2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
30 #include "assigned-dev.h"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/module.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <trace/events/kvm.h>
58 #define CREATE_TRACE_POINTS
61 #include <asm/debugreg.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
71 #define MAX_IO_MSRS 256
72 #define KVM_MAX_MCE_BANKS 32
73 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
75 #define emul_to_vcpu(ctxt) \
76 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
79 * - enable syscall per default because its emulated by KVM
80 * - enable LME and LMA per default on 64 bit KVM
84 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
86 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
89 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
90 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
92 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
93 static void process_nmi(struct kvm_vcpu *vcpu);
94 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
96 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
97 EXPORT_SYMBOL_GPL(kvm_x86_ops);
99 static bool __read_mostly ignore_msrs = 0;
100 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
102 unsigned int min_timer_period_us = 500;
103 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
105 static bool __read_mostly kvmclock_periodic_sync = true;
106 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
108 bool __read_mostly kvm_has_tsc_control;
109 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
110 u32 __read_mostly kvm_max_guest_tsc_khz;
111 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
112 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
113 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
114 u64 __read_mostly kvm_max_tsc_scaling_ratio;
115 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
116 static u64 __read_mostly kvm_default_tsc_scaling_ratio;
118 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
119 static u32 __read_mostly tsc_tolerance_ppm = 250;
120 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
122 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
123 unsigned int __read_mostly lapic_timer_advance_ns = 0;
124 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
126 static bool __read_mostly vector_hashing = true;
127 module_param(vector_hashing, bool, S_IRUGO);
129 static bool __read_mostly backwards_tsc_observed = false;
131 #define KVM_NR_SHARED_MSRS 16
133 struct kvm_shared_msrs_global {
135 u32 msrs[KVM_NR_SHARED_MSRS];
138 struct kvm_shared_msrs {
139 struct user_return_notifier urn;
141 struct kvm_shared_msr_values {
144 } values[KVM_NR_SHARED_MSRS];
147 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
148 static struct kvm_shared_msrs __percpu *shared_msrs;
150 struct kvm_stats_debugfs_item debugfs_entries[] = {
151 { "pf_fixed", VCPU_STAT(pf_fixed) },
152 { "pf_guest", VCPU_STAT(pf_guest) },
153 { "tlb_flush", VCPU_STAT(tlb_flush) },
154 { "invlpg", VCPU_STAT(invlpg) },
155 { "exits", VCPU_STAT(exits) },
156 { "io_exits", VCPU_STAT(io_exits) },
157 { "mmio_exits", VCPU_STAT(mmio_exits) },
158 { "signal_exits", VCPU_STAT(signal_exits) },
159 { "irq_window", VCPU_STAT(irq_window_exits) },
160 { "nmi_window", VCPU_STAT(nmi_window_exits) },
161 { "halt_exits", VCPU_STAT(halt_exits) },
162 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
163 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
164 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
165 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
166 { "hypercalls", VCPU_STAT(hypercalls) },
167 { "request_irq", VCPU_STAT(request_irq_exits) },
168 { "irq_exits", VCPU_STAT(irq_exits) },
169 { "host_state_reload", VCPU_STAT(host_state_reload) },
170 { "efer_reload", VCPU_STAT(efer_reload) },
171 { "fpu_reload", VCPU_STAT(fpu_reload) },
172 { "insn_emulation", VCPU_STAT(insn_emulation) },
173 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
174 { "irq_injections", VCPU_STAT(irq_injections) },
175 { "nmi_injections", VCPU_STAT(nmi_injections) },
176 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
177 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
178 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
179 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
180 { "mmu_flooded", VM_STAT(mmu_flooded) },
181 { "mmu_recycled", VM_STAT(mmu_recycled) },
182 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
183 { "mmu_unsync", VM_STAT(mmu_unsync) },
184 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
185 { "largepages", VM_STAT(lpages) },
189 u64 __read_mostly host_xcr0;
191 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
193 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
196 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
197 vcpu->arch.apf.gfns[i] = ~0;
200 static void kvm_on_user_return(struct user_return_notifier *urn)
203 struct kvm_shared_msrs *locals
204 = container_of(urn, struct kvm_shared_msrs, urn);
205 struct kvm_shared_msr_values *values;
207 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
208 values = &locals->values[slot];
209 if (values->host != values->curr) {
210 wrmsrl(shared_msrs_global.msrs[slot], values->host);
211 values->curr = values->host;
214 locals->registered = false;
215 user_return_notifier_unregister(urn);
218 static void shared_msr_update(unsigned slot, u32 msr)
221 unsigned int cpu = smp_processor_id();
222 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
224 /* only read, and nobody should modify it at this time,
225 * so don't need lock */
226 if (slot >= shared_msrs_global.nr) {
227 printk(KERN_ERR "kvm: invalid MSR slot!");
230 rdmsrl_safe(msr, &value);
231 smsr->values[slot].host = value;
232 smsr->values[slot].curr = value;
235 void kvm_define_shared_msr(unsigned slot, u32 msr)
237 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
238 shared_msrs_global.msrs[slot] = msr;
239 if (slot >= shared_msrs_global.nr)
240 shared_msrs_global.nr = slot + 1;
242 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
244 static void kvm_shared_msr_cpu_online(void)
248 for (i = 0; i < shared_msrs_global.nr; ++i)
249 shared_msr_update(i, shared_msrs_global.msrs[i]);
252 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
254 unsigned int cpu = smp_processor_id();
255 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
258 if (((value ^ smsr->values[slot].curr) & mask) == 0)
260 smsr->values[slot].curr = value;
261 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
265 if (!smsr->registered) {
266 smsr->urn.on_user_return = kvm_on_user_return;
267 user_return_notifier_register(&smsr->urn);
268 smsr->registered = true;
272 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
274 static void drop_user_return_notifiers(void)
276 unsigned int cpu = smp_processor_id();
277 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
279 if (smsr->registered)
280 kvm_on_user_return(&smsr->urn);
283 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
285 return vcpu->arch.apic_base;
287 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
289 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
291 u64 old_state = vcpu->arch.apic_base &
292 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
293 u64 new_state = msr_info->data &
294 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
295 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
296 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
298 if (!msr_info->host_initiated &&
299 ((msr_info->data & reserved_bits) != 0 ||
300 new_state == X2APIC_ENABLE ||
301 (new_state == MSR_IA32_APICBASE_ENABLE &&
302 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
303 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
307 kvm_lapic_set_base(vcpu, msr_info->data);
310 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
312 asmlinkage __visible void kvm_spurious_fault(void)
314 /* Fault while not rebooting. We want the trace. */
317 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
319 #define EXCPT_BENIGN 0
320 #define EXCPT_CONTRIBUTORY 1
323 static int exception_class(int vector)
333 return EXCPT_CONTRIBUTORY;
340 #define EXCPT_FAULT 0
342 #define EXCPT_ABORT 2
343 #define EXCPT_INTERRUPT 3
345 static int exception_type(int vector)
349 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
350 return EXCPT_INTERRUPT;
354 /* #DB is trap, as instruction watchpoints are handled elsewhere */
355 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
358 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
361 /* Reserved exceptions will result in fault */
365 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
366 unsigned nr, bool has_error, u32 error_code,
372 kvm_make_request(KVM_REQ_EVENT, vcpu);
374 if (!vcpu->arch.exception.pending) {
376 if (has_error && !is_protmode(vcpu))
378 vcpu->arch.exception.pending = true;
379 vcpu->arch.exception.has_error_code = has_error;
380 vcpu->arch.exception.nr = nr;
381 vcpu->arch.exception.error_code = error_code;
382 vcpu->arch.exception.reinject = reinject;
386 /* to check exception */
387 prev_nr = vcpu->arch.exception.nr;
388 if (prev_nr == DF_VECTOR) {
389 /* triple fault -> shutdown */
390 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
393 class1 = exception_class(prev_nr);
394 class2 = exception_class(nr);
395 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
396 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
397 /* generate double fault per SDM Table 5-5 */
398 vcpu->arch.exception.pending = true;
399 vcpu->arch.exception.has_error_code = true;
400 vcpu->arch.exception.nr = DF_VECTOR;
401 vcpu->arch.exception.error_code = 0;
403 /* replace previous exception with a new one in a hope
404 that instruction re-execution will regenerate lost
409 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
411 kvm_multiple_exception(vcpu, nr, false, 0, false);
413 EXPORT_SYMBOL_GPL(kvm_queue_exception);
415 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
417 kvm_multiple_exception(vcpu, nr, false, 0, true);
419 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
421 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
424 kvm_inject_gp(vcpu, 0);
426 kvm_x86_ops->skip_emulated_instruction(vcpu);
428 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
430 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
432 ++vcpu->stat.pf_guest;
433 vcpu->arch.cr2 = fault->address;
434 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
436 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
438 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
440 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
441 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
443 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
445 return fault->nested_page_fault;
448 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
450 atomic_inc(&vcpu->arch.nmi_queued);
451 kvm_make_request(KVM_REQ_NMI, vcpu);
453 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
455 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
457 kvm_multiple_exception(vcpu, nr, true, error_code, false);
459 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
461 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
463 kvm_multiple_exception(vcpu, nr, true, error_code, true);
465 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
468 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
469 * a #GP and return false.
471 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
473 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
475 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
478 EXPORT_SYMBOL_GPL(kvm_require_cpl);
480 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
482 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
485 kvm_queue_exception(vcpu, UD_VECTOR);
488 EXPORT_SYMBOL_GPL(kvm_require_dr);
491 * This function will be used to read from the physical memory of the currently
492 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
493 * can read from guest physical or from the guest's guest physical memory.
495 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
496 gfn_t ngfn, void *data, int offset, int len,
499 struct x86_exception exception;
503 ngpa = gfn_to_gpa(ngfn);
504 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
505 if (real_gfn == UNMAPPED_GVA)
508 real_gfn = gpa_to_gfn(real_gfn);
510 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
512 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
514 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
515 void *data, int offset, int len, u32 access)
517 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
518 data, offset, len, access);
522 * Load the pae pdptrs. Return true is they are all valid.
524 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
526 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
527 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
530 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
532 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
533 offset * sizeof(u64), sizeof(pdpte),
534 PFERR_USER_MASK|PFERR_WRITE_MASK);
539 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
540 if (is_present_gpte(pdpte[i]) &&
542 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
549 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
550 __set_bit(VCPU_EXREG_PDPTR,
551 (unsigned long *)&vcpu->arch.regs_avail);
552 __set_bit(VCPU_EXREG_PDPTR,
553 (unsigned long *)&vcpu->arch.regs_dirty);
558 EXPORT_SYMBOL_GPL(load_pdptrs);
560 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
562 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
568 if (is_long_mode(vcpu) || !is_pae(vcpu))
571 if (!test_bit(VCPU_EXREG_PDPTR,
572 (unsigned long *)&vcpu->arch.regs_avail))
575 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
576 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
577 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
578 PFERR_USER_MASK | PFERR_WRITE_MASK);
581 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
587 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
589 unsigned long old_cr0 = kvm_read_cr0(vcpu);
590 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
595 if (cr0 & 0xffffffff00000000UL)
599 cr0 &= ~CR0_RESERVED_BITS;
601 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
604 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
607 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
609 if ((vcpu->arch.efer & EFER_LME)) {
614 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
619 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
624 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
627 kvm_x86_ops->set_cr0(vcpu, cr0);
629 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
630 kvm_clear_async_pf_completion_queue(vcpu);
631 kvm_async_pf_hash_reset(vcpu);
634 if ((cr0 ^ old_cr0) & update_bits)
635 kvm_mmu_reset_context(vcpu);
637 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
638 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
639 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
640 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
644 EXPORT_SYMBOL_GPL(kvm_set_cr0);
646 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
648 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
650 EXPORT_SYMBOL_GPL(kvm_lmsw);
652 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
654 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
655 !vcpu->guest_xcr0_loaded) {
656 /* kvm_set_xcr() also depends on this */
657 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
658 vcpu->guest_xcr0_loaded = 1;
662 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
664 if (vcpu->guest_xcr0_loaded) {
665 if (vcpu->arch.xcr0 != host_xcr0)
666 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
667 vcpu->guest_xcr0_loaded = 0;
671 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
674 u64 old_xcr0 = vcpu->arch.xcr0;
677 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
678 if (index != XCR_XFEATURE_ENABLED_MASK)
680 if (!(xcr0 & XFEATURE_MASK_FP))
682 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
686 * Do not allow the guest to set bits that we do not support
687 * saving. However, xcr0 bit 0 is always set, even if the
688 * emulated CPU does not support XSAVE (see fx_init).
690 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
691 if (xcr0 & ~valid_bits)
694 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
695 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
698 if (xcr0 & XFEATURE_MASK_AVX512) {
699 if (!(xcr0 & XFEATURE_MASK_YMM))
701 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
704 vcpu->arch.xcr0 = xcr0;
706 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
707 kvm_update_cpuid(vcpu);
711 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
713 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
714 __kvm_set_xcr(vcpu, index, xcr)) {
715 kvm_inject_gp(vcpu, 0);
720 EXPORT_SYMBOL_GPL(kvm_set_xcr);
722 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
724 unsigned long old_cr4 = kvm_read_cr4(vcpu);
725 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
726 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
728 if (cr4 & CR4_RESERVED_BITS)
731 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
734 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
737 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
740 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
743 if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
746 if (is_long_mode(vcpu)) {
747 if (!(cr4 & X86_CR4_PAE))
749 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
750 && ((cr4 ^ old_cr4) & pdptr_bits)
751 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
755 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
756 if (!guest_cpuid_has_pcid(vcpu))
759 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
760 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
764 if (kvm_x86_ops->set_cr4(vcpu, cr4))
767 if (((cr4 ^ old_cr4) & pdptr_bits) ||
768 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
769 kvm_mmu_reset_context(vcpu);
771 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
772 kvm_update_cpuid(vcpu);
776 EXPORT_SYMBOL_GPL(kvm_set_cr4);
778 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
781 cr3 &= ~CR3_PCID_INVD;
784 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
785 kvm_mmu_sync_roots(vcpu);
786 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
790 if (is_long_mode(vcpu)) {
791 if (cr3 & CR3_L_MODE_RESERVED_BITS)
793 } else if (is_pae(vcpu) && is_paging(vcpu) &&
794 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
797 vcpu->arch.cr3 = cr3;
798 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
799 kvm_mmu_new_cr3(vcpu);
802 EXPORT_SYMBOL_GPL(kvm_set_cr3);
804 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
806 if (cr8 & CR8_RESERVED_BITS)
808 if (lapic_in_kernel(vcpu))
809 kvm_lapic_set_tpr(vcpu, cr8);
811 vcpu->arch.cr8 = cr8;
814 EXPORT_SYMBOL_GPL(kvm_set_cr8);
816 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
818 if (lapic_in_kernel(vcpu))
819 return kvm_lapic_get_cr8(vcpu);
821 return vcpu->arch.cr8;
823 EXPORT_SYMBOL_GPL(kvm_get_cr8);
825 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
829 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
830 for (i = 0; i < KVM_NR_DB_REGS; i++)
831 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
832 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
836 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
838 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
839 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
842 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
846 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
847 dr7 = vcpu->arch.guest_debug_dr7;
849 dr7 = vcpu->arch.dr7;
850 kvm_x86_ops->set_dr7(vcpu, dr7);
851 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
852 if (dr7 & DR7_BP_EN_MASK)
853 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
856 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
858 u64 fixed = DR6_FIXED_1;
860 if (!guest_cpuid_has_rtm(vcpu))
865 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
869 vcpu->arch.db[dr] = val;
870 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
871 vcpu->arch.eff_db[dr] = val;
876 if (val & 0xffffffff00000000ULL)
878 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
879 kvm_update_dr6(vcpu);
884 if (val & 0xffffffff00000000ULL)
886 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
887 kvm_update_dr7(vcpu);
894 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
896 if (__kvm_set_dr(vcpu, dr, val)) {
897 kvm_inject_gp(vcpu, 0);
902 EXPORT_SYMBOL_GPL(kvm_set_dr);
904 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
908 *val = vcpu->arch.db[dr];
913 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
914 *val = vcpu->arch.dr6;
916 *val = kvm_x86_ops->get_dr6(vcpu);
921 *val = vcpu->arch.dr7;
926 EXPORT_SYMBOL_GPL(kvm_get_dr);
928 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
930 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
934 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
937 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
938 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
941 EXPORT_SYMBOL_GPL(kvm_rdpmc);
944 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
945 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
947 * This list is modified at module load time to reflect the
948 * capabilities of the host cpu. This capabilities test skips MSRs that are
949 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
950 * may depend on host virtualization features rather than host cpu features.
953 static u32 msrs_to_save[] = {
954 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
957 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
959 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
960 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
963 static unsigned num_msrs_to_save;
965 static u32 emulated_msrs[] = {
966 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
967 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
968 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
969 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
970 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
971 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
974 HV_X64_MSR_VP_RUNTIME,
976 HV_X64_MSR_STIMER0_CONFIG,
977 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
981 MSR_IA32_TSCDEADLINE,
982 MSR_IA32_MISC_ENABLE,
988 static unsigned num_emulated_msrs;
990 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
992 if (efer & efer_reserved_bits)
995 if (efer & EFER_FFXSR) {
996 struct kvm_cpuid_entry2 *feat;
998 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
999 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
1003 if (efer & EFER_SVME) {
1004 struct kvm_cpuid_entry2 *feat;
1006 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1007 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1013 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1015 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1017 u64 old_efer = vcpu->arch.efer;
1019 if (!kvm_valid_efer(vcpu, efer))
1023 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1027 efer |= vcpu->arch.efer & EFER_LMA;
1029 kvm_x86_ops->set_efer(vcpu, efer);
1031 /* Update reserved bits */
1032 if ((efer ^ old_efer) & EFER_NX)
1033 kvm_mmu_reset_context(vcpu);
1038 void kvm_enable_efer_bits(u64 mask)
1040 efer_reserved_bits &= ~mask;
1042 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1045 * Writes msr value into into the appropriate "register".
1046 * Returns 0 on success, non-0 otherwise.
1047 * Assumes vcpu_load() was already called.
1049 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1051 switch (msr->index) {
1054 case MSR_KERNEL_GS_BASE:
1057 if (is_noncanonical_address(msr->data))
1060 case MSR_IA32_SYSENTER_EIP:
1061 case MSR_IA32_SYSENTER_ESP:
1063 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1064 * non-canonical address is written on Intel but not on
1065 * AMD (which ignores the top 32-bits, because it does
1066 * not implement 64-bit SYSENTER).
1068 * 64-bit code should hence be able to write a non-canonical
1069 * value on AMD. Making the address canonical ensures that
1070 * vmentry does not fail on Intel after writing a non-canonical
1071 * value, and that something deterministic happens if the guest
1072 * invokes 64-bit SYSENTER.
1074 msr->data = get_canonical(msr->data);
1076 return kvm_x86_ops->set_msr(vcpu, msr);
1078 EXPORT_SYMBOL_GPL(kvm_set_msr);
1081 * Adapt set_msr() to msr_io()'s calling convention
1083 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1085 struct msr_data msr;
1089 msr.host_initiated = true;
1090 r = kvm_get_msr(vcpu, &msr);
1098 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1100 struct msr_data msr;
1104 msr.host_initiated = true;
1105 return kvm_set_msr(vcpu, &msr);
1108 #ifdef CONFIG_X86_64
1109 struct pvclock_gtod_data {
1112 struct { /* extract of a clocksource struct */
1124 static struct pvclock_gtod_data pvclock_gtod_data;
1126 static void update_pvclock_gtod(struct timekeeper *tk)
1128 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1131 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1133 write_seqcount_begin(&vdata->seq);
1135 /* copy pvclock gtod data */
1136 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1137 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1138 vdata->clock.mask = tk->tkr_mono.mask;
1139 vdata->clock.mult = tk->tkr_mono.mult;
1140 vdata->clock.shift = tk->tkr_mono.shift;
1142 vdata->boot_ns = boot_ns;
1143 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1145 write_seqcount_end(&vdata->seq);
1149 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1152 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1153 * vcpu_enter_guest. This function is only called from
1154 * the physical CPU that is running vcpu.
1156 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1159 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1163 struct pvclock_wall_clock wc;
1164 struct timespec boot;
1169 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1174 ++version; /* first time write, random junk */
1178 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1182 * The guest calculates current wall clock time by adding
1183 * system time (updated by kvm_guest_time_update below) to the
1184 * wall clock specified here. guest system time equals host
1185 * system time for us, thus we must fill in host boot time here.
