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 unsigned int min_timer_period_us = 500;
111 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
113 static bool __read_mostly kvmclock_periodic_sync = true;
114 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
116 bool __read_mostly kvm_has_tsc_control;
117 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
118 u32 __read_mostly kvm_max_guest_tsc_khz;
119 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
120 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
121 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
122 u64 __read_mostly kvm_max_tsc_scaling_ratio;
123 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
124 u64 __read_mostly kvm_default_tsc_scaling_ratio;
125 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
127 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
128 static u32 __read_mostly tsc_tolerance_ppm = 250;
129 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
131 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
132 unsigned int __read_mostly lapic_timer_advance_ns = 0;
133 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
135 static bool __read_mostly vector_hashing = true;
136 module_param(vector_hashing, bool, S_IRUGO);
138 #define KVM_NR_SHARED_MSRS 16
140 struct kvm_shared_msrs_global {
142 u32 msrs[KVM_NR_SHARED_MSRS];
145 struct kvm_shared_msrs {
146 struct user_return_notifier urn;
148 struct kvm_shared_msr_values {
151 } values[KVM_NR_SHARED_MSRS];
154 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
155 static struct kvm_shared_msrs __percpu *shared_msrs;
157 struct kvm_stats_debugfs_item debugfs_entries[] = {
158 { "pf_fixed", VCPU_STAT(pf_fixed) },
159 { "pf_guest", VCPU_STAT(pf_guest) },
160 { "tlb_flush", VCPU_STAT(tlb_flush) },
161 { "invlpg", VCPU_STAT(invlpg) },
162 { "exits", VCPU_STAT(exits) },
163 { "io_exits", VCPU_STAT(io_exits) },
164 { "mmio_exits", VCPU_STAT(mmio_exits) },
165 { "signal_exits", VCPU_STAT(signal_exits) },
166 { "irq_window", VCPU_STAT(irq_window_exits) },
167 { "nmi_window", VCPU_STAT(nmi_window_exits) },
168 { "halt_exits", VCPU_STAT(halt_exits) },
169 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
170 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
171 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
172 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
173 { "hypercalls", VCPU_STAT(hypercalls) },
174 { "request_irq", VCPU_STAT(request_irq_exits) },
175 { "irq_exits", VCPU_STAT(irq_exits) },
176 { "host_state_reload", VCPU_STAT(host_state_reload) },
177 { "efer_reload", VCPU_STAT(efer_reload) },
178 { "fpu_reload", VCPU_STAT(fpu_reload) },
179 { "insn_emulation", VCPU_STAT(insn_emulation) },
180 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
181 { "irq_injections", VCPU_STAT(irq_injections) },
182 { "nmi_injections", VCPU_STAT(nmi_injections) },
183 { "req_event", VCPU_STAT(req_event) },
184 { "l1d_flush", VCPU_STAT(l1d_flush) },
185 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
186 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
187 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
188 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
189 { "mmu_flooded", VM_STAT(mmu_flooded) },
190 { "mmu_recycled", VM_STAT(mmu_recycled) },
191 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
192 { "mmu_unsync", VM_STAT(mmu_unsync) },
193 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
194 { "largepages", VM_STAT(lpages) },
195 { "max_mmu_page_hash_collisions",
196 VM_STAT(max_mmu_page_hash_collisions) },
200 u64 __read_mostly host_xcr0;
202 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
204 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
207 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
208 vcpu->arch.apf.gfns[i] = ~0;
211 static void kvm_on_user_return(struct user_return_notifier *urn)
214 struct kvm_shared_msrs *locals
215 = container_of(urn, struct kvm_shared_msrs, urn);
216 struct kvm_shared_msr_values *values;
220 * Disabling irqs at this point since the following code could be
221 * interrupted and executed through kvm_arch_hardware_disable()
223 local_irq_save(flags);
224 if (locals->registered) {
225 locals->registered = false;
226 user_return_notifier_unregister(urn);
228 local_irq_restore(flags);
229 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
230 values = &locals->values[slot];
231 if (values->host != values->curr) {
232 wrmsrl(shared_msrs_global.msrs[slot], values->host);
233 values->curr = values->host;
238 static void shared_msr_update(unsigned slot, u32 msr)
241 unsigned int cpu = smp_processor_id();
242 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
244 /* only read, and nobody should modify it at this time,
245 * so don't need lock */
246 if (slot >= shared_msrs_global.nr) {
247 printk(KERN_ERR "kvm: invalid MSR slot!");
250 rdmsrl_safe(msr, &value);
251 smsr->values[slot].host = value;
252 smsr->values[slot].curr = value;
255 void kvm_define_shared_msr(unsigned slot, u32 msr)
257 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
258 shared_msrs_global.msrs[slot] = msr;
259 if (slot >= shared_msrs_global.nr)
260 shared_msrs_global.nr = slot + 1;
262 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
264 static void kvm_shared_msr_cpu_online(void)
268 for (i = 0; i < shared_msrs_global.nr; ++i)
269 shared_msr_update(i, shared_msrs_global.msrs[i]);
272 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
274 unsigned int cpu = smp_processor_id();
275 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
278 if (((value ^ smsr->values[slot].curr) & mask) == 0)
280 smsr->values[slot].curr = value;
281 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
285 if (!smsr->registered) {
286 smsr->urn.on_user_return = kvm_on_user_return;
287 user_return_notifier_register(&smsr->urn);
288 smsr->registered = true;
292 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
294 static void drop_user_return_notifiers(void)
296 unsigned int cpu = smp_processor_id();
297 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
299 if (smsr->registered)
300 kvm_on_user_return(&smsr->urn);
303 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
305 return vcpu->arch.apic_base;
307 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
309 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
311 u64 old_state = vcpu->arch.apic_base &
312 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
313 u64 new_state = msr_info->data &
314 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
315 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
316 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
318 if ((msr_info->data & reserved_bits) || new_state == X2APIC_ENABLE)
320 if (!msr_info->host_initiated &&
321 ((new_state == MSR_IA32_APICBASE_ENABLE &&
322 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
323 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
327 kvm_lapic_set_base(vcpu, msr_info->data);
330 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
332 asmlinkage __visible void kvm_spurious_fault(void)
334 /* Fault while not rebooting. We want the trace. */
337 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
339 #define EXCPT_BENIGN 0
340 #define EXCPT_CONTRIBUTORY 1
343 static int exception_class(int vector)
353 return EXCPT_CONTRIBUTORY;
360 #define EXCPT_FAULT 0
362 #define EXCPT_ABORT 2
363 #define EXCPT_INTERRUPT 3
365 static int exception_type(int vector)
369 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
370 return EXCPT_INTERRUPT;
374 /* #DB is trap, as instruction watchpoints are handled elsewhere */
375 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
378 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
381 /* Reserved exceptions will result in fault */
385 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
386 unsigned nr, bool has_error, u32 error_code,
392 kvm_make_request(KVM_REQ_EVENT, vcpu);
394 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
396 if (has_error && !is_protmode(vcpu))
400 * On vmentry, vcpu->arch.exception.pending is only
401 * true if an event injection was blocked by
402 * nested_run_pending. In that case, however,
403 * vcpu_enter_guest requests an immediate exit,
404 * and the guest shouldn't proceed far enough to
407 WARN_ON_ONCE(vcpu->arch.exception.pending);
408 vcpu->arch.exception.injected = true;
410 vcpu->arch.exception.pending = true;
411 vcpu->arch.exception.injected = false;
413 vcpu->arch.exception.has_error_code = has_error;
414 vcpu->arch.exception.nr = nr;
415 vcpu->arch.exception.error_code = error_code;
419 /* to check exception */
420 prev_nr = vcpu->arch.exception.nr;
421 if (prev_nr == DF_VECTOR) {
422 /* triple fault -> shutdown */
423 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
426 class1 = exception_class(prev_nr);
427 class2 = exception_class(nr);
428 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
429 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
431 * Generate double fault per SDM Table 5-5. Set
432 * exception.pending = true so that the double fault
433 * can trigger a nested vmexit.
435 vcpu->arch.exception.pending = true;
436 vcpu->arch.exception.injected = false;
437 vcpu->arch.exception.has_error_code = true;
438 vcpu->arch.exception.nr = DF_VECTOR;
439 vcpu->arch.exception.error_code = 0;
441 /* replace previous exception with a new one in a hope
442 that instruction re-execution will regenerate lost
447 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
449 kvm_multiple_exception(vcpu, nr, false, 0, false);
451 EXPORT_SYMBOL_GPL(kvm_queue_exception);
453 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
455 kvm_multiple_exception(vcpu, nr, false, 0, true);
457 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
459 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
462 kvm_inject_gp(vcpu, 0);
464 return kvm_skip_emulated_instruction(vcpu);
468 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
470 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
472 ++vcpu->stat.pf_guest;
473 vcpu->arch.exception.nested_apf =
474 is_guest_mode(vcpu) && fault->async_page_fault;
475 if (vcpu->arch.exception.nested_apf)
476 vcpu->arch.apf.nested_apf_token = fault->address;
478 vcpu->arch.cr2 = fault->address;
479 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
481 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
483 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
485 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
486 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
488 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
490 return fault->nested_page_fault;
493 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
495 atomic_inc(&vcpu->arch.nmi_queued);
496 kvm_make_request(KVM_REQ_NMI, vcpu);
498 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
500 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
502 kvm_multiple_exception(vcpu, nr, true, error_code, false);
504 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
506 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
508 kvm_multiple_exception(vcpu, nr, true, error_code, true);
510 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
513 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
514 * a #GP and return false.
516 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
518 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
520 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
523 EXPORT_SYMBOL_GPL(kvm_require_cpl);
525 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
527 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
530 kvm_queue_exception(vcpu, UD_VECTOR);
533 EXPORT_SYMBOL_GPL(kvm_require_dr);
536 * This function will be used to read from the physical memory of the currently
537 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
538 * can read from guest physical or from the guest's guest physical memory.
540 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
541 gfn_t ngfn, void *data, int offset, int len,
544 struct x86_exception exception;
548 ngpa = gfn_to_gpa(ngfn);
549 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
550 if (real_gfn == UNMAPPED_GVA)
553 real_gfn = gpa_to_gfn(real_gfn);
555 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
557 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
559 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
560 void *data, int offset, int len, u32 access)
562 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
563 data, offset, len, access);
567 * Load the pae pdptrs. Return true is they are all valid.
569 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
571 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
572 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
575 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
577 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
578 offset * sizeof(u64), sizeof(pdpte),
579 PFERR_USER_MASK|PFERR_WRITE_MASK);
584 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
585 if ((pdpte[i] & PT_PRESENT_MASK) &&
587 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
594 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
595 __set_bit(VCPU_EXREG_PDPTR,
596 (unsigned long *)&vcpu->arch.regs_avail);
597 __set_bit(VCPU_EXREG_PDPTR,
598 (unsigned long *)&vcpu->arch.regs_dirty);
603 EXPORT_SYMBOL_GPL(load_pdptrs);
605 bool pdptrs_changed(struct kvm_vcpu *vcpu)
607 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
613 if (is_long_mode(vcpu) || !is_pae(vcpu))
616 if (!test_bit(VCPU_EXREG_PDPTR,
617 (unsigned long *)&vcpu->arch.regs_avail))
620 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
621 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
622 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
623 PFERR_USER_MASK | PFERR_WRITE_MASK);
626 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
631 EXPORT_SYMBOL_GPL(pdptrs_changed);
633 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
635 unsigned long old_cr0 = kvm_read_cr0(vcpu);
636 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
641 if (cr0 & 0xffffffff00000000UL)
645 cr0 &= ~CR0_RESERVED_BITS;
647 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
650 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
653 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
655 if ((vcpu->arch.efer & EFER_LME)) {
660 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
665 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
670 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
673 kvm_x86_ops->set_cr0(vcpu, cr0);
675 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
676 kvm_clear_async_pf_completion_queue(vcpu);
677 kvm_async_pf_hash_reset(vcpu);
680 if ((cr0 ^ old_cr0) & update_bits)
681 kvm_mmu_reset_context(vcpu);
683 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
684 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
685 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
686 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
690 EXPORT_SYMBOL_GPL(kvm_set_cr0);
692 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
694 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
696 EXPORT_SYMBOL_GPL(kvm_lmsw);
698 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
700 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
701 !vcpu->guest_xcr0_loaded) {
702 /* kvm_set_xcr() also depends on this */
703 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
704 vcpu->guest_xcr0_loaded = 1;
708 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
710 if (vcpu->guest_xcr0_loaded) {
711 if (vcpu->arch.xcr0 != host_xcr0)
712 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
713 vcpu->guest_xcr0_loaded = 0;
717 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
720 u64 old_xcr0 = vcpu->arch.xcr0;
723 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
724 if (index != XCR_XFEATURE_ENABLED_MASK)
726 if (!(xcr0 & XFEATURE_MASK_FP))
728 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
732 * Do not allow the guest to set bits that we do not support
733 * saving. However, xcr0 bit 0 is always set, even if the
734 * emulated CPU does not support XSAVE (see fx_init).
736 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
737 if (xcr0 & ~valid_bits)
740 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
741 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
744 if (xcr0 & XFEATURE_MASK_AVX512) {
745 if (!(xcr0 & XFEATURE_MASK_YMM))
747 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
750 vcpu->arch.xcr0 = xcr0;
752 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
753 kvm_update_cpuid(vcpu);
757 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
759 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
760 __kvm_set_xcr(vcpu, index, xcr)) {
761 kvm_inject_gp(vcpu, 0);
766 EXPORT_SYMBOL_GPL(kvm_set_xcr);
768 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
770 unsigned long old_cr4 = kvm_read_cr4(vcpu);
771 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
772 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
774 if (cr4 & CR4_RESERVED_BITS)
777 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
780 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
783 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
786 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
789 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
792 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
795 if (is_long_mode(vcpu)) {
796 if (!(cr4 & X86_CR4_PAE))
798 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
799 && ((cr4 ^ old_cr4) & pdptr_bits)
800 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
804 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
805 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
808 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
809 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
813 if (kvm_x86_ops->set_cr4(vcpu, cr4))
816 if (((cr4 ^ old_cr4) & pdptr_bits) ||
817 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
818 kvm_mmu_reset_context(vcpu);
820 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
821 kvm_update_cpuid(vcpu);
825 EXPORT_SYMBOL_GPL(kvm_set_cr4);
827 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
830 cr3 &= ~CR3_PCID_INVD;
833 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
834 kvm_mmu_sync_roots(vcpu);
835 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
839 if (is_long_mode(vcpu) &&
840 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
842 else if (is_pae(vcpu) && is_paging(vcpu) &&
843 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
846 vcpu->arch.cr3 = cr3;
847 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
848 kvm_mmu_new_cr3(vcpu);
851 EXPORT_SYMBOL_GPL(kvm_set_cr3);
853 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
855 if (cr8 & CR8_RESERVED_BITS)
857 if (lapic_in_kernel(vcpu))
858 kvm_lapic_set_tpr(vcpu, cr8);
860 vcpu->arch.cr8 = cr8;
863 EXPORT_SYMBOL_GPL(kvm_set_cr8);
865 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
867 if (lapic_in_kernel(vcpu))
868 return kvm_lapic_get_cr8(vcpu);
870 return vcpu->arch.cr8;
872 EXPORT_SYMBOL_GPL(kvm_get_cr8);
874 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
878 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
879 for (i = 0; i < KVM_NR_DB_REGS; i++)
880 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
881 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
885 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
887 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
888 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
891 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
895 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
896 dr7 = vcpu->arch.guest_debug_dr7;
898 dr7 = vcpu->arch.dr7;
899 kvm_x86_ops->set_dr7(vcpu, dr7);
900 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
901 if (dr7 & DR7_BP_EN_MASK)
902 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
905 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
907 u64 fixed = DR6_FIXED_1;
909 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
914 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
918 vcpu->arch.db[dr] = val;
919 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
920 vcpu->arch.eff_db[dr] = val;
925 if (val & 0xffffffff00000000ULL)
927 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
928 kvm_update_dr6(vcpu);
933 if (val & 0xffffffff00000000ULL)
935 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
936 kvm_update_dr7(vcpu);
943 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
945 if (__kvm_set_dr(vcpu, dr, val)) {
946 kvm_inject_gp(vcpu, 0);
951 EXPORT_SYMBOL_GPL(kvm_set_dr);
953 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
957 *val = vcpu->arch.db[dr];
962 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
963 *val = vcpu->arch.dr6;
965 *val = kvm_x86_ops->get_dr6(vcpu);
970 *val = vcpu->arch.dr7;
975 EXPORT_SYMBOL_GPL(kvm_get_dr);
977 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
979 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
983 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
986 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
987 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
990 EXPORT_SYMBOL_GPL(kvm_rdpmc);
993 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
994 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
996 * This list is modified at module load time to reflect the
997 * capabilities of the host cpu. This capabilities test skips MSRs that are
998 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
999 * may depend on host virtualization features rather than host cpu features.
1002 static u32 msrs_to_save[] = {
1003 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1005 #ifdef CONFIG_X86_64
1006 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1008 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1009 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1010 MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES
1013 static unsigned num_msrs_to_save;
1015 static u32 emulated_msrs[] = {
1016 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1017 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1018 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1019 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1020 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1021 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1022 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1024 HV_X64_MSR_VP_INDEX,
1025 HV_X64_MSR_VP_RUNTIME,
1026 HV_X64_MSR_SCONTROL,
1027 HV_X64_MSR_STIMER0_CONFIG,
1028 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1031 MSR_IA32_TSC_ADJUST,
1032 MSR_IA32_TSCDEADLINE,
1033 MSR_IA32_MISC_ENABLE,
1034 MSR_IA32_MCG_STATUS,
1036 MSR_IA32_MCG_EXT_CTL,
1039 MSR_MISC_FEATURES_ENABLES,
1040 MSR_AMD64_VIRT_SPEC_CTRL,
1043 static unsigned num_emulated_msrs;
1046 * List of msr numbers which are used to expose MSR-based features that
1047 * can be used by a hypervisor to validate requested CPU features.
1049 static u32 msr_based_features[] = {
1052 MSR_IA32_ARCH_CAPABILITIES,
1055 static unsigned int num_msr_based_features;
1057 u64 kvm_get_arch_capabilities(void)
1061 rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES, &data);
1064 * If we're doing cache flushes (either "always" or "cond")
1065 * we will do one whenever the guest does a vmlaunch/vmresume.
1066 * If an outer hypervisor is doing the cache flush for us
1067 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1068 * capability to the guest too, and if EPT is disabled we're not
1069 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1070 * require a nested hypervisor to do a flush of its own.
1072 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1073 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1077 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities);
1079 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1081 switch (msr->index) {
1082 case MSR_IA32_ARCH_CAPABILITIES:
1083 msr->data = kvm_get_arch_capabilities();
1085 case MSR_IA32_UCODE_REV:
1086 rdmsrl_safe(msr->index, &msr->data);
1089 if (kvm_x86_ops->get_msr_feature(msr))
1095 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1097 struct kvm_msr_entry msr;
1101 r = kvm_get_msr_feature(&msr);
1110 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1112 if (efer & efer_reserved_bits)
1115 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1118 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1123 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1125 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1127 u64 old_efer = vcpu->arch.efer;
1129 if (!kvm_valid_efer(vcpu, efer))
1133 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1137 efer |= vcpu->arch.efer & EFER_LMA;
1139 kvm_x86_ops->set_efer(vcpu, efer);
1141 /* Update reserved bits */
1142 if ((efer ^ old_efer) & EFER_NX)
1143 kvm_mmu_reset_context(vcpu);
1148 void kvm_enable_efer_bits(u64 mask)
1150 efer_reserved_bits &= ~mask;
1152 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1155 * Writes msr value into into the appropriate "register".
1156 * Returns 0 on success, non-0 otherwise.
1157 * Assumes vcpu_load() was already called.
1159 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1161 switch (msr->index) {
1164 case MSR_KERNEL_GS_BASE:
1167 if (is_noncanonical_address(msr->data, vcpu))
1170 case MSR_IA32_SYSENTER_EIP:
1171 case MSR_IA32_SYSENTER_ESP:
1173 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1174 * non-canonical address is written on Intel but not on
1175 * AMD (which ignores the top 32-bits, because it does
1176 * not implement 64-bit SYSENTER).
1178 * 64-bit code should hence be able to write a non-canonical
1179 * value on AMD. Making the address canonical ensures that
1180 * vmentry does not fail on Intel after writing a non-canonical
1181 * value, and that something deterministic happens if the guest
1182 * invokes 64-bit SYSENTER.
1184 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1186 return kvm_x86_ops->set_msr(vcpu, msr);
1188 EXPORT_SYMBOL_GPL(kvm_set_msr);
1191 * Adapt set_msr() to msr_io()'s calling convention
1193 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1195 struct msr_data msr;
1199 msr.host_initiated = true;
1200 r = kvm_get_msr(vcpu, &msr);
1208 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1210 struct msr_data msr;
1214 msr.host_initiated = true;
1215 return kvm_set_msr(vcpu, &msr);
1218 #ifdef CONFIG_X86_64
1219 struct pvclock_gtod_data {
1222 struct { /* extract of a clocksource struct */
1235 static struct pvclock_gtod_data pvclock_gtod_data;
1237 static void update_pvclock_gtod(struct timekeeper *tk)
1239 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1242 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1244 write_seqcount_begin(&vdata->seq);
1246 /* copy pvclock gtod data */
1247 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1248 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1249 vdata->clock.mask = tk->tkr_mono.mask;
1250 vdata->clock.mult = tk->tkr_mono.mult;
1251 vdata->clock.shift = tk->tkr_mono.shift;
1253 vdata->boot_ns = boot_ns;
1254 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1256 vdata->wall_time_sec = tk->xtime_sec;
1258 write_seqcount_end(&vdata->seq);
1262 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1265 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1266 * vcpu_enter_guest. This function is only called from
1267 * the physical CPU that is running vcpu.
1269 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1272 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1276 struct pvclock_wall_clock wc;
1277 struct timespec64 boot;
1282 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1287 ++version; /* first time write, random junk */
1291 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1295 * The guest calculates current wall clock time by adding
1296 * system time (updated by kvm_guest_time_update below) to the
1297 * wall clock specified here. guest system time equals host
1298 * system time for us, thus we must fill in host boot time here.
