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 <trace/events/kvm.h>
56 #define CREATE_TRACE_POINTS
59 #include <asm/debugreg.h>
63 #include <linux/kernel_stat.h>
64 #include <asm/fpu/internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
68 #define MAX_IO_MSRS 256
69 #define KVM_MAX_MCE_BANKS 32
70 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
72 #define emul_to_vcpu(ctxt) \
73 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
76 * - enable syscall per default because its emulated by KVM
77 * - enable LME and LMA per default on 64 bit KVM
81 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
83 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
86 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
87 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
89 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
90 static void process_nmi(struct kvm_vcpu *vcpu);
91 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
93 struct kvm_x86_ops *kvm_x86_ops;
94 EXPORT_SYMBOL_GPL(kvm_x86_ops);
96 static bool ignore_msrs = 0;
97 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
99 unsigned int min_timer_period_us = 500;
100 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
102 static bool __read_mostly kvmclock_periodic_sync = true;
103 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
105 bool kvm_has_tsc_control;
106 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
107 u32 kvm_max_guest_tsc_khz;
108 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
110 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
111 static u32 tsc_tolerance_ppm = 250;
112 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
114 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
115 unsigned int lapic_timer_advance_ns = 0;
116 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
118 static bool backwards_tsc_observed = false;
120 #define KVM_NR_SHARED_MSRS 16
122 struct kvm_shared_msrs_global {
124 u32 msrs[KVM_NR_SHARED_MSRS];
127 struct kvm_shared_msrs {
128 struct user_return_notifier urn;
130 struct kvm_shared_msr_values {
133 } values[KVM_NR_SHARED_MSRS];
136 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
137 static struct kvm_shared_msrs __percpu *shared_msrs;
139 struct kvm_stats_debugfs_item debugfs_entries[] = {
140 { "pf_fixed", VCPU_STAT(pf_fixed) },
141 { "pf_guest", VCPU_STAT(pf_guest) },
142 { "tlb_flush", VCPU_STAT(tlb_flush) },
143 { "invlpg", VCPU_STAT(invlpg) },
144 { "exits", VCPU_STAT(exits) },
145 { "io_exits", VCPU_STAT(io_exits) },
146 { "mmio_exits", VCPU_STAT(mmio_exits) },
147 { "signal_exits", VCPU_STAT(signal_exits) },
148 { "irq_window", VCPU_STAT(irq_window_exits) },
149 { "nmi_window", VCPU_STAT(nmi_window_exits) },
150 { "halt_exits", VCPU_STAT(halt_exits) },
151 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
152 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
153 { "hypercalls", VCPU_STAT(hypercalls) },
154 { "request_irq", VCPU_STAT(request_irq_exits) },
155 { "irq_exits", VCPU_STAT(irq_exits) },
156 { "host_state_reload", VCPU_STAT(host_state_reload) },
157 { "efer_reload", VCPU_STAT(efer_reload) },
158 { "fpu_reload", VCPU_STAT(fpu_reload) },
159 { "insn_emulation", VCPU_STAT(insn_emulation) },
160 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
161 { "irq_injections", VCPU_STAT(irq_injections) },
162 { "nmi_injections", VCPU_STAT(nmi_injections) },
163 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
164 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
165 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
166 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
167 { "mmu_flooded", VM_STAT(mmu_flooded) },
168 { "mmu_recycled", VM_STAT(mmu_recycled) },
169 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
170 { "mmu_unsync", VM_STAT(mmu_unsync) },
171 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
172 { "largepages", VM_STAT(lpages) },
176 u64 __read_mostly host_xcr0;
178 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
180 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
183 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
184 vcpu->arch.apf.gfns[i] = ~0;
187 static void kvm_on_user_return(struct user_return_notifier *urn)
190 struct kvm_shared_msrs *locals
191 = container_of(urn, struct kvm_shared_msrs, urn);
192 struct kvm_shared_msr_values *values;
194 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
195 values = &locals->values[slot];
196 if (values->host != values->curr) {
197 wrmsrl(shared_msrs_global.msrs[slot], values->host);
198 values->curr = values->host;
201 locals->registered = false;
202 user_return_notifier_unregister(urn);
205 static void shared_msr_update(unsigned slot, u32 msr)
208 unsigned int cpu = smp_processor_id();
209 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
211 /* only read, and nobody should modify it at this time,
212 * so don't need lock */
213 if (slot >= shared_msrs_global.nr) {
214 printk(KERN_ERR "kvm: invalid MSR slot!");
217 rdmsrl_safe(msr, &value);
218 smsr->values[slot].host = value;
219 smsr->values[slot].curr = value;
222 void kvm_define_shared_msr(unsigned slot, u32 msr)
224 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
225 shared_msrs_global.msrs[slot] = msr;
226 if (slot >= shared_msrs_global.nr)
227 shared_msrs_global.nr = slot + 1;
229 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
231 static void kvm_shared_msr_cpu_online(void)
235 for (i = 0; i < shared_msrs_global.nr; ++i)
236 shared_msr_update(i, shared_msrs_global.msrs[i]);
239 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
241 unsigned int cpu = smp_processor_id();
242 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
245 if (((value ^ smsr->values[slot].curr) & mask) == 0)
247 smsr->values[slot].curr = value;
248 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
252 if (!smsr->registered) {
253 smsr->urn.on_user_return = kvm_on_user_return;
254 user_return_notifier_register(&smsr->urn);
255 smsr->registered = true;
259 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
261 static void drop_user_return_notifiers(void)
263 unsigned int cpu = smp_processor_id();
264 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
266 if (smsr->registered)
267 kvm_on_user_return(&smsr->urn);
270 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
272 return vcpu->arch.apic_base;
274 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
276 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
278 u64 old_state = vcpu->arch.apic_base &
279 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
280 u64 new_state = msr_info->data &
281 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
282 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
283 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
285 if (!msr_info->host_initiated &&
286 ((msr_info->data & reserved_bits) != 0 ||
287 new_state == X2APIC_ENABLE ||
288 (new_state == MSR_IA32_APICBASE_ENABLE &&
289 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
290 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
294 kvm_lapic_set_base(vcpu, msr_info->data);
297 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
299 asmlinkage __visible void kvm_spurious_fault(void)
301 /* Fault while not rebooting. We want the trace. */
304 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
306 #define EXCPT_BENIGN 0
307 #define EXCPT_CONTRIBUTORY 1
310 static int exception_class(int vector)
320 return EXCPT_CONTRIBUTORY;
327 #define EXCPT_FAULT 0
329 #define EXCPT_ABORT 2
330 #define EXCPT_INTERRUPT 3
332 static int exception_type(int vector)
336 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
337 return EXCPT_INTERRUPT;
341 /* #DB is trap, as instruction watchpoints are handled elsewhere */
342 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
345 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
348 /* Reserved exceptions will result in fault */
352 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
353 unsigned nr, bool has_error, u32 error_code,
359 kvm_make_request(KVM_REQ_EVENT, vcpu);
361 if (!vcpu->arch.exception.pending) {
363 if (has_error && !is_protmode(vcpu))
365 vcpu->arch.exception.pending = true;
366 vcpu->arch.exception.has_error_code = has_error;
367 vcpu->arch.exception.nr = nr;
368 vcpu->arch.exception.error_code = error_code;
369 vcpu->arch.exception.reinject = reinject;
373 /* to check exception */
374 prev_nr = vcpu->arch.exception.nr;
375 if (prev_nr == DF_VECTOR) {
376 /* triple fault -> shutdown */
377 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
380 class1 = exception_class(prev_nr);
381 class2 = exception_class(nr);
382 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
383 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
384 /* generate double fault per SDM Table 5-5 */
385 vcpu->arch.exception.pending = true;
386 vcpu->arch.exception.has_error_code = true;
387 vcpu->arch.exception.nr = DF_VECTOR;
388 vcpu->arch.exception.error_code = 0;
390 /* replace previous exception with a new one in a hope
391 that instruction re-execution will regenerate lost
396 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
398 kvm_multiple_exception(vcpu, nr, false, 0, false);
400 EXPORT_SYMBOL_GPL(kvm_queue_exception);
402 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
404 kvm_multiple_exception(vcpu, nr, false, 0, true);
406 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
408 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
411 kvm_inject_gp(vcpu, 0);
413 kvm_x86_ops->skip_emulated_instruction(vcpu);
415 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
417 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
419 ++vcpu->stat.pf_guest;
420 vcpu->arch.cr2 = fault->address;
421 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
423 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
425 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
427 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
428 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
430 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
432 return fault->nested_page_fault;
435 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
437 atomic_inc(&vcpu->arch.nmi_queued);
438 kvm_make_request(KVM_REQ_NMI, vcpu);
440 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
442 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
444 kvm_multiple_exception(vcpu, nr, true, error_code, false);
446 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
448 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
450 kvm_multiple_exception(vcpu, nr, true, error_code, true);
452 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
455 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
456 * a #GP and return false.
458 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
460 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
462 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
465 EXPORT_SYMBOL_GPL(kvm_require_cpl);
467 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
469 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
472 kvm_queue_exception(vcpu, UD_VECTOR);
475 EXPORT_SYMBOL_GPL(kvm_require_dr);
478 * This function will be used to read from the physical memory of the currently
479 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
480 * can read from guest physical or from the guest's guest physical memory.
482 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
483 gfn_t ngfn, void *data, int offset, int len,
486 struct x86_exception exception;
490 ngpa = gfn_to_gpa(ngfn);
491 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
492 if (real_gfn == UNMAPPED_GVA)
495 real_gfn = gpa_to_gfn(real_gfn);
497 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
499 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
501 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
502 void *data, int offset, int len, u32 access)
504 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
505 data, offset, len, access);
509 * Load the pae pdptrs. Return true is they are all valid.
511 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
513 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
514 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
517 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
519 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
520 offset * sizeof(u64), sizeof(pdpte),
521 PFERR_USER_MASK|PFERR_WRITE_MASK);
526 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
527 if (is_present_gpte(pdpte[i]) &&
528 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
535 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
536 __set_bit(VCPU_EXREG_PDPTR,
537 (unsigned long *)&vcpu->arch.regs_avail);
538 __set_bit(VCPU_EXREG_PDPTR,
539 (unsigned long *)&vcpu->arch.regs_dirty);
544 EXPORT_SYMBOL_GPL(load_pdptrs);
546 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
548 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
554 if (is_long_mode(vcpu) || !is_pae(vcpu))
557 if (!test_bit(VCPU_EXREG_PDPTR,
558 (unsigned long *)&vcpu->arch.regs_avail))
561 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
562 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
563 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
564 PFERR_USER_MASK | PFERR_WRITE_MASK);
567 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
573 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
575 unsigned long old_cr0 = kvm_read_cr0(vcpu);
576 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
581 if (cr0 & 0xffffffff00000000UL)
585 cr0 &= ~CR0_RESERVED_BITS;
587 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
590 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
593 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
595 if ((vcpu->arch.efer & EFER_LME)) {
600 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
605 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
610 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
613 kvm_x86_ops->set_cr0(vcpu, cr0);
615 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
616 kvm_clear_async_pf_completion_queue(vcpu);
617 kvm_async_pf_hash_reset(vcpu);
620 if ((cr0 ^ old_cr0) & update_bits)
621 kvm_mmu_reset_context(vcpu);
623 if ((cr0 ^ old_cr0) & X86_CR0_CD)
624 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
628 EXPORT_SYMBOL_GPL(kvm_set_cr0);
630 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
632 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
634 EXPORT_SYMBOL_GPL(kvm_lmsw);
636 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
638 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
639 !vcpu->guest_xcr0_loaded) {
640 /* kvm_set_xcr() also depends on this */
641 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
642 vcpu->guest_xcr0_loaded = 1;
646 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
648 if (vcpu->guest_xcr0_loaded) {
649 if (vcpu->arch.xcr0 != host_xcr0)
650 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
651 vcpu->guest_xcr0_loaded = 0;
655 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
658 u64 old_xcr0 = vcpu->arch.xcr0;
661 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
662 if (index != XCR_XFEATURE_ENABLED_MASK)
664 if (!(xcr0 & XSTATE_FP))
666 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
670 * Do not allow the guest to set bits that we do not support
671 * saving. However, xcr0 bit 0 is always set, even if the
672 * emulated CPU does not support XSAVE (see fx_init).
674 valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
675 if (xcr0 & ~valid_bits)
678 if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
681 if (xcr0 & XSTATE_AVX512) {
682 if (!(xcr0 & XSTATE_YMM))
684 if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512)
687 kvm_put_guest_xcr0(vcpu);
688 vcpu->arch.xcr0 = xcr0;
690 if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
691 kvm_update_cpuid(vcpu);
695 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
697 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
698 __kvm_set_xcr(vcpu, index, xcr)) {
699 kvm_inject_gp(vcpu, 0);
704 EXPORT_SYMBOL_GPL(kvm_set_xcr);
706 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
708 unsigned long old_cr4 = kvm_read_cr4(vcpu);
709 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
710 X86_CR4_SMEP | X86_CR4_SMAP;
712 if (cr4 & CR4_RESERVED_BITS)
715 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
718 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
721 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
724 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
727 if (is_long_mode(vcpu)) {
728 if (!(cr4 & X86_CR4_PAE))
730 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
731 && ((cr4 ^ old_cr4) & pdptr_bits)
732 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
736 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
737 if (!guest_cpuid_has_pcid(vcpu))
740 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
741 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
745 if (kvm_x86_ops->set_cr4(vcpu, cr4))
748 if (((cr4 ^ old_cr4) & pdptr_bits) ||
749 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
750 kvm_mmu_reset_context(vcpu);
752 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
753 kvm_update_cpuid(vcpu);
757 EXPORT_SYMBOL_GPL(kvm_set_cr4);
759 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
762 cr3 &= ~CR3_PCID_INVD;
765 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
766 kvm_mmu_sync_roots(vcpu);
767 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
771 if (is_long_mode(vcpu)) {
772 if (cr3 & CR3_L_MODE_RESERVED_BITS)
774 } else if (is_pae(vcpu) && is_paging(vcpu) &&
775 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
778 vcpu->arch.cr3 = cr3;
779 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
780 kvm_mmu_new_cr3(vcpu);
783 EXPORT_SYMBOL_GPL(kvm_set_cr3);
785 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
787 if (cr8 & CR8_RESERVED_BITS)
789 if (irqchip_in_kernel(vcpu->kvm))
790 kvm_lapic_set_tpr(vcpu, cr8);
792 vcpu->arch.cr8 = cr8;
795 EXPORT_SYMBOL_GPL(kvm_set_cr8);
797 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
799 if (irqchip_in_kernel(vcpu->kvm))
800 return kvm_lapic_get_cr8(vcpu);
802 return vcpu->arch.cr8;
804 EXPORT_SYMBOL_GPL(kvm_get_cr8);
806 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
810 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
811 for (i = 0; i < KVM_NR_DB_REGS; i++)
812 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
813 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
817 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
819 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
820 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
823 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
827 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
828 dr7 = vcpu->arch.guest_debug_dr7;
830 dr7 = vcpu->arch.dr7;
831 kvm_x86_ops->set_dr7(vcpu, dr7);
832 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
833 if (dr7 & DR7_BP_EN_MASK)
834 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
837 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
839 u64 fixed = DR6_FIXED_1;
841 if (!guest_cpuid_has_rtm(vcpu))
846 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
850 vcpu->arch.db[dr] = val;
851 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
852 vcpu->arch.eff_db[dr] = val;
857 if (val & 0xffffffff00000000ULL)
859 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
860 kvm_update_dr6(vcpu);
865 if (val & 0xffffffff00000000ULL)
867 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
868 kvm_update_dr7(vcpu);
875 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
877 if (__kvm_set_dr(vcpu, dr, val)) {
878 kvm_inject_gp(vcpu, 0);
883 EXPORT_SYMBOL_GPL(kvm_set_dr);
885 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
889 *val = vcpu->arch.db[dr];
894 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
895 *val = vcpu->arch.dr6;
897 *val = kvm_x86_ops->get_dr6(vcpu);
902 *val = vcpu->arch.dr7;
907 EXPORT_SYMBOL_GPL(kvm_get_dr);
909 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
911 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
915 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
918 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
919 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
922 EXPORT_SYMBOL_GPL(kvm_rdpmc);
925 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
926 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
928 * This list is modified at module load time to reflect the
929 * capabilities of the host cpu. This capabilities test skips MSRs that are
930 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
931 * may depend on host virtualization features rather than host cpu features.
934 static u32 msrs_to_save[] = {
935 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
938 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
940 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
941 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
944 static unsigned num_msrs_to_save;
946 static u32 emulated_msrs[] = {
947 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
948 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
949 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
950 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
951 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
952 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
953 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
957 MSR_IA32_TSCDEADLINE,
958 MSR_IA32_MISC_ENABLE,
964 static unsigned num_emulated_msrs;
966 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
968 if (efer & efer_reserved_bits)
971 if (efer & EFER_FFXSR) {
972 struct kvm_cpuid_entry2 *feat;
974 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
975 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
979 if (efer & EFER_SVME) {
980 struct kvm_cpuid_entry2 *feat;
982 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
983 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
989 EXPORT_SYMBOL_GPL(kvm_valid_efer);
991 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
993 u64 old_efer = vcpu->arch.efer;
995 if (!kvm_valid_efer(vcpu, efer))
999 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1003 efer |= vcpu->arch.efer & EFER_LMA;
1005 kvm_x86_ops->set_efer(vcpu, efer);
1007 /* Update reserved bits */
1008 if ((efer ^ old_efer) & EFER_NX)
1009 kvm_mmu_reset_context(vcpu);
1014 void kvm_enable_efer_bits(u64 mask)
1016 efer_reserved_bits &= ~mask;
1018 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1021 * Writes msr value into into the appropriate "register".
1022 * Returns 0 on success, non-0 otherwise.
1023 * Assumes vcpu_load() was already called.
1025 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1027 switch (msr->index) {
1030 case MSR_KERNEL_GS_BASE:
1033 if (is_noncanonical_address(msr->data))
1036 case MSR_IA32_SYSENTER_EIP:
1037 case MSR_IA32_SYSENTER_ESP:
1039 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1040 * non-canonical address is written on Intel but not on
1041 * AMD (which ignores the top 32-bits, because it does
1042 * not implement 64-bit SYSENTER).
1044 * 64-bit code should hence be able to write a non-canonical
1045 * value on AMD. Making the address canonical ensures that
1046 * vmentry does not fail on Intel after writing a non-canonical
1047 * value, and that something deterministic happens if the guest
1048 * invokes 64-bit SYSENTER.
1050 msr->data = get_canonical(msr->data);
1052 return kvm_x86_ops->set_msr(vcpu, msr);
1054 EXPORT_SYMBOL_GPL(kvm_set_msr);
1057 * Adapt set_msr() to msr_io()'s calling convention
1059 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1061 struct msr_data msr;
1065 msr.host_initiated = true;
1066 r = kvm_get_msr(vcpu, &msr);
1074 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1076 struct msr_data msr;
1080 msr.host_initiated = true;
1081 return kvm_set_msr(vcpu, &msr);
1084 #ifdef CONFIG_X86_64
1085 struct pvclock_gtod_data {
1088 struct { /* extract of a clocksource struct */
1100 static struct pvclock_gtod_data pvclock_gtod_data;
1102 static void update_pvclock_gtod(struct timekeeper *tk)
1104 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1107 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1109 write_seqcount_begin(&vdata->seq);
1111 /* copy pvclock gtod data */
1112 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1113 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1114 vdata->clock.mask = tk->tkr_mono.mask;
1115 vdata->clock.mult = tk->tkr_mono.mult;
1116 vdata->clock.shift = tk->tkr_mono.shift;
1118 vdata->boot_ns = boot_ns;
1119 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1121 write_seqcount_end(&vdata->seq);
1125 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1128 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1129 * vcpu_enter_guest. This function is only called from
1130 * the physical CPU that is running vcpu.
1132 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1135 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1139 struct pvclock_wall_clock wc;
1140 struct timespec boot;
1145 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1150 ++version; /* first time write, random junk */
1154 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1157 * The guest calculates current wall clock time by adding
1158 * system time (updated by kvm_guest_time_update below) to the
1159 * wall clock specified here. guest system time equals host
1160 * system time for us, thus we must fill in host boot time here.
