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
743 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
746 if (xcr0 & XFEATURE_MASK_AVX512) {
747 if (!(xcr0 & XFEATURE_MASK_YMM))
749 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
752 vcpu->arch.xcr0 = xcr0;
754 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
755 kvm_update_cpuid(vcpu);
759 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
761 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
762 __kvm_set_xcr(vcpu, index, xcr)) {
763 kvm_inject_gp(vcpu, 0);
768 EXPORT_SYMBOL_GPL(kvm_set_xcr);
770 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
772 unsigned long old_cr4 = kvm_read_cr4(vcpu);
773 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
774 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
776 if (cr4 & CR4_RESERVED_BITS)
779 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
782 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
785 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
788 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
791 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
794 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
797 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
800 if (is_long_mode(vcpu)) {
801 if (!(cr4 & X86_CR4_PAE))
803 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
804 && ((cr4 ^ old_cr4) & pdptr_bits)
805 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
809 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
810 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
813 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
814 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
818 if (kvm_x86_ops->set_cr4(vcpu, cr4))
821 if (((cr4 ^ old_cr4) & pdptr_bits) ||
822 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
823 kvm_mmu_reset_context(vcpu);
825 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
826 kvm_update_cpuid(vcpu);
830 EXPORT_SYMBOL_GPL(kvm_set_cr4);
832 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
835 cr3 &= ~CR3_PCID_INVD;
838 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
839 kvm_mmu_sync_roots(vcpu);
840 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
844 if (is_long_mode(vcpu) &&
845 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 62)))
847 else if (is_pae(vcpu) && is_paging(vcpu) &&
848 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
851 vcpu->arch.cr3 = cr3;
852 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
853 kvm_mmu_new_cr3(vcpu);
856 EXPORT_SYMBOL_GPL(kvm_set_cr3);
858 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
860 if (cr8 & CR8_RESERVED_BITS)
862 if (lapic_in_kernel(vcpu))
863 kvm_lapic_set_tpr(vcpu, cr8);
865 vcpu->arch.cr8 = cr8;
868 EXPORT_SYMBOL_GPL(kvm_set_cr8);
870 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
872 if (lapic_in_kernel(vcpu))
873 return kvm_lapic_get_cr8(vcpu);
875 return vcpu->arch.cr8;
877 EXPORT_SYMBOL_GPL(kvm_get_cr8);
879 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
883 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
884 for (i = 0; i < KVM_NR_DB_REGS; i++)
885 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
886 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
890 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
892 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
893 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
896 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
900 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
901 dr7 = vcpu->arch.guest_debug_dr7;
903 dr7 = vcpu->arch.dr7;
904 kvm_x86_ops->set_dr7(vcpu, dr7);
905 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
906 if (dr7 & DR7_BP_EN_MASK)
907 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
910 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
912 u64 fixed = DR6_FIXED_1;
914 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
919 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
923 vcpu->arch.db[dr] = val;
924 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
925 vcpu->arch.eff_db[dr] = val;
930 if (val & 0xffffffff00000000ULL)
932 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
933 kvm_update_dr6(vcpu);
938 if (val & 0xffffffff00000000ULL)
940 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
941 kvm_update_dr7(vcpu);
948 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
950 if (__kvm_set_dr(vcpu, dr, val)) {
951 kvm_inject_gp(vcpu, 0);
956 EXPORT_SYMBOL_GPL(kvm_set_dr);
958 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
962 *val = vcpu->arch.db[dr];
967 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
968 *val = vcpu->arch.dr6;
970 *val = kvm_x86_ops->get_dr6(vcpu);
975 *val = vcpu->arch.dr7;
980 EXPORT_SYMBOL_GPL(kvm_get_dr);
982 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
984 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
988 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
991 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
992 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
995 EXPORT_SYMBOL_GPL(kvm_rdpmc);
998 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
999 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1001 * This list is modified at module load time to reflect the
1002 * capabilities of the host cpu. This capabilities test skips MSRs that are
1003 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1004 * may depend on host virtualization features rather than host cpu features.
1007 static u32 msrs_to_save[] = {
1008 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1010 #ifdef CONFIG_X86_64
1011 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1013 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1014 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1017 static unsigned num_msrs_to_save;
1019 static u32 emulated_msrs[] = {
1020 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1021 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1022 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1023 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1024 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1025 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1026 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1028 HV_X64_MSR_VP_INDEX,
1029 HV_X64_MSR_VP_RUNTIME,
1030 HV_X64_MSR_SCONTROL,
1031 HV_X64_MSR_STIMER0_CONFIG,
1032 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1035 MSR_IA32_TSC_ADJUST,
1036 MSR_IA32_TSCDEADLINE,
1037 MSR_IA32_MISC_ENABLE,
1038 MSR_IA32_MCG_STATUS,
1040 MSR_IA32_MCG_EXT_CTL,
1044 MSR_MISC_FEATURES_ENABLES,
1047 static unsigned num_emulated_msrs;
1049 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1051 if (efer & efer_reserved_bits)
1054 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1057 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1062 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1064 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1066 u64 old_efer = vcpu->arch.efer;
1068 if (!kvm_valid_efer(vcpu, efer))
1072 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1076 efer |= vcpu->arch.efer & EFER_LMA;
1078 kvm_x86_ops->set_efer(vcpu, efer);
1080 /* Update reserved bits */
1081 if ((efer ^ old_efer) & EFER_NX)
1082 kvm_mmu_reset_context(vcpu);
1087 void kvm_enable_efer_bits(u64 mask)
1089 efer_reserved_bits &= ~mask;
1091 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1094 * Writes msr value into into the appropriate "register".
1095 * Returns 0 on success, non-0 otherwise.
1096 * Assumes vcpu_load() was already called.
1098 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1100 switch (msr->index) {
1103 case MSR_KERNEL_GS_BASE:
1106 if (is_noncanonical_address(msr->data, vcpu))
1109 case MSR_IA32_SYSENTER_EIP:
1110 case MSR_IA32_SYSENTER_ESP:
1112 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1113 * non-canonical address is written on Intel but not on
1114 * AMD (which ignores the top 32-bits, because it does
1115 * not implement 64-bit SYSENTER).
1117 * 64-bit code should hence be able to write a non-canonical
1118 * value on AMD. Making the address canonical ensures that
1119 * vmentry does not fail on Intel after writing a non-canonical
1120 * value, and that something deterministic happens if the guest
1121 * invokes 64-bit SYSENTER.
1123 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1125 return kvm_x86_ops->set_msr(vcpu, msr);
1127 EXPORT_SYMBOL_GPL(kvm_set_msr);
1130 * Adapt set_msr() to msr_io()'s calling convention
1132 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1134 struct msr_data msr;
1138 msr.host_initiated = true;
1139 r = kvm_get_msr(vcpu, &msr);
1147 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1149 struct msr_data msr;
1153 msr.host_initiated = true;
1154 return kvm_set_msr(vcpu, &msr);
1157 #ifdef CONFIG_X86_64
1158 struct pvclock_gtod_data {
1161 struct { /* extract of a clocksource struct */
1174 static struct pvclock_gtod_data pvclock_gtod_data;
1176 static void update_pvclock_gtod(struct timekeeper *tk)
1178 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1181 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1183 write_seqcount_begin(&vdata->seq);
1185 /* copy pvclock gtod data */
1186 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1187 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1188 vdata->clock.mask = tk->tkr_mono.mask;
1189 vdata->clock.mult = tk->tkr_mono.mult;
1190 vdata->clock.shift = tk->tkr_mono.shift;
1192 vdata->boot_ns = boot_ns;
1193 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1195 vdata->wall_time_sec = tk->xtime_sec;
1197 write_seqcount_end(&vdata->seq);
1201 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1204 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1205 * vcpu_enter_guest. This function is only called from
1206 * the physical CPU that is running vcpu.
1208 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1211 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1215 struct pvclock_wall_clock wc;
1216 struct timespec64 boot;
1221 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1226 ++version; /* first time write, random junk */
1230 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1234 * The guest calculates current wall clock time by adding
1235 * system time (updated by kvm_guest_time_update below) to the
1236 * wall clock specified here. guest system time equals host
1237 * system time for us, thus we must fill in host boot time here.
1239 getboottime64(&boot);
1241 if (kvm->arch.kvmclock_offset) {
1242 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1243 boot = timespec64_sub(boot, ts);
1245 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1246 wc.nsec = boot.tv_nsec;
1247 wc.version = version;
1249 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1252 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1255 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1257 do_shl32_div32(dividend, divisor);
1261 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1262 s8 *pshift, u32 *pmultiplier)
1270 scaled64 = scaled_hz;
1271 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1276 tps32 = (uint32_t)tps64;
1277 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1278 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1286 *pmultiplier = div_frac(scaled64, tps32);
1288 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1289 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1292 #ifdef CONFIG_X86_64
1293 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1296 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1297 static unsigned long max_tsc_khz;
1299 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1301 u64 v = (u64)khz * (1000000 + ppm);
1306 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1310 /* Guest TSC same frequency as host TSC? */
1312 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1316 /* TSC scaling supported? */
1317 if (!kvm_has_tsc_control) {
1318 if (user_tsc_khz > tsc_khz) {
1319 vcpu->arch.tsc_catchup = 1;
1320 vcpu->arch.tsc_always_catchup = 1;
1323 WARN(1, "user requested TSC rate below hardware speed\n");
1328 /* TSC scaling required - calculate ratio */
1329 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1330 user_tsc_khz, tsc_khz);
1332 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1333 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1338 vcpu->arch.tsc_scaling_ratio = ratio;
1342 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1344 u32 thresh_lo, thresh_hi;
1345 int use_scaling = 0;
1347 /* tsc_khz can be zero if TSC calibration fails */
1348 if (user_tsc_khz == 0) {
1349 /* set tsc_scaling_ratio to a safe value */
1350 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1354 /* Compute a scale to convert nanoseconds in TSC cycles */
1355 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1356 &vcpu->arch.virtual_tsc_shift,
1357 &vcpu->arch.virtual_tsc_mult);
1358 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1361 * Compute the variation in TSC rate which is acceptable
1362 * within the range of tolerance and decide if the
1363 * rate being applied is within that bounds of the hardware
1364 * rate. If so, no scaling or compensation need be done.
1366 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1367 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1368 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1369 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1372 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1375 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1377 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1378 vcpu->arch.virtual_tsc_mult,
1379 vcpu->arch.virtual_tsc_shift);
1380 tsc += vcpu->arch.this_tsc_write;
1384 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1386 #ifdef CONFIG_X86_64
1388 struct kvm_arch *ka = &vcpu->kvm->arch;
1389 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1391 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1392 atomic_read(&vcpu->kvm->online_vcpus));
1395 * Once the masterclock is enabled, always perform request in
1396 * order to update it.
1398 * In order to enable masterclock, the host clocksource must be TSC
1399 * and the vcpus need to have matched TSCs. When that happens,
1400 * perform request to enable masterclock.
1402 if (ka->use_master_clock ||
1403 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1404 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1406 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1407 atomic_read(&vcpu->kvm->online_vcpus),
1408 ka->use_master_clock, gtod->clock.vclock_mode);
1412 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1414 u64 curr_offset = vcpu->arch.tsc_offset;
1415 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1419 * Multiply tsc by a fixed point number represented by ratio.
1421 * The most significant 64-N bits (mult) of ratio represent the
1422 * integral part of the fixed point number; the remaining N bits
1423 * (frac) represent the fractional part, ie. ratio represents a fixed
1424 * point number (mult + frac * 2^(-N)).
1426 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1428 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1430 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1433 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1436 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1438 if (ratio != kvm_default_tsc_scaling_ratio)
1439 _tsc = __scale_tsc(ratio, tsc);
1443 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1445 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1449 tsc = kvm_scale_tsc(vcpu, rdtsc());
1451 return target_tsc - tsc;
1454 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1456 return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1458 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1460 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1462 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1463 vcpu->arch.tsc_offset = offset;
1466 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1468 struct kvm *kvm = vcpu->kvm;
1469 u64 offset, ns, elapsed;
1470 unsigned long flags;
1472 bool already_matched;
1473 u64 data = msr->data;
1474 bool synchronizing = false;
1476 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1477 offset = kvm_compute_tsc_offset(vcpu, data);
1478 ns = ktime_get_boot_ns();
1479 elapsed = ns - kvm->arch.last_tsc_nsec;
1481 if (vcpu->arch.virtual_tsc_khz) {
1482 if (data == 0 && msr->host_initiated) {
1484 * detection of vcpu initialization -- need to sync
1485 * with other vCPUs. This particularly helps to keep
1486 * kvm_clock stable after CPU hotplug
1488 synchronizing = true;
1490 u64 tsc_exp = kvm->arch.last_tsc_write +
1491 nsec_to_cycles(vcpu, elapsed);
1492 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1494 * Special case: TSC write with a small delta (1 second)
1495 * of virtual cycle time against real time is
1496 * interpreted as an attempt to synchronize the CPU.
1498 synchronizing = data < tsc_exp + tsc_hz &&
1499 data + tsc_hz > tsc_exp;
1504 * For a reliable TSC, we can match TSC offsets, and for an unstable
1505 * TSC, we add elapsed time in this computation. We could let the
1506 * compensation code attempt to catch up if we fall behind, but
1507 * it's better to try to match offsets from the beginning.
1509 if (synchronizing &&
1510 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1511 if (!check_tsc_unstable()) {
1512 offset = kvm->arch.cur_tsc_offset;
1513 pr_debug("kvm: matched tsc offset for %llu\n", data);
1515 u64 delta = nsec_to_cycles(vcpu, elapsed);
1517 offset = kvm_compute_tsc_offset(vcpu, data);
1518 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1521 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1524 * We split periods of matched TSC writes into generations.
1525 * For each generation, we track the original measured
1526 * nanosecond time, offset, and write, so if TSCs are in
1527 * sync, we can match exact offset, and if not, we can match
1528 * exact software computation in compute_guest_tsc()
1530 * These values are tracked in kvm->arch.cur_xxx variables.
1532 kvm->arch.cur_tsc_generation++;
1533 kvm->arch.cur_tsc_nsec = ns;
1534 kvm->arch.cur_tsc_write = data;
1535 kvm->arch.cur_tsc_offset = offset;
1537 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1538 kvm->arch.cur_tsc_generation, data);
1542 * We also track th most recent recorded KHZ, write and time to
1543 * allow the matching interval to be extended at each write.
1545 kvm->arch.last_tsc_nsec = ns;
1546 kvm->arch.last_tsc_write = data;
1547 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1549 vcpu->arch.last_guest_tsc = data;
1551 /* Keep track of which generation this VCPU has synchronized to */
1552 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1553 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1554 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1556 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1557 update_ia32_tsc_adjust_msr(vcpu, offset);
1559 kvm_vcpu_write_tsc_offset(vcpu, offset);
1560 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1562 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1564 kvm->arch.nr_vcpus_matched_tsc = 0;
1565 } else if (!already_matched) {
1566 kvm->arch.nr_vcpus_matched_tsc++;
1569 kvm_track_tsc_matching(vcpu);
1570 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1573 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1575 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1578 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1581 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1583 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1584 WARN_ON(adjustment < 0);
1585 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1586 adjust_tsc_offset_guest(vcpu, adjustment);
1589 #ifdef CONFIG_X86_64
1591 static u64 read_tsc(void)
1593 u64 ret = (u64)rdtsc_ordered();
1594 u64 last = pvclock_gtod_data.clock.cycle_last;
1596 if (likely(ret >= last))
1600 * GCC likes to generate cmov here, but this branch is extremely
1601 * predictable (it's just a function of time and the likely is
1602 * very likely) and there's a data dependence, so force GCC
1603 * to generate a branch instead. I don't barrier() because
1604 * we don't actually need a barrier, and if this function
1605 * ever gets inlined it will generate worse code.
1611 static inline u64 vgettsc(u64 *cycle_now)
1614 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1616 *cycle_now = read_tsc();
1618 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1619 return v * gtod->clock.mult;
1622 static int do_monotonic_boot(s64 *t, u64 *cycle_now)
1624 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1630 seq = read_seqcount_begin(>od->seq);
1631 mode = gtod->clock.vclock_mode;
1632 ns = gtod->nsec_base;
1633 ns += vgettsc(cycle_now);
1634 ns >>= gtod->clock.shift;
1635 ns += gtod->boot_ns;
1636 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1642 static int do_realtime(struct timespec *ts, u64 *cycle_now)
1644 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1650 seq = read_seqcount_begin(>od->seq);
1651 mode = gtod->clock.vclock_mode;
1652 ts->tv_sec = gtod->wall_time_sec;
1653 ns = gtod->nsec_base;
1654 ns += vgettsc(cycle_now);
1655 ns >>= gtod->clock.shift;
1656 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1658 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1664 /* returns true if host is using tsc clocksource */
1665 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
1667 /* checked again under seqlock below */
1668 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1671 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1674 /* returns true if host is using tsc clocksource */
1675 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1678 /* checked again under seqlock below */
1679 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1682 return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1688 * Assuming a stable TSC across physical CPUS, and a stable TSC
1689 * across virtual CPUs, the following condition is possible.
1690 * Each numbered line represents an event visible to both
1691 * CPUs at the next numbered event.
1693 * "timespecX" represents host monotonic time. "tscX" represents
1696 * VCPU0 on CPU0 | VCPU1 on CPU1
1698 * 1. read timespec0,tsc0
1699 * 2. | timespec1 = timespec0 + N
1701 * 3. transition to guest | transition to guest
1702 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1703 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1704 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1706 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1709 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1711 * - 0 < N - M => M < N
1713 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1714 * always the case (the difference between two distinct xtime instances
1715 * might be smaller then the difference between corresponding TSC reads,
1716 * when updating guest vcpus pvclock areas).
1718 * To avoid that problem, do not allow visibility of distinct
1719 * system_timestamp/tsc_timestamp values simultaneously: use a master
1720 * copy of host monotonic time values. Update that master copy
1723 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1727 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1729 #ifdef CONFIG_X86_64
1730 struct kvm_arch *ka = &kvm->arch;
1732 bool host_tsc_clocksource, vcpus_matched;
1734 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1735 atomic_read(&kvm->online_vcpus));
1738 * If the host uses TSC clock, then passthrough TSC as stable
1741 host_tsc_clocksource = kvm_get_time_and_clockread(
1742 &ka->master_kernel_ns,
1743 &ka->master_cycle_now);
1745 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1746 && !ka->backwards_tsc_observed
1747 && !ka->boot_vcpu_runs_old_kvmclock;
1749 if (ka->use_master_clock)
1750 atomic_set(&kvm_guest_has_master_clock, 1);
1752 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1753 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1758 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1760 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1763 static void kvm_gen_update_masterclock(struct kvm *kvm)
1765 #ifdef CONFIG_X86_64
1767 struct kvm_vcpu *vcpu;
1768 struct kvm_arch *ka = &kvm->arch;
1770 spin_lock(&ka->pvclock_gtod_sync_lock);
1771 kvm_make_mclock_inprogress_request(kvm);
1772 /* no guest entries from this point */
1773 pvclock_update_vm_gtod_copy(kvm);
1775 kvm_for_each_vcpu(i, vcpu, kvm)
1776 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1778 /* guest entries allowed */
1779 kvm_for_each_vcpu(i, vcpu, kvm)
1780 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1782 spin_unlock(&ka->pvclock_gtod_sync_lock);
1786 u64 get_kvmclock_ns(struct kvm *kvm)
1788 struct kvm_arch *ka = &kvm->arch;
1789 struct pvclock_vcpu_time_info hv_clock;
1792 spin_lock(&ka->pvclock_gtod_sync_lock);
1793 if (!ka->use_master_clock) {
1794 spin_unlock(&ka->pvclock_gtod_sync_lock);
1795 return ktime_get_boot_ns() + ka->kvmclock_offset;
1798 hv_clock.tsc_timestamp = ka->master_cycle_now;
1799 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1800 spin_unlock(&ka->pvclock_gtod_sync_lock);
1802 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1805 if (__this_cpu_read(cpu_tsc_khz)) {
1806 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1807 &hv_clock.tsc_shift,
1808 &hv_clock.tsc_to_system_mul);
1809 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1811 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
1818 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1820 struct kvm_vcpu_arch *vcpu = &v->arch;
1821 struct pvclock_vcpu_time_info guest_hv_clock;
1823 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1824 &guest_hv_clock, sizeof(guest_hv_clock))))
1827 /* This VCPU is paused, but it's legal for a guest to read another
1828 * VCPU's kvmclock, so we really have to follow the specification where
1829 * it says that version is odd if data is being modified, and even after
1832 * Version field updates must be kept separate. This is because
1833 * kvm_write_guest_cached might use a "rep movs" instruction, and
1834 * writes within a string instruction are weakly ordered. So there
1835 * are three writes overall.