1189 if (kvm->arch.kvmclock_offset) {
1190 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1191 boot = timespec_sub(boot, ts);
1193 wc.sec = boot.tv_sec;
1194 wc.nsec = boot.tv_nsec;
1195 wc.version = version;
1197 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1200 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1203 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1205 do_shl32_div32(dividend, divisor);
1209 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1210 s8 *pshift, u32 *pmultiplier)
1218 scaled64 = scaled_hz;
1219 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1224 tps32 = (uint32_t)tps64;
1225 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1226 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1234 *pmultiplier = div_frac(scaled64, tps32);
1236 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1237 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1240 #ifdef CONFIG_X86_64
1241 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1244 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1245 static unsigned long max_tsc_khz;
1247 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1249 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1250 vcpu->arch.virtual_tsc_shift);
1253 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1255 u64 v = (u64)khz * (1000000 + ppm);
1260 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1264 /* Guest TSC same frequency as host TSC? */
1266 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1270 /* TSC scaling supported? */
1271 if (!kvm_has_tsc_control) {
1272 if (user_tsc_khz > tsc_khz) {
1273 vcpu->arch.tsc_catchup = 1;
1274 vcpu->arch.tsc_always_catchup = 1;
1277 WARN(1, "user requested TSC rate below hardware speed\n");
1282 /* TSC scaling required - calculate ratio */
1283 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1284 user_tsc_khz, tsc_khz);
1286 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1287 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1292 vcpu->arch.tsc_scaling_ratio = ratio;
1296 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1298 u32 thresh_lo, thresh_hi;
1299 int use_scaling = 0;
1301 /* tsc_khz can be zero if TSC calibration fails */
1302 if (user_tsc_khz == 0) {
1303 /* set tsc_scaling_ratio to a safe value */
1304 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1308 /* Compute a scale to convert nanoseconds in TSC cycles */
1309 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1310 &vcpu->arch.virtual_tsc_shift,
1311 &vcpu->arch.virtual_tsc_mult);
1312 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1315 * Compute the variation in TSC rate which is acceptable
1316 * within the range of tolerance and decide if the
1317 * rate being applied is within that bounds of the hardware
1318 * rate. If so, no scaling or compensation need be done.
1320 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1321 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1322 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1323 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1326 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1329 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1331 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1332 vcpu->arch.virtual_tsc_mult,
1333 vcpu->arch.virtual_tsc_shift);
1334 tsc += vcpu->arch.this_tsc_write;
1338 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1340 #ifdef CONFIG_X86_64
1342 struct kvm_arch *ka = &vcpu->kvm->arch;
1343 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1345 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1346 atomic_read(&vcpu->kvm->online_vcpus));
1349 * Once the masterclock is enabled, always perform request in
1350 * order to update it.
1352 * In order to enable masterclock, the host clocksource must be TSC
1353 * and the vcpus need to have matched TSCs. When that happens,
1354 * perform request to enable masterclock.
1356 if (ka->use_master_clock ||
1357 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1358 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1360 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1361 atomic_read(&vcpu->kvm->online_vcpus),
1362 ka->use_master_clock, gtod->clock.vclock_mode);
1366 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1368 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1369 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1373 * Multiply tsc by a fixed point number represented by ratio.
1375 * The most significant 64-N bits (mult) of ratio represent the
1376 * integral part of the fixed point number; the remaining N bits
1377 * (frac) represent the fractional part, ie. ratio represents a fixed
1378 * point number (mult + frac * 2^(-N)).
1380 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1382 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1384 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1387 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1390 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1392 if (ratio != kvm_default_tsc_scaling_ratio)
1393 _tsc = __scale_tsc(ratio, tsc);
1397 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1399 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1403 tsc = kvm_scale_tsc(vcpu, rdtsc());
1405 return target_tsc - tsc;
1408 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1410 return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
1412 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1414 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1416 struct kvm *kvm = vcpu->kvm;
1417 u64 offset, ns, elapsed;
1418 unsigned long flags;
1421 bool already_matched;
1422 u64 data = msr->data;
1424 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1425 offset = kvm_compute_tsc_offset(vcpu, data);
1426 ns = get_kernel_ns();
1427 elapsed = ns - kvm->arch.last_tsc_nsec;
1429 if (vcpu->arch.virtual_tsc_khz) {
1432 /* n.b - signed multiplication and division required */
1433 usdiff = data - kvm->arch.last_tsc_write;
1434 #ifdef CONFIG_X86_64
1435 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1437 /* do_div() only does unsigned */
1438 asm("1: idivl %[divisor]\n"
1439 "2: xor %%edx, %%edx\n"
1440 " movl $0, %[faulted]\n"
1442 ".section .fixup,\"ax\"\n"
1443 "4: movl $1, %[faulted]\n"
1447 _ASM_EXTABLE(1b, 4b)
1449 : "=A"(usdiff), [faulted] "=r" (faulted)
1450 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1453 do_div(elapsed, 1000);
1458 /* idivl overflow => difference is larger than USEC_PER_SEC */
1460 usdiff = USEC_PER_SEC;
1462 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1465 * Special case: TSC write with a small delta (1 second) of virtual
1466 * cycle time against real time is interpreted as an attempt to
1467 * synchronize the CPU.
1469 * For a reliable TSC, we can match TSC offsets, and for an unstable
1470 * TSC, we add elapsed time in this computation. We could let the
1471 * compensation code attempt to catch up if we fall behind, but
1472 * it's better to try to match offsets from the beginning.
1474 if (usdiff < USEC_PER_SEC &&
1475 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1476 if (!check_tsc_unstable()) {
1477 offset = kvm->arch.cur_tsc_offset;
1478 pr_debug("kvm: matched tsc offset for %llu\n", data);
1480 u64 delta = nsec_to_cycles(vcpu, elapsed);
1482 offset = kvm_compute_tsc_offset(vcpu, data);
1483 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1486 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1489 * We split periods of matched TSC writes into generations.
1490 * For each generation, we track the original measured
1491 * nanosecond time, offset, and write, so if TSCs are in
1492 * sync, we can match exact offset, and if not, we can match
1493 * exact software computation in compute_guest_tsc()
1495 * These values are tracked in kvm->arch.cur_xxx variables.
1497 kvm->arch.cur_tsc_generation++;
1498 kvm->arch.cur_tsc_nsec = ns;
1499 kvm->arch.cur_tsc_write = data;
1500 kvm->arch.cur_tsc_offset = offset;
1502 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1503 kvm->arch.cur_tsc_generation, data);
1507 * We also track th most recent recorded KHZ, write and time to
1508 * allow the matching interval to be extended at each write.
1510 kvm->arch.last_tsc_nsec = ns;
1511 kvm->arch.last_tsc_write = data;
1512 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1514 vcpu->arch.last_guest_tsc = data;
1516 /* Keep track of which generation this VCPU has synchronized to */
1517 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1518 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1519 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1521 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1522 update_ia32_tsc_adjust_msr(vcpu, offset);
1523 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1524 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1526 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1528 kvm->arch.nr_vcpus_matched_tsc = 0;
1529 } else if (!already_matched) {
1530 kvm->arch.nr_vcpus_matched_tsc++;
1533 kvm_track_tsc_matching(vcpu);
1534 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1537 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1539 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1542 kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1545 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1547 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1548 WARN_ON(adjustment < 0);
1549 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1550 kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1553 #ifdef CONFIG_X86_64
1555 static cycle_t read_tsc(void)
1557 cycle_t ret = (cycle_t)rdtsc_ordered();
1558 u64 last = pvclock_gtod_data.clock.cycle_last;
1560 if (likely(ret >= last))
1564 * GCC likes to generate cmov here, but this branch is extremely
1565 * predictable (it's just a function of time and the likely is
1566 * very likely) and there's a data dependence, so force GCC
1567 * to generate a branch instead. I don't barrier() because
1568 * we don't actually need a barrier, and if this function
1569 * ever gets inlined it will generate worse code.
1575 static inline u64 vgettsc(cycle_t *cycle_now)
1578 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1580 *cycle_now = read_tsc();
1582 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1583 return v * gtod->clock.mult;
1586 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1588 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1594 seq = read_seqcount_begin(>od->seq);
1595 mode = gtod->clock.vclock_mode;
1596 ns = gtod->nsec_base;
1597 ns += vgettsc(cycle_now);
1598 ns >>= gtod->clock.shift;
1599 ns += gtod->boot_ns;
1600 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1606 /* returns true if host is using tsc clocksource */
1607 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1609 /* checked again under seqlock below */
1610 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1613 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1619 * Assuming a stable TSC across physical CPUS, and a stable TSC
1620 * across virtual CPUs, the following condition is possible.
1621 * Each numbered line represents an event visible to both
1622 * CPUs at the next numbered event.
1624 * "timespecX" represents host monotonic time. "tscX" represents
1627 * VCPU0 on CPU0 | VCPU1 on CPU1
1629 * 1. read timespec0,tsc0
1630 * 2. | timespec1 = timespec0 + N
1632 * 3. transition to guest | transition to guest
1633 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1634 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1635 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1637 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1640 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1642 * - 0 < N - M => M < N
1644 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1645 * always the case (the difference between two distinct xtime instances
1646 * might be smaller then the difference between corresponding TSC reads,
1647 * when updating guest vcpus pvclock areas).
1649 * To avoid that problem, do not allow visibility of distinct
1650 * system_timestamp/tsc_timestamp values simultaneously: use a master
1651 * copy of host monotonic time values. Update that master copy
1654 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1658 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1660 #ifdef CONFIG_X86_64
1661 struct kvm_arch *ka = &kvm->arch;
1663 bool host_tsc_clocksource, vcpus_matched;
1665 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1666 atomic_read(&kvm->online_vcpus));
1669 * If the host uses TSC clock, then passthrough TSC as stable
1672 host_tsc_clocksource = kvm_get_time_and_clockread(
1673 &ka->master_kernel_ns,
1674 &ka->master_cycle_now);
1676 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1677 && !backwards_tsc_observed
1678 && !ka->boot_vcpu_runs_old_kvmclock;
1680 if (ka->use_master_clock)
1681 atomic_set(&kvm_guest_has_master_clock, 1);
1683 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1684 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1689 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1691 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1694 static void kvm_gen_update_masterclock(struct kvm *kvm)
1696 #ifdef CONFIG_X86_64
1698 struct kvm_vcpu *vcpu;
1699 struct kvm_arch *ka = &kvm->arch;
1701 spin_lock(&ka->pvclock_gtod_sync_lock);
1702 kvm_make_mclock_inprogress_request(kvm);
1703 /* no guest entries from this point */
1704 pvclock_update_vm_gtod_copy(kvm);
1706 kvm_for_each_vcpu(i, vcpu, kvm)
1707 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1709 /* guest entries allowed */
1710 kvm_for_each_vcpu(i, vcpu, kvm)
1711 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1713 spin_unlock(&ka->pvclock_gtod_sync_lock);
1717 static int kvm_guest_time_update(struct kvm_vcpu *v)
1719 unsigned long flags, tgt_tsc_khz;
1720 struct kvm_vcpu_arch *vcpu = &v->arch;
1721 struct kvm_arch *ka = &v->kvm->arch;
1723 u64 tsc_timestamp, host_tsc;
1724 struct pvclock_vcpu_time_info guest_hv_clock;
1726 bool use_master_clock;
1732 * If the host uses TSC clock, then passthrough TSC as stable
1735 spin_lock(&ka->pvclock_gtod_sync_lock);
1736 use_master_clock = ka->use_master_clock;
1737 if (use_master_clock) {
1738 host_tsc = ka->master_cycle_now;
1739 kernel_ns = ka->master_kernel_ns;
1741 spin_unlock(&ka->pvclock_gtod_sync_lock);
1743 /* Keep irq disabled to prevent changes to the clock */
1744 local_irq_save(flags);
1745 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1746 if (unlikely(tgt_tsc_khz == 0)) {
1747 local_irq_restore(flags);
1748 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1751 if (!use_master_clock) {
1753 kernel_ns = get_kernel_ns();
1756 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1759 * We may have to catch up the TSC to match elapsed wall clock
1760 * time for two reasons, even if kvmclock is used.
1761 * 1) CPU could have been running below the maximum TSC rate
1762 * 2) Broken TSC compensation resets the base at each VCPU
1763 * entry to avoid unknown leaps of TSC even when running
1764 * again on the same CPU. This may cause apparent elapsed
1765 * time to disappear, and the guest to stand still or run
1768 if (vcpu->tsc_catchup) {
1769 u64 tsc = compute_guest_tsc(v, kernel_ns);
1770 if (tsc > tsc_timestamp) {
1771 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1772 tsc_timestamp = tsc;
1776 local_irq_restore(flags);
1778 if (!vcpu->pv_time_enabled)
1781 if (kvm_has_tsc_control)
1782 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1784 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1785 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1786 &vcpu->hv_clock.tsc_shift,
1787 &vcpu->hv_clock.tsc_to_system_mul);
1788 vcpu->hw_tsc_khz = tgt_tsc_khz;
1791 /* With all the info we got, fill in the values */
1792 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1793 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1794 vcpu->last_guest_tsc = tsc_timestamp;
1796 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1797 &guest_hv_clock, sizeof(guest_hv_clock))))
1800 /* This VCPU is paused, but it's legal for a guest to read another
1801 * VCPU's kvmclock, so we really have to follow the specification where
1802 * it says that version is odd if data is being modified, and even after
1805 * Version field updates must be kept separate. This is because
1806 * kvm_write_guest_cached might use a "rep movs" instruction, and
1807 * writes within a string instruction are weakly ordered. So there
1808 * are three writes overall.
1810 * As a small optimization, only write the version field in the first
1811 * and third write. The vcpu->pv_time cache is still valid, because the
1812 * version field is the first in the struct.
1814 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1816 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1817 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1819 sizeof(vcpu->hv_clock.version));
1823 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1824 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1826 if (vcpu->pvclock_set_guest_stopped_request) {
1827 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1828 vcpu->pvclock_set_guest_stopped_request = false;
1831 /* If the host uses TSC clocksource, then it is stable */
1832 if (use_master_clock)
1833 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1835 vcpu->hv_clock.flags = pvclock_flags;
1837 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1839 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1841 sizeof(vcpu->hv_clock));
1845 vcpu->hv_clock.version++;
1846 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1848 sizeof(vcpu->hv_clock.version));
1853 * kvmclock updates which are isolated to a given vcpu, such as
1854 * vcpu->cpu migration, should not allow system_timestamp from
1855 * the rest of the vcpus to remain static. Otherwise ntp frequency
1856 * correction applies to one vcpu's system_timestamp but not
1859 * So in those cases, request a kvmclock update for all vcpus.
1860 * We need to rate-limit these requests though, as they can
1861 * considerably slow guests that have a large number of vcpus.
1862 * The time for a remote vcpu to update its kvmclock is bound
1863 * by the delay we use to rate-limit the updates.
1866 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1868 static void kvmclock_update_fn(struct work_struct *work)
1871 struct delayed_work *dwork = to_delayed_work(work);
1872 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1873 kvmclock_update_work);
1874 struct kvm *kvm = container_of(ka, struct kvm, arch);
1875 struct kvm_vcpu *vcpu;
1877 kvm_for_each_vcpu(i, vcpu, kvm) {
1878 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1879 kvm_vcpu_kick(vcpu);
1883 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1885 struct kvm *kvm = v->kvm;
1887 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1888 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1889 KVMCLOCK_UPDATE_DELAY);
1892 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1894 static void kvmclock_sync_fn(struct work_struct *work)
1896 struct delayed_work *dwork = to_delayed_work(work);
1897 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1898 kvmclock_sync_work);
1899 struct kvm *kvm = container_of(ka, struct kvm, arch);
1901 if (!kvmclock_periodic_sync)
1904 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1905 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1906 KVMCLOCK_SYNC_PERIOD);
1909 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1911 u64 mcg_cap = vcpu->arch.mcg_cap;
1912 unsigned bank_num = mcg_cap & 0xff;
1915 case MSR_IA32_MCG_STATUS:
1916 vcpu->arch.mcg_status = data;
1918 case MSR_IA32_MCG_CTL:
1919 if (!(mcg_cap & MCG_CTL_P))
1921 if (data != 0 && data != ~(u64)0)
1923 vcpu->arch.mcg_ctl = data;
1926 if (msr >= MSR_IA32_MC0_CTL &&
1927 msr < MSR_IA32_MCx_CTL(bank_num)) {
1928 u32 offset = msr - MSR_IA32_MC0_CTL;
1929 /* only 0 or all 1s can be written to IA32_MCi_CTL
1930 * some Linux kernels though clear bit 10 in bank 4 to
1931 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1932 * this to avoid an uncatched #GP in the guest
1934 if ((offset & 0x3) == 0 &&
1935 data != 0 && (data | (1 << 10)) != ~(u64)0)
1937 vcpu->arch.mce_banks[offset] = data;
1945 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1947 struct kvm *kvm = vcpu->kvm;
1948 int lm = is_long_mode(vcpu);
1949 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1950 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1951 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1952 : kvm->arch.xen_hvm_config.blob_size_32;
1953 u32 page_num = data & ~PAGE_MASK;
1954 u64 page_addr = data & PAGE_MASK;
1959 if (page_num >= blob_size)
1962 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1967 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1976 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1978 gpa_t gpa = data & ~0x3f;
1980 /* Bits 2:5 are reserved, Should be zero */
1984 vcpu->arch.apf.msr_val = data;
1986 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1987 kvm_clear_async_pf_completion_queue(vcpu);
1988 kvm_async_pf_hash_reset(vcpu);
1992 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1996 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1997 kvm_async_pf_wakeup_all(vcpu);
2001 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2003 vcpu->arch.pv_time_enabled = false;
2006 static void record_steal_time(struct kvm_vcpu *vcpu)
2008 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2011 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2012 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2015 if (vcpu->arch.st.steal.version & 1)
2016 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2018 vcpu->arch.st.steal.version += 1;
2020 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2021 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2025 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2026 vcpu->arch.st.last_steal;
2027 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2029 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2030 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2034 vcpu->arch.st.steal.version += 1;
2036 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2037 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2040 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2043 u32 msr = msr_info->index;
2044 u64 data = msr_info->data;
2047 case MSR_AMD64_NB_CFG:
2048 case MSR_IA32_UCODE_REV:
2049 case MSR_IA32_UCODE_WRITE:
2050 case MSR_VM_HSAVE_PA:
2051 case MSR_AMD64_PATCH_LOADER:
2052 case MSR_AMD64_BU_CFG2:
2056 return set_efer(vcpu, data);
2058 data &= ~(u64)0x40; /* ignore flush filter disable */
2059 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2060 data &= ~(u64)0x8; /* ignore TLB cache disable */
2061 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2063 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2068 case MSR_FAM10H_MMIO_CONF_BASE:
2070 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2075 case MSR_IA32_DEBUGCTLMSR:
2077 /* We support the non-activated case already */
2079 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2080 /* Values other than LBR and BTF are vendor-specific,
2081 thus reserved and should throw a #GP */
2084 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2087 case 0x200 ... 0x2ff:
2088 return kvm_mtrr_set_msr(vcpu, msr, data);
2089 case MSR_IA32_APICBASE:
2090 return kvm_set_apic_base(vcpu, msr_info);
2091 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2092 return kvm_x2apic_msr_write(vcpu, msr, data);
2093 case MSR_IA32_TSCDEADLINE:
2094 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2096 case MSR_IA32_TSC_ADJUST:
2097 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2098 if (!msr_info->host_initiated) {
2099 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2100 adjust_tsc_offset_guest(vcpu, adj);
2102 vcpu->arch.ia32_tsc_adjust_msr = data;
2105 case MSR_IA32_MISC_ENABLE:
2106 vcpu->arch.ia32_misc_enable_msr = data;
2108 case MSR_IA32_SMBASE:
2109 if (!msr_info->host_initiated)
2111 vcpu->arch.smbase = data;
2113 case MSR_KVM_WALL_CLOCK_NEW:
2114 case MSR_KVM_WALL_CLOCK:
2115 vcpu->kvm->arch.wall_clock = data;
2116 kvm_write_wall_clock(vcpu->kvm, data);
2118 case MSR_KVM_SYSTEM_TIME_NEW:
2119 case MSR_KVM_SYSTEM_TIME: {
2121 struct kvm_arch *ka = &vcpu->kvm->arch;
2123 kvmclock_reset(vcpu);
2125 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2126 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2128 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2129 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2132 ka->boot_vcpu_runs_old_kvmclock = tmp;
2135 vcpu->arch.time = data;
2136 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2138 /* we verify if the enable bit is set... */
2142 gpa_offset = data & ~(PAGE_MASK | 1);
2144 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2145 &vcpu->arch.pv_time, data & ~1ULL,
2146 sizeof(struct pvclock_vcpu_time_info)))
2147 vcpu->arch.pv_time_enabled = false;
2149 vcpu->arch.pv_time_enabled = true;
2153 case MSR_KVM_ASYNC_PF_EN:
2154 if (kvm_pv_enable_async_pf(vcpu, data))
2157 case MSR_KVM_STEAL_TIME:
2159 if (unlikely(!sched_info_on()))
2162 if (data & KVM_STEAL_RESERVED_MASK)
2165 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2166 data & KVM_STEAL_VALID_BITS,
2167 sizeof(struct kvm_steal_time)))
2170 vcpu->arch.st.msr_val = data;
2172 if (!(data & KVM_MSR_ENABLED))
2175 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2178 case MSR_KVM_PV_EOI_EN:
2179 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2183 case MSR_IA32_MCG_CTL:
2184 case MSR_IA32_MCG_STATUS:
2185 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2186 return set_msr_mce(vcpu, msr, data);
2188 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2189 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2190 pr = true; /* fall through */
2191 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2192 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2193 if (kvm_pmu_is_valid_msr(vcpu, msr))
2194 return kvm_pmu_set_msr(vcpu, msr_info);
2196 if (pr || data != 0)
2197 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2198 "0x%x data 0x%llx\n", msr, data);
2200 case MSR_K7_CLK_CTL:
2202 * Ignore all writes to this no longer documented MSR.