1164 if (kvm->arch.kvmclock_offset) {
1165 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1166 boot = timespec_sub(boot, ts);
1168 wc.sec = boot.tv_sec;
1169 wc.nsec = boot.tv_nsec;
1170 wc.version = version;
1172 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1175 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1178 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1180 uint32_t quotient, remainder;
1182 /* Don't try to replace with do_div(), this one calculates
1183 * "(dividend << 32) / divisor" */
1185 : "=a" (quotient), "=d" (remainder)
1186 : "0" (0), "1" (dividend), "r" (divisor) );
1190 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1191 s8 *pshift, u32 *pmultiplier)
1198 tps64 = base_khz * 1000LL;
1199 scaled64 = scaled_khz * 1000LL;
1200 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1205 tps32 = (uint32_t)tps64;
1206 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1207 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1215 *pmultiplier = div_frac(scaled64, tps32);
1217 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1218 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1221 #ifdef CONFIG_X86_64
1222 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1225 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1226 static unsigned long max_tsc_khz;
1228 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1230 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1231 vcpu->arch.virtual_tsc_shift);
1234 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1236 u64 v = (u64)khz * (1000000 + ppm);
1241 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1243 u32 thresh_lo, thresh_hi;
1244 int use_scaling = 0;
1246 /* tsc_khz can be zero if TSC calibration fails */
1247 if (this_tsc_khz == 0)
1250 /* Compute a scale to convert nanoseconds in TSC cycles */
1251 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1252 &vcpu->arch.virtual_tsc_shift,
1253 &vcpu->arch.virtual_tsc_mult);
1254 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1257 * Compute the variation in TSC rate which is acceptable
1258 * within the range of tolerance and decide if the
1259 * rate being applied is within that bounds of the hardware
1260 * rate. If so, no scaling or compensation need be done.
1262 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1263 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1264 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1265 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1268 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1271 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1273 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1274 vcpu->arch.virtual_tsc_mult,
1275 vcpu->arch.virtual_tsc_shift);
1276 tsc += vcpu->arch.this_tsc_write;
1280 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1282 #ifdef CONFIG_X86_64
1284 struct kvm_arch *ka = &vcpu->kvm->arch;
1285 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1287 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1288 atomic_read(&vcpu->kvm->online_vcpus));
1291 * Once the masterclock is enabled, always perform request in
1292 * order to update it.
1294 * In order to enable masterclock, the host clocksource must be TSC
1295 * and the vcpus need to have matched TSCs. When that happens,
1296 * perform request to enable masterclock.
1298 if (ka->use_master_clock ||
1299 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1300 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1302 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1303 atomic_read(&vcpu->kvm->online_vcpus),
1304 ka->use_master_clock, gtod->clock.vclock_mode);
1308 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1310 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1311 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1314 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1316 struct kvm *kvm = vcpu->kvm;
1317 u64 offset, ns, elapsed;
1318 unsigned long flags;
1321 bool already_matched;
1322 u64 data = msr->data;
1324 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1325 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1326 ns = get_kernel_ns();
1327 elapsed = ns - kvm->arch.last_tsc_nsec;
1329 if (vcpu->arch.virtual_tsc_khz) {
1332 /* n.b - signed multiplication and division required */
1333 usdiff = data - kvm->arch.last_tsc_write;
1334 #ifdef CONFIG_X86_64
1335 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1337 /* do_div() only does unsigned */
1338 asm("1: idivl %[divisor]\n"
1339 "2: xor %%edx, %%edx\n"
1340 " movl $0, %[faulted]\n"
1342 ".section .fixup,\"ax\"\n"
1343 "4: movl $1, %[faulted]\n"
1347 _ASM_EXTABLE(1b, 4b)
1349 : "=A"(usdiff), [faulted] "=r" (faulted)
1350 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1353 do_div(elapsed, 1000);
1358 /* idivl overflow => difference is larger than USEC_PER_SEC */
1360 usdiff = USEC_PER_SEC;
1362 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1365 * Special case: TSC write with a small delta (1 second) of virtual
1366 * cycle time against real time is interpreted as an attempt to
1367 * synchronize the CPU.
1369 * For a reliable TSC, we can match TSC offsets, and for an unstable
1370 * TSC, we add elapsed time in this computation. We could let the
1371 * compensation code attempt to catch up if we fall behind, but
1372 * it's better to try to match offsets from the beginning.
1374 if (usdiff < USEC_PER_SEC &&
1375 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1376 if (!check_tsc_unstable()) {
1377 offset = kvm->arch.cur_tsc_offset;
1378 pr_debug("kvm: matched tsc offset for %llu\n", data);
1380 u64 delta = nsec_to_cycles(vcpu, elapsed);
1382 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1383 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1386 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1389 * We split periods of matched TSC writes into generations.
1390 * For each generation, we track the original measured
1391 * nanosecond time, offset, and write, so if TSCs are in
1392 * sync, we can match exact offset, and if not, we can match
1393 * exact software computation in compute_guest_tsc()
1395 * These values are tracked in kvm->arch.cur_xxx variables.
1397 kvm->arch.cur_tsc_generation++;
1398 kvm->arch.cur_tsc_nsec = ns;
1399 kvm->arch.cur_tsc_write = data;
1400 kvm->arch.cur_tsc_offset = offset;
1402 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1403 kvm->arch.cur_tsc_generation, data);
1407 * We also track th most recent recorded KHZ, write and time to
1408 * allow the matching interval to be extended at each write.
1410 kvm->arch.last_tsc_nsec = ns;
1411 kvm->arch.last_tsc_write = data;
1412 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1414 vcpu->arch.last_guest_tsc = data;
1416 /* Keep track of which generation this VCPU has synchronized to */
1417 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1418 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1419 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1421 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1422 update_ia32_tsc_adjust_msr(vcpu, offset);
1423 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1424 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1426 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1428 kvm->arch.nr_vcpus_matched_tsc = 0;
1429 } else if (!already_matched) {
1430 kvm->arch.nr_vcpus_matched_tsc++;
1433 kvm_track_tsc_matching(vcpu);
1434 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1437 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1439 #ifdef CONFIG_X86_64
1441 static cycle_t read_tsc(void)
1447 * Empirically, a fence (of type that depends on the CPU)
1448 * before rdtsc is enough to ensure that rdtsc is ordered
1449 * with respect to loads. The various CPU manuals are unclear
1450 * as to whether rdtsc can be reordered with later loads,
1451 * but no one has ever seen it happen.
1454 ret = (cycle_t)vget_cycles();
1456 last = pvclock_gtod_data.clock.cycle_last;
1458 if (likely(ret >= last))
1462 * GCC likes to generate cmov here, but this branch is extremely
1463 * predictable (it's just a funciton of time and the likely is
1464 * very likely) and there's a data dependence, so force GCC
1465 * to generate a branch instead. I don't barrier() because
1466 * we don't actually need a barrier, and if this function
1467 * ever gets inlined it will generate worse code.
1473 static inline u64 vgettsc(cycle_t *cycle_now)
1476 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1478 *cycle_now = read_tsc();
1480 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1481 return v * gtod->clock.mult;
1484 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1486 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1492 seq = read_seqcount_begin(>od->seq);
1493 mode = gtod->clock.vclock_mode;
1494 ns = gtod->nsec_base;
1495 ns += vgettsc(cycle_now);
1496 ns >>= gtod->clock.shift;
1497 ns += gtod->boot_ns;
1498 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1504 /* returns true if host is using tsc clocksource */
1505 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1507 /* checked again under seqlock below */
1508 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1511 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1517 * Assuming a stable TSC across physical CPUS, and a stable TSC
1518 * across virtual CPUs, the following condition is possible.
1519 * Each numbered line represents an event visible to both
1520 * CPUs at the next numbered event.
1522 * "timespecX" represents host monotonic time. "tscX" represents
1525 * VCPU0 on CPU0 | VCPU1 on CPU1
1527 * 1. read timespec0,tsc0
1528 * 2. | timespec1 = timespec0 + N
1530 * 3. transition to guest | transition to guest
1531 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1532 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1533 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1535 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1538 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1540 * - 0 < N - M => M < N
1542 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1543 * always the case (the difference between two distinct xtime instances
1544 * might be smaller then the difference between corresponding TSC reads,
1545 * when updating guest vcpus pvclock areas).
1547 * To avoid that problem, do not allow visibility of distinct
1548 * system_timestamp/tsc_timestamp values simultaneously: use a master
1549 * copy of host monotonic time values. Update that master copy
1552 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1556 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1558 #ifdef CONFIG_X86_64
1559 struct kvm_arch *ka = &kvm->arch;
1561 bool host_tsc_clocksource, vcpus_matched;
1563 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1564 atomic_read(&kvm->online_vcpus));
1567 * If the host uses TSC clock, then passthrough TSC as stable
1570 host_tsc_clocksource = kvm_get_time_and_clockread(
1571 &ka->master_kernel_ns,
1572 &ka->master_cycle_now);
1574 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1575 && !backwards_tsc_observed
1576 && !ka->boot_vcpu_runs_old_kvmclock;
1578 if (ka->use_master_clock)
1579 atomic_set(&kvm_guest_has_master_clock, 1);
1581 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1582 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1587 static void kvm_gen_update_masterclock(struct kvm *kvm)
1589 #ifdef CONFIG_X86_64
1591 struct kvm_vcpu *vcpu;
1592 struct kvm_arch *ka = &kvm->arch;
1594 spin_lock(&ka->pvclock_gtod_sync_lock);
1595 kvm_make_mclock_inprogress_request(kvm);
1596 /* no guest entries from this point */
1597 pvclock_update_vm_gtod_copy(kvm);
1599 kvm_for_each_vcpu(i, vcpu, kvm)
1600 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1602 /* guest entries allowed */
1603 kvm_for_each_vcpu(i, vcpu, kvm)
1604 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1606 spin_unlock(&ka->pvclock_gtod_sync_lock);
1610 static int kvm_guest_time_update(struct kvm_vcpu *v)
1612 unsigned long flags, this_tsc_khz;
1613 struct kvm_vcpu_arch *vcpu = &v->arch;
1614 struct kvm_arch *ka = &v->kvm->arch;
1616 u64 tsc_timestamp, host_tsc;
1617 struct pvclock_vcpu_time_info guest_hv_clock;
1619 bool use_master_clock;
1625 * If the host uses TSC clock, then passthrough TSC as stable
1628 spin_lock(&ka->pvclock_gtod_sync_lock);
1629 use_master_clock = ka->use_master_clock;
1630 if (use_master_clock) {
1631 host_tsc = ka->master_cycle_now;
1632 kernel_ns = ka->master_kernel_ns;
1634 spin_unlock(&ka->pvclock_gtod_sync_lock);
1636 /* Keep irq disabled to prevent changes to the clock */
1637 local_irq_save(flags);
1638 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1639 if (unlikely(this_tsc_khz == 0)) {
1640 local_irq_restore(flags);
1641 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1644 if (!use_master_clock) {
1645 host_tsc = native_read_tsc();
1646 kernel_ns = get_kernel_ns();
1649 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1652 * We may have to catch up the TSC to match elapsed wall clock
1653 * time for two reasons, even if kvmclock is used.
1654 * 1) CPU could have been running below the maximum TSC rate
1655 * 2) Broken TSC compensation resets the base at each VCPU
1656 * entry to avoid unknown leaps of TSC even when running
1657 * again on the same CPU. This may cause apparent elapsed
1658 * time to disappear, and the guest to stand still or run
1661 if (vcpu->tsc_catchup) {
1662 u64 tsc = compute_guest_tsc(v, kernel_ns);
1663 if (tsc > tsc_timestamp) {
1664 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1665 tsc_timestamp = tsc;
1669 local_irq_restore(flags);
1671 if (!vcpu->pv_time_enabled)
1674 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1675 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1676 &vcpu->hv_clock.tsc_shift,
1677 &vcpu->hv_clock.tsc_to_system_mul);
1678 vcpu->hw_tsc_khz = this_tsc_khz;
1681 /* With all the info we got, fill in the values */
1682 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1683 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1684 vcpu->last_guest_tsc = tsc_timestamp;
1686 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1687 &guest_hv_clock, sizeof(guest_hv_clock))))
1690 /* This VCPU is paused, but it's legal for a guest to read another
1691 * VCPU's kvmclock, so we really have to follow the specification where
1692 * it says that version is odd if data is being modified, and even after
1695 * Version field updates must be kept separate. This is because
1696 * kvm_write_guest_cached might use a "rep movs" instruction, and
1697 * writes within a string instruction are weakly ordered. So there
1698 * are three writes overall.
1700 * As a small optimization, only write the version field in the first
1701 * and third write. The vcpu->pv_time cache is still valid, because the
1702 * version field is the first in the struct.
1704 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1706 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1707 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1709 sizeof(vcpu->hv_clock.version));
1713 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1714 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1716 if (vcpu->pvclock_set_guest_stopped_request) {
1717 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1718 vcpu->pvclock_set_guest_stopped_request = false;
1721 pvclock_flags |= PVCLOCK_COUNTS_FROM_ZERO;
1723 /* If the host uses TSC clocksource, then it is stable */
1724 if (use_master_clock)
1725 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1727 vcpu->hv_clock.flags = pvclock_flags;
1729 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1731 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1733 sizeof(vcpu->hv_clock));
1737 vcpu->hv_clock.version++;
1738 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1740 sizeof(vcpu->hv_clock.version));
1745 * kvmclock updates which are isolated to a given vcpu, such as
1746 * vcpu->cpu migration, should not allow system_timestamp from
1747 * the rest of the vcpus to remain static. Otherwise ntp frequency
1748 * correction applies to one vcpu's system_timestamp but not
1751 * So in those cases, request a kvmclock update for all vcpus.
1752 * We need to rate-limit these requests though, as they can
1753 * considerably slow guests that have a large number of vcpus.
1754 * The time for a remote vcpu to update its kvmclock is bound
1755 * by the delay we use to rate-limit the updates.
1758 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1760 static void kvmclock_update_fn(struct work_struct *work)
1763 struct delayed_work *dwork = to_delayed_work(work);
1764 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1765 kvmclock_update_work);
1766 struct kvm *kvm = container_of(ka, struct kvm, arch);
1767 struct kvm_vcpu *vcpu;
1769 kvm_for_each_vcpu(i, vcpu, kvm) {
1770 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1771 kvm_vcpu_kick(vcpu);
1775 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1777 struct kvm *kvm = v->kvm;
1779 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1780 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1781 KVMCLOCK_UPDATE_DELAY);
1784 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1786 static void kvmclock_sync_fn(struct work_struct *work)
1788 struct delayed_work *dwork = to_delayed_work(work);
1789 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1790 kvmclock_sync_work);
1791 struct kvm *kvm = container_of(ka, struct kvm, arch);
1793 if (!kvmclock_periodic_sync)
1796 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1797 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1798 KVMCLOCK_SYNC_PERIOD);
1801 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1803 u64 mcg_cap = vcpu->arch.mcg_cap;
1804 unsigned bank_num = mcg_cap & 0xff;
1807 case MSR_IA32_MCG_STATUS:
1808 vcpu->arch.mcg_status = data;
1810 case MSR_IA32_MCG_CTL:
1811 if (!(mcg_cap & MCG_CTL_P))
1813 if (data != 0 && data != ~(u64)0)
1815 vcpu->arch.mcg_ctl = data;
1818 if (msr >= MSR_IA32_MC0_CTL &&
1819 msr < MSR_IA32_MCx_CTL(bank_num)) {
1820 u32 offset = msr - MSR_IA32_MC0_CTL;
1821 /* only 0 or all 1s can be written to IA32_MCi_CTL
1822 * some Linux kernels though clear bit 10 in bank 4 to
1823 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1824 * this to avoid an uncatched #GP in the guest
1826 if ((offset & 0x3) == 0 &&
1827 data != 0 && (data | (1 << 10)) != ~(u64)0)
1829 vcpu->arch.mce_banks[offset] = data;
1837 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1839 struct kvm *kvm = vcpu->kvm;
1840 int lm = is_long_mode(vcpu);
1841 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1842 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1843 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1844 : kvm->arch.xen_hvm_config.blob_size_32;
1845 u32 page_num = data & ~PAGE_MASK;
1846 u64 page_addr = data & PAGE_MASK;
1851 if (page_num >= blob_size)
1854 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1859 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1868 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1870 gpa_t gpa = data & ~0x3f;
1872 /* Bits 2:5 are reserved, Should be zero */
1876 vcpu->arch.apf.msr_val = data;
1878 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1879 kvm_clear_async_pf_completion_queue(vcpu);
1880 kvm_async_pf_hash_reset(vcpu);
1884 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1888 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1889 kvm_async_pf_wakeup_all(vcpu);
1893 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1895 vcpu->arch.pv_time_enabled = false;
1898 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1902 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1905 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1906 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1907 vcpu->arch.st.accum_steal = delta;
1910 static void record_steal_time(struct kvm_vcpu *vcpu)
1912 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1915 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1916 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1919 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1920 vcpu->arch.st.steal.version += 2;
1921 vcpu->arch.st.accum_steal = 0;
1923 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1924 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1927 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1930 u32 msr = msr_info->index;
1931 u64 data = msr_info->data;
1934 case MSR_AMD64_NB_CFG:
1935 case MSR_IA32_UCODE_REV:
1936 case MSR_IA32_UCODE_WRITE:
1937 case MSR_VM_HSAVE_PA:
1938 case MSR_AMD64_PATCH_LOADER:
1939 case MSR_AMD64_BU_CFG2:
1943 return set_efer(vcpu, data);
1945 data &= ~(u64)0x40; /* ignore flush filter disable */
1946 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1947 data &= ~(u64)0x8; /* ignore TLB cache disable */
1948 data &= ~(u64)0x40000; /* ignore Mc status write enable */
1950 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1955 case MSR_FAM10H_MMIO_CONF_BASE:
1957 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1962 case MSR_IA32_DEBUGCTLMSR:
1964 /* We support the non-activated case already */
1966 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1967 /* Values other than LBR and BTF are vendor-specific,
1968 thus reserved and should throw a #GP */
1971 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1974 case 0x200 ... 0x2ff:
1975 return kvm_mtrr_set_msr(vcpu, msr, data);
1976 case MSR_IA32_APICBASE:
1977 return kvm_set_apic_base(vcpu, msr_info);
1978 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1979 return kvm_x2apic_msr_write(vcpu, msr, data);
1980 case MSR_IA32_TSCDEADLINE:
1981 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1983 case MSR_IA32_TSC_ADJUST:
1984 if (guest_cpuid_has_tsc_adjust(vcpu)) {
1985 if (!msr_info->host_initiated) {
1986 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
1987 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
1989 vcpu->arch.ia32_tsc_adjust_msr = data;
1992 case MSR_IA32_MISC_ENABLE:
1993 vcpu->arch.ia32_misc_enable_msr = data;
1995 case MSR_IA32_SMBASE:
1996 if (!msr_info->host_initiated)
1998 vcpu->arch.smbase = data;
2000 case MSR_KVM_WALL_CLOCK_NEW:
2001 case MSR_KVM_WALL_CLOCK:
2002 vcpu->kvm->arch.wall_clock = data;
2003 kvm_write_wall_clock(vcpu->kvm, data);
2005 case MSR_KVM_SYSTEM_TIME_NEW:
2006 case MSR_KVM_SYSTEM_TIME: {
2008 struct kvm_arch *ka = &vcpu->kvm->arch;
2010 kvmclock_reset(vcpu);
2012 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2013 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2015 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2016 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2019 ka->boot_vcpu_runs_old_kvmclock = tmp;
2021 ka->kvmclock_offset = -get_kernel_ns();
2024 vcpu->arch.time = data;
2025 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2027 /* we verify if the enable bit is set... */
2031 gpa_offset = data & ~(PAGE_MASK | 1);
2033 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2034 &vcpu->arch.pv_time, data & ~1ULL,
2035 sizeof(struct pvclock_vcpu_time_info)))
2036 vcpu->arch.pv_time_enabled = false;
2038 vcpu->arch.pv_time_enabled = true;
2042 case MSR_KVM_ASYNC_PF_EN:
2043 if (kvm_pv_enable_async_pf(vcpu, data))
2046 case MSR_KVM_STEAL_TIME:
2048 if (unlikely(!sched_info_on()))
2051 if (data & KVM_STEAL_RESERVED_MASK)
2054 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2055 data & KVM_STEAL_VALID_BITS,
2056 sizeof(struct kvm_steal_time)))
2059 vcpu->arch.st.msr_val = data;
2061 if (!(data & KVM_MSR_ENABLED))
2064 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2067 accumulate_steal_time(vcpu);
2070 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2073 case MSR_KVM_PV_EOI_EN:
2074 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2078 case MSR_IA32_MCG_CTL:
2079 case MSR_IA32_MCG_STATUS:
2080 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2081 return set_msr_mce(vcpu, msr, data);
2083 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2084 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2085 pr = true; /* fall through */
2086 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2087 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2088 if (kvm_pmu_is_valid_msr(vcpu, msr))
2089 return kvm_pmu_set_msr(vcpu, msr_info);
2091 if (pr || data != 0)
2092 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2093 "0x%x data 0x%llx\n", msr, data);
2095 case MSR_K7_CLK_CTL:
2097 * Ignore all writes to this no longer documented MSR.
