2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
30 #include "assigned-dev.h"
32 #include <linux/clocksource.h>
33 #include <linux/interrupt.h>
34 #include <linux/kvm.h>
36 #include <linux/vmalloc.h>
37 #include <linux/module.h>
38 #include <linux/mman.h>
39 #include <linux/highmem.h>
40 #include <linux/iommu.h>
41 #include <linux/intel-iommu.h>
42 #include <linux/cpufreq.h>
43 #include <linux/user-return-notifier.h>
44 #include <linux/srcu.h>
45 #include <linux/slab.h>
46 #include <linux/perf_event.h>
47 #include <linux/uaccess.h>
48 #include <linux/hash.h>
49 #include <linux/pci.h>
50 #include <linux/timekeeper_internal.h>
51 #include <linux/pvclock_gtod.h>
52 #include <trace/events/kvm.h>
54 #define CREATE_TRACE_POINTS
57 #include <asm/debugreg.h>
63 #include <asm/fpu-internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
68 #define MAX_IO_MSRS 256
69 #define KVM_MAX_MCE_BANKS 32
70 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
72 #define emul_to_vcpu(ctxt) \
73 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
76 * - enable syscall per default because its emulated by KVM
77 * - enable LME and LMA per default on 64 bit KVM
81 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
83 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
86 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
87 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
89 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
90 static void process_nmi(struct kvm_vcpu *vcpu);
91 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
93 struct kvm_x86_ops *kvm_x86_ops;
94 EXPORT_SYMBOL_GPL(kvm_x86_ops);
96 static bool ignore_msrs = 0;
97 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
99 unsigned int min_timer_period_us = 500;
100 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
102 static bool __read_mostly kvmclock_periodic_sync = true;
103 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
105 bool kvm_has_tsc_control;
106 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
107 u32 kvm_max_guest_tsc_khz;
108 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
110 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
111 static u32 tsc_tolerance_ppm = 250;
112 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
114 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
115 unsigned int lapic_timer_advance_ns = 0;
116 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
118 static bool backwards_tsc_observed = false;
120 #define KVM_NR_SHARED_MSRS 16
122 struct kvm_shared_msrs_global {
124 u32 msrs[KVM_NR_SHARED_MSRS];
127 struct kvm_shared_msrs {
128 struct user_return_notifier urn;
130 struct kvm_shared_msr_values {
133 } values[KVM_NR_SHARED_MSRS];
136 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
137 static struct kvm_shared_msrs __percpu *shared_msrs;
139 struct kvm_stats_debugfs_item debugfs_entries[] = {
140 { "pf_fixed", VCPU_STAT(pf_fixed) },
141 { "pf_guest", VCPU_STAT(pf_guest) },
142 { "tlb_flush", VCPU_STAT(tlb_flush) },
143 { "invlpg", VCPU_STAT(invlpg) },
144 { "exits", VCPU_STAT(exits) },
145 { "io_exits", VCPU_STAT(io_exits) },
146 { "mmio_exits", VCPU_STAT(mmio_exits) },
147 { "signal_exits", VCPU_STAT(signal_exits) },
148 { "irq_window", VCPU_STAT(irq_window_exits) },
149 { "nmi_window", VCPU_STAT(nmi_window_exits) },
150 { "halt_exits", VCPU_STAT(halt_exits) },
151 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
152 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
153 { "hypercalls", VCPU_STAT(hypercalls) },
154 { "request_irq", VCPU_STAT(request_irq_exits) },
155 { "irq_exits", VCPU_STAT(irq_exits) },
156 { "host_state_reload", VCPU_STAT(host_state_reload) },
157 { "efer_reload", VCPU_STAT(efer_reload) },
158 { "fpu_reload", VCPU_STAT(fpu_reload) },
159 { "insn_emulation", VCPU_STAT(insn_emulation) },
160 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
161 { "irq_injections", VCPU_STAT(irq_injections) },
162 { "nmi_injections", VCPU_STAT(nmi_injections) },
163 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
164 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
165 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
166 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
167 { "mmu_flooded", VM_STAT(mmu_flooded) },
168 { "mmu_recycled", VM_STAT(mmu_recycled) },
169 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
170 { "mmu_unsync", VM_STAT(mmu_unsync) },
171 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
172 { "largepages", VM_STAT(lpages) },
176 u64 __read_mostly host_xcr0;
178 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
180 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
183 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
184 vcpu->arch.apf.gfns[i] = ~0;
187 static void kvm_on_user_return(struct user_return_notifier *urn)
190 struct kvm_shared_msrs *locals
191 = container_of(urn, struct kvm_shared_msrs, urn);
192 struct kvm_shared_msr_values *values;
194 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
195 values = &locals->values[slot];
196 if (values->host != values->curr) {
197 wrmsrl(shared_msrs_global.msrs[slot], values->host);
198 values->curr = values->host;
201 locals->registered = false;
202 user_return_notifier_unregister(urn);
205 static void shared_msr_update(unsigned slot, u32 msr)
208 unsigned int cpu = smp_processor_id();
209 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
211 /* only read, and nobody should modify it at this time,
212 * so don't need lock */
213 if (slot >= shared_msrs_global.nr) {
214 printk(KERN_ERR "kvm: invalid MSR slot!");
217 rdmsrl_safe(msr, &value);
218 smsr->values[slot].host = value;
219 smsr->values[slot].curr = value;