2203 * Writes are only relevant for old K7 processors,
2204 * all pre-dating SVM, but a recommended workaround from
2205 * AMD for these chips. It is possible to specify the
2206 * affected processor models on the command line, hence
2207 * the need to ignore the workaround.
2210 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2211 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2212 case HV_X64_MSR_CRASH_CTL:
2213 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2214 return kvm_hv_set_msr_common(vcpu, msr, data,
2215 msr_info->host_initiated);
2216 case MSR_IA32_BBL_CR_CTL3:
2217 /* Drop writes to this legacy MSR -- see rdmsr
2218 * counterpart for further detail.
2220 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2222 case MSR_AMD64_OSVW_ID_LENGTH:
2223 if (!guest_cpuid_has_osvw(vcpu))
2225 vcpu->arch.osvw.length = data;
2227 case MSR_AMD64_OSVW_STATUS:
2228 if (!guest_cpuid_has_osvw(vcpu))
2230 vcpu->arch.osvw.status = data;
2233 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2234 return xen_hvm_config(vcpu, data);
2235 if (kvm_pmu_is_valid_msr(vcpu, msr))
2236 return kvm_pmu_set_msr(vcpu, msr_info);
2238 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2242 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2249 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2253 * Reads an msr value (of 'msr_index') into 'pdata'.
2254 * Returns 0 on success, non-0 otherwise.
2255 * Assumes vcpu_load() was already called.
2257 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2259 return kvm_x86_ops->get_msr(vcpu, msr);
2261 EXPORT_SYMBOL_GPL(kvm_get_msr);
2263 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2266 u64 mcg_cap = vcpu->arch.mcg_cap;
2267 unsigned bank_num = mcg_cap & 0xff;
2270 case MSR_IA32_P5_MC_ADDR:
2271 case MSR_IA32_P5_MC_TYPE:
2274 case MSR_IA32_MCG_CAP:
2275 data = vcpu->arch.mcg_cap;
2277 case MSR_IA32_MCG_CTL:
2278 if (!(mcg_cap & MCG_CTL_P))
2280 data = vcpu->arch.mcg_ctl;
2282 case MSR_IA32_MCG_STATUS:
2283 data = vcpu->arch.mcg_status;
2286 if (msr >= MSR_IA32_MC0_CTL &&
2287 msr < MSR_IA32_MCx_CTL(bank_num)) {
2288 u32 offset = msr - MSR_IA32_MC0_CTL;
2289 data = vcpu->arch.mce_banks[offset];
2298 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2300 switch (msr_info->index) {
2301 case MSR_IA32_PLATFORM_ID:
2302 case MSR_IA32_EBL_CR_POWERON:
2303 case MSR_IA32_DEBUGCTLMSR:
2304 case MSR_IA32_LASTBRANCHFROMIP:
2305 case MSR_IA32_LASTBRANCHTOIP:
2306 case MSR_IA32_LASTINTFROMIP:
2307 case MSR_IA32_LASTINTTOIP:
2309 case MSR_K8_TSEG_ADDR:
2310 case MSR_K8_TSEG_MASK:
2312 case MSR_VM_HSAVE_PA:
2313 case MSR_K8_INT_PENDING_MSG:
2314 case MSR_AMD64_NB_CFG:
2315 case MSR_FAM10H_MMIO_CONF_BASE:
2316 case MSR_AMD64_BU_CFG2:
2317 case MSR_IA32_PERF_CTL:
2320 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2321 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2322 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2323 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2324 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2325 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2328 case MSR_IA32_UCODE_REV:
2329 msr_info->data = 0x100000000ULL;
2332 case 0x200 ... 0x2ff:
2333 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2334 case 0xcd: /* fsb frequency */
2338 * MSR_EBC_FREQUENCY_ID
2339 * Conservative value valid for even the basic CPU models.
2340 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2341 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2342 * and 266MHz for model 3, or 4. Set Core Clock
2343 * Frequency to System Bus Frequency Ratio to 1 (bits
2344 * 31:24) even though these are only valid for CPU
2345 * models > 2, however guests may end up dividing or
2346 * multiplying by zero otherwise.
2348 case MSR_EBC_FREQUENCY_ID:
2349 msr_info->data = 1 << 24;
2351 case MSR_IA32_APICBASE:
2352 msr_info->data = kvm_get_apic_base(vcpu);
2354 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2355 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2357 case MSR_IA32_TSCDEADLINE:
2358 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2360 case MSR_IA32_TSC_ADJUST:
2361 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2363 case MSR_IA32_MISC_ENABLE:
2364 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2366 case MSR_IA32_SMBASE:
2367 if (!msr_info->host_initiated)
2369 msr_info->data = vcpu->arch.smbase;
2371 case MSR_IA32_PERF_STATUS:
2372 /* TSC increment by tick */
2373 msr_info->data = 1000ULL;
2374 /* CPU multiplier */
2375 msr_info->data |= (((uint64_t)4ULL) << 40);
2378 msr_info->data = vcpu->arch.efer;
2380 case MSR_KVM_WALL_CLOCK:
2381 case MSR_KVM_WALL_CLOCK_NEW:
2382 msr_info->data = vcpu->kvm->arch.wall_clock;
2384 case MSR_KVM_SYSTEM_TIME:
2385 case MSR_KVM_SYSTEM_TIME_NEW:
2386 msr_info->data = vcpu->arch.time;
2388 case MSR_KVM_ASYNC_PF_EN:
2389 msr_info->data = vcpu->arch.apf.msr_val;
2391 case MSR_KVM_STEAL_TIME:
2392 msr_info->data = vcpu->arch.st.msr_val;
2394 case MSR_KVM_PV_EOI_EN:
2395 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2397 case MSR_IA32_P5_MC_ADDR:
2398 case MSR_IA32_P5_MC_TYPE:
2399 case MSR_IA32_MCG_CAP:
2400 case MSR_IA32_MCG_CTL:
2401 case MSR_IA32_MCG_STATUS:
2402 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2403 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2404 case MSR_K7_CLK_CTL:
2406 * Provide expected ramp-up count for K7. All other
2407 * are set to zero, indicating minimum divisors for
2410 * This prevents guest kernels on AMD host with CPU
2411 * type 6, model 8 and higher from exploding due to
2412 * the rdmsr failing.
2414 msr_info->data = 0x20000000;
2416 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2417 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2418 case HV_X64_MSR_CRASH_CTL:
2419 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2420 return kvm_hv_get_msr_common(vcpu,
2421 msr_info->index, &msr_info->data);
2423 case MSR_IA32_BBL_CR_CTL3:
2424 /* This legacy MSR exists but isn't fully documented in current
2425 * silicon. It is however accessed by winxp in very narrow
2426 * scenarios where it sets bit #19, itself documented as
2427 * a "reserved" bit. Best effort attempt to source coherent
2428 * read data here should the balance of the register be
2429 * interpreted by the guest:
2431 * L2 cache control register 3: 64GB range, 256KB size,
2432 * enabled, latency 0x1, configured
2434 msr_info->data = 0xbe702111;
2436 case MSR_AMD64_OSVW_ID_LENGTH:
2437 if (!guest_cpuid_has_osvw(vcpu))
2439 msr_info->data = vcpu->arch.osvw.length;
2441 case MSR_AMD64_OSVW_STATUS:
2442 if (!guest_cpuid_has_osvw(vcpu))
2444 msr_info->data = vcpu->arch.osvw.status;
2447 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2448 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2450 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2453 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2460 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2463 * Read or write a bunch of msrs. All parameters are kernel addresses.
2465 * @return number of msrs set successfully.
2467 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2468 struct kvm_msr_entry *entries,
2469 int (*do_msr)(struct kvm_vcpu *vcpu,
2470 unsigned index, u64 *data))
2474 idx = srcu_read_lock(&vcpu->kvm->srcu);
2475 for (i = 0; i < msrs->nmsrs; ++i)
2476 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2478 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2484 * Read or write a bunch of msrs. Parameters are user addresses.
2486 * @return number of msrs set successfully.
2488 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2489 int (*do_msr)(struct kvm_vcpu *vcpu,
2490 unsigned index, u64 *data),
2493 struct kvm_msrs msrs;
2494 struct kvm_msr_entry *entries;
2499 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2503 if (msrs.nmsrs >= MAX_IO_MSRS)
2506 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2507 entries = memdup_user(user_msrs->entries, size);
2508 if (IS_ERR(entries)) {
2509 r = PTR_ERR(entries);
2513 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2518 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2529 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2534 case KVM_CAP_IRQCHIP:
2536 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2537 case KVM_CAP_SET_TSS_ADDR:
2538 case KVM_CAP_EXT_CPUID:
2539 case KVM_CAP_EXT_EMUL_CPUID:
2540 case KVM_CAP_CLOCKSOURCE:
2542 case KVM_CAP_NOP_IO_DELAY:
2543 case KVM_CAP_MP_STATE:
2544 case KVM_CAP_SYNC_MMU:
2545 case KVM_CAP_USER_NMI:
2546 case KVM_CAP_REINJECT_CONTROL:
2547 case KVM_CAP_IRQ_INJECT_STATUS:
2548 case KVM_CAP_IOEVENTFD:
2549 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2551 case KVM_CAP_PIT_STATE2:
2552 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2553 case KVM_CAP_XEN_HVM:
2554 case KVM_CAP_ADJUST_CLOCK:
2555 case KVM_CAP_VCPU_EVENTS:
2556 case KVM_CAP_HYPERV:
2557 case KVM_CAP_HYPERV_VAPIC:
2558 case KVM_CAP_HYPERV_SPIN:
2559 case KVM_CAP_HYPERV_SYNIC:
2560 case KVM_CAP_PCI_SEGMENT:
2561 case KVM_CAP_DEBUGREGS:
2562 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2564 case KVM_CAP_ASYNC_PF:
2565 case KVM_CAP_GET_TSC_KHZ:
2566 case KVM_CAP_KVMCLOCK_CTRL:
2567 case KVM_CAP_READONLY_MEM:
2568 case KVM_CAP_HYPERV_TIME:
2569 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2570 case KVM_CAP_TSC_DEADLINE_TIMER:
2571 case KVM_CAP_ENABLE_CAP_VM:
2572 case KVM_CAP_DISABLE_QUIRKS:
2573 case KVM_CAP_SET_BOOT_CPU_ID:
2574 case KVM_CAP_SPLIT_IRQCHIP:
2575 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2576 case KVM_CAP_ASSIGN_DEV_IRQ:
2577 case KVM_CAP_PCI_2_3:
2581 case KVM_CAP_X86_SMM:
2582 /* SMBASE is usually relocated above 1M on modern chipsets,
2583 * and SMM handlers might indeed rely on 4G segment limits,
2584 * so do not report SMM to be available if real mode is
2585 * emulated via vm86 mode. Still, do not go to great lengths
2586 * to avoid userspace's usage of the feature, because it is a
2587 * fringe case that is not enabled except via specific settings
2588 * of the module parameters.
2590 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2592 case KVM_CAP_COALESCED_MMIO:
2593 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2596 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2598 case KVM_CAP_NR_VCPUS:
2599 r = KVM_SOFT_MAX_VCPUS;
2601 case KVM_CAP_MAX_VCPUS:
2604 case KVM_CAP_NR_MEMSLOTS:
2605 r = KVM_USER_MEM_SLOTS;
2607 case KVM_CAP_PV_MMU: /* obsolete */
2610 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2612 r = iommu_present(&pci_bus_type);
2616 r = KVM_MAX_MCE_BANKS;
2619 r = boot_cpu_has(X86_FEATURE_XSAVE);
2621 case KVM_CAP_TSC_CONTROL:
2622 r = kvm_has_tsc_control;
2632 long kvm_arch_dev_ioctl(struct file *filp,
2633 unsigned int ioctl, unsigned long arg)
2635 void __user *argp = (void __user *)arg;
2639 case KVM_GET_MSR_INDEX_LIST: {
2640 struct kvm_msr_list __user *user_msr_list = argp;
2641 struct kvm_msr_list msr_list;
2645 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2648 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2649 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2652 if (n < msr_list.nmsrs)
2655 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2656 num_msrs_to_save * sizeof(u32)))
2658 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2660 num_emulated_msrs * sizeof(u32)))
2665 case KVM_GET_SUPPORTED_CPUID:
2666 case KVM_GET_EMULATED_CPUID: {
2667 struct kvm_cpuid2 __user *cpuid_arg = argp;
2668 struct kvm_cpuid2 cpuid;
2671 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2674 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2680 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2685 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2688 mce_cap = KVM_MCE_CAP_SUPPORTED;
2690 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2702 static void wbinvd_ipi(void *garbage)
2707 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2709 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2712 static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
2714 set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
2717 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2719 /* Address WBINVD may be executed by guest */
2720 if (need_emulate_wbinvd(vcpu)) {
2721 if (kvm_x86_ops->has_wbinvd_exit())
2722 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2723 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2724 smp_call_function_single(vcpu->cpu,
2725 wbinvd_ipi, NULL, 1);
2728 kvm_x86_ops->vcpu_load(vcpu, cpu);
2730 /* Apply any externally detected TSC adjustments (due to suspend) */
2731 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2732 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2733 vcpu->arch.tsc_offset_adjustment = 0;
2734 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2737 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2738 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2739 rdtsc() - vcpu->arch.last_host_tsc;
2741 mark_tsc_unstable("KVM discovered backwards TSC");
2742 if (check_tsc_unstable()) {
2743 u64 offset = kvm_compute_tsc_offset(vcpu,
2744 vcpu->arch.last_guest_tsc);
2745 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2746 vcpu->arch.tsc_catchup = 1;
2749 * On a host with synchronized TSC, there is no need to update
2750 * kvmclock on vcpu->cpu migration
2752 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2753 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2754 if (vcpu->cpu != cpu)
2755 kvm_migrate_timers(vcpu);
2759 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2762 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2764 kvm_x86_ops->vcpu_put(vcpu);
2765 kvm_put_guest_fpu(vcpu);
2766 vcpu->arch.last_host_tsc = rdtsc();
2769 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2770 struct kvm_lapic_state *s)
2772 if (vcpu->arch.apicv_active)
2773 kvm_x86_ops->sync_pir_to_irr(vcpu);
2775 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2780 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2781 struct kvm_lapic_state *s)
2783 kvm_apic_post_state_restore(vcpu, s);
2784 update_cr8_intercept(vcpu);
2789 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2791 return (!lapic_in_kernel(vcpu) ||
2792 kvm_apic_accept_pic_intr(vcpu));
2796 * if userspace requested an interrupt window, check that the
2797 * interrupt window is open.
2799 * No need to exit to userspace if we already have an interrupt queued.
2801 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2803 return kvm_arch_interrupt_allowed(vcpu) &&
2804 !kvm_cpu_has_interrupt(vcpu) &&
2805 !kvm_event_needs_reinjection(vcpu) &&
2806 kvm_cpu_accept_dm_intr(vcpu);
2809 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2810 struct kvm_interrupt *irq)
2812 if (irq->irq >= KVM_NR_INTERRUPTS)
2815 if (!irqchip_in_kernel(vcpu->kvm)) {
2816 kvm_queue_interrupt(vcpu, irq->irq, false);
2817 kvm_make_request(KVM_REQ_EVENT, vcpu);
2822 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2823 * fail for in-kernel 8259.
2825 if (pic_in_kernel(vcpu->kvm))
2828 if (vcpu->arch.pending_external_vector != -1)
2831 vcpu->arch.pending_external_vector = irq->irq;
2832 kvm_make_request(KVM_REQ_EVENT, vcpu);
2836 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2838 kvm_inject_nmi(vcpu);
2843 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2845 kvm_make_request(KVM_REQ_SMI, vcpu);
2850 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2851 struct kvm_tpr_access_ctl *tac)
2855 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2859 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2863 unsigned bank_num = mcg_cap & 0xff, bank;
2866 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2868 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2871 vcpu->arch.mcg_cap = mcg_cap;
2872 /* Init IA32_MCG_CTL to all 1s */
2873 if (mcg_cap & MCG_CTL_P)
2874 vcpu->arch.mcg_ctl = ~(u64)0;
2875 /* Init IA32_MCi_CTL to all 1s */
2876 for (bank = 0; bank < bank_num; bank++)
2877 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2882 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2883 struct kvm_x86_mce *mce)
2885 u64 mcg_cap = vcpu->arch.mcg_cap;
2886 unsigned bank_num = mcg_cap & 0xff;
2887 u64 *banks = vcpu->arch.mce_banks;
2889 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2892 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2893 * reporting is disabled
2895 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2896 vcpu->arch.mcg_ctl != ~(u64)0)
2898 banks += 4 * mce->bank;
2900 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2901 * reporting is disabled for the bank
2903 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2905 if (mce->status & MCI_STATUS_UC) {
2906 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2907 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2908 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2911 if (banks[1] & MCI_STATUS_VAL)
2912 mce->status |= MCI_STATUS_OVER;
2913 banks[2] = mce->addr;
2914 banks[3] = mce->misc;
2915 vcpu->arch.mcg_status = mce->mcg_status;
2916 banks[1] = mce->status;
2917 kvm_queue_exception(vcpu, MC_VECTOR);
2918 } else if (!(banks[1] & MCI_STATUS_VAL)
2919 || !(banks[1] & MCI_STATUS_UC)) {
2920 if (banks[1] & MCI_STATUS_VAL)
2921 mce->status |= MCI_STATUS_OVER;
2922 banks[2] = mce->addr;
2923 banks[3] = mce->misc;
2924 banks[1] = mce->status;
2926 banks[1] |= MCI_STATUS_OVER;
2930 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2931 struct kvm_vcpu_events *events)
2934 events->exception.injected =
2935 vcpu->arch.exception.pending &&
2936 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2937 events->exception.nr = vcpu->arch.exception.nr;
2938 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2939 events->exception.pad = 0;
2940 events->exception.error_code = vcpu->arch.exception.error_code;
2942 events->interrupt.injected =
2943 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2944 events->interrupt.nr = vcpu->arch.interrupt.nr;
2945 events->interrupt.soft = 0;
2946 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2948 events->nmi.injected = vcpu->arch.nmi_injected;
2949 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2950 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2951 events->nmi.pad = 0;
2953 events->sipi_vector = 0; /* never valid when reporting to user space */
2955 events->smi.smm = is_smm(vcpu);
2956 events->smi.pending = vcpu->arch.smi_pending;
2957 events->smi.smm_inside_nmi =
2958 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2959 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2961 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2962 | KVM_VCPUEVENT_VALID_SHADOW
2963 | KVM_VCPUEVENT_VALID_SMM);
2964 memset(&events->reserved, 0, sizeof(events->reserved));
2967 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2968 struct kvm_vcpu_events *events)
2970 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2971 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2972 | KVM_VCPUEVENT_VALID_SHADOW
2973 | KVM_VCPUEVENT_VALID_SMM))
2976 if (events->exception.injected &&
2977 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
2981 vcpu->arch.exception.pending = events->exception.injected;
2982 vcpu->arch.exception.nr = events->exception.nr;
2983 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2984 vcpu->arch.exception.error_code = events->exception.error_code;
2986 vcpu->arch.interrupt.pending = events->interrupt.injected;
2987 vcpu->arch.interrupt.nr = events->interrupt.nr;
2988 vcpu->arch.interrupt.soft = events->interrupt.soft;
2989 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2990 kvm_x86_ops->set_interrupt_shadow(vcpu,
2991 events->interrupt.shadow);
2993 vcpu->arch.nmi_injected = events->nmi.injected;
2994 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2995 vcpu->arch.nmi_pending = events->nmi.pending;
2996 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2998 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2999 lapic_in_kernel(vcpu))
3000 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3002 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3003 if (events->smi.smm)
3004 vcpu->arch.hflags |= HF_SMM_MASK;
3006 vcpu->arch.hflags &= ~HF_SMM_MASK;
3007 vcpu->arch.smi_pending = events->smi.pending;
3008 if (events->smi.smm_inside_nmi)
3009 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3011 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3012 if (lapic_in_kernel(vcpu)) {
3013 if (events->smi.latched_init)
3014 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3016 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3020 kvm_make_request(KVM_REQ_EVENT, vcpu);
3025 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3026 struct kvm_debugregs *dbgregs)
3030 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3031 kvm_get_dr(vcpu, 6, &val);
3033 dbgregs->dr7 = vcpu->arch.dr7;
3035 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3038 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3039 struct kvm_debugregs *dbgregs)
3044 if (dbgregs->dr6 & ~0xffffffffull)
3046 if (dbgregs->dr7 & ~0xffffffffull)
3049 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3050 kvm_update_dr0123(vcpu);
3051 vcpu->arch.dr6 = dbgregs->dr6;
3052 kvm_update_dr6(vcpu);
3053 vcpu->arch.dr7 = dbgregs->dr7;
3054 kvm_update_dr7(vcpu);
3059 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3061 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3063 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3064 u64 xstate_bv = xsave->header.xfeatures;
3068 * Copy legacy XSAVE area, to avoid complications with CPUID
3069 * leaves 0 and 1 in the loop below.