2098 * Writes are only relevant for old K7 processors,
2099 * all pre-dating SVM, but a recommended workaround from
2100 * AMD for these chips. It is possible to specify the
2101 * affected processor models on the command line, hence
2102 * the need to ignore the workaround.
2105 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2106 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2107 case HV_X64_MSR_CRASH_CTL:
2108 return kvm_hv_set_msr_common(vcpu, msr, data,
2109 msr_info->host_initiated);
2110 case MSR_IA32_BBL_CR_CTL3:
2111 /* Drop writes to this legacy MSR -- see rdmsr
2112 * counterpart for further detail.
2114 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2116 case MSR_AMD64_OSVW_ID_LENGTH:
2117 if (!guest_cpuid_has_osvw(vcpu))
2119 vcpu->arch.osvw.length = data;
2121 case MSR_AMD64_OSVW_STATUS:
2122 if (!guest_cpuid_has_osvw(vcpu))
2124 vcpu->arch.osvw.status = data;
2127 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2128 return xen_hvm_config(vcpu, data);
2129 if (kvm_pmu_is_valid_msr(vcpu, msr))
2130 return kvm_pmu_set_msr(vcpu, msr_info);
2132 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2136 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2143 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2147 * Reads an msr value (of 'msr_index') into 'pdata'.
2148 * Returns 0 on success, non-0 otherwise.
2149 * Assumes vcpu_load() was already called.
2151 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2153 return kvm_x86_ops->get_msr(vcpu, msr);
2155 EXPORT_SYMBOL_GPL(kvm_get_msr);
2157 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2160 u64 mcg_cap = vcpu->arch.mcg_cap;
2161 unsigned bank_num = mcg_cap & 0xff;
2164 case MSR_IA32_P5_MC_ADDR:
2165 case MSR_IA32_P5_MC_TYPE:
2168 case MSR_IA32_MCG_CAP:
2169 data = vcpu->arch.mcg_cap;
2171 case MSR_IA32_MCG_CTL:
2172 if (!(mcg_cap & MCG_CTL_P))
2174 data = vcpu->arch.mcg_ctl;
2176 case MSR_IA32_MCG_STATUS:
2177 data = vcpu->arch.mcg_status;
2180 if (msr >= MSR_IA32_MC0_CTL &&
2181 msr < MSR_IA32_MCx_CTL(bank_num)) {
2182 u32 offset = msr - MSR_IA32_MC0_CTL;
2183 data = vcpu->arch.mce_banks[offset];
2192 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2194 switch (msr_info->index) {
2195 case MSR_IA32_PLATFORM_ID:
2196 case MSR_IA32_EBL_CR_POWERON:
2197 case MSR_IA32_DEBUGCTLMSR:
2198 case MSR_IA32_LASTBRANCHFROMIP:
2199 case MSR_IA32_LASTBRANCHTOIP:
2200 case MSR_IA32_LASTINTFROMIP:
2201 case MSR_IA32_LASTINTTOIP:
2204 case MSR_VM_HSAVE_PA:
2205 case MSR_K8_INT_PENDING_MSG:
2206 case MSR_AMD64_NB_CFG:
2207 case MSR_FAM10H_MMIO_CONF_BASE:
2208 case MSR_AMD64_BU_CFG2:
2211 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2212 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2213 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2214 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2215 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2216 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2219 case MSR_IA32_UCODE_REV:
2220 msr_info->data = 0x100000000ULL;
2223 case 0x200 ... 0x2ff:
2224 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2225 case 0xcd: /* fsb frequency */
2229 * MSR_EBC_FREQUENCY_ID
2230 * Conservative value valid for even the basic CPU models.
2231 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2232 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2233 * and 266MHz for model 3, or 4. Set Core Clock
2234 * Frequency to System Bus Frequency Ratio to 1 (bits
2235 * 31:24) even though these are only valid for CPU
2236 * models > 2, however guests may end up dividing or
2237 * multiplying by zero otherwise.
2239 case MSR_EBC_FREQUENCY_ID:
2240 msr_info->data = 1 << 24;
2242 case MSR_IA32_APICBASE:
2243 msr_info->data = kvm_get_apic_base(vcpu);
2245 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2246 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2248 case MSR_IA32_TSCDEADLINE:
2249 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2251 case MSR_IA32_TSC_ADJUST:
2252 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2254 case MSR_IA32_MISC_ENABLE:
2255 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2257 case MSR_IA32_SMBASE:
2258 if (!msr_info->host_initiated)
2260 msr_info->data = vcpu->arch.smbase;
2262 case MSR_IA32_PERF_STATUS:
2263 /* TSC increment by tick */
2264 msr_info->data = 1000ULL;
2265 /* CPU multiplier */
2266 msr_info->data |= (((uint64_t)4ULL) << 40);
2269 msr_info->data = vcpu->arch.efer;
2271 case MSR_KVM_WALL_CLOCK:
2272 case MSR_KVM_WALL_CLOCK_NEW:
2273 msr_info->data = vcpu->kvm->arch.wall_clock;
2275 case MSR_KVM_SYSTEM_TIME:
2276 case MSR_KVM_SYSTEM_TIME_NEW:
2277 msr_info->data = vcpu->arch.time;
2279 case MSR_KVM_ASYNC_PF_EN:
2280 msr_info->data = vcpu->arch.apf.msr_val;
2282 case MSR_KVM_STEAL_TIME:
2283 msr_info->data = vcpu->arch.st.msr_val;
2285 case MSR_KVM_PV_EOI_EN:
2286 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2288 case MSR_IA32_P5_MC_ADDR:
2289 case MSR_IA32_P5_MC_TYPE:
2290 case MSR_IA32_MCG_CAP:
2291 case MSR_IA32_MCG_CTL:
2292 case MSR_IA32_MCG_STATUS:
2293 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2294 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2295 case MSR_K7_CLK_CTL:
2297 * Provide expected ramp-up count for K7. All other
2298 * are set to zero, indicating minimum divisors for
2301 * This prevents guest kernels on AMD host with CPU
2302 * type 6, model 8 and higher from exploding due to
2303 * the rdmsr failing.
2305 msr_info->data = 0x20000000;
2307 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2308 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2309 case HV_X64_MSR_CRASH_CTL:
2310 return kvm_hv_get_msr_common(vcpu,
2311 msr_info->index, &msr_info->data);
2313 case MSR_IA32_BBL_CR_CTL3:
2314 /* This legacy MSR exists but isn't fully documented in current
2315 * silicon. It is however accessed by winxp in very narrow
2316 * scenarios where it sets bit #19, itself documented as
2317 * a "reserved" bit. Best effort attempt to source coherent
2318 * read data here should the balance of the register be
2319 * interpreted by the guest:
2321 * L2 cache control register 3: 64GB range, 256KB size,
2322 * enabled, latency 0x1, configured
2324 msr_info->data = 0xbe702111;
2326 case MSR_AMD64_OSVW_ID_LENGTH:
2327 if (!guest_cpuid_has_osvw(vcpu))
2329 msr_info->data = vcpu->arch.osvw.length;
2331 case MSR_AMD64_OSVW_STATUS:
2332 if (!guest_cpuid_has_osvw(vcpu))
2334 msr_info->data = vcpu->arch.osvw.status;
2337 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2338 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2340 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2343 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2350 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2353 * Read or write a bunch of msrs. All parameters are kernel addresses.
2355 * @return number of msrs set successfully.
2357 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2358 struct kvm_msr_entry *entries,
2359 int (*do_msr)(struct kvm_vcpu *vcpu,
2360 unsigned index, u64 *data))
2364 idx = srcu_read_lock(&vcpu->kvm->srcu);
2365 for (i = 0; i < msrs->nmsrs; ++i)
2366 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2368 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2374 * Read or write a bunch of msrs. Parameters are user addresses.
2376 * @return number of msrs set successfully.
2378 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2379 int (*do_msr)(struct kvm_vcpu *vcpu,
2380 unsigned index, u64 *data),
2383 struct kvm_msrs msrs;
2384 struct kvm_msr_entry *entries;
2389 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2393 if (msrs.nmsrs >= MAX_IO_MSRS)
2396 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2397 entries = memdup_user(user_msrs->entries, size);
2398 if (IS_ERR(entries)) {
2399 r = PTR_ERR(entries);
2403 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2408 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2419 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2424 case KVM_CAP_IRQCHIP:
2426 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2427 case KVM_CAP_SET_TSS_ADDR:
2428 case KVM_CAP_EXT_CPUID:
2429 case KVM_CAP_EXT_EMUL_CPUID:
2430 case KVM_CAP_CLOCKSOURCE:
2432 case KVM_CAP_NOP_IO_DELAY:
2433 case KVM_CAP_MP_STATE:
2434 case KVM_CAP_SYNC_MMU:
2435 case KVM_CAP_USER_NMI:
2436 case KVM_CAP_REINJECT_CONTROL:
2437 case KVM_CAP_IRQ_INJECT_STATUS:
2438 case KVM_CAP_IOEVENTFD:
2439 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2441 case KVM_CAP_PIT_STATE2:
2442 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2443 case KVM_CAP_XEN_HVM:
2444 case KVM_CAP_ADJUST_CLOCK:
2445 case KVM_CAP_VCPU_EVENTS:
2446 case KVM_CAP_HYPERV:
2447 case KVM_CAP_HYPERV_VAPIC:
2448 case KVM_CAP_HYPERV_SPIN:
2449 case KVM_CAP_PCI_SEGMENT:
2450 case KVM_CAP_DEBUGREGS:
2451 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2453 case KVM_CAP_ASYNC_PF:
2454 case KVM_CAP_GET_TSC_KHZ:
2455 case KVM_CAP_KVMCLOCK_CTRL:
2456 case KVM_CAP_READONLY_MEM:
2457 case KVM_CAP_HYPERV_TIME:
2458 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2459 case KVM_CAP_TSC_DEADLINE_TIMER:
2460 case KVM_CAP_ENABLE_CAP_VM:
2461 case KVM_CAP_DISABLE_QUIRKS:
2462 case KVM_CAP_SET_BOOT_CPU_ID:
2463 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2464 case KVM_CAP_ASSIGN_DEV_IRQ:
2465 case KVM_CAP_PCI_2_3:
2469 case KVM_CAP_X86_SMM:
2470 /* SMBASE is usually relocated above 1M on modern chipsets,
2471 * and SMM handlers might indeed rely on 4G segment limits,
2472 * so do not report SMM to be available if real mode is
2473 * emulated via vm86 mode. Still, do not go to great lengths
2474 * to avoid userspace's usage of the feature, because it is a
2475 * fringe case that is not enabled except via specific settings
2476 * of the module parameters.
2478 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2480 case KVM_CAP_COALESCED_MMIO:
2481 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2484 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2486 case KVM_CAP_NR_VCPUS:
2487 r = KVM_SOFT_MAX_VCPUS;
2489 case KVM_CAP_MAX_VCPUS:
2492 case KVM_CAP_NR_MEMSLOTS:
2493 r = KVM_USER_MEM_SLOTS;
2495 case KVM_CAP_PV_MMU: /* obsolete */
2498 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2500 r = iommu_present(&pci_bus_type);
2504 r = KVM_MAX_MCE_BANKS;
2509 case KVM_CAP_TSC_CONTROL:
2510 r = kvm_has_tsc_control;
2520 long kvm_arch_dev_ioctl(struct file *filp,
2521 unsigned int ioctl, unsigned long arg)
2523 void __user *argp = (void __user *)arg;
2527 case KVM_GET_MSR_INDEX_LIST: {
2528 struct kvm_msr_list __user *user_msr_list = argp;
2529 struct kvm_msr_list msr_list;
2533 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2536 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2537 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2540 if (n < msr_list.nmsrs)
2543 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2544 num_msrs_to_save * sizeof(u32)))
2546 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2548 num_emulated_msrs * sizeof(u32)))
2553 case KVM_GET_SUPPORTED_CPUID:
2554 case KVM_GET_EMULATED_CPUID: {
2555 struct kvm_cpuid2 __user *cpuid_arg = argp;
2556 struct kvm_cpuid2 cpuid;
2559 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2562 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2568 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2573 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2576 mce_cap = KVM_MCE_CAP_SUPPORTED;
2578 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2590 static void wbinvd_ipi(void *garbage)
2595 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2597 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2600 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2602 /* Address WBINVD may be executed by guest */
2603 if (need_emulate_wbinvd(vcpu)) {
2604 if (kvm_x86_ops->has_wbinvd_exit())
2605 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2606 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2607 smp_call_function_single(vcpu->cpu,
2608 wbinvd_ipi, NULL, 1);
2611 kvm_x86_ops->vcpu_load(vcpu, cpu);
2613 /* Apply any externally detected TSC adjustments (due to suspend) */
2614 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2615 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2616 vcpu->arch.tsc_offset_adjustment = 0;
2617 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2620 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2621 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2622 native_read_tsc() - vcpu->arch.last_host_tsc;
2624 mark_tsc_unstable("KVM discovered backwards TSC");
2625 if (check_tsc_unstable()) {
2626 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2627 vcpu->arch.last_guest_tsc);
2628 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2629 vcpu->arch.tsc_catchup = 1;
2632 * On a host with synchronized TSC, there is no need to update
2633 * kvmclock on vcpu->cpu migration
2635 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2636 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2637 if (vcpu->cpu != cpu)
2638 kvm_migrate_timers(vcpu);
2642 accumulate_steal_time(vcpu);
2643 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2646 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2648 kvm_x86_ops->vcpu_put(vcpu);
2649 kvm_put_guest_fpu(vcpu);
2650 vcpu->arch.last_host_tsc = native_read_tsc();
2653 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2654 struct kvm_lapic_state *s)
2656 kvm_x86_ops->sync_pir_to_irr(vcpu);
2657 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2662 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2663 struct kvm_lapic_state *s)
2665 kvm_apic_post_state_restore(vcpu, s);
2666 update_cr8_intercept(vcpu);
2671 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2672 struct kvm_interrupt *irq)
2674 if (irq->irq >= KVM_NR_INTERRUPTS)
2676 if (irqchip_in_kernel(vcpu->kvm))
2679 kvm_queue_interrupt(vcpu, irq->irq, false);
2680 kvm_make_request(KVM_REQ_EVENT, vcpu);
2685 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2687 kvm_inject_nmi(vcpu);
2692 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2694 kvm_make_request(KVM_REQ_SMI, vcpu);
2699 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2700 struct kvm_tpr_access_ctl *tac)
2704 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2708 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2712 unsigned bank_num = mcg_cap & 0xff, bank;
2715 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2717 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2720 vcpu->arch.mcg_cap = mcg_cap;
2721 /* Init IA32_MCG_CTL to all 1s */
2722 if (mcg_cap & MCG_CTL_P)
2723 vcpu->arch.mcg_ctl = ~(u64)0;
2724 /* Init IA32_MCi_CTL to all 1s */
2725 for (bank = 0; bank < bank_num; bank++)
2726 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2731 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2732 struct kvm_x86_mce *mce)
2734 u64 mcg_cap = vcpu->arch.mcg_cap;
2735 unsigned bank_num = mcg_cap & 0xff;
2736 u64 *banks = vcpu->arch.mce_banks;
2738 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2741 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2742 * reporting is disabled
2744 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2745 vcpu->arch.mcg_ctl != ~(u64)0)
2747 banks += 4 * mce->bank;
2749 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2750 * reporting is disabled for the bank
2752 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2754 if (mce->status & MCI_STATUS_UC) {
2755 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2756 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2757 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2760 if (banks[1] & MCI_STATUS_VAL)
2761 mce->status |= MCI_STATUS_OVER;
2762 banks[2] = mce->addr;
2763 banks[3] = mce->misc;
2764 vcpu->arch.mcg_status = mce->mcg_status;
2765 banks[1] = mce->status;
2766 kvm_queue_exception(vcpu, MC_VECTOR);
2767 } else if (!(banks[1] & MCI_STATUS_VAL)
2768 || !(banks[1] & MCI_STATUS_UC)) {
2769 if (banks[1] & MCI_STATUS_VAL)
2770 mce->status |= MCI_STATUS_OVER;
2771 banks[2] = mce->addr;
2772 banks[3] = mce->misc;
2773 banks[1] = mce->status;
2775 banks[1] |= MCI_STATUS_OVER;
2779 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2780 struct kvm_vcpu_events *events)
2783 events->exception.injected =
2784 vcpu->arch.exception.pending &&
2785 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2786 events->exception.nr = vcpu->arch.exception.nr;
2787 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2788 events->exception.pad = 0;
2789 events->exception.error_code = vcpu->arch.exception.error_code;
2791 events->interrupt.injected =
2792 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2793 events->interrupt.nr = vcpu->arch.interrupt.nr;
2794 events->interrupt.soft = 0;
2795 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2797 events->nmi.injected = vcpu->arch.nmi_injected;
2798 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2799 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2800 events->nmi.pad = 0;
2802 events->sipi_vector = 0; /* never valid when reporting to user space */
2804 events->smi.smm = is_smm(vcpu);
2805 events->smi.pending = vcpu->arch.smi_pending;
2806 events->smi.smm_inside_nmi =
2807 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2808 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2810 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2811 | KVM_VCPUEVENT_VALID_SHADOW
2812 | KVM_VCPUEVENT_VALID_SMM);
2813 memset(&events->reserved, 0, sizeof(events->reserved));
2816 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2817 struct kvm_vcpu_events *events)
2819 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2820 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2821 | KVM_VCPUEVENT_VALID_SHADOW
2822 | KVM_VCPUEVENT_VALID_SMM))
2826 vcpu->arch.exception.pending = events->exception.injected;
2827 vcpu->arch.exception.nr = events->exception.nr;
2828 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2829 vcpu->arch.exception.error_code = events->exception.error_code;
2831 vcpu->arch.interrupt.pending = events->interrupt.injected;
2832 vcpu->arch.interrupt.nr = events->interrupt.nr;
2833 vcpu->arch.interrupt.soft = events->interrupt.soft;
2834 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2835 kvm_x86_ops->set_interrupt_shadow(vcpu,
2836 events->interrupt.shadow);
2838 vcpu->arch.nmi_injected = events->nmi.injected;
2839 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2840 vcpu->arch.nmi_pending = events->nmi.pending;
2841 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2843 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2844 kvm_vcpu_has_lapic(vcpu))
2845 vcpu->arch.apic->sipi_vector = events->sipi_vector;
2847 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
2848 if (events->smi.smm)
2849 vcpu->arch.hflags |= HF_SMM_MASK;
2851 vcpu->arch.hflags &= ~HF_SMM_MASK;
2852 vcpu->arch.smi_pending = events->smi.pending;
2853 if (events->smi.smm_inside_nmi)
2854 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
2856 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
2857 if (kvm_vcpu_has_lapic(vcpu)) {
2858 if (events->smi.latched_init)
2859 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2861 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2865 kvm_make_request(KVM_REQ_EVENT, vcpu);
2870 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2871 struct kvm_debugregs *dbgregs)
2875 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2876 kvm_get_dr(vcpu, 6, &val);
2878 dbgregs->dr7 = vcpu->arch.dr7;
2880 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2883 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2884 struct kvm_debugregs *dbgregs)
2889 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2890 kvm_update_dr0123(vcpu);
2891 vcpu->arch.dr6 = dbgregs->dr6;
2892 kvm_update_dr6(vcpu);
2893 vcpu->arch.dr7 = dbgregs->dr7;
2894 kvm_update_dr7(vcpu);
2899 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
2901 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
2903 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
2904 u64 xstate_bv = xsave->header.xfeatures;
2908 * Copy legacy XSAVE area, to avoid complications with CPUID
2909 * leaves 0 and 1 in the loop below.
