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"
31 #include <linux/clocksource.h>
32 #include <linux/interrupt.h>
33 #include <linux/kvm.h>
35 #include <linux/vmalloc.h>
36 #include <linux/module.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <trace/events/kvm.h>
53 #define CREATE_TRACE_POINTS
56 #include <asm/debugreg.h>
62 #include <asm/fpu-internal.h> /* Ugh! */
64 #include <asm/pvclock.h>
65 #include <asm/div64.h>
67 #define MAX_IO_MSRS 256
68 #define KVM_MAX_MCE_BANKS 32
69 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
71 #define emul_to_vcpu(ctxt) \
72 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
75 * - enable syscall per default because its emulated by KVM
76 * - enable LME and LMA per default on 64 bit KVM
80 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
82 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
85 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
86 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
88 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
89 static void process_nmi(struct kvm_vcpu *vcpu);
90 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
92 struct kvm_x86_ops *kvm_x86_ops;
93 EXPORT_SYMBOL_GPL(kvm_x86_ops);
95 static bool ignore_msrs = 0;
96 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
98 unsigned int min_timer_period_us = 500;
99 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
101 bool kvm_has_tsc_control;
102 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
103 u32 kvm_max_guest_tsc_khz;
104 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
106 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
107 static u32 tsc_tolerance_ppm = 250;
108 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
110 static bool backwards_tsc_observed = false;
112 #define KVM_NR_SHARED_MSRS 16
114 struct kvm_shared_msrs_global {
116 u32 msrs[KVM_NR_SHARED_MSRS];
119 struct kvm_shared_msrs {
120 struct user_return_notifier urn;
122 struct kvm_shared_msr_values {
125 } values[KVM_NR_SHARED_MSRS];
128 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
129 static struct kvm_shared_msrs __percpu *shared_msrs;
131 struct kvm_stats_debugfs_item debugfs_entries[] = {
132 { "pf_fixed", VCPU_STAT(pf_fixed) },
133 { "pf_guest", VCPU_STAT(pf_guest) },
134 { "tlb_flush", VCPU_STAT(tlb_flush) },
135 { "invlpg", VCPU_STAT(invlpg) },
136 { "exits", VCPU_STAT(exits) },
137 { "io_exits", VCPU_STAT(io_exits) },
138 { "mmio_exits", VCPU_STAT(mmio_exits) },
139 { "signal_exits", VCPU_STAT(signal_exits) },
140 { "irq_window", VCPU_STAT(irq_window_exits) },
141 { "nmi_window", VCPU_STAT(nmi_window_exits) },
142 { "halt_exits", VCPU_STAT(halt_exits) },
143 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
144 { "hypercalls", VCPU_STAT(hypercalls) },
145 { "request_irq", VCPU_STAT(request_irq_exits) },
146 { "irq_exits", VCPU_STAT(irq_exits) },
147 { "host_state_reload", VCPU_STAT(host_state_reload) },
148 { "efer_reload", VCPU_STAT(efer_reload) },
149 { "fpu_reload", VCPU_STAT(fpu_reload) },
150 { "insn_emulation", VCPU_STAT(insn_emulation) },
151 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
152 { "irq_injections", VCPU_STAT(irq_injections) },
153 { "nmi_injections", VCPU_STAT(nmi_injections) },
154 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
155 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
156 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
157 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
158 { "mmu_flooded", VM_STAT(mmu_flooded) },
159 { "mmu_recycled", VM_STAT(mmu_recycled) },
160 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
161 { "mmu_unsync", VM_STAT(mmu_unsync) },
162 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
163 { "largepages", VM_STAT(lpages) },
167 u64 __read_mostly host_xcr0;
169 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
171 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
174 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
175 vcpu->arch.apf.gfns[i] = ~0;
178 static void kvm_on_user_return(struct user_return_notifier *urn)
181 struct kvm_shared_msrs *locals
182 = container_of(urn, struct kvm_shared_msrs, urn);
183 struct kvm_shared_msr_values *values;
185 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
186 values = &locals->values[slot];
187 if (values->host != values->curr) {
188 wrmsrl(shared_msrs_global.msrs[slot], values->host);
189 values->curr = values->host;
192 locals->registered = false;
193 user_return_notifier_unregister(urn);
196 static void shared_msr_update(unsigned slot, u32 msr)
199 unsigned int cpu = smp_processor_id();
200 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
202 /* only read, and nobody should modify it at this time,
203 * so don't need lock */
204 if (slot >= shared_msrs_global.nr) {
205 printk(KERN_ERR "kvm: invalid MSR slot!");
208 rdmsrl_safe(msr, &value);
209 smsr->values[slot].host = value;
210 smsr->values[slot].curr = value;
213 void kvm_define_shared_msr(unsigned slot, u32 msr)
215 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
216 if (slot >= shared_msrs_global.nr)
217 shared_msrs_global.nr = slot + 1;
218 shared_msrs_global.msrs[slot] = msr;
219 /* we need ensured the shared_msr_global have been updated */
222 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
224 static void kvm_shared_msr_cpu_online(void)
228 for (i = 0; i < shared_msrs_global.nr; ++i)
229 shared_msr_update(i, shared_msrs_global.msrs[i]);
232 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
234 unsigned int cpu = smp_processor_id();
235 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
238 if (((value ^ smsr->values[slot].curr) & mask) == 0)
240 smsr->values[slot].curr = value;
241 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
245 if (!smsr->registered) {
246 smsr->urn.on_user_return = kvm_on_user_return;
247 user_return_notifier_register(&smsr->urn);
248 smsr->registered = true;
252 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
254 static void drop_user_return_notifiers(void)
256 unsigned int cpu = smp_processor_id();
257 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
259 if (smsr->registered)
260 kvm_on_user_return(&smsr->urn);
263 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
265 return vcpu->arch.apic_base;
267 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
269 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
271 u64 old_state = vcpu->arch.apic_base &
272 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
273 u64 new_state = msr_info->data &
274 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
275 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
276 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
278 if (!msr_info->host_initiated &&
279 ((msr_info->data & reserved_bits) != 0 ||
280 new_state == X2APIC_ENABLE ||
281 (new_state == MSR_IA32_APICBASE_ENABLE &&
282 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
283 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
287 kvm_lapic_set_base(vcpu, msr_info->data);
290 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
292 asmlinkage __visible void kvm_spurious_fault(void)
294 /* Fault while not rebooting. We want the trace. */
297 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
299 #define EXCPT_BENIGN 0
300 #define EXCPT_CONTRIBUTORY 1
303 static int exception_class(int vector)
313 return EXCPT_CONTRIBUTORY;
320 #define EXCPT_FAULT 0
322 #define EXCPT_ABORT 2
323 #define EXCPT_INTERRUPT 3
325 static int exception_type(int vector)
329 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
330 return EXCPT_INTERRUPT;
334 /* #DB is trap, as instruction watchpoints are handled elsewhere */
335 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
338 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
341 /* Reserved exceptions will result in fault */
345 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
346 unsigned nr, bool has_error, u32 error_code,
352 kvm_make_request(KVM_REQ_EVENT, vcpu);
354 if (!vcpu->arch.exception.pending) {
356 vcpu->arch.exception.pending = true;
357 vcpu->arch.exception.has_error_code = has_error;
358 vcpu->arch.exception.nr = nr;
359 vcpu->arch.exception.error_code = error_code;
360 vcpu->arch.exception.reinject = reinject;
364 /* to check exception */
365 prev_nr = vcpu->arch.exception.nr;
366 if (prev_nr == DF_VECTOR) {
367 /* triple fault -> shutdown */
368 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
371 class1 = exception_class(prev_nr);
372 class2 = exception_class(nr);
373 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
374 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
375 /* generate double fault per SDM Table 5-5 */
376 vcpu->arch.exception.pending = true;
377 vcpu->arch.exception.has_error_code = true;
378 vcpu->arch.exception.nr = DF_VECTOR;
379 vcpu->arch.exception.error_code = 0;
381 /* replace previous exception with a new one in a hope
382 that instruction re-execution will regenerate lost
387 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
389 kvm_multiple_exception(vcpu, nr, false, 0, false);
391 EXPORT_SYMBOL_GPL(kvm_queue_exception);
393 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
395 kvm_multiple_exception(vcpu, nr, false, 0, true);
397 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
399 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
402 kvm_inject_gp(vcpu, 0);
404 kvm_x86_ops->skip_emulated_instruction(vcpu);
406 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
408 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
410 ++vcpu->stat.pf_guest;
411 vcpu->arch.cr2 = fault->address;
412 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
414 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
416 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
418 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
419 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
421 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
423 return fault->nested_page_fault;
426 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
428 atomic_inc(&vcpu->arch.nmi_queued);
429 kvm_make_request(KVM_REQ_NMI, vcpu);
431 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
433 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
435 kvm_multiple_exception(vcpu, nr, true, error_code, false);
437 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
439 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
441 kvm_multiple_exception(vcpu, nr, true, error_code, true);
443 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
446 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
447 * a #GP and return false.
449 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
451 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
453 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
456 EXPORT_SYMBOL_GPL(kvm_require_cpl);
459 * This function will be used to read from the physical memory of the currently
460 * running guest. The difference to kvm_read_guest_page is that this function
461 * can read from guest physical or from the guest's guest physical memory.
463 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
464 gfn_t ngfn, void *data, int offset, int len,
467 struct x86_exception exception;
471 ngpa = gfn_to_gpa(ngfn);
472 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
473 if (real_gfn == UNMAPPED_GVA)
476 real_gfn = gpa_to_gfn(real_gfn);
478 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
480 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
482 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
483 void *data, int offset, int len, u32 access)
485 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
486 data, offset, len, access);
490 * Load the pae pdptrs. Return true is they are all valid.
492 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
494 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
495 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
498 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
500 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
501 offset * sizeof(u64), sizeof(pdpte),
502 PFERR_USER_MASK|PFERR_WRITE_MASK);
507 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
508 if (is_present_gpte(pdpte[i]) &&
509 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
516 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
517 __set_bit(VCPU_EXREG_PDPTR,
518 (unsigned long *)&vcpu->arch.regs_avail);
519 __set_bit(VCPU_EXREG_PDPTR,
520 (unsigned long *)&vcpu->arch.regs_dirty);
525 EXPORT_SYMBOL_GPL(load_pdptrs);
527 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
529 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
535 if (is_long_mode(vcpu) || !is_pae(vcpu))
538 if (!test_bit(VCPU_EXREG_PDPTR,
539 (unsigned long *)&vcpu->arch.regs_avail))
542 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
543 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
544 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
545 PFERR_USER_MASK | PFERR_WRITE_MASK);
548 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
554 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
556 unsigned long old_cr0 = kvm_read_cr0(vcpu);
557 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
558 X86_CR0_CD | X86_CR0_NW;
563 if (cr0 & 0xffffffff00000000UL)
567 cr0 &= ~CR0_RESERVED_BITS;
569 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
572 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
575 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
577 if ((vcpu->arch.efer & EFER_LME)) {
582 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
587 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
592 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
595 kvm_x86_ops->set_cr0(vcpu, cr0);
597 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
598 kvm_clear_async_pf_completion_queue(vcpu);
599 kvm_async_pf_hash_reset(vcpu);
602 if ((cr0 ^ old_cr0) & update_bits)
603 kvm_mmu_reset_context(vcpu);
606 EXPORT_SYMBOL_GPL(kvm_set_cr0);
608 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
610 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
612 EXPORT_SYMBOL_GPL(kvm_lmsw);
614 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
616 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
617 !vcpu->guest_xcr0_loaded) {
618 /* kvm_set_xcr() also depends on this */
619 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
620 vcpu->guest_xcr0_loaded = 1;
624 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
626 if (vcpu->guest_xcr0_loaded) {
627 if (vcpu->arch.xcr0 != host_xcr0)
628 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
629 vcpu->guest_xcr0_loaded = 0;
633 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
636 u64 old_xcr0 = vcpu->arch.xcr0;
639 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
640 if (index != XCR_XFEATURE_ENABLED_MASK)
642 if (!(xcr0 & XSTATE_FP))
644 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
648 * Do not allow the guest to set bits that we do not support
649 * saving. However, xcr0 bit 0 is always set, even if the
650 * emulated CPU does not support XSAVE (see fx_init).
652 valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
653 if (xcr0 & ~valid_bits)
656 if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
659 kvm_put_guest_xcr0(vcpu);
660 vcpu->arch.xcr0 = xcr0;
662 if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
663 kvm_update_cpuid(vcpu);
667 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
669 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
670 __kvm_set_xcr(vcpu, index, xcr)) {
671 kvm_inject_gp(vcpu, 0);
676 EXPORT_SYMBOL_GPL(kvm_set_xcr);
678 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
680 unsigned long old_cr4 = kvm_read_cr4(vcpu);
681 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
682 X86_CR4_PAE | X86_CR4_SMEP;
683 if (cr4 & CR4_RESERVED_BITS)
686 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
689 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
692 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
695 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
698 if (is_long_mode(vcpu)) {
699 if (!(cr4 & X86_CR4_PAE))
701 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
702 && ((cr4 ^ old_cr4) & pdptr_bits)
703 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
707 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
708 if (!guest_cpuid_has_pcid(vcpu))
711 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
712 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
716 if (kvm_x86_ops->set_cr4(vcpu, cr4))
719 if (((cr4 ^ old_cr4) & pdptr_bits) ||
720 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
721 kvm_mmu_reset_context(vcpu);
723 if ((cr4 ^ old_cr4) & X86_CR4_SMAP)
724 update_permission_bitmask(vcpu, vcpu->arch.walk_mmu, false);
726 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
727 kvm_update_cpuid(vcpu);
731 EXPORT_SYMBOL_GPL(kvm_set_cr4);
733 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
735 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
736 kvm_mmu_sync_roots(vcpu);
737 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
741 if (is_long_mode(vcpu)) {
742 if (cr3 & CR3_L_MODE_RESERVED_BITS)
744 } else if (is_pae(vcpu) && is_paging(vcpu) &&
745 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
748 vcpu->arch.cr3 = cr3;
749 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
750 kvm_mmu_new_cr3(vcpu);
753 EXPORT_SYMBOL_GPL(kvm_set_cr3);
755 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
757 if (cr8 & CR8_RESERVED_BITS)
759 if (irqchip_in_kernel(vcpu->kvm))
760 kvm_lapic_set_tpr(vcpu, cr8);
762 vcpu->arch.cr8 = cr8;
765 EXPORT_SYMBOL_GPL(kvm_set_cr8);
767 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
769 if (irqchip_in_kernel(vcpu->kvm))
770 return kvm_lapic_get_cr8(vcpu);
772 return vcpu->arch.cr8;
774 EXPORT_SYMBOL_GPL(kvm_get_cr8);
776 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
778 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
779 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
782 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
786 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
787 dr7 = vcpu->arch.guest_debug_dr7;
789 dr7 = vcpu->arch.dr7;
790 kvm_x86_ops->set_dr7(vcpu, dr7);
791 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
792 if (dr7 & DR7_BP_EN_MASK)
793 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
796 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
798 u64 fixed = DR6_FIXED_1;
800 if (!guest_cpuid_has_rtm(vcpu))
805 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
809 vcpu->arch.db[dr] = val;
810 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
811 vcpu->arch.eff_db[dr] = val;
814 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
818 if (val & 0xffffffff00000000ULL)
820 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
821 kvm_update_dr6(vcpu);
824 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
828 if (val & 0xffffffff00000000ULL)
830 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
831 kvm_update_dr7(vcpu);
838 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
842 res = __kvm_set_dr(vcpu, dr, val);
844 kvm_queue_exception(vcpu, UD_VECTOR);
846 kvm_inject_gp(vcpu, 0);
850 EXPORT_SYMBOL_GPL(kvm_set_dr);
852 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
856 *val = vcpu->arch.db[dr];
859 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
863 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
864 *val = vcpu->arch.dr6;
866 *val = kvm_x86_ops->get_dr6(vcpu);
869 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
873 *val = vcpu->arch.dr7;
880 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
882 if (_kvm_get_dr(vcpu, dr, val)) {
883 kvm_queue_exception(vcpu, UD_VECTOR);
888 EXPORT_SYMBOL_GPL(kvm_get_dr);
890 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
892 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
896 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
899 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
900 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
903 EXPORT_SYMBOL_GPL(kvm_rdpmc);
906 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
907 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
909 * This list is modified at module load time to reflect the
910 * capabilities of the host cpu. This capabilities test skips MSRs that are
911 * kvm-specific. Those are put in the beginning of the list.
914 #define KVM_SAVE_MSRS_BEGIN 12
915 static u32 msrs_to_save[] = {
916 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
917 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
918 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
919 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
920 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
922 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
925 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
927 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
928 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
931 static unsigned num_msrs_to_save;
933 static const u32 emulated_msrs[] = {
935 MSR_IA32_TSCDEADLINE,
936 MSR_IA32_MISC_ENABLE,
941 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
943 if (efer & efer_reserved_bits)
946 if (efer & EFER_FFXSR) {
947 struct kvm_cpuid_entry2 *feat;
949 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
950 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
954 if (efer & EFER_SVME) {
955 struct kvm_cpuid_entry2 *feat;
957 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
958 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
964 EXPORT_SYMBOL_GPL(kvm_valid_efer);
966 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
968 u64 old_efer = vcpu->arch.efer;
970 if (!kvm_valid_efer(vcpu, efer))
974 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
978 efer |= vcpu->arch.efer & EFER_LMA;
980 kvm_x86_ops->set_efer(vcpu, efer);
982 /* Update reserved bits */
983 if ((efer ^ old_efer) & EFER_NX)
984 kvm_mmu_reset_context(vcpu);
989 void kvm_enable_efer_bits(u64 mask)
991 efer_reserved_bits &= ~mask;
993 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
996 * Writes msr value into into the appropriate "register".
997 * Returns 0 on success, non-0 otherwise.
998 * Assumes vcpu_load() was already called.
1000 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1002 switch (msr->index) {
1005 case MSR_KERNEL_GS_BASE:
1008 if (is_noncanonical_address(msr->data))
1011 case MSR_IA32_SYSENTER_EIP:
1012 case MSR_IA32_SYSENTER_ESP:
1014 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1015 * non-canonical address is written on Intel but not on
1016 * AMD (which ignores the top 32-bits, because it does
1017 * not implement 64-bit SYSENTER).
1019 * 64-bit code should hence be able to write a non-canonical
1020 * value on AMD. Making the address canonical ensures that
1021 * vmentry does not fail on Intel after writing a non-canonical
1022 * value, and that something deterministic happens if the guest
1023 * invokes 64-bit SYSENTER.