3071 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3074 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3077 * Copy each region from the possibly compacted offset to the
3078 * non-compacted offset.
3080 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3082 u64 feature = valid & -valid;
3083 int index = fls64(feature) - 1;
3084 void *src = get_xsave_addr(xsave, feature);
3087 u32 size, offset, ecx, edx;
3088 cpuid_count(XSTATE_CPUID, index,
3089 &size, &offset, &ecx, &edx);
3090 memcpy(dest + offset, src, size);
3097 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3099 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3100 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3104 * Copy legacy XSAVE area, to avoid complications with CPUID
3105 * leaves 0 and 1 in the loop below.
3107 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3109 /* Set XSTATE_BV and possibly XCOMP_BV. */
3110 xsave->header.xfeatures = xstate_bv;
3111 if (boot_cpu_has(X86_FEATURE_XSAVES))
3112 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3115 * Copy each region from the non-compacted offset to the
3116 * possibly compacted offset.
3118 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3120 u64 feature = valid & -valid;
3121 int index = fls64(feature) - 1;
3122 void *dest = get_xsave_addr(xsave, feature);
3125 u32 size, offset, ecx, edx;
3126 cpuid_count(XSTATE_CPUID, index,
3127 &size, &offset, &ecx, &edx);
3128 memcpy(dest, src + offset, size);
3135 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3136 struct kvm_xsave *guest_xsave)
3138 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3139 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3140 fill_xsave((u8 *) guest_xsave->region, vcpu);
3142 memcpy(guest_xsave->region,
3143 &vcpu->arch.guest_fpu.state.fxsave,
3144 sizeof(struct fxregs_state));
3145 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3146 XFEATURE_MASK_FPSSE;
3150 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3151 struct kvm_xsave *guest_xsave)
3154 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3156 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3158 * Here we allow setting states that are not present in
3159 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3160 * with old userspace.
3162 if (xstate_bv & ~kvm_supported_xcr0())
3164 load_xsave(vcpu, (u8 *)guest_xsave->region);
3166 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3168 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3169 guest_xsave->region, sizeof(struct fxregs_state));
3174 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3175 struct kvm_xcrs *guest_xcrs)
3177 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3178 guest_xcrs->nr_xcrs = 0;
3182 guest_xcrs->nr_xcrs = 1;
3183 guest_xcrs->flags = 0;
3184 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3185 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3188 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3189 struct kvm_xcrs *guest_xcrs)
3193 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3196 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3199 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3200 /* Only support XCR0 currently */
3201 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3202 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3203 guest_xcrs->xcrs[i].value);
3212 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3213 * stopped by the hypervisor. This function will be called from the host only.
3214 * EINVAL is returned when the host attempts to set the flag for a guest that
3215 * does not support pv clocks.
3217 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3219 if (!vcpu->arch.pv_time_enabled)
3221 vcpu->arch.pvclock_set_guest_stopped_request = true;
3222 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3226 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3227 struct kvm_enable_cap *cap)
3233 case KVM_CAP_HYPERV_SYNIC:
3234 return kvm_hv_activate_synic(vcpu);
3240 long kvm_arch_vcpu_ioctl(struct file *filp,
3241 unsigned int ioctl, unsigned long arg)
3243 struct kvm_vcpu *vcpu = filp->private_data;
3244 void __user *argp = (void __user *)arg;
3247 struct kvm_lapic_state *lapic;
3248 struct kvm_xsave *xsave;
3249 struct kvm_xcrs *xcrs;
3255 case KVM_GET_LAPIC: {
3257 if (!lapic_in_kernel(vcpu))
3259 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3264 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3268 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3273 case KVM_SET_LAPIC: {
3275 if (!lapic_in_kernel(vcpu))
3277 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3278 if (IS_ERR(u.lapic))
3279 return PTR_ERR(u.lapic);
3281 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3284 case KVM_INTERRUPT: {
3285 struct kvm_interrupt irq;
3288 if (copy_from_user(&irq, argp, sizeof irq))
3290 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3294 r = kvm_vcpu_ioctl_nmi(vcpu);
3298 r = kvm_vcpu_ioctl_smi(vcpu);
3301 case KVM_SET_CPUID: {
3302 struct kvm_cpuid __user *cpuid_arg = argp;
3303 struct kvm_cpuid cpuid;
3306 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3308 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3311 case KVM_SET_CPUID2: {
3312 struct kvm_cpuid2 __user *cpuid_arg = argp;
3313 struct kvm_cpuid2 cpuid;
3316 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3318 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3319 cpuid_arg->entries);
3322 case KVM_GET_CPUID2: {
3323 struct kvm_cpuid2 __user *cpuid_arg = argp;
3324 struct kvm_cpuid2 cpuid;
3327 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3329 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3330 cpuid_arg->entries);
3334 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3340 r = msr_io(vcpu, argp, do_get_msr, 1);
3343 r = msr_io(vcpu, argp, do_set_msr, 0);
3345 case KVM_TPR_ACCESS_REPORTING: {
3346 struct kvm_tpr_access_ctl tac;
3349 if (copy_from_user(&tac, argp, sizeof tac))
3351 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3355 if (copy_to_user(argp, &tac, sizeof tac))
3360 case KVM_SET_VAPIC_ADDR: {
3361 struct kvm_vapic_addr va;
3364 if (!lapic_in_kernel(vcpu))
3367 if (copy_from_user(&va, argp, sizeof va))
3369 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3372 case KVM_X86_SETUP_MCE: {
3376 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3378 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3381 case KVM_X86_SET_MCE: {
3382 struct kvm_x86_mce mce;
3385 if (copy_from_user(&mce, argp, sizeof mce))
3387 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3390 case KVM_GET_VCPU_EVENTS: {
3391 struct kvm_vcpu_events events;
3393 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3396 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3401 case KVM_SET_VCPU_EVENTS: {
3402 struct kvm_vcpu_events events;
3405 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3408 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3411 case KVM_GET_DEBUGREGS: {
3412 struct kvm_debugregs dbgregs;
3414 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3417 if (copy_to_user(argp, &dbgregs,
3418 sizeof(struct kvm_debugregs)))
3423 case KVM_SET_DEBUGREGS: {
3424 struct kvm_debugregs dbgregs;
3427 if (copy_from_user(&dbgregs, argp,
3428 sizeof(struct kvm_debugregs)))
3431 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3434 case KVM_GET_XSAVE: {
3435 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3440 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3443 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3448 case KVM_SET_XSAVE: {
3449 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3450 if (IS_ERR(u.xsave))
3451 return PTR_ERR(u.xsave);
3453 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3456 case KVM_GET_XCRS: {
3457 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3462 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3465 if (copy_to_user(argp, u.xcrs,
3466 sizeof(struct kvm_xcrs)))
3471 case KVM_SET_XCRS: {
3472 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3474 return PTR_ERR(u.xcrs);
3476 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3479 case KVM_SET_TSC_KHZ: {
3483 user_tsc_khz = (u32)arg;
3485 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3488 if (user_tsc_khz == 0)
3489 user_tsc_khz = tsc_khz;
3491 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3496 case KVM_GET_TSC_KHZ: {
3497 r = vcpu->arch.virtual_tsc_khz;
3500 case KVM_KVMCLOCK_CTRL: {
3501 r = kvm_set_guest_paused(vcpu);
3504 case KVM_ENABLE_CAP: {
3505 struct kvm_enable_cap cap;
3508 if (copy_from_user(&cap, argp, sizeof(cap)))
3510 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3521 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3523 return VM_FAULT_SIGBUS;
3526 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3530 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3532 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3536 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3539 kvm->arch.ept_identity_map_addr = ident_addr;
3543 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3544 u32 kvm_nr_mmu_pages)
3546 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3549 mutex_lock(&kvm->slots_lock);
3551 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3552 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3554 mutex_unlock(&kvm->slots_lock);
3558 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3560 return kvm->arch.n_max_mmu_pages;
3563 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3568 switch (chip->chip_id) {
3569 case KVM_IRQCHIP_PIC_MASTER:
3570 memcpy(&chip->chip.pic,
3571 &pic_irqchip(kvm)->pics[0],
3572 sizeof(struct kvm_pic_state));
3574 case KVM_IRQCHIP_PIC_SLAVE:
3575 memcpy(&chip->chip.pic,
3576 &pic_irqchip(kvm)->pics[1],
3577 sizeof(struct kvm_pic_state));
3579 case KVM_IRQCHIP_IOAPIC:
3580 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3589 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3594 switch (chip->chip_id) {
3595 case KVM_IRQCHIP_PIC_MASTER:
3596 spin_lock(&pic_irqchip(kvm)->lock);
3597 memcpy(&pic_irqchip(kvm)->pics[0],
3599 sizeof(struct kvm_pic_state));
3600 spin_unlock(&pic_irqchip(kvm)->lock);
3602 case KVM_IRQCHIP_PIC_SLAVE:
3603 spin_lock(&pic_irqchip(kvm)->lock);
3604 memcpy(&pic_irqchip(kvm)->pics[1],
3606 sizeof(struct kvm_pic_state));
3607 spin_unlock(&pic_irqchip(kvm)->lock);
3609 case KVM_IRQCHIP_IOAPIC:
3610 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3616 kvm_pic_update_irq(pic_irqchip(kvm));
3620 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3622 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3624 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3626 mutex_lock(&kps->lock);
3627 memcpy(ps, &kps->channels, sizeof(*ps));
3628 mutex_unlock(&kps->lock);
3632 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3635 struct kvm_pit *pit = kvm->arch.vpit;
3637 mutex_lock(&pit->pit_state.lock);
3638 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3639 for (i = 0; i < 3; i++)
3640 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3641 mutex_unlock(&pit->pit_state.lock);
3645 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3647 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3648 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3649 sizeof(ps->channels));
3650 ps->flags = kvm->arch.vpit->pit_state.flags;
3651 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3652 memset(&ps->reserved, 0, sizeof(ps->reserved));
3656 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3660 u32 prev_legacy, cur_legacy;
3661 struct kvm_pit *pit = kvm->arch.vpit;
3663 mutex_lock(&pit->pit_state.lock);
3664 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3665 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3666 if (!prev_legacy && cur_legacy)
3668 memcpy(&pit->pit_state.channels, &ps->channels,
3669 sizeof(pit->pit_state.channels));
3670 pit->pit_state.flags = ps->flags;
3671 for (i = 0; i < 3; i++)
3672 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3674 mutex_unlock(&pit->pit_state.lock);
3678 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3679 struct kvm_reinject_control *control)
3681 struct kvm_pit *pit = kvm->arch.vpit;
3686 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3687 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3688 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3690 mutex_lock(&pit->pit_state.lock);
3691 kvm_pit_set_reinject(pit, control->pit_reinject);
3692 mutex_unlock(&pit->pit_state.lock);
3698 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3699 * @kvm: kvm instance
3700 * @log: slot id and address to which we copy the log
3702 * Steps 1-4 below provide general overview of dirty page logging. See
3703 * kvm_get_dirty_log_protect() function description for additional details.
3705 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3706 * always flush the TLB (step 4) even if previous step failed and the dirty
3707 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3708 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3709 * writes will be marked dirty for next log read.
3711 * 1. Take a snapshot of the bit and clear it if needed.
3712 * 2. Write protect the corresponding page.
3713 * 3. Copy the snapshot to the userspace.
3714 * 4. Flush TLB's if needed.
3716 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3718 bool is_dirty = false;
3721 mutex_lock(&kvm->slots_lock);
3724 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3726 if (kvm_x86_ops->flush_log_dirty)
3727 kvm_x86_ops->flush_log_dirty(kvm);
3729 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3732 * All the TLBs can be flushed out of mmu lock, see the comments in
3733 * kvm_mmu_slot_remove_write_access().
3735 lockdep_assert_held(&kvm->slots_lock);
3737 kvm_flush_remote_tlbs(kvm);
3739 mutex_unlock(&kvm->slots_lock);
3743 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3746 if (!irqchip_in_kernel(kvm))
3749 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3750 irq_event->irq, irq_event->level,
3755 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3756 struct kvm_enable_cap *cap)
3764 case KVM_CAP_DISABLE_QUIRKS:
3765 kvm->arch.disabled_quirks = cap->args[0];
3768 case KVM_CAP_SPLIT_IRQCHIP: {
3769 mutex_lock(&kvm->lock);
3771 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3772 goto split_irqchip_unlock;
3774 if (irqchip_in_kernel(kvm))
3775 goto split_irqchip_unlock;
3776 if (atomic_read(&kvm->online_vcpus))
3777 goto split_irqchip_unlock;
3778 r = kvm_setup_empty_irq_routing(kvm);
3780 goto split_irqchip_unlock;
3781 /* Pairs with irqchip_in_kernel. */
3783 kvm->arch.irqchip_split = true;
3784 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3786 split_irqchip_unlock:
3787 mutex_unlock(&kvm->lock);
3797 long kvm_arch_vm_ioctl(struct file *filp,
3798 unsigned int ioctl, unsigned long arg)
3800 struct kvm *kvm = filp->private_data;
3801 void __user *argp = (void __user *)arg;
3804 * This union makes it completely explicit to gcc-3.x
3805 * that these two variables' stack usage should be
3806 * combined, not added together.
3809 struct kvm_pit_state ps;
3810 struct kvm_pit_state2 ps2;
3811 struct kvm_pit_config pit_config;
3815 case KVM_SET_TSS_ADDR:
3816 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3818 case KVM_SET_IDENTITY_MAP_ADDR: {
3822 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3824 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3827 case KVM_SET_NR_MMU_PAGES:
3828 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3830 case KVM_GET_NR_MMU_PAGES:
3831 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3833 case KVM_CREATE_IRQCHIP: {
3834 struct kvm_pic *vpic;
3836 mutex_lock(&kvm->lock);
3839 goto create_irqchip_unlock;
3841 if (atomic_read(&kvm->online_vcpus))
3842 goto create_irqchip_unlock;
3844 vpic = kvm_create_pic(kvm);
3846 r = kvm_ioapic_init(kvm);
3848 mutex_lock(&kvm->slots_lock);
3849 kvm_destroy_pic(vpic);
3850 mutex_unlock(&kvm->slots_lock);
3851 goto create_irqchip_unlock;
3854 goto create_irqchip_unlock;
3855 r = kvm_setup_default_irq_routing(kvm);
3857 mutex_lock(&kvm->slots_lock);
3858 mutex_lock(&kvm->irq_lock);
3859 kvm_ioapic_destroy(kvm);
3860 kvm_destroy_pic(vpic);
3861 mutex_unlock(&kvm->irq_lock);
3862 mutex_unlock(&kvm->slots_lock);
3863 goto create_irqchip_unlock;
3865 /* Write kvm->irq_routing before kvm->arch.vpic. */
3867 kvm->arch.vpic = vpic;
3868 create_irqchip_unlock:
3869 mutex_unlock(&kvm->lock);
3872 case KVM_CREATE_PIT:
3873 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3875 case KVM_CREATE_PIT2:
3877 if (copy_from_user(&u.pit_config, argp,
3878 sizeof(struct kvm_pit_config)))
3881 mutex_lock(&kvm->slots_lock);
3884 goto create_pit_unlock;
3886 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3890 mutex_unlock(&kvm->slots_lock);
3892 case KVM_GET_IRQCHIP: {
3893 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3894 struct kvm_irqchip *chip;
3896 chip = memdup_user(argp, sizeof(*chip));
3903 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3904 goto get_irqchip_out;
3905 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3907 goto get_irqchip_out;
3909 if (copy_to_user(argp, chip, sizeof *chip))
3910 goto get_irqchip_out;
3916 case KVM_SET_IRQCHIP: {
3917 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3918 struct kvm_irqchip *chip;
3920 chip = memdup_user(argp, sizeof(*chip));
3927 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3928 goto set_irqchip_out;
3929 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3931 goto set_irqchip_out;
3939 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3942 if (!kvm->arch.vpit)
3944 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3948 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3955 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3958 if (!kvm->arch.vpit)
3960 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3963 case KVM_GET_PIT2: {
3965 if (!kvm->arch.vpit)
3967 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3971 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3976 case KVM_SET_PIT2: {
3978 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3981 if (!kvm->arch.vpit)
3983 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3986 case KVM_REINJECT_CONTROL: {
3987 struct kvm_reinject_control control;
3989 if (copy_from_user(&control, argp, sizeof(control)))
3991 r = kvm_vm_ioctl_reinject(kvm, &control);
3994 case KVM_SET_BOOT_CPU_ID:
3996 mutex_lock(&kvm->lock);
3997 if (atomic_read(&kvm->online_vcpus) != 0)
4000 kvm->arch.bsp_vcpu_id = arg;
4001 mutex_unlock(&kvm->lock);
4003 case KVM_XEN_HVM_CONFIG: {
4005 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4006 sizeof(struct kvm_xen_hvm_config)))
4009 if (kvm->arch.xen_hvm_config.flags)
4014 case KVM_SET_CLOCK: {
4015 struct kvm_clock_data user_ns;
4020 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4028 local_irq_disable();
4029 now_ns = get_kernel_ns();
4030 delta = user_ns.clock - now_ns;
4032 kvm->arch.kvmclock_offset = delta;
4033 kvm_gen_update_masterclock(kvm);
4036 case KVM_GET_CLOCK: {
4037 struct kvm_clock_data user_ns;
4040 local_irq_disable();
4041 now_ns = get_kernel_ns();
4042 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4045 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4048 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4053 case KVM_ENABLE_CAP: {
4054 struct kvm_enable_cap cap;
4057 if (copy_from_user(&cap, argp, sizeof(cap)))
4059 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4063 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4069 static void kvm_init_msr_list(void)
4074 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4075 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4079 * Even MSRs that are valid in the host may not be exposed
4080 * to the guests in some cases.
4082 switch (msrs_to_save[i]) {
4083 case MSR_IA32_BNDCFGS:
4084 if (!kvm_x86_ops->mpx_supported())
4088 if (!kvm_x86_ops->rdtscp_supported())
4096 msrs_to_save[j] = msrs_to_save[i];
4099 num_msrs_to_save = j;
4101 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4102 switch (emulated_msrs[i]) {
4103 case MSR_IA32_SMBASE:
4104 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4112 emulated_msrs[j] = emulated_msrs[i];
4115 num_emulated_msrs = j;
4118 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4126 if (!(lapic_in_kernel(vcpu) &&
4127 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4128 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4139 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4146 if (!(lapic_in_kernel(vcpu) &&
4147 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4149 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4151 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4161 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4162 struct kvm_segment *var, int seg)
4164 kvm_x86_ops->set_segment(vcpu, var, seg);
4167 void kvm_get_segment(struct kvm_vcpu *vcpu,
4168 struct kvm_segment *var, int seg)
4170 kvm_x86_ops->get_segment(vcpu, var, seg);
4173 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4174 struct x86_exception *exception)
4178 BUG_ON(!mmu_is_nested(vcpu));
4180 /* NPT walks are always user-walks */
4181 access |= PFERR_USER_MASK;
4182 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4187 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4188 struct x86_exception *exception)
4190 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4191 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4194 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4195 struct x86_exception *exception)
4197 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4198 access |= PFERR_FETCH_MASK;
4199 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4202 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4203 struct x86_exception *exception)
4205 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4206 access |= PFERR_WRITE_MASK;
4207 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4210 /* uses this to access any guest's mapped memory without checking CPL */
4211 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4212 struct x86_exception *exception)
4214 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4217 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4218 struct kvm_vcpu *vcpu, u32 access,
4219 struct x86_exception *exception)
4222 int r = X86EMUL_CONTINUE;
4225 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4227 unsigned offset = addr & (PAGE_SIZE-1);
4228 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4231 if (gpa == UNMAPPED_GVA)
4232 return X86EMUL_PROPAGATE_FAULT;
4233 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4236 r = X86EMUL_IO_NEEDED;
4248 /* used for instruction fetching */
4249 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4250 gva_t addr, void *val, unsigned int bytes,
4251 struct x86_exception *exception)
4253 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4254 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4258 /* Inline kvm_read_guest_virt_helper for speed. */
4259 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4261 if (unlikely(gpa == UNMAPPED_GVA))
4262 return X86EMUL_PROPAGATE_FAULT;
4264 offset = addr & (PAGE_SIZE-1);
4265 if (WARN_ON(offset + bytes > PAGE_SIZE))
4266 bytes = (unsigned)PAGE_SIZE - offset;
4267 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4269 if (unlikely(ret < 0))
4270 return X86EMUL_IO_NEEDED;
4272 return X86EMUL_CONTINUE;
4275 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4276 gva_t addr, void *val, unsigned int bytes,
4277 struct x86_exception *exception)
4279 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4280 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4282 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4285 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4287 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4288 gva_t addr, void *val, unsigned int bytes,
4289 struct x86_exception *exception)
4291 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4292 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4295 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4296 unsigned long addr, void *val, unsigned int bytes)
4298 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4299 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4301 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4304 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4305 gva_t addr, void *val,
4307 struct x86_exception *exception)
4309 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4311 int r = X86EMUL_CONTINUE;
4314 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4317 unsigned offset = addr & (PAGE_SIZE-1);
4318 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4321 if (gpa == UNMAPPED_GVA)
4322 return X86EMUL_PROPAGATE_FAULT;
4323 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4325 r = X86EMUL_IO_NEEDED;
4336 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4338 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4339 gpa_t *gpa, struct x86_exception *exception,
4342 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4343 | (write ? PFERR_WRITE_MASK : 0);
4346 * currently PKRU is only applied to ept enabled guest so
4347 * there is no pkey in EPT page table for L1 guest or EPT
4348 * shadow page table for L2 guest.