1300 getboottime64(&boot);
1302 if (kvm->arch.kvmclock_offset) {
1303 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1304 boot = timespec64_sub(boot, ts);
1306 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1307 wc.nsec = boot.tv_nsec;
1308 wc.version = version;
1310 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1313 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1316 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1318 do_shl32_div32(dividend, divisor);
1322 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1323 s8 *pshift, u32 *pmultiplier)
1331 scaled64 = scaled_hz;
1332 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1337 tps32 = (uint32_t)tps64;
1338 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1339 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1347 *pmultiplier = div_frac(scaled64, tps32);
1349 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1350 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1353 #ifdef CONFIG_X86_64
1354 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1357 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1358 static unsigned long max_tsc_khz;
1360 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1362 u64 v = (u64)khz * (1000000 + ppm);
1367 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1371 /* Guest TSC same frequency as host TSC? */
1373 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1377 /* TSC scaling supported? */
1378 if (!kvm_has_tsc_control) {
1379 if (user_tsc_khz > tsc_khz) {
1380 vcpu->arch.tsc_catchup = 1;
1381 vcpu->arch.tsc_always_catchup = 1;
1384 WARN(1, "user requested TSC rate below hardware speed\n");
1389 /* TSC scaling required - calculate ratio */
1390 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1391 user_tsc_khz, tsc_khz);
1393 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1394 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1399 vcpu->arch.tsc_scaling_ratio = ratio;
1403 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1405 u32 thresh_lo, thresh_hi;
1406 int use_scaling = 0;
1408 /* tsc_khz can be zero if TSC calibration fails */
1409 if (user_tsc_khz == 0) {
1410 /* set tsc_scaling_ratio to a safe value */
1411 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1415 /* Compute a scale to convert nanoseconds in TSC cycles */
1416 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1417 &vcpu->arch.virtual_tsc_shift,
1418 &vcpu->arch.virtual_tsc_mult);
1419 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1422 * Compute the variation in TSC rate which is acceptable
1423 * within the range of tolerance and decide if the
1424 * rate being applied is within that bounds of the hardware
1425 * rate. If so, no scaling or compensation need be done.
1427 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1428 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1429 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1430 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1433 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1436 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1438 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1439 vcpu->arch.virtual_tsc_mult,
1440 vcpu->arch.virtual_tsc_shift);
1441 tsc += vcpu->arch.this_tsc_write;
1445 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1447 #ifdef CONFIG_X86_64
1449 struct kvm_arch *ka = &vcpu->kvm->arch;
1450 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1452 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1453 atomic_read(&vcpu->kvm->online_vcpus));
1456 * Once the masterclock is enabled, always perform request in
1457 * order to update it.
1459 * In order to enable masterclock, the host clocksource must be TSC
1460 * and the vcpus need to have matched TSCs. When that happens,
1461 * perform request to enable masterclock.
1463 if (ka->use_master_clock ||
1464 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1465 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1467 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1468 atomic_read(&vcpu->kvm->online_vcpus),
1469 ka->use_master_clock, gtod->clock.vclock_mode);
1473 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1475 u64 curr_offset = vcpu->arch.tsc_offset;
1476 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1480 * Multiply tsc by a fixed point number represented by ratio.
1482 * The most significant 64-N bits (mult) of ratio represent the
1483 * integral part of the fixed point number; the remaining N bits
1484 * (frac) represent the fractional part, ie. ratio represents a fixed
1485 * point number (mult + frac * 2^(-N)).
1487 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1489 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1491 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1494 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1497 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1499 if (ratio != kvm_default_tsc_scaling_ratio)
1500 _tsc = __scale_tsc(ratio, tsc);
1504 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1506 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1510 tsc = kvm_scale_tsc(vcpu, rdtsc());
1512 return target_tsc - tsc;
1515 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1517 return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1519 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1521 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1523 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1524 vcpu->arch.tsc_offset = offset;
1527 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1529 struct kvm *kvm = vcpu->kvm;
1530 u64 offset, ns, elapsed;
1531 unsigned long flags;
1533 bool already_matched;
1534 u64 data = msr->data;
1535 bool synchronizing = false;
1537 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1538 offset = kvm_compute_tsc_offset(vcpu, data);
1539 ns = ktime_get_boot_ns();
1540 elapsed = ns - kvm->arch.last_tsc_nsec;
1542 if (vcpu->arch.virtual_tsc_khz) {
1543 if (data == 0 && msr->host_initiated) {
1545 * detection of vcpu initialization -- need to sync
1546 * with other vCPUs. This particularly helps to keep
1547 * kvm_clock stable after CPU hotplug
1549 synchronizing = true;
1551 u64 tsc_exp = kvm->arch.last_tsc_write +
1552 nsec_to_cycles(vcpu, elapsed);
1553 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1555 * Special case: TSC write with a small delta (1 second)
1556 * of virtual cycle time against real time is
1557 * interpreted as an attempt to synchronize the CPU.
1559 synchronizing = data < tsc_exp + tsc_hz &&
1560 data + tsc_hz > tsc_exp;
1565 * For a reliable TSC, we can match TSC offsets, and for an unstable
1566 * TSC, we add elapsed time in this computation. We could let the
1567 * compensation code attempt to catch up if we fall behind, but
1568 * it's better to try to match offsets from the beginning.
1570 if (synchronizing &&
1571 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1572 if (!check_tsc_unstable()) {
1573 offset = kvm->arch.cur_tsc_offset;
1574 pr_debug("kvm: matched tsc offset for %llu\n", data);
1576 u64 delta = nsec_to_cycles(vcpu, elapsed);
1578 offset = kvm_compute_tsc_offset(vcpu, data);
1579 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1582 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1585 * We split periods of matched TSC writes into generations.
1586 * For each generation, we track the original measured
1587 * nanosecond time, offset, and write, so if TSCs are in
1588 * sync, we can match exact offset, and if not, we can match
1589 * exact software computation in compute_guest_tsc()
1591 * These values are tracked in kvm->arch.cur_xxx variables.
1593 kvm->arch.cur_tsc_generation++;
1594 kvm->arch.cur_tsc_nsec = ns;
1595 kvm->arch.cur_tsc_write = data;
1596 kvm->arch.cur_tsc_offset = offset;
1598 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1599 kvm->arch.cur_tsc_generation, data);
1603 * We also track th most recent recorded KHZ, write and time to
1604 * allow the matching interval to be extended at each write.
1606 kvm->arch.last_tsc_nsec = ns;
1607 kvm->arch.last_tsc_write = data;
1608 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1610 vcpu->arch.last_guest_tsc = data;
1612 /* Keep track of which generation this VCPU has synchronized to */
1613 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1614 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1615 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1617 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1618 update_ia32_tsc_adjust_msr(vcpu, offset);
1620 kvm_vcpu_write_tsc_offset(vcpu, offset);
1621 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1623 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1625 kvm->arch.nr_vcpus_matched_tsc = 0;
1626 } else if (!already_matched) {
1627 kvm->arch.nr_vcpus_matched_tsc++;
1630 kvm_track_tsc_matching(vcpu);
1631 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1634 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1636 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1639 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1642 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1644 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1645 WARN_ON(adjustment < 0);
1646 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1647 adjust_tsc_offset_guest(vcpu, adjustment);
1650 #ifdef CONFIG_X86_64
1652 static u64 read_tsc(void)
1654 u64 ret = (u64)rdtsc_ordered();
1655 u64 last = pvclock_gtod_data.clock.cycle_last;
1657 if (likely(ret >= last))
1661 * GCC likes to generate cmov here, but this branch is extremely
1662 * predictable (it's just a function of time and the likely is
1663 * very likely) and there's a data dependence, so force GCC
1664 * to generate a branch instead. I don't barrier() because
1665 * we don't actually need a barrier, and if this function
1666 * ever gets inlined it will generate worse code.
1672 static inline u64 vgettsc(u64 *cycle_now)
1675 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1677 *cycle_now = read_tsc();
1679 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1680 return v * gtod->clock.mult;
1683 static int do_monotonic_boot(s64 *t, u64 *cycle_now)
1685 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1691 seq = read_seqcount_begin(>od->seq);
1692 mode = gtod->clock.vclock_mode;
1693 ns = gtod->nsec_base;
1694 ns += vgettsc(cycle_now);
1695 ns >>= gtod->clock.shift;
1696 ns += gtod->boot_ns;
1697 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1703 static int do_realtime(struct timespec *ts, u64 *cycle_now)
1705 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1711 seq = read_seqcount_begin(>od->seq);
1712 mode = gtod->clock.vclock_mode;
1713 ts->tv_sec = gtod->wall_time_sec;
1714 ns = gtod->nsec_base;
1715 ns += vgettsc(cycle_now);
1716 ns >>= gtod->clock.shift;
1717 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1719 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1725 /* returns true if host is using tsc clocksource */
1726 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
1728 /* checked again under seqlock below */
1729 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1732 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1735 /* returns true if host is using tsc clocksource */
1736 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1739 /* checked again under seqlock below */
1740 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1743 return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1749 * Assuming a stable TSC across physical CPUS, and a stable TSC
1750 * across virtual CPUs, the following condition is possible.
1751 * Each numbered line represents an event visible to both
1752 * CPUs at the next numbered event.
1754 * "timespecX" represents host monotonic time. "tscX" represents
1757 * VCPU0 on CPU0 | VCPU1 on CPU1
1759 * 1. read timespec0,tsc0
1760 * 2. | timespec1 = timespec0 + N
1762 * 3. transition to guest | transition to guest
1763 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1764 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1765 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1767 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1770 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1772 * - 0 < N - M => M < N
1774 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1775 * always the case (the difference between two distinct xtime instances
1776 * might be smaller then the difference between corresponding TSC reads,
1777 * when updating guest vcpus pvclock areas).
1779 * To avoid that problem, do not allow visibility of distinct
1780 * system_timestamp/tsc_timestamp values simultaneously: use a master
1781 * copy of host monotonic time values. Update that master copy
1784 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1788 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1790 #ifdef CONFIG_X86_64
1791 struct kvm_arch *ka = &kvm->arch;
1793 bool host_tsc_clocksource, vcpus_matched;
1795 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1796 atomic_read(&kvm->online_vcpus));
1799 * If the host uses TSC clock, then passthrough TSC as stable
1802 host_tsc_clocksource = kvm_get_time_and_clockread(
1803 &ka->master_kernel_ns,
1804 &ka->master_cycle_now);
1806 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1807 && !ka->backwards_tsc_observed
1808 && !ka->boot_vcpu_runs_old_kvmclock;
1810 if (ka->use_master_clock)
1811 atomic_set(&kvm_guest_has_master_clock, 1);
1813 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1814 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1819 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1821 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1824 static void kvm_gen_update_masterclock(struct kvm *kvm)
1826 #ifdef CONFIG_X86_64
1828 struct kvm_vcpu *vcpu;
1829 struct kvm_arch *ka = &kvm->arch;
1831 spin_lock(&ka->pvclock_gtod_sync_lock);
1832 kvm_make_mclock_inprogress_request(kvm);
1833 /* no guest entries from this point */
1834 pvclock_update_vm_gtod_copy(kvm);
1836 kvm_for_each_vcpu(i, vcpu, kvm)
1837 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1839 /* guest entries allowed */
1840 kvm_for_each_vcpu(i, vcpu, kvm)
1841 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1843 spin_unlock(&ka->pvclock_gtod_sync_lock);
1847 u64 get_kvmclock_ns(struct kvm *kvm)
1849 struct kvm_arch *ka = &kvm->arch;
1850 struct pvclock_vcpu_time_info hv_clock;
1853 spin_lock(&ka->pvclock_gtod_sync_lock);
1854 if (!ka->use_master_clock) {
1855 spin_unlock(&ka->pvclock_gtod_sync_lock);
1856 return ktime_get_boot_ns() + ka->kvmclock_offset;
1859 hv_clock.tsc_timestamp = ka->master_cycle_now;
1860 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1861 spin_unlock(&ka->pvclock_gtod_sync_lock);
1863 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1866 if (__this_cpu_read(cpu_tsc_khz)) {
1867 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1868 &hv_clock.tsc_shift,
1869 &hv_clock.tsc_to_system_mul);
1870 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1872 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
1879 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1881 struct kvm_vcpu_arch *vcpu = &v->arch;
1882 struct pvclock_vcpu_time_info guest_hv_clock;
1884 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1885 &guest_hv_clock, sizeof(guest_hv_clock))))
1888 /* This VCPU is paused, but it's legal for a guest to read another
1889 * VCPU's kvmclock, so we really have to follow the specification where
1890 * it says that version is odd if data is being modified, and even after
1893 * Version field updates must be kept separate. This is because
1894 * kvm_write_guest_cached might use a "rep movs" instruction, and
1895 * writes within a string instruction are weakly ordered. So there
1896 * are three writes overall.
1898 * As a small optimization, only write the version field in the first
1899 * and third write. The vcpu->pv_time cache is still valid, because the
1900 * version field is the first in the struct.
1902 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1904 if (guest_hv_clock.version & 1)
1905 ++guest_hv_clock.version; /* first time write, random junk */
1907 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1908 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1910 sizeof(vcpu->hv_clock.version));
1914 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1915 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1917 if (vcpu->pvclock_set_guest_stopped_request) {
1918 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1919 vcpu->pvclock_set_guest_stopped_request = false;
1922 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1924 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1926 sizeof(vcpu->hv_clock));
1930 vcpu->hv_clock.version++;
1931 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1933 sizeof(vcpu->hv_clock.version));
1936 static int kvm_guest_time_update(struct kvm_vcpu *v)
1938 unsigned long flags, tgt_tsc_khz;
1939 struct kvm_vcpu_arch *vcpu = &v->arch;
1940 struct kvm_arch *ka = &v->kvm->arch;
1942 u64 tsc_timestamp, host_tsc;
1944 bool use_master_clock;
1950 * If the host uses TSC clock, then passthrough TSC as stable
1953 spin_lock(&ka->pvclock_gtod_sync_lock);
1954 use_master_clock = ka->use_master_clock;
1955 if (use_master_clock) {
1956 host_tsc = ka->master_cycle_now;
1957 kernel_ns = ka->master_kernel_ns;
1959 spin_unlock(&ka->pvclock_gtod_sync_lock);
1961 /* Keep irq disabled to prevent changes to the clock */
1962 local_irq_save(flags);
1963 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1964 if (unlikely(tgt_tsc_khz == 0)) {
1965 local_irq_restore(flags);
1966 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1969 if (!use_master_clock) {
1971 kernel_ns = ktime_get_boot_ns();
1974 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1977 * We may have to catch up the TSC to match elapsed wall clock
1978 * time for two reasons, even if kvmclock is used.
1979 * 1) CPU could have been running below the maximum TSC rate
1980 * 2) Broken TSC compensation resets the base at each VCPU
1981 * entry to avoid unknown leaps of TSC even when running
1982 * again on the same CPU. This may cause apparent elapsed
1983 * time to disappear, and the guest to stand still or run
1986 if (vcpu->tsc_catchup) {
1987 u64 tsc = compute_guest_tsc(v, kernel_ns);
1988 if (tsc > tsc_timestamp) {
1989 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1990 tsc_timestamp = tsc;
1994 local_irq_restore(flags);
1996 /* With all the info we got, fill in the values */
1998 if (kvm_has_tsc_control)
1999 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2001 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2002 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2003 &vcpu->hv_clock.tsc_shift,
2004 &vcpu->hv_clock.tsc_to_system_mul);
2005 vcpu->hw_tsc_khz = tgt_tsc_khz;
2008 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2009 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2010 vcpu->last_guest_tsc = tsc_timestamp;
2012 /* If the host uses TSC clocksource, then it is stable */
2014 if (use_master_clock)
2015 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2017 vcpu->hv_clock.flags = pvclock_flags;
2019 if (vcpu->pv_time_enabled)
2020 kvm_setup_pvclock_page(v);
2021 if (v == kvm_get_vcpu(v->kvm, 0))
2022 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2027 * kvmclock updates which are isolated to a given vcpu, such as
2028 * vcpu->cpu migration, should not allow system_timestamp from
2029 * the rest of the vcpus to remain static. Otherwise ntp frequency
2030 * correction applies to one vcpu's system_timestamp but not
2033 * So in those cases, request a kvmclock update for all vcpus.
2034 * We need to rate-limit these requests though, as they can
2035 * considerably slow guests that have a large number of vcpus.
2036 * The time for a remote vcpu to update its kvmclock is bound
2037 * by the delay we use to rate-limit the updates.
2040 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2042 static void kvmclock_update_fn(struct work_struct *work)
2045 struct delayed_work *dwork = to_delayed_work(work);
2046 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2047 kvmclock_update_work);
2048 struct kvm *kvm = container_of(ka, struct kvm, arch);
2049 struct kvm_vcpu *vcpu;
2051 kvm_for_each_vcpu(i, vcpu, kvm) {
2052 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2053 kvm_vcpu_kick(vcpu);
2057 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2059 struct kvm *kvm = v->kvm;
2061 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2062 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2063 KVMCLOCK_UPDATE_DELAY);
2066 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2068 static void kvmclock_sync_fn(struct work_struct *work)
2070 struct delayed_work *dwork = to_delayed_work(work);
2071 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2072 kvmclock_sync_work);
2073 struct kvm *kvm = container_of(ka, struct kvm, arch);
2075 if (!kvmclock_periodic_sync)
2078 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2079 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2080 KVMCLOCK_SYNC_PERIOD);
2083 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2085 u64 mcg_cap = vcpu->arch.mcg_cap;
2086 unsigned bank_num = mcg_cap & 0xff;
2089 case MSR_IA32_MCG_STATUS:
2090 vcpu->arch.mcg_status = data;
2092 case MSR_IA32_MCG_CTL:
2093 if (!(mcg_cap & MCG_CTL_P))
2095 if (data != 0 && data != ~(u64)0)
2097 vcpu->arch.mcg_ctl = data;
2100 if (msr >= MSR_IA32_MC0_CTL &&
2101 msr < MSR_IA32_MCx_CTL(bank_num)) {
2102 u32 offset = msr - MSR_IA32_MC0_CTL;
2103 /* only 0 or all 1s can be written to IA32_MCi_CTL
2104 * some Linux kernels though clear bit 10 in bank 4 to
2105 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2106 * this to avoid an uncatched #GP in the guest
2108 if ((offset & 0x3) == 0 &&
2109 data != 0 && (data | (1 << 10)) != ~(u64)0)
2111 vcpu->arch.mce_banks[offset] = data;
2119 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2121 struct kvm *kvm = vcpu->kvm;
2122 int lm = is_long_mode(vcpu);
2123 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2124 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2125 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2126 : kvm->arch.xen_hvm_config.blob_size_32;
2127 u32 page_num = data & ~PAGE_MASK;
2128 u64 page_addr = data & PAGE_MASK;
2133 if (page_num >= blob_size)
2136 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2141 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2150 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2152 gpa_t gpa = data & ~0x3f;
2154 /* Bits 3:5 are reserved, Should be zero */
2158 vcpu->arch.apf.msr_val = data;
2160 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2161 kvm_clear_async_pf_completion_queue(vcpu);
2162 kvm_async_pf_hash_reset(vcpu);
2166 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2170 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2171 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2172 kvm_async_pf_wakeup_all(vcpu);
2176 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2178 vcpu->arch.pv_time_enabled = false;
2181 static void record_steal_time(struct kvm_vcpu *vcpu)
2183 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2186 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2187 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2190 vcpu->arch.st.steal.preempted = 0;
2192 if (vcpu->arch.st.steal.version & 1)
2193 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2195 vcpu->arch.st.steal.version += 1;
2197 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2198 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2202 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2203 vcpu->arch.st.last_steal;
2204 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2206 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2207 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2211 vcpu->arch.st.steal.version += 1;
2213 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2214 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2217 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2220 u32 msr = msr_info->index;
2221 u64 data = msr_info->data;
2224 case MSR_AMD64_NB_CFG:
2225 case MSR_IA32_UCODE_WRITE:
2226 case MSR_VM_HSAVE_PA:
2227 case MSR_AMD64_PATCH_LOADER:
2228 case MSR_AMD64_BU_CFG2:
2229 case MSR_AMD64_DC_CFG:
2230 case MSR_F15H_EX_CFG:
2233 case MSR_IA32_UCODE_REV:
2234 if (msr_info->host_initiated)
2235 vcpu->arch.microcode_version = data;
2238 return set_efer(vcpu, data);
2240 data &= ~(u64)0x40; /* ignore flush filter disable */
2241 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2242 data &= ~(u64)0x8; /* ignore TLB cache disable */
2243 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2245 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2250 case MSR_FAM10H_MMIO_CONF_BASE:
2252 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2257 case MSR_IA32_DEBUGCTLMSR:
2259 /* We support the non-activated case already */
2261 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2262 /* Values other than LBR and BTF are vendor-specific,
2263 thus reserved and should throw a #GP */
2266 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2269 case 0x200 ... 0x2ff:
2270 return kvm_mtrr_set_msr(vcpu, msr, data);
2271 case MSR_IA32_APICBASE:
2272 return kvm_set_apic_base(vcpu, msr_info);
2273 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2274 return kvm_x2apic_msr_write(vcpu, msr, data);
2275 case MSR_IA32_TSCDEADLINE:
2276 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2278 case MSR_IA32_TSC_ADJUST:
2279 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2280 if (!msr_info->host_initiated) {
2281 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2282 adjust_tsc_offset_guest(vcpu, adj);
2284 vcpu->arch.ia32_tsc_adjust_msr = data;
2287 case MSR_IA32_MISC_ENABLE:
2288 vcpu->arch.ia32_misc_enable_msr = data;
2290 case MSR_IA32_SMBASE:
2291 if (!msr_info->host_initiated)
2293 vcpu->arch.smbase = data;
2295 case MSR_KVM_WALL_CLOCK_NEW:
2296 case MSR_KVM_WALL_CLOCK:
2297 vcpu->kvm->arch.wall_clock = data;
2298 kvm_write_wall_clock(vcpu->kvm, data);
2300 case MSR_KVM_SYSTEM_TIME_NEW:
2301 case MSR_KVM_SYSTEM_TIME: {
2302 struct kvm_arch *ka = &vcpu->kvm->arch;
2304 kvmclock_reset(vcpu);
2306 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2307 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2309 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2310 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2312 ka->boot_vcpu_runs_old_kvmclock = tmp;
2315 vcpu->arch.time = data;
2316 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2318 /* we verify if the enable bit is set... */
2322 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2323 &vcpu->arch.pv_time, data & ~1ULL,
2324 sizeof(struct pvclock_vcpu_time_info)))
2325 vcpu->arch.pv_time_enabled = false;
2327 vcpu->arch.pv_time_enabled = true;
2331 case MSR_KVM_ASYNC_PF_EN:
2332 if (kvm_pv_enable_async_pf(vcpu, data))
2335 case MSR_KVM_STEAL_TIME:
2337 if (unlikely(!sched_info_on()))
2340 if (data & KVM_STEAL_RESERVED_MASK)
2343 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2344 data & KVM_STEAL_VALID_BITS,
2345 sizeof(struct kvm_steal_time)))
2348 vcpu->arch.st.msr_val = data;
2350 if (!(data & KVM_MSR_ENABLED))
2353 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2356 case MSR_KVM_PV_EOI_EN:
2357 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2361 case MSR_IA32_MCG_CTL:
2362 case MSR_IA32_MCG_STATUS:
2363 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2364 return set_msr_mce(vcpu, msr, data);
2366 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2367 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2368 pr = true; /* fall through */
2369 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2370 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2371 if (kvm_pmu_is_valid_msr(vcpu, msr))
2372 return kvm_pmu_set_msr(vcpu, msr_info);
2374 if (pr || data != 0)
2375 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2376 "0x%x data 0x%llx\n", msr, data);
2378 case MSR_K7_CLK_CTL:
2380 * Ignore all writes to this no longer documented MSR.