1025 msr->data = get_canonical(msr->data);
1027 return kvm_x86_ops->set_msr(vcpu, msr);
1029 EXPORT_SYMBOL_GPL(kvm_set_msr);
1032 * Adapt set_msr() to msr_io()'s calling convention
1034 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1036 struct msr_data msr;
1040 msr.host_initiated = true;
1041 return kvm_set_msr(vcpu, &msr);
1044 #ifdef CONFIG_X86_64
1045 struct pvclock_gtod_data {
1048 struct { /* extract of a clocksource struct */
1060 static struct pvclock_gtod_data pvclock_gtod_data;
1062 static void update_pvclock_gtod(struct timekeeper *tk)
1064 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1067 boot_ns = ktime_to_ns(ktime_add(tk->tkr.base_mono, tk->offs_boot));
1069 write_seqcount_begin(&vdata->seq);
1071 /* copy pvclock gtod data */
1072 vdata->clock.vclock_mode = tk->tkr.clock->archdata.vclock_mode;
1073 vdata->clock.cycle_last = tk->tkr.cycle_last;
1074 vdata->clock.mask = tk->tkr.mask;
1075 vdata->clock.mult = tk->tkr.mult;
1076 vdata->clock.shift = tk->tkr.shift;
1078 vdata->boot_ns = boot_ns;
1079 vdata->nsec_base = tk->tkr.xtime_nsec;
1081 write_seqcount_end(&vdata->seq);
1086 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1090 struct pvclock_wall_clock wc;
1091 struct timespec boot;
1096 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1101 ++version; /* first time write, random junk */
1105 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1108 * The guest calculates current wall clock time by adding
1109 * system time (updated by kvm_guest_time_update below) to the
1110 * wall clock specified here. guest system time equals host
1111 * system time for us, thus we must fill in host boot time here.
1115 if (kvm->arch.kvmclock_offset) {
1116 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1117 boot = timespec_sub(boot, ts);
1119 wc.sec = boot.tv_sec;
1120 wc.nsec = boot.tv_nsec;
1121 wc.version = version;
1123 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1126 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1129 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1131 uint32_t quotient, remainder;
1133 /* Don't try to replace with do_div(), this one calculates
1134 * "(dividend << 32) / divisor" */
1136 : "=a" (quotient), "=d" (remainder)
1137 : "0" (0), "1" (dividend), "r" (divisor) );
1141 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1142 s8 *pshift, u32 *pmultiplier)
1149 tps64 = base_khz * 1000LL;
1150 scaled64 = scaled_khz * 1000LL;
1151 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1156 tps32 = (uint32_t)tps64;
1157 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1158 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1166 *pmultiplier = div_frac(scaled64, tps32);
1168 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1169 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1172 static inline u64 get_kernel_ns(void)
1174 return ktime_get_boot_ns();
1177 #ifdef CONFIG_X86_64
1178 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1181 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1182 unsigned long max_tsc_khz;
1184 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1186 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1187 vcpu->arch.virtual_tsc_shift);
1190 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1192 u64 v = (u64)khz * (1000000 + ppm);
1197 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1199 u32 thresh_lo, thresh_hi;
1200 int use_scaling = 0;
1202 /* tsc_khz can be zero if TSC calibration fails */
1203 if (this_tsc_khz == 0)
1206 /* Compute a scale to convert nanoseconds in TSC cycles */
1207 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1208 &vcpu->arch.virtual_tsc_shift,
1209 &vcpu->arch.virtual_tsc_mult);
1210 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1213 * Compute the variation in TSC rate which is acceptable
1214 * within the range of tolerance and decide if the
1215 * rate being applied is within that bounds of the hardware
1216 * rate. If so, no scaling or compensation need be done.
1218 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1219 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1220 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1221 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1224 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1227 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1229 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1230 vcpu->arch.virtual_tsc_mult,
1231 vcpu->arch.virtual_tsc_shift);
1232 tsc += vcpu->arch.this_tsc_write;
1236 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1238 #ifdef CONFIG_X86_64
1240 bool do_request = false;
1241 struct kvm_arch *ka = &vcpu->kvm->arch;
1242 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1244 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1245 atomic_read(&vcpu->kvm->online_vcpus));
1247 if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1248 if (!ka->use_master_clock)
1251 if (!vcpus_matched && ka->use_master_clock)
1255 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1257 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1258 atomic_read(&vcpu->kvm->online_vcpus),
1259 ka->use_master_clock, gtod->clock.vclock_mode);
1263 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1265 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1266 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1269 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1271 struct kvm *kvm = vcpu->kvm;
1272 u64 offset, ns, elapsed;
1273 unsigned long flags;
1276 bool already_matched;
1277 u64 data = msr->data;
1279 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1280 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1281 ns = get_kernel_ns();
1282 elapsed = ns - kvm->arch.last_tsc_nsec;
1284 if (vcpu->arch.virtual_tsc_khz) {
1287 /* n.b - signed multiplication and division required */
1288 usdiff = data - kvm->arch.last_tsc_write;
1289 #ifdef CONFIG_X86_64
1290 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1292 /* do_div() only does unsigned */
1293 asm("1: idivl %[divisor]\n"
1294 "2: xor %%edx, %%edx\n"
1295 " movl $0, %[faulted]\n"
1297 ".section .fixup,\"ax\"\n"
1298 "4: movl $1, %[faulted]\n"
1302 _ASM_EXTABLE(1b, 4b)
1304 : "=A"(usdiff), [faulted] "=r" (faulted)
1305 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1308 do_div(elapsed, 1000);
1313 /* idivl overflow => difference is larger than USEC_PER_SEC */
1315 usdiff = USEC_PER_SEC;
1317 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1320 * Special case: TSC write with a small delta (1 second) of virtual
1321 * cycle time against real time is interpreted as an attempt to
1322 * synchronize the CPU.
1324 * For a reliable TSC, we can match TSC offsets, and for an unstable
1325 * TSC, we add elapsed time in this computation. We could let the
1326 * compensation code attempt to catch up if we fall behind, but
1327 * it's better to try to match offsets from the beginning.
1329 if (usdiff < USEC_PER_SEC &&
1330 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1331 if (!check_tsc_unstable()) {
1332 offset = kvm->arch.cur_tsc_offset;
1333 pr_debug("kvm: matched tsc offset for %llu\n", data);
1335 u64 delta = nsec_to_cycles(vcpu, elapsed);
1337 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1338 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1341 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1344 * We split periods of matched TSC writes into generations.
1345 * For each generation, we track the original measured
1346 * nanosecond time, offset, and write, so if TSCs are in
1347 * sync, we can match exact offset, and if not, we can match
1348 * exact software computation in compute_guest_tsc()
1350 * These values are tracked in kvm->arch.cur_xxx variables.
1352 kvm->arch.cur_tsc_generation++;
1353 kvm->arch.cur_tsc_nsec = ns;
1354 kvm->arch.cur_tsc_write = data;
1355 kvm->arch.cur_tsc_offset = offset;
1357 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1358 kvm->arch.cur_tsc_generation, data);
1362 * We also track th most recent recorded KHZ, write and time to
1363 * allow the matching interval to be extended at each write.
1365 kvm->arch.last_tsc_nsec = ns;
1366 kvm->arch.last_tsc_write = data;
1367 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1369 vcpu->arch.last_guest_tsc = data;
1371 /* Keep track of which generation this VCPU has synchronized to */
1372 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1373 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1374 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1376 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1377 update_ia32_tsc_adjust_msr(vcpu, offset);
1378 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1379 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1381 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1383 kvm->arch.nr_vcpus_matched_tsc = 0;
1384 } else if (!already_matched) {
1385 kvm->arch.nr_vcpus_matched_tsc++;
1388 kvm_track_tsc_matching(vcpu);
1389 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1392 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1394 #ifdef CONFIG_X86_64
1396 static cycle_t read_tsc(void)
1402 * Empirically, a fence (of type that depends on the CPU)
1403 * before rdtsc is enough to ensure that rdtsc is ordered
1404 * with respect to loads. The various CPU manuals are unclear
1405 * as to whether rdtsc can be reordered with later loads,
1406 * but no one has ever seen it happen.
1409 ret = (cycle_t)vget_cycles();
1411 last = pvclock_gtod_data.clock.cycle_last;
1413 if (likely(ret >= last))
1417 * GCC likes to generate cmov here, but this branch is extremely
1418 * predictable (it's just a funciton of time and the likely is
1419 * very likely) and there's a data dependence, so force GCC
1420 * to generate a branch instead. I don't barrier() because
1421 * we don't actually need a barrier, and if this function
1422 * ever gets inlined it will generate worse code.
1428 static inline u64 vgettsc(cycle_t *cycle_now)
1431 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1433 *cycle_now = read_tsc();
1435 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1436 return v * gtod->clock.mult;
1439 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1441 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1447 seq = read_seqcount_begin(>od->seq);
1448 mode = gtod->clock.vclock_mode;
1449 ns = gtod->nsec_base;
1450 ns += vgettsc(cycle_now);
1451 ns >>= gtod->clock.shift;
1452 ns += gtod->boot_ns;
1453 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1459 /* returns true if host is using tsc clocksource */
1460 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1462 /* checked again under seqlock below */
1463 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1466 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1472 * Assuming a stable TSC across physical CPUS, and a stable TSC
1473 * across virtual CPUs, the following condition is possible.
1474 * Each numbered line represents an event visible to both
1475 * CPUs at the next numbered event.
1477 * "timespecX" represents host monotonic time. "tscX" represents
1480 * VCPU0 on CPU0 | VCPU1 on CPU1
1482 * 1. read timespec0,tsc0
1483 * 2. | timespec1 = timespec0 + N
1485 * 3. transition to guest | transition to guest
1486 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1487 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1488 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1490 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1493 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1495 * - 0 < N - M => M < N
1497 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1498 * always the case (the difference between two distinct xtime instances
1499 * might be smaller then the difference between corresponding TSC reads,
1500 * when updating guest vcpus pvclock areas).
1502 * To avoid that problem, do not allow visibility of distinct
1503 * system_timestamp/tsc_timestamp values simultaneously: use a master
1504 * copy of host monotonic time values. Update that master copy
1507 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1511 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1513 #ifdef CONFIG_X86_64
1514 struct kvm_arch *ka = &kvm->arch;
1516 bool host_tsc_clocksource, vcpus_matched;
1518 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1519 atomic_read(&kvm->online_vcpus));
1522 * If the host uses TSC clock, then passthrough TSC as stable
1525 host_tsc_clocksource = kvm_get_time_and_clockread(
1526 &ka->master_kernel_ns,
1527 &ka->master_cycle_now);
1529 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1530 && !backwards_tsc_observed;
1532 if (ka->use_master_clock)
1533 atomic_set(&kvm_guest_has_master_clock, 1);
1535 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1536 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1541 static void kvm_gen_update_masterclock(struct kvm *kvm)
1543 #ifdef CONFIG_X86_64
1545 struct kvm_vcpu *vcpu;
1546 struct kvm_arch *ka = &kvm->arch;
1548 spin_lock(&ka->pvclock_gtod_sync_lock);
1549 kvm_make_mclock_inprogress_request(kvm);
1550 /* no guest entries from this point */
1551 pvclock_update_vm_gtod_copy(kvm);
1553 kvm_for_each_vcpu(i, vcpu, kvm)
1554 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1556 /* guest entries allowed */
1557 kvm_for_each_vcpu(i, vcpu, kvm)
1558 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1560 spin_unlock(&ka->pvclock_gtod_sync_lock);
1564 static int kvm_guest_time_update(struct kvm_vcpu *v)
1566 unsigned long flags, this_tsc_khz;
1567 struct kvm_vcpu_arch *vcpu = &v->arch;
1568 struct kvm_arch *ka = &v->kvm->arch;
1570 u64 tsc_timestamp, host_tsc;
1571 struct pvclock_vcpu_time_info guest_hv_clock;
1573 bool use_master_clock;
1579 * If the host uses TSC clock, then passthrough TSC as stable
1582 spin_lock(&ka->pvclock_gtod_sync_lock);
1583 use_master_clock = ka->use_master_clock;
1584 if (use_master_clock) {
1585 host_tsc = ka->master_cycle_now;
1586 kernel_ns = ka->master_kernel_ns;
1588 spin_unlock(&ka->pvclock_gtod_sync_lock);
1590 /* Keep irq disabled to prevent changes to the clock */
1591 local_irq_save(flags);
1592 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1593 if (unlikely(this_tsc_khz == 0)) {
1594 local_irq_restore(flags);
1595 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1598 if (!use_master_clock) {
1599 host_tsc = native_read_tsc();
1600 kernel_ns = get_kernel_ns();
1603 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1606 * We may have to catch up the TSC to match elapsed wall clock
1607 * time for two reasons, even if kvmclock is used.
1608 * 1) CPU could have been running below the maximum TSC rate
1609 * 2) Broken TSC compensation resets the base at each VCPU
1610 * entry to avoid unknown leaps of TSC even when running
1611 * again on the same CPU. This may cause apparent elapsed
1612 * time to disappear, and the guest to stand still or run
1615 if (vcpu->tsc_catchup) {
1616 u64 tsc = compute_guest_tsc(v, kernel_ns);
1617 if (tsc > tsc_timestamp) {
1618 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1619 tsc_timestamp = tsc;
1623 local_irq_restore(flags);
1625 if (!vcpu->pv_time_enabled)
1628 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1629 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1630 &vcpu->hv_clock.tsc_shift,
1631 &vcpu->hv_clock.tsc_to_system_mul);
1632 vcpu->hw_tsc_khz = this_tsc_khz;
1635 /* With all the info we got, fill in the values */
1636 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1637 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1638 vcpu->last_guest_tsc = tsc_timestamp;
1641 * The interface expects us to write an even number signaling that the
1642 * update is finished. Since the guest won't see the intermediate
1643 * state, we just increase by 2 at the end.
1645 vcpu->hv_clock.version += 2;
1647 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1648 &guest_hv_clock, sizeof(guest_hv_clock))))
1651 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1652 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1654 if (vcpu->pvclock_set_guest_stopped_request) {
1655 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1656 vcpu->pvclock_set_guest_stopped_request = false;
1659 /* If the host uses TSC clocksource, then it is stable */
1660 if (use_master_clock)
1661 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1663 vcpu->hv_clock.flags = pvclock_flags;
1665 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1667 sizeof(vcpu->hv_clock));
1672 * kvmclock updates which are isolated to a given vcpu, such as
1673 * vcpu->cpu migration, should not allow system_timestamp from
1674 * the rest of the vcpus to remain static. Otherwise ntp frequency
1675 * correction applies to one vcpu's system_timestamp but not
1678 * So in those cases, request a kvmclock update for all vcpus.
1679 * We need to rate-limit these requests though, as they can
1680 * considerably slow guests that have a large number of vcpus.
1681 * The time for a remote vcpu to update its kvmclock is bound
1682 * by the delay we use to rate-limit the updates.