4350 if (vcpu_match_mmio_gva(vcpu, gva)
4351 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4352 vcpu->arch.access, 0, access)) {
4353 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4354 (gva & (PAGE_SIZE - 1));
4355 trace_vcpu_match_mmio(gva, *gpa, write, false);
4359 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4361 if (*gpa == UNMAPPED_GVA)
4364 /* For APIC access vmexit */
4365 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4368 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4369 trace_vcpu_match_mmio(gva, *gpa, write, true);
4376 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4377 const void *val, int bytes)
4381 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4384 kvm_page_track_write(vcpu, gpa, val, bytes);
4388 struct read_write_emulator_ops {
4389 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4391 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4392 void *val, int bytes);
4393 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4394 int bytes, void *val);
4395 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4396 void *val, int bytes);
4400 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4402 if (vcpu->mmio_read_completed) {
4403 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4404 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4405 vcpu->mmio_read_completed = 0;
4412 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4413 void *val, int bytes)
4415 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4418 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4419 void *val, int bytes)
4421 return emulator_write_phys(vcpu, gpa, val, bytes);
4424 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4426 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4427 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4430 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4431 void *val, int bytes)
4433 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4434 return X86EMUL_IO_NEEDED;
4437 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4438 void *val, int bytes)
4440 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4442 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4443 return X86EMUL_CONTINUE;
4446 static const struct read_write_emulator_ops read_emultor = {
4447 .read_write_prepare = read_prepare,
4448 .read_write_emulate = read_emulate,
4449 .read_write_mmio = vcpu_mmio_read,
4450 .read_write_exit_mmio = read_exit_mmio,
4453 static const struct read_write_emulator_ops write_emultor = {
4454 .read_write_emulate = write_emulate,
4455 .read_write_mmio = write_mmio,
4456 .read_write_exit_mmio = write_exit_mmio,
4460 static int emulator_read_write_onepage(unsigned long addr, void *val,
4462 struct x86_exception *exception,
4463 struct kvm_vcpu *vcpu,
4464 const struct read_write_emulator_ops *ops)
4468 bool write = ops->write;
4469 struct kvm_mmio_fragment *frag;
4471 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4474 return X86EMUL_PROPAGATE_FAULT;
4476 /* For APIC access vmexit */
4480 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4481 return X86EMUL_CONTINUE;
4485 * Is this MMIO handled locally?
4487 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4488 if (handled == bytes)
4489 return X86EMUL_CONTINUE;
4495 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4496 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4500 return X86EMUL_CONTINUE;
4503 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4505 void *val, unsigned int bytes,
4506 struct x86_exception *exception,
4507 const struct read_write_emulator_ops *ops)
4509 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4513 if (ops->read_write_prepare &&
4514 ops->read_write_prepare(vcpu, val, bytes))
4515 return X86EMUL_CONTINUE;
4517 vcpu->mmio_nr_fragments = 0;
4519 /* Crossing a page boundary? */
4520 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4523 now = -addr & ~PAGE_MASK;
4524 rc = emulator_read_write_onepage(addr, val, now, exception,
4527 if (rc != X86EMUL_CONTINUE)
4530 if (ctxt->mode != X86EMUL_MODE_PROT64)
4536 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4538 if (rc != X86EMUL_CONTINUE)
4541 if (!vcpu->mmio_nr_fragments)
4544 gpa = vcpu->mmio_fragments[0].gpa;
4546 vcpu->mmio_needed = 1;
4547 vcpu->mmio_cur_fragment = 0;
4549 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4550 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4551 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4552 vcpu->run->mmio.phys_addr = gpa;
4554 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4557 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4561 struct x86_exception *exception)
4563 return emulator_read_write(ctxt, addr, val, bytes,
4564 exception, &read_emultor);
4567 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4571 struct x86_exception *exception)
4573 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4574 exception, &write_emultor);
4577 #define CMPXCHG_TYPE(t, ptr, old, new) \
4578 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4580 #ifdef CONFIG_X86_64
4581 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4583 # define CMPXCHG64(ptr, old, new) \
4584 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4587 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4592 struct x86_exception *exception)
4594 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4600 /* guests cmpxchg8b have to be emulated atomically */
4601 if (bytes > 8 || (bytes & (bytes - 1)))
4604 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4606 if (gpa == UNMAPPED_GVA ||
4607 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4610 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4613 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4614 if (is_error_page(page))
4617 kaddr = kmap_atomic(page);
4618 kaddr += offset_in_page(gpa);
4621 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4624 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4627 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4630 exchanged = CMPXCHG64(kaddr, old, new);
4635 kunmap_atomic(kaddr);
4636 kvm_release_page_dirty(page);
4639 return X86EMUL_CMPXCHG_FAILED;
4641 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4642 kvm_page_track_write(vcpu, gpa, new, bytes);
4644 return X86EMUL_CONTINUE;
4647 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4649 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4652 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4654 /* TODO: String I/O for in kernel device */
4657 if (vcpu->arch.pio.in)
4658 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4659 vcpu->arch.pio.size, pd);
4661 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4662 vcpu->arch.pio.port, vcpu->arch.pio.size,
4667 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4668 unsigned short port, void *val,
4669 unsigned int count, bool in)
4671 vcpu->arch.pio.port = port;
4672 vcpu->arch.pio.in = in;
4673 vcpu->arch.pio.count = count;
4674 vcpu->arch.pio.size = size;
4676 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4677 vcpu->arch.pio.count = 0;
4681 vcpu->run->exit_reason = KVM_EXIT_IO;
4682 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4683 vcpu->run->io.size = size;
4684 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4685 vcpu->run->io.count = count;
4686 vcpu->run->io.port = port;
4691 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4692 int size, unsigned short port, void *val,
4695 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4698 if (vcpu->arch.pio.count)
4701 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4704 memcpy(val, vcpu->arch.pio_data, size * count);
4705 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4706 vcpu->arch.pio.count = 0;
4713 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4714 int size, unsigned short port,
4715 const void *val, unsigned int count)
4717 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4719 memcpy(vcpu->arch.pio_data, val, size * count);
4720 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4721 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4724 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4726 return kvm_x86_ops->get_segment_base(vcpu, seg);
4729 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4731 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4734 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4736 if (!need_emulate_wbinvd(vcpu))
4737 return X86EMUL_CONTINUE;
4739 if (kvm_x86_ops->has_wbinvd_exit()) {
4740 int cpu = get_cpu();
4742 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4743 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4744 wbinvd_ipi, NULL, 1);
4746 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4749 return X86EMUL_CONTINUE;
4752 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4754 kvm_x86_ops->skip_emulated_instruction(vcpu);
4755 return kvm_emulate_wbinvd_noskip(vcpu);
4757 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4761 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4763 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4766 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4767 unsigned long *dest)
4769 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4772 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4773 unsigned long value)
4776 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4779 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4781 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4784 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4786 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4787 unsigned long value;
4791 value = kvm_read_cr0(vcpu);
4794 value = vcpu->arch.cr2;
4797 value = kvm_read_cr3(vcpu);
4800 value = kvm_read_cr4(vcpu);
4803 value = kvm_get_cr8(vcpu);
4806 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4813 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4815 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4820 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4823 vcpu->arch.cr2 = val;
4826 res = kvm_set_cr3(vcpu, val);
4829 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4832 res = kvm_set_cr8(vcpu, val);
4835 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4842 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4844 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4847 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4849 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4852 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4854 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4857 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4859 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4862 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4864 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4867 static unsigned long emulator_get_cached_segment_base(
4868 struct x86_emulate_ctxt *ctxt, int seg)
4870 return get_segment_base(emul_to_vcpu(ctxt), seg);
4873 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4874 struct desc_struct *desc, u32 *base3,
4877 struct kvm_segment var;
4879 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4880 *selector = var.selector;
4883 memset(desc, 0, sizeof(*desc));
4889 set_desc_limit(desc, var.limit);
4890 set_desc_base(desc, (unsigned long)var.base);
4891 #ifdef CONFIG_X86_64
4893 *base3 = var.base >> 32;
4895 desc->type = var.type;
4897 desc->dpl = var.dpl;
4898 desc->p = var.present;
4899 desc->avl = var.avl;
4907 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4908 struct desc_struct *desc, u32 base3,
4911 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4912 struct kvm_segment var;
4914 var.selector = selector;
4915 var.base = get_desc_base(desc);
4916 #ifdef CONFIG_X86_64
4917 var.base |= ((u64)base3) << 32;
4919 var.limit = get_desc_limit(desc);
4921 var.limit = (var.limit << 12) | 0xfff;
4922 var.type = desc->type;
4923 var.dpl = desc->dpl;
4928 var.avl = desc->avl;
4929 var.present = desc->p;
4930 var.unusable = !var.present;
4933 kvm_set_segment(vcpu, &var, seg);
4937 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4938 u32 msr_index, u64 *pdata)
4940 struct msr_data msr;
4943 msr.index = msr_index;
4944 msr.host_initiated = false;
4945 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4953 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4954 u32 msr_index, u64 data)
4956 struct msr_data msr;
4959 msr.index = msr_index;
4960 msr.host_initiated = false;
4961 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4964 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4966 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4968 return vcpu->arch.smbase;
4971 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4973 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4975 vcpu->arch.smbase = smbase;
4978 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4981 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4984 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4985 u32 pmc, u64 *pdata)
4987 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4990 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4992 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4995 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4998 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5000 * CR0.TS may reference the host fpu state, not the guest fpu state,
5001 * so it may be clear at this point.
5006 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5011 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5012 struct x86_instruction_info *info,
5013 enum x86_intercept_stage stage)
5015 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5018 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5019 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5021 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5024 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5026 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5029 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5031 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5034 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5036 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5039 static const struct x86_emulate_ops emulate_ops = {
5040 .read_gpr = emulator_read_gpr,
5041 .write_gpr = emulator_write_gpr,
5042 .read_std = kvm_read_guest_virt_system,
5043 .write_std = kvm_write_guest_virt_system,
5044 .read_phys = kvm_read_guest_phys_system,
5045 .fetch = kvm_fetch_guest_virt,
5046 .read_emulated = emulator_read_emulated,
5047 .write_emulated = emulator_write_emulated,
5048 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5049 .invlpg = emulator_invlpg,
5050 .pio_in_emulated = emulator_pio_in_emulated,
5051 .pio_out_emulated = emulator_pio_out_emulated,
5052 .get_segment = emulator_get_segment,
5053 .set_segment = emulator_set_segment,
5054 .get_cached_segment_base = emulator_get_cached_segment_base,
5055 .get_gdt = emulator_get_gdt,
5056 .get_idt = emulator_get_idt,
5057 .set_gdt = emulator_set_gdt,
5058 .set_idt = emulator_set_idt,
5059 .get_cr = emulator_get_cr,
5060 .set_cr = emulator_set_cr,
5061 .cpl = emulator_get_cpl,
5062 .get_dr = emulator_get_dr,
5063 .set_dr = emulator_set_dr,
5064 .get_smbase = emulator_get_smbase,
5065 .set_smbase = emulator_set_smbase,
5066 .set_msr = emulator_set_msr,
5067 .get_msr = emulator_get_msr,
5068 .check_pmc = emulator_check_pmc,
5069 .read_pmc = emulator_read_pmc,
5070 .halt = emulator_halt,
5071 .wbinvd = emulator_wbinvd,
5072 .fix_hypercall = emulator_fix_hypercall,
5073 .get_fpu = emulator_get_fpu,
5074 .put_fpu = emulator_put_fpu,
5075 .intercept = emulator_intercept,
5076 .get_cpuid = emulator_get_cpuid,
5077 .set_nmi_mask = emulator_set_nmi_mask,
5080 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5082 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5084 * an sti; sti; sequence only disable interrupts for the first
5085 * instruction. So, if the last instruction, be it emulated or
5086 * not, left the system with the INT_STI flag enabled, it
5087 * means that the last instruction is an sti. We should not
5088 * leave the flag on in this case. The same goes for mov ss
5090 if (int_shadow & mask)
5092 if (unlikely(int_shadow || mask)) {
5093 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5095 kvm_make_request(KVM_REQ_EVENT, vcpu);
5099 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5101 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5102 if (ctxt->exception.vector == PF_VECTOR)
5103 return kvm_propagate_fault(vcpu, &ctxt->exception);
5105 if (ctxt->exception.error_code_valid)
5106 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5107 ctxt->exception.error_code);
5109 kvm_queue_exception(vcpu, ctxt->exception.vector);
5113 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5115 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5118 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5120 ctxt->eflags = kvm_get_rflags(vcpu);
5121 ctxt->eip = kvm_rip_read(vcpu);
5122 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5123 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5124 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5125 cs_db ? X86EMUL_MODE_PROT32 :
5126 X86EMUL_MODE_PROT16;
5127 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5128 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5129 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5130 ctxt->emul_flags = vcpu->arch.hflags;
5132 init_decode_cache(ctxt);
5133 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5136 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5138 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5141 init_emulate_ctxt(vcpu);
5145 ctxt->_eip = ctxt->eip + inc_eip;
5146 ret = emulate_int_real(ctxt, irq);
5148 if (ret != X86EMUL_CONTINUE)
5149 return EMULATE_FAIL;
5151 ctxt->eip = ctxt->_eip;
5152 kvm_rip_write(vcpu, ctxt->eip);
5153 kvm_set_rflags(vcpu, ctxt->eflags);
5155 if (irq == NMI_VECTOR)
5156 vcpu->arch.nmi_pending = 0;
5158 vcpu->arch.interrupt.pending = false;
5160 return EMULATE_DONE;
5162 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5164 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5166 int r = EMULATE_DONE;
5168 ++vcpu->stat.insn_emulation_fail;
5169 trace_kvm_emulate_insn_failed(vcpu);
5170 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5171 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5172 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5173 vcpu->run->internal.ndata = 0;
5176 kvm_queue_exception(vcpu, UD_VECTOR);
5181 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5182 bool write_fault_to_shadow_pgtable,
5188 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5191 if (!vcpu->arch.mmu.direct_map) {
5193 * Write permission should be allowed since only
5194 * write access need to be emulated.
5196 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5199 * If the mapping is invalid in guest, let cpu retry
5200 * it to generate fault.
5202 if (gpa == UNMAPPED_GVA)
5207 * Do not retry the unhandleable instruction if it faults on the
5208 * readonly host memory, otherwise it will goto a infinite loop:
5209 * retry instruction -> write #PF -> emulation fail -> retry
5210 * instruction -> ...
5212 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5215 * If the instruction failed on the error pfn, it can not be fixed,
5216 * report the error to userspace.
5218 if (is_error_noslot_pfn(pfn))
5221 kvm_release_pfn_clean(pfn);
5223 /* The instructions are well-emulated on direct mmu. */
5224 if (vcpu->arch.mmu.direct_map) {
5225 unsigned int indirect_shadow_pages;
5227 spin_lock(&vcpu->kvm->mmu_lock);
5228 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5229 spin_unlock(&vcpu->kvm->mmu_lock);
5231 if (indirect_shadow_pages)
5232 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5238 * if emulation was due to access to shadowed page table
5239 * and it failed try to unshadow page and re-enter the
5240 * guest to let CPU execute the instruction.
5242 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5245 * If the access faults on its page table, it can not
5246 * be fixed by unprotecting shadow page and it should
5247 * be reported to userspace.
5249 return !write_fault_to_shadow_pgtable;
5252 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5253 unsigned long cr2, int emulation_type)
5255 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5256 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5258 last_retry_eip = vcpu->arch.last_retry_eip;
5259 last_retry_addr = vcpu->arch.last_retry_addr;
5262 * If the emulation is caused by #PF and it is non-page_table
5263 * writing instruction, it means the VM-EXIT is caused by shadow
5264 * page protected, we can zap the shadow page and retry this
5265 * instruction directly.
5267 * Note: if the guest uses a non-page-table modifying instruction
5268 * on the PDE that points to the instruction, then we will unmap
5269 * the instruction and go to an infinite loop. So, we cache the
5270 * last retried eip and the last fault address, if we meet the eip
5271 * and the address again, we can break out of the potential infinite
5274 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5276 if (!(emulation_type & EMULTYPE_RETRY))
5279 if (x86_page_table_writing_insn(ctxt))
5282 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5285 vcpu->arch.last_retry_eip = ctxt->eip;
5286 vcpu->arch.last_retry_addr = cr2;
5288 if (!vcpu->arch.mmu.direct_map)
5289 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5291 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5296 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5297 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5299 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5301 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5302 /* This is a good place to trace that we are exiting SMM. */
5303 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5305 if (unlikely(vcpu->arch.smi_pending)) {
5306 kvm_make_request(KVM_REQ_SMI, vcpu);
5307 vcpu->arch.smi_pending = 0;
5309 /* Process a latched INIT, if any. */
5310 kvm_make_request(KVM_REQ_EVENT, vcpu);
5314 kvm_mmu_reset_context(vcpu);
5317 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5319 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5321 vcpu->arch.hflags = emul_flags;
5323 if (changed & HF_SMM_MASK)
5324 kvm_smm_changed(vcpu);
5327 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5336 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5337 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5342 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5344 struct kvm_run *kvm_run = vcpu->run;
5347 * rflags is the old, "raw" value of the flags. The new value has
5348 * not been saved yet.
5350 * This is correct even for TF set by the guest, because "the
5351 * processor will not generate this exception after the instruction
5352 * that sets the TF flag".
5354 if (unlikely(rflags & X86_EFLAGS_TF)) {
5355 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5356 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5358 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5359 kvm_run->debug.arch.exception = DB_VECTOR;
5360 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5361 *r = EMULATE_USER_EXIT;
5363 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5365 * "Certain debug exceptions may clear bit 0-3. The
5366 * remaining contents of the DR6 register are never
5367 * cleared by the processor".