2381 * Writes are only relevant for old K7 processors,
2382 * all pre-dating SVM, but a recommended workaround from
2383 * AMD for these chips. It is possible to specify the
2384 * affected processor models on the command line, hence
2385 * the need to ignore the workaround.
2388 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2389 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2390 case HV_X64_MSR_CRASH_CTL:
2391 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2392 return kvm_hv_set_msr_common(vcpu, msr, data,
2393 msr_info->host_initiated);
2394 case MSR_IA32_BBL_CR_CTL3:
2395 /* Drop writes to this legacy MSR -- see rdmsr
2396 * counterpart for further detail.
2398 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", msr, data);
2400 case MSR_AMD64_OSVW_ID_LENGTH:
2401 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2403 vcpu->arch.osvw.length = data;
2405 case MSR_AMD64_OSVW_STATUS:
2406 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2408 vcpu->arch.osvw.status = data;
2410 case MSR_PLATFORM_INFO:
2411 if (!msr_info->host_initiated ||
2412 data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2413 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2414 cpuid_fault_enabled(vcpu)))
2416 vcpu->arch.msr_platform_info = data;
2418 case MSR_MISC_FEATURES_ENABLES:
2419 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2420 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2421 !supports_cpuid_fault(vcpu)))
2423 vcpu->arch.msr_misc_features_enables = data;
2426 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2427 return xen_hvm_config(vcpu, data);
2428 if (kvm_pmu_is_valid_msr(vcpu, msr))
2429 return kvm_pmu_set_msr(vcpu, msr_info);
2431 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2435 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2442 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2446 * Reads an msr value (of 'msr_index') into 'pdata'.
2447 * Returns 0 on success, non-0 otherwise.
2448 * Assumes vcpu_load() was already called.
2450 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2452 return kvm_x86_ops->get_msr(vcpu, msr);
2454 EXPORT_SYMBOL_GPL(kvm_get_msr);
2456 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2459 u64 mcg_cap = vcpu->arch.mcg_cap;
2460 unsigned bank_num = mcg_cap & 0xff;
2463 case MSR_IA32_P5_MC_ADDR:
2464 case MSR_IA32_P5_MC_TYPE:
2467 case MSR_IA32_MCG_CAP:
2468 data = vcpu->arch.mcg_cap;
2470 case MSR_IA32_MCG_CTL:
2471 if (!(mcg_cap & MCG_CTL_P))
2473 data = vcpu->arch.mcg_ctl;
2475 case MSR_IA32_MCG_STATUS:
2476 data = vcpu->arch.mcg_status;
2479 if (msr >= MSR_IA32_MC0_CTL &&
2480 msr < MSR_IA32_MCx_CTL(bank_num)) {
2481 u32 offset = msr - MSR_IA32_MC0_CTL;
2482 data = vcpu->arch.mce_banks[offset];
2491 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2493 switch (msr_info->index) {
2494 case MSR_IA32_PLATFORM_ID:
2495 case MSR_IA32_EBL_CR_POWERON:
2496 case MSR_IA32_DEBUGCTLMSR:
2497 case MSR_IA32_LASTBRANCHFROMIP:
2498 case MSR_IA32_LASTBRANCHTOIP:
2499 case MSR_IA32_LASTINTFROMIP:
2500 case MSR_IA32_LASTINTTOIP:
2502 case MSR_K8_TSEG_ADDR:
2503 case MSR_K8_TSEG_MASK:
2505 case MSR_VM_HSAVE_PA:
2506 case MSR_K8_INT_PENDING_MSG:
2507 case MSR_AMD64_NB_CFG:
2508 case MSR_FAM10H_MMIO_CONF_BASE:
2509 case MSR_AMD64_BU_CFG2:
2510 case MSR_IA32_PERF_CTL:
2511 case MSR_AMD64_DC_CFG:
2512 case MSR_F15H_EX_CFG:
2515 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2516 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2517 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2518 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2519 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2520 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2523 case MSR_IA32_UCODE_REV:
2524 msr_info->data = vcpu->arch.microcode_version;
2527 case 0x200 ... 0x2ff:
2528 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2529 case 0xcd: /* fsb frequency */
2533 * MSR_EBC_FREQUENCY_ID
2534 * Conservative value valid for even the basic CPU models.
2535 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2536 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2537 * and 266MHz for model 3, or 4. Set Core Clock
2538 * Frequency to System Bus Frequency Ratio to 1 (bits
2539 * 31:24) even though these are only valid for CPU
2540 * models > 2, however guests may end up dividing or
2541 * multiplying by zero otherwise.
2543 case MSR_EBC_FREQUENCY_ID:
2544 msr_info->data = 1 << 24;
2546 case MSR_IA32_APICBASE:
2547 msr_info->data = kvm_get_apic_base(vcpu);
2549 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2550 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2552 case MSR_IA32_TSCDEADLINE:
2553 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2555 case MSR_IA32_TSC_ADJUST:
2556 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2558 case MSR_IA32_MISC_ENABLE:
2559 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2561 case MSR_IA32_SMBASE:
2562 if (!msr_info->host_initiated)
2564 msr_info->data = vcpu->arch.smbase;
2566 case MSR_IA32_PERF_STATUS:
2567 /* TSC increment by tick */
2568 msr_info->data = 1000ULL;
2569 /* CPU multiplier */
2570 msr_info->data |= (((uint64_t)4ULL) << 40);
2573 msr_info->data = vcpu->arch.efer;
2575 case MSR_KVM_WALL_CLOCK:
2576 case MSR_KVM_WALL_CLOCK_NEW:
2577 msr_info->data = vcpu->kvm->arch.wall_clock;
2579 case MSR_KVM_SYSTEM_TIME:
2580 case MSR_KVM_SYSTEM_TIME_NEW:
2581 msr_info->data = vcpu->arch.time;
2583 case MSR_KVM_ASYNC_PF_EN:
2584 msr_info->data = vcpu->arch.apf.msr_val;
2586 case MSR_KVM_STEAL_TIME:
2587 msr_info->data = vcpu->arch.st.msr_val;
2589 case MSR_KVM_PV_EOI_EN:
2590 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2592 case MSR_IA32_P5_MC_ADDR:
2593 case MSR_IA32_P5_MC_TYPE:
2594 case MSR_IA32_MCG_CAP:
2595 case MSR_IA32_MCG_CTL:
2596 case MSR_IA32_MCG_STATUS:
2597 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2598 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2599 case MSR_K7_CLK_CTL:
2601 * Provide expected ramp-up count for K7. All other
2602 * are set to zero, indicating minimum divisors for
2605 * This prevents guest kernels on AMD host with CPU
2606 * type 6, model 8 and higher from exploding due to
2607 * the rdmsr failing.
2609 msr_info->data = 0x20000000;
2611 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2612 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2613 case HV_X64_MSR_CRASH_CTL:
2614 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2615 return kvm_hv_get_msr_common(vcpu,
2616 msr_info->index, &msr_info->data);
2618 case MSR_IA32_BBL_CR_CTL3:
2619 /* This legacy MSR exists but isn't fully documented in current
2620 * silicon. It is however accessed by winxp in very narrow
2621 * scenarios where it sets bit #19, itself documented as
2622 * a "reserved" bit. Best effort attempt to source coherent
2623 * read data here should the balance of the register be
2624 * interpreted by the guest:
2626 * L2 cache control register 3: 64GB range, 256KB size,
2627 * enabled, latency 0x1, configured
2629 msr_info->data = 0xbe702111;
2631 case MSR_AMD64_OSVW_ID_LENGTH:
2632 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2634 msr_info->data = vcpu->arch.osvw.length;
2636 case MSR_AMD64_OSVW_STATUS:
2637 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2639 msr_info->data = vcpu->arch.osvw.status;
2641 case MSR_PLATFORM_INFO:
2642 msr_info->data = vcpu->arch.msr_platform_info;
2644 case MSR_MISC_FEATURES_ENABLES:
2645 msr_info->data = vcpu->arch.msr_misc_features_enables;
2648 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2649 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2651 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2655 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2662 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2665 * Read or write a bunch of msrs. All parameters are kernel addresses.
2667 * @return number of msrs set successfully.
2669 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2670 struct kvm_msr_entry *entries,
2671 int (*do_msr)(struct kvm_vcpu *vcpu,
2672 unsigned index, u64 *data))
2676 for (i = 0; i < msrs->nmsrs; ++i)
2677 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2684 * Read or write a bunch of msrs. Parameters are user addresses.
2686 * @return number of msrs set successfully.
2688 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2689 int (*do_msr)(struct kvm_vcpu *vcpu,
2690 unsigned index, u64 *data),
2693 struct kvm_msrs msrs;
2694 struct kvm_msr_entry *entries;
2699 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2703 if (msrs.nmsrs >= MAX_IO_MSRS)
2706 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2707 entries = memdup_user(user_msrs->entries, size);
2708 if (IS_ERR(entries)) {
2709 r = PTR_ERR(entries);
2713 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2718 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2729 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2734 case KVM_CAP_IRQCHIP:
2736 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2737 case KVM_CAP_SET_TSS_ADDR:
2738 case KVM_CAP_EXT_CPUID:
2739 case KVM_CAP_EXT_EMUL_CPUID:
2740 case KVM_CAP_CLOCKSOURCE:
2742 case KVM_CAP_NOP_IO_DELAY:
2743 case KVM_CAP_MP_STATE:
2744 case KVM_CAP_SYNC_MMU:
2745 case KVM_CAP_USER_NMI:
2746 case KVM_CAP_REINJECT_CONTROL:
2747 case KVM_CAP_IRQ_INJECT_STATUS:
2748 case KVM_CAP_IOEVENTFD:
2749 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2751 case KVM_CAP_PIT_STATE2:
2752 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2753 case KVM_CAP_XEN_HVM:
2754 case KVM_CAP_VCPU_EVENTS:
2755 case KVM_CAP_HYPERV:
2756 case KVM_CAP_HYPERV_VAPIC:
2757 case KVM_CAP_HYPERV_SPIN:
2758 case KVM_CAP_HYPERV_SYNIC:
2759 case KVM_CAP_HYPERV_SYNIC2:
2760 case KVM_CAP_HYPERV_VP_INDEX:
2761 case KVM_CAP_PCI_SEGMENT:
2762 case KVM_CAP_DEBUGREGS:
2763 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2765 case KVM_CAP_ASYNC_PF:
2766 case KVM_CAP_GET_TSC_KHZ:
2767 case KVM_CAP_KVMCLOCK_CTRL:
2768 case KVM_CAP_READONLY_MEM:
2769 case KVM_CAP_HYPERV_TIME:
2770 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2771 case KVM_CAP_TSC_DEADLINE_TIMER:
2772 case KVM_CAP_ENABLE_CAP_VM:
2773 case KVM_CAP_DISABLE_QUIRKS:
2774 case KVM_CAP_SET_BOOT_CPU_ID:
2775 case KVM_CAP_SPLIT_IRQCHIP:
2776 case KVM_CAP_IMMEDIATE_EXIT:
2777 case KVM_CAP_GET_MSR_FEATURES:
2780 case KVM_CAP_ADJUST_CLOCK:
2781 r = KVM_CLOCK_TSC_STABLE;
2783 case KVM_CAP_X86_GUEST_MWAIT:
2784 r = kvm_mwait_in_guest();
2786 case KVM_CAP_X86_SMM:
2787 /* SMBASE is usually relocated above 1M on modern chipsets,
2788 * and SMM handlers might indeed rely on 4G segment limits,
2789 * so do not report SMM to be available if real mode is
2790 * emulated via vm86 mode. Still, do not go to great lengths
2791 * to avoid userspace's usage of the feature, because it is a
2792 * fringe case that is not enabled except via specific settings
2793 * of the module parameters.
2795 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
2798 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2800 case KVM_CAP_NR_VCPUS:
2801 r = KVM_SOFT_MAX_VCPUS;
2803 case KVM_CAP_MAX_VCPUS:
2806 case KVM_CAP_NR_MEMSLOTS:
2807 r = KVM_USER_MEM_SLOTS;
2809 case KVM_CAP_PV_MMU: /* obsolete */
2813 r = KVM_MAX_MCE_BANKS;
2816 r = boot_cpu_has(X86_FEATURE_XSAVE);
2818 case KVM_CAP_TSC_CONTROL:
2819 r = kvm_has_tsc_control;
2821 case KVM_CAP_X2APIC_API:
2822 r = KVM_X2APIC_API_VALID_FLAGS;
2832 long kvm_arch_dev_ioctl(struct file *filp,
2833 unsigned int ioctl, unsigned long arg)
2835 void __user *argp = (void __user *)arg;
2839 case KVM_GET_MSR_INDEX_LIST: {
2840 struct kvm_msr_list __user *user_msr_list = argp;
2841 struct kvm_msr_list msr_list;
2845 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2848 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2849 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2852 if (n < msr_list.nmsrs)
2855 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2856 num_msrs_to_save * sizeof(u32)))
2858 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2860 num_emulated_msrs * sizeof(u32)))
2865 case KVM_GET_SUPPORTED_CPUID:
2866 case KVM_GET_EMULATED_CPUID: {
2867 struct kvm_cpuid2 __user *cpuid_arg = argp;
2868 struct kvm_cpuid2 cpuid;
2871 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2874 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2880 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2885 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2887 if (copy_to_user(argp, &kvm_mce_cap_supported,
2888 sizeof(kvm_mce_cap_supported)))
2892 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
2893 struct kvm_msr_list __user *user_msr_list = argp;
2894 struct kvm_msr_list msr_list;
2898 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
2901 msr_list.nmsrs = num_msr_based_features;
2902 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
2905 if (n < msr_list.nmsrs)
2908 if (copy_to_user(user_msr_list->indices, &msr_based_features,
2909 num_msr_based_features * sizeof(u32)))
2915 r = msr_io(NULL, argp, do_get_msr_feature, 1);
2925 static void wbinvd_ipi(void *garbage)
2930 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2932 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2935 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2937 /* Address WBINVD may be executed by guest */
2938 if (need_emulate_wbinvd(vcpu)) {
2939 if (kvm_x86_ops->has_wbinvd_exit())
2940 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2941 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2942 smp_call_function_single(vcpu->cpu,
2943 wbinvd_ipi, NULL, 1);
2946 kvm_x86_ops->vcpu_load(vcpu, cpu);
2948 /* Apply any externally detected TSC adjustments (due to suspend) */
2949 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2950 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2951 vcpu->arch.tsc_offset_adjustment = 0;
2952 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2955 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2956 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2957 rdtsc() - vcpu->arch.last_host_tsc;
2959 mark_tsc_unstable("KVM discovered backwards TSC");
2961 if (check_tsc_unstable()) {
2962 u64 offset = kvm_compute_tsc_offset(vcpu,
2963 vcpu->arch.last_guest_tsc);
2964 kvm_vcpu_write_tsc_offset(vcpu, offset);
2965 vcpu->arch.tsc_catchup = 1;
2968 if (kvm_lapic_hv_timer_in_use(vcpu))
2969 kvm_lapic_restart_hv_timer(vcpu);
2972 * On a host with synchronized TSC, there is no need to update
2973 * kvmclock on vcpu->cpu migration
2975 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2976 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2977 if (vcpu->cpu != cpu)
2978 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
2982 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2985 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2987 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2990 vcpu->arch.st.steal.preempted = 1;
2992 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
2993 &vcpu->arch.st.steal.preempted,
2994 offsetof(struct kvm_steal_time, preempted),
2995 sizeof(vcpu->arch.st.steal.preempted));
2998 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3002 if (vcpu->preempted)
3003 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3006 * Disable page faults because we're in atomic context here.
3007 * kvm_write_guest_offset_cached() would call might_fault()
3008 * that relies on pagefault_disable() to tell if there's a
3009 * bug. NOTE: the write to guest memory may not go through if
3010 * during postcopy live migration or if there's heavy guest
3013 pagefault_disable();
3015 * kvm_memslots() will be called by
3016 * kvm_write_guest_offset_cached() so take the srcu lock.
3018 idx = srcu_read_lock(&vcpu->kvm->srcu);
3019 kvm_steal_time_set_preempted(vcpu);
3020 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3022 kvm_x86_ops->vcpu_put(vcpu);
3023 vcpu->arch.last_host_tsc = rdtsc();
3025 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3026 * on every vmexit, but if not, we might have a stale dr6 from the
3027 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3032 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3033 struct kvm_lapic_state *s)
3035 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
3036 kvm_x86_ops->sync_pir_to_irr(vcpu);
3038 return kvm_apic_get_state(vcpu, s);
3041 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3042 struct kvm_lapic_state *s)
3046 r = kvm_apic_set_state(vcpu, s);
3049 update_cr8_intercept(vcpu);
3054 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3056 return (!lapic_in_kernel(vcpu) ||
3057 kvm_apic_accept_pic_intr(vcpu));
3061 * if userspace requested an interrupt window, check that the
3062 * interrupt window is open.
3064 * No need to exit to userspace if we already have an interrupt queued.
3066 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3068 return kvm_arch_interrupt_allowed(vcpu) &&
3069 !kvm_cpu_has_interrupt(vcpu) &&
3070 !kvm_event_needs_reinjection(vcpu) &&
3071 kvm_cpu_accept_dm_intr(vcpu);
3074 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3075 struct kvm_interrupt *irq)
3077 if (irq->irq >= KVM_NR_INTERRUPTS)
3080 if (!irqchip_in_kernel(vcpu->kvm)) {
3081 kvm_queue_interrupt(vcpu, irq->irq, false);
3082 kvm_make_request(KVM_REQ_EVENT, vcpu);
3087 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3088 * fail for in-kernel 8259.
3090 if (pic_in_kernel(vcpu->kvm))
3093 if (vcpu->arch.pending_external_vector != -1)
3096 vcpu->arch.pending_external_vector = irq->irq;
3097 kvm_make_request(KVM_REQ_EVENT, vcpu);
3101 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3103 kvm_inject_nmi(vcpu);
3108 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3110 kvm_make_request(KVM_REQ_SMI, vcpu);
3115 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3116 struct kvm_tpr_access_ctl *tac)
3120 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3124 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3128 unsigned bank_num = mcg_cap & 0xff, bank;
3131 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3133 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3136 vcpu->arch.mcg_cap = mcg_cap;
3137 /* Init IA32_MCG_CTL to all 1s */
3138 if (mcg_cap & MCG_CTL_P)
3139 vcpu->arch.mcg_ctl = ~(u64)0;
3140 /* Init IA32_MCi_CTL to all 1s */
3141 for (bank = 0; bank < bank_num; bank++)
3142 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3144 if (kvm_x86_ops->setup_mce)
3145 kvm_x86_ops->setup_mce(vcpu);
3150 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3151 struct kvm_x86_mce *mce)
3153 u64 mcg_cap = vcpu->arch.mcg_cap;
3154 unsigned bank_num = mcg_cap & 0xff;
3155 u64 *banks = vcpu->arch.mce_banks;
3157 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3160 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3161 * reporting is disabled
3163 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3164 vcpu->arch.mcg_ctl != ~(u64)0)
3166 banks += 4 * mce->bank;
3168 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3169 * reporting is disabled for the bank
3171 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3173 if (mce->status & MCI_STATUS_UC) {
3174 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3175 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3176 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3179 if (banks[1] & MCI_STATUS_VAL)
3180 mce->status |= MCI_STATUS_OVER;
3181 banks[2] = mce->addr;
3182 banks[3] = mce->misc;
3183 vcpu->arch.mcg_status = mce->mcg_status;
3184 banks[1] = mce->status;
3185 kvm_queue_exception(vcpu, MC_VECTOR);
3186 } else if (!(banks[1] & MCI_STATUS_VAL)
3187 || !(banks[1] & MCI_STATUS_UC)) {
3188 if (banks[1] & MCI_STATUS_VAL)
3189 mce->status |= MCI_STATUS_OVER;
3190 banks[2] = mce->addr;
3191 banks[3] = mce->misc;
3192 banks[1] = mce->status;
3194 banks[1] |= MCI_STATUS_OVER;
3198 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3199 struct kvm_vcpu_events *events)
3203 * FIXME: pass injected and pending separately. This is only
3204 * needed for nested virtualization, whose state cannot be
3205 * migrated yet. For now we can combine them.