1685 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1687 static void kvmclock_update_fn(struct work_struct *work)
1690 struct delayed_work *dwork = to_delayed_work(work);
1691 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1692 kvmclock_update_work);
1693 struct kvm *kvm = container_of(ka, struct kvm, arch);
1694 struct kvm_vcpu *vcpu;
1696 kvm_for_each_vcpu(i, vcpu, kvm) {
1697 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1698 kvm_vcpu_kick(vcpu);
1702 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1704 struct kvm *kvm = v->kvm;
1706 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1707 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1708 KVMCLOCK_UPDATE_DELAY);
1711 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1713 static void kvmclock_sync_fn(struct work_struct *work)
1715 struct delayed_work *dwork = to_delayed_work(work);
1716 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1717 kvmclock_sync_work);
1718 struct kvm *kvm = container_of(ka, struct kvm, arch);
1720 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1721 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1722 KVMCLOCK_SYNC_PERIOD);
1725 static bool msr_mtrr_valid(unsigned msr)
1728 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1729 case MSR_MTRRfix64K_00000:
1730 case MSR_MTRRfix16K_80000:
1731 case MSR_MTRRfix16K_A0000:
1732 case MSR_MTRRfix4K_C0000:
1733 case MSR_MTRRfix4K_C8000:
1734 case MSR_MTRRfix4K_D0000:
1735 case MSR_MTRRfix4K_D8000:
1736 case MSR_MTRRfix4K_E0000:
1737 case MSR_MTRRfix4K_E8000:
1738 case MSR_MTRRfix4K_F0000:
1739 case MSR_MTRRfix4K_F8000:
1740 case MSR_MTRRdefType:
1741 case MSR_IA32_CR_PAT:
1749 static bool valid_pat_type(unsigned t)
1751 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1754 static bool valid_mtrr_type(unsigned t)
1756 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1759 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1764 if (!msr_mtrr_valid(msr))
1767 if (msr == MSR_IA32_CR_PAT) {
1768 for (i = 0; i < 8; i++)
1769 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1772 } else if (msr == MSR_MTRRdefType) {
1775 return valid_mtrr_type(data & 0xff);
1776 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1777 for (i = 0; i < 8 ; i++)
1778 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1783 /* variable MTRRs */
1784 WARN_ON(!(msr >= 0x200 && msr < 0x200 + 2 * KVM_NR_VAR_MTRR));
1786 mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
1787 if ((msr & 1) == 0) {
1789 if (!valid_mtrr_type(data & 0xff))
1796 kvm_inject_gp(vcpu, 0);
1802 EXPORT_SYMBOL_GPL(kvm_mtrr_valid);
1804 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1806 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1808 if (!kvm_mtrr_valid(vcpu, msr, data))
1811 if (msr == MSR_MTRRdefType) {
1812 vcpu->arch.mtrr_state.def_type = data;
1813 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1814 } else if (msr == MSR_MTRRfix64K_00000)
1816 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1817 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1818 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1819 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1820 else if (msr == MSR_IA32_CR_PAT)
1821 vcpu->arch.pat = data;
1822 else { /* Variable MTRRs */
1823 int idx, is_mtrr_mask;
1826 idx = (msr - 0x200) / 2;
1827 is_mtrr_mask = msr - 0x200 - 2 * idx;
1830 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1833 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1837 kvm_mmu_reset_context(vcpu);
1841 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1843 u64 mcg_cap = vcpu->arch.mcg_cap;
1844 unsigned bank_num = mcg_cap & 0xff;
1847 case MSR_IA32_MCG_STATUS:
1848 vcpu->arch.mcg_status = data;
1850 case MSR_IA32_MCG_CTL:
1851 if (!(mcg_cap & MCG_CTL_P))
1853 if (data != 0 && data != ~(u64)0)
1855 vcpu->arch.mcg_ctl = data;
1858 if (msr >= MSR_IA32_MC0_CTL &&
1859 msr < MSR_IA32_MCx_CTL(bank_num)) {
1860 u32 offset = msr - MSR_IA32_MC0_CTL;
1861 /* only 0 or all 1s can be written to IA32_MCi_CTL
1862 * some Linux kernels though clear bit 10 in bank 4 to
1863 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1864 * this to avoid an uncatched #GP in the guest
1866 if ((offset & 0x3) == 0 &&
1867 data != 0 && (data | (1 << 10)) != ~(u64)0)
1869 vcpu->arch.mce_banks[offset] = data;
1877 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1879 struct kvm *kvm = vcpu->kvm;
1880 int lm = is_long_mode(vcpu);
1881 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1882 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1883 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1884 : kvm->arch.xen_hvm_config.blob_size_32;
1885 u32 page_num = data & ~PAGE_MASK;
1886 u64 page_addr = data & PAGE_MASK;
1891 if (page_num >= blob_size)
1894 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1899 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1908 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1910 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1913 static bool kvm_hv_msr_partition_wide(u32 msr)
1917 case HV_X64_MSR_GUEST_OS_ID:
1918 case HV_X64_MSR_HYPERCALL:
1919 case HV_X64_MSR_REFERENCE_TSC:
1920 case HV_X64_MSR_TIME_REF_COUNT:
1928 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1930 struct kvm *kvm = vcpu->kvm;
1933 case HV_X64_MSR_GUEST_OS_ID:
1934 kvm->arch.hv_guest_os_id = data;
1935 /* setting guest os id to zero disables hypercall page */
1936 if (!kvm->arch.hv_guest_os_id)
1937 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1939 case HV_X64_MSR_HYPERCALL: {
1944 /* if guest os id is not set hypercall should remain disabled */
1945 if (!kvm->arch.hv_guest_os_id)
1947 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1948 kvm->arch.hv_hypercall = data;
1951 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1952 addr = gfn_to_hva(kvm, gfn);
1953 if (kvm_is_error_hva(addr))
1955 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1956 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1957 if (__copy_to_user((void __user *)addr, instructions, 4))
1959 kvm->arch.hv_hypercall = data;
1960 mark_page_dirty(kvm, gfn);
1963 case HV_X64_MSR_REFERENCE_TSC: {
1965 HV_REFERENCE_TSC_PAGE tsc_ref;
1966 memset(&tsc_ref, 0, sizeof(tsc_ref));
1967 kvm->arch.hv_tsc_page = data;
1968 if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1970 gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1971 if (kvm_write_guest(kvm, gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
1972 &tsc_ref, sizeof(tsc_ref)))
1974 mark_page_dirty(kvm, gfn);
1978 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1979 "data 0x%llx\n", msr, data);
1985 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1988 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1992 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1993 vcpu->arch.hv_vapic = data;
1994 if (kvm_lapic_enable_pv_eoi(vcpu, 0))
1998 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
1999 addr = gfn_to_hva(vcpu->kvm, gfn);
2000 if (kvm_is_error_hva(addr))
2002 if (__clear_user((void __user *)addr, PAGE_SIZE))
2004 vcpu->arch.hv_vapic = data;
2005 mark_page_dirty(vcpu->kvm, gfn);
2006 if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
2010 case HV_X64_MSR_EOI:
2011 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
2012 case HV_X64_MSR_ICR:
2013 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
2014 case HV_X64_MSR_TPR:
2015 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
2017 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2018 "data 0x%llx\n", msr, data);
2025 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2027 gpa_t gpa = data & ~0x3f;
2029 /* Bits 2:5 are reserved, Should be zero */
2033 vcpu->arch.apf.msr_val = data;
2035 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2036 kvm_clear_async_pf_completion_queue(vcpu);
2037 kvm_async_pf_hash_reset(vcpu);
2041 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2045 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2046 kvm_async_pf_wakeup_all(vcpu);
2050 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2052 vcpu->arch.pv_time_enabled = false;
2055 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2059 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2062 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2063 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2064 vcpu->arch.st.accum_steal = delta;
2067 static void record_steal_time(struct kvm_vcpu *vcpu)
2069 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2072 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2073 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2076 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2077 vcpu->arch.st.steal.version += 2;
2078 vcpu->arch.st.accum_steal = 0;
2080 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2081 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2084 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2087 u32 msr = msr_info->index;
2088 u64 data = msr_info->data;
2091 case MSR_AMD64_NB_CFG:
2092 case MSR_IA32_UCODE_REV:
2093 case MSR_IA32_UCODE_WRITE:
2094 case MSR_VM_HSAVE_PA:
2095 case MSR_AMD64_PATCH_LOADER:
2096 case MSR_AMD64_BU_CFG2:
2100 return set_efer(vcpu, data);
2102 data &= ~(u64)0x40; /* ignore flush filter disable */
2103 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2104 data &= ~(u64)0x8; /* ignore TLB cache disable */
2105 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2107 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2112 case MSR_FAM10H_MMIO_CONF_BASE:
2114 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2119 case MSR_IA32_DEBUGCTLMSR:
2121 /* We support the non-activated case already */
2123 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2124 /* Values other than LBR and BTF are vendor-specific,
2125 thus reserved and should throw a #GP */
2128 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2131 case 0x200 ... 0x2ff:
2132 return set_msr_mtrr(vcpu, msr, data);
2133 case MSR_IA32_APICBASE:
2134 return kvm_set_apic_base(vcpu, msr_info);
2135 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2136 return kvm_x2apic_msr_write(vcpu, msr, data);
2137 case MSR_IA32_TSCDEADLINE:
2138 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2140 case MSR_IA32_TSC_ADJUST:
2141 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2142 if (!msr_info->host_initiated) {
2143 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2144 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2146 vcpu->arch.ia32_tsc_adjust_msr = data;
2149 case MSR_IA32_MISC_ENABLE:
2150 vcpu->arch.ia32_misc_enable_msr = data;
2152 case MSR_KVM_WALL_CLOCK_NEW:
2153 case MSR_KVM_WALL_CLOCK:
2154 vcpu->kvm->arch.wall_clock = data;
2155 kvm_write_wall_clock(vcpu->kvm, data);
2157 case MSR_KVM_SYSTEM_TIME_NEW:
2158 case MSR_KVM_SYSTEM_TIME: {
2160 kvmclock_reset(vcpu);
2162 vcpu->arch.time = data;
2163 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2165 /* we verify if the enable bit is set... */
2169 gpa_offset = data & ~(PAGE_MASK | 1);
2171 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2172 &vcpu->arch.pv_time, data & ~1ULL,
2173 sizeof(struct pvclock_vcpu_time_info)))
2174 vcpu->arch.pv_time_enabled = false;
2176 vcpu->arch.pv_time_enabled = true;
2180 case MSR_KVM_ASYNC_PF_EN:
2181 if (kvm_pv_enable_async_pf(vcpu, data))
2184 case MSR_KVM_STEAL_TIME:
2186 if (unlikely(!sched_info_on()))
2189 if (data & KVM_STEAL_RESERVED_MASK)
2192 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2193 data & KVM_STEAL_VALID_BITS,
2194 sizeof(struct kvm_steal_time)))
2197 vcpu->arch.st.msr_val = data;
2199 if (!(data & KVM_MSR_ENABLED))
2202 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2205 accumulate_steal_time(vcpu);
2208 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2211 case MSR_KVM_PV_EOI_EN:
2212 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2216 case MSR_IA32_MCG_CTL:
2217 case MSR_IA32_MCG_STATUS:
2218 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2219 return set_msr_mce(vcpu, msr, data);
2221 /* Performance counters are not protected by a CPUID bit,
2222 * so we should check all of them in the generic path for the sake of
2223 * cross vendor migration.
2224 * Writing a zero into the event select MSRs disables them,
2225 * which we perfectly emulate ;-). Any other value should be at least
2226 * reported, some guests depend on them.
2228 case MSR_K7_EVNTSEL0:
2229 case MSR_K7_EVNTSEL1:
2230 case MSR_K7_EVNTSEL2:
2231 case MSR_K7_EVNTSEL3:
2233 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2234 "0x%x data 0x%llx\n", msr, data);
2236 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2237 * so we ignore writes to make it happy.
2239 case MSR_K7_PERFCTR0:
2240 case MSR_K7_PERFCTR1:
2241 case MSR_K7_PERFCTR2:
2242 case MSR_K7_PERFCTR3:
2243 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2244 "0x%x data 0x%llx\n", msr, data);
2246 case MSR_P6_PERFCTR0:
2247 case MSR_P6_PERFCTR1:
2249 case MSR_P6_EVNTSEL0:
2250 case MSR_P6_EVNTSEL1:
2251 if (kvm_pmu_msr(vcpu, msr))
2252 return kvm_pmu_set_msr(vcpu, msr_info);
2254 if (pr || data != 0)
2255 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2256 "0x%x data 0x%llx\n", msr, data);
2258 case MSR_K7_CLK_CTL:
2260 * Ignore all writes to this no longer documented MSR.
2261 * Writes are only relevant for old K7 processors,
2262 * all pre-dating SVM, but a recommended workaround from
2263 * AMD for these chips. It is possible to specify the
2264 * affected processor models on the command line, hence
2265 * the need to ignore the workaround.
2268 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2269 if (kvm_hv_msr_partition_wide(msr)) {
2271 mutex_lock(&vcpu->kvm->lock);
2272 r = set_msr_hyperv_pw(vcpu, msr, data);
2273 mutex_unlock(&vcpu->kvm->lock);
2276 return set_msr_hyperv(vcpu, msr, data);
2278 case MSR_IA32_BBL_CR_CTL3:
2279 /* Drop writes to this legacy MSR -- see rdmsr
2280 * counterpart for further detail.
2282 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2284 case MSR_AMD64_OSVW_ID_LENGTH:
2285 if (!guest_cpuid_has_osvw(vcpu))
2287 vcpu->arch.osvw.length = data;
2289 case MSR_AMD64_OSVW_STATUS:
2290 if (!guest_cpuid_has_osvw(vcpu))
2292 vcpu->arch.osvw.status = data;
2295 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2296 return xen_hvm_config(vcpu, data);
2297 if (kvm_pmu_msr(vcpu, msr))
2298 return kvm_pmu_set_msr(vcpu, msr_info);
2300 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2304 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2311 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2315 * Reads an msr value (of 'msr_index') into 'pdata'.
2316 * Returns 0 on success, non-0 otherwise.
2317 * Assumes vcpu_load() was already called.
2319 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2321 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2324 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2326 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2328 if (!msr_mtrr_valid(msr))
2331 if (msr == MSR_MTRRdefType)
2332 *pdata = vcpu->arch.mtrr_state.def_type +
2333 (vcpu->arch.mtrr_state.enabled << 10);
2334 else if (msr == MSR_MTRRfix64K_00000)
2336 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2337 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2338 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2339 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2340 else if (msr == MSR_IA32_CR_PAT)
2341 *pdata = vcpu->arch.pat;
2342 else { /* Variable MTRRs */
2343 int idx, is_mtrr_mask;
2346 idx = (msr - 0x200) / 2;
2347 is_mtrr_mask = msr - 0x200 - 2 * idx;
2350 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2353 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2360 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2363 u64 mcg_cap = vcpu->arch.mcg_cap;
2364 unsigned bank_num = mcg_cap & 0xff;
2367 case MSR_IA32_P5_MC_ADDR:
2368 case MSR_IA32_P5_MC_TYPE:
2371 case MSR_IA32_MCG_CAP:
2372 data = vcpu->arch.mcg_cap;
2374 case MSR_IA32_MCG_CTL:
2375 if (!(mcg_cap & MCG_CTL_P))
2377 data = vcpu->arch.mcg_ctl;
2379 case MSR_IA32_MCG_STATUS:
2380 data = vcpu->arch.mcg_status;
2383 if (msr >= MSR_IA32_MC0_CTL &&
2384 msr < MSR_IA32_MCx_CTL(bank_num)) {
2385 u32 offset = msr - MSR_IA32_MC0_CTL;
2386 data = vcpu->arch.mce_banks[offset];
2395 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2398 struct kvm *kvm = vcpu->kvm;
2401 case HV_X64_MSR_GUEST_OS_ID:
2402 data = kvm->arch.hv_guest_os_id;
2404 case HV_X64_MSR_HYPERCALL:
2405 data = kvm->arch.hv_hypercall;
2407 case HV_X64_MSR_TIME_REF_COUNT: {
2409 div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2412 case HV_X64_MSR_REFERENCE_TSC:
2413 data = kvm->arch.hv_tsc_page;
2416 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2424 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2429 case HV_X64_MSR_VP_INDEX: {
2432 kvm_for_each_vcpu(r, v, vcpu->kvm) {
2440 case HV_X64_MSR_EOI:
2441 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2442 case HV_X64_MSR_ICR:
2443 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2444 case HV_X64_MSR_TPR:
2445 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2446 case HV_X64_MSR_APIC_ASSIST_PAGE:
2447 data = vcpu->arch.hv_vapic;
2450 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2457 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2462 case MSR_IA32_PLATFORM_ID:
2463 case MSR_IA32_EBL_CR_POWERON:
2464 case MSR_IA32_DEBUGCTLMSR:
2465 case MSR_IA32_LASTBRANCHFROMIP:
2466 case MSR_IA32_LASTBRANCHTOIP:
2467 case MSR_IA32_LASTINTFROMIP:
2468 case MSR_IA32_LASTINTTOIP:
2471 case MSR_VM_HSAVE_PA:
2472 case MSR_K7_EVNTSEL0:
2473 case MSR_K7_EVNTSEL1:
2474 case MSR_K7_EVNTSEL2:
2475 case MSR_K7_EVNTSEL3:
2476 case MSR_K7_PERFCTR0:
2477 case MSR_K7_PERFCTR1:
2478 case MSR_K7_PERFCTR2:
2479 case MSR_K7_PERFCTR3:
2480 case MSR_K8_INT_PENDING_MSG:
2481 case MSR_AMD64_NB_CFG:
2482 case MSR_FAM10H_MMIO_CONF_BASE:
2483 case MSR_AMD64_BU_CFG2:
2486 case MSR_P6_PERFCTR0:
2487 case MSR_P6_PERFCTR1:
2488 case MSR_P6_EVNTSEL0:
2489 case MSR_P6_EVNTSEL1:
2490 if (kvm_pmu_msr(vcpu, msr))
2491 return kvm_pmu_get_msr(vcpu, msr, pdata);
2494 case MSR_IA32_UCODE_REV:
2495 data = 0x100000000ULL;
2498 data = 0x500 | KVM_NR_VAR_MTRR;
2500 case 0x200 ... 0x2ff:
2501 return get_msr_mtrr(vcpu, msr, pdata);
2502 case 0xcd: /* fsb frequency */
2506 * MSR_EBC_FREQUENCY_ID
2507 * Conservative value valid for even the basic CPU models.
2508 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2509 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2510 * and 266MHz for model 3, or 4. Set Core Clock
2511 * Frequency to System Bus Frequency Ratio to 1 (bits
2512 * 31:24) even though these are only valid for CPU
2513 * models > 2, however guests may end up dividing or
2514 * multiplying by zero otherwise.
2516 case MSR_EBC_FREQUENCY_ID:
2519 case MSR_IA32_APICBASE:
2520 data = kvm_get_apic_base(vcpu);
2522 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2523 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2525 case MSR_IA32_TSCDEADLINE:
2526 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2528 case MSR_IA32_TSC_ADJUST:
2529 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2531 case MSR_IA32_MISC_ENABLE:
2532 data = vcpu->arch.ia32_misc_enable_msr;
2534 case MSR_IA32_PERF_STATUS:
2535 /* TSC increment by tick */
2537 /* CPU multiplier */
2538 data |= (((uint64_t)4ULL) << 40);
2541 data = vcpu->arch.efer;
2543 case MSR_KVM_WALL_CLOCK:
2544 case MSR_KVM_WALL_CLOCK_NEW:
2545 data = vcpu->kvm->arch.wall_clock;
2547 case MSR_KVM_SYSTEM_TIME:
2548 case MSR_KVM_SYSTEM_TIME_NEW:
2549 data = vcpu->arch.time;
2551 case MSR_KVM_ASYNC_PF_EN:
2552 data = vcpu->arch.apf.msr_val;
2554 case MSR_KVM_STEAL_TIME:
2555 data = vcpu->arch.st.msr_val;
2557 case MSR_KVM_PV_EOI_EN:
2558 data = vcpu->arch.pv_eoi.msr_val;
2560 case MSR_IA32_P5_MC_ADDR:
2561 case MSR_IA32_P5_MC_TYPE:
2562 case MSR_IA32_MCG_CAP:
2563 case MSR_IA32_MCG_CTL:
2564 case MSR_IA32_MCG_STATUS:
2565 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2566 return get_msr_mce(vcpu, msr, pdata);
2567 case MSR_K7_CLK_CTL:
2569 * Provide expected ramp-up count for K7. All other
2570 * are set to zero, indicating minimum divisors for
2573 * This prevents guest kernels on AMD host with CPU
2574 * type 6, model 8 and higher from exploding due to
2575 * the rdmsr failing.
2579 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2580 if (kvm_hv_msr_partition_wide(msr)) {
2582 mutex_lock(&vcpu->kvm->lock);
2583 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2584 mutex_unlock(&vcpu->kvm->lock);
2587 return get_msr_hyperv(vcpu, msr, pdata);
2589 case MSR_IA32_BBL_CR_CTL3:
2590 /* This legacy MSR exists but isn't fully documented in current
2591 * silicon. It is however accessed by winxp in very narrow
2592 * scenarios where it sets bit #19, itself documented as
2593 * a "reserved" bit. Best effort attempt to source coherent
2594 * read data here should the balance of the register be
2595 * interpreted by the guest:
2597 * L2 cache control register 3: 64GB range, 256KB size,
2598 * enabled, latency 0x1, configured
2602 case MSR_AMD64_OSVW_ID_LENGTH:
2603 if (!guest_cpuid_has_osvw(vcpu))
2605 data = vcpu->arch.osvw.length;
2607 case MSR_AMD64_OSVW_STATUS:
2608 if (!guest_cpuid_has_osvw(vcpu))
2610 data = vcpu->arch.osvw.status;
2613 if (kvm_pmu_msr(vcpu, msr))
2614 return kvm_pmu_get_msr(vcpu, msr, pdata);
2616 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2619 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2627 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2630 * Read or write a bunch of msrs. All parameters are kernel addresses.
2632 * @return number of msrs set successfully.
2634 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2635 struct kvm_msr_entry *entries,
2636 int (*do_msr)(struct kvm_vcpu *vcpu,
2637 unsigned index, u64 *data))
2641 idx = srcu_read_lock(&vcpu->kvm->srcu);
2642 for (i = 0; i < msrs->nmsrs; ++i)
2643 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2645 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2651 * Read or write a bunch of msrs. Parameters are user addresses.
2653 * @return number of msrs set successfully.