5369 vcpu->arch.dr6 &= ~15;
5370 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5371 kvm_queue_exception(vcpu, DB_VECTOR);
5376 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5378 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5379 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5380 struct kvm_run *kvm_run = vcpu->run;
5381 unsigned long eip = kvm_get_linear_rip(vcpu);
5382 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5383 vcpu->arch.guest_debug_dr7,
5387 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5388 kvm_run->debug.arch.pc = eip;
5389 kvm_run->debug.arch.exception = DB_VECTOR;
5390 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5391 *r = EMULATE_USER_EXIT;
5396 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5397 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5398 unsigned long eip = kvm_get_linear_rip(vcpu);
5399 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5404 vcpu->arch.dr6 &= ~15;
5405 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5406 kvm_queue_exception(vcpu, DB_VECTOR);
5415 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5422 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5423 bool writeback = true;
5424 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5427 * Clear write_fault_to_shadow_pgtable here to ensure it is
5430 vcpu->arch.write_fault_to_shadow_pgtable = false;
5431 kvm_clear_exception_queue(vcpu);
5433 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5434 init_emulate_ctxt(vcpu);
5437 * We will reenter on the same instruction since
5438 * we do not set complete_userspace_io. This does not
5439 * handle watchpoints yet, those would be handled in
5442 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5445 ctxt->interruptibility = 0;
5446 ctxt->have_exception = false;
5447 ctxt->exception.vector = -1;
5448 ctxt->perm_ok = false;
5450 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5452 r = x86_decode_insn(ctxt, insn, insn_len);
5454 trace_kvm_emulate_insn_start(vcpu);
5455 ++vcpu->stat.insn_emulation;
5456 if (r != EMULATION_OK) {
5457 if (emulation_type & EMULTYPE_TRAP_UD)
5458 return EMULATE_FAIL;
5459 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5461 return EMULATE_DONE;
5462 if (emulation_type & EMULTYPE_SKIP)
5463 return EMULATE_FAIL;
5464 return handle_emulation_failure(vcpu);
5468 if (emulation_type & EMULTYPE_SKIP) {
5469 kvm_rip_write(vcpu, ctxt->_eip);
5470 if (ctxt->eflags & X86_EFLAGS_RF)
5471 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5472 return EMULATE_DONE;
5475 if (retry_instruction(ctxt, cr2, emulation_type))
5476 return EMULATE_DONE;
5478 /* this is needed for vmware backdoor interface to work since it
5479 changes registers values during IO operation */
5480 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5481 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5482 emulator_invalidate_register_cache(ctxt);
5486 r = x86_emulate_insn(ctxt);
5488 if (r == EMULATION_INTERCEPTED)
5489 return EMULATE_DONE;
5491 if (r == EMULATION_FAILED) {
5492 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5494 return EMULATE_DONE;
5496 return handle_emulation_failure(vcpu);
5499 if (ctxt->have_exception) {
5501 if (inject_emulated_exception(vcpu))
5503 } else if (vcpu->arch.pio.count) {
5504 if (!vcpu->arch.pio.in) {
5505 /* FIXME: return into emulator if single-stepping. */
5506 vcpu->arch.pio.count = 0;
5509 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5511 r = EMULATE_USER_EXIT;
5512 } else if (vcpu->mmio_needed) {
5513 if (!vcpu->mmio_is_write)
5515 r = EMULATE_USER_EXIT;
5516 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5517 } else if (r == EMULATION_RESTART)
5523 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5524 toggle_interruptibility(vcpu, ctxt->interruptibility);
5525 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5526 if (vcpu->arch.hflags != ctxt->emul_flags)
5527 kvm_set_hflags(vcpu, ctxt->emul_flags);
5528 kvm_rip_write(vcpu, ctxt->eip);
5529 if (r == EMULATE_DONE)
5530 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5531 if (!ctxt->have_exception ||
5532 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5533 __kvm_set_rflags(vcpu, ctxt->eflags);
5536 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5537 * do nothing, and it will be requested again as soon as
5538 * the shadow expires. But we still need to check here,
5539 * because POPF has no interrupt shadow.
5541 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5542 kvm_make_request(KVM_REQ_EVENT, vcpu);
5544 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5548 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5550 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5552 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5553 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5554 size, port, &val, 1);
5555 /* do not return to emulator after return from userspace */
5556 vcpu->arch.pio.count = 0;
5559 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5561 static void tsc_bad(void *info)
5563 __this_cpu_write(cpu_tsc_khz, 0);
5566 static void tsc_khz_changed(void *data)
5568 struct cpufreq_freqs *freq = data;
5569 unsigned long khz = 0;
5573 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5574 khz = cpufreq_quick_get(raw_smp_processor_id());
5577 __this_cpu_write(cpu_tsc_khz, khz);
5580 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5583 struct cpufreq_freqs *freq = data;
5585 struct kvm_vcpu *vcpu;
5586 int i, send_ipi = 0;
5589 * We allow guests to temporarily run on slowing clocks,
5590 * provided we notify them after, or to run on accelerating
5591 * clocks, provided we notify them before. Thus time never
5594 * However, we have a problem. We can't atomically update
5595 * the frequency of a given CPU from this function; it is
5596 * merely a notifier, which can be called from any CPU.
5597 * Changing the TSC frequency at arbitrary points in time
5598 * requires a recomputation of local variables related to
5599 * the TSC for each VCPU. We must flag these local variables
5600 * to be updated and be sure the update takes place with the
5601 * new frequency before any guests proceed.
5603 * Unfortunately, the combination of hotplug CPU and frequency
5604 * change creates an intractable locking scenario; the order
5605 * of when these callouts happen is undefined with respect to
5606 * CPU hotplug, and they can race with each other. As such,
5607 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5608 * undefined; you can actually have a CPU frequency change take
5609 * place in between the computation of X and the setting of the
5610 * variable. To protect against this problem, all updates of
5611 * the per_cpu tsc_khz variable are done in an interrupt
5612 * protected IPI, and all callers wishing to update the value
5613 * must wait for a synchronous IPI to complete (which is trivial
5614 * if the caller is on the CPU already). This establishes the
5615 * necessary total order on variable updates.
5617 * Note that because a guest time update may take place
5618 * anytime after the setting of the VCPU's request bit, the
5619 * correct TSC value must be set before the request. However,
5620 * to ensure the update actually makes it to any guest which
5621 * starts running in hardware virtualization between the set
5622 * and the acquisition of the spinlock, we must also ping the
5623 * CPU after setting the request bit.
5627 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5629 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5632 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5634 spin_lock(&kvm_lock);
5635 list_for_each_entry(kvm, &vm_list, vm_list) {
5636 kvm_for_each_vcpu(i, vcpu, kvm) {
5637 if (vcpu->cpu != freq->cpu)
5639 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5640 if (vcpu->cpu != smp_processor_id())
5644 spin_unlock(&kvm_lock);
5646 if (freq->old < freq->new && send_ipi) {
5648 * We upscale the frequency. Must make the guest
5649 * doesn't see old kvmclock values while running with
5650 * the new frequency, otherwise we risk the guest sees
5651 * time go backwards.
5653 * In case we update the frequency for another cpu
5654 * (which might be in guest context) send an interrupt
5655 * to kick the cpu out of guest context. Next time
5656 * guest context is entered kvmclock will be updated,
5657 * so the guest will not see stale values.
5659 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5664 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5665 .notifier_call = kvmclock_cpufreq_notifier
5668 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5669 unsigned long action, void *hcpu)
5671 unsigned int cpu = (unsigned long)hcpu;
5675 case CPU_DOWN_FAILED:
5676 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5678 case CPU_DOWN_PREPARE:
5679 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5685 static struct notifier_block kvmclock_cpu_notifier_block = {
5686 .notifier_call = kvmclock_cpu_notifier,
5687 .priority = -INT_MAX
5690 static void kvm_timer_init(void)
5694 max_tsc_khz = tsc_khz;
5696 cpu_notifier_register_begin();
5697 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5698 #ifdef CONFIG_CPU_FREQ
5699 struct cpufreq_policy policy;
5700 memset(&policy, 0, sizeof(policy));
5702 cpufreq_get_policy(&policy, cpu);
5703 if (policy.cpuinfo.max_freq)
5704 max_tsc_khz = policy.cpuinfo.max_freq;
5707 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5708 CPUFREQ_TRANSITION_NOTIFIER);
5710 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5711 for_each_online_cpu(cpu)
5712 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5714 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5715 cpu_notifier_register_done();
5719 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5721 int kvm_is_in_guest(void)
5723 return __this_cpu_read(current_vcpu) != NULL;
5726 static int kvm_is_user_mode(void)
5730 if (__this_cpu_read(current_vcpu))
5731 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5733 return user_mode != 0;
5736 static unsigned long kvm_get_guest_ip(void)
5738 unsigned long ip = 0;
5740 if (__this_cpu_read(current_vcpu))
5741 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5746 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5747 .is_in_guest = kvm_is_in_guest,
5748 .is_user_mode = kvm_is_user_mode,
5749 .get_guest_ip = kvm_get_guest_ip,
5752 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5754 __this_cpu_write(current_vcpu, vcpu);
5756 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5758 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5760 __this_cpu_write(current_vcpu, NULL);
5762 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5764 static void kvm_set_mmio_spte_mask(void)
5767 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5770 * Set the reserved bits and the present bit of an paging-structure
5771 * entry to generate page fault with PFER.RSV = 1.
5773 /* Mask the reserved physical address bits. */
5774 mask = rsvd_bits(maxphyaddr, 51);
5776 /* Bit 62 is always reserved for 32bit host. */
5777 mask |= 0x3ull << 62;
5779 /* Set the present bit. */
5782 #ifdef CONFIG_X86_64
5784 * If reserved bit is not supported, clear the present bit to disable
5787 if (maxphyaddr == 52)
5791 kvm_mmu_set_mmio_spte_mask(mask);
5794 #ifdef CONFIG_X86_64
5795 static void pvclock_gtod_update_fn(struct work_struct *work)
5799 struct kvm_vcpu *vcpu;
5802 spin_lock(&kvm_lock);
5803 list_for_each_entry(kvm, &vm_list, vm_list)
5804 kvm_for_each_vcpu(i, vcpu, kvm)
5805 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5806 atomic_set(&kvm_guest_has_master_clock, 0);
5807 spin_unlock(&kvm_lock);
5810 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5813 * Notification about pvclock gtod data update.
5815 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5818 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5819 struct timekeeper *tk = priv;
5821 update_pvclock_gtod(tk);
5823 /* disable master clock if host does not trust, or does not
5824 * use, TSC clocksource
5826 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5827 atomic_read(&kvm_guest_has_master_clock) != 0)
5828 queue_work(system_long_wq, &pvclock_gtod_work);
5833 static struct notifier_block pvclock_gtod_notifier = {
5834 .notifier_call = pvclock_gtod_notify,
5838 int kvm_arch_init(void *opaque)
5841 struct kvm_x86_ops *ops = opaque;
5844 printk(KERN_ERR "kvm: already loaded the other module\n");
5849 if (!ops->cpu_has_kvm_support()) {
5850 printk(KERN_ERR "kvm: no hardware support\n");
5854 if (ops->disabled_by_bios()) {
5855 printk(KERN_ERR "kvm: disabled by bios\n");
5861 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5863 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5867 r = kvm_mmu_module_init();
5869 goto out_free_percpu;
5871 kvm_set_mmio_spte_mask();
5875 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5876 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5880 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5882 if (boot_cpu_has(X86_FEATURE_XSAVE))
5883 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5886 #ifdef CONFIG_X86_64
5887 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5893 free_percpu(shared_msrs);
5898 void kvm_arch_exit(void)
5900 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5902 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5903 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5904 CPUFREQ_TRANSITION_NOTIFIER);
5905 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5906 #ifdef CONFIG_X86_64
5907 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5910 kvm_mmu_module_exit();
5911 free_percpu(shared_msrs);
5914 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5916 ++vcpu->stat.halt_exits;
5917 if (lapic_in_kernel(vcpu)) {
5918 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5921 vcpu->run->exit_reason = KVM_EXIT_HLT;
5925 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5927 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5929 kvm_x86_ops->skip_emulated_instruction(vcpu);
5930 return kvm_vcpu_halt(vcpu);
5932 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5935 * kvm_pv_kick_cpu_op: Kick a vcpu.
5937 * @apicid - apicid of vcpu to be kicked.
5939 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5941 struct kvm_lapic_irq lapic_irq;
5943 lapic_irq.shorthand = 0;
5944 lapic_irq.dest_mode = 0;
5945 lapic_irq.dest_id = apicid;
5946 lapic_irq.msi_redir_hint = false;
5948 lapic_irq.delivery_mode = APIC_DM_REMRD;
5949 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5952 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
5954 vcpu->arch.apicv_active = false;
5955 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
5958 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5960 unsigned long nr, a0, a1, a2, a3, ret;
5961 int op_64_bit, r = 1;
5963 kvm_x86_ops->skip_emulated_instruction(vcpu);
5965 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5966 return kvm_hv_hypercall(vcpu);
5968 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5969 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5970 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5971 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5972 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5974 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5976 op_64_bit = is_64_bit_mode(vcpu);
5985 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5991 case KVM_HC_VAPIC_POLL_IRQ:
5994 case KVM_HC_KICK_CPU:
5995 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6005 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6006 ++vcpu->stat.hypercalls;
6009 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6011 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6013 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6014 char instruction[3];
6015 unsigned long rip = kvm_rip_read(vcpu);
6017 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6019 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
6022 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6024 return vcpu->run->request_interrupt_window &&
6025 likely(!pic_in_kernel(vcpu->kvm));
6028 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6030 struct kvm_run *kvm_run = vcpu->run;
6032 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6033 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6034 kvm_run->cr8 = kvm_get_cr8(vcpu);
6035 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6036 kvm_run->ready_for_interrupt_injection =
6037 pic_in_kernel(vcpu->kvm) ||
6038 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6041 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6045 if (!kvm_x86_ops->update_cr8_intercept)
6048 if (!lapic_in_kernel(vcpu))
6051 if (vcpu->arch.apicv_active)
6054 if (!vcpu->arch.apic->vapic_addr)
6055 max_irr = kvm_lapic_find_highest_irr(vcpu);
6062 tpr = kvm_lapic_get_cr8(vcpu);
6064 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6067 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6071 /* try to reinject previous events if any */
6072 if (vcpu->arch.exception.pending) {
6073 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6074 vcpu->arch.exception.has_error_code,
6075 vcpu->arch.exception.error_code);
6077 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6078 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6081 if (vcpu->arch.exception.nr == DB_VECTOR &&
6082 (vcpu->arch.dr7 & DR7_GD)) {
6083 vcpu->arch.dr7 &= ~DR7_GD;
6084 kvm_update_dr7(vcpu);
6087 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6088 vcpu->arch.exception.has_error_code,
6089 vcpu->arch.exception.error_code,
6090 vcpu->arch.exception.reinject);
6094 if (vcpu->arch.nmi_injected) {
6095 kvm_x86_ops->set_nmi(vcpu);
6099 if (vcpu->arch.interrupt.pending) {
6100 kvm_x86_ops->set_irq(vcpu);
6104 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6105 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6110 /* try to inject new event if pending */
6111 if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6112 --vcpu->arch.nmi_pending;
6113 vcpu->arch.nmi_injected = true;
6114 kvm_x86_ops->set_nmi(vcpu);
6115 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6117 * Because interrupts can be injected asynchronously, we are
6118 * calling check_nested_events again here to avoid a race condition.
6119 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6120 * proposal and current concerns. Perhaps we should be setting
6121 * KVM_REQ_EVENT only on certain events and not unconditionally?
6123 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6124 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6128 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6129 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6131 kvm_x86_ops->set_irq(vcpu);
6137 static void process_nmi(struct kvm_vcpu *vcpu)
6142 * x86 is limited to one NMI running, and one NMI pending after it.
6143 * If an NMI is already in progress, limit further NMIs to just one.
6144 * Otherwise, allow two (and we'll inject the first one immediately).
6146 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6149 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6150 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6151 kvm_make_request(KVM_REQ_EVENT, vcpu);
6154 #define put_smstate(type, buf, offset, val) \
6155 *(type *)((buf) + (offset) - 0x7e00) = val
6157 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
6160 flags |= seg->g << 23;
6161 flags |= seg->db << 22;
6162 flags |= seg->l << 21;
6163 flags |= seg->avl << 20;
6164 flags |= seg->present << 15;
6165 flags |= seg->dpl << 13;
6166 flags |= seg->s << 12;
6167 flags |= seg->type << 8;
6171 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6173 struct kvm_segment seg;
6176 kvm_get_segment(vcpu, &seg, n);
6177 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6180 offset = 0x7f84 + n * 12;
6182 offset = 0x7f2c + (n - 3) * 12;
6184 put_smstate(u32, buf, offset + 8, seg.base);
6185 put_smstate(u32, buf, offset + 4, seg.limit);
6186 put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
6189 #ifdef CONFIG_X86_64
6190 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6192 struct kvm_segment seg;
6196 kvm_get_segment(vcpu, &seg, n);
6197 offset = 0x7e00 + n * 16;
6199 flags = process_smi_get_segment_flags(&seg) >> 8;
6200 put_smstate(u16, buf, offset, seg.selector);
6201 put_smstate(u16, buf, offset + 2, flags);
6202 put_smstate(u32, buf, offset + 4, seg.limit);
6203 put_smstate(u64, buf, offset + 8, seg.base);
6207 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6210 struct kvm_segment seg;
6214 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6215 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6216 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6217 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6219 for (i = 0; i < 8; i++)
6220 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6222 kvm_get_dr(vcpu, 6, &val);
6223 put_smstate(u32, buf, 0x7fcc, (u32)val);
6224 kvm_get_dr(vcpu, 7, &val);
6225 put_smstate(u32, buf, 0x7fc8, (u32)val);
6227 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6228 put_smstate(u32, buf, 0x7fc4, seg.selector);
6229 put_smstate(u32, buf, 0x7f64, seg.base);
6230 put_smstate(u32, buf, 0x7f60, seg.limit);
6231 put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
6233 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6234 put_smstate(u32, buf, 0x7fc0, seg.selector);
6235 put_smstate(u32, buf, 0x7f80, seg.base);
6236 put_smstate(u32, buf, 0x7f7c, seg.limit);
6237 put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6239 kvm_x86_ops->get_gdt(vcpu, &dt);
6240 put_smstate(u32, buf, 0x7f74, dt.address);
6241 put_smstate(u32, buf, 0x7f70, dt.size);
6243 kvm_x86_ops->get_idt(vcpu, &dt);
6244 put_smstate(u32, buf, 0x7f58, dt.address);
6245 put_smstate(u32, buf, 0x7f54, dt.size);
6247 for (i = 0; i < 6; i++)
6248 process_smi_save_seg_32(vcpu, buf, i);
6250 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6253 put_smstate(u32, buf, 0x7efc, 0x00020000);
6254 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6257 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6259 #ifdef CONFIG_X86_64
6261 struct kvm_segment seg;
6265 for (i = 0; i < 16; i++)
6266 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6268 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6269 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6271 kvm_get_dr(vcpu, 6, &val);
6272 put_smstate(u64, buf, 0x7f68, val);
6273 kvm_get_dr(vcpu, 7, &val);
6274 put_smstate(u64, buf, 0x7f60, val);
6276 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6277 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6278 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6280 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6283 put_smstate(u32, buf, 0x7efc, 0x00020064);
6285 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6287 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6288 put_smstate(u16, buf, 0x7e90, seg.selector);
6289 put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6290 put_smstate(u32, buf, 0x7e94, seg.limit);
6291 put_smstate(u64, buf, 0x7e98, seg.base);
6293 kvm_x86_ops->get_idt(vcpu, &dt);
6294 put_smstate(u32, buf, 0x7e84, dt.size);
6295 put_smstate(u64, buf, 0x7e88, dt.address);
6297 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6298 put_smstate(u16, buf, 0x7e70, seg.selector);
6299 put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6300 put_smstate(u32, buf, 0x7e74, seg.limit);
6301 put_smstate(u64, buf, 0x7e78, seg.base);
6303 kvm_x86_ops->get_gdt(vcpu, &dt);
6304 put_smstate(u32, buf, 0x7e64, dt.size);
6305 put_smstate(u64, buf, 0x7e68, dt.address);
6307 for (i = 0; i < 6; i++)
6308 process_smi_save_seg_64(vcpu, buf, i);
6314 static void process_smi(struct kvm_vcpu *vcpu)
6316 struct kvm_segment cs, ds;
6322 vcpu->arch.smi_pending = true;
6326 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6327 vcpu->arch.hflags |= HF_SMM_MASK;
6328 memset(buf, 0, 512);
6329 if (guest_cpuid_has_longmode(vcpu))
6330 process_smi_save_state_64(vcpu, buf);
6332 process_smi_save_state_32(vcpu, buf);
6334 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6336 if (kvm_x86_ops->get_nmi_mask(vcpu))
6337 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6339 kvm_x86_ops->set_nmi_mask(vcpu, true);
6341 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6342 kvm_rip_write(vcpu, 0x8000);
6344 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6345 kvm_x86_ops->set_cr0(vcpu, cr0);
6346 vcpu->arch.cr0 = cr0;
6348 kvm_x86_ops->set_cr4(vcpu, 0);
6350 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6351 dt.address = dt.size = 0;
6352 kvm_x86_ops->set_idt(vcpu, &dt);
6354 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6356 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6357 cs.base = vcpu->arch.smbase;
6362 cs.limit = ds.limit = 0xffffffff;
6363 cs.type = ds.type = 0x3;
6364 cs.dpl = ds.dpl = 0;
6369 cs.avl = ds.avl = 0;
6370 cs.present = ds.present = 1;
6371 cs.unusable = ds.unusable = 0;
6372 cs.padding = ds.padding = 0;
6374 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6375 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6376 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6377 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6378 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6379 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6381 if (guest_cpuid_has_longmode(vcpu))
6382 kvm_x86_ops->set_efer(vcpu, 0);
6384 kvm_update_cpuid(vcpu);
6385 kvm_mmu_reset_context(vcpu);
6388 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6390 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6393 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6395 u64 eoi_exit_bitmap[4];
6397 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6400 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6402 if (irqchip_split(vcpu->kvm))
6403 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6405 if (vcpu->arch.apicv_active)
6406 kvm_x86_ops->sync_pir_to_irr(vcpu);
6407 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6409 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6410 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6411 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6414 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6416 ++vcpu->stat.tlb_flush;
6417 kvm_x86_ops->tlb_flush(vcpu);
6420 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6422 struct page *page = NULL;
6424 if (!lapic_in_kernel(vcpu))
6427 if (!kvm_x86_ops->set_apic_access_page_addr)
6430 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6431 if (is_error_page(page))
6433 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6436 * Do not pin apic access page in memory, the MMU notifier
6437 * will call us again if it is migrated or swapped out.