3207 events->exception.injected =
3208 (vcpu->arch.exception.pending ||
3209 vcpu->arch.exception.injected) &&
3210 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3211 events->exception.nr = vcpu->arch.exception.nr;
3212 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3213 events->exception.pad = 0;
3214 events->exception.error_code = vcpu->arch.exception.error_code;
3216 events->interrupt.injected =
3217 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3218 events->interrupt.nr = vcpu->arch.interrupt.nr;
3219 events->interrupt.soft = 0;
3220 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3222 events->nmi.injected = vcpu->arch.nmi_injected;
3223 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3224 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3225 events->nmi.pad = 0;
3227 events->sipi_vector = 0; /* never valid when reporting to user space */
3229 events->smi.smm = is_smm(vcpu);
3230 events->smi.pending = vcpu->arch.smi_pending;
3231 events->smi.smm_inside_nmi =
3232 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3233 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3235 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3236 | KVM_VCPUEVENT_VALID_SHADOW
3237 | KVM_VCPUEVENT_VALID_SMM);
3238 memset(&events->reserved, 0, sizeof(events->reserved));
3241 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3243 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3244 struct kvm_vcpu_events *events)
3246 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3247 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3248 | KVM_VCPUEVENT_VALID_SHADOW
3249 | KVM_VCPUEVENT_VALID_SMM))
3252 if (events->exception.injected &&
3253 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3254 is_guest_mode(vcpu)))
3257 /* INITs are latched while in SMM */
3258 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3259 (events->smi.smm || events->smi.pending) &&
3260 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3264 vcpu->arch.exception.injected = false;
3265 vcpu->arch.exception.pending = events->exception.injected;
3266 vcpu->arch.exception.nr = events->exception.nr;
3267 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3268 vcpu->arch.exception.error_code = events->exception.error_code;
3270 vcpu->arch.interrupt.pending = events->interrupt.injected;
3271 vcpu->arch.interrupt.nr = events->interrupt.nr;
3272 vcpu->arch.interrupt.soft = events->interrupt.soft;
3273 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3274 kvm_x86_ops->set_interrupt_shadow(vcpu,
3275 events->interrupt.shadow);
3277 vcpu->arch.nmi_injected = events->nmi.injected;
3278 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3279 vcpu->arch.nmi_pending = events->nmi.pending;
3280 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3282 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3283 lapic_in_kernel(vcpu))
3284 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3286 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3287 u32 hflags = vcpu->arch.hflags;
3288 if (events->smi.smm)
3289 hflags |= HF_SMM_MASK;
3291 hflags &= ~HF_SMM_MASK;
3292 kvm_set_hflags(vcpu, hflags);
3294 vcpu->arch.smi_pending = events->smi.pending;
3296 if (events->smi.smm) {
3297 if (events->smi.smm_inside_nmi)
3298 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3300 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3301 if (lapic_in_kernel(vcpu)) {
3302 if (events->smi.latched_init)
3303 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3305 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3310 kvm_make_request(KVM_REQ_EVENT, vcpu);
3315 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3316 struct kvm_debugregs *dbgregs)
3320 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3321 kvm_get_dr(vcpu, 6, &val);
3323 dbgregs->dr7 = vcpu->arch.dr7;
3325 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3328 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3329 struct kvm_debugregs *dbgregs)
3334 if (dbgregs->dr6 & ~0xffffffffull)
3336 if (dbgregs->dr7 & ~0xffffffffull)
3339 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3340 kvm_update_dr0123(vcpu);
3341 vcpu->arch.dr6 = dbgregs->dr6;
3342 kvm_update_dr6(vcpu);
3343 vcpu->arch.dr7 = dbgregs->dr7;
3344 kvm_update_dr7(vcpu);
3349 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3351 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3353 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3354 u64 xstate_bv = xsave->header.xfeatures;
3358 * Copy legacy XSAVE area, to avoid complications with CPUID
3359 * leaves 0 and 1 in the loop below.
3361 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3364 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3365 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3368 * Copy each region from the possibly compacted offset to the
3369 * non-compacted offset.
3371 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3373 u64 feature = valid & -valid;
3374 int index = fls64(feature) - 1;
3375 void *src = get_xsave_addr(xsave, feature);
3378 u32 size, offset, ecx, edx;
3379 cpuid_count(XSTATE_CPUID, index,
3380 &size, &offset, &ecx, &edx);
3381 if (feature == XFEATURE_MASK_PKRU)
3382 memcpy(dest + offset, &vcpu->arch.pkru,
3383 sizeof(vcpu->arch.pkru));
3385 memcpy(dest + offset, src, size);
3393 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3395 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3396 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3400 * Copy legacy XSAVE area, to avoid complications with CPUID
3401 * leaves 0 and 1 in the loop below.
3403 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3405 /* Set XSTATE_BV and possibly XCOMP_BV. */
3406 xsave->header.xfeatures = xstate_bv;
3407 if (boot_cpu_has(X86_FEATURE_XSAVES))
3408 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3411 * Copy each region from the non-compacted offset to the
3412 * possibly compacted offset.
3414 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3416 u64 feature = valid & -valid;
3417 int index = fls64(feature) - 1;
3418 void *dest = get_xsave_addr(xsave, feature);
3421 u32 size, offset, ecx, edx;
3422 cpuid_count(XSTATE_CPUID, index,
3423 &size, &offset, &ecx, &edx);
3424 if (feature == XFEATURE_MASK_PKRU)
3425 memcpy(&vcpu->arch.pkru, src + offset,
3426 sizeof(vcpu->arch.pkru));
3428 memcpy(dest, src + offset, size);
3435 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3436 struct kvm_xsave *guest_xsave)
3438 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3439 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3440 fill_xsave((u8 *) guest_xsave->region, vcpu);
3442 memcpy(guest_xsave->region,
3443 &vcpu->arch.guest_fpu.state.fxsave,
3444 sizeof(struct fxregs_state));
3445 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3446 XFEATURE_MASK_FPSSE;
3450 #define XSAVE_MXCSR_OFFSET 24
3452 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3453 struct kvm_xsave *guest_xsave)
3456 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3457 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3459 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3461 * Here we allow setting states that are not present in
3462 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3463 * with old userspace.
3465 if (xstate_bv & ~kvm_supported_xcr0() ||
3466 mxcsr & ~mxcsr_feature_mask)
3468 load_xsave(vcpu, (u8 *)guest_xsave->region);
3470 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3471 mxcsr & ~mxcsr_feature_mask)
3473 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3474 guest_xsave->region, sizeof(struct fxregs_state));
3479 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3480 struct kvm_xcrs *guest_xcrs)
3482 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3483 guest_xcrs->nr_xcrs = 0;
3487 guest_xcrs->nr_xcrs = 1;
3488 guest_xcrs->flags = 0;
3489 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3490 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3493 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3494 struct kvm_xcrs *guest_xcrs)
3498 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3501 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3504 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3505 /* Only support XCR0 currently */
3506 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3507 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3508 guest_xcrs->xcrs[i].value);
3517 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3518 * stopped by the hypervisor. This function will be called from the host only.
3519 * EINVAL is returned when the host attempts to set the flag for a guest that
3520 * does not support pv clocks.
3522 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3524 if (!vcpu->arch.pv_time_enabled)
3526 vcpu->arch.pvclock_set_guest_stopped_request = true;
3527 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3531 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3532 struct kvm_enable_cap *cap)
3538 case KVM_CAP_HYPERV_SYNIC2:
3541 case KVM_CAP_HYPERV_SYNIC:
3542 if (!irqchip_in_kernel(vcpu->kvm))
3544 return kvm_hv_activate_synic(vcpu, cap->cap ==
3545 KVM_CAP_HYPERV_SYNIC2);
3551 long kvm_arch_vcpu_ioctl(struct file *filp,
3552 unsigned int ioctl, unsigned long arg)
3554 struct kvm_vcpu *vcpu = filp->private_data;
3555 void __user *argp = (void __user *)arg;
3558 struct kvm_lapic_state *lapic;
3559 struct kvm_xsave *xsave;
3560 struct kvm_xcrs *xcrs;
3566 case KVM_GET_LAPIC: {
3568 if (!lapic_in_kernel(vcpu))
3570 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3575 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3579 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3584 case KVM_SET_LAPIC: {
3586 if (!lapic_in_kernel(vcpu))
3588 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3589 if (IS_ERR(u.lapic))
3590 return PTR_ERR(u.lapic);
3592 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3595 case KVM_INTERRUPT: {
3596 struct kvm_interrupt irq;
3599 if (copy_from_user(&irq, argp, sizeof irq))
3601 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3605 r = kvm_vcpu_ioctl_nmi(vcpu);
3609 r = kvm_vcpu_ioctl_smi(vcpu);
3612 case KVM_SET_CPUID: {
3613 struct kvm_cpuid __user *cpuid_arg = argp;
3614 struct kvm_cpuid cpuid;
3617 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3619 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3622 case KVM_SET_CPUID2: {
3623 struct kvm_cpuid2 __user *cpuid_arg = argp;
3624 struct kvm_cpuid2 cpuid;
3627 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3629 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3630 cpuid_arg->entries);
3633 case KVM_GET_CPUID2: {
3634 struct kvm_cpuid2 __user *cpuid_arg = argp;
3635 struct kvm_cpuid2 cpuid;
3638 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3640 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3641 cpuid_arg->entries);
3645 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3650 case KVM_GET_MSRS: {
3651 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3652 r = msr_io(vcpu, argp, do_get_msr, 1);
3653 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3656 case KVM_SET_MSRS: {
3657 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3658 r = msr_io(vcpu, argp, do_set_msr, 0);
3659 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3662 case KVM_TPR_ACCESS_REPORTING: {
3663 struct kvm_tpr_access_ctl tac;
3666 if (copy_from_user(&tac, argp, sizeof tac))
3668 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3672 if (copy_to_user(argp, &tac, sizeof tac))
3677 case KVM_SET_VAPIC_ADDR: {
3678 struct kvm_vapic_addr va;
3682 if (!lapic_in_kernel(vcpu))
3685 if (copy_from_user(&va, argp, sizeof va))
3687 idx = srcu_read_lock(&vcpu->kvm->srcu);
3688 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3689 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3692 case KVM_X86_SETUP_MCE: {
3696 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3698 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3701 case KVM_X86_SET_MCE: {
3702 struct kvm_x86_mce mce;
3705 if (copy_from_user(&mce, argp, sizeof mce))
3707 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3710 case KVM_GET_VCPU_EVENTS: {
3711 struct kvm_vcpu_events events;
3713 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3716 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3721 case KVM_SET_VCPU_EVENTS: {
3722 struct kvm_vcpu_events events;
3725 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3728 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3731 case KVM_GET_DEBUGREGS: {
3732 struct kvm_debugregs dbgregs;
3734 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3737 if (copy_to_user(argp, &dbgregs,
3738 sizeof(struct kvm_debugregs)))
3743 case KVM_SET_DEBUGREGS: {
3744 struct kvm_debugregs dbgregs;
3747 if (copy_from_user(&dbgregs, argp,
3748 sizeof(struct kvm_debugregs)))
3751 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3754 case KVM_GET_XSAVE: {
3755 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3760 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3763 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3768 case KVM_SET_XSAVE: {
3769 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3770 if (IS_ERR(u.xsave))
3771 return PTR_ERR(u.xsave);
3773 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3776 case KVM_GET_XCRS: {
3777 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3782 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3785 if (copy_to_user(argp, u.xcrs,
3786 sizeof(struct kvm_xcrs)))
3791 case KVM_SET_XCRS: {
3792 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3794 return PTR_ERR(u.xcrs);
3796 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3799 case KVM_SET_TSC_KHZ: {
3803 user_tsc_khz = (u32)arg;
3805 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3808 if (user_tsc_khz == 0)
3809 user_tsc_khz = tsc_khz;
3811 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3816 case KVM_GET_TSC_KHZ: {
3817 r = vcpu->arch.virtual_tsc_khz;
3820 case KVM_KVMCLOCK_CTRL: {
3821 r = kvm_set_guest_paused(vcpu);
3824 case KVM_ENABLE_CAP: {
3825 struct kvm_enable_cap cap;
3828 if (copy_from_user(&cap, argp, sizeof(cap)))
3830 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3841 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3843 return VM_FAULT_SIGBUS;
3846 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3850 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3852 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3856 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3859 kvm->arch.ept_identity_map_addr = ident_addr;
3863 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3864 u32 kvm_nr_mmu_pages)
3866 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3869 mutex_lock(&kvm->slots_lock);
3871 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3872 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3874 mutex_unlock(&kvm->slots_lock);
3878 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3880 return kvm->arch.n_max_mmu_pages;
3883 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3885 struct kvm_pic *pic = kvm->arch.vpic;
3889 switch (chip->chip_id) {
3890 case KVM_IRQCHIP_PIC_MASTER:
3891 memcpy(&chip->chip.pic, &pic->pics[0],
3892 sizeof(struct kvm_pic_state));
3894 case KVM_IRQCHIP_PIC_SLAVE:
3895 memcpy(&chip->chip.pic, &pic->pics[1],
3896 sizeof(struct kvm_pic_state));
3898 case KVM_IRQCHIP_IOAPIC:
3899 kvm_get_ioapic(kvm, &chip->chip.ioapic);
3908 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3910 struct kvm_pic *pic = kvm->arch.vpic;
3914 switch (chip->chip_id) {
3915 case KVM_IRQCHIP_PIC_MASTER:
3916 spin_lock(&pic->lock);
3917 memcpy(&pic->pics[0], &chip->chip.pic,
3918 sizeof(struct kvm_pic_state));
3919 spin_unlock(&pic->lock);
3921 case KVM_IRQCHIP_PIC_SLAVE:
3922 spin_lock(&pic->lock);
3923 memcpy(&pic->pics[1], &chip->chip.pic,
3924 sizeof(struct kvm_pic_state));
3925 spin_unlock(&pic->lock);
3927 case KVM_IRQCHIP_IOAPIC:
3928 kvm_set_ioapic(kvm, &chip->chip.ioapic);
3934 kvm_pic_update_irq(pic);
3938 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3940 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3942 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3944 mutex_lock(&kps->lock);
3945 memcpy(ps, &kps->channels, sizeof(*ps));
3946 mutex_unlock(&kps->lock);
3950 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3953 struct kvm_pit *pit = kvm->arch.vpit;
3955 mutex_lock(&pit->pit_state.lock);
3956 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3957 for (i = 0; i < 3; i++)
3958 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3959 mutex_unlock(&pit->pit_state.lock);
3963 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3965 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3966 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3967 sizeof(ps->channels));
3968 ps->flags = kvm->arch.vpit->pit_state.flags;
3969 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3970 memset(&ps->reserved, 0, sizeof(ps->reserved));
3974 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3978 u32 prev_legacy, cur_legacy;
3979 struct kvm_pit *pit = kvm->arch.vpit;
3981 mutex_lock(&pit->pit_state.lock);
3982 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3983 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3984 if (!prev_legacy && cur_legacy)
3986 memcpy(&pit->pit_state.channels, &ps->channels,
3987 sizeof(pit->pit_state.channels));
3988 pit->pit_state.flags = ps->flags;
3989 for (i = 0; i < 3; i++)
3990 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3992 mutex_unlock(&pit->pit_state.lock);
3996 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3997 struct kvm_reinject_control *control)
3999 struct kvm_pit *pit = kvm->arch.vpit;
4004 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4005 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4006 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4008 mutex_lock(&pit->pit_state.lock);
4009 kvm_pit_set_reinject(pit, control->pit_reinject);
4010 mutex_unlock(&pit->pit_state.lock);
4016 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4017 * @kvm: kvm instance
4018 * @log: slot id and address to which we copy the log
4020 * Steps 1-4 below provide general overview of dirty page logging. See
4021 * kvm_get_dirty_log_protect() function description for additional details.
4023 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4024 * always flush the TLB (step 4) even if previous step failed and the dirty
4025 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4026 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4027 * writes will be marked dirty for next log read.
4029 * 1. Take a snapshot of the bit and clear it if needed.
4030 * 2. Write protect the corresponding page.
4031 * 3. Copy the snapshot to the userspace.
4032 * 4. Flush TLB's if needed.
4034 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4036 bool is_dirty = false;
4039 mutex_lock(&kvm->slots_lock);
4042 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4044 if (kvm_x86_ops->flush_log_dirty)
4045 kvm_x86_ops->flush_log_dirty(kvm);
4047 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
4050 * All the TLBs can be flushed out of mmu lock, see the comments in
4051 * kvm_mmu_slot_remove_write_access().
4053 lockdep_assert_held(&kvm->slots_lock);
4055 kvm_flush_remote_tlbs(kvm);
4057 mutex_unlock(&kvm->slots_lock);
4061 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4064 if (!irqchip_in_kernel(kvm))
4067 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4068 irq_event->irq, irq_event->level,
4073 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4074 struct kvm_enable_cap *cap)
4082 case KVM_CAP_DISABLE_QUIRKS:
4083 kvm->arch.disabled_quirks = cap->args[0];
4086 case KVM_CAP_SPLIT_IRQCHIP: {
4087 mutex_lock(&kvm->lock);
4089 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4090 goto split_irqchip_unlock;
4092 if (irqchip_in_kernel(kvm))
4093 goto split_irqchip_unlock;
4094 if (kvm->created_vcpus)
4095 goto split_irqchip_unlock;
4096 r = kvm_setup_empty_irq_routing(kvm);
4098 goto split_irqchip_unlock;
4099 /* Pairs with irqchip_in_kernel. */
4101 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4102 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4104 split_irqchip_unlock:
4105 mutex_unlock(&kvm->lock);
4108 case KVM_CAP_X2APIC_API:
4110 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4113 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4114 kvm->arch.x2apic_format = true;
4115 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4116 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4127 long kvm_arch_vm_ioctl(struct file *filp,
4128 unsigned int ioctl, unsigned long arg)
4130 struct kvm *kvm = filp->private_data;
4131 void __user *argp = (void __user *)arg;
4134 * This union makes it completely explicit to gcc-3.x
4135 * that these two variables' stack usage should be
4136 * combined, not added together.
4139 struct kvm_pit_state ps;
4140 struct kvm_pit_state2 ps2;
4141 struct kvm_pit_config pit_config;
4145 case KVM_SET_TSS_ADDR:
4146 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4148 case KVM_SET_IDENTITY_MAP_ADDR: {
4152 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4154 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4157 case KVM_SET_NR_MMU_PAGES:
4158 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4160 case KVM_GET_NR_MMU_PAGES:
4161 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4163 case KVM_CREATE_IRQCHIP: {
4164 mutex_lock(&kvm->lock);
4167 if (irqchip_in_kernel(kvm))
4168 goto create_irqchip_unlock;
4171 if (kvm->created_vcpus)
4172 goto create_irqchip_unlock;
4174 r = kvm_pic_init(kvm);
4176 goto create_irqchip_unlock;
4178 r = kvm_ioapic_init(kvm);
4180 kvm_pic_destroy(kvm);
4181 goto create_irqchip_unlock;
4184 r = kvm_setup_default_irq_routing(kvm);
4186 kvm_ioapic_destroy(kvm);
4187 kvm_pic_destroy(kvm);
4188 goto create_irqchip_unlock;
4190 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4192 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4193 create_irqchip_unlock:
4194 mutex_unlock(&kvm->lock);
4197 case KVM_CREATE_PIT:
4198 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4200 case KVM_CREATE_PIT2:
4202 if (copy_from_user(&u.pit_config, argp,
4203 sizeof(struct kvm_pit_config)))
4206 mutex_lock(&kvm->lock);
4209 goto create_pit_unlock;
4211 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4215 mutex_unlock(&kvm->lock);
4217 case KVM_GET_IRQCHIP: {
4218 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4219 struct kvm_irqchip *chip;
4221 chip = memdup_user(argp, sizeof(*chip));
4228 if (!irqchip_kernel(kvm))
4229 goto get_irqchip_out;
4230 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4232 goto get_irqchip_out;
4234 if (copy_to_user(argp, chip, sizeof *chip))
4235 goto get_irqchip_out;
4241 case KVM_SET_IRQCHIP: {
4242 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4243 struct kvm_irqchip *chip;
4245 chip = memdup_user(argp, sizeof(*chip));
4252 if (!irqchip_kernel(kvm))
4253 goto set_irqchip_out;
4254 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4256 goto set_irqchip_out;
4264 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4267 if (!kvm->arch.vpit)
4269 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4273 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4280 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4283 if (!kvm->arch.vpit)
4285 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4288 case KVM_GET_PIT2: {
4290 if (!kvm->arch.vpit)
4292 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4296 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4301 case KVM_SET_PIT2: {
4303 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4306 if (!kvm->arch.vpit)
4308 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4311 case KVM_REINJECT_CONTROL: {
4312 struct kvm_reinject_control control;
4314 if (copy_from_user(&control, argp, sizeof(control)))
4316 r = kvm_vm_ioctl_reinject(kvm, &control);
4319 case KVM_SET_BOOT_CPU_ID:
4321 mutex_lock(&kvm->lock);
4322 if (kvm->created_vcpus)
4325 kvm->arch.bsp_vcpu_id = arg;
4326 mutex_unlock(&kvm->lock);
4328 case KVM_XEN_HVM_CONFIG: {
4329 struct kvm_xen_hvm_config xhc;
4331 if (copy_from_user(&xhc, argp, sizeof(xhc)))
4336 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4340 case KVM_SET_CLOCK: {
4341 struct kvm_clock_data user_ns;
4345 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4354 * TODO: userspace has to take care of races with VCPU_RUN, so
4355 * kvm_gen_update_masterclock() can be cut down to locked
4356 * pvclock_update_vm_gtod_copy().
4358 kvm_gen_update_masterclock(kvm);
4359 now_ns = get_kvmclock_ns(kvm);
4360 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4361 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4364 case KVM_GET_CLOCK: {
4365 struct kvm_clock_data user_ns;
4368 now_ns = get_kvmclock_ns(kvm);
4369 user_ns.clock = now_ns;
4370 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4371 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4374 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4379 case KVM_ENABLE_CAP: {
4380 struct kvm_enable_cap cap;
4383 if (copy_from_user(&cap, argp, sizeof(cap)))
4385 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4395 static void kvm_init_msr_list(void)
4400 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4401 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4405 * Even MSRs that are valid in the host may not be exposed
4406 * to the guests in some cases.
4408 switch (msrs_to_save[i]) {
4409 case MSR_IA32_BNDCFGS:
4410 if (!kvm_x86_ops->mpx_supported())
4414 if (!kvm_x86_ops->rdtscp_supported())
4422 msrs_to_save[j] = msrs_to_save[i];
4425 num_msrs_to_save = j;
4427 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4428 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4432 emulated_msrs[j] = emulated_msrs[i];
4435 num_emulated_msrs = j;
4437 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4438 struct kvm_msr_entry msr;
4440 msr.index = msr_based_features[i];
4441 if (kvm_get_msr_feature(&msr))
4445 msr_based_features[j] = msr_based_features[i];
4448 num_msr_based_features = j;
4451 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4459 if (!(lapic_in_kernel(vcpu) &&
4460 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4461 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4472 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4479 if (!(lapic_in_kernel(vcpu) &&
4480 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4482 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4484 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4494 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4495 struct kvm_segment *var, int seg)
4497 kvm_x86_ops->set_segment(vcpu, var, seg);
4500 void kvm_get_segment(struct kvm_vcpu *vcpu,
4501 struct kvm_segment *var, int seg)
4503 kvm_x86_ops->get_segment(vcpu, var, seg);
4506 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4507 struct x86_exception *exception)
4511 BUG_ON(!mmu_is_nested(vcpu));
4513 /* NPT walks are always user-walks */
4514 access |= PFERR_USER_MASK;
4515 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4520 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4521 struct x86_exception *exception)
4523 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4524 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4527 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4528 struct x86_exception *exception)
4530 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4531 access |= PFERR_FETCH_MASK;
4532 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4535 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4536 struct x86_exception *exception)
4538 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4539 access |= PFERR_WRITE_MASK;
4540 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4543 /* uses this to access any guest's mapped memory without checking CPL */
4544 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4545 struct x86_exception *exception)
4547 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4550 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4551 struct kvm_vcpu *vcpu, u32 access,
4552 struct x86_exception *exception)
4555 int r = X86EMUL_CONTINUE;
4558 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4560 unsigned offset = addr & (PAGE_SIZE-1);
4561 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4564 if (gpa == UNMAPPED_GVA)
4565 return X86EMUL_PROPAGATE_FAULT;
4566 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4569 r = X86EMUL_IO_NEEDED;
4581 /* used for instruction fetching */
4582 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4583 gva_t addr, void *val, unsigned int bytes,
4584 struct x86_exception *exception)
4586 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4587 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4591 /* Inline kvm_read_guest_virt_helper for speed. */
4592 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4594 if (unlikely(gpa == UNMAPPED_GVA))
4595 return X86EMUL_PROPAGATE_FAULT;
4597 offset = addr & (PAGE_SIZE-1);
4598 if (WARN_ON(offset + bytes > PAGE_SIZE))
4599 bytes = (unsigned)PAGE_SIZE - offset;
4600 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4602 if (unlikely(ret < 0))
4603 return X86EMUL_IO_NEEDED;
4605 return X86EMUL_CONTINUE;
4608 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
4609 gva_t addr, void *val, unsigned int bytes,
4610 struct x86_exception *exception)
4612 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4615 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
4616 * is returned, but our callers are not ready for that and they blindly
4617 * call kvm_inject_page_fault. Ensure that they at least do not leak
4618 * uninitialized kernel stack memory into cr2 and error code.