2655 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2656 int (*do_msr)(struct kvm_vcpu *vcpu,
2657 unsigned index, u64 *data),
2660 struct kvm_msrs msrs;
2661 struct kvm_msr_entry *entries;
2666 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2670 if (msrs.nmsrs >= MAX_IO_MSRS)
2673 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2674 entries = memdup_user(user_msrs->entries, size);
2675 if (IS_ERR(entries)) {
2676 r = PTR_ERR(entries);
2680 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2685 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2696 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2701 case KVM_CAP_IRQCHIP:
2703 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2704 case KVM_CAP_SET_TSS_ADDR:
2705 case KVM_CAP_EXT_CPUID:
2706 case KVM_CAP_EXT_EMUL_CPUID:
2707 case KVM_CAP_CLOCKSOURCE:
2709 case KVM_CAP_NOP_IO_DELAY:
2710 case KVM_CAP_MP_STATE:
2711 case KVM_CAP_SYNC_MMU:
2712 case KVM_CAP_USER_NMI:
2713 case KVM_CAP_REINJECT_CONTROL:
2714 case KVM_CAP_IRQ_INJECT_STATUS:
2716 case KVM_CAP_IOEVENTFD:
2717 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2719 case KVM_CAP_PIT_STATE2:
2720 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2721 case KVM_CAP_XEN_HVM:
2722 case KVM_CAP_ADJUST_CLOCK:
2723 case KVM_CAP_VCPU_EVENTS:
2724 case KVM_CAP_HYPERV:
2725 case KVM_CAP_HYPERV_VAPIC:
2726 case KVM_CAP_HYPERV_SPIN:
2727 case KVM_CAP_PCI_SEGMENT:
2728 case KVM_CAP_DEBUGREGS:
2729 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2731 case KVM_CAP_ASYNC_PF:
2732 case KVM_CAP_GET_TSC_KHZ:
2733 case KVM_CAP_KVMCLOCK_CTRL:
2734 case KVM_CAP_READONLY_MEM:
2735 case KVM_CAP_HYPERV_TIME:
2736 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2737 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2738 case KVM_CAP_ASSIGN_DEV_IRQ:
2739 case KVM_CAP_PCI_2_3:
2743 case KVM_CAP_COALESCED_MMIO:
2744 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2747 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2749 case KVM_CAP_NR_VCPUS:
2750 r = KVM_SOFT_MAX_VCPUS;
2752 case KVM_CAP_MAX_VCPUS:
2755 case KVM_CAP_NR_MEMSLOTS:
2756 r = KVM_USER_MEM_SLOTS;
2758 case KVM_CAP_PV_MMU: /* obsolete */
2761 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2763 r = iommu_present(&pci_bus_type);
2767 r = KVM_MAX_MCE_BANKS;
2772 case KVM_CAP_TSC_CONTROL:
2773 r = kvm_has_tsc_control;
2775 case KVM_CAP_TSC_DEADLINE_TIMER:
2776 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2786 long kvm_arch_dev_ioctl(struct file *filp,
2787 unsigned int ioctl, unsigned long arg)
2789 void __user *argp = (void __user *)arg;
2793 case KVM_GET_MSR_INDEX_LIST: {
2794 struct kvm_msr_list __user *user_msr_list = argp;
2795 struct kvm_msr_list msr_list;
2799 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2802 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2803 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2806 if (n < msr_list.nmsrs)
2809 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2810 num_msrs_to_save * sizeof(u32)))
2812 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2814 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2819 case KVM_GET_SUPPORTED_CPUID:
2820 case KVM_GET_EMULATED_CPUID: {
2821 struct kvm_cpuid2 __user *cpuid_arg = argp;
2822 struct kvm_cpuid2 cpuid;
2825 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2828 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2834 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2839 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2842 mce_cap = KVM_MCE_CAP_SUPPORTED;
2844 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2856 static void wbinvd_ipi(void *garbage)
2861 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2863 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2866 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2868 /* Address WBINVD may be executed by guest */
2869 if (need_emulate_wbinvd(vcpu)) {
2870 if (kvm_x86_ops->has_wbinvd_exit())
2871 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2872 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2873 smp_call_function_single(vcpu->cpu,
2874 wbinvd_ipi, NULL, 1);
2877 kvm_x86_ops->vcpu_load(vcpu, cpu);
2879 /* Apply any externally detected TSC adjustments (due to suspend) */
2880 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2881 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2882 vcpu->arch.tsc_offset_adjustment = 0;
2883 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2886 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2887 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2888 native_read_tsc() - vcpu->arch.last_host_tsc;
2890 mark_tsc_unstable("KVM discovered backwards TSC");
2891 if (check_tsc_unstable()) {
2892 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2893 vcpu->arch.last_guest_tsc);
2894 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2895 vcpu->arch.tsc_catchup = 1;
2898 * On a host with synchronized TSC, there is no need to update
2899 * kvmclock on vcpu->cpu migration
2901 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2902 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2903 if (vcpu->cpu != cpu)
2904 kvm_migrate_timers(vcpu);
2908 accumulate_steal_time(vcpu);
2909 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2912 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2914 kvm_x86_ops->vcpu_put(vcpu);
2915 kvm_put_guest_fpu(vcpu);
2916 vcpu->arch.last_host_tsc = native_read_tsc();
2919 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2920 struct kvm_lapic_state *s)
2922 kvm_x86_ops->sync_pir_to_irr(vcpu);
2923 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2928 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2929 struct kvm_lapic_state *s)
2931 kvm_apic_post_state_restore(vcpu, s);
2932 update_cr8_intercept(vcpu);
2937 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2938 struct kvm_interrupt *irq)
2940 if (irq->irq >= KVM_NR_INTERRUPTS)
2942 if (irqchip_in_kernel(vcpu->kvm))
2945 kvm_queue_interrupt(vcpu, irq->irq, false);
2946 kvm_make_request(KVM_REQ_EVENT, vcpu);
2951 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2953 kvm_inject_nmi(vcpu);
2958 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2959 struct kvm_tpr_access_ctl *tac)
2963 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2967 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2971 unsigned bank_num = mcg_cap & 0xff, bank;
2974 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2976 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2979 vcpu->arch.mcg_cap = mcg_cap;
2980 /* Init IA32_MCG_CTL to all 1s */
2981 if (mcg_cap & MCG_CTL_P)
2982 vcpu->arch.mcg_ctl = ~(u64)0;
2983 /* Init IA32_MCi_CTL to all 1s */
2984 for (bank = 0; bank < bank_num; bank++)
2985 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2990 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2991 struct kvm_x86_mce *mce)
2993 u64 mcg_cap = vcpu->arch.mcg_cap;
2994 unsigned bank_num = mcg_cap & 0xff;
2995 u64 *banks = vcpu->arch.mce_banks;
2997 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3000 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3001 * reporting is disabled
3003 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3004 vcpu->arch.mcg_ctl != ~(u64)0)
3006 banks += 4 * mce->bank;
3008 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3009 * reporting is disabled for the bank
3011 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3013 if (mce->status & MCI_STATUS_UC) {
3014 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3015 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3016 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3019 if (banks[1] & MCI_STATUS_VAL)
3020 mce->status |= MCI_STATUS_OVER;
3021 banks[2] = mce->addr;
3022 banks[3] = mce->misc;
3023 vcpu->arch.mcg_status = mce->mcg_status;
3024 banks[1] = mce->status;
3025 kvm_queue_exception(vcpu, MC_VECTOR);
3026 } else if (!(banks[1] & MCI_STATUS_VAL)
3027 || !(banks[1] & MCI_STATUS_UC)) {
3028 if (banks[1] & MCI_STATUS_VAL)
3029 mce->status |= MCI_STATUS_OVER;
3030 banks[2] = mce->addr;
3031 banks[3] = mce->misc;
3032 banks[1] = mce->status;
3034 banks[1] |= MCI_STATUS_OVER;
3038 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3039 struct kvm_vcpu_events *events)
3042 events->exception.injected =
3043 vcpu->arch.exception.pending &&
3044 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3045 events->exception.nr = vcpu->arch.exception.nr;
3046 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3047 events->exception.pad = 0;
3048 events->exception.error_code = vcpu->arch.exception.error_code;
3050 events->interrupt.injected =
3051 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3052 events->interrupt.nr = vcpu->arch.interrupt.nr;
3053 events->interrupt.soft = 0;
3054 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3056 events->nmi.injected = vcpu->arch.nmi_injected;
3057 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3058 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3059 events->nmi.pad = 0;
3061 events->sipi_vector = 0; /* never valid when reporting to user space */
3063 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3064 | KVM_VCPUEVENT_VALID_SHADOW);
3065 memset(&events->reserved, 0, sizeof(events->reserved));
3068 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3069 struct kvm_vcpu_events *events)
3071 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3072 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3073 | KVM_VCPUEVENT_VALID_SHADOW))
3077 vcpu->arch.exception.pending = events->exception.injected;
3078 vcpu->arch.exception.nr = events->exception.nr;
3079 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3080 vcpu->arch.exception.error_code = events->exception.error_code;
3082 vcpu->arch.interrupt.pending = events->interrupt.injected;
3083 vcpu->arch.interrupt.nr = events->interrupt.nr;
3084 vcpu->arch.interrupt.soft = events->interrupt.soft;
3085 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3086 kvm_x86_ops->set_interrupt_shadow(vcpu,
3087 events->interrupt.shadow);
3089 vcpu->arch.nmi_injected = events->nmi.injected;
3090 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3091 vcpu->arch.nmi_pending = events->nmi.pending;
3092 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3094 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3095 kvm_vcpu_has_lapic(vcpu))
3096 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3098 kvm_make_request(KVM_REQ_EVENT, vcpu);
3103 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3104 struct kvm_debugregs *dbgregs)
3108 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3109 _kvm_get_dr(vcpu, 6, &val);
3111 dbgregs->dr7 = vcpu->arch.dr7;
3113 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3116 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3117 struct kvm_debugregs *dbgregs)
3122 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3123 vcpu->arch.dr6 = dbgregs->dr6;
3124 kvm_update_dr6(vcpu);
3125 vcpu->arch.dr7 = dbgregs->dr7;
3126 kvm_update_dr7(vcpu);
3131 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3132 struct kvm_xsave *guest_xsave)
3134 if (cpu_has_xsave) {
3135 memcpy(guest_xsave->region,
3136 &vcpu->arch.guest_fpu.state->xsave,
3137 vcpu->arch.guest_xstate_size);
3138 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] &=
3139 vcpu->arch.guest_supported_xcr0 | XSTATE_FPSSE;
3141 memcpy(guest_xsave->region,
3142 &vcpu->arch.guest_fpu.state->fxsave,
3143 sizeof(struct i387_fxsave_struct));
3144 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3149 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3150 struct kvm_xsave *guest_xsave)
3153 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3155 if (cpu_has_xsave) {
3157 * Here we allow setting states that are not present in
3158 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3159 * with old userspace.
3161 if (xstate_bv & ~kvm_supported_xcr0())
3163 memcpy(&vcpu->arch.guest_fpu.state->xsave,
3164 guest_xsave->region, vcpu->arch.guest_xstate_size);
3166 if (xstate_bv & ~XSTATE_FPSSE)
3168 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3169 guest_xsave->region, sizeof(struct i387_fxsave_struct));
3174 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3175 struct kvm_xcrs *guest_xcrs)
3177 if (!cpu_has_xsave) {
3178 guest_xcrs->nr_xcrs = 0;
3182 guest_xcrs->nr_xcrs = 1;
3183 guest_xcrs->flags = 0;
3184 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3185 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3188 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3189 struct kvm_xcrs *guest_xcrs)
3196 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3199 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3200 /* Only support XCR0 currently */
3201 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3202 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3203 guest_xcrs->xcrs[i].value);
3212 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3213 * stopped by the hypervisor. This function will be called from the host only.
3214 * EINVAL is returned when the host attempts to set the flag for a guest that
3215 * does not support pv clocks.
3217 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3219 if (!vcpu->arch.pv_time_enabled)
3221 vcpu->arch.pvclock_set_guest_stopped_request = true;
3222 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3226 long kvm_arch_vcpu_ioctl(struct file *filp,
3227 unsigned int ioctl, unsigned long arg)
3229 struct kvm_vcpu *vcpu = filp->private_data;
3230 void __user *argp = (void __user *)arg;
3233 struct kvm_lapic_state *lapic;
3234 struct kvm_xsave *xsave;
3235 struct kvm_xcrs *xcrs;
3241 case KVM_GET_LAPIC: {
3243 if (!vcpu->arch.apic)
3245 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3250 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3254 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3259 case KVM_SET_LAPIC: {
3261 if (!vcpu->arch.apic)
3263 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3264 if (IS_ERR(u.lapic))
3265 return PTR_ERR(u.lapic);
3267 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3270 case KVM_INTERRUPT: {
3271 struct kvm_interrupt irq;
3274 if (copy_from_user(&irq, argp, sizeof irq))
3276 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3280 r = kvm_vcpu_ioctl_nmi(vcpu);
3283 case KVM_SET_CPUID: {
3284 struct kvm_cpuid __user *cpuid_arg = argp;
3285 struct kvm_cpuid cpuid;
3288 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3290 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3293 case KVM_SET_CPUID2: {
3294 struct kvm_cpuid2 __user *cpuid_arg = argp;
3295 struct kvm_cpuid2 cpuid;
3298 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3300 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3301 cpuid_arg->entries);
3304 case KVM_GET_CPUID2: {
3305 struct kvm_cpuid2 __user *cpuid_arg = argp;
3306 struct kvm_cpuid2 cpuid;
3309 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3311 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3312 cpuid_arg->entries);
3316 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3322 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3325 r = msr_io(vcpu, argp, do_set_msr, 0);
3327 case KVM_TPR_ACCESS_REPORTING: {
3328 struct kvm_tpr_access_ctl tac;
3331 if (copy_from_user(&tac, argp, sizeof tac))
3333 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3337 if (copy_to_user(argp, &tac, sizeof tac))
3342 case KVM_SET_VAPIC_ADDR: {
3343 struct kvm_vapic_addr va;
3346 if (!irqchip_in_kernel(vcpu->kvm))
3349 if (copy_from_user(&va, argp, sizeof va))
3351 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3354 case KVM_X86_SETUP_MCE: {
3358 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3360 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3363 case KVM_X86_SET_MCE: {
3364 struct kvm_x86_mce mce;
3367 if (copy_from_user(&mce, argp, sizeof mce))
3369 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3372 case KVM_GET_VCPU_EVENTS: {
3373 struct kvm_vcpu_events events;
3375 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3378 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3383 case KVM_SET_VCPU_EVENTS: {
3384 struct kvm_vcpu_events events;
3387 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3390 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3393 case KVM_GET_DEBUGREGS: {
3394 struct kvm_debugregs dbgregs;
3396 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3399 if (copy_to_user(argp, &dbgregs,
3400 sizeof(struct kvm_debugregs)))
3405 case KVM_SET_DEBUGREGS: {
3406 struct kvm_debugregs dbgregs;
3409 if (copy_from_user(&dbgregs, argp,
3410 sizeof(struct kvm_debugregs)))
3413 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3416 case KVM_GET_XSAVE: {
3417 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3422 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3425 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3430 case KVM_SET_XSAVE: {
3431 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3432 if (IS_ERR(u.xsave))
3433 return PTR_ERR(u.xsave);
3435 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3438 case KVM_GET_XCRS: {
3439 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3444 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3447 if (copy_to_user(argp, u.xcrs,
3448 sizeof(struct kvm_xcrs)))
3453 case KVM_SET_XCRS: {
3454 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3456 return PTR_ERR(u.xcrs);
3458 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3461 case KVM_SET_TSC_KHZ: {
3465 user_tsc_khz = (u32)arg;
3467 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3470 if (user_tsc_khz == 0)
3471 user_tsc_khz = tsc_khz;
3473 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3478 case KVM_GET_TSC_KHZ: {
3479 r = vcpu->arch.virtual_tsc_khz;
3482 case KVM_KVMCLOCK_CTRL: {
3483 r = kvm_set_guest_paused(vcpu);
3494 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3496 return VM_FAULT_SIGBUS;
3499 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3503 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3505 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3509 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3512 kvm->arch.ept_identity_map_addr = ident_addr;
3516 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3517 u32 kvm_nr_mmu_pages)
3519 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3522 mutex_lock(&kvm->slots_lock);
3524 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3525 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3527 mutex_unlock(&kvm->slots_lock);
3531 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3533 return kvm->arch.n_max_mmu_pages;
3536 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3541 switch (chip->chip_id) {
3542 case KVM_IRQCHIP_PIC_MASTER:
3543 memcpy(&chip->chip.pic,
3544 &pic_irqchip(kvm)->pics[0],
3545 sizeof(struct kvm_pic_state));
3547 case KVM_IRQCHIP_PIC_SLAVE:
3548 memcpy(&chip->chip.pic,
3549 &pic_irqchip(kvm)->pics[1],
3550 sizeof(struct kvm_pic_state));
3552 case KVM_IRQCHIP_IOAPIC:
3553 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3562 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3567 switch (chip->chip_id) {
3568 case KVM_IRQCHIP_PIC_MASTER:
3569 spin_lock(&pic_irqchip(kvm)->lock);
3570 memcpy(&pic_irqchip(kvm)->pics[0],
3572 sizeof(struct kvm_pic_state));
3573 spin_unlock(&pic_irqchip(kvm)->lock);
3575 case KVM_IRQCHIP_PIC_SLAVE:
3576 spin_lock(&pic_irqchip(kvm)->lock);
3577 memcpy(&pic_irqchip(kvm)->pics[1],
3579 sizeof(struct kvm_pic_state));
3580 spin_unlock(&pic_irqchip(kvm)->lock);
3582 case KVM_IRQCHIP_IOAPIC:
3583 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3589 kvm_pic_update_irq(pic_irqchip(kvm));
3593 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3597 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3598 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3599 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3603 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3607 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3608 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3609 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3610 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3614 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3618 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3619 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3620 sizeof(ps->channels));
3621 ps->flags = kvm->arch.vpit->pit_state.flags;
3622 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3623 memset(&ps->reserved, 0, sizeof(ps->reserved));
3627 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3629 int r = 0, start = 0;
3630 u32 prev_legacy, cur_legacy;
3631 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3632 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3633 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3634 if (!prev_legacy && cur_legacy)
3636 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3637 sizeof(kvm->arch.vpit->pit_state.channels));
3638 kvm->arch.vpit->pit_state.flags = ps->flags;
3639 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3640 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3644 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3645 struct kvm_reinject_control *control)
3647 if (!kvm->arch.vpit)
3649 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3650 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3651 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3656 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3657 * @kvm: kvm instance
3658 * @log: slot id and address to which we copy the log
3660 * We need to keep it in mind that VCPU threads can write to the bitmap
3661 * concurrently. So, to avoid losing data, we keep the following order for
3664 * 1. Take a snapshot of the bit and clear it if needed.
3665 * 2. Write protect the corresponding page.
3666 * 3. Flush TLB's if needed.
3667 * 4. Copy the snapshot to the userspace.
3669 * Between 2 and 3, the guest may write to the page using the remaining TLB
3670 * entry. This is not a problem because the page will be reported dirty at
3671 * step 4 using the snapshot taken before and step 3 ensures that successive
3672 * writes will be logged for the next call.
3674 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3677 struct kvm_memory_slot *memslot;
3679 unsigned long *dirty_bitmap;
3680 unsigned long *dirty_bitmap_buffer;
3681 bool is_dirty = false;
3683 mutex_lock(&kvm->slots_lock);
3686 if (log->slot >= KVM_USER_MEM_SLOTS)
3689 memslot = id_to_memslot(kvm->memslots, log->slot);
3691 dirty_bitmap = memslot->dirty_bitmap;
3696 n = kvm_dirty_bitmap_bytes(memslot);
3698 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3699 memset(dirty_bitmap_buffer, 0, n);
3701 spin_lock(&kvm->mmu_lock);
3703 for (i = 0; i < n / sizeof(long); i++) {
3707 if (!dirty_bitmap[i])
3712 mask = xchg(&dirty_bitmap[i], 0);
3713 dirty_bitmap_buffer[i] = mask;
3715 offset = i * BITS_PER_LONG;
3716 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3719 spin_unlock(&kvm->mmu_lock);
3721 /* See the comments in kvm_mmu_slot_remove_write_access(). */
3722 lockdep_assert_held(&kvm->slots_lock);
3725 * All the TLBs can be flushed out of mmu lock, see the comments in
3726 * kvm_mmu_slot_remove_write_access().
3729 kvm_flush_remote_tlbs(kvm);
3732 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3737 mutex_unlock(&kvm->slots_lock);
3741 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3744 if (!irqchip_in_kernel(kvm))
3747 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3748 irq_event->irq, irq_event->level,
3753 long kvm_arch_vm_ioctl(struct file *filp,
3754 unsigned int ioctl, unsigned long arg)
3756 struct kvm *kvm = filp->private_data;
3757 void __user *argp = (void __user *)arg;
3760 * This union makes it completely explicit to gcc-3.x
3761 * that these two variables' stack usage should be
3762 * combined, not added together.