2911 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
2914 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
2917 * Copy each region from the possibly compacted offset to the
2918 * non-compacted offset.
2920 valid = xstate_bv & ~XSTATE_FPSSE;
2922 u64 feature = valid & -valid;
2923 int index = fls64(feature) - 1;
2924 void *src = get_xsave_addr(xsave, feature);
2927 u32 size, offset, ecx, edx;
2928 cpuid_count(XSTATE_CPUID, index,
2929 &size, &offset, &ecx, &edx);
2930 memcpy(dest + offset, src, size);
2937 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
2939 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
2940 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
2944 * Copy legacy XSAVE area, to avoid complications with CPUID
2945 * leaves 0 and 1 in the loop below.
2947 memcpy(xsave, src, XSAVE_HDR_OFFSET);
2949 /* Set XSTATE_BV and possibly XCOMP_BV. */
2950 xsave->header.xfeatures = xstate_bv;
2952 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
2955 * Copy each region from the non-compacted offset to the
2956 * possibly compacted offset.
2958 valid = xstate_bv & ~XSTATE_FPSSE;
2960 u64 feature = valid & -valid;
2961 int index = fls64(feature) - 1;
2962 void *dest = get_xsave_addr(xsave, feature);
2965 u32 size, offset, ecx, edx;
2966 cpuid_count(XSTATE_CPUID, index,
2967 &size, &offset, &ecx, &edx);
2968 memcpy(dest, src + offset, size);
2975 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2976 struct kvm_xsave *guest_xsave)
2978 if (cpu_has_xsave) {
2979 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
2980 fill_xsave((u8 *) guest_xsave->region, vcpu);
2982 memcpy(guest_xsave->region,
2983 &vcpu->arch.guest_fpu.state.fxsave,
2984 sizeof(struct fxregs_state));
2985 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2990 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2991 struct kvm_xsave *guest_xsave)
2994 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2996 if (cpu_has_xsave) {
2998 * Here we allow setting states that are not present in
2999 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3000 * with old userspace.
3002 if (xstate_bv & ~kvm_supported_xcr0())
3004 load_xsave(vcpu, (u8 *)guest_xsave->region);
3006 if (xstate_bv & ~XSTATE_FPSSE)
3008 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3009 guest_xsave->region, sizeof(struct fxregs_state));
3014 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3015 struct kvm_xcrs *guest_xcrs)
3017 if (!cpu_has_xsave) {
3018 guest_xcrs->nr_xcrs = 0;
3022 guest_xcrs->nr_xcrs = 1;
3023 guest_xcrs->flags = 0;
3024 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3025 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3028 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3029 struct kvm_xcrs *guest_xcrs)
3036 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3039 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3040 /* Only support XCR0 currently */
3041 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3042 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3043 guest_xcrs->xcrs[i].value);
3052 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3053 * stopped by the hypervisor. This function will be called from the host only.
3054 * EINVAL is returned when the host attempts to set the flag for a guest that
3055 * does not support pv clocks.
3057 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3059 if (!vcpu->arch.pv_time_enabled)
3061 vcpu->arch.pvclock_set_guest_stopped_request = true;
3062 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3066 long kvm_arch_vcpu_ioctl(struct file *filp,
3067 unsigned int ioctl, unsigned long arg)
3069 struct kvm_vcpu *vcpu = filp->private_data;
3070 void __user *argp = (void __user *)arg;
3073 struct kvm_lapic_state *lapic;
3074 struct kvm_xsave *xsave;
3075 struct kvm_xcrs *xcrs;
3081 case KVM_GET_LAPIC: {
3083 if (!vcpu->arch.apic)
3085 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3090 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3094 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3099 case KVM_SET_LAPIC: {
3101 if (!vcpu->arch.apic)
3103 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3104 if (IS_ERR(u.lapic))
3105 return PTR_ERR(u.lapic);
3107 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3110 case KVM_INTERRUPT: {
3111 struct kvm_interrupt irq;
3114 if (copy_from_user(&irq, argp, sizeof irq))
3116 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3120 r = kvm_vcpu_ioctl_nmi(vcpu);
3124 r = kvm_vcpu_ioctl_smi(vcpu);
3127 case KVM_SET_CPUID: {
3128 struct kvm_cpuid __user *cpuid_arg = argp;
3129 struct kvm_cpuid cpuid;
3132 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3134 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3137 case KVM_SET_CPUID2: {
3138 struct kvm_cpuid2 __user *cpuid_arg = argp;
3139 struct kvm_cpuid2 cpuid;
3142 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3144 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3145 cpuid_arg->entries);
3148 case KVM_GET_CPUID2: {
3149 struct kvm_cpuid2 __user *cpuid_arg = argp;
3150 struct kvm_cpuid2 cpuid;
3153 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3155 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3156 cpuid_arg->entries);
3160 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3166 r = msr_io(vcpu, argp, do_get_msr, 1);
3169 r = msr_io(vcpu, argp, do_set_msr, 0);
3171 case KVM_TPR_ACCESS_REPORTING: {
3172 struct kvm_tpr_access_ctl tac;
3175 if (copy_from_user(&tac, argp, sizeof tac))
3177 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3181 if (copy_to_user(argp, &tac, sizeof tac))
3186 case KVM_SET_VAPIC_ADDR: {
3187 struct kvm_vapic_addr va;
3190 if (!irqchip_in_kernel(vcpu->kvm))
3193 if (copy_from_user(&va, argp, sizeof va))
3195 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3198 case KVM_X86_SETUP_MCE: {
3202 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3204 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3207 case KVM_X86_SET_MCE: {
3208 struct kvm_x86_mce mce;
3211 if (copy_from_user(&mce, argp, sizeof mce))
3213 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3216 case KVM_GET_VCPU_EVENTS: {
3217 struct kvm_vcpu_events events;
3219 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3222 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3227 case KVM_SET_VCPU_EVENTS: {
3228 struct kvm_vcpu_events events;
3231 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3234 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3237 case KVM_GET_DEBUGREGS: {
3238 struct kvm_debugregs dbgregs;
3240 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3243 if (copy_to_user(argp, &dbgregs,
3244 sizeof(struct kvm_debugregs)))
3249 case KVM_SET_DEBUGREGS: {
3250 struct kvm_debugregs dbgregs;
3253 if (copy_from_user(&dbgregs, argp,
3254 sizeof(struct kvm_debugregs)))
3257 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3260 case KVM_GET_XSAVE: {
3261 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3266 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3269 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3274 case KVM_SET_XSAVE: {
3275 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3276 if (IS_ERR(u.xsave))
3277 return PTR_ERR(u.xsave);
3279 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3282 case KVM_GET_XCRS: {
3283 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3288 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3291 if (copy_to_user(argp, u.xcrs,
3292 sizeof(struct kvm_xcrs)))
3297 case KVM_SET_XCRS: {
3298 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3300 return PTR_ERR(u.xcrs);
3302 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3305 case KVM_SET_TSC_KHZ: {
3309 user_tsc_khz = (u32)arg;
3311 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3314 if (user_tsc_khz == 0)
3315 user_tsc_khz = tsc_khz;
3317 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3322 case KVM_GET_TSC_KHZ: {
3323 r = vcpu->arch.virtual_tsc_khz;
3326 case KVM_KVMCLOCK_CTRL: {
3327 r = kvm_set_guest_paused(vcpu);
3338 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3340 return VM_FAULT_SIGBUS;
3343 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3347 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3349 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3353 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3356 kvm->arch.ept_identity_map_addr = ident_addr;
3360 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3361 u32 kvm_nr_mmu_pages)
3363 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3366 mutex_lock(&kvm->slots_lock);
3368 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3369 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3371 mutex_unlock(&kvm->slots_lock);
3375 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3377 return kvm->arch.n_max_mmu_pages;
3380 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3385 switch (chip->chip_id) {
3386 case KVM_IRQCHIP_PIC_MASTER:
3387 memcpy(&chip->chip.pic,
3388 &pic_irqchip(kvm)->pics[0],
3389 sizeof(struct kvm_pic_state));
3391 case KVM_IRQCHIP_PIC_SLAVE:
3392 memcpy(&chip->chip.pic,
3393 &pic_irqchip(kvm)->pics[1],
3394 sizeof(struct kvm_pic_state));
3396 case KVM_IRQCHIP_IOAPIC:
3397 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3406 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3411 switch (chip->chip_id) {
3412 case KVM_IRQCHIP_PIC_MASTER:
3413 spin_lock(&pic_irqchip(kvm)->lock);
3414 memcpy(&pic_irqchip(kvm)->pics[0],
3416 sizeof(struct kvm_pic_state));
3417 spin_unlock(&pic_irqchip(kvm)->lock);
3419 case KVM_IRQCHIP_PIC_SLAVE:
3420 spin_lock(&pic_irqchip(kvm)->lock);
3421 memcpy(&pic_irqchip(kvm)->pics[1],
3423 sizeof(struct kvm_pic_state));
3424 spin_unlock(&pic_irqchip(kvm)->lock);
3426 case KVM_IRQCHIP_IOAPIC:
3427 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3433 kvm_pic_update_irq(pic_irqchip(kvm));
3437 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3441 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3442 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3443 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3447 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3451 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3452 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3453 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3454 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3458 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3462 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3463 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3464 sizeof(ps->channels));
3465 ps->flags = kvm->arch.vpit->pit_state.flags;
3466 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3467 memset(&ps->reserved, 0, sizeof(ps->reserved));
3471 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3473 int r = 0, start = 0;
3474 u32 prev_legacy, cur_legacy;
3475 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3476 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3477 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3478 if (!prev_legacy && cur_legacy)
3480 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3481 sizeof(kvm->arch.vpit->pit_state.channels));
3482 kvm->arch.vpit->pit_state.flags = ps->flags;
3483 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3484 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3488 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3489 struct kvm_reinject_control *control)
3491 if (!kvm->arch.vpit)
3493 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3494 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3495 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3500 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3501 * @kvm: kvm instance
3502 * @log: slot id and address to which we copy the log
3504 * Steps 1-4 below provide general overview of dirty page logging. See
3505 * kvm_get_dirty_log_protect() function description for additional details.
3507 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3508 * always flush the TLB (step 4) even if previous step failed and the dirty
3509 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3510 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3511 * writes will be marked dirty for next log read.
3513 * 1. Take a snapshot of the bit and clear it if needed.
3514 * 2. Write protect the corresponding page.
3515 * 3. Copy the snapshot to the userspace.
3516 * 4. Flush TLB's if needed.
3518 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3520 bool is_dirty = false;
3523 mutex_lock(&kvm->slots_lock);
3526 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3528 if (kvm_x86_ops->flush_log_dirty)
3529 kvm_x86_ops->flush_log_dirty(kvm);
3531 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3534 * All the TLBs can be flushed out of mmu lock, see the comments in
3535 * kvm_mmu_slot_remove_write_access().
3537 lockdep_assert_held(&kvm->slots_lock);
3539 kvm_flush_remote_tlbs(kvm);
3541 mutex_unlock(&kvm->slots_lock);
3545 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3548 if (!irqchip_in_kernel(kvm))
3551 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3552 irq_event->irq, irq_event->level,
3557 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3558 struct kvm_enable_cap *cap)
3566 case KVM_CAP_DISABLE_QUIRKS:
3567 kvm->arch.disabled_quirks = cap->args[0];
3577 long kvm_arch_vm_ioctl(struct file *filp,
3578 unsigned int ioctl, unsigned long arg)
3580 struct kvm *kvm = filp->private_data;
3581 void __user *argp = (void __user *)arg;
3584 * This union makes it completely explicit to gcc-3.x
3585 * that these two variables' stack usage should be
3586 * combined, not added together.
3589 struct kvm_pit_state ps;
3590 struct kvm_pit_state2 ps2;
3591 struct kvm_pit_config pit_config;
3595 case KVM_SET_TSS_ADDR:
3596 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3598 case KVM_SET_IDENTITY_MAP_ADDR: {
3602 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3604 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3607 case KVM_SET_NR_MMU_PAGES:
3608 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3610 case KVM_GET_NR_MMU_PAGES:
3611 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3613 case KVM_CREATE_IRQCHIP: {
3614 struct kvm_pic *vpic;
3616 mutex_lock(&kvm->lock);
3619 goto create_irqchip_unlock;
3621 if (atomic_read(&kvm->online_vcpus))
3622 goto create_irqchip_unlock;
3624 vpic = kvm_create_pic(kvm);
3626 r = kvm_ioapic_init(kvm);
3628 mutex_lock(&kvm->slots_lock);
3629 kvm_destroy_pic(vpic);
3630 mutex_unlock(&kvm->slots_lock);
3631 goto create_irqchip_unlock;
3634 goto create_irqchip_unlock;
3635 r = kvm_setup_default_irq_routing(kvm);
3637 mutex_lock(&kvm->slots_lock);
3638 mutex_lock(&kvm->irq_lock);
3639 kvm_ioapic_destroy(kvm);
3640 kvm_destroy_pic(vpic);
3641 mutex_unlock(&kvm->irq_lock);
3642 mutex_unlock(&kvm->slots_lock);
3643 goto create_irqchip_unlock;
3645 /* Write kvm->irq_routing before kvm->arch.vpic. */
3647 kvm->arch.vpic = vpic;
3648 create_irqchip_unlock:
3649 mutex_unlock(&kvm->lock);
3652 case KVM_CREATE_PIT:
3653 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3655 case KVM_CREATE_PIT2:
3657 if (copy_from_user(&u.pit_config, argp,
3658 sizeof(struct kvm_pit_config)))
3661 mutex_lock(&kvm->slots_lock);
3664 goto create_pit_unlock;
3666 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3670 mutex_unlock(&kvm->slots_lock);
3672 case KVM_GET_IRQCHIP: {
3673 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3674 struct kvm_irqchip *chip;
3676 chip = memdup_user(argp, sizeof(*chip));
3683 if (!irqchip_in_kernel(kvm))
3684 goto get_irqchip_out;
3685 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3687 goto get_irqchip_out;
3689 if (copy_to_user(argp, chip, sizeof *chip))
3690 goto get_irqchip_out;
3696 case KVM_SET_IRQCHIP: {
3697 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3698 struct kvm_irqchip *chip;
3700 chip = memdup_user(argp, sizeof(*chip));
3707 if (!irqchip_in_kernel(kvm))
3708 goto set_irqchip_out;
3709 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3711 goto set_irqchip_out;
3719 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3722 if (!kvm->arch.vpit)
3724 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3728 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3735 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3738 if (!kvm->arch.vpit)
3740 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3743 case KVM_GET_PIT2: {
3745 if (!kvm->arch.vpit)
3747 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3751 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3756 case KVM_SET_PIT2: {
3758 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3761 if (!kvm->arch.vpit)
3763 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3766 case KVM_REINJECT_CONTROL: {
3767 struct kvm_reinject_control control;
3769 if (copy_from_user(&control, argp, sizeof(control)))
3771 r = kvm_vm_ioctl_reinject(kvm, &control);
3774 case KVM_SET_BOOT_CPU_ID:
3776 mutex_lock(&kvm->lock);
3777 if (atomic_read(&kvm->online_vcpus) != 0)
3780 kvm->arch.bsp_vcpu_id = arg;
3781 mutex_unlock(&kvm->lock);
3783 case KVM_XEN_HVM_CONFIG: {
3785 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3786 sizeof(struct kvm_xen_hvm_config)))
3789 if (kvm->arch.xen_hvm_config.flags)
3794 case KVM_SET_CLOCK: {
3795 struct kvm_clock_data user_ns;
3800 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3808 local_irq_disable();
3809 now_ns = get_kernel_ns();
3810 delta = user_ns.clock - now_ns;
3812 kvm->arch.kvmclock_offset = delta;
3813 kvm_gen_update_masterclock(kvm);
3816 case KVM_GET_CLOCK: {
3817 struct kvm_clock_data user_ns;
3820 local_irq_disable();
3821 now_ns = get_kernel_ns();
3822 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3825 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3828 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3833 case KVM_ENABLE_CAP: {
3834 struct kvm_enable_cap cap;
3837 if (copy_from_user(&cap, argp, sizeof(cap)))
3839 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
3843 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
3849 static void kvm_init_msr_list(void)
3854 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
3855 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3859 * Even MSRs that are valid in the host may not be exposed
3860 * to the guests in some cases. We could work around this
3861 * in VMX with the generic MSR save/load machinery, but it
3862 * is not really worthwhile since it will really only
3863 * happen with nested virtualization.
3865 switch (msrs_to_save[i]) {
3866 case MSR_IA32_BNDCFGS:
3867 if (!kvm_x86_ops->mpx_supported())
3875 msrs_to_save[j] = msrs_to_save[i];
3878 num_msrs_to_save = j;
3880 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
3881 switch (emulated_msrs[i]) {
3882 case MSR_IA32_SMBASE:
3883 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
3891 emulated_msrs[j] = emulated_msrs[i];
3894 num_emulated_msrs = j;
3897 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3905 if (!(vcpu->arch.apic &&
3906 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
3907 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
3918 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3925 if (!(vcpu->arch.apic &&
3926 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
3928 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
3930 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3940 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3941 struct kvm_segment *var, int seg)
3943 kvm_x86_ops->set_segment(vcpu, var, seg);
3946 void kvm_get_segment(struct kvm_vcpu *vcpu,
3947 struct kvm_segment *var, int seg)
3949 kvm_x86_ops->get_segment(vcpu, var, seg);
3952 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
3953 struct x86_exception *exception)
3957 BUG_ON(!mmu_is_nested(vcpu));
3959 /* NPT walks are always user-walks */
3960 access |= PFERR_USER_MASK;
3961 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
3966 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3967 struct x86_exception *exception)
3969 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3970 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3973 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3974 struct x86_exception *exception)
3976 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3977 access |= PFERR_FETCH_MASK;
3978 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3981 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3982 struct x86_exception *exception)
3984 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3985 access |= PFERR_WRITE_MASK;
3986 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3989 /* uses this to access any guest's mapped memory without checking CPL */
3990 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3991 struct x86_exception *exception)
3993 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3996 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3997 struct kvm_vcpu *vcpu, u32 access,
3998 struct x86_exception *exception)
4001 int r = X86EMUL_CONTINUE;
4004 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4006 unsigned offset = addr & (PAGE_SIZE-1);
4007 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4010 if (gpa == UNMAPPED_GVA)
4011 return X86EMUL_PROPAGATE_FAULT;
4012 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4015 r = X86EMUL_IO_NEEDED;
4027 /* used for instruction fetching */
4028 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4029 gva_t addr, void *val, unsigned int bytes,
4030 struct x86_exception *exception)
4032 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4033 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4037 /* Inline kvm_read_guest_virt_helper for speed. */
4038 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4040 if (unlikely(gpa == UNMAPPED_GVA))
4041 return X86EMUL_PROPAGATE_FAULT;
4043 offset = addr & (PAGE_SIZE-1);
4044 if (WARN_ON(offset + bytes > PAGE_SIZE))
4045 bytes = (unsigned)PAGE_SIZE - offset;
4046 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4048 if (unlikely(ret < 0))
4049 return X86EMUL_IO_NEEDED;
4051 return X86EMUL_CONTINUE;
4054 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4055 gva_t addr, void *val, unsigned int bytes,
4056 struct x86_exception *exception)
4058 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4059 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4061 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4064 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4066 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4067 gva_t addr, void *val, unsigned int bytes,
4068 struct x86_exception *exception)
4070 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4071 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4074 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4075 gva_t addr, void *val,
4077 struct x86_exception *exception)
4079 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4081 int r = X86EMUL_CONTINUE;
4084 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4087 unsigned offset = addr & (PAGE_SIZE-1);
4088 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4091 if (gpa == UNMAPPED_GVA)
4092 return X86EMUL_PROPAGATE_FAULT;
4093 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4095 r = X86EMUL_IO_NEEDED;
4106 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4108 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4109 gpa_t *gpa, struct x86_exception *exception,
4112 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4113 | (write ? PFERR_WRITE_MASK : 0);
4115 if (vcpu_match_mmio_gva(vcpu, gva)
4116 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4117 vcpu->arch.access, access)) {
4118 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4119 (gva & (PAGE_SIZE - 1));
4120 trace_vcpu_match_mmio(gva, *gpa, write, false);
4124 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4126 if (*gpa == UNMAPPED_GVA)
4129 /* For APIC access vmexit */
4130 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4133 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4134 trace_vcpu_match_mmio(gva, *gpa, write, true);
4141 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4142 const void *val, int bytes)
4146 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4149 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4153 struct read_write_emulator_ops {
4154 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4156 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4157 void *val, int bytes);
4158 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4159 int bytes, void *val);
4160 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4161 void *val, int bytes);
4165 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4167 if (vcpu->mmio_read_completed) {
4168 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4169 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4170 vcpu->mmio_read_completed = 0;
4177 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4178 void *val, int bytes)
4180 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4183 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4184 void *val, int bytes)
4186 return emulator_write_phys(vcpu, gpa, val, bytes);
4189 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4191 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4192 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4195 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4196 void *val, int bytes)
4198 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4199 return X86EMUL_IO_NEEDED;
4202 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4203 void *val, int bytes)
4205 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4207 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4208 return X86EMUL_CONTINUE;
4211 static const struct read_write_emulator_ops read_emultor = {
4212 .read_write_prepare = read_prepare,
4213 .read_write_emulate = read_emulate,
4214 .read_write_mmio = vcpu_mmio_read,
4215 .read_write_exit_mmio = read_exit_mmio,
4218 static const struct read_write_emulator_ops write_emultor = {
4219 .read_write_emulate = write_emulate,
4220 .read_write_mmio = write_mmio,
4221 .read_write_exit_mmio = write_exit_mmio,
4225 static int emulator_read_write_onepage(unsigned long addr, void *val,
4227 struct x86_exception *exception,
4228 struct kvm_vcpu *vcpu,
4229 const struct read_write_emulator_ops *ops)
4233 bool write = ops->write;
4234 struct kvm_mmio_fragment *frag;
4236 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4239 return X86EMUL_PROPAGATE_FAULT;
4241 /* For APIC access vmexit */
4245 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4246 return X86EMUL_CONTINUE;
4250 * Is this MMIO handled locally?