1837 * As a small optimization, only write the version field in the first
1838 * and third write. The vcpu->pv_time cache is still valid, because the
1839 * version field is the first in the struct.
1841 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1843 if (guest_hv_clock.version & 1)
1844 ++guest_hv_clock.version; /* first time write, random junk */
1846 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1847 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1849 sizeof(vcpu->hv_clock.version));
1853 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1854 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1856 if (vcpu->pvclock_set_guest_stopped_request) {
1857 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1858 vcpu->pvclock_set_guest_stopped_request = false;
1861 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1863 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1865 sizeof(vcpu->hv_clock));
1869 vcpu->hv_clock.version++;
1870 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1872 sizeof(vcpu->hv_clock.version));
1875 static int kvm_guest_time_update(struct kvm_vcpu *v)
1877 unsigned long flags, tgt_tsc_khz;
1878 struct kvm_vcpu_arch *vcpu = &v->arch;
1879 struct kvm_arch *ka = &v->kvm->arch;
1881 u64 tsc_timestamp, host_tsc;
1883 bool use_master_clock;
1889 * If the host uses TSC clock, then passthrough TSC as stable
1892 spin_lock(&ka->pvclock_gtod_sync_lock);
1893 use_master_clock = ka->use_master_clock;
1894 if (use_master_clock) {
1895 host_tsc = ka->master_cycle_now;
1896 kernel_ns = ka->master_kernel_ns;
1898 spin_unlock(&ka->pvclock_gtod_sync_lock);
1900 /* Keep irq disabled to prevent changes to the clock */
1901 local_irq_save(flags);
1902 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1903 if (unlikely(tgt_tsc_khz == 0)) {
1904 local_irq_restore(flags);
1905 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1908 if (!use_master_clock) {
1910 kernel_ns = ktime_get_boot_ns();
1913 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1916 * We may have to catch up the TSC to match elapsed wall clock
1917 * time for two reasons, even if kvmclock is used.
1918 * 1) CPU could have been running below the maximum TSC rate
1919 * 2) Broken TSC compensation resets the base at each VCPU
1920 * entry to avoid unknown leaps of TSC even when running
1921 * again on the same CPU. This may cause apparent elapsed
1922 * time to disappear, and the guest to stand still or run
1925 if (vcpu->tsc_catchup) {
1926 u64 tsc = compute_guest_tsc(v, kernel_ns);
1927 if (tsc > tsc_timestamp) {
1928 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1929 tsc_timestamp = tsc;
1933 local_irq_restore(flags);
1935 /* With all the info we got, fill in the values */
1937 if (kvm_has_tsc_control)
1938 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1940 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1941 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1942 &vcpu->hv_clock.tsc_shift,
1943 &vcpu->hv_clock.tsc_to_system_mul);
1944 vcpu->hw_tsc_khz = tgt_tsc_khz;
1947 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1948 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1949 vcpu->last_guest_tsc = tsc_timestamp;
1951 /* If the host uses TSC clocksource, then it is stable */
1953 if (use_master_clock)
1954 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1956 vcpu->hv_clock.flags = pvclock_flags;
1958 if (vcpu->pv_time_enabled)
1959 kvm_setup_pvclock_page(v);
1960 if (v == kvm_get_vcpu(v->kvm, 0))
1961 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
1966 * kvmclock updates which are isolated to a given vcpu, such as
1967 * vcpu->cpu migration, should not allow system_timestamp from
1968 * the rest of the vcpus to remain static. Otherwise ntp frequency
1969 * correction applies to one vcpu's system_timestamp but not
1972 * So in those cases, request a kvmclock update for all vcpus.
1973 * We need to rate-limit these requests though, as they can
1974 * considerably slow guests that have a large number of vcpus.
1975 * The time for a remote vcpu to update its kvmclock is bound
1976 * by the delay we use to rate-limit the updates.
1979 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1981 static void kvmclock_update_fn(struct work_struct *work)
1984 struct delayed_work *dwork = to_delayed_work(work);
1985 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1986 kvmclock_update_work);
1987 struct kvm *kvm = container_of(ka, struct kvm, arch);
1988 struct kvm_vcpu *vcpu;
1990 kvm_for_each_vcpu(i, vcpu, kvm) {
1991 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1992 kvm_vcpu_kick(vcpu);
1996 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1998 struct kvm *kvm = v->kvm;
2000 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2001 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2002 KVMCLOCK_UPDATE_DELAY);
2005 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2007 static void kvmclock_sync_fn(struct work_struct *work)
2009 struct delayed_work *dwork = to_delayed_work(work);
2010 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2011 kvmclock_sync_work);
2012 struct kvm *kvm = container_of(ka, struct kvm, arch);
2014 if (!kvmclock_periodic_sync)
2017 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2018 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2019 KVMCLOCK_SYNC_PERIOD);
2022 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2024 u64 mcg_cap = vcpu->arch.mcg_cap;
2025 unsigned bank_num = mcg_cap & 0xff;
2026 u32 msr = msr_info->index;
2027 u64 data = msr_info->data;
2030 case MSR_IA32_MCG_STATUS:
2031 vcpu->arch.mcg_status = data;
2033 case MSR_IA32_MCG_CTL:
2034 if (!(mcg_cap & MCG_CTL_P))
2036 if (data != 0 && data != ~(u64)0)
2038 vcpu->arch.mcg_ctl = data;
2041 if (msr >= MSR_IA32_MC0_CTL &&
2042 msr < MSR_IA32_MCx_CTL(bank_num)) {
2043 u32 offset = msr - MSR_IA32_MC0_CTL;
2044 /* only 0 or all 1s can be written to IA32_MCi_CTL
2045 * some Linux kernels though clear bit 10 in bank 4 to
2046 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2047 * this to avoid an uncatched #GP in the guest
2049 if ((offset & 0x3) == 0 &&
2050 data != 0 && (data | (1 << 10)) != ~(u64)0)
2052 if (!msr_info->host_initiated &&
2053 (offset & 0x3) == 1 && data != 0)
2055 vcpu->arch.mce_banks[offset] = data;
2063 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2065 struct kvm *kvm = vcpu->kvm;
2066 int lm = is_long_mode(vcpu);
2067 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2068 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2069 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2070 : kvm->arch.xen_hvm_config.blob_size_32;
2071 u32 page_num = data & ~PAGE_MASK;
2072 u64 page_addr = data & PAGE_MASK;
2077 if (page_num >= blob_size)
2080 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2085 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2094 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2096 gpa_t gpa = data & ~0x3f;
2098 /* Bits 3:5 are reserved, Should be zero */
2102 vcpu->arch.apf.msr_val = data;
2104 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2105 kvm_clear_async_pf_completion_queue(vcpu);
2106 kvm_async_pf_hash_reset(vcpu);
2110 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2114 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2115 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2116 kvm_async_pf_wakeup_all(vcpu);
2120 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2122 vcpu->arch.pv_time_enabled = false;
2125 static void record_steal_time(struct kvm_vcpu *vcpu)
2127 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2130 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2131 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2134 vcpu->arch.st.steal.preempted = 0;
2136 if (vcpu->arch.st.steal.version & 1)
2137 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2139 vcpu->arch.st.steal.version += 1;
2141 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2142 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2146 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2147 vcpu->arch.st.last_steal;
2148 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2150 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2151 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2155 vcpu->arch.st.steal.version += 1;
2157 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2158 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2161 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2164 u32 msr = msr_info->index;
2165 u64 data = msr_info->data;
2168 case MSR_AMD64_NB_CFG:
2169 case MSR_IA32_UCODE_REV:
2170 case MSR_IA32_UCODE_WRITE:
2171 case MSR_VM_HSAVE_PA:
2172 case MSR_AMD64_PATCH_LOADER:
2173 case MSR_AMD64_BU_CFG2:
2174 case MSR_AMD64_DC_CFG:
2178 return set_efer(vcpu, data);
2180 data &= ~(u64)0x40; /* ignore flush filter disable */
2181 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2182 data &= ~(u64)0x8; /* ignore TLB cache disable */
2183 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2185 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2190 case MSR_FAM10H_MMIO_CONF_BASE:
2192 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2197 case MSR_IA32_DEBUGCTLMSR:
2199 /* We support the non-activated case already */
2201 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2202 /* Values other than LBR and BTF are vendor-specific,
2203 thus reserved and should throw a #GP */
2206 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2209 case 0x200 ... 0x2ff:
2210 return kvm_mtrr_set_msr(vcpu, msr, data);
2211 case MSR_IA32_APICBASE:
2212 return kvm_set_apic_base(vcpu, msr_info);
2213 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2214 return kvm_x2apic_msr_write(vcpu, msr, data);
2215 case MSR_IA32_TSCDEADLINE:
2216 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2218 case MSR_IA32_TSC_ADJUST:
2219 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2220 if (!msr_info->host_initiated) {
2221 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2222 adjust_tsc_offset_guest(vcpu, adj);
2224 vcpu->arch.ia32_tsc_adjust_msr = data;
2227 case MSR_IA32_MISC_ENABLE:
2228 vcpu->arch.ia32_misc_enable_msr = data;
2230 case MSR_IA32_SMBASE:
2231 if (!msr_info->host_initiated)
2233 vcpu->arch.smbase = data;
2236 if (!msr_info->host_initiated)
2238 vcpu->arch.smi_count = data;
2240 case MSR_KVM_WALL_CLOCK_NEW:
2241 case MSR_KVM_WALL_CLOCK:
2242 vcpu->kvm->arch.wall_clock = data;
2243 kvm_write_wall_clock(vcpu->kvm, data);
2245 case MSR_KVM_SYSTEM_TIME_NEW:
2246 case MSR_KVM_SYSTEM_TIME: {
2247 struct kvm_arch *ka = &vcpu->kvm->arch;
2249 kvmclock_reset(vcpu);
2251 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2252 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2254 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2255 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2257 ka->boot_vcpu_runs_old_kvmclock = tmp;
2260 vcpu->arch.time = data;
2261 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2263 /* we verify if the enable bit is set... */
2267 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2268 &vcpu->arch.pv_time, data & ~1ULL,
2269 sizeof(struct pvclock_vcpu_time_info)))
2270 vcpu->arch.pv_time_enabled = false;
2272 vcpu->arch.pv_time_enabled = true;
2276 case MSR_KVM_ASYNC_PF_EN:
2277 if (kvm_pv_enable_async_pf(vcpu, data))
2280 case MSR_KVM_STEAL_TIME:
2282 if (unlikely(!sched_info_on()))
2285 if (data & KVM_STEAL_RESERVED_MASK)
2288 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2289 data & KVM_STEAL_VALID_BITS,
2290 sizeof(struct kvm_steal_time)))
2293 vcpu->arch.st.msr_val = data;
2295 if (!(data & KVM_MSR_ENABLED))
2298 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2301 case MSR_KVM_PV_EOI_EN:
2302 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2306 case MSR_IA32_MCG_CTL:
2307 case MSR_IA32_MCG_STATUS:
2308 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2309 return set_msr_mce(vcpu, msr_info);
2311 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2312 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2313 pr = true; /* fall through */
2314 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2315 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2316 if (kvm_pmu_is_valid_msr(vcpu, msr))
2317 return kvm_pmu_set_msr(vcpu, msr_info);
2319 if (pr || data != 0)
2320 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2321 "0x%x data 0x%llx\n", msr, data);
2323 case MSR_K7_CLK_CTL:
2325 * Ignore all writes to this no longer documented MSR.
2326 * Writes are only relevant for old K7 processors,
2327 * all pre-dating SVM, but a recommended workaround from
2328 * AMD for these chips. It is possible to specify the
2329 * affected processor models on the command line, hence
2330 * the need to ignore the workaround.
2333 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2334 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2335 case HV_X64_MSR_CRASH_CTL:
2336 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2337 return kvm_hv_set_msr_common(vcpu, msr, data,
2338 msr_info->host_initiated);
2339 case MSR_IA32_BBL_CR_CTL3:
2340 /* Drop writes to this legacy MSR -- see rdmsr
2341 * counterpart for further detail.
2343 if (report_ignored_msrs)
2344 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2347 case MSR_AMD64_OSVW_ID_LENGTH:
2348 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2350 vcpu->arch.osvw.length = data;
2352 case MSR_AMD64_OSVW_STATUS:
2353 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2355 vcpu->arch.osvw.status = data;
2357 case MSR_PLATFORM_INFO:
2358 if (!msr_info->host_initiated ||
2359 data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2360 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2361 cpuid_fault_enabled(vcpu)))
2363 vcpu->arch.msr_platform_info = data;
2365 case MSR_MISC_FEATURES_ENABLES:
2366 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2367 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2368 !supports_cpuid_fault(vcpu)))
2370 vcpu->arch.msr_misc_features_enables = data;
2373 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2374 return xen_hvm_config(vcpu, data);
2375 if (kvm_pmu_is_valid_msr(vcpu, msr))
2376 return kvm_pmu_set_msr(vcpu, msr_info);
2378 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2382 if (report_ignored_msrs)
2384 "ignored wrmsr: 0x%x data 0x%llx\n",
2391 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2395 * Reads an msr value (of 'msr_index') into 'pdata'.
2396 * Returns 0 on success, non-0 otherwise.
2397 * Assumes vcpu_load() was already called.
2399 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2401 return kvm_x86_ops->get_msr(vcpu, msr);
2403 EXPORT_SYMBOL_GPL(kvm_get_msr);
2405 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2408 u64 mcg_cap = vcpu->arch.mcg_cap;
2409 unsigned bank_num = mcg_cap & 0xff;
2412 case MSR_IA32_P5_MC_ADDR:
2413 case MSR_IA32_P5_MC_TYPE:
2416 case MSR_IA32_MCG_CAP:
2417 data = vcpu->arch.mcg_cap;
2419 case MSR_IA32_MCG_CTL:
2420 if (!(mcg_cap & MCG_CTL_P))
2422 data = vcpu->arch.mcg_ctl;
2424 case MSR_IA32_MCG_STATUS:
2425 data = vcpu->arch.mcg_status;
2428 if (msr >= MSR_IA32_MC0_CTL &&
2429 msr < MSR_IA32_MCx_CTL(bank_num)) {
2430 u32 offset = msr - MSR_IA32_MC0_CTL;
2431 data = vcpu->arch.mce_banks[offset];
2440 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2442 switch (msr_info->index) {
2443 case MSR_IA32_PLATFORM_ID:
2444 case MSR_IA32_EBL_CR_POWERON:
2445 case MSR_IA32_DEBUGCTLMSR:
2446 case MSR_IA32_LASTBRANCHFROMIP:
2447 case MSR_IA32_LASTBRANCHTOIP:
2448 case MSR_IA32_LASTINTFROMIP:
2449 case MSR_IA32_LASTINTTOIP:
2451 case MSR_K8_TSEG_ADDR:
2452 case MSR_K8_TSEG_MASK:
2454 case MSR_VM_HSAVE_PA:
2455 case MSR_K8_INT_PENDING_MSG:
2456 case MSR_AMD64_NB_CFG:
2457 case MSR_FAM10H_MMIO_CONF_BASE:
2458 case MSR_AMD64_BU_CFG2:
2459 case MSR_IA32_PERF_CTL:
2460 case MSR_AMD64_DC_CFG:
2463 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2464 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2465 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2466 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2467 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2468 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2471 case MSR_IA32_UCODE_REV:
2472 msr_info->data = 0x100000000ULL;
2475 case 0x200 ... 0x2ff:
2476 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2477 case 0xcd: /* fsb frequency */
2481 * MSR_EBC_FREQUENCY_ID
2482 * Conservative value valid for even the basic CPU models.
2483 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2484 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2485 * and 266MHz for model 3, or 4. Set Core Clock
2486 * Frequency to System Bus Frequency Ratio to 1 (bits
2487 * 31:24) even though these are only valid for CPU
2488 * models > 2, however guests may end up dividing or
2489 * multiplying by zero otherwise.
2491 case MSR_EBC_FREQUENCY_ID:
2492 msr_info->data = 1 << 24;
2494 case MSR_IA32_APICBASE:
2495 msr_info->data = kvm_get_apic_base(vcpu);
2497 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2498 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2500 case MSR_IA32_TSCDEADLINE:
2501 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2503 case MSR_IA32_TSC_ADJUST:
2504 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2506 case MSR_IA32_MISC_ENABLE:
2507 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2509 case MSR_IA32_SMBASE:
2510 if (!msr_info->host_initiated)
2512 msr_info->data = vcpu->arch.smbase;
2515 msr_info->data = vcpu->arch.smi_count;
2517 case MSR_IA32_PERF_STATUS:
2518 /* TSC increment by tick */
2519 msr_info->data = 1000ULL;
2520 /* CPU multiplier */
2521 msr_info->data |= (((uint64_t)4ULL) << 40);
2524 msr_info->data = vcpu->arch.efer;
2526 case MSR_KVM_WALL_CLOCK:
2527 case MSR_KVM_WALL_CLOCK_NEW:
2528 msr_info->data = vcpu->kvm->arch.wall_clock;
2530 case MSR_KVM_SYSTEM_TIME:
2531 case MSR_KVM_SYSTEM_TIME_NEW:
2532 msr_info->data = vcpu->arch.time;
2534 case MSR_KVM_ASYNC_PF_EN:
2535 msr_info->data = vcpu->arch.apf.msr_val;
2537 case MSR_KVM_STEAL_TIME:
2538 msr_info->data = vcpu->arch.st.msr_val;
2540 case MSR_KVM_PV_EOI_EN:
2541 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2543 case MSR_IA32_P5_MC_ADDR:
2544 case MSR_IA32_P5_MC_TYPE:
2545 case MSR_IA32_MCG_CAP:
2546 case MSR_IA32_MCG_CTL:
2547 case MSR_IA32_MCG_STATUS:
2548 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2549 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2550 case MSR_K7_CLK_CTL:
2552 * Provide expected ramp-up count for K7. All other
2553 * are set to zero, indicating minimum divisors for
2556 * This prevents guest kernels on AMD host with CPU
2557 * type 6, model 8 and higher from exploding due to
2558 * the rdmsr failing.
2560 msr_info->data = 0x20000000;
2562 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2563 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2564 case HV_X64_MSR_CRASH_CTL:
2565 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2566 return kvm_hv_get_msr_common(vcpu,
2567 msr_info->index, &msr_info->data);
2569 case MSR_IA32_BBL_CR_CTL3:
2570 /* This legacy MSR exists but isn't fully documented in current
2571 * silicon. It is however accessed by winxp in very narrow
2572 * scenarios where it sets bit #19, itself documented as
2573 * a "reserved" bit. Best effort attempt to source coherent
2574 * read data here should the balance of the register be
2575 * interpreted by the guest:
2577 * L2 cache control register 3: 64GB range, 256KB size,
2578 * enabled, latency 0x1, configured
2580 msr_info->data = 0xbe702111;
2582 case MSR_AMD64_OSVW_ID_LENGTH:
2583 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2585 msr_info->data = vcpu->arch.osvw.length;
2587 case MSR_AMD64_OSVW_STATUS:
2588 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2590 msr_info->data = vcpu->arch.osvw.status;
2592 case MSR_PLATFORM_INFO:
2593 msr_info->data = vcpu->arch.msr_platform_info;
2595 case MSR_MISC_FEATURES_ENABLES:
2596 msr_info->data = vcpu->arch.msr_misc_features_enables;
2599 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2600 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2602 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2606 if (report_ignored_msrs)
2607 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2615 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2618 * Read or write a bunch of msrs. All parameters are kernel addresses.
2620 * @return number of msrs set successfully.
2622 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2623 struct kvm_msr_entry *entries,
2624 int (*do_msr)(struct kvm_vcpu *vcpu,
2625 unsigned index, u64 *data))
2629 idx = srcu_read_lock(&vcpu->kvm->srcu);
2630 for (i = 0; i < msrs->nmsrs; ++i)
2631 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2633 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2639 * Read or write a bunch of msrs. Parameters are user addresses.
2641 * @return number of msrs set successfully.