4620 memset(exception, 0, sizeof(*exception));
4621 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4624 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4626 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
4627 gva_t addr, void *val, unsigned int bytes,
4628 struct x86_exception *exception, bool system)
4630 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4633 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
4634 access |= PFERR_USER_MASK;
4636 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
4639 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4640 unsigned long addr, void *val, unsigned int bytes)
4642 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4643 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4645 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4648 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4649 struct kvm_vcpu *vcpu, u32 access,
4650 struct x86_exception *exception)
4653 int r = X86EMUL_CONTINUE;
4656 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4659 unsigned offset = addr & (PAGE_SIZE-1);
4660 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4663 if (gpa == UNMAPPED_GVA)
4664 return X86EMUL_PROPAGATE_FAULT;
4665 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4667 r = X86EMUL_IO_NEEDED;
4679 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
4680 unsigned int bytes, struct x86_exception *exception,
4683 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4684 u32 access = PFERR_WRITE_MASK;
4686 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
4687 access |= PFERR_USER_MASK;
4689 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
4693 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
4694 unsigned int bytes, struct x86_exception *exception)
4696 /* kvm_write_guest_virt_system can pull in tons of pages. */
4697 vcpu->arch.l1tf_flush_l1d = true;
4699 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
4700 PFERR_WRITE_MASK, exception);
4702 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4704 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4705 gpa_t gpa, bool write)
4707 /* For APIC access vmexit */
4708 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4711 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4712 trace_vcpu_match_mmio(gva, gpa, write, true);
4719 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4720 gpa_t *gpa, struct x86_exception *exception,
4723 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4724 | (write ? PFERR_WRITE_MASK : 0);
4727 * currently PKRU is only applied to ept enabled guest so
4728 * there is no pkey in EPT page table for L1 guest or EPT
4729 * shadow page table for L2 guest.
4731 if (vcpu_match_mmio_gva(vcpu, gva)
4732 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4733 vcpu->arch.access, 0, access)) {
4734 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4735 (gva & (PAGE_SIZE - 1));
4736 trace_vcpu_match_mmio(gva, *gpa, write, false);
4740 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4742 if (*gpa == UNMAPPED_GVA)
4745 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4748 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4749 const void *val, int bytes)
4753 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4756 kvm_page_track_write(vcpu, gpa, val, bytes);
4760 struct read_write_emulator_ops {
4761 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4763 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4764 void *val, int bytes);
4765 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4766 int bytes, void *val);
4767 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4768 void *val, int bytes);
4772 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4774 if (vcpu->mmio_read_completed) {
4775 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4776 vcpu->mmio_fragments[0].gpa, val);
4777 vcpu->mmio_read_completed = 0;
4784 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4785 void *val, int bytes)
4787 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4790 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4791 void *val, int bytes)
4793 return emulator_write_phys(vcpu, gpa, val, bytes);
4796 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4798 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
4799 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4802 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4803 void *val, int bytes)
4805 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
4806 return X86EMUL_IO_NEEDED;
4809 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4810 void *val, int bytes)
4812 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4814 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4815 return X86EMUL_CONTINUE;
4818 static const struct read_write_emulator_ops read_emultor = {
4819 .read_write_prepare = read_prepare,
4820 .read_write_emulate = read_emulate,
4821 .read_write_mmio = vcpu_mmio_read,
4822 .read_write_exit_mmio = read_exit_mmio,
4825 static const struct read_write_emulator_ops write_emultor = {
4826 .read_write_emulate = write_emulate,
4827 .read_write_mmio = write_mmio,
4828 .read_write_exit_mmio = write_exit_mmio,
4832 static int emulator_read_write_onepage(unsigned long addr, void *val,
4834 struct x86_exception *exception,
4835 struct kvm_vcpu *vcpu,
4836 const struct read_write_emulator_ops *ops)
4840 bool write = ops->write;
4841 struct kvm_mmio_fragment *frag;
4842 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4845 * If the exit was due to a NPF we may already have a GPA.
4846 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4847 * Note, this cannot be used on string operations since string
4848 * operation using rep will only have the initial GPA from the NPF
4851 if (vcpu->arch.gpa_available &&
4852 emulator_can_use_gpa(ctxt) &&
4853 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
4854 gpa = vcpu->arch.gpa_val;
4855 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
4857 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4859 return X86EMUL_PROPAGATE_FAULT;
4862 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
4863 return X86EMUL_CONTINUE;
4866 * Is this MMIO handled locally?
4868 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4869 if (handled == bytes)
4870 return X86EMUL_CONTINUE;
4876 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4877 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4881 return X86EMUL_CONTINUE;
4884 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4886 void *val, unsigned int bytes,
4887 struct x86_exception *exception,
4888 const struct read_write_emulator_ops *ops)
4890 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4894 if (ops->read_write_prepare &&
4895 ops->read_write_prepare(vcpu, val, bytes))
4896 return X86EMUL_CONTINUE;
4898 vcpu->mmio_nr_fragments = 0;
4900 /* Crossing a page boundary? */
4901 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4904 now = -addr & ~PAGE_MASK;
4905 rc = emulator_read_write_onepage(addr, val, now, exception,
4908 if (rc != X86EMUL_CONTINUE)
4911 if (ctxt->mode != X86EMUL_MODE_PROT64)
4917 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4919 if (rc != X86EMUL_CONTINUE)
4922 if (!vcpu->mmio_nr_fragments)
4925 gpa = vcpu->mmio_fragments[0].gpa;
4927 vcpu->mmio_needed = 1;
4928 vcpu->mmio_cur_fragment = 0;
4930 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4931 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4932 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4933 vcpu->run->mmio.phys_addr = gpa;
4935 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4938 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4942 struct x86_exception *exception)
4944 return emulator_read_write(ctxt, addr, val, bytes,
4945 exception, &read_emultor);
4948 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4952 struct x86_exception *exception)
4954 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4955 exception, &write_emultor);
4958 #define CMPXCHG_TYPE(t, ptr, old, new) \
4959 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4961 #ifdef CONFIG_X86_64
4962 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4964 # define CMPXCHG64(ptr, old, new) \
4965 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4968 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4973 struct x86_exception *exception)
4975 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4981 /* guests cmpxchg8b have to be emulated atomically */
4982 if (bytes > 8 || (bytes & (bytes - 1)))
4985 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4987 if (gpa == UNMAPPED_GVA ||
4988 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4991 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4994 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4995 if (is_error_page(page))
4998 kaddr = kmap_atomic(page);
4999 kaddr += offset_in_page(gpa);
5002 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5005 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5008 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5011 exchanged = CMPXCHG64(kaddr, old, new);
5016 kunmap_atomic(kaddr);
5017 kvm_release_page_dirty(page);
5020 return X86EMUL_CMPXCHG_FAILED;
5022 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5023 kvm_page_track_write(vcpu, gpa, new, bytes);
5025 return X86EMUL_CONTINUE;
5028 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5030 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5033 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5037 for (i = 0; i < vcpu->arch.pio.count; i++) {
5038 if (vcpu->arch.pio.in)
5039 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5040 vcpu->arch.pio.size, pd);
5042 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5043 vcpu->arch.pio.port, vcpu->arch.pio.size,
5047 pd += vcpu->arch.pio.size;
5052 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5053 unsigned short port, void *val,
5054 unsigned int count, bool in)
5056 vcpu->arch.pio.port = port;
5057 vcpu->arch.pio.in = in;
5058 vcpu->arch.pio.count = count;
5059 vcpu->arch.pio.size = size;
5061 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5062 vcpu->arch.pio.count = 0;
5066 vcpu->run->exit_reason = KVM_EXIT_IO;
5067 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5068 vcpu->run->io.size = size;
5069 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5070 vcpu->run->io.count = count;
5071 vcpu->run->io.port = port;
5076 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5077 int size, unsigned short port, void *val,
5080 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5083 if (vcpu->arch.pio.count)
5086 memset(vcpu->arch.pio_data, 0, size * count);
5088 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5091 memcpy(val, vcpu->arch.pio_data, size * count);
5092 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5093 vcpu->arch.pio.count = 0;
5100 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5101 int size, unsigned short port,
5102 const void *val, unsigned int count)
5104 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5106 memcpy(vcpu->arch.pio_data, val, size * count);
5107 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5108 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5111 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5113 return kvm_x86_ops->get_segment_base(vcpu, seg);
5116 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5118 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5121 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5123 if (!need_emulate_wbinvd(vcpu))
5124 return X86EMUL_CONTINUE;
5126 if (kvm_x86_ops->has_wbinvd_exit()) {
5127 int cpu = get_cpu();
5129 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5130 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5131 wbinvd_ipi, NULL, 1);
5133 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5136 return X86EMUL_CONTINUE;
5139 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5141 kvm_emulate_wbinvd_noskip(vcpu);
5142 return kvm_skip_emulated_instruction(vcpu);
5144 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5148 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5150 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5153 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5154 unsigned long *dest)
5156 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5159 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5160 unsigned long value)
5163 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5166 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5168 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5171 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5173 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5174 unsigned long value;
5178 value = kvm_read_cr0(vcpu);
5181 value = vcpu->arch.cr2;
5184 value = kvm_read_cr3(vcpu);
5187 value = kvm_read_cr4(vcpu);
5190 value = kvm_get_cr8(vcpu);
5193 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5200 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5202 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5207 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5210 vcpu->arch.cr2 = val;
5213 res = kvm_set_cr3(vcpu, val);
5216 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5219 res = kvm_set_cr8(vcpu, val);
5222 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5229 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5231 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5234 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5236 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5239 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5241 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5244 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5246 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5249 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5251 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5254 static unsigned long emulator_get_cached_segment_base(
5255 struct x86_emulate_ctxt *ctxt, int seg)
5257 return get_segment_base(emul_to_vcpu(ctxt), seg);
5260 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5261 struct desc_struct *desc, u32 *base3,
5264 struct kvm_segment var;
5266 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5267 *selector = var.selector;
5270 memset(desc, 0, sizeof(*desc));
5278 set_desc_limit(desc, var.limit);
5279 set_desc_base(desc, (unsigned long)var.base);
5280 #ifdef CONFIG_X86_64
5282 *base3 = var.base >> 32;
5284 desc->type = var.type;
5286 desc->dpl = var.dpl;
5287 desc->p = var.present;
5288 desc->avl = var.avl;
5296 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5297 struct desc_struct *desc, u32 base3,
5300 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5301 struct kvm_segment var;
5303 var.selector = selector;
5304 var.base = get_desc_base(desc);
5305 #ifdef CONFIG_X86_64
5306 var.base |= ((u64)base3) << 32;
5308 var.limit = get_desc_limit(desc);
5310 var.limit = (var.limit << 12) | 0xfff;
5311 var.type = desc->type;
5312 var.dpl = desc->dpl;
5317 var.avl = desc->avl;
5318 var.present = desc->p;
5319 var.unusable = !var.present;
5322 kvm_set_segment(vcpu, &var, seg);
5326 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5327 u32 msr_index, u64 *pdata)
5329 struct msr_data msr;
5332 msr.index = msr_index;
5333 msr.host_initiated = false;
5334 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5342 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5343 u32 msr_index, u64 data)
5345 struct msr_data msr;
5348 msr.index = msr_index;
5349 msr.host_initiated = false;
5350 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5353 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5355 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5357 return vcpu->arch.smbase;
5360 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5362 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5364 vcpu->arch.smbase = smbase;
5367 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5370 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5373 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5374 u32 pmc, u64 *pdata)
5376 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5379 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5381 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5384 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5388 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5392 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5393 struct x86_instruction_info *info,
5394 enum x86_intercept_stage stage)
5396 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5399 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5400 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5402 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5405 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5407 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5410 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5412 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5415 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5417 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5420 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5422 return emul_to_vcpu(ctxt)->arch.hflags;
5425 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5427 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5430 static const struct x86_emulate_ops emulate_ops = {
5431 .read_gpr = emulator_read_gpr,
5432 .write_gpr = emulator_write_gpr,
5433 .read_std = emulator_read_std,
5434 .write_std = emulator_write_std,
5435 .read_phys = kvm_read_guest_phys_system,
5436 .fetch = kvm_fetch_guest_virt,
5437 .read_emulated = emulator_read_emulated,
5438 .write_emulated = emulator_write_emulated,
5439 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5440 .invlpg = emulator_invlpg,
5441 .pio_in_emulated = emulator_pio_in_emulated,
5442 .pio_out_emulated = emulator_pio_out_emulated,
5443 .get_segment = emulator_get_segment,
5444 .set_segment = emulator_set_segment,
5445 .get_cached_segment_base = emulator_get_cached_segment_base,
5446 .get_gdt = emulator_get_gdt,
5447 .get_idt = emulator_get_idt,
5448 .set_gdt = emulator_set_gdt,
5449 .set_idt = emulator_set_idt,
5450 .get_cr = emulator_get_cr,
5451 .set_cr = emulator_set_cr,
5452 .cpl = emulator_get_cpl,
5453 .get_dr = emulator_get_dr,
5454 .set_dr = emulator_set_dr,
5455 .get_smbase = emulator_get_smbase,
5456 .set_smbase = emulator_set_smbase,
5457 .set_msr = emulator_set_msr,
5458 .get_msr = emulator_get_msr,
5459 .check_pmc = emulator_check_pmc,
5460 .read_pmc = emulator_read_pmc,
5461 .halt = emulator_halt,
5462 .wbinvd = emulator_wbinvd,
5463 .fix_hypercall = emulator_fix_hypercall,
5464 .get_fpu = emulator_get_fpu,
5465 .put_fpu = emulator_put_fpu,
5466 .intercept = emulator_intercept,
5467 .get_cpuid = emulator_get_cpuid,
5468 .set_nmi_mask = emulator_set_nmi_mask,
5469 .get_hflags = emulator_get_hflags,
5470 .set_hflags = emulator_set_hflags,
5473 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5475 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5477 * an sti; sti; sequence only disable interrupts for the first
5478 * instruction. So, if the last instruction, be it emulated or
5479 * not, left the system with the INT_STI flag enabled, it
5480 * means that the last instruction is an sti. We should not
5481 * leave the flag on in this case. The same goes for mov ss
5483 if (int_shadow & mask)
5485 if (unlikely(int_shadow || mask)) {
5486 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5488 kvm_make_request(KVM_REQ_EVENT, vcpu);
5492 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5494 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5495 if (ctxt->exception.vector == PF_VECTOR)
5496 return kvm_propagate_fault(vcpu, &ctxt->exception);
5498 if (ctxt->exception.error_code_valid)
5499 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5500 ctxt->exception.error_code);
5502 kvm_queue_exception(vcpu, ctxt->exception.vector);
5506 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5508 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5511 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5513 ctxt->eflags = kvm_get_rflags(vcpu);
5514 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5516 ctxt->eip = kvm_rip_read(vcpu);
5517 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5518 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5519 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5520 cs_db ? X86EMUL_MODE_PROT32 :
5521 X86EMUL_MODE_PROT16;
5522 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5523 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5524 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5526 init_decode_cache(ctxt);
5527 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5530 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5532 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5535 init_emulate_ctxt(vcpu);
5539 ctxt->_eip = ctxt->eip + inc_eip;
5540 ret = emulate_int_real(ctxt, irq);
5542 if (ret != X86EMUL_CONTINUE)
5543 return EMULATE_FAIL;
5545 ctxt->eip = ctxt->_eip;
5546 kvm_rip_write(vcpu, ctxt->eip);
5547 kvm_set_rflags(vcpu, ctxt->eflags);
5549 if (irq == NMI_VECTOR)
5550 vcpu->arch.nmi_pending = 0;
5552 vcpu->arch.interrupt.pending = false;
5554 return EMULATE_DONE;
5556 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5558 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5560 int r = EMULATE_DONE;
5562 ++vcpu->stat.insn_emulation_fail;
5563 trace_kvm_emulate_insn_failed(vcpu);
5564 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5565 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5566 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5567 vcpu->run->internal.ndata = 0;
5568 r = EMULATE_USER_EXIT;
5570 kvm_queue_exception(vcpu, UD_VECTOR);
5575 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5576 bool write_fault_to_shadow_pgtable,
5582 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5585 if (!vcpu->arch.mmu.direct_map) {
5587 * Write permission should be allowed since only
5588 * write access need to be emulated.
5590 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5593 * If the mapping is invalid in guest, let cpu retry
5594 * it to generate fault.
5596 if (gpa == UNMAPPED_GVA)
5601 * Do not retry the unhandleable instruction if it faults on the
5602 * readonly host memory, otherwise it will goto a infinite loop:
5603 * retry instruction -> write #PF -> emulation fail -> retry
5604 * instruction -> ...
5606 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5609 * If the instruction failed on the error pfn, it can not be fixed,
5610 * report the error to userspace.
5612 if (is_error_noslot_pfn(pfn))
5615 kvm_release_pfn_clean(pfn);
5617 /* The instructions are well-emulated on direct mmu. */
5618 if (vcpu->arch.mmu.direct_map) {
5619 unsigned int indirect_shadow_pages;
5621 spin_lock(&vcpu->kvm->mmu_lock);
5622 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5623 spin_unlock(&vcpu->kvm->mmu_lock);
5625 if (indirect_shadow_pages)
5626 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5632 * if emulation was due to access to shadowed page table
5633 * and it failed try to unshadow page and re-enter the
5634 * guest to let CPU execute the instruction.
5636 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5639 * If the access faults on its page table, it can not
5640 * be fixed by unprotecting shadow page and it should
5641 * be reported to userspace.
5643 return !write_fault_to_shadow_pgtable;
5646 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5647 unsigned long cr2, int emulation_type)
5649 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5650 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5652 last_retry_eip = vcpu->arch.last_retry_eip;
5653 last_retry_addr = vcpu->arch.last_retry_addr;
5656 * If the emulation is caused by #PF and it is non-page_table
5657 * writing instruction, it means the VM-EXIT is caused by shadow
5658 * page protected, we can zap the shadow page and retry this
5659 * instruction directly.
5661 * Note: if the guest uses a non-page-table modifying instruction
5662 * on the PDE that points to the instruction, then we will unmap
5663 * the instruction and go to an infinite loop. So, we cache the
5664 * last retried eip and the last fault address, if we meet the eip
5665 * and the address again, we can break out of the potential infinite
5668 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5670 if (!(emulation_type & EMULTYPE_RETRY))
5673 if (x86_page_table_writing_insn(ctxt))
5676 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5679 vcpu->arch.last_retry_eip = ctxt->eip;
5680 vcpu->arch.last_retry_addr = cr2;
5682 if (!vcpu->arch.mmu.direct_map)
5683 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5685 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5690 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5691 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5693 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5695 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5696 /* This is a good place to trace that we are exiting SMM. */
5697 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5699 /* Process a latched INIT or SMI, if any. */
5700 kvm_make_request(KVM_REQ_EVENT, vcpu);
5703 kvm_mmu_reset_context(vcpu);
5706 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5708 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5710 vcpu->arch.hflags = emul_flags;
5712 if (changed & HF_SMM_MASK)
5713 kvm_smm_changed(vcpu);
5716 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5725 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5726 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5731 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
5733 struct kvm_run *kvm_run = vcpu->run;
5735 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5736 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5737 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5738 kvm_run->debug.arch.exception = DB_VECTOR;
5739 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5740 *r = EMULATE_USER_EXIT;
5743 * "Certain debug exceptions may clear bit 0-3. The
5744 * remaining contents of the DR6 register are never
5745 * cleared by the processor".
5747 vcpu->arch.dr6 &= ~15;
5748 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5749 kvm_queue_exception(vcpu, DB_VECTOR);
5753 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5755 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5756 int r = EMULATE_DONE;
5758 kvm_x86_ops->skip_emulated_instruction(vcpu);
5761 * rflags is the old, "raw" value of the flags. The new value has
5762 * not been saved yet.
5764 * This is correct even for TF set by the guest, because "the
5765 * processor will not generate this exception after the instruction
5766 * that sets the TF flag".