4252 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4253 if (handled == bytes)
4254 return X86EMUL_CONTINUE;
4260 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4261 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4265 return X86EMUL_CONTINUE;
4268 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4270 void *val, unsigned int bytes,
4271 struct x86_exception *exception,
4272 const struct read_write_emulator_ops *ops)
4274 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4278 if (ops->read_write_prepare &&
4279 ops->read_write_prepare(vcpu, val, bytes))
4280 return X86EMUL_CONTINUE;
4282 vcpu->mmio_nr_fragments = 0;
4284 /* Crossing a page boundary? */
4285 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4288 now = -addr & ~PAGE_MASK;
4289 rc = emulator_read_write_onepage(addr, val, now, exception,
4292 if (rc != X86EMUL_CONTINUE)
4295 if (ctxt->mode != X86EMUL_MODE_PROT64)
4301 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4303 if (rc != X86EMUL_CONTINUE)
4306 if (!vcpu->mmio_nr_fragments)
4309 gpa = vcpu->mmio_fragments[0].gpa;
4311 vcpu->mmio_needed = 1;
4312 vcpu->mmio_cur_fragment = 0;
4314 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4315 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4316 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4317 vcpu->run->mmio.phys_addr = gpa;
4319 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4322 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4326 struct x86_exception *exception)
4328 return emulator_read_write(ctxt, addr, val, bytes,
4329 exception, &read_emultor);
4332 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4336 struct x86_exception *exception)
4338 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4339 exception, &write_emultor);
4342 #define CMPXCHG_TYPE(t, ptr, old, new) \
4343 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4345 #ifdef CONFIG_X86_64
4346 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4348 # define CMPXCHG64(ptr, old, new) \
4349 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4352 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4357 struct x86_exception *exception)
4359 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4365 /* guests cmpxchg8b have to be emulated atomically */
4366 if (bytes > 8 || (bytes & (bytes - 1)))
4369 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4371 if (gpa == UNMAPPED_GVA ||
4372 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4375 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4378 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4379 if (is_error_page(page))
4382 kaddr = kmap_atomic(page);
4383 kaddr += offset_in_page(gpa);
4386 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4389 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4392 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4395 exchanged = CMPXCHG64(kaddr, old, new);
4400 kunmap_atomic(kaddr);
4401 kvm_release_page_dirty(page);
4404 return X86EMUL_CMPXCHG_FAILED;
4406 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4407 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4409 return X86EMUL_CONTINUE;
4412 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4414 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4417 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4419 /* TODO: String I/O for in kernel device */
4422 if (vcpu->arch.pio.in)
4423 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4424 vcpu->arch.pio.size, pd);
4426 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4427 vcpu->arch.pio.port, vcpu->arch.pio.size,
4432 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4433 unsigned short port, void *val,
4434 unsigned int count, bool in)
4436 vcpu->arch.pio.port = port;
4437 vcpu->arch.pio.in = in;
4438 vcpu->arch.pio.count = count;
4439 vcpu->arch.pio.size = size;
4441 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4442 vcpu->arch.pio.count = 0;
4446 vcpu->run->exit_reason = KVM_EXIT_IO;
4447 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4448 vcpu->run->io.size = size;
4449 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4450 vcpu->run->io.count = count;
4451 vcpu->run->io.port = port;
4456 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4457 int size, unsigned short port, void *val,
4460 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4463 if (vcpu->arch.pio.count)
4466 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4469 memcpy(val, vcpu->arch.pio_data, size * count);
4470 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4471 vcpu->arch.pio.count = 0;
4478 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4479 int size, unsigned short port,
4480 const void *val, unsigned int count)
4482 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4484 memcpy(vcpu->arch.pio_data, val, size * count);
4485 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4486 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4489 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4491 return kvm_x86_ops->get_segment_base(vcpu, seg);
4494 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4496 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4499 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4501 if (!need_emulate_wbinvd(vcpu))
4502 return X86EMUL_CONTINUE;
4504 if (kvm_x86_ops->has_wbinvd_exit()) {
4505 int cpu = get_cpu();
4507 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4508 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4509 wbinvd_ipi, NULL, 1);
4511 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4514 return X86EMUL_CONTINUE;
4517 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4519 kvm_x86_ops->skip_emulated_instruction(vcpu);
4520 return kvm_emulate_wbinvd_noskip(vcpu);
4522 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4526 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4528 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4531 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4532 unsigned long *dest)
4534 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4537 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4538 unsigned long value)
4541 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4544 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4546 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4549 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4551 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4552 unsigned long value;
4556 value = kvm_read_cr0(vcpu);
4559 value = vcpu->arch.cr2;
4562 value = kvm_read_cr3(vcpu);
4565 value = kvm_read_cr4(vcpu);
4568 value = kvm_get_cr8(vcpu);
4571 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4578 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4580 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4585 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4588 vcpu->arch.cr2 = val;
4591 res = kvm_set_cr3(vcpu, val);
4594 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4597 res = kvm_set_cr8(vcpu, val);
4600 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4607 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4609 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4612 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4614 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4617 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4619 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4622 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4624 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4627 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4629 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4632 static unsigned long emulator_get_cached_segment_base(
4633 struct x86_emulate_ctxt *ctxt, int seg)
4635 return get_segment_base(emul_to_vcpu(ctxt), seg);
4638 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4639 struct desc_struct *desc, u32 *base3,
4642 struct kvm_segment var;
4644 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4645 *selector = var.selector;
4648 memset(desc, 0, sizeof(*desc));
4654 set_desc_limit(desc, var.limit);
4655 set_desc_base(desc, (unsigned long)var.base);
4656 #ifdef CONFIG_X86_64
4658 *base3 = var.base >> 32;
4660 desc->type = var.type;
4662 desc->dpl = var.dpl;
4663 desc->p = var.present;
4664 desc->avl = var.avl;
4672 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4673 struct desc_struct *desc, u32 base3,
4676 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4677 struct kvm_segment var;
4679 var.selector = selector;
4680 var.base = get_desc_base(desc);
4681 #ifdef CONFIG_X86_64
4682 var.base |= ((u64)base3) << 32;
4684 var.limit = get_desc_limit(desc);
4686 var.limit = (var.limit << 12) | 0xfff;
4687 var.type = desc->type;
4688 var.dpl = desc->dpl;
4693 var.avl = desc->avl;
4694 var.present = desc->p;
4695 var.unusable = !var.present;
4698 kvm_set_segment(vcpu, &var, seg);
4702 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4703 u32 msr_index, u64 *pdata)
4705 struct msr_data msr;
4708 msr.index = msr_index;
4709 msr.host_initiated = false;
4710 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4718 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4719 u32 msr_index, u64 data)
4721 struct msr_data msr;
4724 msr.index = msr_index;
4725 msr.host_initiated = false;
4726 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4729 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4731 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4733 return vcpu->arch.smbase;
4736 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4738 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4740 vcpu->arch.smbase = smbase;
4743 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4746 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4749 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4750 u32 pmc, u64 *pdata)
4752 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4755 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4757 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4760 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4763 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4765 * CR0.TS may reference the host fpu state, not the guest fpu state,
4766 * so it may be clear at this point.
4771 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4776 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4777 struct x86_instruction_info *info,
4778 enum x86_intercept_stage stage)
4780 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4783 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4784 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4786 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4789 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4791 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4794 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4796 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4799 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
4801 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
4804 static const struct x86_emulate_ops emulate_ops = {
4805 .read_gpr = emulator_read_gpr,
4806 .write_gpr = emulator_write_gpr,
4807 .read_std = kvm_read_guest_virt_system,
4808 .write_std = kvm_write_guest_virt_system,
4809 .fetch = kvm_fetch_guest_virt,
4810 .read_emulated = emulator_read_emulated,
4811 .write_emulated = emulator_write_emulated,
4812 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4813 .invlpg = emulator_invlpg,
4814 .pio_in_emulated = emulator_pio_in_emulated,
4815 .pio_out_emulated = emulator_pio_out_emulated,
4816 .get_segment = emulator_get_segment,
4817 .set_segment = emulator_set_segment,
4818 .get_cached_segment_base = emulator_get_cached_segment_base,
4819 .get_gdt = emulator_get_gdt,
4820 .get_idt = emulator_get_idt,
4821 .set_gdt = emulator_set_gdt,
4822 .set_idt = emulator_set_idt,
4823 .get_cr = emulator_get_cr,
4824 .set_cr = emulator_set_cr,
4825 .cpl = emulator_get_cpl,
4826 .get_dr = emulator_get_dr,
4827 .set_dr = emulator_set_dr,
4828 .get_smbase = emulator_get_smbase,
4829 .set_smbase = emulator_set_smbase,
4830 .set_msr = emulator_set_msr,
4831 .get_msr = emulator_get_msr,
4832 .check_pmc = emulator_check_pmc,
4833 .read_pmc = emulator_read_pmc,
4834 .halt = emulator_halt,
4835 .wbinvd = emulator_wbinvd,
4836 .fix_hypercall = emulator_fix_hypercall,
4837 .get_fpu = emulator_get_fpu,
4838 .put_fpu = emulator_put_fpu,
4839 .intercept = emulator_intercept,
4840 .get_cpuid = emulator_get_cpuid,
4841 .set_nmi_mask = emulator_set_nmi_mask,
4844 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4846 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
4848 * an sti; sti; sequence only disable interrupts for the first
4849 * instruction. So, if the last instruction, be it emulated or
4850 * not, left the system with the INT_STI flag enabled, it
4851 * means that the last instruction is an sti. We should not
4852 * leave the flag on in this case. The same goes for mov ss
4854 if (int_shadow & mask)
4856 if (unlikely(int_shadow || mask)) {
4857 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4859 kvm_make_request(KVM_REQ_EVENT, vcpu);
4863 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
4865 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4866 if (ctxt->exception.vector == PF_VECTOR)
4867 return kvm_propagate_fault(vcpu, &ctxt->exception);
4869 if (ctxt->exception.error_code_valid)
4870 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4871 ctxt->exception.error_code);
4873 kvm_queue_exception(vcpu, ctxt->exception.vector);
4877 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4879 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4882 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4884 ctxt->eflags = kvm_get_rflags(vcpu);
4885 ctxt->eip = kvm_rip_read(vcpu);
4886 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4887 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4888 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
4889 cs_db ? X86EMUL_MODE_PROT32 :
4890 X86EMUL_MODE_PROT16;
4891 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
4892 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
4893 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
4894 ctxt->emul_flags = vcpu->arch.hflags;
4896 init_decode_cache(ctxt);
4897 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4900 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4902 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4905 init_emulate_ctxt(vcpu);
4909 ctxt->_eip = ctxt->eip + inc_eip;
4910 ret = emulate_int_real(ctxt, irq);
4912 if (ret != X86EMUL_CONTINUE)
4913 return EMULATE_FAIL;
4915 ctxt->eip = ctxt->_eip;
4916 kvm_rip_write(vcpu, ctxt->eip);
4917 kvm_set_rflags(vcpu, ctxt->eflags);
4919 if (irq == NMI_VECTOR)
4920 vcpu->arch.nmi_pending = 0;
4922 vcpu->arch.interrupt.pending = false;
4924 return EMULATE_DONE;
4926 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4928 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4930 int r = EMULATE_DONE;
4932 ++vcpu->stat.insn_emulation_fail;
4933 trace_kvm_emulate_insn_failed(vcpu);
4934 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
4935 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4936 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4937 vcpu->run->internal.ndata = 0;
4940 kvm_queue_exception(vcpu, UD_VECTOR);
4945 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4946 bool write_fault_to_shadow_pgtable,
4952 if (emulation_type & EMULTYPE_NO_REEXECUTE)
4955 if (!vcpu->arch.mmu.direct_map) {
4957 * Write permission should be allowed since only
4958 * write access need to be emulated.
4960 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4963 * If the mapping is invalid in guest, let cpu retry
4964 * it to generate fault.
4966 if (gpa == UNMAPPED_GVA)
4971 * Do not retry the unhandleable instruction if it faults on the
4972 * readonly host memory, otherwise it will goto a infinite loop:
4973 * retry instruction -> write #PF -> emulation fail -> retry
4974 * instruction -> ...
4976 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4979 * If the instruction failed on the error pfn, it can not be fixed,
4980 * report the error to userspace.
4982 if (is_error_noslot_pfn(pfn))
4985 kvm_release_pfn_clean(pfn);
4987 /* The instructions are well-emulated on direct mmu. */
4988 if (vcpu->arch.mmu.direct_map) {
4989 unsigned int indirect_shadow_pages;
4991 spin_lock(&vcpu->kvm->mmu_lock);
4992 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
4993 spin_unlock(&vcpu->kvm->mmu_lock);
4995 if (indirect_shadow_pages)
4996 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5002 * if emulation was due to access to shadowed page table
5003 * and it failed try to unshadow page and re-enter the
5004 * guest to let CPU execute the instruction.
5006 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5009 * If the access faults on its page table, it can not
5010 * be fixed by unprotecting shadow page and it should
5011 * be reported to userspace.
5013 return !write_fault_to_shadow_pgtable;
5016 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5017 unsigned long cr2, int emulation_type)
5019 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5020 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5022 last_retry_eip = vcpu->arch.last_retry_eip;
5023 last_retry_addr = vcpu->arch.last_retry_addr;
5026 * If the emulation is caused by #PF and it is non-page_table
5027 * writing instruction, it means the VM-EXIT is caused by shadow
5028 * page protected, we can zap the shadow page and retry this
5029 * instruction directly.
5031 * Note: if the guest uses a non-page-table modifying instruction
5032 * on the PDE that points to the instruction, then we will unmap
5033 * the instruction and go to an infinite loop. So, we cache the
5034 * last retried eip and the last fault address, if we meet the eip
5035 * and the address again, we can break out of the potential infinite
5038 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5040 if (!(emulation_type & EMULTYPE_RETRY))
5043 if (x86_page_table_writing_insn(ctxt))
5046 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5049 vcpu->arch.last_retry_eip = ctxt->eip;
5050 vcpu->arch.last_retry_addr = cr2;
5052 if (!vcpu->arch.mmu.direct_map)
5053 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5055 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5060 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5061 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5063 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5065 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5066 /* This is a good place to trace that we are exiting SMM. */
5067 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5069 if (unlikely(vcpu->arch.smi_pending)) {
5070 kvm_make_request(KVM_REQ_SMI, vcpu);
5071 vcpu->arch.smi_pending = 0;
5073 /* Process a latched INIT, if any. */
5074 kvm_make_request(KVM_REQ_EVENT, vcpu);
5078 kvm_mmu_reset_context(vcpu);
5081 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5083 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5085 vcpu->arch.hflags = emul_flags;
5087 if (changed & HF_SMM_MASK)
5088 kvm_smm_changed(vcpu);
5091 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5100 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5101 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5106 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5108 struct kvm_run *kvm_run = vcpu->run;
5111 * rflags is the old, "raw" value of the flags. The new value has
5112 * not been saved yet.
5114 * This is correct even for TF set by the guest, because "the
5115 * processor will not generate this exception after the instruction
5116 * that sets the TF flag".
5118 if (unlikely(rflags & X86_EFLAGS_TF)) {
5119 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5120 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5122 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5123 kvm_run->debug.arch.exception = DB_VECTOR;
5124 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5125 *r = EMULATE_USER_EXIT;
5127 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5129 * "Certain debug exceptions may clear bit 0-3. The
5130 * remaining contents of the DR6 register are never
5131 * cleared by the processor".