2643 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2644 int (*do_msr)(struct kvm_vcpu *vcpu,
2645 unsigned index, u64 *data),
2648 struct kvm_msrs msrs;
2649 struct kvm_msr_entry *entries;
2654 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2658 if (msrs.nmsrs >= MAX_IO_MSRS)
2661 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2662 entries = memdup_user(user_msrs->entries, size);
2663 if (IS_ERR(entries)) {
2664 r = PTR_ERR(entries);
2668 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2673 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2684 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2689 case KVM_CAP_IRQCHIP:
2691 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2692 case KVM_CAP_SET_TSS_ADDR:
2693 case KVM_CAP_EXT_CPUID:
2694 case KVM_CAP_EXT_EMUL_CPUID:
2695 case KVM_CAP_CLOCKSOURCE:
2697 case KVM_CAP_NOP_IO_DELAY:
2698 case KVM_CAP_MP_STATE:
2699 case KVM_CAP_SYNC_MMU:
2700 case KVM_CAP_USER_NMI:
2701 case KVM_CAP_REINJECT_CONTROL:
2702 case KVM_CAP_IRQ_INJECT_STATUS:
2703 case KVM_CAP_IOEVENTFD:
2704 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2706 case KVM_CAP_PIT_STATE2:
2707 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2708 case KVM_CAP_XEN_HVM:
2709 case KVM_CAP_VCPU_EVENTS:
2710 case KVM_CAP_HYPERV:
2711 case KVM_CAP_HYPERV_VAPIC:
2712 case KVM_CAP_HYPERV_SPIN:
2713 case KVM_CAP_HYPERV_SYNIC:
2714 case KVM_CAP_HYPERV_SYNIC2:
2715 case KVM_CAP_HYPERV_VP_INDEX:
2716 case KVM_CAP_PCI_SEGMENT:
2717 case KVM_CAP_DEBUGREGS:
2718 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2720 case KVM_CAP_ASYNC_PF:
2721 case KVM_CAP_GET_TSC_KHZ:
2722 case KVM_CAP_KVMCLOCK_CTRL:
2723 case KVM_CAP_READONLY_MEM:
2724 case KVM_CAP_HYPERV_TIME:
2725 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2726 case KVM_CAP_TSC_DEADLINE_TIMER:
2727 case KVM_CAP_ENABLE_CAP_VM:
2728 case KVM_CAP_DISABLE_QUIRKS:
2729 case KVM_CAP_SET_BOOT_CPU_ID:
2730 case KVM_CAP_SPLIT_IRQCHIP:
2731 case KVM_CAP_IMMEDIATE_EXIT:
2734 case KVM_CAP_ADJUST_CLOCK:
2735 r = KVM_CLOCK_TSC_STABLE;
2737 case KVM_CAP_X86_GUEST_MWAIT:
2738 r = kvm_mwait_in_guest();
2740 case KVM_CAP_X86_SMM:
2741 /* SMBASE is usually relocated above 1M on modern chipsets,
2742 * and SMM handlers might indeed rely on 4G segment limits,
2743 * so do not report SMM to be available if real mode is
2744 * emulated via vm86 mode. Still, do not go to great lengths
2745 * to avoid userspace's usage of the feature, because it is a
2746 * fringe case that is not enabled except via specific settings
2747 * of the module parameters.
2749 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2752 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2754 case KVM_CAP_NR_VCPUS:
2755 r = KVM_SOFT_MAX_VCPUS;
2757 case KVM_CAP_MAX_VCPUS:
2760 case KVM_CAP_NR_MEMSLOTS:
2761 r = KVM_USER_MEM_SLOTS;
2763 case KVM_CAP_PV_MMU: /* obsolete */
2767 r = KVM_MAX_MCE_BANKS;
2770 r = boot_cpu_has(X86_FEATURE_XSAVE);
2772 case KVM_CAP_TSC_CONTROL:
2773 r = kvm_has_tsc_control;
2775 case KVM_CAP_X2APIC_API:
2776 r = KVM_X2APIC_API_VALID_FLAGS;
2786 long kvm_arch_dev_ioctl(struct file *filp,
2787 unsigned int ioctl, unsigned long arg)
2789 void __user *argp = (void __user *)arg;
2793 case KVM_GET_MSR_INDEX_LIST: {
2794 struct kvm_msr_list __user *user_msr_list = argp;
2795 struct kvm_msr_list msr_list;
2799 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2802 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2803 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2806 if (n < msr_list.nmsrs)
2809 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2810 num_msrs_to_save * sizeof(u32)))
2812 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2814 num_emulated_msrs * sizeof(u32)))
2819 case KVM_GET_SUPPORTED_CPUID:
2820 case KVM_GET_EMULATED_CPUID: {
2821 struct kvm_cpuid2 __user *cpuid_arg = argp;
2822 struct kvm_cpuid2 cpuid;
2825 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2828 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2834 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2839 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2841 if (copy_to_user(argp, &kvm_mce_cap_supported,
2842 sizeof(kvm_mce_cap_supported)))
2854 static void wbinvd_ipi(void *garbage)
2859 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2861 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2864 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2866 /* Address WBINVD may be executed by guest */
2867 if (need_emulate_wbinvd(vcpu)) {
2868 if (kvm_x86_ops->has_wbinvd_exit())
2869 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2870 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2871 smp_call_function_single(vcpu->cpu,
2872 wbinvd_ipi, NULL, 1);
2875 kvm_x86_ops->vcpu_load(vcpu, cpu);
2877 /* Apply any externally detected TSC adjustments (due to suspend) */
2878 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2879 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2880 vcpu->arch.tsc_offset_adjustment = 0;
2881 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2884 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2885 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2886 rdtsc() - vcpu->arch.last_host_tsc;
2888 mark_tsc_unstable("KVM discovered backwards TSC");
2890 if (check_tsc_unstable()) {
2891 u64 offset = kvm_compute_tsc_offset(vcpu,
2892 vcpu->arch.last_guest_tsc);
2893 kvm_vcpu_write_tsc_offset(vcpu, offset);
2894 vcpu->arch.tsc_catchup = 1;
2897 if (kvm_lapic_hv_timer_in_use(vcpu))
2898 kvm_lapic_restart_hv_timer(vcpu);
2901 * On a host with synchronized TSC, there is no need to update
2902 * kvmclock on vcpu->cpu migration
2904 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2905 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2906 if (vcpu->cpu != cpu)
2907 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
2911 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2914 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2916 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2919 vcpu->arch.st.steal.preempted = 1;
2921 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
2922 &vcpu->arch.st.steal.preempted,
2923 offsetof(struct kvm_steal_time, preempted),
2924 sizeof(vcpu->arch.st.steal.preempted));
2927 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2931 if (vcpu->preempted)
2932 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
2935 * Disable page faults because we're in atomic context here.
2936 * kvm_write_guest_offset_cached() would call might_fault()
2937 * that relies on pagefault_disable() to tell if there's a
2938 * bug. NOTE: the write to guest memory may not go through if
2939 * during postcopy live migration or if there's heavy guest
2942 pagefault_disable();
2944 * kvm_memslots() will be called by
2945 * kvm_write_guest_offset_cached() so take the srcu lock.
2947 idx = srcu_read_lock(&vcpu->kvm->srcu);
2948 kvm_steal_time_set_preempted(vcpu);
2949 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2951 kvm_x86_ops->vcpu_put(vcpu);
2952 vcpu->arch.last_host_tsc = rdtsc();
2955 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2956 struct kvm_lapic_state *s)
2958 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
2959 kvm_x86_ops->sync_pir_to_irr(vcpu);
2961 return kvm_apic_get_state(vcpu, s);
2964 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2965 struct kvm_lapic_state *s)
2969 r = kvm_apic_set_state(vcpu, s);
2972 update_cr8_intercept(vcpu);
2977 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2979 return (!lapic_in_kernel(vcpu) ||
2980 kvm_apic_accept_pic_intr(vcpu));
2984 * if userspace requested an interrupt window, check that the
2985 * interrupt window is open.
2987 * No need to exit to userspace if we already have an interrupt queued.
2989 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2991 return kvm_arch_interrupt_allowed(vcpu) &&
2992 !kvm_cpu_has_interrupt(vcpu) &&
2993 !kvm_event_needs_reinjection(vcpu) &&
2994 kvm_cpu_accept_dm_intr(vcpu);
2997 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2998 struct kvm_interrupt *irq)
3000 if (irq->irq >= KVM_NR_INTERRUPTS)
3003 if (!irqchip_in_kernel(vcpu->kvm)) {
3004 kvm_queue_interrupt(vcpu, irq->irq, false);
3005 kvm_make_request(KVM_REQ_EVENT, vcpu);
3010 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3011 * fail for in-kernel 8259.
3013 if (pic_in_kernel(vcpu->kvm))
3016 if (vcpu->arch.pending_external_vector != -1)
3019 vcpu->arch.pending_external_vector = irq->irq;
3020 kvm_make_request(KVM_REQ_EVENT, vcpu);
3024 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3026 kvm_inject_nmi(vcpu);
3031 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3033 kvm_make_request(KVM_REQ_SMI, vcpu);
3038 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3039 struct kvm_tpr_access_ctl *tac)
3043 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3047 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3051 unsigned bank_num = mcg_cap & 0xff, bank;
3054 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3056 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3059 vcpu->arch.mcg_cap = mcg_cap;
3060 /* Init IA32_MCG_CTL to all 1s */
3061 if (mcg_cap & MCG_CTL_P)
3062 vcpu->arch.mcg_ctl = ~(u64)0;
3063 /* Init IA32_MCi_CTL to all 1s */
3064 for (bank = 0; bank < bank_num; bank++)
3065 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3067 if (kvm_x86_ops->setup_mce)
3068 kvm_x86_ops->setup_mce(vcpu);
3073 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3074 struct kvm_x86_mce *mce)
3076 u64 mcg_cap = vcpu->arch.mcg_cap;
3077 unsigned bank_num = mcg_cap & 0xff;
3078 u64 *banks = vcpu->arch.mce_banks;
3080 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3083 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3084 * reporting is disabled
3086 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3087 vcpu->arch.mcg_ctl != ~(u64)0)
3089 banks += 4 * mce->bank;
3091 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3092 * reporting is disabled for the bank
3094 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3096 if (mce->status & MCI_STATUS_UC) {
3097 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3098 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3099 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3102 if (banks[1] & MCI_STATUS_VAL)
3103 mce->status |= MCI_STATUS_OVER;
3104 banks[2] = mce->addr;
3105 banks[3] = mce->misc;
3106 vcpu->arch.mcg_status = mce->mcg_status;
3107 banks[1] = mce->status;
3108 kvm_queue_exception(vcpu, MC_VECTOR);
3109 } else if (!(banks[1] & MCI_STATUS_VAL)
3110 || !(banks[1] & MCI_STATUS_UC)) {
3111 if (banks[1] & MCI_STATUS_VAL)
3112 mce->status |= MCI_STATUS_OVER;
3113 banks[2] = mce->addr;
3114 banks[3] = mce->misc;
3115 banks[1] = mce->status;
3117 banks[1] |= MCI_STATUS_OVER;
3121 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3122 struct kvm_vcpu_events *events)
3126 * FIXME: pass injected and pending separately. This is only
3127 * needed for nested virtualization, whose state cannot be
3128 * migrated yet. For now we can combine them.
3130 events->exception.injected =
3131 (vcpu->arch.exception.pending ||
3132 vcpu->arch.exception.injected) &&
3133 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3134 events->exception.nr = vcpu->arch.exception.nr;
3135 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3136 events->exception.pad = 0;
3137 events->exception.error_code = vcpu->arch.exception.error_code;
3139 events->interrupt.injected =
3140 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3141 events->interrupt.nr = vcpu->arch.interrupt.nr;
3142 events->interrupt.soft = 0;
3143 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3145 events->nmi.injected = vcpu->arch.nmi_injected;
3146 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3147 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3148 events->nmi.pad = 0;
3150 events->sipi_vector = 0; /* never valid when reporting to user space */
3152 events->smi.smm = is_smm(vcpu);
3153 events->smi.pending = vcpu->arch.smi_pending;
3154 events->smi.smm_inside_nmi =
3155 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3156 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3158 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3159 | KVM_VCPUEVENT_VALID_SHADOW
3160 | KVM_VCPUEVENT_VALID_SMM);
3161 memset(&events->reserved, 0, sizeof(events->reserved));
3164 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3166 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3167 struct kvm_vcpu_events *events)
3169 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3170 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3171 | KVM_VCPUEVENT_VALID_SHADOW
3172 | KVM_VCPUEVENT_VALID_SMM))
3175 if (events->exception.injected &&
3176 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3177 is_guest_mode(vcpu)))
3180 /* INITs are latched while in SMM */
3181 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3182 (events->smi.smm || events->smi.pending) &&
3183 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3187 vcpu->arch.exception.injected = false;
3188 vcpu->arch.exception.pending = events->exception.injected;
3189 vcpu->arch.exception.nr = events->exception.nr;
3190 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3191 vcpu->arch.exception.error_code = events->exception.error_code;
3193 vcpu->arch.interrupt.pending = events->interrupt.injected;
3194 vcpu->arch.interrupt.nr = events->interrupt.nr;
3195 vcpu->arch.interrupt.soft = events->interrupt.soft;
3196 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3197 kvm_x86_ops->set_interrupt_shadow(vcpu,
3198 events->interrupt.shadow);
3200 vcpu->arch.nmi_injected = events->nmi.injected;
3201 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3202 vcpu->arch.nmi_pending = events->nmi.pending;
3203 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3205 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3206 lapic_in_kernel(vcpu))
3207 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3209 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3210 u32 hflags = vcpu->arch.hflags;
3211 if (events->smi.smm)
3212 hflags |= HF_SMM_MASK;
3214 hflags &= ~HF_SMM_MASK;
3215 kvm_set_hflags(vcpu, hflags);
3217 vcpu->arch.smi_pending = events->smi.pending;
3219 if (events->smi.smm) {
3220 if (events->smi.smm_inside_nmi)
3221 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3223 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3224 if (lapic_in_kernel(vcpu)) {
3225 if (events->smi.latched_init)
3226 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3228 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3233 kvm_make_request(KVM_REQ_EVENT, vcpu);
3238 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3239 struct kvm_debugregs *dbgregs)
3243 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3244 kvm_get_dr(vcpu, 6, &val);
3246 dbgregs->dr7 = vcpu->arch.dr7;
3248 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3251 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3252 struct kvm_debugregs *dbgregs)
3257 if (dbgregs->dr6 & ~0xffffffffull)
3259 if (dbgregs->dr7 & ~0xffffffffull)
3262 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3263 kvm_update_dr0123(vcpu);
3264 vcpu->arch.dr6 = dbgregs->dr6;
3265 kvm_update_dr6(vcpu);
3266 vcpu->arch.dr7 = dbgregs->dr7;
3267 kvm_update_dr7(vcpu);
3272 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3274 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3276 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3277 u64 xstate_bv = xsave->header.xfeatures;
3281 * Copy legacy XSAVE area, to avoid complications with CPUID
3282 * leaves 0 and 1 in the loop below.
3284 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3287 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3288 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3291 * Copy each region from the possibly compacted offset to the
3292 * non-compacted offset.
3294 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3296 u64 feature = valid & -valid;
3297 int index = fls64(feature) - 1;
3298 void *src = get_xsave_addr(xsave, feature);
3301 u32 size, offset, ecx, edx;
3302 cpuid_count(XSTATE_CPUID, index,
3303 &size, &offset, &ecx, &edx);
3304 if (feature == XFEATURE_MASK_PKRU)
3305 memcpy(dest + offset, &vcpu->arch.pkru,
3306 sizeof(vcpu->arch.pkru));
3308 memcpy(dest + offset, src, size);
3316 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3318 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3319 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3323 * Copy legacy XSAVE area, to avoid complications with CPUID
3324 * leaves 0 and 1 in the loop below.
3326 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3328 /* Set XSTATE_BV and possibly XCOMP_BV. */
3329 xsave->header.xfeatures = xstate_bv;
3330 if (boot_cpu_has(X86_FEATURE_XSAVES))
3331 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3334 * Copy each region from the non-compacted offset to the
3335 * possibly compacted offset.
3337 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3339 u64 feature = valid & -valid;
3340 int index = fls64(feature) - 1;
3341 void *dest = get_xsave_addr(xsave, feature);
3344 u32 size, offset, ecx, edx;
3345 cpuid_count(XSTATE_CPUID, index,
3346 &size, &offset, &ecx, &edx);
3347 if (feature == XFEATURE_MASK_PKRU)
3348 memcpy(&vcpu->arch.pkru, src + offset,
3349 sizeof(vcpu->arch.pkru));
3351 memcpy(dest, src + offset, size);
3358 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3359 struct kvm_xsave *guest_xsave)
3361 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3362 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3363 fill_xsave((u8 *) guest_xsave->region, vcpu);
3365 memcpy(guest_xsave->region,
3366 &vcpu->arch.guest_fpu.state.fxsave,
3367 sizeof(struct fxregs_state));
3368 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3369 XFEATURE_MASK_FPSSE;
3373 #define XSAVE_MXCSR_OFFSET 24
3375 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3376 struct kvm_xsave *guest_xsave)
3379 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3380 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3382 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3384 * Here we allow setting states that are not present in
3385 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3386 * with old userspace.
3388 if (xstate_bv & ~kvm_supported_xcr0() ||
3389 mxcsr & ~mxcsr_feature_mask)
3391 load_xsave(vcpu, (u8 *)guest_xsave->region);
3393 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3394 mxcsr & ~mxcsr_feature_mask)
3396 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3397 guest_xsave->region, sizeof(struct fxregs_state));
3402 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3403 struct kvm_xcrs *guest_xcrs)
3405 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3406 guest_xcrs->nr_xcrs = 0;
3410 guest_xcrs->nr_xcrs = 1;
3411 guest_xcrs->flags = 0;
3412 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3413 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3416 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3417 struct kvm_xcrs *guest_xcrs)
3421 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3424 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3427 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3428 /* Only support XCR0 currently */
3429 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3430 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3431 guest_xcrs->xcrs[i].value);
3440 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3441 * stopped by the hypervisor. This function will be called from the host only.
3442 * EINVAL is returned when the host attempts to set the flag for a guest that
3443 * does not support pv clocks.