5768 if (unlikely(rflags & X86_EFLAGS_TF))
5769 kvm_vcpu_do_singlestep(vcpu, &r);
5770 return r == EMULATE_DONE;
5772 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5774 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5776 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5777 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5778 struct kvm_run *kvm_run = vcpu->run;
5779 unsigned long eip = kvm_get_linear_rip(vcpu);
5780 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5781 vcpu->arch.guest_debug_dr7,
5785 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5786 kvm_run->debug.arch.pc = eip;
5787 kvm_run->debug.arch.exception = DB_VECTOR;
5788 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5789 *r = EMULATE_USER_EXIT;
5794 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5795 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5796 unsigned long eip = kvm_get_linear_rip(vcpu);
5797 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5802 vcpu->arch.dr6 &= ~15;
5803 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5804 kvm_queue_exception(vcpu, DB_VECTOR);
5813 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5820 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5821 bool writeback = true;
5822 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5824 vcpu->arch.l1tf_flush_l1d = true;
5827 * Clear write_fault_to_shadow_pgtable here to ensure it is
5830 vcpu->arch.write_fault_to_shadow_pgtable = false;
5831 kvm_clear_exception_queue(vcpu);
5833 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5834 init_emulate_ctxt(vcpu);
5837 * We will reenter on the same instruction since
5838 * we do not set complete_userspace_io. This does not
5839 * handle watchpoints yet, those would be handled in
5842 if (!(emulation_type & EMULTYPE_SKIP) &&
5843 kvm_vcpu_check_breakpoint(vcpu, &r))
5846 ctxt->interruptibility = 0;
5847 ctxt->have_exception = false;
5848 ctxt->exception.vector = -1;
5849 ctxt->perm_ok = false;
5851 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5853 r = x86_decode_insn(ctxt, insn, insn_len);
5855 trace_kvm_emulate_insn_start(vcpu);
5856 ++vcpu->stat.insn_emulation;
5857 if (r != EMULATION_OK) {
5858 if (emulation_type & EMULTYPE_TRAP_UD)
5859 return EMULATE_FAIL;
5860 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5862 return EMULATE_DONE;
5863 if (ctxt->have_exception && inject_emulated_exception(vcpu))
5864 return EMULATE_DONE;
5865 if (emulation_type & EMULTYPE_SKIP)
5866 return EMULATE_FAIL;
5867 return handle_emulation_failure(vcpu);
5871 if (emulation_type & EMULTYPE_SKIP) {
5872 kvm_rip_write(vcpu, ctxt->_eip);
5873 if (ctxt->eflags & X86_EFLAGS_RF)
5874 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5875 return EMULATE_DONE;
5878 if (retry_instruction(ctxt, cr2, emulation_type))
5879 return EMULATE_DONE;
5881 /* this is needed for vmware backdoor interface to work since it
5882 changes registers values during IO operation */
5883 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5884 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5885 emulator_invalidate_register_cache(ctxt);
5889 /* Save the faulting GPA (cr2) in the address field */
5890 ctxt->exception.address = cr2;
5892 r = x86_emulate_insn(ctxt);
5894 if (r == EMULATION_INTERCEPTED)
5895 return EMULATE_DONE;
5897 if (r == EMULATION_FAILED) {
5898 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5900 return EMULATE_DONE;
5902 return handle_emulation_failure(vcpu);
5905 if (ctxt->have_exception) {
5907 if (inject_emulated_exception(vcpu))
5909 } else if (vcpu->arch.pio.count) {
5910 if (!vcpu->arch.pio.in) {
5911 /* FIXME: return into emulator if single-stepping. */
5912 vcpu->arch.pio.count = 0;
5915 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5917 r = EMULATE_USER_EXIT;
5918 } else if (vcpu->mmio_needed) {
5919 if (!vcpu->mmio_is_write)
5921 r = EMULATE_USER_EXIT;
5922 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5923 } else if (r == EMULATION_RESTART)
5929 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5930 toggle_interruptibility(vcpu, ctxt->interruptibility);
5931 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5932 kvm_rip_write(vcpu, ctxt->eip);
5933 if (r == EMULATE_DONE && ctxt->tf)
5934 kvm_vcpu_do_singlestep(vcpu, &r);
5935 if (!ctxt->have_exception ||
5936 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5937 __kvm_set_rflags(vcpu, ctxt->eflags);
5940 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5941 * do nothing, and it will be requested again as soon as
5942 * the shadow expires. But we still need to check here,
5943 * because POPF has no interrupt shadow.
5945 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5946 kvm_make_request(KVM_REQ_EVENT, vcpu);
5948 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5952 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5954 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5956 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5957 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5958 size, port, &val, 1);
5959 /* do not return to emulator after return from userspace */
5960 vcpu->arch.pio.count = 0;
5963 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5965 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5969 /* We should only ever be called with arch.pio.count equal to 1 */
5970 BUG_ON(vcpu->arch.pio.count != 1);
5972 /* For size less than 4 we merge, else we zero extend */
5973 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5977 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5978 * the copy and tracing
5980 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5981 vcpu->arch.pio.port, &val, 1);
5982 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5987 int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5992 /* For size less than 4 we merge, else we zero extend */
5993 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5995 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5998 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6002 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6006 EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
6008 static int kvmclock_cpu_down_prep(unsigned int cpu)
6010 __this_cpu_write(cpu_tsc_khz, 0);
6014 static void tsc_khz_changed(void *data)
6016 struct cpufreq_freqs *freq = data;
6017 unsigned long khz = 0;
6021 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6022 khz = cpufreq_quick_get(raw_smp_processor_id());
6025 __this_cpu_write(cpu_tsc_khz, khz);
6028 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6031 struct cpufreq_freqs *freq = data;
6033 struct kvm_vcpu *vcpu;
6034 int i, send_ipi = 0;
6037 * We allow guests to temporarily run on slowing clocks,
6038 * provided we notify them after, or to run on accelerating
6039 * clocks, provided we notify them before. Thus time never
6042 * However, we have a problem. We can't atomically update
6043 * the frequency of a given CPU from this function; it is
6044 * merely a notifier, which can be called from any CPU.
6045 * Changing the TSC frequency at arbitrary points in time
6046 * requires a recomputation of local variables related to
6047 * the TSC for each VCPU. We must flag these local variables
6048 * to be updated and be sure the update takes place with the
6049 * new frequency before any guests proceed.
6051 * Unfortunately, the combination of hotplug CPU and frequency
6052 * change creates an intractable locking scenario; the order
6053 * of when these callouts happen is undefined with respect to
6054 * CPU hotplug, and they can race with each other. As such,
6055 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6056 * undefined; you can actually have a CPU frequency change take
6057 * place in between the computation of X and the setting of the
6058 * variable. To protect against this problem, all updates of
6059 * the per_cpu tsc_khz variable are done in an interrupt
6060 * protected IPI, and all callers wishing to update the value
6061 * must wait for a synchronous IPI to complete (which is trivial
6062 * if the caller is on the CPU already). This establishes the
6063 * necessary total order on variable updates.
6065 * Note that because a guest time update may take place
6066 * anytime after the setting of the VCPU's request bit, the
6067 * correct TSC value must be set before the request. However,
6068 * to ensure the update actually makes it to any guest which
6069 * starts running in hardware virtualization between the set
6070 * and the acquisition of the spinlock, we must also ping the
6071 * CPU after setting the request bit.
6075 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6077 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6080 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6082 spin_lock(&kvm_lock);
6083 list_for_each_entry(kvm, &vm_list, vm_list) {
6084 kvm_for_each_vcpu(i, vcpu, kvm) {
6085 if (vcpu->cpu != freq->cpu)
6087 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6088 if (vcpu->cpu != smp_processor_id())
6092 spin_unlock(&kvm_lock);
6094 if (freq->old < freq->new && send_ipi) {
6096 * We upscale the frequency. Must make the guest
6097 * doesn't see old kvmclock values while running with
6098 * the new frequency, otherwise we risk the guest sees
6099 * time go backwards.
6101 * In case we update the frequency for another cpu
6102 * (which might be in guest context) send an interrupt
6103 * to kick the cpu out of guest context. Next time
6104 * guest context is entered kvmclock will be updated,
6105 * so the guest will not see stale values.
6107 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6112 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6113 .notifier_call = kvmclock_cpufreq_notifier
6116 static int kvmclock_cpu_online(unsigned int cpu)
6118 tsc_khz_changed(NULL);
6122 static void kvm_timer_init(void)
6124 max_tsc_khz = tsc_khz;
6126 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6127 #ifdef CONFIG_CPU_FREQ
6128 struct cpufreq_policy policy;
6131 memset(&policy, 0, sizeof(policy));
6133 cpufreq_get_policy(&policy, cpu);
6134 if (policy.cpuinfo.max_freq)
6135 max_tsc_khz = policy.cpuinfo.max_freq;
6138 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6139 CPUFREQ_TRANSITION_NOTIFIER);
6141 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6143 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6144 kvmclock_cpu_online, kvmclock_cpu_down_prep);
6147 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6149 int kvm_is_in_guest(void)
6151 return __this_cpu_read(current_vcpu) != NULL;
6154 static int kvm_is_user_mode(void)
6158 if (__this_cpu_read(current_vcpu))
6159 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6161 return user_mode != 0;
6164 static unsigned long kvm_get_guest_ip(void)
6166 unsigned long ip = 0;
6168 if (__this_cpu_read(current_vcpu))
6169 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6174 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6175 .is_in_guest = kvm_is_in_guest,
6176 .is_user_mode = kvm_is_user_mode,
6177 .get_guest_ip = kvm_get_guest_ip,
6180 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
6182 __this_cpu_write(current_vcpu, vcpu);
6184 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
6186 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
6188 __this_cpu_write(current_vcpu, NULL);
6190 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
6192 static void kvm_set_mmio_spte_mask(void)
6195 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6198 * Set the reserved bits and the present bit of an paging-structure
6199 * entry to generate page fault with PFER.RSV = 1.
6203 * Mask the uppermost physical address bit, which would be reserved as
6204 * long as the supported physical address width is less than 52.
6208 /* Set the present bit. */
6212 * If reserved bit is not supported, clear the present bit to disable
6215 if (IS_ENABLED(CONFIG_X86_64) && maxphyaddr == 52)
6218 kvm_mmu_set_mmio_spte_mask(mask, mask);
6221 #ifdef CONFIG_X86_64
6222 static void pvclock_gtod_update_fn(struct work_struct *work)
6226 struct kvm_vcpu *vcpu;
6229 spin_lock(&kvm_lock);
6230 list_for_each_entry(kvm, &vm_list, vm_list)
6231 kvm_for_each_vcpu(i, vcpu, kvm)
6232 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6233 atomic_set(&kvm_guest_has_master_clock, 0);
6234 spin_unlock(&kvm_lock);
6237 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6240 * Notification about pvclock gtod data update.
6242 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6245 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6246 struct timekeeper *tk = priv;
6248 update_pvclock_gtod(tk);
6250 /* disable master clock if host does not trust, or does not
6251 * use, TSC clocksource
6253 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6254 atomic_read(&kvm_guest_has_master_clock) != 0)
6255 queue_work(system_long_wq, &pvclock_gtod_work);
6260 static struct notifier_block pvclock_gtod_notifier = {
6261 .notifier_call = pvclock_gtod_notify,
6265 int kvm_arch_init(void *opaque)
6268 struct kvm_x86_ops *ops = opaque;
6271 printk(KERN_ERR "kvm: already loaded the other module\n");
6276 if (!ops->cpu_has_kvm_support()) {
6277 printk(KERN_ERR "kvm: no hardware support\n");
6281 if (ops->disabled_by_bios()) {
6282 printk(KERN_ERR "kvm: disabled by bios\n");
6288 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6290 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6294 r = kvm_mmu_module_init();
6296 goto out_free_percpu;
6298 kvm_set_mmio_spte_mask();
6302 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6303 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6304 PT_PRESENT_MASK, 0, sme_me_mask);
6307 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6309 if (boot_cpu_has(X86_FEATURE_XSAVE))
6310 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6313 #ifdef CONFIG_X86_64
6314 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6320 free_percpu(shared_msrs);
6325 void kvm_arch_exit(void)
6328 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6330 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6331 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6332 CPUFREQ_TRANSITION_NOTIFIER);
6333 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6334 #ifdef CONFIG_X86_64
6335 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6338 kvm_mmu_module_exit();
6339 free_percpu(shared_msrs);
6342 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6344 ++vcpu->stat.halt_exits;
6345 if (lapic_in_kernel(vcpu)) {
6346 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6349 vcpu->run->exit_reason = KVM_EXIT_HLT;
6353 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6355 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6357 int ret = kvm_skip_emulated_instruction(vcpu);
6359 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6360 * KVM_EXIT_DEBUG here.
6362 return kvm_vcpu_halt(vcpu) && ret;
6364 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6366 #ifdef CONFIG_X86_64
6367 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6368 unsigned long clock_type)
6370 struct kvm_clock_pairing clock_pairing;
6375 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6376 return -KVM_EOPNOTSUPP;
6378 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6379 return -KVM_EOPNOTSUPP;
6381 clock_pairing.sec = ts.tv_sec;
6382 clock_pairing.nsec = ts.tv_nsec;
6383 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6384 clock_pairing.flags = 0;
6385 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
6388 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6389 sizeof(struct kvm_clock_pairing)))
6397 * kvm_pv_kick_cpu_op: Kick a vcpu.
6399 * @apicid - apicid of vcpu to be kicked.
6401 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6403 struct kvm_lapic_irq lapic_irq;
6405 lapic_irq.shorthand = 0;
6406 lapic_irq.dest_mode = 0;
6407 lapic_irq.level = 0;
6408 lapic_irq.dest_id = apicid;
6409 lapic_irq.msi_redir_hint = false;
6411 lapic_irq.delivery_mode = APIC_DM_REMRD;
6412 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6415 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6417 vcpu->arch.apicv_active = false;
6418 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6421 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6423 unsigned long nr, a0, a1, a2, a3, ret;
6426 if (kvm_hv_hypercall_enabled(vcpu->kvm)) {
6427 if (!kvm_hv_hypercall(vcpu))
6432 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6433 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6434 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6435 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6436 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6438 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6440 op_64_bit = is_64_bit_mode(vcpu);
6449 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6455 case KVM_HC_VAPIC_POLL_IRQ:
6458 case KVM_HC_KICK_CPU:
6459 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6462 #ifdef CONFIG_X86_64
6463 case KVM_HC_CLOCK_PAIRING:
6464 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6474 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6477 ++vcpu->stat.hypercalls;
6478 return kvm_skip_emulated_instruction(vcpu);
6480 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6482 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6484 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6485 char instruction[3];
6486 unsigned long rip = kvm_rip_read(vcpu);
6488 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6490 return emulator_write_emulated(ctxt, rip, instruction, 3,
6494 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6496 return vcpu->run->request_interrupt_window &&
6497 likely(!pic_in_kernel(vcpu->kvm));
6500 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6502 struct kvm_run *kvm_run = vcpu->run;
6504 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6505 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6506 kvm_run->cr8 = kvm_get_cr8(vcpu);
6507 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6508 kvm_run->ready_for_interrupt_injection =
6509 pic_in_kernel(vcpu->kvm) ||
6510 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6513 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6517 if (!kvm_x86_ops->update_cr8_intercept)
6520 if (!lapic_in_kernel(vcpu))
6523 if (vcpu->arch.apicv_active)
6526 if (!vcpu->arch.apic->vapic_addr)
6527 max_irr = kvm_lapic_find_highest_irr(vcpu);
6534 tpr = kvm_lapic_get_cr8(vcpu);
6536 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6539 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6543 /* try to reinject previous events if any */
6544 if (vcpu->arch.exception.injected) {
6545 kvm_x86_ops->queue_exception(vcpu);
6550 * Exceptions must be injected immediately, or the exception
6551 * frame will have the address of the NMI or interrupt handler.
6553 if (!vcpu->arch.exception.pending) {
6554 if (vcpu->arch.nmi_injected) {
6555 kvm_x86_ops->set_nmi(vcpu);
6559 if (vcpu->arch.interrupt.pending) {
6560 kvm_x86_ops->set_irq(vcpu);
6565 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6566 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6571 /* try to inject new event if pending */
6572 if (vcpu->arch.exception.pending) {
6573 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6574 vcpu->arch.exception.has_error_code,
6575 vcpu->arch.exception.error_code);
6577 vcpu->arch.exception.pending = false;
6578 vcpu->arch.exception.injected = true;
6580 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6581 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6584 if (vcpu->arch.exception.nr == DB_VECTOR &&
6585 (vcpu->arch.dr7 & DR7_GD)) {
6586 vcpu->arch.dr7 &= ~DR7_GD;
6587 kvm_update_dr7(vcpu);
6590 kvm_x86_ops->queue_exception(vcpu);
6591 } else if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6592 vcpu->arch.smi_pending = false;
6594 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6595 --vcpu->arch.nmi_pending;
6596 vcpu->arch.nmi_injected = true;
6597 kvm_x86_ops->set_nmi(vcpu);
6598 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6600 * Because interrupts can be injected asynchronously, we are
6601 * calling check_nested_events again here to avoid a race condition.
6602 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6603 * proposal and current concerns. Perhaps we should be setting
6604 * KVM_REQ_EVENT only on certain events and not unconditionally?
6606 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6607 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6611 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6612 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6614 kvm_x86_ops->set_irq(vcpu);
6621 static void process_nmi(struct kvm_vcpu *vcpu)
6626 * x86 is limited to one NMI running, and one NMI pending after it.
6627 * If an NMI is already in progress, limit further NMIs to just one.
6628 * Otherwise, allow two (and we'll inject the first one immediately).
6630 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6633 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6634 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6635 kvm_make_request(KVM_REQ_EVENT, vcpu);
6638 #define put_smstate(type, buf, offset, val) \
6639 *(type *)((buf) + (offset) - 0x7e00) = val
6641 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6644 flags |= seg->g << 23;
6645 flags |= seg->db << 22;
6646 flags |= seg->l << 21;
6647 flags |= seg->avl << 20;
6648 flags |= seg->present << 15;
6649 flags |= seg->dpl << 13;
6650 flags |= seg->s << 12;
6651 flags |= seg->type << 8;
6655 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6657 struct kvm_segment seg;
6660 kvm_get_segment(vcpu, &seg, n);
6661 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6664 offset = 0x7f84 + n * 12;
6666 offset = 0x7f2c + (n - 3) * 12;
6668 put_smstate(u32, buf, offset + 8, seg.base);
6669 put_smstate(u32, buf, offset + 4, seg.limit);
6670 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6673 #ifdef CONFIG_X86_64
6674 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6676 struct kvm_segment seg;
6680 kvm_get_segment(vcpu, &seg, n);
6681 offset = 0x7e00 + n * 16;
6683 flags = enter_smm_get_segment_flags(&seg) >> 8;
6684 put_smstate(u16, buf, offset, seg.selector);
6685 put_smstate(u16, buf, offset + 2, flags);
6686 put_smstate(u32, buf, offset + 4, seg.limit);
6687 put_smstate(u64, buf, offset + 8, seg.base);
6691 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6694 struct kvm_segment seg;
6698 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6699 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6700 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6701 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6703 for (i = 0; i < 8; i++)
6704 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6706 kvm_get_dr(vcpu, 6, &val);
6707 put_smstate(u32, buf, 0x7fcc, (u32)val);
6708 kvm_get_dr(vcpu, 7, &val);
6709 put_smstate(u32, buf, 0x7fc8, (u32)val);
6711 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6712 put_smstate(u32, buf, 0x7fc4, seg.selector);
6713 put_smstate(u32, buf, 0x7f64, seg.base);
6714 put_smstate(u32, buf, 0x7f60, seg.limit);
6715 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6717 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6718 put_smstate(u32, buf, 0x7fc0, seg.selector);
6719 put_smstate(u32, buf, 0x7f80, seg.base);
6720 put_smstate(u32, buf, 0x7f7c, seg.limit);
6721 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6723 kvm_x86_ops->get_gdt(vcpu, &dt);
6724 put_smstate(u32, buf, 0x7f74, dt.address);
6725 put_smstate(u32, buf, 0x7f70, dt.size);
6727 kvm_x86_ops->get_idt(vcpu, &dt);
6728 put_smstate(u32, buf, 0x7f58, dt.address);
6729 put_smstate(u32, buf, 0x7f54, dt.size);
6731 for (i = 0; i < 6; i++)
6732 enter_smm_save_seg_32(vcpu, buf, i);
6734 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6737 put_smstate(u32, buf, 0x7efc, 0x00020000);
6738 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6741 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6743 #ifdef CONFIG_X86_64
6745 struct kvm_segment seg;
6749 for (i = 0; i < 16; i++)
6750 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6752 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6753 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6755 kvm_get_dr(vcpu, 6, &val);
6756 put_smstate(u64, buf, 0x7f68, val);
6757 kvm_get_dr(vcpu, 7, &val);
6758 put_smstate(u64, buf, 0x7f60, val);
6760 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6761 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6762 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6764 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6767 put_smstate(u32, buf, 0x7efc, 0x00020064);
6769 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6771 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6772 put_smstate(u16, buf, 0x7e90, seg.selector);
6773 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6774 put_smstate(u32, buf, 0x7e94, seg.limit);
6775 put_smstate(u64, buf, 0x7e98, seg.base);
6777 kvm_x86_ops->get_idt(vcpu, &dt);
6778 put_smstate(u32, buf, 0x7e84, dt.size);
6779 put_smstate(u64, buf, 0x7e88, dt.address);
6781 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6782 put_smstate(u16, buf, 0x7e70, seg.selector);
6783 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6784 put_smstate(u32, buf, 0x7e74, seg.limit);
6785 put_smstate(u64, buf, 0x7e78, seg.base);
6787 kvm_x86_ops->get_gdt(vcpu, &dt);
6788 put_smstate(u32, buf, 0x7e64, dt.size);
6789 put_smstate(u64, buf, 0x7e68, dt.address);
6791 for (i = 0; i < 6; i++)
6792 enter_smm_save_seg_64(vcpu, buf, i);
6798 static void enter_smm(struct kvm_vcpu *vcpu)
6800 struct kvm_segment cs, ds;
6805 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6806 vcpu->arch.hflags |= HF_SMM_MASK;
6807 memset(buf, 0, 512);
6808 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
6809 enter_smm_save_state_64(vcpu, buf);
6811 enter_smm_save_state_32(vcpu, buf);
6813 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6815 if (kvm_x86_ops->get_nmi_mask(vcpu))
6816 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6818 kvm_x86_ops->set_nmi_mask(vcpu, true);
6820 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6821 kvm_rip_write(vcpu, 0x8000);
6823 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6824 kvm_x86_ops->set_cr0(vcpu, cr0);
6825 vcpu->arch.cr0 = cr0;
6827 kvm_x86_ops->set_cr4(vcpu, 0);
6829 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6830 dt.address = dt.size = 0;
6831 kvm_x86_ops->set_idt(vcpu, &dt);
6833 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6835 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6836 cs.base = vcpu->arch.smbase;
6841 cs.limit = ds.limit = 0xffffffff;
6842 cs.type = ds.type = 0x3;
6843 cs.dpl = ds.dpl = 0;
6848 cs.avl = ds.avl = 0;
6849 cs.present = ds.present = 1;
6850 cs.unusable = ds.unusable = 0;
6851 cs.padding = ds.padding = 0;
6853 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6854 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6855 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6856 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6857 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6858 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6860 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
6861 kvm_x86_ops->set_efer(vcpu, 0);
6863 kvm_update_cpuid(vcpu);
6864 kvm_mmu_reset_context(vcpu);
6867 static void process_smi(struct kvm_vcpu *vcpu)
6869 vcpu->arch.smi_pending = true;
6870 kvm_make_request(KVM_REQ_EVENT, vcpu);
6873 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6875 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6878 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6880 u64 eoi_exit_bitmap[4];
6882 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6885 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6887 if (irqchip_split(vcpu->kvm))
6888 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6890 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6891 kvm_x86_ops->sync_pir_to_irr(vcpu);
6892 if (ioapic_in_kernel(vcpu->kvm))
6893 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6895 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6896 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6897 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6900 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6902 ++vcpu->stat.tlb_flush;
6903 kvm_x86_ops->tlb_flush(vcpu);
6906 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
6907 unsigned long start, unsigned long end)
6909 unsigned long apic_address;
6912 * The physical address of apic access page is stored in the VMCS.