5133 vcpu->arch.dr6 &= ~15;
5134 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5135 kvm_queue_exception(vcpu, DB_VECTOR);
5140 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5142 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5143 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5144 struct kvm_run *kvm_run = vcpu->run;
5145 unsigned long eip = kvm_get_linear_rip(vcpu);
5146 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5147 vcpu->arch.guest_debug_dr7,
5151 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5152 kvm_run->debug.arch.pc = eip;
5153 kvm_run->debug.arch.exception = DB_VECTOR;
5154 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5155 *r = EMULATE_USER_EXIT;
5160 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5161 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5162 unsigned long eip = kvm_get_linear_rip(vcpu);
5163 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5168 vcpu->arch.dr6 &= ~15;
5169 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5170 kvm_queue_exception(vcpu, DB_VECTOR);
5179 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5186 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5187 bool writeback = true;
5188 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5191 * Clear write_fault_to_shadow_pgtable here to ensure it is
5194 vcpu->arch.write_fault_to_shadow_pgtable = false;
5195 kvm_clear_exception_queue(vcpu);
5197 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5198 init_emulate_ctxt(vcpu);
5201 * We will reenter on the same instruction since
5202 * we do not set complete_userspace_io. This does not
5203 * handle watchpoints yet, those would be handled in
5206 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5209 ctxt->interruptibility = 0;
5210 ctxt->have_exception = false;
5211 ctxt->exception.vector = -1;
5212 ctxt->perm_ok = false;
5214 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5216 r = x86_decode_insn(ctxt, insn, insn_len);
5218 trace_kvm_emulate_insn_start(vcpu);
5219 ++vcpu->stat.insn_emulation;
5220 if (r != EMULATION_OK) {
5221 if (emulation_type & EMULTYPE_TRAP_UD)
5222 return EMULATE_FAIL;
5223 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5225 return EMULATE_DONE;
5226 if (emulation_type & EMULTYPE_SKIP)
5227 return EMULATE_FAIL;
5228 return handle_emulation_failure(vcpu);
5232 if (emulation_type & EMULTYPE_SKIP) {
5233 kvm_rip_write(vcpu, ctxt->_eip);
5234 if (ctxt->eflags & X86_EFLAGS_RF)
5235 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5236 return EMULATE_DONE;
5239 if (retry_instruction(ctxt, cr2, emulation_type))
5240 return EMULATE_DONE;
5242 /* this is needed for vmware backdoor interface to work since it
5243 changes registers values during IO operation */
5244 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5245 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5246 emulator_invalidate_register_cache(ctxt);
5250 r = x86_emulate_insn(ctxt);
5252 if (r == EMULATION_INTERCEPTED)
5253 return EMULATE_DONE;
5255 if (r == EMULATION_FAILED) {
5256 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5258 return EMULATE_DONE;
5260 return handle_emulation_failure(vcpu);
5263 if (ctxt->have_exception) {
5265 if (inject_emulated_exception(vcpu))
5267 } else if (vcpu->arch.pio.count) {
5268 if (!vcpu->arch.pio.in) {
5269 /* FIXME: return into emulator if single-stepping. */
5270 vcpu->arch.pio.count = 0;
5273 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5275 r = EMULATE_USER_EXIT;
5276 } else if (vcpu->mmio_needed) {
5277 if (!vcpu->mmio_is_write)
5279 r = EMULATE_USER_EXIT;
5280 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5281 } else if (r == EMULATION_RESTART)
5287 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5288 toggle_interruptibility(vcpu, ctxt->interruptibility);
5289 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5290 if (vcpu->arch.hflags != ctxt->emul_flags)
5291 kvm_set_hflags(vcpu, ctxt->emul_flags);
5292 kvm_rip_write(vcpu, ctxt->eip);
5293 if (r == EMULATE_DONE)
5294 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5295 if (!ctxt->have_exception ||
5296 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5297 __kvm_set_rflags(vcpu, ctxt->eflags);
5300 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5301 * do nothing, and it will be requested again as soon as
5302 * the shadow expires. But we still need to check here,
5303 * because POPF has no interrupt shadow.
5305 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5306 kvm_make_request(KVM_REQ_EVENT, vcpu);
5308 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5312 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5314 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5316 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5317 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5318 size, port, &val, 1);
5319 /* do not return to emulator after return from userspace */
5320 vcpu->arch.pio.count = 0;
5323 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5325 static void tsc_bad(void *info)
5327 __this_cpu_write(cpu_tsc_khz, 0);
5330 static void tsc_khz_changed(void *data)
5332 struct cpufreq_freqs *freq = data;
5333 unsigned long khz = 0;
5337 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5338 khz = cpufreq_quick_get(raw_smp_processor_id());
5341 __this_cpu_write(cpu_tsc_khz, khz);
5344 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5347 struct cpufreq_freqs *freq = data;
5349 struct kvm_vcpu *vcpu;
5350 int i, send_ipi = 0;
5353 * We allow guests to temporarily run on slowing clocks,
5354 * provided we notify them after, or to run on accelerating
5355 * clocks, provided we notify them before. Thus time never
5358 * However, we have a problem. We can't atomically update
5359 * the frequency of a given CPU from this function; it is
5360 * merely a notifier, which can be called from any CPU.
5361 * Changing the TSC frequency at arbitrary points in time
5362 * requires a recomputation of local variables related to
5363 * the TSC for each VCPU. We must flag these local variables
5364 * to be updated and be sure the update takes place with the
5365 * new frequency before any guests proceed.
5367 * Unfortunately, the combination of hotplug CPU and frequency
5368 * change creates an intractable locking scenario; the order
5369 * of when these callouts happen is undefined with respect to
5370 * CPU hotplug, and they can race with each other. As such,
5371 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5372 * undefined; you can actually have a CPU frequency change take
5373 * place in between the computation of X and the setting of the
5374 * variable. To protect against this problem, all updates of
5375 * the per_cpu tsc_khz variable are done in an interrupt
5376 * protected IPI, and all callers wishing to update the value
5377 * must wait for a synchronous IPI to complete (which is trivial
5378 * if the caller is on the CPU already). This establishes the
5379 * necessary total order on variable updates.
5381 * Note that because a guest time update may take place
5382 * anytime after the setting of the VCPU's request bit, the
5383 * correct TSC value must be set before the request. However,
5384 * to ensure the update actually makes it to any guest which
5385 * starts running in hardware virtualization between the set
5386 * and the acquisition of the spinlock, we must also ping the
5387 * CPU after setting the request bit.
5391 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5393 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5396 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5398 spin_lock(&kvm_lock);
5399 list_for_each_entry(kvm, &vm_list, vm_list) {
5400 kvm_for_each_vcpu(i, vcpu, kvm) {
5401 if (vcpu->cpu != freq->cpu)
5403 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5404 if (vcpu->cpu != smp_processor_id())
5408 spin_unlock(&kvm_lock);
5410 if (freq->old < freq->new && send_ipi) {
5412 * We upscale the frequency. Must make the guest
5413 * doesn't see old kvmclock values while running with
5414 * the new frequency, otherwise we risk the guest sees
5415 * time go backwards.
5417 * In case we update the frequency for another cpu
5418 * (which might be in guest context) send an interrupt
5419 * to kick the cpu out of guest context. Next time
5420 * guest context is entered kvmclock will be updated,
5421 * so the guest will not see stale values.
5423 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5428 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5429 .notifier_call = kvmclock_cpufreq_notifier
5432 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5433 unsigned long action, void *hcpu)
5435 unsigned int cpu = (unsigned long)hcpu;
5439 case CPU_DOWN_FAILED:
5440 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5442 case CPU_DOWN_PREPARE:
5443 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5449 static struct notifier_block kvmclock_cpu_notifier_block = {
5450 .notifier_call = kvmclock_cpu_notifier,
5451 .priority = -INT_MAX
5454 static void kvm_timer_init(void)
5458 max_tsc_khz = tsc_khz;
5460 cpu_notifier_register_begin();
5461 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5462 #ifdef CONFIG_CPU_FREQ
5463 struct cpufreq_policy policy;
5464 memset(&policy, 0, sizeof(policy));
5466 cpufreq_get_policy(&policy, cpu);
5467 if (policy.cpuinfo.max_freq)
5468 max_tsc_khz = policy.cpuinfo.max_freq;
5471 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5472 CPUFREQ_TRANSITION_NOTIFIER);
5474 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5475 for_each_online_cpu(cpu)
5476 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5478 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5479 cpu_notifier_register_done();
5483 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5485 int kvm_is_in_guest(void)
5487 return __this_cpu_read(current_vcpu) != NULL;
5490 static int kvm_is_user_mode(void)
5494 if (__this_cpu_read(current_vcpu))
5495 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5497 return user_mode != 0;
5500 static unsigned long kvm_get_guest_ip(void)
5502 unsigned long ip = 0;
5504 if (__this_cpu_read(current_vcpu))
5505 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5510 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5511 .is_in_guest = kvm_is_in_guest,
5512 .is_user_mode = kvm_is_user_mode,
5513 .get_guest_ip = kvm_get_guest_ip,
5516 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5518 __this_cpu_write(current_vcpu, vcpu);
5520 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5522 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5524 __this_cpu_write(current_vcpu, NULL);
5526 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5528 static void kvm_set_mmio_spte_mask(void)
5531 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5534 * Set the reserved bits and the present bit of an paging-structure
5535 * entry to generate page fault with PFER.RSV = 1.
5537 /* Mask the reserved physical address bits. */
5538 mask = rsvd_bits(maxphyaddr, 51);
5540 /* Bit 62 is always reserved for 32bit host. */
5541 mask |= 0x3ull << 62;
5543 /* Set the present bit. */
5546 #ifdef CONFIG_X86_64
5548 * If reserved bit is not supported, clear the present bit to disable
5551 if (maxphyaddr == 52)
5555 kvm_mmu_set_mmio_spte_mask(mask);
5558 #ifdef CONFIG_X86_64
5559 static void pvclock_gtod_update_fn(struct work_struct *work)
5563 struct kvm_vcpu *vcpu;
5566 spin_lock(&kvm_lock);
5567 list_for_each_entry(kvm, &vm_list, vm_list)
5568 kvm_for_each_vcpu(i, vcpu, kvm)
5569 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5570 atomic_set(&kvm_guest_has_master_clock, 0);
5571 spin_unlock(&kvm_lock);
5574 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5577 * Notification about pvclock gtod data update.
5579 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5582 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5583 struct timekeeper *tk = priv;
5585 update_pvclock_gtod(tk);
5587 /* disable master clock if host does not trust, or does not
5588 * use, TSC clocksource
5590 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5591 atomic_read(&kvm_guest_has_master_clock) != 0)
5592 queue_work(system_long_wq, &pvclock_gtod_work);
5597 static struct notifier_block pvclock_gtod_notifier = {
5598 .notifier_call = pvclock_gtod_notify,
5602 int kvm_arch_init(void *opaque)
5605 struct kvm_x86_ops *ops = opaque;
5608 printk(KERN_ERR "kvm: already loaded the other module\n");
5613 if (!ops->cpu_has_kvm_support()) {
5614 printk(KERN_ERR "kvm: no hardware support\n");
5618 if (ops->disabled_by_bios()) {
5619 printk(KERN_ERR "kvm: disabled by bios\n");
5625 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5627 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5631 r = kvm_mmu_module_init();
5633 goto out_free_percpu;
5635 kvm_set_mmio_spte_mask();
5639 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5640 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5644 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5647 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5650 #ifdef CONFIG_X86_64
5651 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5657 free_percpu(shared_msrs);
5662 void kvm_arch_exit(void)
5664 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5666 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5667 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5668 CPUFREQ_TRANSITION_NOTIFIER);
5669 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5670 #ifdef CONFIG_X86_64
5671 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5674 kvm_mmu_module_exit();
5675 free_percpu(shared_msrs);
5678 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5680 ++vcpu->stat.halt_exits;
5681 if (irqchip_in_kernel(vcpu->kvm)) {
5682 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5685 vcpu->run->exit_reason = KVM_EXIT_HLT;
5689 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5691 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5693 kvm_x86_ops->skip_emulated_instruction(vcpu);
5694 return kvm_vcpu_halt(vcpu);
5696 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5699 * kvm_pv_kick_cpu_op: Kick a vcpu.
5701 * @apicid - apicid of vcpu to be kicked.
5703 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5705 struct kvm_lapic_irq lapic_irq;
5707 lapic_irq.shorthand = 0;
5708 lapic_irq.dest_mode = 0;
5709 lapic_irq.dest_id = apicid;
5710 lapic_irq.msi_redir_hint = false;
5712 lapic_irq.delivery_mode = APIC_DM_REMRD;
5713 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5716 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5718 unsigned long nr, a0, a1, a2, a3, ret;
5719 int op_64_bit, r = 1;
5721 kvm_x86_ops->skip_emulated_instruction(vcpu);
5723 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5724 return kvm_hv_hypercall(vcpu);
5726 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5727 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5728 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5729 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5730 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5732 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5734 op_64_bit = is_64_bit_mode(vcpu);
5743 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5749 case KVM_HC_VAPIC_POLL_IRQ:
5752 case KVM_HC_KICK_CPU:
5753 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5763 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5764 ++vcpu->stat.hypercalls;
5767 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5769 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5771 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5772 char instruction[3];
5773 unsigned long rip = kvm_rip_read(vcpu);
5775 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5777 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5781 * Check if userspace requested an interrupt window, and that the
5782 * interrupt window is open.
5784 * No need to exit to userspace if we already have an interrupt queued.
5786 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5788 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5789 vcpu->run->request_interrupt_window &&
5790 kvm_arch_interrupt_allowed(vcpu));
5793 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5795 struct kvm_run *kvm_run = vcpu->run;
5797 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5798 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
5799 kvm_run->cr8 = kvm_get_cr8(vcpu);
5800 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5801 if (irqchip_in_kernel(vcpu->kvm))
5802 kvm_run->ready_for_interrupt_injection = 1;
5804 kvm_run->ready_for_interrupt_injection =
5805 kvm_arch_interrupt_allowed(vcpu) &&
5806 !kvm_cpu_has_interrupt(vcpu) &&
5807 !kvm_event_needs_reinjection(vcpu);
5810 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5814 if (!kvm_x86_ops->update_cr8_intercept)
5817 if (!vcpu->arch.apic)
5820 if (!vcpu->arch.apic->vapic_addr)
5821 max_irr = kvm_lapic_find_highest_irr(vcpu);
5828 tpr = kvm_lapic_get_cr8(vcpu);
5830 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5833 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
5837 /* try to reinject previous events if any */
5838 if (vcpu->arch.exception.pending) {
5839 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5840 vcpu->arch.exception.has_error_code,
5841 vcpu->arch.exception.error_code);
5843 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
5844 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
5847 if (vcpu->arch.exception.nr == DB_VECTOR &&
5848 (vcpu->arch.dr7 & DR7_GD)) {
5849 vcpu->arch.dr7 &= ~DR7_GD;
5850 kvm_update_dr7(vcpu);
5853 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5854 vcpu->arch.exception.has_error_code,
5855 vcpu->arch.exception.error_code,
5856 vcpu->arch.exception.reinject);
5860 if (vcpu->arch.nmi_injected) {
5861 kvm_x86_ops->set_nmi(vcpu);
5865 if (vcpu->arch.interrupt.pending) {
5866 kvm_x86_ops->set_irq(vcpu);
5870 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5871 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5876 /* try to inject new event if pending */
5877 if (vcpu->arch.nmi_pending) {
5878 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5879 --vcpu->arch.nmi_pending;
5880 vcpu->arch.nmi_injected = true;
5881 kvm_x86_ops->set_nmi(vcpu);
5883 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
5885 * Because interrupts can be injected asynchronously, we are
5886 * calling check_nested_events again here to avoid a race condition.
5887 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
5888 * proposal and current concerns. Perhaps we should be setting
5889 * KVM_REQ_EVENT only on certain events and not unconditionally?
5891 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5892 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5896 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5897 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5899 kvm_x86_ops->set_irq(vcpu);
5905 static void process_nmi(struct kvm_vcpu *vcpu)
5910 * x86 is limited to one NMI running, and one NMI pending after it.
5911 * If an NMI is already in progress, limit further NMIs to just one.
5912 * Otherwise, allow two (and we'll inject the first one immediately).
5914 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5917 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5918 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5919 kvm_make_request(KVM_REQ_EVENT, vcpu);
5922 #define put_smstate(type, buf, offset, val) \
5923 *(type *)((buf) + (offset) - 0x7e00) = val
5925 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
5928 flags |= seg->g << 23;
5929 flags |= seg->db << 22;
5930 flags |= seg->l << 21;
5931 flags |= seg->avl << 20;
5932 flags |= seg->present << 15;
5933 flags |= seg->dpl << 13;
5934 flags |= seg->s << 12;
5935 flags |= seg->type << 8;
5939 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
5941 struct kvm_segment seg;
5944 kvm_get_segment(vcpu, &seg, n);
5945 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
5948 offset = 0x7f84 + n * 12;
5950 offset = 0x7f2c + (n - 3) * 12;
5952 put_smstate(u32, buf, offset + 8, seg.base);
5953 put_smstate(u32, buf, offset + 4, seg.limit);
5954 put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
5957 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
5959 struct kvm_segment seg;
5963 kvm_get_segment(vcpu, &seg, n);
5964 offset = 0x7e00 + n * 16;
5966 flags = process_smi_get_segment_flags(&seg) >> 8;
5967 put_smstate(u16, buf, offset, seg.selector);
5968 put_smstate(u16, buf, offset + 2, flags);
5969 put_smstate(u32, buf, offset + 4, seg.limit);
5970 put_smstate(u64, buf, offset + 8, seg.base);
5973 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
5976 struct kvm_segment seg;
5980 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
5981 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
5982 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
5983 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
5985 for (i = 0; i < 8; i++)
5986 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
5988 kvm_get_dr(vcpu, 6, &val);
5989 put_smstate(u32, buf, 0x7fcc, (u32)val);
5990 kvm_get_dr(vcpu, 7, &val);
5991 put_smstate(u32, buf, 0x7fc8, (u32)val);
5993 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
5994 put_smstate(u32, buf, 0x7fc4, seg.selector);
5995 put_smstate(u32, buf, 0x7f64, seg.base);
5996 put_smstate(u32, buf, 0x7f60, seg.limit);
5997 put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
5999 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6000 put_smstate(u32, buf, 0x7fc0, seg.selector);
6001 put_smstate(u32, buf, 0x7f80, seg.base);
6002 put_smstate(u32, buf, 0x7f7c, seg.limit);
6003 put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6005 kvm_x86_ops->get_gdt(vcpu, &dt);
6006 put_smstate(u32, buf, 0x7f74, dt.address);
6007 put_smstate(u32, buf, 0x7f70, dt.size);
6009 kvm_x86_ops->get_idt(vcpu, &dt);
6010 put_smstate(u32, buf, 0x7f58, dt.address);
6011 put_smstate(u32, buf, 0x7f54, dt.size);
6013 for (i = 0; i < 6; i++)
6014 process_smi_save_seg_32(vcpu, buf, i);
6016 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6019 put_smstate(u32, buf, 0x7efc, 0x00020000);
6020 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6023 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6025 #ifdef CONFIG_X86_64
6027 struct kvm_segment seg;
6031 for (i = 0; i < 16; i++)
6032 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6034 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6035 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6037 kvm_get_dr(vcpu, 6, &val);
6038 put_smstate(u64, buf, 0x7f68, val);
6039 kvm_get_dr(vcpu, 7, &val);
6040 put_smstate(u64, buf, 0x7f60, val);
6042 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6043 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6044 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6046 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6049 put_smstate(u32, buf, 0x7efc, 0x00020064);
6051 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6053 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6054 put_smstate(u16, buf, 0x7e90, seg.selector);
6055 put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6056 put_smstate(u32, buf, 0x7e94, seg.limit);
6057 put_smstate(u64, buf, 0x7e98, seg.base);
6059 kvm_x86_ops->get_idt(vcpu, &dt);
6060 put_smstate(u32, buf, 0x7e84, dt.size);
6061 put_smstate(u64, buf, 0x7e88, dt.address);
6063 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6064 put_smstate(u16, buf, 0x7e70, seg.selector);
6065 put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6066 put_smstate(u32, buf, 0x7e74, seg.limit);
6067 put_smstate(u64, buf, 0x7e78, seg.base);
6069 kvm_x86_ops->get_gdt(vcpu, &dt);
6070 put_smstate(u32, buf, 0x7e64, dt.size);
6071 put_smstate(u64, buf, 0x7e68, dt.address);
6073 for (i = 0; i < 6; i++)
6074 process_smi_save_seg_64(vcpu, buf, i);
6080 static void process_smi(struct kvm_vcpu *vcpu)
6082 struct kvm_segment cs, ds;
6087 vcpu->arch.smi_pending = true;
6091 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6092 vcpu->arch.hflags |= HF_SMM_MASK;
6093 memset(buf, 0, 512);
6094 if (guest_cpuid_has_longmode(vcpu))
6095 process_smi_save_state_64(vcpu, buf);
6097 process_smi_save_state_32(vcpu, buf);
6099 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6101 if (kvm_x86_ops->get_nmi_mask(vcpu))
6102 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6104 kvm_x86_ops->set_nmi_mask(vcpu, true);
6106 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6107 kvm_rip_write(vcpu, 0x8000);
6109 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6110 kvm_x86_ops->set_cr0(vcpu, cr0);
6111 vcpu->arch.cr0 = cr0;
6113 kvm_x86_ops->set_cr4(vcpu, 0);
6115 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6117 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6118 cs.base = vcpu->arch.smbase;
6123 cs.limit = ds.limit = 0xffffffff;
6124 cs.type = ds.type = 0x3;
6125 cs.dpl = ds.dpl = 0;
6130 cs.avl = ds.avl = 0;
6131 cs.present = ds.present = 1;
6132 cs.unusable = ds.unusable = 0;
6133 cs.padding = ds.padding = 0;
6135 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6136 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6137 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6138 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6139 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6140 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6142 if (guest_cpuid_has_longmode(vcpu))
6143 kvm_x86_ops->set_efer(vcpu, 0);
6145 kvm_update_cpuid(vcpu);
6146 kvm_mmu_reset_context(vcpu);
6149 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6151 u64 eoi_exit_bitmap[4];
6154 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6157 memset(eoi_exit_bitmap, 0, 32);
6160 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6161 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6162 kvm_apic_update_tmr(vcpu, tmr);
6165 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6167 ++vcpu->stat.tlb_flush;
6168 kvm_x86_ops->tlb_flush(vcpu);
6171 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6173 struct page *page = NULL;
6175 if (!irqchip_in_kernel(vcpu->kvm))
6178 if (!kvm_x86_ops->set_apic_access_page_addr)
6181 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6182 if (is_error_page(page))
6184 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6187 * Do not pin apic access page in memory, the MMU notifier
6188 * will call us again if it is migrated or swapped out.