6441 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6443 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6444 unsigned long address)
6447 * The physical address of apic access page is stored in the VMCS.
6448 * Update it when it becomes invalid.
6450 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6451 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6455 * Returns 1 to let vcpu_run() continue the guest execution loop without
6456 * exiting to the userspace. Otherwise, the value will be returned to the
6459 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6463 dm_request_for_irq_injection(vcpu) &&
6464 kvm_cpu_accept_dm_intr(vcpu);
6466 bool req_immediate_exit = false;
6468 if (vcpu->requests) {
6469 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6470 kvm_mmu_unload(vcpu);
6471 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6472 __kvm_migrate_timers(vcpu);
6473 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6474 kvm_gen_update_masterclock(vcpu->kvm);
6475 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6476 kvm_gen_kvmclock_update(vcpu);
6477 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6478 r = kvm_guest_time_update(vcpu);
6482 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6483 kvm_mmu_sync_roots(vcpu);
6484 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6485 kvm_vcpu_flush_tlb(vcpu);
6486 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6487 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6491 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6492 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6496 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6497 vcpu->fpu_active = 0;
6498 kvm_x86_ops->fpu_deactivate(vcpu);
6500 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6501 /* Page is swapped out. Do synthetic halt */
6502 vcpu->arch.apf.halted = true;
6506 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6507 record_steal_time(vcpu);
6508 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6510 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6512 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6513 kvm_pmu_handle_event(vcpu);
6514 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6515 kvm_pmu_deliver_pmi(vcpu);
6516 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6517 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6518 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6519 vcpu->arch.ioapic_handled_vectors)) {
6520 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6521 vcpu->run->eoi.vector =
6522 vcpu->arch.pending_ioapic_eoi;
6527 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6528 vcpu_scan_ioapic(vcpu);
6529 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6530 kvm_vcpu_reload_apic_access_page(vcpu);
6531 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6532 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6533 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6537 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6538 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6539 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6543 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6544 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6545 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6551 * KVM_REQ_HV_STIMER has to be processed after
6552 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6553 * depend on the guest clock being up-to-date
6555 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6556 kvm_hv_process_stimers(vcpu);
6560 * KVM_REQ_EVENT is not set when posted interrupts are set by
6561 * VT-d hardware, so we have to update RVI unconditionally.
6563 if (kvm_lapic_enabled(vcpu)) {
6565 * Update architecture specific hints for APIC
6566 * virtual interrupt delivery.
6568 if (vcpu->arch.apicv_active)
6569 kvm_x86_ops->hwapic_irr_update(vcpu,
6570 kvm_lapic_find_highest_irr(vcpu));
6573 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6574 kvm_apic_accept_events(vcpu);
6575 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6580 if (inject_pending_event(vcpu, req_int_win) != 0)
6581 req_immediate_exit = true;
6582 /* enable NMI/IRQ window open exits if needed */
6584 if (vcpu->arch.nmi_pending)
6585 kvm_x86_ops->enable_nmi_window(vcpu);
6586 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6587 kvm_x86_ops->enable_irq_window(vcpu);
6590 if (kvm_lapic_enabled(vcpu)) {
6591 update_cr8_intercept(vcpu);
6592 kvm_lapic_sync_to_vapic(vcpu);
6596 r = kvm_mmu_reload(vcpu);
6598 goto cancel_injection;
6603 kvm_x86_ops->prepare_guest_switch(vcpu);
6604 if (vcpu->fpu_active)
6605 kvm_load_guest_fpu(vcpu);
6606 vcpu->mode = IN_GUEST_MODE;
6608 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6611 * We should set ->mode before check ->requests,
6612 * Please see the comment in kvm_make_all_cpus_request.
6613 * This also orders the write to mode from any reads
6614 * to the page tables done while the VCPU is running.
6615 * Please see the comment in kvm_flush_remote_tlbs.
6617 smp_mb__after_srcu_read_unlock();
6619 local_irq_disable();
6621 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6622 || need_resched() || signal_pending(current)) {
6623 vcpu->mode = OUTSIDE_GUEST_MODE;
6627 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6629 goto cancel_injection;
6632 kvm_load_guest_xcr0(vcpu);
6634 if (req_immediate_exit)
6635 smp_send_reschedule(vcpu->cpu);
6637 trace_kvm_entry(vcpu->vcpu_id);
6638 wait_lapic_expire(vcpu);
6639 __kvm_guest_enter();
6641 if (unlikely(vcpu->arch.switch_db_regs)) {
6643 set_debugreg(vcpu->arch.eff_db[0], 0);
6644 set_debugreg(vcpu->arch.eff_db[1], 1);
6645 set_debugreg(vcpu->arch.eff_db[2], 2);
6646 set_debugreg(vcpu->arch.eff_db[3], 3);
6647 set_debugreg(vcpu->arch.dr6, 6);
6648 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6651 kvm_x86_ops->run(vcpu);
6654 * Do this here before restoring debug registers on the host. And
6655 * since we do this before handling the vmexit, a DR access vmexit
6656 * can (a) read the correct value of the debug registers, (b) set
6657 * KVM_DEBUGREG_WONT_EXIT again.
6659 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6660 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6661 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6662 kvm_update_dr0123(vcpu);
6663 kvm_update_dr6(vcpu);
6664 kvm_update_dr7(vcpu);
6665 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6669 * If the guest has used debug registers, at least dr7
6670 * will be disabled while returning to the host.
6671 * If we don't have active breakpoints in the host, we don't
6672 * care about the messed up debug address registers. But if
6673 * we have some of them active, restore the old state.
6675 if (hw_breakpoint_active())
6676 hw_breakpoint_restore();
6678 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6680 vcpu->mode = OUTSIDE_GUEST_MODE;
6683 kvm_put_guest_xcr0(vcpu);
6685 /* Interrupt is enabled by handle_external_intr() */
6686 kvm_x86_ops->handle_external_intr(vcpu);
6691 * We must have an instruction between local_irq_enable() and
6692 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6693 * the interrupt shadow. The stat.exits increment will do nicely.
6694 * But we need to prevent reordering, hence this barrier():
6702 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6705 * Profile KVM exit RIPs:
6707 if (unlikely(prof_on == KVM_PROFILING)) {
6708 unsigned long rip = kvm_rip_read(vcpu);
6709 profile_hit(KVM_PROFILING, (void *)rip);
6712 if (unlikely(vcpu->arch.tsc_always_catchup))
6713 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6715 if (vcpu->arch.apic_attention)
6716 kvm_lapic_sync_from_vapic(vcpu);
6718 r = kvm_x86_ops->handle_exit(vcpu);
6722 kvm_x86_ops->cancel_injection(vcpu);
6723 if (unlikely(vcpu->arch.apic_attention))
6724 kvm_lapic_sync_from_vapic(vcpu);
6729 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6731 if (!kvm_arch_vcpu_runnable(vcpu) &&
6732 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6733 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6734 kvm_vcpu_block(vcpu);
6735 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6737 if (kvm_x86_ops->post_block)
6738 kvm_x86_ops->post_block(vcpu);
6740 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6744 kvm_apic_accept_events(vcpu);
6745 switch(vcpu->arch.mp_state) {
6746 case KVM_MP_STATE_HALTED:
6747 vcpu->arch.pv.pv_unhalted = false;
6748 vcpu->arch.mp_state =
6749 KVM_MP_STATE_RUNNABLE;
6750 case KVM_MP_STATE_RUNNABLE:
6751 vcpu->arch.apf.halted = false;
6753 case KVM_MP_STATE_INIT_RECEIVED:
6762 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6764 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6765 !vcpu->arch.apf.halted);
6768 static int vcpu_run(struct kvm_vcpu *vcpu)
6771 struct kvm *kvm = vcpu->kvm;
6773 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6776 if (kvm_vcpu_running(vcpu)) {
6777 r = vcpu_enter_guest(vcpu);
6779 r = vcpu_block(kvm, vcpu);
6785 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6786 if (kvm_cpu_has_pending_timer(vcpu))
6787 kvm_inject_pending_timer_irqs(vcpu);
6789 if (dm_request_for_irq_injection(vcpu) &&
6790 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
6792 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6793 ++vcpu->stat.request_irq_exits;
6797 kvm_check_async_pf_completion(vcpu);
6799 if (signal_pending(current)) {
6801 vcpu->run->exit_reason = KVM_EXIT_INTR;
6802 ++vcpu->stat.signal_exits;
6805 if (need_resched()) {
6806 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6808 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6812 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6817 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6820 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6821 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6822 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6823 if (r != EMULATE_DONE)
6828 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6830 BUG_ON(!vcpu->arch.pio.count);
6832 return complete_emulated_io(vcpu);
6836 * Implements the following, as a state machine:
6840 * for each mmio piece in the fragment
6848 * for each mmio piece in the fragment
6853 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6855 struct kvm_run *run = vcpu->run;
6856 struct kvm_mmio_fragment *frag;
6859 BUG_ON(!vcpu->mmio_needed);
6861 /* Complete previous fragment */
6862 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6863 len = min(8u, frag->len);
6864 if (!vcpu->mmio_is_write)
6865 memcpy(frag->data, run->mmio.data, len);
6867 if (frag->len <= 8) {
6868 /* Switch to the next fragment. */
6870 vcpu->mmio_cur_fragment++;
6872 /* Go forward to the next mmio piece. */
6878 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6879 vcpu->mmio_needed = 0;
6881 /* FIXME: return into emulator if single-stepping. */
6882 if (vcpu->mmio_is_write)
6884 vcpu->mmio_read_completed = 1;
6885 return complete_emulated_io(vcpu);
6888 run->exit_reason = KVM_EXIT_MMIO;
6889 run->mmio.phys_addr = frag->gpa;
6890 if (vcpu->mmio_is_write)
6891 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6892 run->mmio.len = min(8u, frag->len);
6893 run->mmio.is_write = vcpu->mmio_is_write;
6894 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6899 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6901 struct fpu *fpu = ¤t->thread.fpu;
6905 fpu__activate_curr(fpu);
6907 if (vcpu->sigset_active)
6908 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6910 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6911 kvm_vcpu_block(vcpu);
6912 kvm_apic_accept_events(vcpu);
6913 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6918 /* re-sync apic's tpr */
6919 if (!lapic_in_kernel(vcpu)) {
6920 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6926 if (unlikely(vcpu->arch.complete_userspace_io)) {
6927 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6928 vcpu->arch.complete_userspace_io = NULL;
6933 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6938 post_kvm_run_save(vcpu);
6939 if (vcpu->sigset_active)
6940 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6945 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6947 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6949 * We are here if userspace calls get_regs() in the middle of
6950 * instruction emulation. Registers state needs to be copied
6951 * back from emulation context to vcpu. Userspace shouldn't do
6952 * that usually, but some bad designed PV devices (vmware
6953 * backdoor interface) need this to work
6955 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6956 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6958 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6959 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6960 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6961 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6962 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6963 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6964 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6965 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6966 #ifdef CONFIG_X86_64
6967 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6968 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6969 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6970 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6971 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6972 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6973 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6974 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6977 regs->rip = kvm_rip_read(vcpu);
6978 regs->rflags = kvm_get_rflags(vcpu);
6983 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6985 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6986 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6988 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6989 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6990 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6991 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6992 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6993 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6994 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6995 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6996 #ifdef CONFIG_X86_64
6997 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6998 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6999 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7000 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7001 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7002 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7003 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7004 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7007 kvm_rip_write(vcpu, regs->rip);
7008 kvm_set_rflags(vcpu, regs->rflags);
7010 vcpu->arch.exception.pending = false;
7012 kvm_make_request(KVM_REQ_EVENT, vcpu);
7017 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7019 struct kvm_segment cs;
7021 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7025 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7027 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7028 struct kvm_sregs *sregs)
7032 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7033 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7034 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7035 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7036 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7037 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7039 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7040 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7042 kvm_x86_ops->get_idt(vcpu, &dt);
7043 sregs->idt.limit = dt.size;
7044 sregs->idt.base = dt.address;
7045 kvm_x86_ops->get_gdt(vcpu, &dt);
7046 sregs->gdt.limit = dt.size;
7047 sregs->gdt.base = dt.address;
7049 sregs->cr0 = kvm_read_cr0(vcpu);
7050 sregs->cr2 = vcpu->arch.cr2;
7051 sregs->cr3 = kvm_read_cr3(vcpu);
7052 sregs->cr4 = kvm_read_cr4(vcpu);
7053 sregs->cr8 = kvm_get_cr8(vcpu);
7054 sregs->efer = vcpu->arch.efer;
7055 sregs->apic_base = kvm_get_apic_base(vcpu);
7057 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7059 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7060 set_bit(vcpu->arch.interrupt.nr,
7061 (unsigned long *)sregs->interrupt_bitmap);
7066 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7067 struct kvm_mp_state *mp_state)
7069 kvm_apic_accept_events(vcpu);
7070 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7071 vcpu->arch.pv.pv_unhalted)
7072 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7074 mp_state->mp_state = vcpu->arch.mp_state;
7079 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7080 struct kvm_mp_state *mp_state)
7082 if (!lapic_in_kernel(vcpu) &&
7083 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7086 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7087 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7088 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7090 vcpu->arch.mp_state = mp_state->mp_state;
7091 kvm_make_request(KVM_REQ_EVENT, vcpu);
7095 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7096 int reason, bool has_error_code, u32 error_code)
7098 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7101 init_emulate_ctxt(vcpu);
7103 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7104 has_error_code, error_code);
7107 return EMULATE_FAIL;
7109 kvm_rip_write(vcpu, ctxt->eip);
7110 kvm_set_rflags(vcpu, ctxt->eflags);
7111 kvm_make_request(KVM_REQ_EVENT, vcpu);
7112 return EMULATE_DONE;
7114 EXPORT_SYMBOL_GPL(kvm_task_switch);
7116 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7117 struct kvm_sregs *sregs)
7119 struct msr_data apic_base_msr;
7120 int mmu_reset_needed = 0;
7121 int pending_vec, max_bits, idx;
7124 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7127 dt.size = sregs->idt.limit;
7128 dt.address = sregs->idt.base;
7129 kvm_x86_ops->set_idt(vcpu, &dt);
7130 dt.size = sregs->gdt.limit;
7131 dt.address = sregs->gdt.base;
7132 kvm_x86_ops->set_gdt(vcpu, &dt);
7134 vcpu->arch.cr2 = sregs->cr2;
7135 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7136 vcpu->arch.cr3 = sregs->cr3;
7137 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7139 kvm_set_cr8(vcpu, sregs->cr8);
7141 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7142 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7143 apic_base_msr.data = sregs->apic_base;
7144 apic_base_msr.host_initiated = true;
7145 kvm_set_apic_base(vcpu, &apic_base_msr);
7147 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7148 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7149 vcpu->arch.cr0 = sregs->cr0;
7151 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7152 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7153 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7154 kvm_update_cpuid(vcpu);
7156 idx = srcu_read_lock(&vcpu->kvm->srcu);
7157 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7158 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7159 mmu_reset_needed = 1;
7161 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7163 if (mmu_reset_needed)
7164 kvm_mmu_reset_context(vcpu);
7166 max_bits = KVM_NR_INTERRUPTS;
7167 pending_vec = find_first_bit(
7168 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7169 if (pending_vec < max_bits) {
7170 kvm_queue_interrupt(vcpu, pending_vec, false);
7171 pr_debug("Set back pending irq %d\n", pending_vec);
7174 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7175 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7176 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7177 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7178 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7179 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7181 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7182 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7184 update_cr8_intercept(vcpu);
7186 /* Older userspace won't unhalt the vcpu on reset. */
7187 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7188 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7190 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7192 kvm_make_request(KVM_REQ_EVENT, vcpu);
7197 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7198 struct kvm_guest_debug *dbg)
7200 unsigned long rflags;
7203 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7205 if (vcpu->arch.exception.pending)
7207 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7208 kvm_queue_exception(vcpu, DB_VECTOR);
7210 kvm_queue_exception(vcpu, BP_VECTOR);
7214 * Read rflags as long as potentially injected trace flags are still
7217 rflags = kvm_get_rflags(vcpu);
7219 vcpu->guest_debug = dbg->control;
7220 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7221 vcpu->guest_debug = 0;
7223 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7224 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7225 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7226 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7228 for (i = 0; i < KVM_NR_DB_REGS; i++)
7229 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7231 kvm_update_dr7(vcpu);
7233 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7234 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7235 get_segment_base(vcpu, VCPU_SREG_CS);
7238 * Trigger an rflags update that will inject or remove the trace
7241 kvm_set_rflags(vcpu, rflags);
7243 kvm_x86_ops->update_bp_intercept(vcpu);
7253 * Translate a guest virtual address to a guest physical address.
7255 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7256 struct kvm_translation *tr)
7258 unsigned long vaddr = tr->linear_address;
7262 idx = srcu_read_lock(&vcpu->kvm->srcu);
7263 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7264 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7265 tr->physical_address = gpa;
7266 tr->valid = gpa != UNMAPPED_GVA;
7273 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7275 struct fxregs_state *fxsave =
7276 &vcpu->arch.guest_fpu.state.fxsave;
7278 memcpy(fpu->fpr, fxsave->st_space, 128);
7279 fpu->fcw = fxsave->cwd;
7280 fpu->fsw = fxsave->swd;
7281 fpu->ftwx = fxsave->twd;
7282 fpu->last_opcode = fxsave->fop;
7283 fpu->last_ip = fxsave->rip;
7284 fpu->last_dp = fxsave->rdp;
7285 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7290 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7292 struct fxregs_state *fxsave =
7293 &vcpu->arch.guest_fpu.state.fxsave;
7295 memcpy(fxsave->st_space, fpu->fpr, 128);
7296 fxsave->cwd = fpu->fcw;
7297 fxsave->swd = fpu->fsw;
7298 fxsave->twd = fpu->ftwx;
7299 fxsave->fop = fpu->last_opcode;
7300 fxsave->rip = fpu->last_ip;
7301 fxsave->rdp = fpu->last_dp;
7302 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7307 static void fx_init(struct kvm_vcpu *vcpu)
7309 fpstate_init(&vcpu->arch.guest_fpu.state);
7310 if (boot_cpu_has(X86_FEATURE_XSAVES))
7311 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7312 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7315 * Ensure guest xcr0 is valid for loading
7317 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7319 vcpu->arch.cr0 |= X86_CR0_ET;
7322 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7324 if (vcpu->guest_fpu_loaded)
7328 * Restore all possible states in the guest,
7329 * and assume host would use all available bits.
7330 * Guest xcr0 would be loaded later.
7332 vcpu->guest_fpu_loaded = 1;
7333 __kernel_fpu_begin();
7334 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7338 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7340 if (!vcpu->guest_fpu_loaded) {
7341 vcpu->fpu_counter = 0;
7345 vcpu->guest_fpu_loaded = 0;
7346 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7348 ++vcpu->stat.fpu_reload;
7350 * If using eager FPU mode, or if the guest is a frequent user
7351 * of the FPU, just leave the FPU active for next time.
7352 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7353 * the FPU in bursts will revert to loading it on demand.