6913 * Update it when it becomes invalid.
6915 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6916 if (start <= apic_address && apic_address < end)
6917 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6920 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6922 struct page *page = NULL;
6924 if (!lapic_in_kernel(vcpu))
6927 if (!kvm_x86_ops->set_apic_access_page_addr)
6930 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6931 if (is_error_page(page))
6933 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6936 * Do not pin apic access page in memory, the MMU notifier
6937 * will call us again if it is migrated or swapped out.
6941 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6944 * Returns 1 to let vcpu_run() continue the guest execution loop without
6945 * exiting to the userspace. Otherwise, the value will be returned to the
6948 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6952 dm_request_for_irq_injection(vcpu) &&
6953 kvm_cpu_accept_dm_intr(vcpu);
6955 bool req_immediate_exit = false;
6957 if (kvm_request_pending(vcpu)) {
6958 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6959 kvm_mmu_unload(vcpu);
6960 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6961 __kvm_migrate_timers(vcpu);
6962 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6963 kvm_gen_update_masterclock(vcpu->kvm);
6964 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6965 kvm_gen_kvmclock_update(vcpu);
6966 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6967 r = kvm_guest_time_update(vcpu);
6971 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6972 kvm_mmu_sync_roots(vcpu);
6973 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6974 kvm_vcpu_flush_tlb(vcpu);
6975 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6976 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6980 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6981 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6982 vcpu->mmio_needed = 0;
6986 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6987 /* Page is swapped out. Do synthetic halt */
6988 vcpu->arch.apf.halted = true;
6992 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6993 record_steal_time(vcpu);
6994 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6996 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6998 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6999 kvm_pmu_handle_event(vcpu);
7000 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7001 kvm_pmu_deliver_pmi(vcpu);
7002 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7003 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7004 if (test_bit(vcpu->arch.pending_ioapic_eoi,
7005 vcpu->arch.ioapic_handled_vectors)) {
7006 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7007 vcpu->run->eoi.vector =
7008 vcpu->arch.pending_ioapic_eoi;
7013 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7014 vcpu_scan_ioapic(vcpu);
7015 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7016 kvm_vcpu_reload_apic_access_page(vcpu);
7017 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7018 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7019 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7023 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7024 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7025 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7029 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7030 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7031 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7037 * KVM_REQ_HV_STIMER has to be processed after
7038 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7039 * depend on the guest clock being up-to-date
7041 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7042 kvm_hv_process_stimers(vcpu);
7045 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7046 ++vcpu->stat.req_event;
7047 kvm_apic_accept_events(vcpu);
7048 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7053 if (inject_pending_event(vcpu, req_int_win) != 0)
7054 req_immediate_exit = true;
7056 /* Enable NMI/IRQ window open exits if needed.
7058 * SMIs have two cases: 1) they can be nested, and
7059 * then there is nothing to do here because RSM will
7060 * cause a vmexit anyway; 2) or the SMI can be pending
7061 * because inject_pending_event has completed the
7062 * injection of an IRQ or NMI from the previous vmexit,
7063 * and then we request an immediate exit to inject the SMI.
7065 if (vcpu->arch.smi_pending && !is_smm(vcpu))
7066 req_immediate_exit = true;
7067 if (vcpu->arch.nmi_pending)
7068 kvm_x86_ops->enable_nmi_window(vcpu);
7069 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7070 kvm_x86_ops->enable_irq_window(vcpu);
7071 WARN_ON(vcpu->arch.exception.pending);
7074 if (kvm_lapic_enabled(vcpu)) {
7075 update_cr8_intercept(vcpu);
7076 kvm_lapic_sync_to_vapic(vcpu);
7080 r = kvm_mmu_reload(vcpu);
7082 goto cancel_injection;
7087 kvm_x86_ops->prepare_guest_switch(vcpu);
7090 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7091 * IPI are then delayed after guest entry, which ensures that they
7092 * result in virtual interrupt delivery.
7094 local_irq_disable();
7095 vcpu->mode = IN_GUEST_MODE;
7097 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7100 * 1) We should set ->mode before checking ->requests. Please see
7101 * the comment in kvm_vcpu_exiting_guest_mode().
7103 * 2) For APICv, we should set ->mode before checking PIR.ON. This
7104 * pairs with the memory barrier implicit in pi_test_and_set_on
7105 * (see vmx_deliver_posted_interrupt).
7107 * 3) This also orders the write to mode from any reads to the page
7108 * tables done while the VCPU is running. Please see the comment
7109 * in kvm_flush_remote_tlbs.
7111 smp_mb__after_srcu_read_unlock();
7114 * This handles the case where a posted interrupt was
7115 * notified with kvm_vcpu_kick.
7117 if (kvm_lapic_enabled(vcpu)) {
7118 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
7119 kvm_x86_ops->sync_pir_to_irr(vcpu);
7122 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7123 || need_resched() || signal_pending(current)) {
7124 vcpu->mode = OUTSIDE_GUEST_MODE;
7128 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7130 goto cancel_injection;
7133 kvm_load_guest_xcr0(vcpu);
7135 if (req_immediate_exit) {
7136 kvm_make_request(KVM_REQ_EVENT, vcpu);
7137 smp_send_reschedule(vcpu->cpu);
7140 trace_kvm_entry(vcpu->vcpu_id);
7141 wait_lapic_expire(vcpu);
7142 guest_enter_irqoff();
7144 if (unlikely(vcpu->arch.switch_db_regs)) {
7146 set_debugreg(vcpu->arch.eff_db[0], 0);
7147 set_debugreg(vcpu->arch.eff_db[1], 1);
7148 set_debugreg(vcpu->arch.eff_db[2], 2);
7149 set_debugreg(vcpu->arch.eff_db[3], 3);
7150 set_debugreg(vcpu->arch.dr6, 6);
7151 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7154 kvm_x86_ops->run(vcpu);
7157 * Do this here before restoring debug registers on the host. And
7158 * since we do this before handling the vmexit, a DR access vmexit
7159 * can (a) read the correct value of the debug registers, (b) set
7160 * KVM_DEBUGREG_WONT_EXIT again.
7162 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7163 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7164 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7165 kvm_update_dr0123(vcpu);
7166 kvm_update_dr6(vcpu);
7167 kvm_update_dr7(vcpu);
7168 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7172 * If the guest has used debug registers, at least dr7
7173 * will be disabled while returning to the host.
7174 * If we don't have active breakpoints in the host, we don't
7175 * care about the messed up debug address registers. But if
7176 * we have some of them active, restore the old state.
7178 if (hw_breakpoint_active())
7179 hw_breakpoint_restore();
7181 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7183 vcpu->mode = OUTSIDE_GUEST_MODE;
7186 kvm_put_guest_xcr0(vcpu);
7188 kvm_x86_ops->handle_external_intr(vcpu);
7192 guest_exit_irqoff();
7197 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7200 * Profile KVM exit RIPs:
7202 if (unlikely(prof_on == KVM_PROFILING)) {
7203 unsigned long rip = kvm_rip_read(vcpu);
7204 profile_hit(KVM_PROFILING, (void *)rip);
7207 if (unlikely(vcpu->arch.tsc_always_catchup))
7208 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7210 if (vcpu->arch.apic_attention)
7211 kvm_lapic_sync_from_vapic(vcpu);
7213 vcpu->arch.gpa_available = false;
7214 r = kvm_x86_ops->handle_exit(vcpu);
7218 kvm_x86_ops->cancel_injection(vcpu);
7219 if (unlikely(vcpu->arch.apic_attention))
7220 kvm_lapic_sync_from_vapic(vcpu);
7225 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7227 if (!kvm_arch_vcpu_runnable(vcpu) &&
7228 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7229 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7230 kvm_vcpu_block(vcpu);
7231 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7233 if (kvm_x86_ops->post_block)
7234 kvm_x86_ops->post_block(vcpu);
7236 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7240 kvm_apic_accept_events(vcpu);
7241 switch(vcpu->arch.mp_state) {
7242 case KVM_MP_STATE_HALTED:
7243 vcpu->arch.pv.pv_unhalted = false;
7244 vcpu->arch.mp_state =
7245 KVM_MP_STATE_RUNNABLE;
7246 case KVM_MP_STATE_RUNNABLE:
7247 vcpu->arch.apf.halted = false;
7249 case KVM_MP_STATE_INIT_RECEIVED:
7258 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7260 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7261 kvm_x86_ops->check_nested_events(vcpu, false);
7263 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7264 !vcpu->arch.apf.halted);
7267 static int vcpu_run(struct kvm_vcpu *vcpu)
7270 struct kvm *kvm = vcpu->kvm;
7272 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7273 vcpu->arch.l1tf_flush_l1d = true;
7276 if (kvm_vcpu_running(vcpu)) {
7277 r = vcpu_enter_guest(vcpu);
7279 r = vcpu_block(kvm, vcpu);
7285 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7286 if (kvm_cpu_has_pending_timer(vcpu))
7287 kvm_inject_pending_timer_irqs(vcpu);
7289 if (dm_request_for_irq_injection(vcpu) &&
7290 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7292 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7293 ++vcpu->stat.request_irq_exits;
7297 kvm_check_async_pf_completion(vcpu);
7299 if (signal_pending(current)) {
7301 vcpu->run->exit_reason = KVM_EXIT_INTR;
7302 ++vcpu->stat.signal_exits;
7305 if (need_resched()) {
7306 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7308 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7312 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7317 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7320 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7321 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7322 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7323 if (r != EMULATE_DONE)
7328 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7330 BUG_ON(!vcpu->arch.pio.count);
7332 return complete_emulated_io(vcpu);
7336 * Implements the following, as a state machine:
7340 * for each mmio piece in the fragment
7348 * for each mmio piece in the fragment
7353 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7355 struct kvm_run *run = vcpu->run;
7356 struct kvm_mmio_fragment *frag;
7359 BUG_ON(!vcpu->mmio_needed);
7361 /* Complete previous fragment */
7362 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7363 len = min(8u, frag->len);
7364 if (!vcpu->mmio_is_write)
7365 memcpy(frag->data, run->mmio.data, len);
7367 if (frag->len <= 8) {
7368 /* Switch to the next fragment. */
7370 vcpu->mmio_cur_fragment++;
7372 /* Go forward to the next mmio piece. */
7378 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7379 vcpu->mmio_needed = 0;
7381 /* FIXME: return into emulator if single-stepping. */
7382 if (vcpu->mmio_is_write)
7384 vcpu->mmio_read_completed = 1;
7385 return complete_emulated_io(vcpu);
7388 run->exit_reason = KVM_EXIT_MMIO;
7389 run->mmio.phys_addr = frag->gpa;
7390 if (vcpu->mmio_is_write)
7391 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7392 run->mmio.len = min(8u, frag->len);
7393 run->mmio.is_write = vcpu->mmio_is_write;
7394 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7399 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7403 kvm_sigset_activate(vcpu);
7405 kvm_load_guest_fpu(vcpu);
7407 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7408 if (kvm_run->immediate_exit) {
7412 kvm_vcpu_block(vcpu);
7413 kvm_apic_accept_events(vcpu);
7414 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7416 if (signal_pending(current)) {
7418 vcpu->run->exit_reason = KVM_EXIT_INTR;
7419 ++vcpu->stat.signal_exits;
7424 /* re-sync apic's tpr */
7425 if (!lapic_in_kernel(vcpu)) {
7426 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7432 if (unlikely(vcpu->arch.complete_userspace_io)) {
7433 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7434 vcpu->arch.complete_userspace_io = NULL;
7439 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7441 if (kvm_run->immediate_exit)
7447 kvm_put_guest_fpu(vcpu);
7448 post_kvm_run_save(vcpu);
7449 kvm_sigset_deactivate(vcpu);
7454 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7456 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7458 * We are here if userspace calls get_regs() in the middle of
7459 * instruction emulation. Registers state needs to be copied
7460 * back from emulation context to vcpu. Userspace shouldn't do
7461 * that usually, but some bad designed PV devices (vmware
7462 * backdoor interface) need this to work
7464 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7465 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7467 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7468 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7469 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7470 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7471 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7472 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7473 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7474 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7475 #ifdef CONFIG_X86_64
7476 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7477 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7478 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7479 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7480 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7481 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7482 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7483 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7486 regs->rip = kvm_rip_read(vcpu);
7487 regs->rflags = kvm_get_rflags(vcpu);
7492 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7494 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7495 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7497 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7498 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7499 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7500 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7501 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7502 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7503 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7504 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7505 #ifdef CONFIG_X86_64
7506 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7507 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7508 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7509 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7510 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7511 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7512 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7513 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7516 kvm_rip_write(vcpu, regs->rip);
7517 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
7519 vcpu->arch.exception.pending = false;
7521 kvm_make_request(KVM_REQ_EVENT, vcpu);
7526 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7528 struct kvm_segment cs;
7530 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7534 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7536 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7537 struct kvm_sregs *sregs)
7541 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7542 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7543 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7544 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7545 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7546 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7548 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7549 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7551 kvm_x86_ops->get_idt(vcpu, &dt);
7552 sregs->idt.limit = dt.size;
7553 sregs->idt.base = dt.address;
7554 kvm_x86_ops->get_gdt(vcpu, &dt);
7555 sregs->gdt.limit = dt.size;
7556 sregs->gdt.base = dt.address;
7558 sregs->cr0 = kvm_read_cr0(vcpu);
7559 sregs->cr2 = vcpu->arch.cr2;
7560 sregs->cr3 = kvm_read_cr3(vcpu);
7561 sregs->cr4 = kvm_read_cr4(vcpu);
7562 sregs->cr8 = kvm_get_cr8(vcpu);
7563 sregs->efer = vcpu->arch.efer;
7564 sregs->apic_base = kvm_get_apic_base(vcpu);
7566 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7568 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7569 set_bit(vcpu->arch.interrupt.nr,
7570 (unsigned long *)sregs->interrupt_bitmap);
7575 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7576 struct kvm_mp_state *mp_state)
7578 kvm_apic_accept_events(vcpu);
7579 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7580 vcpu->arch.pv.pv_unhalted)
7581 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7583 mp_state->mp_state = vcpu->arch.mp_state;
7588 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7589 struct kvm_mp_state *mp_state)
7591 if (!lapic_in_kernel(vcpu) &&
7592 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7595 /* INITs are latched while in SMM */
7596 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7597 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7598 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7601 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7602 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7603 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7605 vcpu->arch.mp_state = mp_state->mp_state;
7606 kvm_make_request(KVM_REQ_EVENT, vcpu);
7610 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7611 int reason, bool has_error_code, u32 error_code)
7613 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7616 init_emulate_ctxt(vcpu);
7618 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7619 has_error_code, error_code);
7622 return EMULATE_FAIL;
7624 kvm_rip_write(vcpu, ctxt->eip);
7625 kvm_set_rflags(vcpu, ctxt->eflags);
7626 kvm_make_request(KVM_REQ_EVENT, vcpu);
7627 return EMULATE_DONE;
7629 EXPORT_SYMBOL_GPL(kvm_task_switch);
7631 int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7633 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
7635 * When EFER.LME and CR0.PG are set, the processor is in
7636 * 64-bit mode (though maybe in a 32-bit code segment).
7637 * CR4.PAE and EFER.LMA must be set.
7639 if (!(sregs->cr4 & X86_CR4_PAE)
7640 || !(sregs->efer & EFER_LMA))
7644 * Not in 64-bit mode: EFER.LMA is clear and the code
7645 * segment cannot be 64-bit.
7647 if (sregs->efer & EFER_LMA || sregs->cs.l)
7654 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7655 struct kvm_sregs *sregs)
7657 struct msr_data apic_base_msr;
7658 int mmu_reset_needed = 0;
7659 int cpuid_update_needed = 0;
7660 int pending_vec, max_bits, idx;
7663 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7664 (sregs->cr4 & X86_CR4_OSXSAVE))
7667 if (kvm_valid_sregs(vcpu, sregs))
7670 apic_base_msr.data = sregs->apic_base;
7671 apic_base_msr.host_initiated = true;
7672 if (kvm_set_apic_base(vcpu, &apic_base_msr))
7675 dt.size = sregs->idt.limit;
7676 dt.address = sregs->idt.base;
7677 kvm_x86_ops->set_idt(vcpu, &dt);
7678 dt.size = sregs->gdt.limit;
7679 dt.address = sregs->gdt.base;
7680 kvm_x86_ops->set_gdt(vcpu, &dt);
7682 vcpu->arch.cr2 = sregs->cr2;
7683 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7684 vcpu->arch.cr3 = sregs->cr3;
7685 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7687 kvm_set_cr8(vcpu, sregs->cr8);
7689 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7690 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7692 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7693 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7694 vcpu->arch.cr0 = sregs->cr0;
7696 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7697 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
7698 (X86_CR4_OSXSAVE | X86_CR4_PKE));
7699 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7700 if (cpuid_update_needed)
7701 kvm_update_cpuid(vcpu);
7703 idx = srcu_read_lock(&vcpu->kvm->srcu);
7704 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7705 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7706 mmu_reset_needed = 1;
7708 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7710 if (mmu_reset_needed)
7711 kvm_mmu_reset_context(vcpu);
7713 max_bits = KVM_NR_INTERRUPTS;
7714 pending_vec = find_first_bit(
7715 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7716 if (pending_vec < max_bits) {
7717 kvm_queue_interrupt(vcpu, pending_vec, false);
7718 pr_debug("Set back pending irq %d\n", pending_vec);
7721 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7722 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7723 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7724 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7725 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7726 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7728 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7729 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7731 update_cr8_intercept(vcpu);
7733 /* Older userspace won't unhalt the vcpu on reset. */
7734 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7735 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7737 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7739 kvm_make_request(KVM_REQ_EVENT, vcpu);
7744 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7745 struct kvm_guest_debug *dbg)
7747 unsigned long rflags;
7750 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7752 if (vcpu->arch.exception.pending)
7754 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7755 kvm_queue_exception(vcpu, DB_VECTOR);
7757 kvm_queue_exception(vcpu, BP_VECTOR);
7761 * Read rflags as long as potentially injected trace flags are still
7764 rflags = kvm_get_rflags(vcpu);
7766 vcpu->guest_debug = dbg->control;
7767 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7768 vcpu->guest_debug = 0;
7770 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7771 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7772 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7773 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7775 for (i = 0; i < KVM_NR_DB_REGS; i++)
7776 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7778 kvm_update_dr7(vcpu);
7780 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7781 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7782 get_segment_base(vcpu, VCPU_SREG_CS);
7785 * Trigger an rflags update that will inject or remove the trace
7788 kvm_set_rflags(vcpu, rflags);
7790 kvm_x86_ops->update_bp_intercept(vcpu);
7800 * Translate a guest virtual address to a guest physical address.
7802 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7803 struct kvm_translation *tr)
7805 unsigned long vaddr = tr->linear_address;
7809 idx = srcu_read_lock(&vcpu->kvm->srcu);
7810 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7811 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7812 tr->physical_address = gpa;
7813 tr->valid = gpa != UNMAPPED_GVA;
7820 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7822 struct fxregs_state *fxsave =
7823 &vcpu->arch.guest_fpu.state.fxsave;
7825 memcpy(fpu->fpr, fxsave->st_space, 128);
7826 fpu->fcw = fxsave->cwd;
7827 fpu->fsw = fxsave->swd;
7828 fpu->ftwx = fxsave->twd;
7829 fpu->last_opcode = fxsave->fop;
7830 fpu->last_ip = fxsave->rip;
7831 fpu->last_dp = fxsave->rdp;
7832 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7837 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7839 struct fxregs_state *fxsave =
7840 &vcpu->arch.guest_fpu.state.fxsave;
7842 memcpy(fxsave->st_space, fpu->fpr, 128);
7843 fxsave->cwd = fpu->fcw;
7844 fxsave->swd = fpu->fsw;
7845 fxsave->twd = fpu->ftwx;
7846 fxsave->fop = fpu->last_opcode;
7847 fxsave->rip = fpu->last_ip;
7848 fxsave->rdp = fpu->last_dp;
7849 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7854 static void fx_init(struct kvm_vcpu *vcpu)
7856 fpstate_init(&vcpu->arch.guest_fpu.state);
7857 if (boot_cpu_has(X86_FEATURE_XSAVES))
7858 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7859 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7862 * Ensure guest xcr0 is valid for loading
7864 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7866 vcpu->arch.cr0 |= X86_CR0_ET;
7869 /* Swap (qemu) user FPU context for the guest FPU context. */
7870 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7873 copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
7874 /* PKRU is separately restored in kvm_x86_ops->run. */
7875 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
7876 ~XFEATURE_MASK_PKRU);
7881 /* When vcpu_run ends, restore user space FPU context. */
7882 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7885 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7886 copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
7888 ++vcpu->stat.fpu_reload;
7892 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7894 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7896 kvmclock_reset(vcpu);
7898 kvm_x86_ops->vcpu_free(vcpu);
7899 free_cpumask_var(wbinvd_dirty_mask);
7902 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7905 struct kvm_vcpu *vcpu;
7907 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7908 printk_once(KERN_WARNING
7909 "kvm: SMP vm created on host with unstable TSC; "
7910 "guest TSC will not be reliable\n");
7912 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7917 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7921 kvm_vcpu_mtrr_init(vcpu);
7922 r = vcpu_load(vcpu);
7925 kvm_vcpu_reset(vcpu, false);
7926 kvm_mmu_setup(vcpu);
7931 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7933 struct msr_data msr;
7934 struct kvm *kvm = vcpu->kvm;
7936 kvm_hv_vcpu_postcreate(vcpu);
7938 if (vcpu_load(vcpu))
7941 msr.index = MSR_IA32_TSC;
7942 msr.host_initiated = true;
7943 kvm_write_tsc(vcpu, &msr);
7946 if (!kvmclock_periodic_sync)
7949 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7950 KVMCLOCK_SYNC_PERIOD);
7953 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7956 vcpu->arch.apf.msr_val = 0;
7958 r = vcpu_load(vcpu);
7960 kvm_mmu_unload(vcpu);
7963 kvm_x86_ops->vcpu_free(vcpu);
7966 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7968 kvm_lapic_reset(vcpu, init_event);
7970 vcpu->arch.hflags = 0;
7972 vcpu->arch.smi_pending = 0;
7973 atomic_set(&vcpu->arch.nmi_queued, 0);
7974 vcpu->arch.nmi_pending = 0;
7975 vcpu->arch.nmi_injected = false;
7976 kvm_clear_interrupt_queue(vcpu);
7977 kvm_clear_exception_queue(vcpu);
7978 vcpu->arch.exception.pending = false;
7980 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7981 kvm_update_dr0123(vcpu);
7982 vcpu->arch.dr6 = DR6_INIT;
7983 kvm_update_dr6(vcpu);
7984 vcpu->arch.dr7 = DR7_FIXED_1;
7985 kvm_update_dr7(vcpu);
7989 kvm_make_request(KVM_REQ_EVENT, vcpu);
7990 vcpu->arch.apf.msr_val = 0;
7991 vcpu->arch.st.msr_val = 0;
7993 kvmclock_reset(vcpu);
7995 kvm_clear_async_pf_completion_queue(vcpu);
7996 kvm_async_pf_hash_reset(vcpu);
7997 vcpu->arch.apf.halted = false;
8000 kvm_pmu_reset(vcpu);
8001 vcpu->arch.smbase = 0x30000;
8003 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8004 vcpu->arch.msr_misc_features_enables = 0;
8007 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8008 vcpu->arch.regs_avail = ~0;
8009 vcpu->arch.regs_dirty = ~0;
8011 kvm_x86_ops->vcpu_reset(vcpu, init_event);
8014 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8016 struct kvm_segment cs;
8018 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8019 cs.selector = vector << 8;
8020 cs.base = vector << 12;
8021 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8022 kvm_rip_write(vcpu, 0);
8025 int kvm_arch_hardware_enable(void)
8028 struct kvm_vcpu *vcpu;
8033 bool stable, backwards_tsc = false;
8035 kvm_shared_msr_cpu_online();
8036 ret = kvm_x86_ops->hardware_enable();
8040 local_tsc = rdtsc();
8041 stable = !check_tsc_unstable();
8042 list_for_each_entry(kvm, &vm_list, vm_list) {
8043 kvm_for_each_vcpu(i, vcpu, kvm) {
8044 if (!stable && vcpu->cpu == smp_processor_id())
8045 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8046 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8047 backwards_tsc = true;
8048 if (vcpu->arch.last_host_tsc > max_tsc)
8049 max_tsc = vcpu->arch.last_host_tsc;
8055 * Sometimes, even reliable TSCs go backwards. This happens on
8056 * platforms that reset TSC during suspend or hibernate actions, but
8057 * maintain synchronization. We must compensate. Fortunately, we can
8058 * detect that condition here, which happens early in CPU bringup,
8059 * before any KVM threads can be running. Unfortunately, we can't
8060 * bring the TSCs fully up to date with real time, as we aren't yet far
8061 * enough into CPU bringup that we know how much real time has actually
8062 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8063 * variables that haven't been updated yet.