6192 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6194 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6195 unsigned long address)
6198 * The physical address of apic access page is stored in the VMCS.
6199 * Update it when it becomes invalid.
6201 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6202 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6206 * Returns 1 to let vcpu_run() continue the guest execution loop without
6207 * exiting to the userspace. Otherwise, the value will be returned to the
6210 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6213 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6214 vcpu->run->request_interrupt_window;
6215 bool req_immediate_exit = false;
6217 if (vcpu->requests) {
6218 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6219 kvm_mmu_unload(vcpu);
6220 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6221 __kvm_migrate_timers(vcpu);
6222 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6223 kvm_gen_update_masterclock(vcpu->kvm);
6224 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6225 kvm_gen_kvmclock_update(vcpu);
6226 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6227 r = kvm_guest_time_update(vcpu);
6231 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6232 kvm_mmu_sync_roots(vcpu);
6233 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6234 kvm_vcpu_flush_tlb(vcpu);
6235 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6236 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6240 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6241 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6245 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6246 vcpu->fpu_active = 0;
6247 kvm_x86_ops->fpu_deactivate(vcpu);
6249 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6250 /* Page is swapped out. Do synthetic halt */
6251 vcpu->arch.apf.halted = true;
6255 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6256 record_steal_time(vcpu);
6257 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6259 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6261 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6262 kvm_pmu_handle_event(vcpu);
6263 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6264 kvm_pmu_deliver_pmi(vcpu);
6265 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6266 vcpu_scan_ioapic(vcpu);
6267 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6268 kvm_vcpu_reload_apic_access_page(vcpu);
6269 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6270 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6271 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6277 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6278 kvm_apic_accept_events(vcpu);
6279 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6284 if (inject_pending_event(vcpu, req_int_win) != 0)
6285 req_immediate_exit = true;
6286 /* enable NMI/IRQ window open exits if needed */
6287 else if (vcpu->arch.nmi_pending)
6288 kvm_x86_ops->enable_nmi_window(vcpu);
6289 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6290 kvm_x86_ops->enable_irq_window(vcpu);
6292 if (kvm_lapic_enabled(vcpu)) {
6294 * Update architecture specific hints for APIC
6295 * virtual interrupt delivery.
6297 if (kvm_x86_ops->hwapic_irr_update)
6298 kvm_x86_ops->hwapic_irr_update(vcpu,
6299 kvm_lapic_find_highest_irr(vcpu));
6300 update_cr8_intercept(vcpu);
6301 kvm_lapic_sync_to_vapic(vcpu);
6305 r = kvm_mmu_reload(vcpu);
6307 goto cancel_injection;
6312 kvm_x86_ops->prepare_guest_switch(vcpu);
6313 if (vcpu->fpu_active)
6314 kvm_load_guest_fpu(vcpu);
6315 kvm_load_guest_xcr0(vcpu);
6317 vcpu->mode = IN_GUEST_MODE;
6319 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6321 /* We should set ->mode before check ->requests,
6322 * see the comment in make_all_cpus_request.
6324 smp_mb__after_srcu_read_unlock();
6326 local_irq_disable();
6328 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6329 || need_resched() || signal_pending(current)) {
6330 vcpu->mode = OUTSIDE_GUEST_MODE;
6334 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6336 goto cancel_injection;
6339 if (req_immediate_exit)
6340 smp_send_reschedule(vcpu->cpu);
6342 __kvm_guest_enter();
6344 if (unlikely(vcpu->arch.switch_db_regs)) {
6346 set_debugreg(vcpu->arch.eff_db[0], 0);
6347 set_debugreg(vcpu->arch.eff_db[1], 1);
6348 set_debugreg(vcpu->arch.eff_db[2], 2);
6349 set_debugreg(vcpu->arch.eff_db[3], 3);
6350 set_debugreg(vcpu->arch.dr6, 6);
6351 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6354 trace_kvm_entry(vcpu->vcpu_id);
6355 wait_lapic_expire(vcpu);
6356 kvm_x86_ops->run(vcpu);
6359 * Do this here before restoring debug registers on the host. And
6360 * since we do this before handling the vmexit, a DR access vmexit
6361 * can (a) read the correct value of the debug registers, (b) set
6362 * KVM_DEBUGREG_WONT_EXIT again.
6364 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6367 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6368 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6369 for (i = 0; i < KVM_NR_DB_REGS; i++)
6370 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6374 * If the guest has used debug registers, at least dr7
6375 * will be disabled while returning to the host.
6376 * If we don't have active breakpoints in the host, we don't
6377 * care about the messed up debug address registers. But if
6378 * we have some of them active, restore the old state.
6380 if (hw_breakpoint_active())
6381 hw_breakpoint_restore();
6383 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6386 vcpu->mode = OUTSIDE_GUEST_MODE;
6389 /* Interrupt is enabled by handle_external_intr() */
6390 kvm_x86_ops->handle_external_intr(vcpu);
6395 * We must have an instruction between local_irq_enable() and
6396 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6397 * the interrupt shadow. The stat.exits increment will do nicely.
6398 * But we need to prevent reordering, hence this barrier():
6406 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6409 * Profile KVM exit RIPs:
6411 if (unlikely(prof_on == KVM_PROFILING)) {
6412 unsigned long rip = kvm_rip_read(vcpu);
6413 profile_hit(KVM_PROFILING, (void *)rip);
6416 if (unlikely(vcpu->arch.tsc_always_catchup))
6417 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6419 if (vcpu->arch.apic_attention)
6420 kvm_lapic_sync_from_vapic(vcpu);
6422 r = kvm_x86_ops->handle_exit(vcpu);
6426 kvm_x86_ops->cancel_injection(vcpu);
6427 if (unlikely(vcpu->arch.apic_attention))
6428 kvm_lapic_sync_from_vapic(vcpu);
6433 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6435 if (!kvm_arch_vcpu_runnable(vcpu)) {
6436 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6437 kvm_vcpu_block(vcpu);
6438 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6439 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6443 kvm_apic_accept_events(vcpu);
6444 switch(vcpu->arch.mp_state) {
6445 case KVM_MP_STATE_HALTED:
6446 vcpu->arch.pv.pv_unhalted = false;
6447 vcpu->arch.mp_state =
6448 KVM_MP_STATE_RUNNABLE;
6449 case KVM_MP_STATE_RUNNABLE:
6450 vcpu->arch.apf.halted = false;
6452 case KVM_MP_STATE_INIT_RECEIVED:
6461 static int vcpu_run(struct kvm_vcpu *vcpu)
6464 struct kvm *kvm = vcpu->kvm;
6466 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6469 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6470 !vcpu->arch.apf.halted)
6471 r = vcpu_enter_guest(vcpu);
6473 r = vcpu_block(kvm, vcpu);
6477 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6478 if (kvm_cpu_has_pending_timer(vcpu))
6479 kvm_inject_pending_timer_irqs(vcpu);
6481 if (dm_request_for_irq_injection(vcpu)) {
6483 vcpu->run->exit_reason = KVM_EXIT_INTR;
6484 ++vcpu->stat.request_irq_exits;
6488 kvm_check_async_pf_completion(vcpu);
6490 if (signal_pending(current)) {
6492 vcpu->run->exit_reason = KVM_EXIT_INTR;
6493 ++vcpu->stat.signal_exits;
6496 if (need_resched()) {
6497 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6499 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6503 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6508 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6511 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6512 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6513 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6514 if (r != EMULATE_DONE)
6519 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6521 BUG_ON(!vcpu->arch.pio.count);
6523 return complete_emulated_io(vcpu);
6527 * Implements the following, as a state machine:
6531 * for each mmio piece in the fragment
6539 * for each mmio piece in the fragment
6544 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6546 struct kvm_run *run = vcpu->run;
6547 struct kvm_mmio_fragment *frag;
6550 BUG_ON(!vcpu->mmio_needed);
6552 /* Complete previous fragment */
6553 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6554 len = min(8u, frag->len);
6555 if (!vcpu->mmio_is_write)
6556 memcpy(frag->data, run->mmio.data, len);
6558 if (frag->len <= 8) {
6559 /* Switch to the next fragment. */
6561 vcpu->mmio_cur_fragment++;
6563 /* Go forward to the next mmio piece. */
6569 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6570 vcpu->mmio_needed = 0;
6572 /* FIXME: return into emulator if single-stepping. */
6573 if (vcpu->mmio_is_write)
6575 vcpu->mmio_read_completed = 1;
6576 return complete_emulated_io(vcpu);
6579 run->exit_reason = KVM_EXIT_MMIO;
6580 run->mmio.phys_addr = frag->gpa;
6581 if (vcpu->mmio_is_write)
6582 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6583 run->mmio.len = min(8u, frag->len);
6584 run->mmio.is_write = vcpu->mmio_is_write;
6585 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6590 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6592 struct fpu *fpu = ¤t->thread.fpu;
6596 fpu__activate_curr(fpu);
6598 if (vcpu->sigset_active)
6599 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6601 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6602 kvm_vcpu_block(vcpu);
6603 kvm_apic_accept_events(vcpu);
6604 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6609 /* re-sync apic's tpr */
6610 if (!irqchip_in_kernel(vcpu->kvm)) {
6611 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6617 if (unlikely(vcpu->arch.complete_userspace_io)) {
6618 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6619 vcpu->arch.complete_userspace_io = NULL;
6624 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6629 post_kvm_run_save(vcpu);
6630 if (vcpu->sigset_active)
6631 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6636 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6638 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6640 * We are here if userspace calls get_regs() in the middle of
6641 * instruction emulation. Registers state needs to be copied
6642 * back from emulation context to vcpu. Userspace shouldn't do
6643 * that usually, but some bad designed PV devices (vmware
6644 * backdoor interface) need this to work
6646 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6647 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6649 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6650 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6651 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6652 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6653 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6654 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6655 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6656 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6657 #ifdef CONFIG_X86_64
6658 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6659 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6660 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6661 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6662 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6663 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6664 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6665 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6668 regs->rip = kvm_rip_read(vcpu);
6669 regs->rflags = kvm_get_rflags(vcpu);
6674 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6676 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6677 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6679 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6680 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6681 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6682 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6683 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6684 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6685 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6686 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6687 #ifdef CONFIG_X86_64
6688 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6689 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6690 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6691 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6692 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6693 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6694 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6695 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6698 kvm_rip_write(vcpu, regs->rip);
6699 kvm_set_rflags(vcpu, regs->rflags);
6701 vcpu->arch.exception.pending = false;
6703 kvm_make_request(KVM_REQ_EVENT, vcpu);
6708 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6710 struct kvm_segment cs;
6712 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6716 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6718 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6719 struct kvm_sregs *sregs)
6723 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6724 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6725 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6726 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6727 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6728 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6730 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6731 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6733 kvm_x86_ops->get_idt(vcpu, &dt);
6734 sregs->idt.limit = dt.size;
6735 sregs->idt.base = dt.address;
6736 kvm_x86_ops->get_gdt(vcpu, &dt);
6737 sregs->gdt.limit = dt.size;
6738 sregs->gdt.base = dt.address;
6740 sregs->cr0 = kvm_read_cr0(vcpu);
6741 sregs->cr2 = vcpu->arch.cr2;
6742 sregs->cr3 = kvm_read_cr3(vcpu);
6743 sregs->cr4 = kvm_read_cr4(vcpu);
6744 sregs->cr8 = kvm_get_cr8(vcpu);
6745 sregs->efer = vcpu->arch.efer;
6746 sregs->apic_base = kvm_get_apic_base(vcpu);
6748 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6750 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6751 set_bit(vcpu->arch.interrupt.nr,
6752 (unsigned long *)sregs->interrupt_bitmap);
6757 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6758 struct kvm_mp_state *mp_state)
6760 kvm_apic_accept_events(vcpu);
6761 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6762 vcpu->arch.pv.pv_unhalted)
6763 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6765 mp_state->mp_state = vcpu->arch.mp_state;
6770 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6771 struct kvm_mp_state *mp_state)
6773 if (!kvm_vcpu_has_lapic(vcpu) &&
6774 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6777 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6778 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6779 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6781 vcpu->arch.mp_state = mp_state->mp_state;
6782 kvm_make_request(KVM_REQ_EVENT, vcpu);
6786 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6787 int reason, bool has_error_code, u32 error_code)
6789 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6792 init_emulate_ctxt(vcpu);
6794 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6795 has_error_code, error_code);
6798 return EMULATE_FAIL;
6800 kvm_rip_write(vcpu, ctxt->eip);
6801 kvm_set_rflags(vcpu, ctxt->eflags);
6802 kvm_make_request(KVM_REQ_EVENT, vcpu);
6803 return EMULATE_DONE;
6805 EXPORT_SYMBOL_GPL(kvm_task_switch);
6807 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6808 struct kvm_sregs *sregs)
6810 struct msr_data apic_base_msr;
6811 int mmu_reset_needed = 0;
6812 int pending_vec, max_bits, idx;
6815 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6818 dt.size = sregs->idt.limit;
6819 dt.address = sregs->idt.base;
6820 kvm_x86_ops->set_idt(vcpu, &dt);
6821 dt.size = sregs->gdt.limit;
6822 dt.address = sregs->gdt.base;
6823 kvm_x86_ops->set_gdt(vcpu, &dt);
6825 vcpu->arch.cr2 = sregs->cr2;
6826 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6827 vcpu->arch.cr3 = sregs->cr3;
6828 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6830 kvm_set_cr8(vcpu, sregs->cr8);
6832 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6833 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6834 apic_base_msr.data = sregs->apic_base;
6835 apic_base_msr.host_initiated = true;
6836 kvm_set_apic_base(vcpu, &apic_base_msr);
6838 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6839 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6840 vcpu->arch.cr0 = sregs->cr0;
6842 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6843 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6844 if (sregs->cr4 & X86_CR4_OSXSAVE)
6845 kvm_update_cpuid(vcpu);
6847 idx = srcu_read_lock(&vcpu->kvm->srcu);
6848 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6849 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6850 mmu_reset_needed = 1;
6852 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6854 if (mmu_reset_needed)
6855 kvm_mmu_reset_context(vcpu);
6857 max_bits = KVM_NR_INTERRUPTS;
6858 pending_vec = find_first_bit(
6859 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6860 if (pending_vec < max_bits) {
6861 kvm_queue_interrupt(vcpu, pending_vec, false);
6862 pr_debug("Set back pending irq %d\n", pending_vec);
6865 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6866 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6867 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6868 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6869 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6870 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6872 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6873 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6875 update_cr8_intercept(vcpu);
6877 /* Older userspace won't unhalt the vcpu on reset. */
6878 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6879 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6881 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6883 kvm_make_request(KVM_REQ_EVENT, vcpu);
6888 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6889 struct kvm_guest_debug *dbg)
6891 unsigned long rflags;
6894 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6896 if (vcpu->arch.exception.pending)
6898 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6899 kvm_queue_exception(vcpu, DB_VECTOR);
6901 kvm_queue_exception(vcpu, BP_VECTOR);
6905 * Read rflags as long as potentially injected trace flags are still
6908 rflags = kvm_get_rflags(vcpu);
6910 vcpu->guest_debug = dbg->control;
6911 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6912 vcpu->guest_debug = 0;
6914 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6915 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6916 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6917 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6919 for (i = 0; i < KVM_NR_DB_REGS; i++)
6920 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6922 kvm_update_dr7(vcpu);
6924 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6925 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6926 get_segment_base(vcpu, VCPU_SREG_CS);
6929 * Trigger an rflags update that will inject or remove the trace
6932 kvm_set_rflags(vcpu, rflags);
6934 kvm_x86_ops->update_db_bp_intercept(vcpu);
6944 * Translate a guest virtual address to a guest physical address.
6946 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6947 struct kvm_translation *tr)
6949 unsigned long vaddr = tr->linear_address;
6953 idx = srcu_read_lock(&vcpu->kvm->srcu);
6954 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6955 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6956 tr->physical_address = gpa;
6957 tr->valid = gpa != UNMAPPED_GVA;
6964 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6966 struct fxregs_state *fxsave =
6967 &vcpu->arch.guest_fpu.state.fxsave;
6969 memcpy(fpu->fpr, fxsave->st_space, 128);
6970 fpu->fcw = fxsave->cwd;
6971 fpu->fsw = fxsave->swd;
6972 fpu->ftwx = fxsave->twd;
6973 fpu->last_opcode = fxsave->fop;
6974 fpu->last_ip = fxsave->rip;
6975 fpu->last_dp = fxsave->rdp;
6976 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6981 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6983 struct fxregs_state *fxsave =
6984 &vcpu->arch.guest_fpu.state.fxsave;
6986 memcpy(fxsave->st_space, fpu->fpr, 128);
6987 fxsave->cwd = fpu->fcw;
6988 fxsave->swd = fpu->fsw;
6989 fxsave->twd = fpu->ftwx;
6990 fxsave->fop = fpu->last_opcode;
6991 fxsave->rip = fpu->last_ip;
6992 fxsave->rdp = fpu->last_dp;
6993 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6998 static void fx_init(struct kvm_vcpu *vcpu)
7000 fpstate_init(&vcpu->arch.guest_fpu.state);
7002 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7003 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7006 * Ensure guest xcr0 is valid for loading
7008 vcpu->arch.xcr0 = XSTATE_FP;
7010 vcpu->arch.cr0 |= X86_CR0_ET;
7013 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7015 if (vcpu->guest_fpu_loaded)
7019 * Restore all possible states in the guest,
7020 * and assume host would use all available bits.
7021 * Guest xcr0 would be loaded later.
7023 kvm_put_guest_xcr0(vcpu);
7024 vcpu->guest_fpu_loaded = 1;
7025 __kernel_fpu_begin();
7026 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7030 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7032 kvm_put_guest_xcr0(vcpu);
7034 if (!vcpu->guest_fpu_loaded) {
7035 vcpu->fpu_counter = 0;
7039 vcpu->guest_fpu_loaded = 0;
7040 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7042 ++vcpu->stat.fpu_reload;
7044 * If using eager FPU mode, or if the guest is a frequent user
7045 * of the FPU, just leave the FPU active for next time.
7046 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7047 * the FPU in bursts will revert to loading it on demand.