3765 struct kvm_pit_state ps;
3766 struct kvm_pit_state2 ps2;
3767 struct kvm_pit_config pit_config;
3771 case KVM_SET_TSS_ADDR:
3772 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3774 case KVM_SET_IDENTITY_MAP_ADDR: {
3778 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3780 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3783 case KVM_SET_NR_MMU_PAGES:
3784 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3786 case KVM_GET_NR_MMU_PAGES:
3787 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3789 case KVM_CREATE_IRQCHIP: {
3790 struct kvm_pic *vpic;
3792 mutex_lock(&kvm->lock);
3795 goto create_irqchip_unlock;
3797 if (atomic_read(&kvm->online_vcpus))
3798 goto create_irqchip_unlock;
3800 vpic = kvm_create_pic(kvm);
3802 r = kvm_ioapic_init(kvm);
3804 mutex_lock(&kvm->slots_lock);
3805 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3807 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3809 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3811 mutex_unlock(&kvm->slots_lock);
3813 goto create_irqchip_unlock;
3816 goto create_irqchip_unlock;
3818 kvm->arch.vpic = vpic;
3820 r = kvm_setup_default_irq_routing(kvm);
3822 mutex_lock(&kvm->slots_lock);
3823 mutex_lock(&kvm->irq_lock);
3824 kvm_ioapic_destroy(kvm);
3825 kvm_destroy_pic(kvm);
3826 mutex_unlock(&kvm->irq_lock);
3827 mutex_unlock(&kvm->slots_lock);
3829 create_irqchip_unlock:
3830 mutex_unlock(&kvm->lock);
3833 case KVM_CREATE_PIT:
3834 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3836 case KVM_CREATE_PIT2:
3838 if (copy_from_user(&u.pit_config, argp,
3839 sizeof(struct kvm_pit_config)))
3842 mutex_lock(&kvm->slots_lock);
3845 goto create_pit_unlock;
3847 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3851 mutex_unlock(&kvm->slots_lock);
3853 case KVM_GET_IRQCHIP: {
3854 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3855 struct kvm_irqchip *chip;
3857 chip = memdup_user(argp, sizeof(*chip));
3864 if (!irqchip_in_kernel(kvm))
3865 goto get_irqchip_out;
3866 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3868 goto get_irqchip_out;
3870 if (copy_to_user(argp, chip, sizeof *chip))
3871 goto get_irqchip_out;
3877 case KVM_SET_IRQCHIP: {
3878 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3879 struct kvm_irqchip *chip;
3881 chip = memdup_user(argp, sizeof(*chip));
3888 if (!irqchip_in_kernel(kvm))
3889 goto set_irqchip_out;
3890 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3892 goto set_irqchip_out;
3900 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3903 if (!kvm->arch.vpit)
3905 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3909 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3916 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3919 if (!kvm->arch.vpit)
3921 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3924 case KVM_GET_PIT2: {
3926 if (!kvm->arch.vpit)
3928 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3932 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3937 case KVM_SET_PIT2: {
3939 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3942 if (!kvm->arch.vpit)
3944 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3947 case KVM_REINJECT_CONTROL: {
3948 struct kvm_reinject_control control;
3950 if (copy_from_user(&control, argp, sizeof(control)))
3952 r = kvm_vm_ioctl_reinject(kvm, &control);
3955 case KVM_XEN_HVM_CONFIG: {
3957 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3958 sizeof(struct kvm_xen_hvm_config)))
3961 if (kvm->arch.xen_hvm_config.flags)
3966 case KVM_SET_CLOCK: {
3967 struct kvm_clock_data user_ns;
3972 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3980 local_irq_disable();
3981 now_ns = get_kernel_ns();
3982 delta = user_ns.clock - now_ns;
3984 kvm->arch.kvmclock_offset = delta;
3985 kvm_gen_update_masterclock(kvm);
3988 case KVM_GET_CLOCK: {
3989 struct kvm_clock_data user_ns;
3992 local_irq_disable();
3993 now_ns = get_kernel_ns();
3994 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3997 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4000 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4013 static void kvm_init_msr_list(void)
4018 /* skip the first msrs in the list. KVM-specific */
4019 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
4020 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4024 * Even MSRs that are valid in the host may not be exposed
4025 * to the guests in some cases. We could work around this
4026 * in VMX with the generic MSR save/load machinery, but it
4027 * is not really worthwhile since it will really only
4028 * happen with nested virtualization.
4030 switch (msrs_to_save[i]) {
4031 case MSR_IA32_BNDCFGS:
4032 if (!kvm_x86_ops->mpx_supported())
4040 msrs_to_save[j] = msrs_to_save[i];
4043 num_msrs_to_save = j;
4046 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4054 if (!(vcpu->arch.apic &&
4055 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
4056 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
4067 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4074 if (!(vcpu->arch.apic &&
4075 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
4076 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
4078 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4088 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4089 struct kvm_segment *var, int seg)
4091 kvm_x86_ops->set_segment(vcpu, var, seg);
4094 void kvm_get_segment(struct kvm_vcpu *vcpu,
4095 struct kvm_segment *var, int seg)
4097 kvm_x86_ops->get_segment(vcpu, var, seg);
4100 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4101 struct x86_exception *exception)
4105 BUG_ON(!mmu_is_nested(vcpu));
4107 /* NPT walks are always user-walks */
4108 access |= PFERR_USER_MASK;
4109 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4114 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4115 struct x86_exception *exception)
4117 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4118 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4121 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4122 struct x86_exception *exception)
4124 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4125 access |= PFERR_FETCH_MASK;
4126 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4129 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4130 struct x86_exception *exception)
4132 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4133 access |= PFERR_WRITE_MASK;
4134 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4137 /* uses this to access any guest's mapped memory without checking CPL */
4138 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4139 struct x86_exception *exception)
4141 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4144 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4145 struct kvm_vcpu *vcpu, u32 access,
4146 struct x86_exception *exception)
4149 int r = X86EMUL_CONTINUE;
4152 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4154 unsigned offset = addr & (PAGE_SIZE-1);
4155 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4158 if (gpa == UNMAPPED_GVA)
4159 return X86EMUL_PROPAGATE_FAULT;
4160 ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, data,
4163 r = X86EMUL_IO_NEEDED;
4175 /* used for instruction fetching */
4176 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4177 gva_t addr, void *val, unsigned int bytes,
4178 struct x86_exception *exception)
4180 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4181 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4185 /* Inline kvm_read_guest_virt_helper for speed. */
4186 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4188 if (unlikely(gpa == UNMAPPED_GVA))
4189 return X86EMUL_PROPAGATE_FAULT;
4191 offset = addr & (PAGE_SIZE-1);
4192 if (WARN_ON(offset + bytes > PAGE_SIZE))
4193 bytes = (unsigned)PAGE_SIZE - offset;
4194 ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, val,
4196 if (unlikely(ret < 0))
4197 return X86EMUL_IO_NEEDED;
4199 return X86EMUL_CONTINUE;
4202 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4203 gva_t addr, void *val, unsigned int bytes,
4204 struct x86_exception *exception)
4206 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4207 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4209 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4212 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4214 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4215 gva_t addr, void *val, unsigned int bytes,
4216 struct x86_exception *exception)
4218 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4219 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4222 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4223 gva_t addr, void *val,
4225 struct x86_exception *exception)
4227 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4229 int r = X86EMUL_CONTINUE;
4232 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4235 unsigned offset = addr & (PAGE_SIZE-1);
4236 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4239 if (gpa == UNMAPPED_GVA)
4240 return X86EMUL_PROPAGATE_FAULT;
4241 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4243 r = X86EMUL_IO_NEEDED;
4254 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4256 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4257 gpa_t *gpa, struct x86_exception *exception,
4260 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4261 | (write ? PFERR_WRITE_MASK : 0);
4263 if (vcpu_match_mmio_gva(vcpu, gva)
4264 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4265 vcpu->arch.access, access)) {
4266 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4267 (gva & (PAGE_SIZE - 1));
4268 trace_vcpu_match_mmio(gva, *gpa, write, false);
4272 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4274 if (*gpa == UNMAPPED_GVA)
4277 /* For APIC access vmexit */
4278 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4281 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4282 trace_vcpu_match_mmio(gva, *gpa, write, true);
4289 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4290 const void *val, int bytes)
4294 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4297 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4301 struct read_write_emulator_ops {
4302 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4304 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4305 void *val, int bytes);
4306 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4307 int bytes, void *val);
4308 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4309 void *val, int bytes);
4313 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4315 if (vcpu->mmio_read_completed) {
4316 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4317 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4318 vcpu->mmio_read_completed = 0;
4325 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4326 void *val, int bytes)
4328 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4331 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4332 void *val, int bytes)
4334 return emulator_write_phys(vcpu, gpa, val, bytes);
4337 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4339 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4340 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4343 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4344 void *val, int bytes)
4346 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4347 return X86EMUL_IO_NEEDED;
4350 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4351 void *val, int bytes)
4353 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4355 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4356 return X86EMUL_CONTINUE;
4359 static const struct read_write_emulator_ops read_emultor = {
4360 .read_write_prepare = read_prepare,
4361 .read_write_emulate = read_emulate,
4362 .read_write_mmio = vcpu_mmio_read,
4363 .read_write_exit_mmio = read_exit_mmio,
4366 static const struct read_write_emulator_ops write_emultor = {
4367 .read_write_emulate = write_emulate,
4368 .read_write_mmio = write_mmio,
4369 .read_write_exit_mmio = write_exit_mmio,
4373 static int emulator_read_write_onepage(unsigned long addr, void *val,
4375 struct x86_exception *exception,
4376 struct kvm_vcpu *vcpu,
4377 const struct read_write_emulator_ops *ops)
4381 bool write = ops->write;
4382 struct kvm_mmio_fragment *frag;
4384 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4387 return X86EMUL_PROPAGATE_FAULT;
4389 /* For APIC access vmexit */
4393 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4394 return X86EMUL_CONTINUE;
4398 * Is this MMIO handled locally?
4400 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4401 if (handled == bytes)
4402 return X86EMUL_CONTINUE;
4408 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4409 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4413 return X86EMUL_CONTINUE;
4416 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4417 void *val, unsigned int bytes,
4418 struct x86_exception *exception,
4419 const struct read_write_emulator_ops *ops)
4421 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4425 if (ops->read_write_prepare &&
4426 ops->read_write_prepare(vcpu, val, bytes))
4427 return X86EMUL_CONTINUE;
4429 vcpu->mmio_nr_fragments = 0;
4431 /* Crossing a page boundary? */
4432 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4435 now = -addr & ~PAGE_MASK;
4436 rc = emulator_read_write_onepage(addr, val, now, exception,
4439 if (rc != X86EMUL_CONTINUE)
4446 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4448 if (rc != X86EMUL_CONTINUE)
4451 if (!vcpu->mmio_nr_fragments)
4454 gpa = vcpu->mmio_fragments[0].gpa;
4456 vcpu->mmio_needed = 1;
4457 vcpu->mmio_cur_fragment = 0;
4459 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4460 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4461 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4462 vcpu->run->mmio.phys_addr = gpa;
4464 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4467 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4471 struct x86_exception *exception)
4473 return emulator_read_write(ctxt, addr, val, bytes,
4474 exception, &read_emultor);
4477 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4481 struct x86_exception *exception)
4483 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4484 exception, &write_emultor);
4487 #define CMPXCHG_TYPE(t, ptr, old, new) \
4488 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4490 #ifdef CONFIG_X86_64
4491 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4493 # define CMPXCHG64(ptr, old, new) \
4494 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4497 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4502 struct x86_exception *exception)
4504 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4510 /* guests cmpxchg8b have to be emulated atomically */
4511 if (bytes > 8 || (bytes & (bytes - 1)))
4514 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4516 if (gpa == UNMAPPED_GVA ||
4517 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4520 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4523 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4524 if (is_error_page(page))
4527 kaddr = kmap_atomic(page);
4528 kaddr += offset_in_page(gpa);
4531 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4534 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4537 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4540 exchanged = CMPXCHG64(kaddr, old, new);
4545 kunmap_atomic(kaddr);
4546 kvm_release_page_dirty(page);
4549 return X86EMUL_CMPXCHG_FAILED;
4551 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
4552 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4554 return X86EMUL_CONTINUE;
4557 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4559 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4562 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4564 /* TODO: String I/O for in kernel device */
4567 if (vcpu->arch.pio.in)
4568 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4569 vcpu->arch.pio.size, pd);
4571 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4572 vcpu->arch.pio.port, vcpu->arch.pio.size,
4577 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4578 unsigned short port, void *val,
4579 unsigned int count, bool in)
4581 vcpu->arch.pio.port = port;
4582 vcpu->arch.pio.in = in;
4583 vcpu->arch.pio.count = count;
4584 vcpu->arch.pio.size = size;
4586 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4587 vcpu->arch.pio.count = 0;
4591 vcpu->run->exit_reason = KVM_EXIT_IO;
4592 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4593 vcpu->run->io.size = size;
4594 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4595 vcpu->run->io.count = count;
4596 vcpu->run->io.port = port;
4601 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4602 int size, unsigned short port, void *val,
4605 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4608 if (vcpu->arch.pio.count)
4611 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4614 memcpy(val, vcpu->arch.pio_data, size * count);
4615 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4616 vcpu->arch.pio.count = 0;
4623 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4624 int size, unsigned short port,
4625 const void *val, unsigned int count)
4627 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4629 memcpy(vcpu->arch.pio_data, val, size * count);
4630 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4631 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4634 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4636 return kvm_x86_ops->get_segment_base(vcpu, seg);
4639 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4641 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4644 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4646 if (!need_emulate_wbinvd(vcpu))
4647 return X86EMUL_CONTINUE;
4649 if (kvm_x86_ops->has_wbinvd_exit()) {
4650 int cpu = get_cpu();
4652 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4653 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4654 wbinvd_ipi, NULL, 1);
4656 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4659 return X86EMUL_CONTINUE;
4661 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4663 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4665 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4668 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4670 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4673 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4676 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4679 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4681 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4684 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4686 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4687 unsigned long value;
4691 value = kvm_read_cr0(vcpu);
4694 value = vcpu->arch.cr2;
4697 value = kvm_read_cr3(vcpu);
4700 value = kvm_read_cr4(vcpu);
4703 value = kvm_get_cr8(vcpu);
4706 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4713 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4715 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4720 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4723 vcpu->arch.cr2 = val;
4726 res = kvm_set_cr3(vcpu, val);
4729 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4732 res = kvm_set_cr8(vcpu, val);
4735 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4742 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4744 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4747 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4749 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4752 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4754 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4757 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4759 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4762 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4764 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4767 static unsigned long emulator_get_cached_segment_base(
4768 struct x86_emulate_ctxt *ctxt, int seg)
4770 return get_segment_base(emul_to_vcpu(ctxt), seg);
4773 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4774 struct desc_struct *desc, u32 *base3,
4777 struct kvm_segment var;
4779 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4780 *selector = var.selector;
4783 memset(desc, 0, sizeof(*desc));
4789 set_desc_limit(desc, var.limit);
4790 set_desc_base(desc, (unsigned long)var.base);
4791 #ifdef CONFIG_X86_64
4793 *base3 = var.base >> 32;
4795 desc->type = var.type;
4797 desc->dpl = var.dpl;
4798 desc->p = var.present;
4799 desc->avl = var.avl;
4807 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4808 struct desc_struct *desc, u32 base3,
4811 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4812 struct kvm_segment var;
4814 var.selector = selector;
4815 var.base = get_desc_base(desc);
4816 #ifdef CONFIG_X86_64
4817 var.base |= ((u64)base3) << 32;
4819 var.limit = get_desc_limit(desc);
4821 var.limit = (var.limit << 12) | 0xfff;
4822 var.type = desc->type;
4823 var.dpl = desc->dpl;
4828 var.avl = desc->avl;
4829 var.present = desc->p;
4830 var.unusable = !var.present;
4833 kvm_set_segment(vcpu, &var, seg);
4837 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4838 u32 msr_index, u64 *pdata)
4840 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4843 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4844 u32 msr_index, u64 data)
4846 struct msr_data msr;
4849 msr.index = msr_index;
4850 msr.host_initiated = false;
4851 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4854 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4857 return kvm_pmu_check_pmc(emul_to_vcpu(ctxt), pmc);
4860 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4861 u32 pmc, u64 *pdata)
4863 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4866 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4868 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4871 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4874 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4876 * CR0.TS may reference the host fpu state, not the guest fpu state,
4877 * so it may be clear at this point.
4882 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4887 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4888 struct x86_instruction_info *info,
4889 enum x86_intercept_stage stage)
4891 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4894 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4895 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4897 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4900 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4902 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4905 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4907 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4910 static const struct x86_emulate_ops emulate_ops = {
4911 .read_gpr = emulator_read_gpr,
4912 .write_gpr = emulator_write_gpr,
4913 .read_std = kvm_read_guest_virt_system,
4914 .write_std = kvm_write_guest_virt_system,
4915 .fetch = kvm_fetch_guest_virt,
4916 .read_emulated = emulator_read_emulated,
4917 .write_emulated = emulator_write_emulated,
4918 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4919 .invlpg = emulator_invlpg,
4920 .pio_in_emulated = emulator_pio_in_emulated,
4921 .pio_out_emulated = emulator_pio_out_emulated,
4922 .get_segment = emulator_get_segment,
4923 .set_segment = emulator_set_segment,
4924 .get_cached_segment_base = emulator_get_cached_segment_base,
4925 .get_gdt = emulator_get_gdt,
4926 .get_idt = emulator_get_idt,
4927 .set_gdt = emulator_set_gdt,
4928 .set_idt = emulator_set_idt,
4929 .get_cr = emulator_get_cr,
4930 .set_cr = emulator_set_cr,
4931 .cpl = emulator_get_cpl,
4932 .get_dr = emulator_get_dr,
4933 .set_dr = emulator_set_dr,
4934 .set_msr = emulator_set_msr,
4935 .get_msr = emulator_get_msr,
4936 .check_pmc = emulator_check_pmc,
4937 .read_pmc = emulator_read_pmc,
4938 .halt = emulator_halt,
4939 .wbinvd = emulator_wbinvd,
4940 .fix_hypercall = emulator_fix_hypercall,
4941 .get_fpu = emulator_get_fpu,
4942 .put_fpu = emulator_put_fpu,
4943 .intercept = emulator_intercept,
4944 .get_cpuid = emulator_get_cpuid,
4947 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4949 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
4951 * an sti; sti; sequence only disable interrupts for the first
4952 * instruction. So, if the last instruction, be it emulated or
4953 * not, left the system with the INT_STI flag enabled, it
4954 * means that the last instruction is an sti. We should not
4955 * leave the flag on in this case. The same goes for mov ss
4957 if (int_shadow & mask)
4959 if (unlikely(int_shadow || mask)) {
4960 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4962 kvm_make_request(KVM_REQ_EVENT, vcpu);
4966 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
4968 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4969 if (ctxt->exception.vector == PF_VECTOR)
4970 return kvm_propagate_fault(vcpu, &ctxt->exception);
4972 if (ctxt->exception.error_code_valid)
4973 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4974 ctxt->exception.error_code);
4976 kvm_queue_exception(vcpu, ctxt->exception.vector);
4980 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4982 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4985 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4987 ctxt->eflags = kvm_get_rflags(vcpu);
4988 ctxt->eip = kvm_rip_read(vcpu);
4989 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4990 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4991 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
4992 cs_db ? X86EMUL_MODE_PROT32 :
4993 X86EMUL_MODE_PROT16;
4994 ctxt->guest_mode = is_guest_mode(vcpu);
4996 init_decode_cache(ctxt);
4997 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5000 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5002 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5005 init_emulate_ctxt(vcpu);
5009 ctxt->_eip = ctxt->eip + inc_eip;
5010 ret = emulate_int_real(ctxt, irq);
5012 if (ret != X86EMUL_CONTINUE)
5013 return EMULATE_FAIL;
5015 ctxt->eip = ctxt->_eip;
5016 kvm_rip_write(vcpu, ctxt->eip);
5017 kvm_set_rflags(vcpu, ctxt->eflags);
5019 if (irq == NMI_VECTOR)
5020 vcpu->arch.nmi_pending = 0;
5022 vcpu->arch.interrupt.pending = false;
5024 return EMULATE_DONE;
5026 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5028 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5030 int r = EMULATE_DONE;
5032 ++vcpu->stat.insn_emulation_fail;
5033 trace_kvm_emulate_insn_failed(vcpu);
5034 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5035 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5036 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5037 vcpu->run->internal.ndata = 0;
5040 kvm_queue_exception(vcpu, UD_VECTOR);
5045 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5046 bool write_fault_to_shadow_pgtable,
5052 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5055 if (!vcpu->arch.mmu.direct_map) {
5057 * Write permission should be allowed since only
5058 * write access need to be emulated.