222 void kvm_define_shared_msr(unsigned slot, u32 msr)
224 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
225 if (slot >= shared_msrs_global.nr)
226 shared_msrs_global.nr = slot + 1;
227 shared_msrs_global.msrs[slot] = msr;
228 /* we need ensured the shared_msr_global have been updated */
231 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
233 static void kvm_shared_msr_cpu_online(void)
237 for (i = 0; i < shared_msrs_global.nr; ++i)
238 shared_msr_update(i, shared_msrs_global.msrs[i]);
241 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
243 unsigned int cpu = smp_processor_id();
244 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
247 if (((value ^ smsr->values[slot].curr) & mask) == 0)
249 smsr->values[slot].curr = value;
250 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
254 if (!smsr->registered) {
255 smsr->urn.on_user_return = kvm_on_user_return;
256 user_return_notifier_register(&smsr->urn);
257 smsr->registered = true;
261 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
263 static void drop_user_return_notifiers(void)
265 unsigned int cpu = smp_processor_id();
266 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
268 if (smsr->registered)
269 kvm_on_user_return(&smsr->urn);
272 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
274 return vcpu->arch.apic_base;
276 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
278 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
280 u64 old_state = vcpu->arch.apic_base &
281 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
282 u64 new_state = msr_info->data &
283 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
284 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
285 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
287 if (!msr_info->host_initiated &&
288 ((msr_info->data & reserved_bits) != 0 ||
289 new_state == X2APIC_ENABLE ||
290 (new_state == MSR_IA32_APICBASE_ENABLE &&
291 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
292 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
296 kvm_lapic_set_base(vcpu, msr_info->data);
299 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
301 asmlinkage __visible void kvm_spurious_fault(void)
303 /* Fault while not rebooting. We want the trace. */
306 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
308 #define EXCPT_BENIGN 0
309 #define EXCPT_CONTRIBUTORY 1
312 static int exception_class(int vector)
322 return EXCPT_CONTRIBUTORY;
329 #define EXCPT_FAULT 0
331 #define EXCPT_ABORT 2
332 #define EXCPT_INTERRUPT 3
334 static int exception_type(int vector)
338 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
339 return EXCPT_INTERRUPT;
343 /* #DB is trap, as instruction watchpoints are handled elsewhere */
344 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
347 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
350 /* Reserved exceptions will result in fault */
354 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
355 unsigned nr, bool has_error, u32 error_code,
361 kvm_make_request(KVM_REQ_EVENT, vcpu);
363 if (!vcpu->arch.exception.pending) {
365 if (has_error && !is_protmode(vcpu))
367 vcpu->arch.exception.pending = true;
368 vcpu->arch.exception.has_error_code = has_error;
369 vcpu->arch.exception.nr = nr;
370 vcpu->arch.exception.error_code = error_code;
371 vcpu->arch.exception.reinject = reinject;
375 /* to check exception */
376 prev_nr = vcpu->arch.exception.nr;
377 if (prev_nr == DF_VECTOR) {
378 /* triple fault -> shutdown */
379 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
382 class1 = exception_class(prev_nr);
383 class2 = exception_class(nr);
384 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
385 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
386 /* generate double fault per SDM Table 5-5 */
387 vcpu->arch.exception.pending = true;
388 vcpu->arch.exception.has_error_code = true;
389 vcpu->arch.exception.nr = DF_VECTOR;
390 vcpu->arch.exception.error_code = 0;
392 /* replace previous exception with a new one in a hope
393 that instruction re-execution will regenerate lost
398 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
400 kvm_multiple_exception(vcpu, nr, false, 0, false);
402 EXPORT_SYMBOL_GPL(kvm_queue_exception);
404 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
406 kvm_multiple_exception(vcpu, nr, false, 0, true);
408 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
410 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
413 kvm_inject_gp(vcpu, 0);
415 kvm_x86_ops->skip_emulated_instruction(vcpu);
417 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
419 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
421 ++vcpu->stat.pf_guest;
422 vcpu->arch.cr2 = fault->address;
423 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
425 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
427 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
429 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
430 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
432 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
434 return fault->nested_page_fault;
437 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
439 atomic_inc(&vcpu->arch.nmi_queued);
440 kvm_make_request(KVM_REQ_NMI, vcpu);
442 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
444 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
446 kvm_multiple_exception(vcpu, nr, true, error_code, false);
448 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
450 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
452 kvm_multiple_exception(vcpu, nr, true, error_code, true);
454 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
457 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
458 * a #GP and return false.