7049 if (!vcpu->arch.eager_fpu) {
7050 if (++vcpu->fpu_counter < 5)
7051 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7056 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7058 kvmclock_reset(vcpu);
7060 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7061 kvm_x86_ops->vcpu_free(vcpu);
7064 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7067 struct kvm_vcpu *vcpu;
7069 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7070 printk_once(KERN_WARNING
7071 "kvm: SMP vm created on host with unstable TSC; "
7072 "guest TSC will not be reliable\n");
7074 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7079 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7083 kvm_vcpu_mtrr_init(vcpu);
7084 r = vcpu_load(vcpu);
7087 kvm_vcpu_reset(vcpu, false);
7088 kvm_mmu_setup(vcpu);
7093 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7095 struct msr_data msr;
7096 struct kvm *kvm = vcpu->kvm;
7098 if (vcpu_load(vcpu))
7101 msr.index = MSR_IA32_TSC;
7102 msr.host_initiated = true;
7103 kvm_write_tsc(vcpu, &msr);
7106 if (!kvmclock_periodic_sync)
7109 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7110 KVMCLOCK_SYNC_PERIOD);
7113 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7116 vcpu->arch.apf.msr_val = 0;
7118 r = vcpu_load(vcpu);
7120 kvm_mmu_unload(vcpu);
7123 kvm_x86_ops->vcpu_free(vcpu);
7126 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7128 vcpu->arch.hflags = 0;
7130 atomic_set(&vcpu->arch.nmi_queued, 0);
7131 vcpu->arch.nmi_pending = 0;
7132 vcpu->arch.nmi_injected = false;
7133 kvm_clear_interrupt_queue(vcpu);
7134 kvm_clear_exception_queue(vcpu);
7136 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7137 kvm_update_dr0123(vcpu);
7138 vcpu->arch.dr6 = DR6_INIT;
7139 kvm_update_dr6(vcpu);
7140 vcpu->arch.dr7 = DR7_FIXED_1;
7141 kvm_update_dr7(vcpu);
7145 kvm_make_request(KVM_REQ_EVENT, vcpu);
7146 vcpu->arch.apf.msr_val = 0;
7147 vcpu->arch.st.msr_val = 0;
7149 kvmclock_reset(vcpu);
7151 kvm_clear_async_pf_completion_queue(vcpu);
7152 kvm_async_pf_hash_reset(vcpu);
7153 vcpu->arch.apf.halted = false;
7156 kvm_pmu_reset(vcpu);
7157 vcpu->arch.smbase = 0x30000;
7160 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7161 vcpu->arch.regs_avail = ~0;
7162 vcpu->arch.regs_dirty = ~0;
7164 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7167 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7169 struct kvm_segment cs;
7171 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7172 cs.selector = vector << 8;
7173 cs.base = vector << 12;
7174 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7175 kvm_rip_write(vcpu, 0);
7178 int kvm_arch_hardware_enable(void)
7181 struct kvm_vcpu *vcpu;
7186 bool stable, backwards_tsc = false;
7188 kvm_shared_msr_cpu_online();
7189 ret = kvm_x86_ops->hardware_enable();
7193 local_tsc = native_read_tsc();
7194 stable = !check_tsc_unstable();
7195 list_for_each_entry(kvm, &vm_list, vm_list) {
7196 kvm_for_each_vcpu(i, vcpu, kvm) {
7197 if (!stable && vcpu->cpu == smp_processor_id())
7198 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7199 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7200 backwards_tsc = true;
7201 if (vcpu->arch.last_host_tsc > max_tsc)
7202 max_tsc = vcpu->arch.last_host_tsc;
7208 * Sometimes, even reliable TSCs go backwards. This happens on
7209 * platforms that reset TSC during suspend or hibernate actions, but
7210 * maintain synchronization. We must compensate. Fortunately, we can
7211 * detect that condition here, which happens early in CPU bringup,
7212 * before any KVM threads can be running. Unfortunately, we can't
7213 * bring the TSCs fully up to date with real time, as we aren't yet far
7214 * enough into CPU bringup that we know how much real time has actually
7215 * elapsed; our helper function, get_kernel_ns() will be using boot
7216 * variables that haven't been updated yet.
7218 * So we simply find the maximum observed TSC above, then record the
7219 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7220 * the adjustment will be applied. Note that we accumulate
7221 * adjustments, in case multiple suspend cycles happen before some VCPU
7222 * gets a chance to run again. In the event that no KVM threads get a
7223 * chance to run, we will miss the entire elapsed period, as we'll have
7224 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7225 * loose cycle time. This isn't too big a deal, since the loss will be
7226 * uniform across all VCPUs (not to mention the scenario is extremely
7227 * unlikely). It is possible that a second hibernate recovery happens
7228 * much faster than a first, causing the observed TSC here to be
7229 * smaller; this would require additional padding adjustment, which is
7230 * why we set last_host_tsc to the local tsc observed here.
7232 * N.B. - this code below runs only on platforms with reliable TSC,
7233 * as that is the only way backwards_tsc is set above. Also note
7234 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7235 * have the same delta_cyc adjustment applied if backwards_tsc
7236 * is detected. Note further, this adjustment is only done once,
7237 * as we reset last_host_tsc on all VCPUs to stop this from being
7238 * called multiple times (one for each physical CPU bringup).
7240 * Platforms with unreliable TSCs don't have to deal with this, they
7241 * will be compensated by the logic in vcpu_load, which sets the TSC to
7242 * catchup mode. This will catchup all VCPUs to real time, but cannot
7243 * guarantee that they stay in perfect synchronization.
7245 if (backwards_tsc) {
7246 u64 delta_cyc = max_tsc - local_tsc;
7247 backwards_tsc_observed = true;
7248 list_for_each_entry(kvm, &vm_list, vm_list) {
7249 kvm_for_each_vcpu(i, vcpu, kvm) {
7250 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7251 vcpu->arch.last_host_tsc = local_tsc;
7252 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7256 * We have to disable TSC offset matching.. if you were
7257 * booting a VM while issuing an S4 host suspend....
7258 * you may have some problem. Solving this issue is
7259 * left as an exercise to the reader.
7261 kvm->arch.last_tsc_nsec = 0;
7262 kvm->arch.last_tsc_write = 0;
7269 void kvm_arch_hardware_disable(void)
7271 kvm_x86_ops->hardware_disable();
7272 drop_user_return_notifiers();
7275 int kvm_arch_hardware_setup(void)
7279 r = kvm_x86_ops->hardware_setup();
7283 kvm_init_msr_list();
7287 void kvm_arch_hardware_unsetup(void)
7289 kvm_x86_ops->hardware_unsetup();
7292 void kvm_arch_check_processor_compat(void *rtn)
7294 kvm_x86_ops->check_processor_compatibility(rtn);
7297 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7299 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7301 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7303 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7305 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7308 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7310 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7313 struct static_key kvm_no_apic_vcpu __read_mostly;
7315 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7321 BUG_ON(vcpu->kvm == NULL);
7324 vcpu->arch.pv.pv_unhalted = false;
7325 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7326 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7327 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7329 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7331 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7336 vcpu->arch.pio_data = page_address(page);
7338 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7340 r = kvm_mmu_create(vcpu);
7342 goto fail_free_pio_data;
7344 if (irqchip_in_kernel(kvm)) {
7345 r = kvm_create_lapic(vcpu);
7347 goto fail_mmu_destroy;
7349 static_key_slow_inc(&kvm_no_apic_vcpu);
7351 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7353 if (!vcpu->arch.mce_banks) {
7355 goto fail_free_lapic;
7357 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7359 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7361 goto fail_free_mce_banks;
7366 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7367 vcpu->arch.pv_time_enabled = false;
7369 vcpu->arch.guest_supported_xcr0 = 0;
7370 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7372 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7374 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7376 kvm_async_pf_hash_reset(vcpu);
7381 fail_free_mce_banks:
7382 kfree(vcpu->arch.mce_banks);
7384 kvm_free_lapic(vcpu);
7386 kvm_mmu_destroy(vcpu);
7388 free_page((unsigned long)vcpu->arch.pio_data);
7393 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7397 kvm_pmu_destroy(vcpu);
7398 kfree(vcpu->arch.mce_banks);
7399 kvm_free_lapic(vcpu);
7400 idx = srcu_read_lock(&vcpu->kvm->srcu);
7401 kvm_mmu_destroy(vcpu);
7402 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7403 free_page((unsigned long)vcpu->arch.pio_data);
7404 if (!irqchip_in_kernel(vcpu->kvm))
7405 static_key_slow_dec(&kvm_no_apic_vcpu);
7408 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7410 kvm_x86_ops->sched_in(vcpu, cpu);
7413 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7418 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7419 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7420 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7421 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7422 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7424 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7425 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7426 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7427 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7428 &kvm->arch.irq_sources_bitmap);
7430 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7431 mutex_init(&kvm->arch.apic_map_lock);
7432 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7434 pvclock_update_vm_gtod_copy(kvm);
7436 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7437 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7442 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7445 r = vcpu_load(vcpu);
7447 kvm_mmu_unload(vcpu);
7451 static void kvm_free_vcpus(struct kvm *kvm)
7454 struct kvm_vcpu *vcpu;
7457 * Unpin any mmu pages first.
7459 kvm_for_each_vcpu(i, vcpu, kvm) {
7460 kvm_clear_async_pf_completion_queue(vcpu);
7461 kvm_unload_vcpu_mmu(vcpu);
7463 kvm_for_each_vcpu(i, vcpu, kvm)
7464 kvm_arch_vcpu_free(vcpu);
7466 mutex_lock(&kvm->lock);
7467 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7468 kvm->vcpus[i] = NULL;
7470 atomic_set(&kvm->online_vcpus, 0);
7471 mutex_unlock(&kvm->lock);
7474 void kvm_arch_sync_events(struct kvm *kvm)
7476 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7477 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7478 kvm_free_all_assigned_devices(kvm);
7482 int __x86_set_memory_region(struct kvm *kvm,
7483 const struct kvm_userspace_memory_region *mem)
7487 /* Called with kvm->slots_lock held. */
7488 BUG_ON(mem->slot >= KVM_MEM_SLOTS_NUM);
7490 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7491 struct kvm_userspace_memory_region m = *mem;
7494 r = __kvm_set_memory_region(kvm, &m);
7501 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7503 int x86_set_memory_region(struct kvm *kvm,
7504 const struct kvm_userspace_memory_region *mem)
7508 mutex_lock(&kvm->slots_lock);
7509 r = __x86_set_memory_region(kvm, mem);
7510 mutex_unlock(&kvm->slots_lock);
7514 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7516 void kvm_arch_destroy_vm(struct kvm *kvm)
7518 if (current->mm == kvm->mm) {
7520 * Free memory regions allocated on behalf of userspace,
7521 * unless the the memory map has changed due to process exit
7524 struct kvm_userspace_memory_region mem;
7525 memset(&mem, 0, sizeof(mem));
7526 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7527 x86_set_memory_region(kvm, &mem);
7529 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7530 x86_set_memory_region(kvm, &mem);
7532 mem.slot = TSS_PRIVATE_MEMSLOT;
7533 x86_set_memory_region(kvm, &mem);
7535 kvm_iommu_unmap_guest(kvm);
7536 kfree(kvm->arch.vpic);
7537 kfree(kvm->arch.vioapic);
7538 kvm_free_vcpus(kvm);
7539 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7542 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7543 struct kvm_memory_slot *dont)
7547 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7548 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7549 kvfree(free->arch.rmap[i]);
7550 free->arch.rmap[i] = NULL;
7555 if (!dont || free->arch.lpage_info[i - 1] !=
7556 dont->arch.lpage_info[i - 1]) {
7557 kvfree(free->arch.lpage_info[i - 1]);
7558 free->arch.lpage_info[i - 1] = NULL;
7563 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7564 unsigned long npages)
7568 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7573 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7574 slot->base_gfn, level) + 1;
7576 slot->arch.rmap[i] =
7577 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7578 if (!slot->arch.rmap[i])
7583 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7584 sizeof(*slot->arch.lpage_info[i - 1]));
7585 if (!slot->arch.lpage_info[i - 1])
7588 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7589 slot->arch.lpage_info[i - 1][0].write_count = 1;
7590 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7591 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7592 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7594 * If the gfn and userspace address are not aligned wrt each
7595 * other, or if explicitly asked to, disable large page
7596 * support for this slot
7598 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7599 !kvm_largepages_enabled()) {
7602 for (j = 0; j < lpages; ++j)
7603 slot->arch.lpage_info[i - 1][j].write_count = 1;
7610 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7611 kvfree(slot->arch.rmap[i]);
7612 slot->arch.rmap[i] = NULL;
7616 kvfree(slot->arch.lpage_info[i - 1]);
7617 slot->arch.lpage_info[i - 1] = NULL;
7622 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7625 * memslots->generation has been incremented.
7626 * mmio generation may have reached its maximum value.
7628 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7631 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7632 struct kvm_memory_slot *memslot,
7633 const struct kvm_userspace_memory_region *mem,
7634 enum kvm_mr_change change)
7637 * Only private memory slots need to be mapped here since
7638 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7640 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7641 unsigned long userspace_addr;
7644 * MAP_SHARED to prevent internal slot pages from being moved
7647 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7648 PROT_READ | PROT_WRITE,
7649 MAP_SHARED | MAP_ANONYMOUS, 0);
7651 if (IS_ERR((void *)userspace_addr))
7652 return PTR_ERR((void *)userspace_addr);
7654 memslot->userspace_addr = userspace_addr;
7660 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7661 struct kvm_memory_slot *new)
7663 /* Still write protect RO slot */
7664 if (new->flags & KVM_MEM_READONLY) {
7665 kvm_mmu_slot_remove_write_access(kvm, new);
7670 * Call kvm_x86_ops dirty logging hooks when they are valid.
7672 * kvm_x86_ops->slot_disable_log_dirty is called when:
7674 * - KVM_MR_CREATE with dirty logging is disabled
7675 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7677 * The reason is, in case of PML, we need to set D-bit for any slots
7678 * with dirty logging disabled in order to eliminate unnecessary GPA
7679 * logging in PML buffer (and potential PML buffer full VMEXT). This
7680 * guarantees leaving PML enabled during guest's lifetime won't have
7681 * any additonal overhead from PML when guest is running with dirty
7682 * logging disabled for memory slots.
7684 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7685 * to dirty logging mode.
7687 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7689 * In case of write protect:
7691 * Write protect all pages for dirty logging.
7693 * All the sptes including the large sptes which point to this
7694 * slot are set to readonly. We can not create any new large
7695 * spte on this slot until the end of the logging.
7697 * See the comments in fast_page_fault().
7699 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7700 if (kvm_x86_ops->slot_enable_log_dirty)
7701 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7703 kvm_mmu_slot_remove_write_access(kvm, new);
7705 if (kvm_x86_ops->slot_disable_log_dirty)
7706 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7710 void kvm_arch_commit_memory_region(struct kvm *kvm,
7711 const struct kvm_userspace_memory_region *mem,
7712 const struct kvm_memory_slot *old,
7713 const struct kvm_memory_slot *new,
7714 enum kvm_mr_change change)
7716 int nr_mmu_pages = 0;
7718 if (change == KVM_MR_DELETE && old->id >= KVM_USER_MEM_SLOTS) {
7721 ret = vm_munmap(old->userspace_addr,
7722 old->npages * PAGE_SIZE);
7725 "kvm_vm_ioctl_set_memory_region: "
7726 "failed to munmap memory\n");
7729 if (!kvm->arch.n_requested_mmu_pages)
7730 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7733 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7736 * Dirty logging tracks sptes in 4k granularity, meaning that large
7737 * sptes have to be split. If live migration is successful, the guest
7738 * in the source machine will be destroyed and large sptes will be
7739 * created in the destination. However, if the guest continues to run
7740 * in the source machine (for example if live migration fails), small
7741 * sptes will remain around and cause bad performance.
7743 * Scan sptes if dirty logging has been stopped, dropping those
7744 * which can be collapsed into a single large-page spte. Later
7745 * page faults will create the large-page sptes.
7747 if ((change != KVM_MR_DELETE) &&
7748 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7749 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7750 kvm_mmu_zap_collapsible_sptes(kvm, new);
7753 * Set up write protection and/or dirty logging for the new slot.
7755 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7756 * been zapped so no dirty logging staff is needed for old slot. For
7757 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7758 * new and it's also covered when dealing with the new slot.
7760 * FIXME: const-ify all uses of struct kvm_memory_slot.
7762 if (change != KVM_MR_DELETE)
7763 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
7766 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7768 kvm_mmu_invalidate_zap_all_pages(kvm);
7771 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7772 struct kvm_memory_slot *slot)
7774 kvm_mmu_invalidate_zap_all_pages(kvm);
7777 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7779 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7780 kvm_x86_ops->check_nested_events(vcpu, false);
7782 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7783 !vcpu->arch.apf.halted)
7784 || !list_empty_careful(&vcpu->async_pf.done)
7785 || kvm_apic_has_events(vcpu)
7786 || vcpu->arch.pv.pv_unhalted
7787 || atomic_read(&vcpu->arch.nmi_queued) ||
7788 (kvm_arch_interrupt_allowed(vcpu) &&
7789 kvm_cpu_has_interrupt(vcpu));
7792 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7794 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7797 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7799 return kvm_x86_ops->interrupt_allowed(vcpu);
7802 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
7804 if (is_64_bit_mode(vcpu))
7805 return kvm_rip_read(vcpu);
7806 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
7807 kvm_rip_read(vcpu));
7809 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
7811 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7813 return kvm_get_linear_rip(vcpu) == linear_rip;
7815 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7817 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7819 unsigned long rflags;
7821 rflags = kvm_x86_ops->get_rflags(vcpu);
7822 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7823 rflags &= ~X86_EFLAGS_TF;
7826 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7828 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7830 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7831 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7832 rflags |= X86_EFLAGS_TF;
7833 kvm_x86_ops->set_rflags(vcpu, rflags);
7836 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7838 __kvm_set_rflags(vcpu, rflags);
7839 kvm_make_request(KVM_REQ_EVENT, vcpu);
7841 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7843 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7847 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7851 r = kvm_mmu_reload(vcpu);
7855 if (!vcpu->arch.mmu.direct_map &&
7856 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7859 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7862 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7864 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7867 static inline u32 kvm_async_pf_next_probe(u32 key)
7869 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7872 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7874 u32 key = kvm_async_pf_hash_fn(gfn);
7876 while (vcpu->arch.apf.gfns[key] != ~0)
7877 key = kvm_async_pf_next_probe(key);
7879 vcpu->arch.apf.gfns[key] = gfn;
7882 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7885 u32 key = kvm_async_pf_hash_fn(gfn);
7887 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7888 (vcpu->arch.apf.gfns[key] != gfn &&
7889 vcpu->arch.apf.gfns[key] != ~0); i++)
7890 key = kvm_async_pf_next_probe(key);
7895 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7897 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7900 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7904 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7906 vcpu->arch.apf.gfns[i] = ~0;
7908 j = kvm_async_pf_next_probe(j);
7909 if (vcpu->arch.apf.gfns[j] == ~0)
7911 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7913 * k lies cyclically in ]i,j]
7915 * |....j i.k.| or |.k..j i...|
7917 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7918 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7923 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7926 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7930 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7931 struct kvm_async_pf *work)
7933 struct x86_exception fault;
7935 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7936 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7938 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7939 (vcpu->arch.apf.send_user_only &&
7940 kvm_x86_ops->get_cpl(vcpu) == 0))
7941 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7942 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7943 fault.vector = PF_VECTOR;
7944 fault.error_code_valid = true;
7945 fault.error_code = 0;
7946 fault.nested_page_fault = false;
7947 fault.address = work->arch.token;
7948 kvm_inject_page_fault(vcpu, &fault);
7952 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7953 struct kvm_async_pf *work)
7955 struct x86_exception fault;
7957 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7958 if (work->wakeup_all)
7959 work->arch.token = ~0; /* broadcast wakeup */
7961 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7963 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7964 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7965 fault.vector = PF_VECTOR;
7966 fault.error_code_valid = true;
7967 fault.error_code = 0;
7968 fault.nested_page_fault = false;
7969 fault.address = work->arch.token;
7970 kvm_inject_page_fault(vcpu, &fault);
7972 vcpu->arch.apf.halted = false;
7973 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7976 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7978 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7981 return !kvm_event_needs_reinjection(vcpu) &&
7982 kvm_x86_ops->interrupt_allowed(vcpu);
7985 void kvm_arch_start_assignment(struct kvm *kvm)
7987 atomic_inc(&kvm->arch.assigned_device_count);
7989 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
7991 void kvm_arch_end_assignment(struct kvm *kvm)
7993 atomic_dec(&kvm->arch.assigned_device_count);
7995 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
7997 bool kvm_arch_has_assigned_device(struct kvm *kvm)
7999 return atomic_read(&kvm->arch.assigned_device_count);
8001 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8003 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8005 atomic_inc(&kvm->arch.noncoherent_dma_count);
8007 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8009 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8011 atomic_dec(&kvm->arch.noncoherent_dma_count);
8013 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8015 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8017 return atomic_read(&kvm->arch.noncoherent_dma_count);
8019 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8021 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8022 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8023 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8024 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8025 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8026 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8027 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8028 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8029 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8030 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8031 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8032 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8033 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8034 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8035 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);