5060 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5063 * If the mapping is invalid in guest, let cpu retry
5064 * it to generate fault.
5066 if (gpa == UNMAPPED_GVA)
5071 * Do not retry the unhandleable instruction if it faults on the
5072 * readonly host memory, otherwise it will goto a infinite loop:
5073 * retry instruction -> write #PF -> emulation fail -> retry
5074 * instruction -> ...
5076 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5079 * If the instruction failed on the error pfn, it can not be fixed,
5080 * report the error to userspace.
5082 if (is_error_noslot_pfn(pfn))
5085 kvm_release_pfn_clean(pfn);
5087 /* The instructions are well-emulated on direct mmu. */
5088 if (vcpu->arch.mmu.direct_map) {
5089 unsigned int indirect_shadow_pages;
5091 spin_lock(&vcpu->kvm->mmu_lock);
5092 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5093 spin_unlock(&vcpu->kvm->mmu_lock);
5095 if (indirect_shadow_pages)
5096 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5102 * if emulation was due to access to shadowed page table
5103 * and it failed try to unshadow page and re-enter the
5104 * guest to let CPU execute the instruction.
5106 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5109 * If the access faults on its page table, it can not
5110 * be fixed by unprotecting shadow page and it should
5111 * be reported to userspace.
5113 return !write_fault_to_shadow_pgtable;
5116 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5117 unsigned long cr2, int emulation_type)
5119 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5120 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5122 last_retry_eip = vcpu->arch.last_retry_eip;
5123 last_retry_addr = vcpu->arch.last_retry_addr;
5126 * If the emulation is caused by #PF and it is non-page_table
5127 * writing instruction, it means the VM-EXIT is caused by shadow
5128 * page protected, we can zap the shadow page and retry this
5129 * instruction directly.
5131 * Note: if the guest uses a non-page-table modifying instruction
5132 * on the PDE that points to the instruction, then we will unmap
5133 * the instruction and go to an infinite loop. So, we cache the
5134 * last retried eip and the last fault address, if we meet the eip
5135 * and the address again, we can break out of the potential infinite
5138 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5140 if (!(emulation_type & EMULTYPE_RETRY))
5143 if (x86_page_table_writing_insn(ctxt))
5146 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5149 vcpu->arch.last_retry_eip = ctxt->eip;
5150 vcpu->arch.last_retry_addr = cr2;
5152 if (!vcpu->arch.mmu.direct_map)
5153 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5155 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5160 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5161 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5163 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5172 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5173 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5178 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5180 struct kvm_run *kvm_run = vcpu->run;
5183 * rflags is the old, "raw" value of the flags. The new value has
5184 * not been saved yet.
5186 * This is correct even for TF set by the guest, because "the
5187 * processor will not generate this exception after the instruction
5188 * that sets the TF flag".
5190 if (unlikely(rflags & X86_EFLAGS_TF)) {
5191 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5192 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5194 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5195 kvm_run->debug.arch.exception = DB_VECTOR;
5196 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5197 *r = EMULATE_USER_EXIT;
5199 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5201 * "Certain debug exceptions may clear bit 0-3. The
5202 * remaining contents of the DR6 register are never
5203 * cleared by the processor".
5205 vcpu->arch.dr6 &= ~15;
5206 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5207 kvm_queue_exception(vcpu, DB_VECTOR);
5212 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5214 struct kvm_run *kvm_run = vcpu->run;
5215 unsigned long eip = vcpu->arch.emulate_ctxt.eip;
5218 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5219 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5220 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5221 vcpu->arch.guest_debug_dr7,
5225 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5226 kvm_run->debug.arch.pc = kvm_rip_read(vcpu) +
5227 get_segment_base(vcpu, VCPU_SREG_CS);
5229 kvm_run->debug.arch.exception = DB_VECTOR;
5230 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5231 *r = EMULATE_USER_EXIT;
5236 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5237 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5238 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5243 vcpu->arch.dr6 &= ~15;
5244 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5245 kvm_queue_exception(vcpu, DB_VECTOR);
5254 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5261 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5262 bool writeback = true;
5263 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5266 * Clear write_fault_to_shadow_pgtable here to ensure it is
5269 vcpu->arch.write_fault_to_shadow_pgtable = false;
5270 kvm_clear_exception_queue(vcpu);
5272 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5273 init_emulate_ctxt(vcpu);
5276 * We will reenter on the same instruction since
5277 * we do not set complete_userspace_io. This does not
5278 * handle watchpoints yet, those would be handled in
5281 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5284 ctxt->interruptibility = 0;
5285 ctxt->have_exception = false;
5286 ctxt->exception.vector = -1;
5287 ctxt->perm_ok = false;
5289 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5291 r = x86_decode_insn(ctxt, insn, insn_len);
5293 trace_kvm_emulate_insn_start(vcpu);
5294 ++vcpu->stat.insn_emulation;
5295 if (r != EMULATION_OK) {
5296 if (emulation_type & EMULTYPE_TRAP_UD)
5297 return EMULATE_FAIL;
5298 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5300 return EMULATE_DONE;
5301 if (emulation_type & EMULTYPE_SKIP)
5302 return EMULATE_FAIL;
5303 return handle_emulation_failure(vcpu);
5307 if (emulation_type & EMULTYPE_SKIP) {
5308 kvm_rip_write(vcpu, ctxt->_eip);
5309 if (ctxt->eflags & X86_EFLAGS_RF)
5310 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5311 return EMULATE_DONE;
5314 if (retry_instruction(ctxt, cr2, emulation_type))
5315 return EMULATE_DONE;
5317 /* this is needed for vmware backdoor interface to work since it
5318 changes registers values during IO operation */
5319 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5320 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5321 emulator_invalidate_register_cache(ctxt);
5325 r = x86_emulate_insn(ctxt);
5327 if (r == EMULATION_INTERCEPTED)
5328 return EMULATE_DONE;
5330 if (r == EMULATION_FAILED) {
5331 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5333 return EMULATE_DONE;
5335 return handle_emulation_failure(vcpu);
5338 if (ctxt->have_exception) {
5340 if (inject_emulated_exception(vcpu))
5342 } else if (vcpu->arch.pio.count) {
5343 if (!vcpu->arch.pio.in) {
5344 /* FIXME: return into emulator if single-stepping. */
5345 vcpu->arch.pio.count = 0;
5348 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5350 r = EMULATE_USER_EXIT;
5351 } else if (vcpu->mmio_needed) {
5352 if (!vcpu->mmio_is_write)
5354 r = EMULATE_USER_EXIT;
5355 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5356 } else if (r == EMULATION_RESTART)
5362 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5363 toggle_interruptibility(vcpu, ctxt->interruptibility);
5364 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5365 kvm_rip_write(vcpu, ctxt->eip);
5366 if (r == EMULATE_DONE)
5367 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5368 __kvm_set_rflags(vcpu, ctxt->eflags);
5371 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5372 * do nothing, and it will be requested again as soon as
5373 * the shadow expires. But we still need to check here,
5374 * because POPF has no interrupt shadow.
5376 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5377 kvm_make_request(KVM_REQ_EVENT, vcpu);
5379 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5383 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5385 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5387 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5388 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5389 size, port, &val, 1);
5390 /* do not return to emulator after return from userspace */
5391 vcpu->arch.pio.count = 0;
5394 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5396 static void tsc_bad(void *info)
5398 __this_cpu_write(cpu_tsc_khz, 0);
5401 static void tsc_khz_changed(void *data)
5403 struct cpufreq_freqs *freq = data;
5404 unsigned long khz = 0;
5408 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5409 khz = cpufreq_quick_get(raw_smp_processor_id());
5412 __this_cpu_write(cpu_tsc_khz, khz);
5415 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5418 struct cpufreq_freqs *freq = data;
5420 struct kvm_vcpu *vcpu;
5421 int i, send_ipi = 0;
5424 * We allow guests to temporarily run on slowing clocks,
5425 * provided we notify them after, or to run on accelerating
5426 * clocks, provided we notify them before. Thus time never
5429 * However, we have a problem. We can't atomically update
5430 * the frequency of a given CPU from this function; it is
5431 * merely a notifier, which can be called from any CPU.
5432 * Changing the TSC frequency at arbitrary points in time
5433 * requires a recomputation of local variables related to
5434 * the TSC for each VCPU. We must flag these local variables
5435 * to be updated and be sure the update takes place with the
5436 * new frequency before any guests proceed.
5438 * Unfortunately, the combination of hotplug CPU and frequency
5439 * change creates an intractable locking scenario; the order
5440 * of when these callouts happen is undefined with respect to
5441 * CPU hotplug, and they can race with each other. As such,
5442 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5443 * undefined; you can actually have a CPU frequency change take
5444 * place in between the computation of X and the setting of the
5445 * variable. To protect against this problem, all updates of
5446 * the per_cpu tsc_khz variable are done in an interrupt
5447 * protected IPI, and all callers wishing to update the value
5448 * must wait for a synchronous IPI to complete (which is trivial
5449 * if the caller is on the CPU already). This establishes the
5450 * necessary total order on variable updates.
5452 * Note that because a guest time update may take place
5453 * anytime after the setting of the VCPU's request bit, the
5454 * correct TSC value must be set before the request. However,
5455 * to ensure the update actually makes it to any guest which
5456 * starts running in hardware virtualization between the set
5457 * and the acquisition of the spinlock, we must also ping the
5458 * CPU after setting the request bit.
5462 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5464 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5467 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5469 spin_lock(&kvm_lock);
5470 list_for_each_entry(kvm, &vm_list, vm_list) {
5471 kvm_for_each_vcpu(i, vcpu, kvm) {
5472 if (vcpu->cpu != freq->cpu)
5474 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5475 if (vcpu->cpu != smp_processor_id())
5479 spin_unlock(&kvm_lock);
5481 if (freq->old < freq->new && send_ipi) {
5483 * We upscale the frequency. Must make the guest
5484 * doesn't see old kvmclock values while running with
5485 * the new frequency, otherwise we risk the guest sees
5486 * time go backwards.
5488 * In case we update the frequency for another cpu
5489 * (which might be in guest context) send an interrupt
5490 * to kick the cpu out of guest context. Next time
5491 * guest context is entered kvmclock will be updated,
5492 * so the guest will not see stale values.
5494 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5499 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5500 .notifier_call = kvmclock_cpufreq_notifier
5503 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5504 unsigned long action, void *hcpu)
5506 unsigned int cpu = (unsigned long)hcpu;
5510 case CPU_DOWN_FAILED:
5511 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5513 case CPU_DOWN_PREPARE:
5514 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5520 static struct notifier_block kvmclock_cpu_notifier_block = {
5521 .notifier_call = kvmclock_cpu_notifier,
5522 .priority = -INT_MAX
5525 static void kvm_timer_init(void)
5529 max_tsc_khz = tsc_khz;
5531 cpu_notifier_register_begin();
5532 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5533 #ifdef CONFIG_CPU_FREQ
5534 struct cpufreq_policy policy;
5535 memset(&policy, 0, sizeof(policy));
5537 cpufreq_get_policy(&policy, cpu);
5538 if (policy.cpuinfo.max_freq)
5539 max_tsc_khz = policy.cpuinfo.max_freq;
5542 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5543 CPUFREQ_TRANSITION_NOTIFIER);
5545 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5546 for_each_online_cpu(cpu)
5547 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5549 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5550 cpu_notifier_register_done();
5554 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5556 int kvm_is_in_guest(void)
5558 return __this_cpu_read(current_vcpu) != NULL;
5561 static int kvm_is_user_mode(void)
5565 if (__this_cpu_read(current_vcpu))
5566 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5568 return user_mode != 0;
5571 static unsigned long kvm_get_guest_ip(void)
5573 unsigned long ip = 0;
5575 if (__this_cpu_read(current_vcpu))
5576 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5581 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5582 .is_in_guest = kvm_is_in_guest,
5583 .is_user_mode = kvm_is_user_mode,
5584 .get_guest_ip = kvm_get_guest_ip,
5587 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5589 __this_cpu_write(current_vcpu, vcpu);
5591 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5593 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5595 __this_cpu_write(current_vcpu, NULL);
5597 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5599 static void kvm_set_mmio_spte_mask(void)
5602 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5605 * Set the reserved bits and the present bit of an paging-structure
5606 * entry to generate page fault with PFER.RSV = 1.
5608 /* Mask the reserved physical address bits. */
5609 mask = rsvd_bits(maxphyaddr, 51);
5611 /* Bit 62 is always reserved for 32bit host. */
5612 mask |= 0x3ull << 62;
5614 /* Set the present bit. */
5617 #ifdef CONFIG_X86_64
5619 * If reserved bit is not supported, clear the present bit to disable
5622 if (maxphyaddr == 52)
5626 kvm_mmu_set_mmio_spte_mask(mask);
5629 #ifdef CONFIG_X86_64
5630 static void pvclock_gtod_update_fn(struct work_struct *work)
5634 struct kvm_vcpu *vcpu;
5637 spin_lock(&kvm_lock);
5638 list_for_each_entry(kvm, &vm_list, vm_list)
5639 kvm_for_each_vcpu(i, vcpu, kvm)
5640 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5641 atomic_set(&kvm_guest_has_master_clock, 0);
5642 spin_unlock(&kvm_lock);
5645 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5648 * Notification about pvclock gtod data update.
5650 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5653 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5654 struct timekeeper *tk = priv;
5656 update_pvclock_gtod(tk);
5658 /* disable master clock if host does not trust, or does not
5659 * use, TSC clocksource
5661 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5662 atomic_read(&kvm_guest_has_master_clock) != 0)
5663 queue_work(system_long_wq, &pvclock_gtod_work);
5668 static struct notifier_block pvclock_gtod_notifier = {
5669 .notifier_call = pvclock_gtod_notify,
5673 int kvm_arch_init(void *opaque)
5676 struct kvm_x86_ops *ops = opaque;
5679 printk(KERN_ERR "kvm: already loaded the other module\n");
5684 if (!ops->cpu_has_kvm_support()) {
5685 printk(KERN_ERR "kvm: no hardware support\n");
5689 if (ops->disabled_by_bios()) {
5690 printk(KERN_ERR "kvm: disabled by bios\n");
5696 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5698 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5702 r = kvm_mmu_module_init();
5704 goto out_free_percpu;
5706 kvm_set_mmio_spte_mask();
5709 kvm_init_msr_list();
5711 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5712 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5716 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5719 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5722 #ifdef CONFIG_X86_64
5723 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5729 free_percpu(shared_msrs);
5734 void kvm_arch_exit(void)
5736 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5738 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5739 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5740 CPUFREQ_TRANSITION_NOTIFIER);
5741 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5742 #ifdef CONFIG_X86_64
5743 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5746 kvm_mmu_module_exit();
5747 free_percpu(shared_msrs);
5750 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5752 ++vcpu->stat.halt_exits;
5753 if (irqchip_in_kernel(vcpu->kvm)) {
5754 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5757 vcpu->run->exit_reason = KVM_EXIT_HLT;
5761 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5763 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5765 u64 param, ingpa, outgpa, ret;
5766 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5767 bool fast, longmode;
5770 * hypercall generates UD from non zero cpl and real mode
5773 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5774 kvm_queue_exception(vcpu, UD_VECTOR);
5778 longmode = is_64_bit_mode(vcpu);
5781 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5782 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5783 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5784 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5785 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5786 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5788 #ifdef CONFIG_X86_64
5790 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5791 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5792 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5796 code = param & 0xffff;
5797 fast = (param >> 16) & 0x1;
5798 rep_cnt = (param >> 32) & 0xfff;
5799 rep_idx = (param >> 48) & 0xfff;
5801 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5804 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5805 kvm_vcpu_on_spin(vcpu);
5808 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5812 ret = res | (((u64)rep_done & 0xfff) << 32);
5814 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5816 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5817 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5824 * kvm_pv_kick_cpu_op: Kick a vcpu.
5826 * @apicid - apicid of vcpu to be kicked.
5828 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5830 struct kvm_lapic_irq lapic_irq;
5832 lapic_irq.shorthand = 0;
5833 lapic_irq.dest_mode = 0;
5834 lapic_irq.dest_id = apicid;
5836 lapic_irq.delivery_mode = APIC_DM_REMRD;
5837 kvm_irq_delivery_to_apic(kvm, 0, &lapic_irq, NULL);
5840 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5842 unsigned long nr, a0, a1, a2, a3, ret;
5843 int op_64_bit, r = 1;
5845 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5846 return kvm_hv_hypercall(vcpu);
5848 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5849 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5850 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5851 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5852 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5854 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5856 op_64_bit = is_64_bit_mode(vcpu);
5865 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5871 case KVM_HC_VAPIC_POLL_IRQ:
5874 case KVM_HC_KICK_CPU:
5875 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5885 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5886 ++vcpu->stat.hypercalls;
5889 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5891 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5893 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5894 char instruction[3];
5895 unsigned long rip = kvm_rip_read(vcpu);
5897 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5899 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5903 * Check if userspace requested an interrupt window, and that the
5904 * interrupt window is open.
5906 * No need to exit to userspace if we already have an interrupt queued.