8065 * So we simply find the maximum observed TSC above, then record the
8066 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
8067 * the adjustment will be applied. Note that we accumulate
8068 * adjustments, in case multiple suspend cycles happen before some VCPU
8069 * gets a chance to run again. In the event that no KVM threads get a
8070 * chance to run, we will miss the entire elapsed period, as we'll have
8071 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8072 * loose cycle time. This isn't too big a deal, since the loss will be
8073 * uniform across all VCPUs (not to mention the scenario is extremely
8074 * unlikely). It is possible that a second hibernate recovery happens
8075 * much faster than a first, causing the observed TSC here to be
8076 * smaller; this would require additional padding adjustment, which is
8077 * why we set last_host_tsc to the local tsc observed here.
8079 * N.B. - this code below runs only on platforms with reliable TSC,
8080 * as that is the only way backwards_tsc is set above. Also note
8081 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8082 * have the same delta_cyc adjustment applied if backwards_tsc
8083 * is detected. Note further, this adjustment is only done once,
8084 * as we reset last_host_tsc on all VCPUs to stop this from being
8085 * called multiple times (one for each physical CPU bringup).
8087 * Platforms with unreliable TSCs don't have to deal with this, they
8088 * will be compensated by the logic in vcpu_load, which sets the TSC to
8089 * catchup mode. This will catchup all VCPUs to real time, but cannot
8090 * guarantee that they stay in perfect synchronization.
8092 if (backwards_tsc) {
8093 u64 delta_cyc = max_tsc - local_tsc;
8094 list_for_each_entry(kvm, &vm_list, vm_list) {
8095 kvm->arch.backwards_tsc_observed = true;
8096 kvm_for_each_vcpu(i, vcpu, kvm) {
8097 vcpu->arch.tsc_offset_adjustment += delta_cyc;
8098 vcpu->arch.last_host_tsc = local_tsc;
8099 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8103 * We have to disable TSC offset matching.. if you were
8104 * booting a VM while issuing an S4 host suspend....
8105 * you may have some problem. Solving this issue is
8106 * left as an exercise to the reader.
8108 kvm->arch.last_tsc_nsec = 0;
8109 kvm->arch.last_tsc_write = 0;
8116 void kvm_arch_hardware_disable(void)
8118 kvm_x86_ops->hardware_disable();
8119 drop_user_return_notifiers();
8122 int kvm_arch_hardware_setup(void)
8126 r = kvm_x86_ops->hardware_setup();
8130 if (kvm_has_tsc_control) {
8132 * Make sure the user can only configure tsc_khz values that
8133 * fit into a signed integer.
8134 * A min value is not calculated needed because it will always
8135 * be 1 on all machines.
8137 u64 max = min(0x7fffffffULL,
8138 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8139 kvm_max_guest_tsc_khz = max;
8141 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8144 kvm_init_msr_list();
8148 void kvm_arch_hardware_unsetup(void)
8150 kvm_x86_ops->hardware_unsetup();
8153 void kvm_arch_check_processor_compat(void *rtn)
8155 kvm_x86_ops->check_processor_compatibility(rtn);
8158 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8160 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8162 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8164 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8166 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8169 struct static_key kvm_no_apic_vcpu __read_mostly;
8170 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8172 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8178 BUG_ON(vcpu->kvm == NULL);
8181 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8182 vcpu->arch.pv.pv_unhalted = false;
8183 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8184 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8185 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8187 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8189 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8194 vcpu->arch.pio_data = page_address(page);
8196 kvm_set_tsc_khz(vcpu, max_tsc_khz);
8198 r = kvm_mmu_create(vcpu);
8200 goto fail_free_pio_data;
8202 if (irqchip_in_kernel(kvm)) {
8203 r = kvm_create_lapic(vcpu);
8205 goto fail_mmu_destroy;
8207 static_key_slow_inc(&kvm_no_apic_vcpu);
8209 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8211 if (!vcpu->arch.mce_banks) {
8213 goto fail_free_lapic;
8215 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8217 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8219 goto fail_free_mce_banks;
8224 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
8225 vcpu->arch.pv_time_enabled = false;
8227 vcpu->arch.guest_supported_xcr0 = 0;
8228 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8230 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8232 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8234 kvm_async_pf_hash_reset(vcpu);
8237 vcpu->arch.pending_external_vector = -1;
8238 vcpu->arch.preempted_in_kernel = false;
8240 kvm_hv_vcpu_init(vcpu);
8244 fail_free_mce_banks:
8245 kfree(vcpu->arch.mce_banks);
8247 kvm_free_lapic(vcpu);
8249 kvm_mmu_destroy(vcpu);
8251 free_page((unsigned long)vcpu->arch.pio_data);
8256 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8260 kvm_hv_vcpu_uninit(vcpu);
8261 kvm_pmu_destroy(vcpu);
8262 kfree(vcpu->arch.mce_banks);
8263 kvm_free_lapic(vcpu);
8264 idx = srcu_read_lock(&vcpu->kvm->srcu);
8265 kvm_mmu_destroy(vcpu);
8266 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8267 free_page((unsigned long)vcpu->arch.pio_data);
8268 if (!lapic_in_kernel(vcpu))
8269 static_key_slow_dec(&kvm_no_apic_vcpu);
8272 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8274 vcpu->arch.l1tf_flush_l1d = true;
8275 kvm_x86_ops->sched_in(vcpu, cpu);
8278 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8283 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8284 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8285 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8286 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8287 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8289 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8290 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8291 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8292 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8293 &kvm->arch.irq_sources_bitmap);
8295 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8296 mutex_init(&kvm->arch.apic_map_lock);
8297 mutex_init(&kvm->arch.hyperv.hv_lock);
8298 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8300 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8301 pvclock_update_vm_gtod_copy(kvm);
8303 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8304 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8306 kvm_page_track_init(kvm);
8307 kvm_mmu_init_vm(kvm);
8309 if (kvm_x86_ops->vm_init)
8310 return kvm_x86_ops->vm_init(kvm);
8315 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8318 r = vcpu_load(vcpu);
8320 kvm_mmu_unload(vcpu);
8324 static void kvm_free_vcpus(struct kvm *kvm)
8327 struct kvm_vcpu *vcpu;
8330 * Unpin any mmu pages first.
8332 kvm_for_each_vcpu(i, vcpu, kvm) {
8333 kvm_clear_async_pf_completion_queue(vcpu);
8334 kvm_unload_vcpu_mmu(vcpu);
8336 kvm_for_each_vcpu(i, vcpu, kvm)
8337 kvm_arch_vcpu_free(vcpu);
8339 mutex_lock(&kvm->lock);
8340 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8341 kvm->vcpus[i] = NULL;
8343 atomic_set(&kvm->online_vcpus, 0);
8344 mutex_unlock(&kvm->lock);
8347 void kvm_arch_sync_events(struct kvm *kvm)
8349 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8350 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8354 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8358 struct kvm_memslots *slots = kvm_memslots(kvm);
8359 struct kvm_memory_slot *slot, old;
8361 /* Called with kvm->slots_lock held. */
8362 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8365 slot = id_to_memslot(slots, id);
8371 * MAP_SHARED to prevent internal slot pages from being moved
8374 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8375 MAP_SHARED | MAP_ANONYMOUS, 0);
8376 if (IS_ERR((void *)hva))
8377 return PTR_ERR((void *)hva);
8386 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8387 struct kvm_userspace_memory_region m;
8389 m.slot = id | (i << 16);
8391 m.guest_phys_addr = gpa;
8392 m.userspace_addr = hva;
8393 m.memory_size = size;
8394 r = __kvm_set_memory_region(kvm, &m);
8400 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8404 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8406 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8410 mutex_lock(&kvm->slots_lock);
8411 r = __x86_set_memory_region(kvm, id, gpa, size);
8412 mutex_unlock(&kvm->slots_lock);
8416 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8418 void kvm_arch_destroy_vm(struct kvm *kvm)
8420 if (current->mm == kvm->mm) {
8422 * Free memory regions allocated on behalf of userspace,
8423 * unless the the memory map has changed due to process exit
8426 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8427 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8428 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8430 if (kvm_x86_ops->vm_destroy)
8431 kvm_x86_ops->vm_destroy(kvm);
8432 kvm_pic_destroy(kvm);
8433 kvm_ioapic_destroy(kvm);
8434 kvm_free_vcpus(kvm);
8435 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8436 kvm_mmu_uninit_vm(kvm);
8437 kvm_page_track_cleanup(kvm);
8440 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8441 struct kvm_memory_slot *dont)
8445 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8446 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8447 kvfree(free->arch.rmap[i]);
8448 free->arch.rmap[i] = NULL;
8453 if (!dont || free->arch.lpage_info[i - 1] !=
8454 dont->arch.lpage_info[i - 1]) {
8455 kvfree(free->arch.lpage_info[i - 1]);
8456 free->arch.lpage_info[i - 1] = NULL;
8460 kvm_page_track_free_memslot(free, dont);
8463 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8464 unsigned long npages)
8468 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8469 struct kvm_lpage_info *linfo;
8474 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8475 slot->base_gfn, level) + 1;
8477 slot->arch.rmap[i] =
8478 kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8479 if (!slot->arch.rmap[i])
8484 linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8488 slot->arch.lpage_info[i - 1] = linfo;
8490 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8491 linfo[0].disallow_lpage = 1;
8492 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8493 linfo[lpages - 1].disallow_lpage = 1;
8494 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8496 * If the gfn and userspace address are not aligned wrt each
8497 * other, or if explicitly asked to, disable large page
8498 * support for this slot
8500 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8501 !kvm_largepages_enabled()) {
8504 for (j = 0; j < lpages; ++j)
8505 linfo[j].disallow_lpage = 1;
8509 if (kvm_page_track_create_memslot(slot, npages))
8515 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8516 kvfree(slot->arch.rmap[i]);
8517 slot->arch.rmap[i] = NULL;
8521 kvfree(slot->arch.lpage_info[i - 1]);
8522 slot->arch.lpage_info[i - 1] = NULL;
8527 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8530 * memslots->generation has been incremented.
8531 * mmio generation may have reached its maximum value.
8533 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8536 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8537 struct kvm_memory_slot *memslot,
8538 const struct kvm_userspace_memory_region *mem,
8539 enum kvm_mr_change change)
8544 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8545 struct kvm_memory_slot *new)
8547 /* Still write protect RO slot */
8548 if (new->flags & KVM_MEM_READONLY) {
8549 kvm_mmu_slot_remove_write_access(kvm, new);
8554 * Call kvm_x86_ops dirty logging hooks when they are valid.
8556 * kvm_x86_ops->slot_disable_log_dirty is called when:
8558 * - KVM_MR_CREATE with dirty logging is disabled
8559 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8561 * The reason is, in case of PML, we need to set D-bit for any slots
8562 * with dirty logging disabled in order to eliminate unnecessary GPA
8563 * logging in PML buffer (and potential PML buffer full VMEXT). This
8564 * guarantees leaving PML enabled during guest's lifetime won't have
8565 * any additonal overhead from PML when guest is running with dirty
8566 * logging disabled for memory slots.
8568 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8569 * to dirty logging mode.
8571 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8573 * In case of write protect:
8575 * Write protect all pages for dirty logging.
8577 * All the sptes including the large sptes which point to this
8578 * slot are set to readonly. We can not create any new large
8579 * spte on this slot until the end of the logging.
8581 * See the comments in fast_page_fault().
8583 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8584 if (kvm_x86_ops->slot_enable_log_dirty)
8585 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8587 kvm_mmu_slot_remove_write_access(kvm, new);
8589 if (kvm_x86_ops->slot_disable_log_dirty)
8590 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8594 void kvm_arch_commit_memory_region(struct kvm *kvm,
8595 const struct kvm_userspace_memory_region *mem,
8596 const struct kvm_memory_slot *old,
8597 const struct kvm_memory_slot *new,
8598 enum kvm_mr_change change)
8600 int nr_mmu_pages = 0;
8602 if (!kvm->arch.n_requested_mmu_pages)
8603 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8606 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8609 * Dirty logging tracks sptes in 4k granularity, meaning that large
8610 * sptes have to be split. If live migration is successful, the guest
8611 * in the source machine will be destroyed and large sptes will be
8612 * created in the destination. However, if the guest continues to run
8613 * in the source machine (for example if live migration fails), small
8614 * sptes will remain around and cause bad performance.
8616 * Scan sptes if dirty logging has been stopped, dropping those
8617 * which can be collapsed into a single large-page spte. Later
8618 * page faults will create the large-page sptes.
8620 if ((change != KVM_MR_DELETE) &&
8621 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8622 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8623 kvm_mmu_zap_collapsible_sptes(kvm, new);
8626 * Set up write protection and/or dirty logging for the new slot.
8628 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8629 * been zapped so no dirty logging staff is needed for old slot. For
8630 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8631 * new and it's also covered when dealing with the new slot.
8633 * FIXME: const-ify all uses of struct kvm_memory_slot.
8635 if (change != KVM_MR_DELETE)
8636 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8639 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8641 kvm_mmu_invalidate_zap_all_pages(kvm);
8644 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8645 struct kvm_memory_slot *slot)
8647 kvm_page_track_flush_slot(kvm, slot);
8650 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8652 if (!list_empty_careful(&vcpu->async_pf.done))
8655 if (kvm_apic_has_events(vcpu))
8658 if (vcpu->arch.pv.pv_unhalted)
8661 if (vcpu->arch.exception.pending)
8664 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
8665 (vcpu->arch.nmi_pending &&
8666 kvm_x86_ops->nmi_allowed(vcpu)))
8669 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
8670 (vcpu->arch.smi_pending && !is_smm(vcpu)))
8673 if (kvm_arch_interrupt_allowed(vcpu) &&
8674 kvm_cpu_has_interrupt(vcpu))
8677 if (kvm_hv_has_stimer_pending(vcpu))
8683 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8685 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8688 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
8690 return vcpu->arch.preempted_in_kernel;
8693 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8695 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8698 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8700 return kvm_x86_ops->interrupt_allowed(vcpu);
8703 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8705 if (is_64_bit_mode(vcpu))
8706 return kvm_rip_read(vcpu);
8707 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8708 kvm_rip_read(vcpu));
8710 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8712 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8714 return kvm_get_linear_rip(vcpu) == linear_rip;
8716 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8718 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8720 unsigned long rflags;
8722 rflags = kvm_x86_ops->get_rflags(vcpu);
8723 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8724 rflags &= ~X86_EFLAGS_TF;
8727 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8729 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8731 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8732 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8733 rflags |= X86_EFLAGS_TF;
8734 kvm_x86_ops->set_rflags(vcpu, rflags);
8737 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8739 __kvm_set_rflags(vcpu, rflags);
8740 kvm_make_request(KVM_REQ_EVENT, vcpu);
8742 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8744 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8748 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8752 r = kvm_mmu_reload(vcpu);
8756 if (!vcpu->arch.mmu.direct_map &&
8757 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8760 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8763 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8765 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8768 static inline u32 kvm_async_pf_next_probe(u32 key)
8770 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8773 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8775 u32 key = kvm_async_pf_hash_fn(gfn);
8777 while (vcpu->arch.apf.gfns[key] != ~0)
8778 key = kvm_async_pf_next_probe(key);
8780 vcpu->arch.apf.gfns[key] = gfn;
8783 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8786 u32 key = kvm_async_pf_hash_fn(gfn);
8788 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8789 (vcpu->arch.apf.gfns[key] != gfn &&
8790 vcpu->arch.apf.gfns[key] != ~0); i++)
8791 key = kvm_async_pf_next_probe(key);
8796 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8798 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8801 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8805 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8807 vcpu->arch.apf.gfns[i] = ~0;
8809 j = kvm_async_pf_next_probe(j);
8810 if (vcpu->arch.apf.gfns[j] == ~0)
8812 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8814 * k lies cyclically in ]i,j]
8816 * |....j i.k.| or |.k..j i...|
8818 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8819 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8824 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8827 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8831 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
8834 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
8838 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8839 struct kvm_async_pf *work)
8841 struct x86_exception fault;
8843 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8844 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8846 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8847 (vcpu->arch.apf.send_user_only &&
8848 kvm_x86_ops->get_cpl(vcpu) == 0))
8849 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8850 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8851 fault.vector = PF_VECTOR;
8852 fault.error_code_valid = true;
8853 fault.error_code = 0;
8854 fault.nested_page_fault = false;
8855 fault.address = work->arch.token;
8856 fault.async_page_fault = true;
8857 kvm_inject_page_fault(vcpu, &fault);
8861 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8862 struct kvm_async_pf *work)
8864 struct x86_exception fault;
8867 if (work->wakeup_all)
8868 work->arch.token = ~0; /* broadcast wakeup */
8870 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8871 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8873 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
8874 !apf_get_user(vcpu, &val)) {
8875 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
8876 vcpu->arch.exception.pending &&
8877 vcpu->arch.exception.nr == PF_VECTOR &&
8878 !apf_put_user(vcpu, 0)) {
8879 vcpu->arch.exception.injected = false;
8880 vcpu->arch.exception.pending = false;
8881 vcpu->arch.exception.nr = 0;
8882 vcpu->arch.exception.has_error_code = false;
8883 vcpu->arch.exception.error_code = 0;
8884 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8885 fault.vector = PF_VECTOR;
8886 fault.error_code_valid = true;
8887 fault.error_code = 0;
8888 fault.nested_page_fault = false;
8889 fault.address = work->arch.token;
8890 fault.async_page_fault = true;
8891 kvm_inject_page_fault(vcpu, &fault);
8894 vcpu->arch.apf.halted = false;
8895 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8898 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8900 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8903 return kvm_can_do_async_pf(vcpu);
8906 void kvm_arch_start_assignment(struct kvm *kvm)
8908 atomic_inc(&kvm->arch.assigned_device_count);
8910 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8912 void kvm_arch_end_assignment(struct kvm *kvm)
8914 atomic_dec(&kvm->arch.assigned_device_count);
8916 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8918 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8920 return atomic_read(&kvm->arch.assigned_device_count);
8922 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8924 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8926 atomic_inc(&kvm->arch.noncoherent_dma_count);
8928 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8930 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8932 atomic_dec(&kvm->arch.noncoherent_dma_count);
8934 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8936 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8938 return atomic_read(&kvm->arch.noncoherent_dma_count);
8940 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8942 bool kvm_arch_has_irq_bypass(void)
8944 return kvm_x86_ops->update_pi_irte != NULL;
8947 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8948 struct irq_bypass_producer *prod)
8950 struct kvm_kernel_irqfd *irqfd =
8951 container_of(cons, struct kvm_kernel_irqfd, consumer);
8953 irqfd->producer = prod;
8955 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8956 prod->irq, irqfd->gsi, 1);
8959 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8960 struct irq_bypass_producer *prod)
8963 struct kvm_kernel_irqfd *irqfd =
8964 container_of(cons, struct kvm_kernel_irqfd, consumer);
8966 WARN_ON(irqfd->producer != prod);
8967 irqfd->producer = NULL;
8970 * When producer of consumer is unregistered, we change back to
8971 * remapped mode, so we can re-use the current implementation
8972 * when the irq is masked/disabled or the consumer side (KVM
8973 * int this case doesn't want to receive the interrupts.
8975 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8977 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8978 " fails: %d\n", irqfd->consumer.token, ret);
8981 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8982 uint32_t guest_irq, bool set)
8984 if (!kvm_x86_ops->update_pi_irte)
8987 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8990 bool kvm_vector_hashing_enabled(void)
8992 return vector_hashing;
8994 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8996 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8997 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8998 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8999 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9000 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9001 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9002 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9003 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9004 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9005 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9006 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9007 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9008 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9009 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9010 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9011 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9012 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9013 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9014 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);