460 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
462 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
464 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
467 EXPORT_SYMBOL_GPL(kvm_require_cpl);
469 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
471 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
474 kvm_queue_exception(vcpu, UD_VECTOR);
477 EXPORT_SYMBOL_GPL(kvm_require_dr);
480 * This function will be used to read from the physical memory of the currently
481 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
482 * can read from guest physical or from the guest's guest physical memory.
484 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
485 gfn_t ngfn, void *data, int offset, int len,
488 struct x86_exception exception;
492 ngpa = gfn_to_gpa(ngfn);
493 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
494 if (real_gfn == UNMAPPED_GVA)
497 real_gfn = gpa_to_gfn(real_gfn);
499 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
501 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
503 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
504 void *data, int offset, int len, u32 access)
506 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
507 data, offset, len, access);
511 * Load the pae pdptrs. Return true is they are all valid.
513 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
515 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
516 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
519 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
521 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
522 offset * sizeof(u64), sizeof(pdpte),
523 PFERR_USER_MASK|PFERR_WRITE_MASK);
528 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
529 if (is_present_gpte(pdpte[i]) &&
530 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
537 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
538 __set_bit(VCPU_EXREG_PDPTR,
539 (unsigned long *)&vcpu->arch.regs_avail);
540 __set_bit(VCPU_EXREG_PDPTR,
541 (unsigned long *)&vcpu->arch.regs_dirty);
546 EXPORT_SYMBOL_GPL(load_pdptrs);
548 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
550 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
556 if (is_long_mode(vcpu) || !is_pae(vcpu))
559 if (!test_bit(VCPU_EXREG_PDPTR,
560 (unsigned long *)&vcpu->arch.regs_avail))
563 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
564 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
565 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
566 PFERR_USER_MASK | PFERR_WRITE_MASK);
569 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
575 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
577 unsigned long old_cr0 = kvm_read_cr0(vcpu);
578 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
583 if (cr0 & 0xffffffff00000000UL)
587 cr0 &= ~CR0_RESERVED_BITS;
589 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
592 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
595 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
597 if ((vcpu->arch.efer & EFER_LME)) {
602 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
607 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
612 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
615 kvm_x86_ops->set_cr0(vcpu, cr0);
617 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
618 kvm_clear_async_pf_completion_queue(vcpu);
619 kvm_async_pf_hash_reset(vcpu);
622 if ((cr0 ^ old_cr0) & update_bits)
623 kvm_mmu_reset_context(vcpu);
626 EXPORT_SYMBOL_GPL(kvm_set_cr0);
628 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
630 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
632 EXPORT_SYMBOL_GPL(kvm_lmsw);
634 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
636 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
637 !vcpu->guest_xcr0_loaded) {
638 /* kvm_set_xcr() also depends on this */
639 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
640 vcpu->guest_xcr0_loaded = 1;
644 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
646 if (vcpu->guest_xcr0_loaded) {
647 if (vcpu->arch.xcr0 != host_xcr0)
648 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
649 vcpu->guest_xcr0_loaded = 0;
653 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
656 u64 old_xcr0 = vcpu->arch.xcr0;
659 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
660 if (index != XCR_XFEATURE_ENABLED_MASK)
662 if (!(xcr0 & XSTATE_FP))
664 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
668 * Do not allow the guest to set bits that we do not support
669 * saving. However, xcr0 bit 0 is always set, even if the
670 * emulated CPU does not support XSAVE (see fx_init).