5908 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5910 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5911 vcpu->run->request_interrupt_window &&
5912 kvm_arch_interrupt_allowed(vcpu));
5915 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5917 struct kvm_run *kvm_run = vcpu->run;
5919 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5920 kvm_run->cr8 = kvm_get_cr8(vcpu);
5921 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5922 if (irqchip_in_kernel(vcpu->kvm))
5923 kvm_run->ready_for_interrupt_injection = 1;
5925 kvm_run->ready_for_interrupt_injection =
5926 kvm_arch_interrupt_allowed(vcpu) &&
5927 !kvm_cpu_has_interrupt(vcpu) &&
5928 !kvm_event_needs_reinjection(vcpu);
5931 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5935 if (!kvm_x86_ops->update_cr8_intercept)
5938 if (!vcpu->arch.apic)
5941 if (!vcpu->arch.apic->vapic_addr)
5942 max_irr = kvm_lapic_find_highest_irr(vcpu);
5949 tpr = kvm_lapic_get_cr8(vcpu);
5951 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5954 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
5958 /* try to reinject previous events if any */
5959 if (vcpu->arch.exception.pending) {
5960 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5961 vcpu->arch.exception.has_error_code,
5962 vcpu->arch.exception.error_code);
5964 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
5965 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
5968 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5969 vcpu->arch.exception.has_error_code,
5970 vcpu->arch.exception.error_code,
5971 vcpu->arch.exception.reinject);
5975 if (vcpu->arch.nmi_injected) {
5976 kvm_x86_ops->set_nmi(vcpu);
5980 if (vcpu->arch.interrupt.pending) {
5981 kvm_x86_ops->set_irq(vcpu);
5985 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5986 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5991 /* try to inject new event if pending */
5992 if (vcpu->arch.nmi_pending) {
5993 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5994 --vcpu->arch.nmi_pending;
5995 vcpu->arch.nmi_injected = true;
5996 kvm_x86_ops->set_nmi(vcpu);
5998 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6000 * Because interrupts can be injected asynchronously, we are
6001 * calling check_nested_events again here to avoid a race condition.
6002 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6003 * proposal and current concerns. Perhaps we should be setting
6004 * KVM_REQ_EVENT only on certain events and not unconditionally?
6006 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6007 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6011 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6012 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6014 kvm_x86_ops->set_irq(vcpu);
6020 static void process_nmi(struct kvm_vcpu *vcpu)
6025 * x86 is limited to one NMI running, and one NMI pending after it.
6026 * If an NMI is already in progress, limit further NMIs to just one.
6027 * Otherwise, allow two (and we'll inject the first one immediately).
6029 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6032 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6033 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6034 kvm_make_request(KVM_REQ_EVENT, vcpu);
6037 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6039 u64 eoi_exit_bitmap[4];
6042 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6045 memset(eoi_exit_bitmap, 0, 32);
6048 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6049 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6050 kvm_apic_update_tmr(vcpu, tmr);
6053 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6055 ++vcpu->stat.tlb_flush;
6056 kvm_x86_ops->tlb_flush(vcpu);
6059 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6061 struct page *page = NULL;
6063 if (!irqchip_in_kernel(vcpu->kvm))
6066 if (!kvm_x86_ops->set_apic_access_page_addr)
6069 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6070 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6073 * Do not pin apic access page in memory, the MMU notifier
6074 * will call us again if it is migrated or swapped out.
6078 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6080 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6081 unsigned long address)
6084 * The physical address of apic access page is stored in the VMCS.
6085 * Update it when it becomes invalid.
6087 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6088 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6092 * Returns 1 to let __vcpu_run() continue the guest execution loop without
6093 * exiting to the userspace. Otherwise, the value will be returned to the
6096 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6099 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6100 vcpu->run->request_interrupt_window;
6101 bool req_immediate_exit = false;
6103 if (vcpu->requests) {
6104 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6105 kvm_mmu_unload(vcpu);
6106 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6107 __kvm_migrate_timers(vcpu);
6108 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6109 kvm_gen_update_masterclock(vcpu->kvm);
6110 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6111 kvm_gen_kvmclock_update(vcpu);
6112 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6113 r = kvm_guest_time_update(vcpu);
6117 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6118 kvm_mmu_sync_roots(vcpu);
6119 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6120 kvm_vcpu_flush_tlb(vcpu);
6121 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6122 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6126 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6127 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6131 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6132 vcpu->fpu_active = 0;
6133 kvm_x86_ops->fpu_deactivate(vcpu);
6135 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6136 /* Page is swapped out. Do synthetic halt */
6137 vcpu->arch.apf.halted = true;
6141 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6142 record_steal_time(vcpu);
6143 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6145 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6146 kvm_handle_pmu_event(vcpu);
6147 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6148 kvm_deliver_pmi(vcpu);
6149 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6150 vcpu_scan_ioapic(vcpu);
6151 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6152 kvm_vcpu_reload_apic_access_page(vcpu);
6155 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6156 kvm_apic_accept_events(vcpu);
6157 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6162 if (inject_pending_event(vcpu, req_int_win) != 0)
6163 req_immediate_exit = true;
6164 /* enable NMI/IRQ window open exits if needed */
6165 else if (vcpu->arch.nmi_pending)
6166 kvm_x86_ops->enable_nmi_window(vcpu);
6167 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6168 kvm_x86_ops->enable_irq_window(vcpu);
6170 if (kvm_lapic_enabled(vcpu)) {
6172 * Update architecture specific hints for APIC
6173 * virtual interrupt delivery.
6175 if (kvm_x86_ops->hwapic_irr_update)
6176 kvm_x86_ops->hwapic_irr_update(vcpu,
6177 kvm_lapic_find_highest_irr(vcpu));
6178 update_cr8_intercept(vcpu);
6179 kvm_lapic_sync_to_vapic(vcpu);
6183 r = kvm_mmu_reload(vcpu);
6185 goto cancel_injection;
6190 kvm_x86_ops->prepare_guest_switch(vcpu);
6191 if (vcpu->fpu_active)
6192 kvm_load_guest_fpu(vcpu);
6193 kvm_load_guest_xcr0(vcpu);
6195 vcpu->mode = IN_GUEST_MODE;
6197 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6199 /* We should set ->mode before check ->requests,
6200 * see the comment in make_all_cpus_request.
6202 smp_mb__after_srcu_read_unlock();
6204 local_irq_disable();
6206 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6207 || need_resched() || signal_pending(current)) {
6208 vcpu->mode = OUTSIDE_GUEST_MODE;
6212 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6214 goto cancel_injection;
6217 if (req_immediate_exit)
6218 smp_send_reschedule(vcpu->cpu);
6222 if (unlikely(vcpu->arch.switch_db_regs)) {
6224 set_debugreg(vcpu->arch.eff_db[0], 0);
6225 set_debugreg(vcpu->arch.eff_db[1], 1);
6226 set_debugreg(vcpu->arch.eff_db[2], 2);
6227 set_debugreg(vcpu->arch.eff_db[3], 3);
6228 set_debugreg(vcpu->arch.dr6, 6);
6231 trace_kvm_entry(vcpu->vcpu_id);
6232 kvm_x86_ops->run(vcpu);
6235 * Do this here before restoring debug registers on the host. And
6236 * since we do this before handling the vmexit, a DR access vmexit
6237 * can (a) read the correct value of the debug registers, (b) set
6238 * KVM_DEBUGREG_WONT_EXIT again.
6240 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6243 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6244 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6245 for (i = 0; i < KVM_NR_DB_REGS; i++)
6246 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6250 * If the guest has used debug registers, at least dr7
6251 * will be disabled while returning to the host.
6252 * If we don't have active breakpoints in the host, we don't
6253 * care about the messed up debug address registers. But if
6254 * we have some of them active, restore the old state.
6256 if (hw_breakpoint_active())
6257 hw_breakpoint_restore();
6259 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6262 vcpu->mode = OUTSIDE_GUEST_MODE;
6265 /* Interrupt is enabled by handle_external_intr() */
6266 kvm_x86_ops->handle_external_intr(vcpu);
6271 * We must have an instruction between local_irq_enable() and
6272 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6273 * the interrupt shadow. The stat.exits increment will do nicely.
6274 * But we need to prevent reordering, hence this barrier():
6282 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6285 * Profile KVM exit RIPs:
6287 if (unlikely(prof_on == KVM_PROFILING)) {
6288 unsigned long rip = kvm_rip_read(vcpu);
6289 profile_hit(KVM_PROFILING, (void *)rip);
6292 if (unlikely(vcpu->arch.tsc_always_catchup))
6293 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6295 if (vcpu->arch.apic_attention)
6296 kvm_lapic_sync_from_vapic(vcpu);
6298 r = kvm_x86_ops->handle_exit(vcpu);
6302 kvm_x86_ops->cancel_injection(vcpu);
6303 if (unlikely(vcpu->arch.apic_attention))
6304 kvm_lapic_sync_from_vapic(vcpu);
6310 static int __vcpu_run(struct kvm_vcpu *vcpu)
6313 struct kvm *kvm = vcpu->kvm;
6315 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6319 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6320 !vcpu->arch.apf.halted)
6321 r = vcpu_enter_guest(vcpu);
6323 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6324 kvm_vcpu_block(vcpu);
6325 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6326 if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
6327 kvm_apic_accept_events(vcpu);
6328 switch(vcpu->arch.mp_state) {
6329 case KVM_MP_STATE_HALTED:
6330 vcpu->arch.pv.pv_unhalted = false;
6331 vcpu->arch.mp_state =
6332 KVM_MP_STATE_RUNNABLE;
6333 case KVM_MP_STATE_RUNNABLE:
6334 vcpu->arch.apf.halted = false;
6336 case KVM_MP_STATE_INIT_RECEIVED:
6348 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6349 if (kvm_cpu_has_pending_timer(vcpu))
6350 kvm_inject_pending_timer_irqs(vcpu);
6352 if (dm_request_for_irq_injection(vcpu)) {
6354 vcpu->run->exit_reason = KVM_EXIT_INTR;
6355 ++vcpu->stat.request_irq_exits;
6358 kvm_check_async_pf_completion(vcpu);
6360 if (signal_pending(current)) {
6362 vcpu->run->exit_reason = KVM_EXIT_INTR;
6363 ++vcpu->stat.signal_exits;
6365 if (need_resched()) {
6366 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6368 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6372 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6377 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6380 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6381 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6382 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6383 if (r != EMULATE_DONE)
6388 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6390 BUG_ON(!vcpu->arch.pio.count);
6392 return complete_emulated_io(vcpu);
6396 * Implements the following, as a state machine:
6400 * for each mmio piece in the fragment
6408 * for each mmio piece in the fragment
6413 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6415 struct kvm_run *run = vcpu->run;
6416 struct kvm_mmio_fragment *frag;
6419 BUG_ON(!vcpu->mmio_needed);
6421 /* Complete previous fragment */
6422 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6423 len = min(8u, frag->len);
6424 if (!vcpu->mmio_is_write)
6425 memcpy(frag->data, run->mmio.data, len);
6427 if (frag->len <= 8) {
6428 /* Switch to the next fragment. */
6430 vcpu->mmio_cur_fragment++;
6432 /* Go forward to the next mmio piece. */
6438 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6439 vcpu->mmio_needed = 0;
6441 /* FIXME: return into emulator if single-stepping. */
6442 if (vcpu->mmio_is_write)
6444 vcpu->mmio_read_completed = 1;
6445 return complete_emulated_io(vcpu);
6448 run->exit_reason = KVM_EXIT_MMIO;
6449 run->mmio.phys_addr = frag->gpa;
6450 if (vcpu->mmio_is_write)
6451 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6452 run->mmio.len = min(8u, frag->len);
6453 run->mmio.is_write = vcpu->mmio_is_write;
6454 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6459 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6464 if (!tsk_used_math(current) && init_fpu(current))
6467 if (vcpu->sigset_active)
6468 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6470 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6471 kvm_vcpu_block(vcpu);
6472 kvm_apic_accept_events(vcpu);
6473 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6478 /* re-sync apic's tpr */
6479 if (!irqchip_in_kernel(vcpu->kvm)) {
6480 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6486 if (unlikely(vcpu->arch.complete_userspace_io)) {
6487 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6488 vcpu->arch.complete_userspace_io = NULL;
6493 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6495 r = __vcpu_run(vcpu);
6498 post_kvm_run_save(vcpu);
6499 if (vcpu->sigset_active)
6500 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6505 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6507 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6509 * We are here if userspace calls get_regs() in the middle of
6510 * instruction emulation. Registers state needs to be copied
6511 * back from emulation context to vcpu. Userspace shouldn't do
6512 * that usually, but some bad designed PV devices (vmware
6513 * backdoor interface) need this to work
6515 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6516 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6518 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6519 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6520 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6521 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6522 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6523 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6524 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6525 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6526 #ifdef CONFIG_X86_64
6527 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6528 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6529 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6530 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6531 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6532 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6533 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6534 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6537 regs->rip = kvm_rip_read(vcpu);
6538 regs->rflags = kvm_get_rflags(vcpu);
6543 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6545 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6546 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6548 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6549 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6550 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6551 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6552 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6553 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6554 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6555 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6556 #ifdef CONFIG_X86_64
6557 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6558 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6559 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6560 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6561 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6562 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6563 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6564 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6567 kvm_rip_write(vcpu, regs->rip);
6568 kvm_set_rflags(vcpu, regs->rflags);
6570 vcpu->arch.exception.pending = false;
6572 kvm_make_request(KVM_REQ_EVENT, vcpu);
6577 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6579 struct kvm_segment cs;
6581 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6585 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6587 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6588 struct kvm_sregs *sregs)
6592 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6593 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6594 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6595 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6596 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6597 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6599 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6600 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6602 kvm_x86_ops->get_idt(vcpu, &dt);
6603 sregs->idt.limit = dt.size;
6604 sregs->idt.base = dt.address;
6605 kvm_x86_ops->get_gdt(vcpu, &dt);
6606 sregs->gdt.limit = dt.size;
6607 sregs->gdt.base = dt.address;
6609 sregs->cr0 = kvm_read_cr0(vcpu);
6610 sregs->cr2 = vcpu->arch.cr2;
6611 sregs->cr3 = kvm_read_cr3(vcpu);
6612 sregs->cr4 = kvm_read_cr4(vcpu);
6613 sregs->cr8 = kvm_get_cr8(vcpu);
6614 sregs->efer = vcpu->arch.efer;
6615 sregs->apic_base = kvm_get_apic_base(vcpu);
6617 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6619 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6620 set_bit(vcpu->arch.interrupt.nr,
6621 (unsigned long *)sregs->interrupt_bitmap);
6626 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6627 struct kvm_mp_state *mp_state)
6629 kvm_apic_accept_events(vcpu);
6630 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6631 vcpu->arch.pv.pv_unhalted)
6632 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6634 mp_state->mp_state = vcpu->arch.mp_state;
6639 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6640 struct kvm_mp_state *mp_state)
6642 if (!kvm_vcpu_has_lapic(vcpu) &&
6643 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6646 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6647 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6648 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6650 vcpu->arch.mp_state = mp_state->mp_state;
6651 kvm_make_request(KVM_REQ_EVENT, vcpu);
6655 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6656 int reason, bool has_error_code, u32 error_code)
6658 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6661 init_emulate_ctxt(vcpu);
6663 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6664 has_error_code, error_code);
6667 return EMULATE_FAIL;
6669 kvm_rip_write(vcpu, ctxt->eip);
6670 kvm_set_rflags(vcpu, ctxt->eflags);
6671 kvm_make_request(KVM_REQ_EVENT, vcpu);
6672 return EMULATE_DONE;
6674 EXPORT_SYMBOL_GPL(kvm_task_switch);
6676 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6677 struct kvm_sregs *sregs)
6679 struct msr_data apic_base_msr;
6680 int mmu_reset_needed = 0;
6681 int pending_vec, max_bits, idx;
6684 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6687 dt.size = sregs->idt.limit;
6688 dt.address = sregs->idt.base;
6689 kvm_x86_ops->set_idt(vcpu, &dt);
6690 dt.size = sregs->gdt.limit;
6691 dt.address = sregs->gdt.base;
6692 kvm_x86_ops->set_gdt(vcpu, &dt);
6694 vcpu->arch.cr2 = sregs->cr2;
6695 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6696 vcpu->arch.cr3 = sregs->cr3;
6697 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6699 kvm_set_cr8(vcpu, sregs->cr8);
6701 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6702 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6703 apic_base_msr.data = sregs->apic_base;
6704 apic_base_msr.host_initiated = true;
6705 kvm_set_apic_base(vcpu, &apic_base_msr);
6707 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6708 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6709 vcpu->arch.cr0 = sregs->cr0;
6711 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6712 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6713 if (sregs->cr4 & X86_CR4_OSXSAVE)
6714 kvm_update_cpuid(vcpu);
6716 idx = srcu_read_lock(&vcpu->kvm->srcu);
6717 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6718 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6719 mmu_reset_needed = 1;
6721 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6723 if (mmu_reset_needed)
6724 kvm_mmu_reset_context(vcpu);
6726 max_bits = KVM_NR_INTERRUPTS;
6727 pending_vec = find_first_bit(
6728 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6729 if (pending_vec < max_bits) {
6730 kvm_queue_interrupt(vcpu, pending_vec, false);
6731 pr_debug("Set back pending irq %d\n", pending_vec);
6734 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6735 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6736 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6737 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6738 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6739 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6741 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6742 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6744 update_cr8_intercept(vcpu);
6746 /* Older userspace won't unhalt the vcpu on reset. */
6747 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6748 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6750 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6752 kvm_make_request(KVM_REQ_EVENT, vcpu);
6757 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6758 struct kvm_guest_debug *dbg)
6760 unsigned long rflags;
6763 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6765 if (vcpu->arch.exception.pending)
6767 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6768 kvm_queue_exception(vcpu, DB_VECTOR);
6770 kvm_queue_exception(vcpu, BP_VECTOR);
6774 * Read rflags as long as potentially injected trace flags are still
6777 rflags = kvm_get_rflags(vcpu);
6779 vcpu->guest_debug = dbg->control;
6780 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6781 vcpu->guest_debug = 0;
6783 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6784 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6785 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6786 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6788 for (i = 0; i < KVM_NR_DB_REGS; i++)
6789 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6791 kvm_update_dr7(vcpu);
6793 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6794 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6795 get_segment_base(vcpu, VCPU_SREG_CS);
6798 * Trigger an rflags update that will inject or remove the trace
6801 kvm_set_rflags(vcpu, rflags);
6803 kvm_x86_ops->update_db_bp_intercept(vcpu);
6813 * Translate a guest virtual address to a guest physical address.
6815 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6816 struct kvm_translation *tr)
6818 unsigned long vaddr = tr->linear_address;
6822 idx = srcu_read_lock(&vcpu->kvm->srcu);
6823 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6824 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6825 tr->physical_address = gpa;
6826 tr->valid = gpa != UNMAPPED_GVA;
6833 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6835 struct i387_fxsave_struct *fxsave =
6836 &vcpu->arch.guest_fpu.state->fxsave;
6838 memcpy(fpu->fpr, fxsave->st_space, 128);
6839 fpu->fcw = fxsave->cwd;
6840 fpu->fsw = fxsave->swd;
6841 fpu->ftwx = fxsave->twd;
6842 fpu->last_opcode = fxsave->fop;
6843 fpu->last_ip = fxsave->rip;
6844 fpu->last_dp = fxsave->rdp;
6845 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6850 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6852 struct i387_fxsave_struct *fxsave =
6853 &vcpu->arch.guest_fpu.state->fxsave;
6855 memcpy(fxsave->st_space, fpu->fpr, 128);
6856 fxsave->cwd = fpu->fcw;
6857 fxsave->swd = fpu->fsw;
6858 fxsave->twd = fpu->ftwx;
6859 fxsave->fop = fpu->last_opcode;
6860 fxsave->rip = fpu->last_ip;
6861 fxsave->rdp = fpu->last_dp;
6862 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6867 int fx_init(struct kvm_vcpu *vcpu)
6871 err = fpu_alloc(&vcpu->arch.guest_fpu);
6875 fpu_finit(&vcpu->arch.guest_fpu);
6878 * Ensure guest xcr0 is valid for loading
6880 vcpu->arch.xcr0 = XSTATE_FP;
6882 vcpu->arch.cr0 |= X86_CR0_ET;
6886 EXPORT_SYMBOL_GPL(fx_init);
6888 static void fx_free(struct kvm_vcpu *vcpu)
6890 fpu_free(&vcpu->arch.guest_fpu);
6893 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6895 if (vcpu->guest_fpu_loaded)
6899 * Restore all possible states in the guest,
6900 * and assume host would use all available bits.