3445 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3447 if (!vcpu->arch.pv_time_enabled)
3449 vcpu->arch.pvclock_set_guest_stopped_request = true;
3450 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3454 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3455 struct kvm_enable_cap *cap)
3461 case KVM_CAP_HYPERV_SYNIC2:
3464 case KVM_CAP_HYPERV_SYNIC:
3465 if (!irqchip_in_kernel(vcpu->kvm))
3467 return kvm_hv_activate_synic(vcpu, cap->cap ==
3468 KVM_CAP_HYPERV_SYNIC2);
3474 long kvm_arch_vcpu_ioctl(struct file *filp,
3475 unsigned int ioctl, unsigned long arg)
3477 struct kvm_vcpu *vcpu = filp->private_data;
3478 void __user *argp = (void __user *)arg;
3481 struct kvm_lapic_state *lapic;
3482 struct kvm_xsave *xsave;
3483 struct kvm_xcrs *xcrs;
3489 case KVM_GET_LAPIC: {
3491 if (!lapic_in_kernel(vcpu))
3493 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3498 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3502 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3507 case KVM_SET_LAPIC: {
3509 if (!lapic_in_kernel(vcpu))
3511 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3512 if (IS_ERR(u.lapic))
3513 return PTR_ERR(u.lapic);
3515 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3518 case KVM_INTERRUPT: {
3519 struct kvm_interrupt irq;
3522 if (copy_from_user(&irq, argp, sizeof irq))
3524 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3528 r = kvm_vcpu_ioctl_nmi(vcpu);
3532 r = kvm_vcpu_ioctl_smi(vcpu);
3535 case KVM_SET_CPUID: {
3536 struct kvm_cpuid __user *cpuid_arg = argp;
3537 struct kvm_cpuid cpuid;
3540 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3542 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3545 case KVM_SET_CPUID2: {
3546 struct kvm_cpuid2 __user *cpuid_arg = argp;
3547 struct kvm_cpuid2 cpuid;
3550 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3552 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3553 cpuid_arg->entries);
3556 case KVM_GET_CPUID2: {
3557 struct kvm_cpuid2 __user *cpuid_arg = argp;
3558 struct kvm_cpuid2 cpuid;
3561 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3563 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3564 cpuid_arg->entries);
3568 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3574 r = msr_io(vcpu, argp, do_get_msr, 1);
3577 r = msr_io(vcpu, argp, do_set_msr, 0);
3579 case KVM_TPR_ACCESS_REPORTING: {
3580 struct kvm_tpr_access_ctl tac;
3583 if (copy_from_user(&tac, argp, sizeof tac))
3585 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3589 if (copy_to_user(argp, &tac, sizeof tac))
3594 case KVM_SET_VAPIC_ADDR: {
3595 struct kvm_vapic_addr va;
3599 if (!lapic_in_kernel(vcpu))
3602 if (copy_from_user(&va, argp, sizeof va))
3604 idx = srcu_read_lock(&vcpu->kvm->srcu);
3605 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3606 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3609 case KVM_X86_SETUP_MCE: {
3613 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3615 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3618 case KVM_X86_SET_MCE: {
3619 struct kvm_x86_mce mce;
3622 if (copy_from_user(&mce, argp, sizeof mce))
3624 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3627 case KVM_GET_VCPU_EVENTS: {
3628 struct kvm_vcpu_events events;
3630 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3633 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3638 case KVM_SET_VCPU_EVENTS: {
3639 struct kvm_vcpu_events events;
3642 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3645 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3648 case KVM_GET_DEBUGREGS: {
3649 struct kvm_debugregs dbgregs;
3651 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3654 if (copy_to_user(argp, &dbgregs,
3655 sizeof(struct kvm_debugregs)))
3660 case KVM_SET_DEBUGREGS: {
3661 struct kvm_debugregs dbgregs;
3664 if (copy_from_user(&dbgregs, argp,
3665 sizeof(struct kvm_debugregs)))
3668 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3671 case KVM_GET_XSAVE: {
3672 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3677 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3680 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3685 case KVM_SET_XSAVE: {
3686 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3687 if (IS_ERR(u.xsave))
3688 return PTR_ERR(u.xsave);
3690 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3693 case KVM_GET_XCRS: {
3694 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3699 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3702 if (copy_to_user(argp, u.xcrs,
3703 sizeof(struct kvm_xcrs)))
3708 case KVM_SET_XCRS: {
3709 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3711 return PTR_ERR(u.xcrs);
3713 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3716 case KVM_SET_TSC_KHZ: {
3720 user_tsc_khz = (u32)arg;
3722 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3725 if (user_tsc_khz == 0)
3726 user_tsc_khz = tsc_khz;
3728 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3733 case KVM_GET_TSC_KHZ: {
3734 r = vcpu->arch.virtual_tsc_khz;
3737 case KVM_KVMCLOCK_CTRL: {
3738 r = kvm_set_guest_paused(vcpu);
3741 case KVM_ENABLE_CAP: {
3742 struct kvm_enable_cap cap;
3745 if (copy_from_user(&cap, argp, sizeof(cap)))
3747 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3758 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3760 return VM_FAULT_SIGBUS;
3763 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3767 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3769 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3773 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3776 kvm->arch.ept_identity_map_addr = ident_addr;
3780 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3781 u32 kvm_nr_mmu_pages)
3783 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3786 mutex_lock(&kvm->slots_lock);
3788 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3789 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3791 mutex_unlock(&kvm->slots_lock);
3795 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3797 return kvm->arch.n_max_mmu_pages;
3800 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3802 struct kvm_pic *pic = kvm->arch.vpic;
3806 switch (chip->chip_id) {
3807 case KVM_IRQCHIP_PIC_MASTER:
3808 memcpy(&chip->chip.pic, &pic->pics[0],
3809 sizeof(struct kvm_pic_state));
3811 case KVM_IRQCHIP_PIC_SLAVE:
3812 memcpy(&chip->chip.pic, &pic->pics[1],
3813 sizeof(struct kvm_pic_state));
3815 case KVM_IRQCHIP_IOAPIC:
3816 kvm_get_ioapic(kvm, &chip->chip.ioapic);
3825 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3827 struct kvm_pic *pic = kvm->arch.vpic;
3831 switch (chip->chip_id) {
3832 case KVM_IRQCHIP_PIC_MASTER:
3833 spin_lock(&pic->lock);
3834 memcpy(&pic->pics[0], &chip->chip.pic,
3835 sizeof(struct kvm_pic_state));
3836 spin_unlock(&pic->lock);
3838 case KVM_IRQCHIP_PIC_SLAVE:
3839 spin_lock(&pic->lock);
3840 memcpy(&pic->pics[1], &chip->chip.pic,
3841 sizeof(struct kvm_pic_state));
3842 spin_unlock(&pic->lock);
3844 case KVM_IRQCHIP_IOAPIC:
3845 kvm_set_ioapic(kvm, &chip->chip.ioapic);
3851 kvm_pic_update_irq(pic);
3855 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3857 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3859 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3861 mutex_lock(&kps->lock);
3862 memcpy(ps, &kps->channels, sizeof(*ps));
3863 mutex_unlock(&kps->lock);
3867 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3870 struct kvm_pit *pit = kvm->arch.vpit;
3872 mutex_lock(&pit->pit_state.lock);
3873 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3874 for (i = 0; i < 3; i++)
3875 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3876 mutex_unlock(&pit->pit_state.lock);
3880 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3882 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3883 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3884 sizeof(ps->channels));
3885 ps->flags = kvm->arch.vpit->pit_state.flags;
3886 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3887 memset(&ps->reserved, 0, sizeof(ps->reserved));
3891 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3895 u32 prev_legacy, cur_legacy;
3896 struct kvm_pit *pit = kvm->arch.vpit;
3898 mutex_lock(&pit->pit_state.lock);
3899 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3900 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3901 if (!prev_legacy && cur_legacy)
3903 memcpy(&pit->pit_state.channels, &ps->channels,
3904 sizeof(pit->pit_state.channels));
3905 pit->pit_state.flags = ps->flags;
3906 for (i = 0; i < 3; i++)
3907 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3909 mutex_unlock(&pit->pit_state.lock);
3913 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3914 struct kvm_reinject_control *control)
3916 struct kvm_pit *pit = kvm->arch.vpit;
3921 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3922 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3923 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3925 mutex_lock(&pit->pit_state.lock);
3926 kvm_pit_set_reinject(pit, control->pit_reinject);
3927 mutex_unlock(&pit->pit_state.lock);
3933 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3934 * @kvm: kvm instance
3935 * @log: slot id and address to which we copy the log
3937 * Steps 1-4 below provide general overview of dirty page logging. See
3938 * kvm_get_dirty_log_protect() function description for additional details.
3940 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3941 * always flush the TLB (step 4) even if previous step failed and the dirty
3942 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3943 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3944 * writes will be marked dirty for next log read.
3946 * 1. Take a snapshot of the bit and clear it if needed.
3947 * 2. Write protect the corresponding page.
3948 * 3. Copy the snapshot to the userspace.
3949 * 4. Flush TLB's if needed.
3951 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3953 bool is_dirty = false;
3956 mutex_lock(&kvm->slots_lock);
3959 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3961 if (kvm_x86_ops->flush_log_dirty)
3962 kvm_x86_ops->flush_log_dirty(kvm);
3964 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3967 * All the TLBs can be flushed out of mmu lock, see the comments in
3968 * kvm_mmu_slot_remove_write_access().
3970 lockdep_assert_held(&kvm->slots_lock);
3972 kvm_flush_remote_tlbs(kvm);
3974 mutex_unlock(&kvm->slots_lock);
3978 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3981 if (!irqchip_in_kernel(kvm))
3984 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3985 irq_event->irq, irq_event->level,
3990 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3991 struct kvm_enable_cap *cap)
3999 case KVM_CAP_DISABLE_QUIRKS:
4000 kvm->arch.disabled_quirks = cap->args[0];
4003 case KVM_CAP_SPLIT_IRQCHIP: {
4004 mutex_lock(&kvm->lock);
4006 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4007 goto split_irqchip_unlock;
4009 if (irqchip_in_kernel(kvm))
4010 goto split_irqchip_unlock;
4011 if (kvm->created_vcpus)
4012 goto split_irqchip_unlock;
4013 r = kvm_setup_empty_irq_routing(kvm);
4015 goto split_irqchip_unlock;
4016 /* Pairs with irqchip_in_kernel. */
4018 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4019 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4021 split_irqchip_unlock:
4022 mutex_unlock(&kvm->lock);
4025 case KVM_CAP_X2APIC_API:
4027 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4030 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4031 kvm->arch.x2apic_format = true;
4032 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4033 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4044 long kvm_arch_vm_ioctl(struct file *filp,
4045 unsigned int ioctl, unsigned long arg)
4047 struct kvm *kvm = filp->private_data;
4048 void __user *argp = (void __user *)arg;
4051 * This union makes it completely explicit to gcc-3.x
4052 * that these two variables' stack usage should be
4053 * combined, not added together.
4056 struct kvm_pit_state ps;
4057 struct kvm_pit_state2 ps2;
4058 struct kvm_pit_config pit_config;
4062 case KVM_SET_TSS_ADDR:
4063 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4065 case KVM_SET_IDENTITY_MAP_ADDR: {
4068 mutex_lock(&kvm->lock);
4070 if (kvm->created_vcpus)
4071 goto set_identity_unlock;
4073 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4074 goto set_identity_unlock;
4075 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4076 set_identity_unlock:
4077 mutex_unlock(&kvm->lock);
4080 case KVM_SET_NR_MMU_PAGES:
4081 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4083 case KVM_GET_NR_MMU_PAGES:
4084 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4086 case KVM_CREATE_IRQCHIP: {
4087 mutex_lock(&kvm->lock);
4090 if (irqchip_in_kernel(kvm))
4091 goto create_irqchip_unlock;
4094 if (kvm->created_vcpus)
4095 goto create_irqchip_unlock;
4097 r = kvm_pic_init(kvm);
4099 goto create_irqchip_unlock;
4101 r = kvm_ioapic_init(kvm);
4103 kvm_pic_destroy(kvm);
4104 goto create_irqchip_unlock;
4107 r = kvm_setup_default_irq_routing(kvm);
4109 kvm_ioapic_destroy(kvm);
4110 kvm_pic_destroy(kvm);
4111 goto create_irqchip_unlock;
4113 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4115 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4116 create_irqchip_unlock:
4117 mutex_unlock(&kvm->lock);
4120 case KVM_CREATE_PIT:
4121 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4123 case KVM_CREATE_PIT2:
4125 if (copy_from_user(&u.pit_config, argp,
4126 sizeof(struct kvm_pit_config)))
4129 mutex_lock(&kvm->lock);
4132 goto create_pit_unlock;
4134 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4138 mutex_unlock(&kvm->lock);
4140 case KVM_GET_IRQCHIP: {
4141 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4142 struct kvm_irqchip *chip;
4144 chip = memdup_user(argp, sizeof(*chip));
4151 if (!irqchip_kernel(kvm))
4152 goto get_irqchip_out;
4153 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4155 goto get_irqchip_out;
4157 if (copy_to_user(argp, chip, sizeof *chip))
4158 goto get_irqchip_out;
4164 case KVM_SET_IRQCHIP: {
4165 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4166 struct kvm_irqchip *chip;
4168 chip = memdup_user(argp, sizeof(*chip));
4175 if (!irqchip_kernel(kvm))
4176 goto set_irqchip_out;
4177 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4179 goto set_irqchip_out;
4187 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4190 if (!kvm->arch.vpit)
4192 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4196 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4203 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4206 if (!kvm->arch.vpit)
4208 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4211 case KVM_GET_PIT2: {
4213 if (!kvm->arch.vpit)
4215 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4219 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4224 case KVM_SET_PIT2: {
4226 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4229 if (!kvm->arch.vpit)
4231 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4234 case KVM_REINJECT_CONTROL: {
4235 struct kvm_reinject_control control;
4237 if (copy_from_user(&control, argp, sizeof(control)))
4239 r = kvm_vm_ioctl_reinject(kvm, &control);
4242 case KVM_SET_BOOT_CPU_ID:
4244 mutex_lock(&kvm->lock);
4245 if (kvm->created_vcpus)
4248 kvm->arch.bsp_vcpu_id = arg;
4249 mutex_unlock(&kvm->lock);
4251 case KVM_XEN_HVM_CONFIG: {
4253 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4254 sizeof(struct kvm_xen_hvm_config)))
4257 if (kvm->arch.xen_hvm_config.flags)
4262 case KVM_SET_CLOCK: {
4263 struct kvm_clock_data user_ns;
4267 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4276 * TODO: userspace has to take care of races with VCPU_RUN, so
4277 * kvm_gen_update_masterclock() can be cut down to locked
4278 * pvclock_update_vm_gtod_copy().
4280 kvm_gen_update_masterclock(kvm);
4281 now_ns = get_kvmclock_ns(kvm);
4282 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4283 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4286 case KVM_GET_CLOCK: {
4287 struct kvm_clock_data user_ns;
4290 now_ns = get_kvmclock_ns(kvm);
4291 user_ns.clock = now_ns;
4292 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4293 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4296 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4301 case KVM_ENABLE_CAP: {
4302 struct kvm_enable_cap cap;
4305 if (copy_from_user(&cap, argp, sizeof(cap)))
4307 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4317 static void kvm_init_msr_list(void)
4322 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4323 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4327 * Even MSRs that are valid in the host may not be exposed
4328 * to the guests in some cases.
4330 switch (msrs_to_save[i]) {
4331 case MSR_IA32_BNDCFGS:
4332 if (!kvm_x86_ops->mpx_supported())
4336 if (!kvm_x86_ops->rdtscp_supported())
4344 msrs_to_save[j] = msrs_to_save[i];
4347 num_msrs_to_save = j;
4349 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4350 switch (emulated_msrs[i]) {
4351 case MSR_IA32_SMBASE:
4352 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4360 emulated_msrs[j] = emulated_msrs[i];
4363 num_emulated_msrs = j;
4366 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4374 if (!(lapic_in_kernel(vcpu) &&
4375 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4376 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4387 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4394 if (!(lapic_in_kernel(vcpu) &&
4395 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4397 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4399 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4409 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4410 struct kvm_segment *var, int seg)
4412 kvm_x86_ops->set_segment(vcpu, var, seg);
4415 void kvm_get_segment(struct kvm_vcpu *vcpu,
4416 struct kvm_segment *var, int seg)
4418 kvm_x86_ops->get_segment(vcpu, var, seg);
4421 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4422 struct x86_exception *exception)
4426 BUG_ON(!mmu_is_nested(vcpu));
4428 /* NPT walks are always user-walks */
4429 access |= PFERR_USER_MASK;
4430 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4435 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4436 struct x86_exception *exception)
4438 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4439 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4442 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4443 struct x86_exception *exception)
4445 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4446 access |= PFERR_FETCH_MASK;
4447 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4450 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4451 struct x86_exception *exception)
4453 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4454 access |= PFERR_WRITE_MASK;
4455 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4458 /* uses this to access any guest's mapped memory without checking CPL */
4459 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4460 struct x86_exception *exception)
4462 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4465 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4466 struct kvm_vcpu *vcpu, u32 access,
4467 struct x86_exception *exception)
4470 int r = X86EMUL_CONTINUE;
4473 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4475 unsigned offset = addr & (PAGE_SIZE-1);
4476 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4479 if (gpa == UNMAPPED_GVA)
4480 return X86EMUL_PROPAGATE_FAULT;
4481 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4484 r = X86EMUL_IO_NEEDED;
4496 /* used for instruction fetching */
4497 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4498 gva_t addr, void *val, unsigned int bytes,
4499 struct x86_exception *exception)
4501 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4502 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4506 /* Inline kvm_read_guest_virt_helper for speed. */
4507 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4509 if (unlikely(gpa == UNMAPPED_GVA))
4510 return X86EMUL_PROPAGATE_FAULT;
4512 offset = addr & (PAGE_SIZE-1);
4513 if (WARN_ON(offset + bytes > PAGE_SIZE))
4514 bytes = (unsigned)PAGE_SIZE - offset;
4515 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4517 if (unlikely(ret < 0))
4518 return X86EMUL_IO_NEEDED;
4520 return X86EMUL_CONTINUE;
4523 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4524 gva_t addr, void *val, unsigned int bytes,
4525 struct x86_exception *exception)
4527 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4528 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4530 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4533 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4535 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4536 gva_t addr, void *val, unsigned int bytes,
4537 struct x86_exception *exception)
4539 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4540 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4543 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4544 unsigned long addr, void *val, unsigned int bytes)
4546 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4547 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4549 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4552 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4553 gva_t addr, void *val,
4555 struct x86_exception *exception)
4557 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4559 int r = X86EMUL_CONTINUE;
4562 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4565 unsigned offset = addr & (PAGE_SIZE-1);
4566 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4569 if (gpa == UNMAPPED_GVA)
4570 return X86EMUL_PROPAGATE_FAULT;
4571 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4573 r = X86EMUL_IO_NEEDED;
4584 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4586 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4587 gpa_t gpa, bool write)
4589 /* For APIC access vmexit */
4590 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4593 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4594 trace_vcpu_match_mmio(gva, gpa, write, true);
4601 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4602 gpa_t *gpa, struct x86_exception *exception,
4605 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4606 | (write ? PFERR_WRITE_MASK : 0);
4609 * currently PKRU is only applied to ept enabled guest so
4610 * there is no pkey in EPT page table for L1 guest or EPT
4611 * shadow page table for L2 guest.
4613 if (vcpu_match_mmio_gva(vcpu, gva)
4614 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4615 vcpu->arch.access, 0, access)) {
4616 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4617 (gva & (PAGE_SIZE - 1));
4618 trace_vcpu_match_mmio(gva, *gpa, write, false);
4622 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4624 if (*gpa == UNMAPPED_GVA)
4627 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4630 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4631 const void *val, int bytes)
4635 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4638 kvm_page_track_write(vcpu, gpa, val, bytes);
4642 struct read_write_emulator_ops {
4643 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4645 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4646 void *val, int bytes);
4647 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4648 int bytes, void *val);
4649 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4650 void *val, int bytes);
4654 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4656 if (vcpu->mmio_read_completed) {
4657 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4658 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4659 vcpu->mmio_read_completed = 0;
4666 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4667 void *val, int bytes)
4669 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4672 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4673 void *val, int bytes)
4675 return emulator_write_phys(vcpu, gpa, val, bytes);
4678 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4680 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4681 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4684 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4685 void *val, int bytes)
4687 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4688 return X86EMUL_IO_NEEDED;
4691 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4692 void *val, int bytes)
4694 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4696 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4697 return X86EMUL_CONTINUE;
4700 static const struct read_write_emulator_ops read_emultor = {
4701 .read_write_prepare = read_prepare,
4702 .read_write_emulate = read_emulate,
4703 .read_write_mmio = vcpu_mmio_read,
4704 .read_write_exit_mmio = read_exit_mmio,
4707 static const struct read_write_emulator_ops write_emultor = {
4708 .read_write_emulate = write_emulate,
4709 .read_write_mmio = write_mmio,
4710 .read_write_exit_mmio = write_exit_mmio,
4714 static int emulator_read_write_onepage(unsigned long addr, void *val,
4716 struct x86_exception *exception,
4717 struct kvm_vcpu *vcpu,
4718 const struct read_write_emulator_ops *ops)
4722 bool write = ops->write;
4723 struct kvm_mmio_fragment *frag;
4724 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4727 * If the exit was due to a NPF we may already have a GPA.
4728 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4729 * Note, this cannot be used on string operations since string
4730 * operation using rep will only have the initial GPA from the NPF
4733 if (vcpu->arch.gpa_available &&
4734 emulator_can_use_gpa(ctxt) &&
4735 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
4736 gpa = vcpu->arch.gpa_val;
4737 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
4739 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4741 return X86EMUL_PROPAGATE_FAULT;
4744 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
4745 return X86EMUL_CONTINUE;
4748 * Is this MMIO handled locally?