7355 if (!use_eager_fpu()) {
7356 if (++vcpu->fpu_counter < 5)
7357 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7362 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7364 kvmclock_reset(vcpu);
7366 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7367 kvm_x86_ops->vcpu_free(vcpu);
7370 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7373 struct kvm_vcpu *vcpu;
7375 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7376 printk_once(KERN_WARNING
7377 "kvm: SMP vm created on host with unstable TSC; "
7378 "guest TSC will not be reliable\n");
7380 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7385 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7389 kvm_vcpu_mtrr_init(vcpu);
7390 r = vcpu_load(vcpu);
7393 kvm_vcpu_reset(vcpu, false);
7394 kvm_mmu_setup(vcpu);
7399 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7401 struct msr_data msr;
7402 struct kvm *kvm = vcpu->kvm;
7404 if (vcpu_load(vcpu))
7407 msr.index = MSR_IA32_TSC;
7408 msr.host_initiated = true;
7409 kvm_write_tsc(vcpu, &msr);
7412 if (!kvmclock_periodic_sync)
7415 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7416 KVMCLOCK_SYNC_PERIOD);
7419 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7422 vcpu->arch.apf.msr_val = 0;
7424 r = vcpu_load(vcpu);
7426 kvm_mmu_unload(vcpu);
7429 kvm_x86_ops->vcpu_free(vcpu);
7432 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7434 vcpu->arch.hflags = 0;
7436 atomic_set(&vcpu->arch.nmi_queued, 0);
7437 vcpu->arch.nmi_pending = 0;
7438 vcpu->arch.nmi_injected = false;
7439 kvm_clear_interrupt_queue(vcpu);
7440 kvm_clear_exception_queue(vcpu);
7442 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7443 kvm_update_dr0123(vcpu);
7444 vcpu->arch.dr6 = DR6_INIT;
7445 kvm_update_dr6(vcpu);
7446 vcpu->arch.dr7 = DR7_FIXED_1;
7447 kvm_update_dr7(vcpu);
7451 kvm_make_request(KVM_REQ_EVENT, vcpu);
7452 vcpu->arch.apf.msr_val = 0;
7453 vcpu->arch.st.msr_val = 0;
7455 kvmclock_reset(vcpu);
7457 kvm_clear_async_pf_completion_queue(vcpu);
7458 kvm_async_pf_hash_reset(vcpu);
7459 vcpu->arch.apf.halted = false;
7462 kvm_pmu_reset(vcpu);
7463 vcpu->arch.smbase = 0x30000;
7466 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7467 vcpu->arch.regs_avail = ~0;
7468 vcpu->arch.regs_dirty = ~0;
7470 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7473 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7475 struct kvm_segment cs;
7477 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7478 cs.selector = vector << 8;
7479 cs.base = vector << 12;
7480 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7481 kvm_rip_write(vcpu, 0);
7484 int kvm_arch_hardware_enable(void)
7487 struct kvm_vcpu *vcpu;
7492 bool stable, backwards_tsc = false;
7494 kvm_shared_msr_cpu_online();
7495 ret = kvm_x86_ops->hardware_enable();
7499 local_tsc = rdtsc();
7500 stable = !check_tsc_unstable();
7501 list_for_each_entry(kvm, &vm_list, vm_list) {
7502 kvm_for_each_vcpu(i, vcpu, kvm) {
7503 if (!stable && vcpu->cpu == smp_processor_id())
7504 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7505 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7506 backwards_tsc = true;
7507 if (vcpu->arch.last_host_tsc > max_tsc)
7508 max_tsc = vcpu->arch.last_host_tsc;
7514 * Sometimes, even reliable TSCs go backwards. This happens on
7515 * platforms that reset TSC during suspend or hibernate actions, but
7516 * maintain synchronization. We must compensate. Fortunately, we can
7517 * detect that condition here, which happens early in CPU bringup,
7518 * before any KVM threads can be running. Unfortunately, we can't
7519 * bring the TSCs fully up to date with real time, as we aren't yet far
7520 * enough into CPU bringup that we know how much real time has actually
7521 * elapsed; our helper function, get_kernel_ns() will be using boot
7522 * variables that haven't been updated yet.
7524 * So we simply find the maximum observed TSC above, then record the
7525 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7526 * the adjustment will be applied. Note that we accumulate
7527 * adjustments, in case multiple suspend cycles happen before some VCPU
7528 * gets a chance to run again. In the event that no KVM threads get a
7529 * chance to run, we will miss the entire elapsed period, as we'll have
7530 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7531 * loose cycle time. This isn't too big a deal, since the loss will be
7532 * uniform across all VCPUs (not to mention the scenario is extremely
7533 * unlikely). It is possible that a second hibernate recovery happens
7534 * much faster than a first, causing the observed TSC here to be
7535 * smaller; this would require additional padding adjustment, which is
7536 * why we set last_host_tsc to the local tsc observed here.
7538 * N.B. - this code below runs only on platforms with reliable TSC,
7539 * as that is the only way backwards_tsc is set above. Also note
7540 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7541 * have the same delta_cyc adjustment applied if backwards_tsc
7542 * is detected. Note further, this adjustment is only done once,
7543 * as we reset last_host_tsc on all VCPUs to stop this from being
7544 * called multiple times (one for each physical CPU bringup).
7546 * Platforms with unreliable TSCs don't have to deal with this, they
7547 * will be compensated by the logic in vcpu_load, which sets the TSC to
7548 * catchup mode. This will catchup all VCPUs to real time, but cannot
7549 * guarantee that they stay in perfect synchronization.
7551 if (backwards_tsc) {
7552 u64 delta_cyc = max_tsc - local_tsc;
7553 backwards_tsc_observed = true;
7554 list_for_each_entry(kvm, &vm_list, vm_list) {
7555 kvm_for_each_vcpu(i, vcpu, kvm) {
7556 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7557 vcpu->arch.last_host_tsc = local_tsc;
7558 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7562 * We have to disable TSC offset matching.. if you were
7563 * booting a VM while issuing an S4 host suspend....
7564 * you may have some problem. Solving this issue is
7565 * left as an exercise to the reader.
7567 kvm->arch.last_tsc_nsec = 0;
7568 kvm->arch.last_tsc_write = 0;
7575 void kvm_arch_hardware_disable(void)
7577 kvm_x86_ops->hardware_disable();
7578 drop_user_return_notifiers();
7581 int kvm_arch_hardware_setup(void)
7585 r = kvm_x86_ops->hardware_setup();
7589 if (kvm_has_tsc_control) {
7591 * Make sure the user can only configure tsc_khz values that
7592 * fit into a signed integer.
7593 * A min value is not calculated needed because it will always
7594 * be 1 on all machines.
7596 u64 max = min(0x7fffffffULL,
7597 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7598 kvm_max_guest_tsc_khz = max;
7600 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7603 kvm_init_msr_list();
7607 void kvm_arch_hardware_unsetup(void)
7609 kvm_x86_ops->hardware_unsetup();
7612 void kvm_arch_check_processor_compat(void *rtn)
7614 kvm_x86_ops->check_processor_compatibility(rtn);
7617 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7619 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7621 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7623 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7625 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7628 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7630 return irqchip_in_kernel(vcpu->kvm) == lapic_in_kernel(vcpu);
7633 struct static_key kvm_no_apic_vcpu __read_mostly;
7634 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7636 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7642 BUG_ON(vcpu->kvm == NULL);
7645 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7646 vcpu->arch.pv.pv_unhalted = false;
7647 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7648 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7649 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7651 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7653 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7658 vcpu->arch.pio_data = page_address(page);
7660 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7662 r = kvm_mmu_create(vcpu);
7664 goto fail_free_pio_data;
7666 if (irqchip_in_kernel(kvm)) {
7667 r = kvm_create_lapic(vcpu);
7669 goto fail_mmu_destroy;
7671 static_key_slow_inc(&kvm_no_apic_vcpu);
7673 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7675 if (!vcpu->arch.mce_banks) {
7677 goto fail_free_lapic;
7679 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7681 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7683 goto fail_free_mce_banks;
7688 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7689 vcpu->arch.pv_time_enabled = false;
7691 vcpu->arch.guest_supported_xcr0 = 0;
7692 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7694 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7696 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7698 kvm_async_pf_hash_reset(vcpu);
7701 vcpu->arch.pending_external_vector = -1;
7703 kvm_hv_vcpu_init(vcpu);
7707 fail_free_mce_banks:
7708 kfree(vcpu->arch.mce_banks);
7710 kvm_free_lapic(vcpu);
7712 kvm_mmu_destroy(vcpu);
7714 free_page((unsigned long)vcpu->arch.pio_data);
7719 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7723 kvm_hv_vcpu_uninit(vcpu);
7724 kvm_pmu_destroy(vcpu);
7725 kfree(vcpu->arch.mce_banks);
7726 kvm_free_lapic(vcpu);
7727 idx = srcu_read_lock(&vcpu->kvm->srcu);
7728 kvm_mmu_destroy(vcpu);
7729 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7730 free_page((unsigned long)vcpu->arch.pio_data);
7731 if (!lapic_in_kernel(vcpu))
7732 static_key_slow_dec(&kvm_no_apic_vcpu);
7735 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7737 kvm_x86_ops->sched_in(vcpu, cpu);
7740 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7745 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7746 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7747 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7748 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7749 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7751 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7752 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7753 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7754 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7755 &kvm->arch.irq_sources_bitmap);
7757 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7758 mutex_init(&kvm->arch.apic_map_lock);
7759 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7761 pvclock_update_vm_gtod_copy(kvm);
7763 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7764 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7766 kvm_page_track_init(kvm);
7767 kvm_mmu_init_vm(kvm);
7769 if (kvm_x86_ops->vm_init)
7770 return kvm_x86_ops->vm_init(kvm);
7775 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7778 r = vcpu_load(vcpu);
7780 kvm_mmu_unload(vcpu);
7784 static void kvm_free_vcpus(struct kvm *kvm)
7787 struct kvm_vcpu *vcpu;
7790 * Unpin any mmu pages first.
7792 kvm_for_each_vcpu(i, vcpu, kvm) {
7793 kvm_clear_async_pf_completion_queue(vcpu);
7794 kvm_unload_vcpu_mmu(vcpu);
7796 kvm_for_each_vcpu(i, vcpu, kvm)
7797 kvm_arch_vcpu_free(vcpu);
7799 mutex_lock(&kvm->lock);
7800 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7801 kvm->vcpus[i] = NULL;
7803 atomic_set(&kvm->online_vcpus, 0);
7804 mutex_unlock(&kvm->lock);
7807 void kvm_arch_sync_events(struct kvm *kvm)
7809 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7810 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7811 kvm_free_all_assigned_devices(kvm);
7815 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7819 struct kvm_memslots *slots = kvm_memslots(kvm);
7820 struct kvm_memory_slot *slot, old;
7822 /* Called with kvm->slots_lock held. */
7823 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7826 slot = id_to_memslot(slots, id);
7832 * MAP_SHARED to prevent internal slot pages from being moved
7835 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7836 MAP_SHARED | MAP_ANONYMOUS, 0);
7837 if (IS_ERR((void *)hva))
7838 return PTR_ERR((void *)hva);
7847 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7848 struct kvm_userspace_memory_region m;
7850 m.slot = id | (i << 16);
7852 m.guest_phys_addr = gpa;
7853 m.userspace_addr = hva;
7854 m.memory_size = size;
7855 r = __kvm_set_memory_region(kvm, &m);
7861 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7867 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7869 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7873 mutex_lock(&kvm->slots_lock);
7874 r = __x86_set_memory_region(kvm, id, gpa, size);
7875 mutex_unlock(&kvm->slots_lock);
7879 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7881 void kvm_arch_destroy_vm(struct kvm *kvm)
7883 if (current->mm == kvm->mm) {
7885 * Free memory regions allocated on behalf of userspace,
7886 * unless the the memory map has changed due to process exit
7889 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7890 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7891 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7893 if (kvm_x86_ops->vm_destroy)
7894 kvm_x86_ops->vm_destroy(kvm);
7895 kvm_iommu_unmap_guest(kvm);
7896 kfree(kvm->arch.vpic);
7897 kfree(kvm->arch.vioapic);
7898 kvm_free_vcpus(kvm);
7899 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7900 kvm_mmu_uninit_vm(kvm);
7903 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7904 struct kvm_memory_slot *dont)
7908 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7909 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7910 kvfree(free->arch.rmap[i]);
7911 free->arch.rmap[i] = NULL;
7916 if (!dont || free->arch.lpage_info[i - 1] !=
7917 dont->arch.lpage_info[i - 1]) {
7918 kvfree(free->arch.lpage_info[i - 1]);
7919 free->arch.lpage_info[i - 1] = NULL;
7923 kvm_page_track_free_memslot(free, dont);
7926 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7927 unsigned long npages)
7931 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7932 struct kvm_lpage_info *linfo;
7937 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7938 slot->base_gfn, level) + 1;
7940 slot->arch.rmap[i] =
7941 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7942 if (!slot->arch.rmap[i])
7947 linfo = kvm_kvzalloc(lpages * sizeof(*linfo));
7951 slot->arch.lpage_info[i - 1] = linfo;
7953 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7954 linfo[0].disallow_lpage = 1;
7955 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7956 linfo[lpages - 1].disallow_lpage = 1;
7957 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7959 * If the gfn and userspace address are not aligned wrt each
7960 * other, or if explicitly asked to, disable large page
7961 * support for this slot
7963 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7964 !kvm_largepages_enabled()) {
7967 for (j = 0; j < lpages; ++j)
7968 linfo[j].disallow_lpage = 1;
7972 if (kvm_page_track_create_memslot(slot, npages))
7978 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7979 kvfree(slot->arch.rmap[i]);
7980 slot->arch.rmap[i] = NULL;
7984 kvfree(slot->arch.lpage_info[i - 1]);
7985 slot->arch.lpage_info[i - 1] = NULL;
7990 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7993 * memslots->generation has been incremented.
7994 * mmio generation may have reached its maximum value.
7996 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7999 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8000 struct kvm_memory_slot *memslot,
8001 const struct kvm_userspace_memory_region *mem,
8002 enum kvm_mr_change change)
8007 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8008 struct kvm_memory_slot *new)
8010 /* Still write protect RO slot */
8011 if (new->flags & KVM_MEM_READONLY) {
8012 kvm_mmu_slot_remove_write_access(kvm, new);
8017 * Call kvm_x86_ops dirty logging hooks when they are valid.
8019 * kvm_x86_ops->slot_disable_log_dirty is called when:
8021 * - KVM_MR_CREATE with dirty logging is disabled
8022 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8024 * The reason is, in case of PML, we need to set D-bit for any slots
8025 * with dirty logging disabled in order to eliminate unnecessary GPA
8026 * logging in PML buffer (and potential PML buffer full VMEXT). This
8027 * guarantees leaving PML enabled during guest's lifetime won't have
8028 * any additonal overhead from PML when guest is running with dirty
8029 * logging disabled for memory slots.
8031 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8032 * to dirty logging mode.
8034 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8036 * In case of write protect:
8038 * Write protect all pages for dirty logging.
8040 * All the sptes including the large sptes which point to this
8041 * slot are set to readonly. We can not create any new large
8042 * spte on this slot until the end of the logging.
8044 * See the comments in fast_page_fault().
8046 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8047 if (kvm_x86_ops->slot_enable_log_dirty)
8048 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8050 kvm_mmu_slot_remove_write_access(kvm, new);
8052 if (kvm_x86_ops->slot_disable_log_dirty)
8053 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8057 void kvm_arch_commit_memory_region(struct kvm *kvm,
8058 const struct kvm_userspace_memory_region *mem,
8059 const struct kvm_memory_slot *old,
8060 const struct kvm_memory_slot *new,
8061 enum kvm_mr_change change)
8063 int nr_mmu_pages = 0;
8065 if (!kvm->arch.n_requested_mmu_pages)
8066 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8069 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8072 * Dirty logging tracks sptes in 4k granularity, meaning that large
8073 * sptes have to be split. If live migration is successful, the guest
8074 * in the source machine will be destroyed and large sptes will be
8075 * created in the destination. However, if the guest continues to run
8076 * in the source machine (for example if live migration fails), small
8077 * sptes will remain around and cause bad performance.
8079 * Scan sptes if dirty logging has been stopped, dropping those
8080 * which can be collapsed into a single large-page spte. Later
8081 * page faults will create the large-page sptes.
8083 if ((change != KVM_MR_DELETE) &&
8084 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8085 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8086 kvm_mmu_zap_collapsible_sptes(kvm, new);
8089 * Set up write protection and/or dirty logging for the new slot.
8091 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8092 * been zapped so no dirty logging staff is needed for old slot. For
8093 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8094 * new and it's also covered when dealing with the new slot.
8096 * FIXME: const-ify all uses of struct kvm_memory_slot.
8098 if (change != KVM_MR_DELETE)
8099 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8102 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8104 kvm_mmu_invalidate_zap_all_pages(kvm);
8107 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8108 struct kvm_memory_slot *slot)
8110 kvm_mmu_invalidate_zap_all_pages(kvm);
8113 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8115 if (!list_empty_careful(&vcpu->async_pf.done))
8118 if (kvm_apic_has_events(vcpu))
8121 if (vcpu->arch.pv.pv_unhalted)
8124 if (atomic_read(&vcpu->arch.nmi_queued))
8127 if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8130 if (kvm_arch_interrupt_allowed(vcpu) &&
8131 kvm_cpu_has_interrupt(vcpu))
8134 if (kvm_hv_has_stimer_pending(vcpu))
8140 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8142 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8143 kvm_x86_ops->check_nested_events(vcpu, false);
8145 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8148 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8150 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8153 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8155 return kvm_x86_ops->interrupt_allowed(vcpu);
8158 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8160 if (is_64_bit_mode(vcpu))
8161 return kvm_rip_read(vcpu);
8162 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8163 kvm_rip_read(vcpu));
8165 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8167 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8169 return kvm_get_linear_rip(vcpu) == linear_rip;
8171 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8173 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8175 unsigned long rflags;
8177 rflags = kvm_x86_ops->get_rflags(vcpu);
8178 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8179 rflags &= ~X86_EFLAGS_TF;
8182 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8184 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8186 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8187 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8188 rflags |= X86_EFLAGS_TF;
8189 kvm_x86_ops->set_rflags(vcpu, rflags);
8192 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8194 __kvm_set_rflags(vcpu, rflags);
8195 kvm_make_request(KVM_REQ_EVENT, vcpu);
8197 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8199 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8203 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8207 r = kvm_mmu_reload(vcpu);
8211 if (!vcpu->arch.mmu.direct_map &&
8212 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8215 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8218 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8220 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8223 static inline u32 kvm_async_pf_next_probe(u32 key)
8225 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8228 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8230 u32 key = kvm_async_pf_hash_fn(gfn);
8232 while (vcpu->arch.apf.gfns[key] != ~0)
8233 key = kvm_async_pf_next_probe(key);
8235 vcpu->arch.apf.gfns[key] = gfn;
8238 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8241 u32 key = kvm_async_pf_hash_fn(gfn);
8243 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8244 (vcpu->arch.apf.gfns[key] != gfn &&
8245 vcpu->arch.apf.gfns[key] != ~0); i++)
8246 key = kvm_async_pf_next_probe(key);
8251 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8253 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8256 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8260 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8262 vcpu->arch.apf.gfns[i] = ~0;
8264 j = kvm_async_pf_next_probe(j);
8265 if (vcpu->arch.apf.gfns[j] == ~0)
8267 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8269 * k lies cyclically in ]i,j]
8271 * |....j i.k.| or |.k..j i...|
8273 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8274 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8279 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8282 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8286 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8287 struct kvm_async_pf *work)
8289 struct x86_exception fault;
8291 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8292 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8294 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8295 (vcpu->arch.apf.send_user_only &&
8296 kvm_x86_ops->get_cpl(vcpu) == 0))
8297 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8298 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8299 fault.vector = PF_VECTOR;
8300 fault.error_code_valid = true;
8301 fault.error_code = 0;
8302 fault.nested_page_fault = false;
8303 fault.address = work->arch.token;
8304 kvm_inject_page_fault(vcpu, &fault);
8308 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8309 struct kvm_async_pf *work)
8311 struct x86_exception fault;
8313 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8314 if (work->wakeup_all)
8315 work->arch.token = ~0; /* broadcast wakeup */
8317 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8319 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8320 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8321 fault.vector = PF_VECTOR;
8322 fault.error_code_valid = true;
8323 fault.error_code = 0;
8324 fault.nested_page_fault = false;
8325 fault.address = work->arch.token;
8326 kvm_inject_page_fault(vcpu, &fault);
8328 vcpu->arch.apf.halted = false;
8329 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8332 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8334 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8337 return !kvm_event_needs_reinjection(vcpu) &&
8338 kvm_x86_ops->interrupt_allowed(vcpu);
8341 void kvm_arch_start_assignment(struct kvm *kvm)
8343 atomic_inc(&kvm->arch.assigned_device_count);
8345 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8347 void kvm_arch_end_assignment(struct kvm *kvm)
8349 atomic_dec(&kvm->arch.assigned_device_count);
8351 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8353 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8355 return atomic_read(&kvm->arch.assigned_device_count);
8357 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8359 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8361 atomic_inc(&kvm->arch.noncoherent_dma_count);
8363 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8365 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8367 atomic_dec(&kvm->arch.noncoherent_dma_count);
8369 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8371 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8373 return atomic_read(&kvm->arch.noncoherent_dma_count);
8375 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8377 bool kvm_arch_has_irq_bypass(void)
8379 return kvm_x86_ops->update_pi_irte != NULL;
8382 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8383 struct irq_bypass_producer *prod)
8385 struct kvm_kernel_irqfd *irqfd =
8386 container_of(cons, struct kvm_kernel_irqfd, consumer);
8388 irqfd->producer = prod;
8390 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8391 prod->irq, irqfd->gsi, 1);
8394 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8395 struct irq_bypass_producer *prod)
8398 struct kvm_kernel_irqfd *irqfd =
8399 container_of(cons, struct kvm_kernel_irqfd, consumer);
8401 WARN_ON(irqfd->producer != prod);
8402 irqfd->producer = NULL;
8405 * When producer of consumer is unregistered, we change back to
8406 * remapped mode, so we can re-use the current implementation
8407 * when the irq is masked/disabed or the consumer side (KVM
8408 * int this case doesn't want to receive the interrupts.
8410 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8412 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8413 " fails: %d\n", irqfd->consumer.token, ret);
8416 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8417 uint32_t guest_irq, bool set)
8419 if (!kvm_x86_ops->update_pi_irte)
8422 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8425 bool kvm_vector_hashing_enabled(void)
8427 return vector_hashing;
8429 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8431 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8432 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8433 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8434 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8435 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8436 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8437 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8438 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8439 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8440 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8441 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8442 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8443 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8444 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8445 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8446 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8447 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8448 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8449 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);