6901 * Guest xcr0 would be loaded later.
6903 kvm_put_guest_xcr0(vcpu);
6904 vcpu->guest_fpu_loaded = 1;
6905 __kernel_fpu_begin();
6906 fpu_restore_checking(&vcpu->arch.guest_fpu);
6910 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6912 kvm_put_guest_xcr0(vcpu);
6914 if (!vcpu->guest_fpu_loaded)
6917 vcpu->guest_fpu_loaded = 0;
6918 fpu_save_init(&vcpu->arch.guest_fpu);
6920 ++vcpu->stat.fpu_reload;
6921 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6925 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6927 kvmclock_reset(vcpu);
6929 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6931 kvm_x86_ops->vcpu_free(vcpu);
6934 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6937 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6938 printk_once(KERN_WARNING
6939 "kvm: SMP vm created on host with unstable TSC; "
6940 "guest TSC will not be reliable\n");
6941 return kvm_x86_ops->vcpu_create(kvm, id);
6944 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6948 vcpu->arch.mtrr_state.have_fixed = 1;
6949 r = vcpu_load(vcpu);
6952 kvm_vcpu_reset(vcpu);
6953 kvm_mmu_setup(vcpu);
6959 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6962 struct msr_data msr;
6963 struct kvm *kvm = vcpu->kvm;
6965 r = vcpu_load(vcpu);
6969 msr.index = MSR_IA32_TSC;
6970 msr.host_initiated = true;
6971 kvm_write_tsc(vcpu, &msr);
6974 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
6975 KVMCLOCK_SYNC_PERIOD);
6980 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6983 vcpu->arch.apf.msr_val = 0;
6985 r = vcpu_load(vcpu);
6987 kvm_mmu_unload(vcpu);
6991 kvm_x86_ops->vcpu_free(vcpu);
6994 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6996 atomic_set(&vcpu->arch.nmi_queued, 0);
6997 vcpu->arch.nmi_pending = 0;
6998 vcpu->arch.nmi_injected = false;
6999 kvm_clear_interrupt_queue(vcpu);
7000 kvm_clear_exception_queue(vcpu);
7002 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7003 vcpu->arch.dr6 = DR6_INIT;
7004 kvm_update_dr6(vcpu);
7005 vcpu->arch.dr7 = DR7_FIXED_1;
7006 kvm_update_dr7(vcpu);
7008 kvm_make_request(KVM_REQ_EVENT, vcpu);
7009 vcpu->arch.apf.msr_val = 0;
7010 vcpu->arch.st.msr_val = 0;
7012 kvmclock_reset(vcpu);
7014 kvm_clear_async_pf_completion_queue(vcpu);
7015 kvm_async_pf_hash_reset(vcpu);
7016 vcpu->arch.apf.halted = false;
7018 kvm_pmu_reset(vcpu);
7020 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7021 vcpu->arch.regs_avail = ~0;
7022 vcpu->arch.regs_dirty = ~0;
7024 kvm_x86_ops->vcpu_reset(vcpu);
7027 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, unsigned int vector)
7029 struct kvm_segment cs;
7031 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7032 cs.selector = vector << 8;
7033 cs.base = vector << 12;
7034 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7035 kvm_rip_write(vcpu, 0);
7038 int kvm_arch_hardware_enable(void)
7041 struct kvm_vcpu *vcpu;
7046 bool stable, backwards_tsc = false;
7048 kvm_shared_msr_cpu_online();
7049 ret = kvm_x86_ops->hardware_enable();
7053 local_tsc = native_read_tsc();
7054 stable = !check_tsc_unstable();
7055 list_for_each_entry(kvm, &vm_list, vm_list) {
7056 kvm_for_each_vcpu(i, vcpu, kvm) {
7057 if (!stable && vcpu->cpu == smp_processor_id())
7058 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7059 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7060 backwards_tsc = true;
7061 if (vcpu->arch.last_host_tsc > max_tsc)
7062 max_tsc = vcpu->arch.last_host_tsc;
7068 * Sometimes, even reliable TSCs go backwards. This happens on
7069 * platforms that reset TSC during suspend or hibernate actions, but
7070 * maintain synchronization. We must compensate. Fortunately, we can
7071 * detect that condition here, which happens early in CPU bringup,
7072 * before any KVM threads can be running. Unfortunately, we can't
7073 * bring the TSCs fully up to date with real time, as we aren't yet far
7074 * enough into CPU bringup that we know how much real time has actually
7075 * elapsed; our helper function, get_kernel_ns() will be using boot
7076 * variables that haven't been updated yet.
7078 * So we simply find the maximum observed TSC above, then record the
7079 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7080 * the adjustment will be applied. Note that we accumulate
7081 * adjustments, in case multiple suspend cycles happen before some VCPU
7082 * gets a chance to run again. In the event that no KVM threads get a
7083 * chance to run, we will miss the entire elapsed period, as we'll have
7084 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7085 * loose cycle time. This isn't too big a deal, since the loss will be
7086 * uniform across all VCPUs (not to mention the scenario is extremely
7087 * unlikely). It is possible that a second hibernate recovery happens
7088 * much faster than a first, causing the observed TSC here to be
7089 * smaller; this would require additional padding adjustment, which is
7090 * why we set last_host_tsc to the local tsc observed here.
7092 * N.B. - this code below runs only on platforms with reliable TSC,
7093 * as that is the only way backwards_tsc is set above. Also note
7094 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7095 * have the same delta_cyc adjustment applied if backwards_tsc
7096 * is detected. Note further, this adjustment is only done once,
7097 * as we reset last_host_tsc on all VCPUs to stop this from being
7098 * called multiple times (one for each physical CPU bringup).
7100 * Platforms with unreliable TSCs don't have to deal with this, they
7101 * will be compensated by the logic in vcpu_load, which sets the TSC to
7102 * catchup mode. This will catchup all VCPUs to real time, but cannot
7103 * guarantee that they stay in perfect synchronization.
7105 if (backwards_tsc) {
7106 u64 delta_cyc = max_tsc - local_tsc;
7107 backwards_tsc_observed = true;
7108 list_for_each_entry(kvm, &vm_list, vm_list) {
7109 kvm_for_each_vcpu(i, vcpu, kvm) {
7110 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7111 vcpu->arch.last_host_tsc = local_tsc;
7112 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7116 * We have to disable TSC offset matching.. if you were
7117 * booting a VM while issuing an S4 host suspend....
7118 * you may have some problem. Solving this issue is
7119 * left as an exercise to the reader.
7121 kvm->arch.last_tsc_nsec = 0;
7122 kvm->arch.last_tsc_write = 0;
7129 void kvm_arch_hardware_disable(void)
7131 kvm_x86_ops->hardware_disable();
7132 drop_user_return_notifiers();
7135 int kvm_arch_hardware_setup(void)
7137 return kvm_x86_ops->hardware_setup();
7140 void kvm_arch_hardware_unsetup(void)
7142 kvm_x86_ops->hardware_unsetup();
7145 void kvm_arch_check_processor_compat(void *rtn)
7147 kvm_x86_ops->check_processor_compatibility(rtn);
7150 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7152 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7155 struct static_key kvm_no_apic_vcpu __read_mostly;
7157 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7163 BUG_ON(vcpu->kvm == NULL);
7166 vcpu->arch.pv.pv_unhalted = false;
7167 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7168 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
7169 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7171 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7173 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7178 vcpu->arch.pio_data = page_address(page);
7180 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7182 r = kvm_mmu_create(vcpu);
7184 goto fail_free_pio_data;
7186 if (irqchip_in_kernel(kvm)) {
7187 r = kvm_create_lapic(vcpu);
7189 goto fail_mmu_destroy;
7191 static_key_slow_inc(&kvm_no_apic_vcpu);
7193 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7195 if (!vcpu->arch.mce_banks) {
7197 goto fail_free_lapic;
7199 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7201 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7203 goto fail_free_mce_banks;
7208 goto fail_free_wbinvd_dirty_mask;
7210 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7211 vcpu->arch.pv_time_enabled = false;
7213 vcpu->arch.guest_supported_xcr0 = 0;
7214 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7216 kvm_async_pf_hash_reset(vcpu);
7220 fail_free_wbinvd_dirty_mask:
7221 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7222 fail_free_mce_banks:
7223 kfree(vcpu->arch.mce_banks);
7225 kvm_free_lapic(vcpu);
7227 kvm_mmu_destroy(vcpu);
7229 free_page((unsigned long)vcpu->arch.pio_data);
7234 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7238 kvm_pmu_destroy(vcpu);
7239 kfree(vcpu->arch.mce_banks);
7240 kvm_free_lapic(vcpu);
7241 idx = srcu_read_lock(&vcpu->kvm->srcu);
7242 kvm_mmu_destroy(vcpu);
7243 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7244 free_page((unsigned long)vcpu->arch.pio_data);
7245 if (!irqchip_in_kernel(vcpu->kvm))
7246 static_key_slow_dec(&kvm_no_apic_vcpu);
7249 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7251 kvm_x86_ops->sched_in(vcpu, cpu);
7254 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7259 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7260 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7261 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7262 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7264 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7265 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7266 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7267 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7268 &kvm->arch.irq_sources_bitmap);
7270 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7271 mutex_init(&kvm->arch.apic_map_lock);
7272 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7274 pvclock_update_vm_gtod_copy(kvm);
7276 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7277 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7282 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7285 r = vcpu_load(vcpu);
7287 kvm_mmu_unload(vcpu);
7291 static void kvm_free_vcpus(struct kvm *kvm)
7294 struct kvm_vcpu *vcpu;
7297 * Unpin any mmu pages first.
7299 kvm_for_each_vcpu(i, vcpu, kvm) {
7300 kvm_clear_async_pf_completion_queue(vcpu);
7301 kvm_unload_vcpu_mmu(vcpu);
7303 kvm_for_each_vcpu(i, vcpu, kvm)
7304 kvm_arch_vcpu_free(vcpu);
7306 mutex_lock(&kvm->lock);
7307 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7308 kvm->vcpus[i] = NULL;
7310 atomic_set(&kvm->online_vcpus, 0);
7311 mutex_unlock(&kvm->lock);
7314 void kvm_arch_sync_events(struct kvm *kvm)
7316 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7317 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7318 kvm_free_all_assigned_devices(kvm);
7322 void kvm_arch_destroy_vm(struct kvm *kvm)
7324 if (current->mm == kvm->mm) {
7326 * Free memory regions allocated on behalf of userspace,
7327 * unless the the memory map has changed due to process exit
7330 struct kvm_userspace_memory_region mem;
7331 memset(&mem, 0, sizeof(mem));
7332 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7333 kvm_set_memory_region(kvm, &mem);
7335 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7336 kvm_set_memory_region(kvm, &mem);
7338 mem.slot = TSS_PRIVATE_MEMSLOT;
7339 kvm_set_memory_region(kvm, &mem);
7341 kvm_iommu_unmap_guest(kvm);
7342 kfree(kvm->arch.vpic);
7343 kfree(kvm->arch.vioapic);
7344 kvm_free_vcpus(kvm);
7345 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7348 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7349 struct kvm_memory_slot *dont)
7353 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7354 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7355 kvm_kvfree(free->arch.rmap[i]);
7356 free->arch.rmap[i] = NULL;
7361 if (!dont || free->arch.lpage_info[i - 1] !=
7362 dont->arch.lpage_info[i - 1]) {
7363 kvm_kvfree(free->arch.lpage_info[i - 1]);
7364 free->arch.lpage_info[i - 1] = NULL;
7369 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7370 unsigned long npages)
7374 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7379 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7380 slot->base_gfn, level) + 1;
7382 slot->arch.rmap[i] =
7383 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7384 if (!slot->arch.rmap[i])
7389 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7390 sizeof(*slot->arch.lpage_info[i - 1]));
7391 if (!slot->arch.lpage_info[i - 1])
7394 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7395 slot->arch.lpage_info[i - 1][0].write_count = 1;
7396 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7397 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7398 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7400 * If the gfn and userspace address are not aligned wrt each
7401 * other, or if explicitly asked to, disable large page
7402 * support for this slot
7404 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7405 !kvm_largepages_enabled()) {
7408 for (j = 0; j < lpages; ++j)
7409 slot->arch.lpage_info[i - 1][j].write_count = 1;
7416 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7417 kvm_kvfree(slot->arch.rmap[i]);
7418 slot->arch.rmap[i] = NULL;
7422 kvm_kvfree(slot->arch.lpage_info[i - 1]);
7423 slot->arch.lpage_info[i - 1] = NULL;
7428 void kvm_arch_memslots_updated(struct kvm *kvm)
7431 * memslots->generation has been incremented.
7432 * mmio generation may have reached its maximum value.
7434 kvm_mmu_invalidate_mmio_sptes(kvm);
7437 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7438 struct kvm_memory_slot *memslot,
7439 struct kvm_userspace_memory_region *mem,
7440 enum kvm_mr_change change)
7443 * Only private memory slots need to be mapped here since
7444 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7446 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7447 unsigned long userspace_addr;
7450 * MAP_SHARED to prevent internal slot pages from being moved
7453 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7454 PROT_READ | PROT_WRITE,
7455 MAP_SHARED | MAP_ANONYMOUS, 0);
7457 if (IS_ERR((void *)userspace_addr))
7458 return PTR_ERR((void *)userspace_addr);
7460 memslot->userspace_addr = userspace_addr;
7466 void kvm_arch_commit_memory_region(struct kvm *kvm,
7467 struct kvm_userspace_memory_region *mem,
7468 const struct kvm_memory_slot *old,
7469 enum kvm_mr_change change)
7472 int nr_mmu_pages = 0;
7474 if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7477 ret = vm_munmap(old->userspace_addr,
7478 old->npages * PAGE_SIZE);
7481 "kvm_vm_ioctl_set_memory_region: "
7482 "failed to munmap memory\n");
7485 if (!kvm->arch.n_requested_mmu_pages)
7486 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7489 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7491 * Write protect all pages for dirty logging.
7493 * All the sptes including the large sptes which point to this
7494 * slot are set to readonly. We can not create any new large
7495 * spte on this slot until the end of the logging.
7497 * See the comments in fast_page_fault().
7499 if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
7500 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7503 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7505 kvm_mmu_invalidate_zap_all_pages(kvm);
7508 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7509 struct kvm_memory_slot *slot)
7511 kvm_mmu_invalidate_zap_all_pages(kvm);
7514 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7516 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7517 kvm_x86_ops->check_nested_events(vcpu, false);
7519 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7520 !vcpu->arch.apf.halted)
7521 || !list_empty_careful(&vcpu->async_pf.done)
7522 || kvm_apic_has_events(vcpu)
7523 || vcpu->arch.pv.pv_unhalted
7524 || atomic_read(&vcpu->arch.nmi_queued) ||
7525 (kvm_arch_interrupt_allowed(vcpu) &&
7526 kvm_cpu_has_interrupt(vcpu));
7529 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7531 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7534 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7536 return kvm_x86_ops->interrupt_allowed(vcpu);
7539 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7541 unsigned long current_rip = kvm_rip_read(vcpu) +
7542 get_segment_base(vcpu, VCPU_SREG_CS);
7544 return current_rip == linear_rip;
7546 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7548 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7550 unsigned long rflags;
7552 rflags = kvm_x86_ops->get_rflags(vcpu);
7553 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7554 rflags &= ~X86_EFLAGS_TF;
7557 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7559 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7561 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7562 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7563 rflags |= X86_EFLAGS_TF;
7564 kvm_x86_ops->set_rflags(vcpu, rflags);
7567 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7569 __kvm_set_rflags(vcpu, rflags);
7570 kvm_make_request(KVM_REQ_EVENT, vcpu);
7572 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7574 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7578 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7582 r = kvm_mmu_reload(vcpu);
7586 if (!vcpu->arch.mmu.direct_map &&
7587 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7590 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7593 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7595 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7598 static inline u32 kvm_async_pf_next_probe(u32 key)
7600 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7603 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7605 u32 key = kvm_async_pf_hash_fn(gfn);
7607 while (vcpu->arch.apf.gfns[key] != ~0)
7608 key = kvm_async_pf_next_probe(key);
7610 vcpu->arch.apf.gfns[key] = gfn;
7613 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7616 u32 key = kvm_async_pf_hash_fn(gfn);
7618 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7619 (vcpu->arch.apf.gfns[key] != gfn &&
7620 vcpu->arch.apf.gfns[key] != ~0); i++)
7621 key = kvm_async_pf_next_probe(key);
7626 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7628 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7631 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7635 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7637 vcpu->arch.apf.gfns[i] = ~0;
7639 j = kvm_async_pf_next_probe(j);
7640 if (vcpu->arch.apf.gfns[j] == ~0)
7642 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7644 * k lies cyclically in ]i,j]
7646 * |....j i.k.| or |.k..j i...|
7648 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7649 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7654 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7657 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7661 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7662 struct kvm_async_pf *work)
7664 struct x86_exception fault;
7666 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7667 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7669 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7670 (vcpu->arch.apf.send_user_only &&
7671 kvm_x86_ops->get_cpl(vcpu) == 0))
7672 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7673 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7674 fault.vector = PF_VECTOR;
7675 fault.error_code_valid = true;
7676 fault.error_code = 0;
7677 fault.nested_page_fault = false;
7678 fault.address = work->arch.token;
7679 kvm_inject_page_fault(vcpu, &fault);
7683 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7684 struct kvm_async_pf *work)
7686 struct x86_exception fault;
7688 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7689 if (work->wakeup_all)
7690 work->arch.token = ~0; /* broadcast wakeup */
7692 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7694 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7695 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7696 fault.vector = PF_VECTOR;
7697 fault.error_code_valid = true;
7698 fault.error_code = 0;
7699 fault.nested_page_fault = false;
7700 fault.address = work->arch.token;
7701 kvm_inject_page_fault(vcpu, &fault);
7703 vcpu->arch.apf.halted = false;
7704 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7707 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7709 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7712 return !kvm_event_needs_reinjection(vcpu) &&
7713 kvm_x86_ops->interrupt_allowed(vcpu);
7716 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
7718 atomic_inc(&kvm->arch.noncoherent_dma_count);
7720 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
7722 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
7724 atomic_dec(&kvm->arch.noncoherent_dma_count);
7726 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
7728 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
7730 return atomic_read(&kvm->arch.noncoherent_dma_count);
7732 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
7734 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7735 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7736 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7737 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7738 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7739 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7740 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7741 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7742 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7743 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7744 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7745 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7746 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7747 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);