4750 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4751 if (handled == bytes)
4752 return X86EMUL_CONTINUE;
4758 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4759 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4763 return X86EMUL_CONTINUE;
4766 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4768 void *val, unsigned int bytes,
4769 struct x86_exception *exception,
4770 const struct read_write_emulator_ops *ops)
4772 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4776 if (ops->read_write_prepare &&
4777 ops->read_write_prepare(vcpu, val, bytes))
4778 return X86EMUL_CONTINUE;
4780 vcpu->mmio_nr_fragments = 0;
4782 /* Crossing a page boundary? */
4783 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4786 now = -addr & ~PAGE_MASK;
4787 rc = emulator_read_write_onepage(addr, val, now, exception,
4790 if (rc != X86EMUL_CONTINUE)
4793 if (ctxt->mode != X86EMUL_MODE_PROT64)
4799 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4801 if (rc != X86EMUL_CONTINUE)
4804 if (!vcpu->mmio_nr_fragments)
4807 gpa = vcpu->mmio_fragments[0].gpa;
4809 vcpu->mmio_needed = 1;
4810 vcpu->mmio_cur_fragment = 0;
4812 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4813 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4814 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4815 vcpu->run->mmio.phys_addr = gpa;
4817 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4820 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4824 struct x86_exception *exception)
4826 return emulator_read_write(ctxt, addr, val, bytes,
4827 exception, &read_emultor);
4830 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4834 struct x86_exception *exception)
4836 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4837 exception, &write_emultor);
4840 #define CMPXCHG_TYPE(t, ptr, old, new) \
4841 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4843 #ifdef CONFIG_X86_64
4844 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4846 # define CMPXCHG64(ptr, old, new) \
4847 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4850 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4855 struct x86_exception *exception)
4857 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4863 /* guests cmpxchg8b have to be emulated atomically */
4864 if (bytes > 8 || (bytes & (bytes - 1)))
4867 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4869 if (gpa == UNMAPPED_GVA ||
4870 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4873 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4876 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4877 if (is_error_page(page))
4880 kaddr = kmap_atomic(page);
4881 kaddr += offset_in_page(gpa);
4884 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4887 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4890 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4893 exchanged = CMPXCHG64(kaddr, old, new);
4898 kunmap_atomic(kaddr);
4899 kvm_release_page_dirty(page);
4902 return X86EMUL_CMPXCHG_FAILED;
4904 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4905 kvm_page_track_write(vcpu, gpa, new, bytes);
4907 return X86EMUL_CONTINUE;
4910 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4912 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4915 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4919 for (i = 0; i < vcpu->arch.pio.count; i++) {
4920 if (vcpu->arch.pio.in)
4921 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4922 vcpu->arch.pio.size, pd);
4924 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4925 vcpu->arch.pio.port, vcpu->arch.pio.size,
4929 pd += vcpu->arch.pio.size;
4934 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4935 unsigned short port, void *val,
4936 unsigned int count, bool in)
4938 vcpu->arch.pio.port = port;
4939 vcpu->arch.pio.in = in;
4940 vcpu->arch.pio.count = count;
4941 vcpu->arch.pio.size = size;
4943 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4944 vcpu->arch.pio.count = 0;
4948 vcpu->run->exit_reason = KVM_EXIT_IO;
4949 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4950 vcpu->run->io.size = size;
4951 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4952 vcpu->run->io.count = count;
4953 vcpu->run->io.port = port;
4958 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4959 int size, unsigned short port, void *val,
4962 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4965 if (vcpu->arch.pio.count)
4968 memset(vcpu->arch.pio_data, 0, size * count);
4970 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4973 memcpy(val, vcpu->arch.pio_data, size * count);
4974 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4975 vcpu->arch.pio.count = 0;
4982 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4983 int size, unsigned short port,
4984 const void *val, unsigned int count)
4986 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4988 memcpy(vcpu->arch.pio_data, val, size * count);
4989 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4990 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4993 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4995 return kvm_x86_ops->get_segment_base(vcpu, seg);
4998 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5000 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5003 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5005 if (!need_emulate_wbinvd(vcpu))
5006 return X86EMUL_CONTINUE;
5008 if (kvm_x86_ops->has_wbinvd_exit()) {
5009 int cpu = get_cpu();
5011 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5012 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5013 wbinvd_ipi, NULL, 1);
5015 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5018 return X86EMUL_CONTINUE;
5021 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5023 kvm_emulate_wbinvd_noskip(vcpu);
5024 return kvm_skip_emulated_instruction(vcpu);
5026 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5030 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5032 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5035 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5036 unsigned long *dest)
5038 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5041 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5042 unsigned long value)
5045 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5048 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5050 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5053 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5055 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5056 unsigned long value;
5060 value = kvm_read_cr0(vcpu);
5063 value = vcpu->arch.cr2;
5066 value = kvm_read_cr3(vcpu);
5069 value = kvm_read_cr4(vcpu);
5072 value = kvm_get_cr8(vcpu);
5075 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5082 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5084 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5089 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5092 vcpu->arch.cr2 = val;
5095 res = kvm_set_cr3(vcpu, val);
5098 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5101 res = kvm_set_cr8(vcpu, val);
5104 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5111 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5113 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5116 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5118 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5121 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5123 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5126 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5128 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5131 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5133 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5136 static unsigned long emulator_get_cached_segment_base(
5137 struct x86_emulate_ctxt *ctxt, int seg)
5139 return get_segment_base(emul_to_vcpu(ctxt), seg);
5142 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5143 struct desc_struct *desc, u32 *base3,
5146 struct kvm_segment var;
5148 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5149 *selector = var.selector;
5152 memset(desc, 0, sizeof(*desc));
5160 set_desc_limit(desc, var.limit);
5161 set_desc_base(desc, (unsigned long)var.base);
5162 #ifdef CONFIG_X86_64
5164 *base3 = var.base >> 32;
5166 desc->type = var.type;
5168 desc->dpl = var.dpl;
5169 desc->p = var.present;
5170 desc->avl = var.avl;
5178 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5179 struct desc_struct *desc, u32 base3,
5182 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5183 struct kvm_segment var;
5185 var.selector = selector;
5186 var.base = get_desc_base(desc);
5187 #ifdef CONFIG_X86_64
5188 var.base |= ((u64)base3) << 32;
5190 var.limit = get_desc_limit(desc);
5192 var.limit = (var.limit << 12) | 0xfff;
5193 var.type = desc->type;
5194 var.dpl = desc->dpl;
5199 var.avl = desc->avl;
5200 var.present = desc->p;
5201 var.unusable = !var.present;
5204 kvm_set_segment(vcpu, &var, seg);
5208 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5209 u32 msr_index, u64 *pdata)
5211 struct msr_data msr;
5214 msr.index = msr_index;
5215 msr.host_initiated = false;
5216 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5224 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5225 u32 msr_index, u64 data)
5227 struct msr_data msr;
5230 msr.index = msr_index;
5231 msr.host_initiated = false;
5232 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5235 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5237 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5239 return vcpu->arch.smbase;
5242 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5244 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5246 vcpu->arch.smbase = smbase;
5249 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5252 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5255 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5256 u32 pmc, u64 *pdata)
5258 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5261 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5263 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5266 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5267 struct x86_instruction_info *info,
5268 enum x86_intercept_stage stage)
5270 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5273 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5274 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5276 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5279 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5281 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5284 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5286 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5289 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5291 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5294 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5296 return emul_to_vcpu(ctxt)->arch.hflags;
5299 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5301 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5304 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5306 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5309 static const struct x86_emulate_ops emulate_ops = {
5310 .read_gpr = emulator_read_gpr,
5311 .write_gpr = emulator_write_gpr,
5312 .read_std = kvm_read_guest_virt_system,
5313 .write_std = kvm_write_guest_virt_system,
5314 .read_phys = kvm_read_guest_phys_system,
5315 .fetch = kvm_fetch_guest_virt,
5316 .read_emulated = emulator_read_emulated,
5317 .write_emulated = emulator_write_emulated,
5318 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5319 .invlpg = emulator_invlpg,
5320 .pio_in_emulated = emulator_pio_in_emulated,
5321 .pio_out_emulated = emulator_pio_out_emulated,
5322 .get_segment = emulator_get_segment,
5323 .set_segment = emulator_set_segment,
5324 .get_cached_segment_base = emulator_get_cached_segment_base,
5325 .get_gdt = emulator_get_gdt,
5326 .get_idt = emulator_get_idt,
5327 .set_gdt = emulator_set_gdt,
5328 .set_idt = emulator_set_idt,
5329 .get_cr = emulator_get_cr,
5330 .set_cr = emulator_set_cr,
5331 .cpl = emulator_get_cpl,
5332 .get_dr = emulator_get_dr,
5333 .set_dr = emulator_set_dr,
5334 .get_smbase = emulator_get_smbase,
5335 .set_smbase = emulator_set_smbase,
5336 .set_msr = emulator_set_msr,
5337 .get_msr = emulator_get_msr,
5338 .check_pmc = emulator_check_pmc,
5339 .read_pmc = emulator_read_pmc,
5340 .halt = emulator_halt,
5341 .wbinvd = emulator_wbinvd,
5342 .fix_hypercall = emulator_fix_hypercall,
5343 .intercept = emulator_intercept,
5344 .get_cpuid = emulator_get_cpuid,
5345 .set_nmi_mask = emulator_set_nmi_mask,
5346 .get_hflags = emulator_get_hflags,
5347 .set_hflags = emulator_set_hflags,
5348 .pre_leave_smm = emulator_pre_leave_smm,
5351 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5353 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5355 * an sti; sti; sequence only disable interrupts for the first
5356 * instruction. So, if the last instruction, be it emulated or
5357 * not, left the system with the INT_STI flag enabled, it
5358 * means that the last instruction is an sti. We should not
5359 * leave the flag on in this case. The same goes for mov ss
5361 if (int_shadow & mask)
5363 if (unlikely(int_shadow || mask)) {
5364 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5366 kvm_make_request(KVM_REQ_EVENT, vcpu);
5370 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5372 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5373 if (ctxt->exception.vector == PF_VECTOR)
5374 return kvm_propagate_fault(vcpu, &ctxt->exception);
5376 if (ctxt->exception.error_code_valid)
5377 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5378 ctxt->exception.error_code);
5380 kvm_queue_exception(vcpu, ctxt->exception.vector);
5384 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5386 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5389 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5391 ctxt->eflags = kvm_get_rflags(vcpu);
5392 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5394 ctxt->eip = kvm_rip_read(vcpu);
5395 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5396 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5397 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5398 cs_db ? X86EMUL_MODE_PROT32 :
5399 X86EMUL_MODE_PROT16;
5400 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5401 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5402 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5404 init_decode_cache(ctxt);
5405 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5408 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5410 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5413 init_emulate_ctxt(vcpu);
5417 ctxt->_eip = ctxt->eip + inc_eip;
5418 ret = emulate_int_real(ctxt, irq);
5420 if (ret != X86EMUL_CONTINUE)
5421 return EMULATE_FAIL;
5423 ctxt->eip = ctxt->_eip;
5424 kvm_rip_write(vcpu, ctxt->eip);
5425 kvm_set_rflags(vcpu, ctxt->eflags);
5427 if (irq == NMI_VECTOR)
5428 vcpu->arch.nmi_pending = 0;
5430 vcpu->arch.interrupt.pending = false;
5432 return EMULATE_DONE;
5434 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5436 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5438 int r = EMULATE_DONE;
5440 ++vcpu->stat.insn_emulation_fail;
5441 trace_kvm_emulate_insn_failed(vcpu);
5442 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5443 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5444 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5445 vcpu->run->internal.ndata = 0;
5446 r = EMULATE_USER_EXIT;
5448 kvm_queue_exception(vcpu, UD_VECTOR);
5453 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5454 bool write_fault_to_shadow_pgtable,
5460 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5463 if (!vcpu->arch.mmu.direct_map) {
5465 * Write permission should be allowed since only
5466 * write access need to be emulated.
5468 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5471 * If the mapping is invalid in guest, let cpu retry
5472 * it to generate fault.
5474 if (gpa == UNMAPPED_GVA)
5479 * Do not retry the unhandleable instruction if it faults on the
5480 * readonly host memory, otherwise it will goto a infinite loop:
5481 * retry instruction -> write #PF -> emulation fail -> retry
5482 * instruction -> ...
5484 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5487 * If the instruction failed on the error pfn, it can not be fixed,
5488 * report the error to userspace.
5490 if (is_error_noslot_pfn(pfn))
5493 kvm_release_pfn_clean(pfn);
5495 /* The instructions are well-emulated on direct mmu. */
5496 if (vcpu->arch.mmu.direct_map) {
5497 unsigned int indirect_shadow_pages;
5499 spin_lock(&vcpu->kvm->mmu_lock);
5500 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5501 spin_unlock(&vcpu->kvm->mmu_lock);
5503 if (indirect_shadow_pages)
5504 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5510 * if emulation was due to access to shadowed page table
5511 * and it failed try to unshadow page and re-enter the
5512 * guest to let CPU execute the instruction.
5514 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5517 * If the access faults on its page table, it can not
5518 * be fixed by unprotecting shadow page and it should
5519 * be reported to userspace.
5521 return !write_fault_to_shadow_pgtable;
5524 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5525 unsigned long cr2, int emulation_type)
5527 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5528 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5530 last_retry_eip = vcpu->arch.last_retry_eip;
5531 last_retry_addr = vcpu->arch.last_retry_addr;
5534 * If the emulation is caused by #PF and it is non-page_table
5535 * writing instruction, it means the VM-EXIT is caused by shadow
5536 * page protected, we can zap the shadow page and retry this
5537 * instruction directly.
5539 * Note: if the guest uses a non-page-table modifying instruction
5540 * on the PDE that points to the instruction, then we will unmap
5541 * the instruction and go to an infinite loop. So, we cache the
5542 * last retried eip and the last fault address, if we meet the eip
5543 * and the address again, we can break out of the potential infinite
5546 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5548 if (!(emulation_type & EMULTYPE_RETRY))
5551 if (x86_page_table_writing_insn(ctxt))
5554 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5557 vcpu->arch.last_retry_eip = ctxt->eip;
5558 vcpu->arch.last_retry_addr = cr2;
5560 if (!vcpu->arch.mmu.direct_map)
5561 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5563 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5568 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5569 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5571 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5573 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5574 /* This is a good place to trace that we are exiting SMM. */
5575 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5577 /* Process a latched INIT or SMI, if any. */
5578 kvm_make_request(KVM_REQ_EVENT, vcpu);
5581 kvm_mmu_reset_context(vcpu);
5584 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5586 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5588 vcpu->arch.hflags = emul_flags;
5590 if (changed & HF_SMM_MASK)
5591 kvm_smm_changed(vcpu);
5594 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5603 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5604 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5609 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
5611 struct kvm_run *kvm_run = vcpu->run;
5613 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5614 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5615 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5616 kvm_run->debug.arch.exception = DB_VECTOR;
5617 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5618 *r = EMULATE_USER_EXIT;
5621 * "Certain debug exceptions may clear bit 0-3. The
5622 * remaining contents of the DR6 register are never
5623 * cleared by the processor".
5625 vcpu->arch.dr6 &= ~15;
5626 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5627 kvm_queue_exception(vcpu, DB_VECTOR);
5631 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5633 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5634 int r = EMULATE_DONE;
5636 kvm_x86_ops->skip_emulated_instruction(vcpu);
5639 * rflags is the old, "raw" value of the flags. The new value has
5640 * not been saved yet.
5642 * This is correct even for TF set by the guest, because "the
5643 * processor will not generate this exception after the instruction
5644 * that sets the TF flag".
5646 if (unlikely(rflags & X86_EFLAGS_TF))
5647 kvm_vcpu_do_singlestep(vcpu, &r);
5648 return r == EMULATE_DONE;
5650 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5652 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5654 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5655 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5656 struct kvm_run *kvm_run = vcpu->run;
5657 unsigned long eip = kvm_get_linear_rip(vcpu);
5658 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5659 vcpu->arch.guest_debug_dr7,
5663 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5664 kvm_run->debug.arch.pc = eip;
5665 kvm_run->debug.arch.exception = DB_VECTOR;
5666 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5667 *r = EMULATE_USER_EXIT;
5672 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5673 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5674 unsigned long eip = kvm_get_linear_rip(vcpu);
5675 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5680 vcpu->arch.dr6 &= ~15;
5681 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5682 kvm_queue_exception(vcpu, DB_VECTOR);
5691 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5698 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5699 bool writeback = true;
5700 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5703 * Clear write_fault_to_shadow_pgtable here to ensure it is
5706 vcpu->arch.write_fault_to_shadow_pgtable = false;
5707 kvm_clear_exception_queue(vcpu);
5709 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5710 init_emulate_ctxt(vcpu);
5713 * We will reenter on the same instruction since
5714 * we do not set complete_userspace_io. This does not
5715 * handle watchpoints yet, those would be handled in
5718 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5721 ctxt->interruptibility = 0;
5722 ctxt->have_exception = false;
5723 ctxt->exception.vector = -1;
5724 ctxt->perm_ok = false;
5726 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5728 r = x86_decode_insn(ctxt, insn, insn_len);
5730 trace_kvm_emulate_insn_start(vcpu);
5731 ++vcpu->stat.insn_emulation;
5732 if (r != EMULATION_OK) {
5733 if (emulation_type & EMULTYPE_TRAP_UD)
5734 return EMULATE_FAIL;
5735 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5737 return EMULATE_DONE;
5738 if (ctxt->have_exception && inject_emulated_exception(vcpu))
5739 return EMULATE_DONE;
5740 if (emulation_type & EMULTYPE_SKIP)
5741 return EMULATE_FAIL;
5742 return handle_emulation_failure(vcpu);
5746 if (emulation_type & EMULTYPE_SKIP) {
5747 kvm_rip_write(vcpu, ctxt->_eip);
5748 if (ctxt->eflags & X86_EFLAGS_RF)
5749 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5750 return EMULATE_DONE;
5753 if (retry_instruction(ctxt, cr2, emulation_type))
5754 return EMULATE_DONE;
5756 /* this is needed for vmware backdoor interface to work since it
5757 changes registers values during IO operation */
5758 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5759 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5760 emulator_invalidate_register_cache(ctxt);
5764 /* Save the faulting GPA (cr2) in the address field */
5765 ctxt->exception.address = cr2;
5767 r = x86_emulate_insn(ctxt);
5769 if (r == EMULATION_INTERCEPTED)
5770 return EMULATE_DONE;
5772 if (r == EMULATION_FAILED) {
5773 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5775 return EMULATE_DONE;
5777 return handle_emulation_failure(vcpu);
5780 if (ctxt->have_exception) {
5782 if (inject_emulated_exception(vcpu))
5784 } else if (vcpu->arch.pio.count) {
5785 if (!vcpu->arch.pio.in) {
5786 /* FIXME: return into emulator if single-stepping. */
5787 vcpu->arch.pio.count = 0;
5790 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5792 r = EMULATE_USER_EXIT;
5793 } else if (vcpu->mmio_needed) {
5794 if (!vcpu->mmio_is_write)
5796 r = EMULATE_USER_EXIT;
5797 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5798 } else if (r == EMULATION_RESTART)
5804 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5805 toggle_interruptibility(vcpu, ctxt->interruptibility);
5806 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5807 kvm_rip_write(vcpu, ctxt->eip);
5808 if (r == EMULATE_DONE &&
5809 (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
5810 kvm_vcpu_do_singlestep(vcpu, &r);
5811 if (!ctxt->have_exception ||
5812 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5813 __kvm_set_rflags(vcpu, ctxt->eflags);
5816 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5817 * do nothing, and it will be requested again as soon as
5818 * the shadow expires. But we still need to check here,
5819 * because POPF has no interrupt shadow.
5821 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5822 kvm_make_request(KVM_REQ_EVENT, vcpu);
5824 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5828 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5830 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5832 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5833 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5834 size, port, &val, 1);
5835 /* do not return to emulator after return from userspace */
5836 vcpu->arch.pio.count = 0;
5839 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5841 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5845 /* We should only ever be called with arch.pio.count equal to 1 */
5846 BUG_ON(vcpu->arch.pio.count != 1);
5848 /* For size less than 4 we merge, else we zero extend */
5849 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5853 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5854 * the copy and tracing
5856 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5857 vcpu->arch.pio.port, &val, 1);
5858 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5863 int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5868 /* For size less than 4 we merge, else we zero extend */
5869 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5871 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5874 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5878 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5882 EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5884 static int kvmclock_cpu_down_prep(unsigned int cpu)
5886 __this_cpu_write(cpu_tsc_khz, 0);
5890 static void tsc_khz_changed(void *data)
5892 struct cpufreq_freqs *freq = data;
5893 unsigned long khz = 0;
5897 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5898 khz = cpufreq_quick_get(raw_smp_processor_id());
5901 __this_cpu_write(cpu_tsc_khz, khz);
5904 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5907 struct cpufreq_freqs *freq = data;
5909 struct kvm_vcpu *vcpu;
5910 int i, send_ipi = 0;
5913 * We allow guests to temporarily run on slowing clocks,
5914 * provided we notify them after, or to run on accelerating
5915 * clocks, provided we notify them before. Thus time never
5918 * However, we have a problem. We can't atomically update
5919 * the frequency of a given CPU from this function; it is
5920 * merely a notifier, which can be called from any CPU.
5921 * Changing the TSC frequency at arbitrary points in time
5922 * requires a recomputation of local variables related to
5923 * the TSC for each VCPU. We must flag these local variables
5924 * to be updated and be sure the update takes place with the
5925 * new frequency before any guests proceed.
5927 * Unfortunately, the combination of hotplug CPU and frequency
5928 * change creates an intractable locking scenario; the order
5929 * of when these callouts happen is undefined with respect to
5930 * CPU hotplug, and they can race with each other. As such,
5931 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5932 * undefined; you can actually have a CPU frequency change take
5933 * place in between the computation of X and the setting of the
5934 * variable. To protect against this problem, all updates of
5935 * the per_cpu tsc_khz variable are done in an interrupt
5936 * protected IPI, and all callers wishing to update the value
5937 * must wait for a synchronous IPI to complete (which is trivial
5938 * if the caller is on the CPU already). This establishes the
5939 * necessary total order on variable updates.
5941 * Note that because a guest time update may take place
5942 * anytime after the setting of the VCPU's request bit, the
5943 * correct TSC value must be set before the request. However,
5944 * to ensure the update actually makes it to any guest which
5945 * starts running in hardware virtualization between the set
5946 * and the acquisition of the spinlock, we must also ping the
5947 * CPU after setting the request bit.
5951 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5953 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5956 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5958 spin_lock(&kvm_lock);
5959 list_for_each_entry(kvm, &vm_list, vm_list) {
5960 kvm_for_each_vcpu(i, vcpu, kvm) {
5961 if (vcpu->cpu != freq->cpu)
5963 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5964 if (vcpu->cpu != smp_processor_id())
5968 spin_unlock(&kvm_lock);
5970 if (freq->old < freq->new && send_ipi) {
5972 * We upscale the frequency. Must make the guest
5973 * doesn't see old kvmclock values while running with
5974 * the new frequency, otherwise we risk the guest sees
5975 * time go backwards.
5977 * In case we update the frequency for another cpu
5978 * (which might be in guest context) send an interrupt
5979 * to kick the cpu out of guest context. Next time
5980 * guest context is entered kvmclock will be updated,
5981 * so the guest will not see stale values.
5983 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5988 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5989 .notifier_call = kvmclock_cpufreq_notifier
5992 static int kvmclock_cpu_online(unsigned int cpu)
5994 tsc_khz_changed(NULL);
5998 static void kvm_timer_init(void)
6000 max_tsc_khz = tsc_khz;
6002 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6003 #ifdef CONFIG_CPU_FREQ
6004 struct cpufreq_policy policy;
6007 memset(&policy, 0, sizeof(policy));
6009 cpufreq_get_policy(&policy, cpu);
6010 if (policy.cpuinfo.max_freq)
6011 max_tsc_khz = policy.cpuinfo.max_freq;
6014 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6015 CPUFREQ_TRANSITION_NOTIFIER);
6017 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6019 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6020 kvmclock_cpu_online, kvmclock_cpu_down_prep);
6023 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6025 int kvm_is_in_guest(void)
6027 return __this_cpu_read(current_vcpu) != NULL;
6030 static int kvm_is_user_mode(void)
6034 if (__this_cpu_read(current_vcpu))
6035 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6037 return user_mode != 0;
6040 static unsigned long kvm_get_guest_ip(void)
6042 unsigned long ip = 0;
6044 if (__this_cpu_read(current_vcpu))
6045 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6050 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6051 .is_in_guest = kvm_is_in_guest,
6052 .is_user_mode = kvm_is_user_mode,
6053 .get_guest_ip = kvm_get_guest_ip,
6056 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
6058 __this_cpu_write(current_vcpu, vcpu);
6060 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
6062 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
6064 __this_cpu_write(current_vcpu, NULL);
6066 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
6068 static void kvm_set_mmio_spte_mask(void)
6071 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6074 * Set the reserved bits and the present bit of an paging-structure
6075 * entry to generate page fault with PFER.RSV = 1.
6077 /* Mask the reserved physical address bits. */
6078 mask = rsvd_bits(maxphyaddr, 51);
6080 /* Set the present bit. */
6083 #ifdef CONFIG_X86_64
6085 * If reserved bit is not supported, clear the present bit to disable
6088 if (maxphyaddr == 52)
6092 kvm_mmu_set_mmio_spte_mask(mask, mask);
6095 #ifdef CONFIG_X86_64
6096 static void pvclock_gtod_update_fn(struct work_struct *work)
6100 struct kvm_vcpu *vcpu;
6103 spin_lock(&kvm_lock);
6104 list_for_each_entry(kvm, &vm_list, vm_list)
6105 kvm_for_each_vcpu(i, vcpu, kvm)
6106 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6107 atomic_set(&kvm_guest_has_master_clock, 0);
6108 spin_unlock(&kvm_lock);
6111 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6114 * Notification about pvclock gtod data update.
6116 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6119 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6120 struct timekeeper *tk = priv;
6122 update_pvclock_gtod(tk);
6124 /* disable master clock if host does not trust, or does not
6125 * use, TSC clocksource
6127 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6128 atomic_read(&kvm_guest_has_master_clock) != 0)
6129 queue_work(system_long_wq, &pvclock_gtod_work);
6134 static struct notifier_block pvclock_gtod_notifier = {
6135 .notifier_call = pvclock_gtod_notify,
6139 int kvm_arch_init(void *opaque)
6142 struct kvm_x86_ops *ops = opaque;
6145 printk(KERN_ERR "kvm: already loaded the other module\n");
6150 if (!ops->cpu_has_kvm_support()) {
6151 printk(KERN_ERR "kvm: no hardware support\n");
6155 if (ops->disabled_by_bios()) {
6156 printk(KERN_ERR "kvm: disabled by bios\n");
6162 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6164 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6168 r = kvm_mmu_module_init();
6170 goto out_free_percpu;
6172 kvm_set_mmio_spte_mask();
6176 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6177 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6178 PT_PRESENT_MASK, 0, sme_me_mask);
6181 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6183 if (boot_cpu_has(X86_FEATURE_XSAVE))
6184 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6187 #ifdef CONFIG_X86_64
6188 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6194 free_percpu(shared_msrs);
6199 void kvm_arch_exit(void)
6202 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6204 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6205 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6206 CPUFREQ_TRANSITION_NOTIFIER);
6207 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6208 #ifdef CONFIG_X86_64
6209 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6212 kvm_mmu_module_exit();
6213 free_percpu(shared_msrs);
6216 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6218 ++vcpu->stat.halt_exits;
6219 if (lapic_in_kernel(vcpu)) {
6220 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6223 vcpu->run->exit_reason = KVM_EXIT_HLT;
6227 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6229 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6231 int ret = kvm_skip_emulated_instruction(vcpu);
6233 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6234 * KVM_EXIT_DEBUG here.
6236 return kvm_vcpu_halt(vcpu) && ret;
6238 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6240 #ifdef CONFIG_X86_64
6241 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6242 unsigned long clock_type)
6244 struct kvm_clock_pairing clock_pairing;
6249 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6250 return -KVM_EOPNOTSUPP;
6252 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6253 return -KVM_EOPNOTSUPP;
6255 clock_pairing.sec = ts.tv_sec;
6256 clock_pairing.nsec = ts.tv_nsec;
6257 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6258 clock_pairing.flags = 0;
6261 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6262 sizeof(struct kvm_clock_pairing)))
6270 * kvm_pv_kick_cpu_op: Kick a vcpu.
6272 * @apicid - apicid of vcpu to be kicked.
6274 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6276 struct kvm_lapic_irq lapic_irq;
6278 lapic_irq.shorthand = 0;
6279 lapic_irq.dest_mode = 0;
6280 lapic_irq.level = 0;
6281 lapic_irq.dest_id = apicid;
6282 lapic_irq.msi_redir_hint = false;
6284 lapic_irq.delivery_mode = APIC_DM_REMRD;
6285 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6288 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6290 vcpu->arch.apicv_active = false;
6291 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6294 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6296 unsigned long nr, a0, a1, a2, a3, ret;
6299 r = kvm_skip_emulated_instruction(vcpu);
6301 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6302 return kvm_hv_hypercall(vcpu);
6304 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6305 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6306 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6307 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6308 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6310 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6312 op_64_bit = is_64_bit_mode(vcpu);
6321 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6327 case KVM_HC_VAPIC_POLL_IRQ:
6330 case KVM_HC_KICK_CPU:
6331 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6334 #ifdef CONFIG_X86_64
6335 case KVM_HC_CLOCK_PAIRING:
6336 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6346 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6347 ++vcpu->stat.hypercalls;
6350 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6352 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6354 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6355 char instruction[3];
6356 unsigned long rip = kvm_rip_read(vcpu);
6358 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6360 return emulator_write_emulated(ctxt, rip, instruction, 3,
6364 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6366 return vcpu->run->request_interrupt_window &&
6367 likely(!pic_in_kernel(vcpu->kvm));
6370 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6372 struct kvm_run *kvm_run = vcpu->run;
6374 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6375 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6376 kvm_run->cr8 = kvm_get_cr8(vcpu);
6377 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6378 kvm_run->ready_for_interrupt_injection =
6379 pic_in_kernel(vcpu->kvm) ||
6380 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6383 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6387 if (!kvm_x86_ops->update_cr8_intercept)
6390 if (!lapic_in_kernel(vcpu))
6393 if (vcpu->arch.apicv_active)
6396 if (!vcpu->arch.apic->vapic_addr)
6397 max_irr = kvm_lapic_find_highest_irr(vcpu);
6404 tpr = kvm_lapic_get_cr8(vcpu);
6406 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6409 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6413 /* try to reinject previous events if any */
6414 if (vcpu->arch.exception.injected) {
6415 kvm_x86_ops->queue_exception(vcpu);
6420 * Exceptions must be injected immediately, or the exception
6421 * frame will have the address of the NMI or interrupt handler.
6423 if (!vcpu->arch.exception.pending) {
6424 if (vcpu->arch.nmi_injected) {
6425 kvm_x86_ops->set_nmi(vcpu);
6429 if (vcpu->arch.interrupt.pending) {
6430 kvm_x86_ops->set_irq(vcpu);
6435 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6436 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6441 /* try to inject new event if pending */
6442 if (vcpu->arch.exception.pending) {
6443 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6444 vcpu->arch.exception.has_error_code,
6445 vcpu->arch.exception.error_code);
6447 vcpu->arch.exception.pending = false;
6448 vcpu->arch.exception.injected = true;
6450 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6451 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6454 if (vcpu->arch.exception.nr == DB_VECTOR &&
6455 (vcpu->arch.dr7 & DR7_GD)) {
6456 vcpu->arch.dr7 &= ~DR7_GD;
6457 kvm_update_dr7(vcpu);
6460 kvm_x86_ops->queue_exception(vcpu);
6461 } else if (vcpu->arch.smi_pending && !is_smm(vcpu) && kvm_x86_ops->smi_allowed(vcpu)) {
6462 vcpu->arch.smi_pending = false;
6463 ++vcpu->arch.smi_count;
6465 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6466 --vcpu->arch.nmi_pending;
6467 vcpu->arch.nmi_injected = true;
6468 kvm_x86_ops->set_nmi(vcpu);
6469 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6471 * Because interrupts can be injected asynchronously, we are
6472 * calling check_nested_events again here to avoid a race condition.
6473 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6474 * proposal and current concerns. Perhaps we should be setting
6475 * KVM_REQ_EVENT only on certain events and not unconditionally?
6477 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6478 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6482 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6483 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6485 kvm_x86_ops->set_irq(vcpu);
6492 static void process_nmi(struct kvm_vcpu *vcpu)
6497 * x86 is limited to one NMI running, and one NMI pending after it.
6498 * If an NMI is already in progress, limit further NMIs to just one.
6499 * Otherwise, allow two (and we'll inject the first one immediately).
6501 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6504 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6505 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6506 kvm_make_request(KVM_REQ_EVENT, vcpu);
6509 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6512 flags |= seg->g << 23;
6513 flags |= seg->db << 22;
6514 flags |= seg->l << 21;
6515 flags |= seg->avl << 20;
6516 flags |= seg->present << 15;
6517 flags |= seg->dpl << 13;
6518 flags |= seg->s << 12;
6519 flags |= seg->type << 8;
6523 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6525 struct kvm_segment seg;
6528 kvm_get_segment(vcpu, &seg, n);
6529 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6532 offset = 0x7f84 + n * 12;
6534 offset = 0x7f2c + (n - 3) * 12;
6536 put_smstate(u32, buf, offset + 8, seg.base);
6537 put_smstate(u32, buf, offset + 4, seg.limit);
6538 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6541 #ifdef CONFIG_X86_64
6542 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6544 struct kvm_segment seg;
6548 kvm_get_segment(vcpu, &seg, n);
6549 offset = 0x7e00 + n * 16;
6551 flags = enter_smm_get_segment_flags(&seg) >> 8;
6552 put_smstate(u16, buf, offset, seg.selector);
6553 put_smstate(u16, buf, offset + 2, flags);
6554 put_smstate(u32, buf, offset + 4, seg.limit);
6555 put_smstate(u64, buf, offset + 8, seg.base);
6559 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6562 struct kvm_segment seg;
6566 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6567 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6568 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6569 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6571 for (i = 0; i < 8; i++)
6572 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6574 kvm_get_dr(vcpu, 6, &val);
6575 put_smstate(u32, buf, 0x7fcc, (u32)val);
6576 kvm_get_dr(vcpu, 7, &val);
6577 put_smstate(u32, buf, 0x7fc8, (u32)val);
6579 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6580 put_smstate(u32, buf, 0x7fc4, seg.selector);
6581 put_smstate(u32, buf, 0x7f64, seg.base);
6582 put_smstate(u32, buf, 0x7f60, seg.limit);
6583 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6585 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6586 put_smstate(u32, buf, 0x7fc0, seg.selector);
6587 put_smstate(u32, buf, 0x7f80, seg.base);
6588 put_smstate(u32, buf, 0x7f7c, seg.limit);
6589 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6591 kvm_x86_ops->get_gdt(vcpu, &dt);
6592 put_smstate(u32, buf, 0x7f74, dt.address);
6593 put_smstate(u32, buf, 0x7f70, dt.size);
6595 kvm_x86_ops->get_idt(vcpu, &dt);
6596 put_smstate(u32, buf, 0x7f58, dt.address);
6597 put_smstate(u32, buf, 0x7f54, dt.size);
6599 for (i = 0; i < 6; i++)
6600 enter_smm_save_seg_32(vcpu, buf, i);
6602 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6605 put_smstate(u32, buf, 0x7efc, 0x00020000);
6606 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6609 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6611 #ifdef CONFIG_X86_64
6613 struct kvm_segment seg;
6617 for (i = 0; i < 16; i++)
6618 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6620 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6621 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6623 kvm_get_dr(vcpu, 6, &val);
6624 put_smstate(u64, buf, 0x7f68, val);
6625 kvm_get_dr(vcpu, 7, &val);
6626 put_smstate(u64, buf, 0x7f60, val);
6628 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6629 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6630 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6632 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6635 put_smstate(u32, buf, 0x7efc, 0x00020064);
6637 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6639 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6640 put_smstate(u16, buf, 0x7e90, seg.selector);
6641 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6642 put_smstate(u32, buf, 0x7e94, seg.limit);
6643 put_smstate(u64, buf, 0x7e98, seg.base);
6645 kvm_x86_ops->get_idt(vcpu, &dt);
6646 put_smstate(u32, buf, 0x7e84, dt.size);
6647 put_smstate(u64, buf, 0x7e88, dt.address);
6649 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6650 put_smstate(u16, buf, 0x7e70, seg.selector);
6651 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6652 put_smstate(u32, buf, 0x7e74, seg.limit);
6653 put_smstate(u64, buf, 0x7e78, seg.base);
6655 kvm_x86_ops->get_gdt(vcpu, &dt);
6656 put_smstate(u32, buf, 0x7e64, dt.size);
6657 put_smstate(u64, buf, 0x7e68, dt.address);
6659 for (i = 0; i < 6; i++)
6660 enter_smm_save_seg_64(vcpu, buf, i);
6666 static void enter_smm(struct kvm_vcpu *vcpu)
6668 struct kvm_segment cs, ds;
6673 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6674 memset(buf, 0, 512);
6675 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
6676 enter_smm_save_state_64(vcpu, buf);
6678 enter_smm_save_state_32(vcpu, buf);
6681 * Give pre_enter_smm() a chance to make ISA-specific changes to the
6682 * vCPU state (e.g. leave guest mode) after we've saved the state into
6683 * the SMM state-save area.
6685 kvm_x86_ops->pre_enter_smm(vcpu, buf);
6687 vcpu->arch.hflags |= HF_SMM_MASK;
6688 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6690 if (kvm_x86_ops->get_nmi_mask(vcpu))
6691 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6693 kvm_x86_ops->set_nmi_mask(vcpu, true);
6695 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6696 kvm_rip_write(vcpu, 0x8000);
6698 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6699 kvm_x86_ops->set_cr0(vcpu, cr0);
6700 vcpu->arch.cr0 = cr0;
6702 kvm_x86_ops->set_cr4(vcpu, 0);
6704 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6705 dt.address = dt.size = 0;
6706 kvm_x86_ops->set_idt(vcpu, &dt);
6708 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6710 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6711 cs.base = vcpu->arch.smbase;
6716 cs.limit = ds.limit = 0xffffffff;
6717 cs.type = ds.type = 0x3;
6718 cs.dpl = ds.dpl = 0;
6723 cs.avl = ds.avl = 0;
6724 cs.present = ds.present = 1;
6725 cs.unusable = ds.unusable = 0;
6726 cs.padding = ds.padding = 0;
6728 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6729 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6730 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6731 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6732 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6733 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6735 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
6736 kvm_x86_ops->set_efer(vcpu, 0);
6738 kvm_update_cpuid(vcpu);
6739 kvm_mmu_reset_context(vcpu);
6742 static void process_smi(struct kvm_vcpu *vcpu)
6744 vcpu->arch.smi_pending = true;
6745 kvm_make_request(KVM_REQ_EVENT, vcpu);
6748 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6750 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6753 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6755 u64 eoi_exit_bitmap[4];
6757 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6760 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6762 if (irqchip_split(vcpu->kvm))
6763 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6765 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6766 kvm_x86_ops->sync_pir_to_irr(vcpu);
6767 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6769 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6770 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6771 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6774 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6776 ++vcpu->stat.tlb_flush;
6777 kvm_x86_ops->tlb_flush(vcpu);
6780 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
6781 unsigned long start, unsigned long end)
6783 unsigned long apic_address;
6786 * The physical address of apic access page is stored in the VMCS.
6787 * Update it when it becomes invalid.
6789 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6790 if (start <= apic_address && apic_address < end)
6791 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6794 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6796 struct page *page = NULL;
6798 if (!lapic_in_kernel(vcpu))
6801 if (!kvm_x86_ops->set_apic_access_page_addr)
6804 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6805 if (is_error_page(page))
6807 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6810 * Do not pin apic access page in memory, the MMU notifier
6811 * will call us again if it is migrated or swapped out.
6815 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6818 * Returns 1 to let vcpu_run() continue the guest execution loop without
6819 * exiting to the userspace. Otherwise, the value will be returned to the
6822 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6826 dm_request_for_irq_injection(vcpu) &&
6827 kvm_cpu_accept_dm_intr(vcpu);
6829 bool req_immediate_exit = false;
6831 if (kvm_request_pending(vcpu)) {
6832 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6833 kvm_mmu_unload(vcpu);
6834 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6835 __kvm_migrate_timers(vcpu);
6836 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6837 kvm_gen_update_masterclock(vcpu->kvm);
6838 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6839 kvm_gen_kvmclock_update(vcpu);
6840 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6841 r = kvm_guest_time_update(vcpu);
6845 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6846 kvm_mmu_sync_roots(vcpu);
6847 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6848 kvm_vcpu_flush_tlb(vcpu);
6849 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6850 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6854 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6855 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6856 vcpu->mmio_needed = 0;
6860 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6861 /* Page is swapped out. Do synthetic halt */
6862 vcpu->arch.apf.halted = true;
6866 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6867 record_steal_time(vcpu);
6868 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6870 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6872 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6873 kvm_pmu_handle_event(vcpu);
6874 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6875 kvm_pmu_deliver_pmi(vcpu);
6876 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6877 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6878 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6879 vcpu->arch.ioapic_handled_vectors)) {
6880 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6881 vcpu->run->eoi.vector =
6882 vcpu->arch.pending_ioapic_eoi;
6887 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6888 vcpu_scan_ioapic(vcpu);
6889 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6890 kvm_vcpu_reload_apic_access_page(vcpu);
6891 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6892 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6893 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6897 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6898 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6899 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6903 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6904 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6905 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6911 * KVM_REQ_HV_STIMER has to be processed after
6912 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6913 * depend on the guest clock being up-to-date
6915 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6916 kvm_hv_process_stimers(vcpu);
6919 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6920 ++vcpu->stat.req_event;
6921 kvm_apic_accept_events(vcpu);
6922 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6927 if (inject_pending_event(vcpu, req_int_win) != 0)
6928 req_immediate_exit = true;
6930 /* Enable SMI/NMI/IRQ window open exits if needed.
6932 * SMIs have three cases:
6933 * 1) They can be nested, and then there is nothing to
6934 * do here because RSM will cause a vmexit anyway.
6935 * 2) There is an ISA-specific reason why SMI cannot be
6936 * injected, and the moment when this changes can be
6938 * 3) Or the SMI can be pending because
6939 * inject_pending_event has completed the injection
6940 * of an IRQ or NMI from the previous vmexit, and
6941 * then we request an immediate exit to inject the
6944 if (vcpu->arch.smi_pending && !is_smm(vcpu))
6945 if (!kvm_x86_ops->enable_smi_window(vcpu))
6946 req_immediate_exit = true;
6947 if (vcpu->arch.nmi_pending)
6948 kvm_x86_ops->enable_nmi_window(vcpu);
6949 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6950 kvm_x86_ops->enable_irq_window(vcpu);
6951 WARN_ON(vcpu->arch.exception.pending);
6954 if (kvm_lapic_enabled(vcpu)) {
6955 update_cr8_intercept(vcpu);
6956 kvm_lapic_sync_to_vapic(vcpu);
6960 r = kvm_mmu_reload(vcpu);
6962 goto cancel_injection;
6967 kvm_x86_ops->prepare_guest_switch(vcpu);
6970 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
6971 * IPI are then delayed after guest entry, which ensures that they
6972 * result in virtual interrupt delivery.
6974 local_irq_disable();
6975 vcpu->mode = IN_GUEST_MODE;
6977 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6980 * 1) We should set ->mode before checking ->requests. Please see
6981 * the comment in kvm_vcpu_exiting_guest_mode().
6983 * 2) For APICv, we should set ->mode before checking PIR.ON. This
6984 * pairs with the memory barrier implicit in pi_test_and_set_on
6985 * (see vmx_deliver_posted_interrupt).
6987 * 3) This also orders the write to mode from any reads to the page
6988 * tables done while the VCPU is running. Please see the comment
6989 * in kvm_flush_remote_tlbs.
6991 smp_mb__after_srcu_read_unlock();
6994 * This handles the case where a posted interrupt was
6995 * notified with kvm_vcpu_kick.
6997 if (kvm_lapic_enabled(vcpu)) {
6998 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6999 kvm_x86_ops->sync_pir_to_irr(vcpu);
7002 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7003 || need_resched() || signal_pending(current)) {
7004 vcpu->mode = OUTSIDE_GUEST_MODE;
7008 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7010 goto cancel_injection;
7013 kvm_load_guest_xcr0(vcpu);
7015 if (req_immediate_exit) {
7016 kvm_make_request(KVM_REQ_EVENT, vcpu);
7017 smp_send_reschedule(vcpu->cpu);
7020 trace_kvm_entry(vcpu->vcpu_id);
7021 if (lapic_timer_advance_ns)
7022 wait_lapic_expire(vcpu);
7023 guest_enter_irqoff();
7025 if (unlikely(vcpu->arch.switch_db_regs)) {
7027 set_debugreg(vcpu->arch.eff_db[0], 0);
7028 set_debugreg(vcpu->arch.eff_db[1], 1);
7029 set_debugreg(vcpu->arch.eff_db[2], 2);
7030 set_debugreg(vcpu->arch.eff_db[3], 3);
7031 set_debugreg(vcpu->arch.dr6, 6);
7032 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7035 kvm_x86_ops->run(vcpu);
7038 * Do this here before restoring debug registers on the host. And
7039 * since we do this before handling the vmexit, a DR access vmexit
7040 * can (a) read the correct value of the debug registers, (b) set
7041 * KVM_DEBUGREG_WONT_EXIT again.
7043 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7044 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7045 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7046 kvm_update_dr0123(vcpu);
7047 kvm_update_dr6(vcpu);
7048 kvm_update_dr7(vcpu);
7049 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7053 * If the guest has used debug registers, at least dr7
7054 * will be disabled while returning to the host.
7055 * If we don't have active breakpoints in the host, we don't
7056 * care about the messed up debug address registers. But if
7057 * we have some of them active, restore the old state.
7059 if (hw_breakpoint_active())
7060 hw_breakpoint_restore();
7062 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7064 vcpu->mode = OUTSIDE_GUEST_MODE;
7067 kvm_put_guest_xcr0(vcpu);
7069 kvm_x86_ops->handle_external_intr(vcpu);
7073 guest_exit_irqoff();
7078 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7081 * Profile KVM exit RIPs:
7083 if (unlikely(prof_on == KVM_PROFILING)) {
7084 unsigned long rip = kvm_rip_read(vcpu);
7085 profile_hit(KVM_PROFILING, (void *)rip);
7088 if (unlikely(vcpu->arch.tsc_always_catchup))
7089 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7091 if (vcpu->arch.apic_attention)
7092 kvm_lapic_sync_from_vapic(vcpu);
7094 vcpu->arch.gpa_available = false;
7095 r = kvm_x86_ops->handle_exit(vcpu);
7099 kvm_x86_ops->cancel_injection(vcpu);
7100 if (unlikely(vcpu->arch.apic_attention))
7101 kvm_lapic_sync_from_vapic(vcpu);
7106 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7108 if (!kvm_arch_vcpu_runnable(vcpu) &&
7109 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7110 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7111 kvm_vcpu_block(vcpu);
7112 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7114 if (kvm_x86_ops->post_block)
7115 kvm_x86_ops->post_block(vcpu);
7117 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7121 kvm_apic_accept_events(vcpu);
7122 switch(vcpu->arch.mp_state) {
7123 case KVM_MP_STATE_HALTED:
7124 vcpu->arch.pv.pv_unhalted = false;
7125 vcpu->arch.mp_state =
7126 KVM_MP_STATE_RUNNABLE;
7127 case KVM_MP_STATE_RUNNABLE:
7128 vcpu->arch.apf.halted = false;
7130 case KVM_MP_STATE_INIT_RECEIVED:
7139 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7141 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7142 kvm_x86_ops->check_nested_events(vcpu, false);
7144 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7145 !vcpu->arch.apf.halted);
7148 static int vcpu_run(struct kvm_vcpu *vcpu)
7151 struct kvm *kvm = vcpu->kvm;
7153 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7156 if (kvm_vcpu_running(vcpu)) {
7157 r = vcpu_enter_guest(vcpu);
7159 r = vcpu_block(kvm, vcpu);
7165 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7166 if (kvm_cpu_has_pending_timer(vcpu))
7167 kvm_inject_pending_timer_irqs(vcpu);
7169 if (dm_request_for_irq_injection(vcpu) &&
7170 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7172 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7173 ++vcpu->stat.request_irq_exits;
7177 kvm_check_async_pf_completion(vcpu);
7179 if (signal_pending(current)) {
7181 vcpu->run->exit_reason = KVM_EXIT_INTR;
7182 ++vcpu->stat.signal_exits;
7185 if (need_resched()) {
7186 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7188 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7192 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7197 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7200 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7201 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7202 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7203 if (r != EMULATE_DONE)
7208 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7210 BUG_ON(!vcpu->arch.pio.count);
7212 return complete_emulated_io(vcpu);
7216 * Implements the following, as a state machine:
7220 * for each mmio piece in the fragment
7228 * for each mmio piece in the fragment
7233 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7235 struct kvm_run *run = vcpu->run;
7236 struct kvm_mmio_fragment *frag;
7239 BUG_ON(!vcpu->mmio_needed);
7241 /* Complete previous fragment */
7242 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7243 len = min(8u, frag->len);
7244 if (!vcpu->mmio_is_write)
7245 memcpy(frag->data, run->mmio.data, len);
7247 if (frag->len <= 8) {
7248 /* Switch to the next fragment. */
7250 vcpu->mmio_cur_fragment++;
7252 /* Go forward to the next mmio piece. */
7258 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7259 vcpu->mmio_needed = 0;
7261 /* FIXME: return into emulator if single-stepping. */
7262 if (vcpu->mmio_is_write)
7264 vcpu->mmio_read_completed = 1;
7265 return complete_emulated_io(vcpu);
7268 run->exit_reason = KVM_EXIT_MMIO;
7269 run->mmio.phys_addr = frag->gpa;
7270 if (vcpu->mmio_is_write)
7271 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7272 run->mmio.len = min(8u, frag->len);
7273 run->mmio.is_write = vcpu->mmio_is_write;
7274 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7279 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7283 kvm_sigset_activate(vcpu);
7285 kvm_load_guest_fpu(vcpu);
7287 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7288 if (kvm_run->immediate_exit) {
7292 kvm_vcpu_block(vcpu);
7293 kvm_apic_accept_events(vcpu);
7294 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7296 if (signal_pending(current)) {
7298 vcpu->run->exit_reason = KVM_EXIT_INTR;
7299 ++vcpu->stat.signal_exits;
7304 /* re-sync apic's tpr */
7305 if (!lapic_in_kernel(vcpu)) {
7306 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7312 if (unlikely(vcpu->arch.complete_userspace_io)) {
7313 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7314 vcpu->arch.complete_userspace_io = NULL;
7319 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7321 if (kvm_run->immediate_exit)
7327 kvm_put_guest_fpu(vcpu);
7328 post_kvm_run_save(vcpu);
7329 kvm_sigset_deactivate(vcpu);
7334 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7336 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7338 * We are here if userspace calls get_regs() in the middle of
7339 * instruction emulation. Registers state needs to be copied
7340 * back from emulation context to vcpu. Userspace shouldn't do
7341 * that usually, but some bad designed PV devices (vmware
7342 * backdoor interface) need this to work
7344 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7345 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7347 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7348 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7349 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7350 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7351 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7352 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7353 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7354 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7355 #ifdef CONFIG_X86_64
7356 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7357 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7358 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7359 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7360 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7361 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7362 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7363 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7366 regs->rip = kvm_rip_read(vcpu);
7367 regs->rflags = kvm_get_rflags(vcpu);
7372 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7374 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7375 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7377 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7378 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7379 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7380 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7381 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7382 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7383 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7384 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7385 #ifdef CONFIG_X86_64
7386 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7387 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7388 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7389 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7390 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7391 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7392 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7393 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7396 kvm_rip_write(vcpu, regs->rip);
7397 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
7399 vcpu->arch.exception.pending = false;
7401 kvm_make_request(KVM_REQ_EVENT, vcpu);
7406 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7408 struct kvm_segment cs;
7410 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7414 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7416 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7417 struct kvm_sregs *sregs)
7421 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7422 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7423 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7424 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7425 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7426 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7428 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7429 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7431 kvm_x86_ops->get_idt(vcpu, &dt);
7432 sregs->idt.limit = dt.size;
7433 sregs->idt.base = dt.address;
7434 kvm_x86_ops->get_gdt(vcpu, &dt);
7435 sregs->gdt.limit = dt.size;
7436 sregs->gdt.base = dt.address;
7438 sregs->cr0 = kvm_read_cr0(vcpu);
7439 sregs->cr2 = vcpu->arch.cr2;
7440 sregs->cr3 = kvm_read_cr3(vcpu);
7441 sregs->cr4 = kvm_read_cr4(vcpu);
7442 sregs->cr8 = kvm_get_cr8(vcpu);
7443 sregs->efer = vcpu->arch.efer;
7444 sregs->apic_base = kvm_get_apic_base(vcpu);
7446 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7448 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7449 set_bit(vcpu->arch.interrupt.nr,
7450 (unsigned long *)sregs->interrupt_bitmap);
7455 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7456 struct kvm_mp_state *mp_state)
7458 kvm_apic_accept_events(vcpu);
7459 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7460 vcpu->arch.pv.pv_unhalted)
7461 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7463 mp_state->mp_state = vcpu->arch.mp_state;
7468 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7469 struct kvm_mp_state *mp_state)
7471 if (!lapic_in_kernel(vcpu) &&
7472 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7475 /* INITs are latched while in SMM */
7476 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7477 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7478 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7481 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7482 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7483 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7485 vcpu->arch.mp_state = mp_state->mp_state;
7486 kvm_make_request(KVM_REQ_EVENT, vcpu);
7490 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7491 int reason, bool has_error_code, u32 error_code)
7493 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7496 init_emulate_ctxt(vcpu);
7498 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7499 has_error_code, error_code);
7502 return EMULATE_FAIL;
7504 kvm_rip_write(vcpu, ctxt->eip);
7505 kvm_set_rflags(vcpu, ctxt->eflags);
7506 kvm_make_request(KVM_REQ_EVENT, vcpu);
7507 return EMULATE_DONE;
7509 EXPORT_SYMBOL_GPL(kvm_task_switch);
7511 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7512 struct kvm_sregs *sregs)
7514 struct msr_data apic_base_msr;
7515 int mmu_reset_needed = 0;
7516 int pending_vec, max_bits, idx;
7519 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7520 (sregs->cr4 & X86_CR4_OSXSAVE))
7523 apic_base_msr.data = sregs->apic_base;
7524 apic_base_msr.host_initiated = true;
7525 if (kvm_set_apic_base(vcpu, &apic_base_msr))
7528 dt.size = sregs->idt.limit;
7529 dt.address = sregs->idt.base;
7530 kvm_x86_ops->set_idt(vcpu, &dt);
7531 dt.size = sregs->gdt.limit;
7532 dt.address = sregs->gdt.base;
7533 kvm_x86_ops->set_gdt(vcpu, &dt);
7535 vcpu->arch.cr2 = sregs->cr2;
7536 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7537 vcpu->arch.cr3 = sregs->cr3;
7538 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7540 kvm_set_cr8(vcpu, sregs->cr8);
7542 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7543 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7545 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7546 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7547 vcpu->arch.cr0 = sregs->cr0;
7549 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7550 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7551 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7552 kvm_update_cpuid(vcpu);
7554 idx = srcu_read_lock(&vcpu->kvm->srcu);
7555 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7556 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7557 mmu_reset_needed = 1;
7559 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7561 if (mmu_reset_needed)
7562 kvm_mmu_reset_context(vcpu);
7564 max_bits = KVM_NR_INTERRUPTS;
7565 pending_vec = find_first_bit(
7566 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7567 if (pending_vec < max_bits) {
7568 kvm_queue_interrupt(vcpu, pending_vec, false);
7569 pr_debug("Set back pending irq %d\n", pending_vec);
7572 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7573 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7574 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7575 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7576 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7577 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7579 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7580 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7582 update_cr8_intercept(vcpu);
7584 /* Older userspace won't unhalt the vcpu on reset. */
7585 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7586 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7588 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7590 kvm_make_request(KVM_REQ_EVENT, vcpu);
7595 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7596 struct kvm_guest_debug *dbg)
7598 unsigned long rflags;
7601 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7603 if (vcpu->arch.exception.pending)
7605 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7606 kvm_queue_exception(vcpu, DB_VECTOR);
7608 kvm_queue_exception(vcpu, BP_VECTOR);
7612 * Read rflags as long as potentially injected trace flags are still
7615 rflags = kvm_get_rflags(vcpu);
7617 vcpu->guest_debug = dbg->control;
7618 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7619 vcpu->guest_debug = 0;
7621 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7622 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7623 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7624 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7626 for (i = 0; i < KVM_NR_DB_REGS; i++)
7627 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7629 kvm_update_dr7(vcpu);
7631 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7632 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7633 get_segment_base(vcpu, VCPU_SREG_CS);
7636 * Trigger an rflags update that will inject or remove the trace
7639 kvm_set_rflags(vcpu, rflags);
7641 kvm_x86_ops->update_bp_intercept(vcpu);
7651 * Translate a guest virtual address to a guest physical address.
7653 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7654 struct kvm_translation *tr)
7656 unsigned long vaddr = tr->linear_address;
7660 idx = srcu_read_lock(&vcpu->kvm->srcu);
7661 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7662 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7663 tr->physical_address = gpa;
7664 tr->valid = gpa != UNMAPPED_GVA;
7671 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7673 struct fxregs_state *fxsave =
7674 &vcpu->arch.guest_fpu.state.fxsave;
7676 memcpy(fpu->fpr, fxsave->st_space, 128);
7677 fpu->fcw = fxsave->cwd;
7678 fpu->fsw = fxsave->swd;
7679 fpu->ftwx = fxsave->twd;
7680 fpu->last_opcode = fxsave->fop;
7681 fpu->last_ip = fxsave->rip;
7682 fpu->last_dp = fxsave->rdp;
7683 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7688 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7690 struct fxregs_state *fxsave =
7691 &vcpu->arch.guest_fpu.state.fxsave;
7693 memcpy(fxsave->st_space, fpu->fpr, 128);
7694 fxsave->cwd = fpu->fcw;
7695 fxsave->swd = fpu->fsw;
7696 fxsave->twd = fpu->ftwx;
7697 fxsave->fop = fpu->last_opcode;
7698 fxsave->rip = fpu->last_ip;
7699 fxsave->rdp = fpu->last_dp;
7700 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7705 static void fx_init(struct kvm_vcpu *vcpu)
7707 fpstate_init(&vcpu->arch.guest_fpu.state);
7708 if (boot_cpu_has(X86_FEATURE_XSAVES))
7709 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7710 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7713 * Ensure guest xcr0 is valid for loading
7715 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7717 vcpu->arch.cr0 |= X86_CR0_ET;
7720 /* Swap (qemu) user FPU context for the guest FPU context. */
7721 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7724 copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
7725 /* PKRU is separately restored in kvm_x86_ops->run. */
7726 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
7727 ~XFEATURE_MASK_PKRU);
7732 /* When vcpu_run ends, restore user space FPU context. */
7733 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7736 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7737 copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
7739 ++vcpu->stat.fpu_reload;
7743 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7745 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7747 kvmclock_reset(vcpu);
7749 kvm_x86_ops->vcpu_free(vcpu);
7750 free_cpumask_var(wbinvd_dirty_mask);
7753 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7756 struct kvm_vcpu *vcpu;
7758 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7759 printk_once(KERN_WARNING
7760 "kvm: SMP vm created on host with unstable TSC; "
7761 "guest TSC will not be reliable\n");
7763 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7768 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7770 kvm_vcpu_mtrr_init(vcpu);
7772 kvm_vcpu_reset(vcpu, false);
7773 kvm_mmu_setup(vcpu);
7778 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7780 struct msr_data msr;
7781 struct kvm *kvm = vcpu->kvm;
7783 kvm_hv_vcpu_postcreate(vcpu);
7785 if (mutex_lock_killable(&vcpu->mutex))
7789 msr.index = MSR_IA32_TSC;
7790 msr.host_initiated = true;
7791 kvm_write_tsc(vcpu, &msr);
7793 mutex_unlock(&vcpu->mutex);
7795 if (!kvmclock_periodic_sync)
7798 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7799 KVMCLOCK_SYNC_PERIOD);
7802 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7804 vcpu->arch.apf.msr_val = 0;
7807 kvm_mmu_unload(vcpu);
7810 kvm_x86_ops->vcpu_free(vcpu);
7813 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7815 vcpu->arch.hflags = 0;
7817 vcpu->arch.smi_pending = 0;
7818 vcpu->arch.smi_count = 0;
7819 atomic_set(&vcpu->arch.nmi_queued, 0);
7820 vcpu->arch.nmi_pending = 0;
7821 vcpu->arch.nmi_injected = false;
7822 kvm_clear_interrupt_queue(vcpu);
7823 kvm_clear_exception_queue(vcpu);
7824 vcpu->arch.exception.pending = false;
7826 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7827 kvm_update_dr0123(vcpu);
7828 vcpu->arch.dr6 = DR6_INIT;
7829 kvm_update_dr6(vcpu);
7830 vcpu->arch.dr7 = DR7_FIXED_1;
7831 kvm_update_dr7(vcpu);
7835 kvm_make_request(KVM_REQ_EVENT, vcpu);
7836 vcpu->arch.apf.msr_val = 0;
7837 vcpu->arch.st.msr_val = 0;
7839 kvmclock_reset(vcpu);
7841 kvm_clear_async_pf_completion_queue(vcpu);
7842 kvm_async_pf_hash_reset(vcpu);
7843 vcpu->arch.apf.halted = false;
7845 if (kvm_mpx_supported()) {
7846 void *mpx_state_buffer;
7849 * To avoid have the INIT path from kvm_apic_has_events() that be
7850 * called with loaded FPU and does not let userspace fix the state.
7853 kvm_put_guest_fpu(vcpu);
7854 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
7855 XFEATURE_MASK_BNDREGS);
7856 if (mpx_state_buffer)
7857 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
7858 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
7859 XFEATURE_MASK_BNDCSR);
7860 if (mpx_state_buffer)
7861 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
7863 kvm_load_guest_fpu(vcpu);
7867 kvm_pmu_reset(vcpu);
7868 vcpu->arch.smbase = 0x30000;
7870 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
7871 vcpu->arch.msr_misc_features_enables = 0;
7873 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7876 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7877 vcpu->arch.regs_avail = ~0;
7878 vcpu->arch.regs_dirty = ~0;
7880 vcpu->arch.ia32_xss = 0;
7882 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7885 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7887 struct kvm_segment cs;
7889 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7890 cs.selector = vector << 8;
7891 cs.base = vector << 12;
7892 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7893 kvm_rip_write(vcpu, 0);
7896 int kvm_arch_hardware_enable(void)
7899 struct kvm_vcpu *vcpu;
7904 bool stable, backwards_tsc = false;
7906 kvm_shared_msr_cpu_online();
7907 ret = kvm_x86_ops->hardware_enable();
7911 local_tsc = rdtsc();
7912 stable = !check_tsc_unstable();
7913 list_for_each_entry(kvm, &vm_list, vm_list) {
7914 kvm_for_each_vcpu(i, vcpu, kvm) {
7915 if (!stable && vcpu->cpu == smp_processor_id())
7916 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7917 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7918 backwards_tsc = true;
7919 if (vcpu->arch.last_host_tsc > max_tsc)
7920 max_tsc = vcpu->arch.last_host_tsc;
7926 * Sometimes, even reliable TSCs go backwards. This happens on
7927 * platforms that reset TSC during suspend or hibernate actions, but
7928 * maintain synchronization. We must compensate. Fortunately, we can
7929 * detect that condition here, which happens early in CPU bringup,
7930 * before any KVM threads can be running. Unfortunately, we can't
7931 * bring the TSCs fully up to date with real time, as we aren't yet far
7932 * enough into CPU bringup that we know how much real time has actually
7933 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
7934 * variables that haven't been updated yet.
7936 * So we simply find the maximum observed TSC above, then record the
7937 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7938 * the adjustment will be applied. Note that we accumulate
7939 * adjustments, in case multiple suspend cycles happen before some VCPU
7940 * gets a chance to run again. In the event that no KVM threads get a
7941 * chance to run, we will miss the entire elapsed period, as we'll have
7942 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7943 * loose cycle time. This isn't too big a deal, since the loss will be
7944 * uniform across all VCPUs (not to mention the scenario is extremely
7945 * unlikely). It is possible that a second hibernate recovery happens
7946 * much faster than a first, causing the observed TSC here to be
7947 * smaller; this would require additional padding adjustment, which is
7948 * why we set last_host_tsc to the local tsc observed here.
7950 * N.B. - this code below runs only on platforms with reliable TSC,
7951 * as that is the only way backwards_tsc is set above. Also note
7952 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7953 * have the same delta_cyc adjustment applied if backwards_tsc
7954 * is detected. Note further, this adjustment is only done once,
7955 * as we reset last_host_tsc on all VCPUs to stop this from being
7956 * called multiple times (one for each physical CPU bringup).
7958 * Platforms with unreliable TSCs don't have to deal with this, they
7959 * will be compensated by the logic in vcpu_load, which sets the TSC to
7960 * catchup mode. This will catchup all VCPUs to real time, but cannot
7961 * guarantee that they stay in perfect synchronization.
7963 if (backwards_tsc) {
7964 u64 delta_cyc = max_tsc - local_tsc;
7965 list_for_each_entry(kvm, &vm_list, vm_list) {
7966 kvm->arch.backwards_tsc_observed = true;
7967 kvm_for_each_vcpu(i, vcpu, kvm) {
7968 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7969 vcpu->arch.last_host_tsc = local_tsc;
7970 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7974 * We have to disable TSC offset matching.. if you were
7975 * booting a VM while issuing an S4 host suspend....
7976 * you may have some problem. Solving this issue is
7977 * left as an exercise to the reader.
7979 kvm->arch.last_tsc_nsec = 0;
7980 kvm->arch.last_tsc_write = 0;
7987 void kvm_arch_hardware_disable(void)
7989 kvm_x86_ops->hardware_disable();
7990 drop_user_return_notifiers();
7993 int kvm_arch_hardware_setup(void)
7997 r = kvm_x86_ops->hardware_setup();
8001 if (kvm_has_tsc_control) {
8003 * Make sure the user can only configure tsc_khz values that
8004 * fit into a signed integer.
8005 * A min value is not calculated needed because it will always
8006 * be 1 on all machines.
8008 u64 max = min(0x7fffffffULL,
8009 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8010 kvm_max_guest_tsc_khz = max;
8012 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8015 kvm_init_msr_list();
8019 void kvm_arch_hardware_unsetup(void)
8021 kvm_x86_ops->hardware_unsetup();
8024 void kvm_arch_check_processor_compat(void *rtn)
8026 kvm_x86_ops->check_processor_compatibility(rtn);
8029 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8031 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8033 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8035 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8037 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8040 struct static_key kvm_no_apic_vcpu __read_mostly;
8041 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8043 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8048 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8049 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8050 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8051 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8053 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8055 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8060 vcpu->arch.pio_data = page_address(page);
8062 kvm_set_tsc_khz(vcpu, max_tsc_khz);
8064 r = kvm_mmu_create(vcpu);
8066 goto fail_free_pio_data;
8068 if (irqchip_in_kernel(vcpu->kvm)) {
8069 r = kvm_create_lapic(vcpu);
8071 goto fail_mmu_destroy;
8073 static_key_slow_inc(&kvm_no_apic_vcpu);
8075 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8077 if (!vcpu->arch.mce_banks) {
8079 goto fail_free_lapic;
8081 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8083 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8085 goto fail_free_mce_banks;
8090 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8092 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8094 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8096 kvm_async_pf_hash_reset(vcpu);
8099 vcpu->arch.pending_external_vector = -1;
8100 vcpu->arch.preempted_in_kernel = false;
8102 kvm_hv_vcpu_init(vcpu);
8106 fail_free_mce_banks:
8107 kfree(vcpu->arch.mce_banks);
8109 kvm_free_lapic(vcpu);
8111 kvm_mmu_destroy(vcpu);
8113 free_page((unsigned long)vcpu->arch.pio_data);
8118 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8122 kvm_hv_vcpu_uninit(vcpu);
8123 kvm_pmu_destroy(vcpu);
8124 kfree(vcpu->arch.mce_banks);
8125 kvm_free_lapic(vcpu);
8126 idx = srcu_read_lock(&vcpu->kvm->srcu);
8127 kvm_mmu_destroy(vcpu);
8128 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8129 free_page((unsigned long)vcpu->arch.pio_data);
8130 if (!lapic_in_kernel(vcpu))
8131 static_key_slow_dec(&kvm_no_apic_vcpu);
8134 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8136 kvm_x86_ops->sched_in(vcpu, cpu);
8139 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8144 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8145 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8146 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8147 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8148 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8150 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8151 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8152 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8153 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8154 &kvm->arch.irq_sources_bitmap);
8156 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8157 mutex_init(&kvm->arch.apic_map_lock);
8158 mutex_init(&kvm->arch.hyperv.hv_lock);
8159 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8161 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8162 pvclock_update_vm_gtod_copy(kvm);
8164 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8165 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8167 kvm_page_track_init(kvm);
8168 kvm_mmu_init_vm(kvm);
8170 if (kvm_x86_ops->vm_init)
8171 return kvm_x86_ops->vm_init(kvm);
8176 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8179 kvm_mmu_unload(vcpu);
8183 static void kvm_free_vcpus(struct kvm *kvm)
8186 struct kvm_vcpu *vcpu;
8189 * Unpin any mmu pages first.
8191 kvm_for_each_vcpu(i, vcpu, kvm) {
8192 kvm_clear_async_pf_completion_queue(vcpu);
8193 kvm_unload_vcpu_mmu(vcpu);
8195 kvm_for_each_vcpu(i, vcpu, kvm)
8196 kvm_arch_vcpu_free(vcpu);
8198 mutex_lock(&kvm->lock);
8199 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8200 kvm->vcpus[i] = NULL;
8202 atomic_set(&kvm->online_vcpus, 0);
8203 mutex_unlock(&kvm->lock);
8206 void kvm_arch_sync_events(struct kvm *kvm)
8208 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8209 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8213 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8217 struct kvm_memslots *slots = kvm_memslots(kvm);
8218 struct kvm_memory_slot *slot, old;
8220 /* Called with kvm->slots_lock held. */
8221 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8224 slot = id_to_memslot(slots, id);
8230 * MAP_SHARED to prevent internal slot pages from being moved
8233 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8234 MAP_SHARED | MAP_ANONYMOUS, 0);
8235 if (IS_ERR((void *)hva))
8236 return PTR_ERR((void *)hva);
8245 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8246 struct kvm_userspace_memory_region m;
8248 m.slot = id | (i << 16);
8250 m.guest_phys_addr = gpa;
8251 m.userspace_addr = hva;
8252 m.memory_size = size;
8253 r = __kvm_set_memory_region(kvm, &m);
8259 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8265 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8267 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8271 mutex_lock(&kvm->slots_lock);
8272 r = __x86_set_memory_region(kvm, id, gpa, size);
8273 mutex_unlock(&kvm->slots_lock);
8277 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8279 void kvm_arch_destroy_vm(struct kvm *kvm)
8281 if (current->mm == kvm->mm) {
8283 * Free memory regions allocated on behalf of userspace,
8284 * unless the the memory map has changed due to process exit
8287 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8288 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8289 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8291 if (kvm_x86_ops->vm_destroy)
8292 kvm_x86_ops->vm_destroy(kvm);
8293 kvm_pic_destroy(kvm);
8294 kvm_ioapic_destroy(kvm);
8295 kvm_free_vcpus(kvm);
8296 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8297 kvm_mmu_uninit_vm(kvm);
8298 kvm_page_track_cleanup(kvm);
8301 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8302 struct kvm_memory_slot *dont)
8306 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8307 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8308 kvfree(free->arch.rmap[i]);
8309 free->arch.rmap[i] = NULL;
8314 if (!dont || free->arch.lpage_info[i - 1] !=
8315 dont->arch.lpage_info[i - 1]) {
8316 kvfree(free->arch.lpage_info[i - 1]);
8317 free->arch.lpage_info[i - 1] = NULL;
8321 kvm_page_track_free_memslot(free, dont);
8324 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8325 unsigned long npages)
8329 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8330 struct kvm_lpage_info *linfo;
8335 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8336 slot->base_gfn, level) + 1;
8338 slot->arch.rmap[i] =
8339 kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8340 if (!slot->arch.rmap[i])
8345 linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8349 slot->arch.lpage_info[i - 1] = linfo;
8351 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8352 linfo[0].disallow_lpage = 1;
8353 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8354 linfo[lpages - 1].disallow_lpage = 1;
8355 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8357 * If the gfn and userspace address are not aligned wrt each
8358 * other, or if explicitly asked to, disable large page
8359 * support for this slot
8361 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8362 !kvm_largepages_enabled()) {
8365 for (j = 0; j < lpages; ++j)
8366 linfo[j].disallow_lpage = 1;
8370 if (kvm_page_track_create_memslot(slot, npages))
8376 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8377 kvfree(slot->arch.rmap[i]);
8378 slot->arch.rmap[i] = NULL;
8382 kvfree(slot->arch.lpage_info[i - 1]);
8383 slot->arch.lpage_info[i - 1] = NULL;
8388 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8391 * memslots->generation has been incremented.
8392 * mmio generation may have reached its maximum value.
8394 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8397 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8398 struct kvm_memory_slot *memslot,
8399 const struct kvm_userspace_memory_region *mem,
8400 enum kvm_mr_change change)
8405 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8406 struct kvm_memory_slot *new)
8408 /* Still write protect RO slot */
8409 if (new->flags & KVM_MEM_READONLY) {
8410 kvm_mmu_slot_remove_write_access(kvm, new);
8415 * Call kvm_x86_ops dirty logging hooks when they are valid.
8417 * kvm_x86_ops->slot_disable_log_dirty is called when:
8419 * - KVM_MR_CREATE with dirty logging is disabled
8420 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8422 * The reason is, in case of PML, we need to set D-bit for any slots
8423 * with dirty logging disabled in order to eliminate unnecessary GPA
8424 * logging in PML buffer (and potential PML buffer full VMEXT). This
8425 * guarantees leaving PML enabled during guest's lifetime won't have
8426 * any additonal overhead from PML when guest is running with dirty
8427 * logging disabled for memory slots.
8429 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8430 * to dirty logging mode.
8432 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8434 * In case of write protect:
8436 * Write protect all pages for dirty logging.
8438 * All the sptes including the large sptes which point to this
8439 * slot are set to readonly. We can not create any new large
8440 * spte on this slot until the end of the logging.
8442 * See the comments in fast_page_fault().
8444 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8445 if (kvm_x86_ops->slot_enable_log_dirty)
8446 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8448 kvm_mmu_slot_remove_write_access(kvm, new);
8450 if (kvm_x86_ops->slot_disable_log_dirty)
8451 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8455 void kvm_arch_commit_memory_region(struct kvm *kvm,
8456 const struct kvm_userspace_memory_region *mem,
8457 const struct kvm_memory_slot *old,
8458 const struct kvm_memory_slot *new,
8459 enum kvm_mr_change change)
8461 int nr_mmu_pages = 0;
8463 if (!kvm->arch.n_requested_mmu_pages)
8464 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8467 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8470 * Dirty logging tracks sptes in 4k granularity, meaning that large
8471 * sptes have to be split. If live migration is successful, the guest
8472 * in the source machine will be destroyed and large sptes will be
8473 * created in the destination. However, if the guest continues to run
8474 * in the source machine (for example if live migration fails), small
8475 * sptes will remain around and cause bad performance.
8477 * Scan sptes if dirty logging has been stopped, dropping those
8478 * which can be collapsed into a single large-page spte. Later
8479 * page faults will create the large-page sptes.
8481 if ((change != KVM_MR_DELETE) &&
8482 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8483 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8484 kvm_mmu_zap_collapsible_sptes(kvm, new);
8487 * Set up write protection and/or dirty logging for the new slot.
8489 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8490 * been zapped so no dirty logging staff is needed for old slot. For
8491 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8492 * new and it's also covered when dealing with the new slot.
8494 * FIXME: const-ify all uses of struct kvm_memory_slot.
8496 if (change != KVM_MR_DELETE)
8497 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8500 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8502 kvm_mmu_invalidate_zap_all_pages(kvm);
8505 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8506 struct kvm_memory_slot *slot)
8508 kvm_page_track_flush_slot(kvm, slot);
8511 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8513 if (!list_empty_careful(&vcpu->async_pf.done))
8516 if (kvm_apic_has_events(vcpu))
8519 if (vcpu->arch.pv.pv_unhalted)
8522 if (vcpu->arch.exception.pending)
8525 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
8526 (vcpu->arch.nmi_pending &&
8527 kvm_x86_ops->nmi_allowed(vcpu)))
8530 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
8531 (vcpu->arch.smi_pending && !is_smm(vcpu)))
8534 if (kvm_arch_interrupt_allowed(vcpu) &&
8535 kvm_cpu_has_interrupt(vcpu))
8538 if (kvm_hv_has_stimer_pending(vcpu))
8544 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8546 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8549 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
8551 return vcpu->arch.preempted_in_kernel;
8554 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8556 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8559 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8561 return kvm_x86_ops->interrupt_allowed(vcpu);
8564 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8566 if (is_64_bit_mode(vcpu))
8567 return kvm_rip_read(vcpu);
8568 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8569 kvm_rip_read(vcpu));
8571 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8573 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8575 return kvm_get_linear_rip(vcpu) == linear_rip;
8577 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8579 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8581 unsigned long rflags;
8583 rflags = kvm_x86_ops->get_rflags(vcpu);
8584 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8585 rflags &= ~X86_EFLAGS_TF;
8588 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8590 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8592 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8593 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8594 rflags |= X86_EFLAGS_TF;
8595 kvm_x86_ops->set_rflags(vcpu, rflags);
8598 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8600 __kvm_set_rflags(vcpu, rflags);
8601 kvm_make_request(KVM_REQ_EVENT, vcpu);
8603 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8605 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8609 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8613 r = kvm_mmu_reload(vcpu);
8617 if (!vcpu->arch.mmu.direct_map &&
8618 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8621 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8624 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8626 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8629 static inline u32 kvm_async_pf_next_probe(u32 key)
8631 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8634 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8636 u32 key = kvm_async_pf_hash_fn(gfn);
8638 while (vcpu->arch.apf.gfns[key] != ~0)
8639 key = kvm_async_pf_next_probe(key);
8641 vcpu->arch.apf.gfns[key] = gfn;
8644 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8647 u32 key = kvm_async_pf_hash_fn(gfn);
8649 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8650 (vcpu->arch.apf.gfns[key] != gfn &&
8651 vcpu->arch.apf.gfns[key] != ~0); i++)
8652 key = kvm_async_pf_next_probe(key);
8657 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8659 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8662 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8666 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8668 vcpu->arch.apf.gfns[i] = ~0;
8670 j = kvm_async_pf_next_probe(j);
8671 if (vcpu->arch.apf.gfns[j] == ~0)
8673 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8675 * k lies cyclically in ]i,j]
8677 * |....j i.k.| or |.k..j i...|
8679 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8680 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8685 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8688 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8692 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
8695 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
8699 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8700 struct kvm_async_pf *work)
8702 struct x86_exception fault;
8704 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8705 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8707 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8708 (vcpu->arch.apf.send_user_only &&
8709 kvm_x86_ops->get_cpl(vcpu) == 0))
8710 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8711 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8712 fault.vector = PF_VECTOR;
8713 fault.error_code_valid = true;
8714 fault.error_code = 0;
8715 fault.nested_page_fault = false;
8716 fault.address = work->arch.token;
8717 fault.async_page_fault = true;
8718 kvm_inject_page_fault(vcpu, &fault);
8722 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8723 struct kvm_async_pf *work)
8725 struct x86_exception fault;
8728 if (work->wakeup_all)
8729 work->arch.token = ~0; /* broadcast wakeup */
8731 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8732 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8734 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
8735 !apf_get_user(vcpu, &val)) {
8736 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
8737 vcpu->arch.exception.pending &&
8738 vcpu->arch.exception.nr == PF_VECTOR &&
8739 !apf_put_user(vcpu, 0)) {
8740 vcpu->arch.exception.injected = false;
8741 vcpu->arch.exception.pending = false;
8742 vcpu->arch.exception.nr = 0;
8743 vcpu->arch.exception.has_error_code = false;
8744 vcpu->arch.exception.error_code = 0;
8745 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8746 fault.vector = PF_VECTOR;
8747 fault.error_code_valid = true;
8748 fault.error_code = 0;
8749 fault.nested_page_fault = false;
8750 fault.address = work->arch.token;
8751 fault.async_page_fault = true;
8752 kvm_inject_page_fault(vcpu, &fault);
8755 vcpu->arch.apf.halted = false;
8756 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8759 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8761 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8764 return kvm_can_do_async_pf(vcpu);
8767 void kvm_arch_start_assignment(struct kvm *kvm)
8769 atomic_inc(&kvm->arch.assigned_device_count);
8771 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8773 void kvm_arch_end_assignment(struct kvm *kvm)
8775 atomic_dec(&kvm->arch.assigned_device_count);
8777 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8779 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8781 return atomic_read(&kvm->arch.assigned_device_count);
8783 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8785 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8787 atomic_inc(&kvm->arch.noncoherent_dma_count);
8789 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8791 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8793 atomic_dec(&kvm->arch.noncoherent_dma_count);
8795 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8797 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8799 return atomic_read(&kvm->arch.noncoherent_dma_count);
8801 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8803 bool kvm_arch_has_irq_bypass(void)
8805 return kvm_x86_ops->update_pi_irte != NULL;
8808 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8809 struct irq_bypass_producer *prod)
8811 struct kvm_kernel_irqfd *irqfd =
8812 container_of(cons, struct kvm_kernel_irqfd, consumer);
8814 irqfd->producer = prod;
8816 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8817 prod->irq, irqfd->gsi, 1);
8820 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8821 struct irq_bypass_producer *prod)
8824 struct kvm_kernel_irqfd *irqfd =
8825 container_of(cons, struct kvm_kernel_irqfd, consumer);
8827 WARN_ON(irqfd->producer != prod);
8828 irqfd->producer = NULL;
8831 * When producer of consumer is unregistered, we change back to
8832 * remapped mode, so we can re-use the current implementation
8833 * when the irq is masked/disabled or the consumer side (KVM
8834 * int this case doesn't want to receive the interrupts.
8836 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8838 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8839 " fails: %d\n", irqfd->consumer.token, ret);
8842 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8843 uint32_t guest_irq, bool set)
8845 if (!kvm_x86_ops->update_pi_irte)
8848 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8851 bool kvm_vector_hashing_enabled(void)
8853 return vector_hashing;
8855 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8857 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8858 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8859 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8860 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8861 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8862 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8863 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8864 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8865 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8866 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8867 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8868 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8869 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8870 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8871 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8872 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8873 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8874 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8875 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);