672 valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
673 if (xcr0 & ~valid_bits)
676 if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
679 if (xcr0 & XSTATE_AVX512) {
680 if (!(xcr0 & XSTATE_YMM))
682 if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512)
685 kvm_put_guest_xcr0(vcpu);
686 vcpu->arch.xcr0 = xcr0;
688 if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
689 kvm_update_cpuid(vcpu);
693 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
695 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
696 __kvm_set_xcr(vcpu, index, xcr)) {
697 kvm_inject_gp(vcpu, 0);
702 EXPORT_SYMBOL_GPL(kvm_set_xcr);
704 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
706 unsigned long old_cr4 = kvm_read_cr4(vcpu);
707 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
708 X86_CR4_SMEP | X86_CR4_SMAP;
710 if (cr4 & CR4_RESERVED_BITS)
713 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
716 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
719 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
722 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
725 if (is_long_mode(vcpu)) {
726 if (!(cr4 & X86_CR4_PAE))
728 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
729 && ((cr4 ^ old_cr4) & pdptr_bits)
730 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
734 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
735 if (!guest_cpuid_has_pcid(vcpu))
738 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
739 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
743 if (kvm_x86_ops->set_cr4(vcpu, cr4))
746 if (((cr4 ^ old_cr4) & pdptr_bits) ||
747 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
748 kvm_mmu_reset_context(vcpu);
750 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
751 kvm_update_cpuid(vcpu);
755 EXPORT_SYMBOL_GPL(kvm_set_cr4);
757 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
760 cr3 &= ~CR3_PCID_INVD;
763 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
764 kvm_mmu_sync_roots(vcpu);
765 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
769 if (is_long_mode(vcpu)) {
770 if (cr3 & CR3_L_MODE_RESERVED_BITS)
772 } else if (is_pae(vcpu) && is_paging(vcpu) &&
773 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
776 vcpu->arch.cr3 = cr3;
777 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
778 kvm_mmu_new_cr3(vcpu);
781 EXPORT_SYMBOL_GPL(kvm_set_cr3);
783 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
785 if (cr8 & CR8_RESERVED_BITS)
787 if (irqchip_in_kernel(vcpu->kvm))
788 kvm_lapic_set_tpr(vcpu, cr8);
790 vcpu->arch.cr8 = cr8;
793 EXPORT_SYMBOL_GPL(kvm_set_cr8);
795 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
797 if (irqchip_in_kernel(vcpu->kvm))
798 return kvm_lapic_get_cr8(vcpu);
800 return vcpu->arch.cr8;
802 EXPORT_SYMBOL_GPL(kvm_get_cr8);
804 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
808 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
809 for (i = 0; i < KVM_NR_DB_REGS; i++)
810 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
811 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
815 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
817 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
818 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
821 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
825 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
826 dr7 = vcpu->arch.guest_debug_dr7;
828 dr7 = vcpu->arch.dr7;
829 kvm_x86_ops->set_dr7(vcpu, dr7);
830 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
831 if (dr7 & DR7_BP_EN_MASK)
832 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
835 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
837 u64 fixed = DR6_FIXED_1;
839 if (!guest_cpuid_has_rtm(vcpu))
844 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
848 vcpu->arch.db[dr] = val;
849 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
850 vcpu->arch.eff_db[dr] = val;
855 if (val & 0xffffffff00000000ULL)
857 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
858 kvm_update_dr6(vcpu);
863 if (val & 0xffffffff00000000ULL)
865 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
866 kvm_update_dr7(vcpu);
873 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
875 if (__kvm_set_dr(vcpu, dr, val)) {
876 kvm_inject_gp(vcpu, 0);
881 EXPORT_SYMBOL_GPL(kvm_set_dr);
883 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
887 *val = vcpu->arch.db[dr];
892 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
893 *val = vcpu->arch.dr6;
895 *val = kvm_x86_ops->get_dr6(vcpu);
900 *val = vcpu->arch.dr7;
905 EXPORT_SYMBOL_GPL(kvm_get_dr);
907 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
909 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
913 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
916 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
917 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
920 EXPORT_SYMBOL_GPL(kvm_rdpmc);
923 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
924 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
926 * This list is modified at module load time to reflect the
927 * capabilities of the host cpu. This capabilities test skips MSRs that are
928 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
929 * may depend on host virtualization features rather than host cpu features.
932 static u32 msrs_to_save[] = {
933 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
936 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
938 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
939 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
942 static unsigned num_msrs_to_save;
944 static u32 emulated_msrs[] = {
945 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
946 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
947 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
948 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
949 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
953 MSR_IA32_TSCDEADLINE,
954 MSR_IA32_MISC_ENABLE,
960 static unsigned num_emulated_msrs;
962 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
964 if (efer & efer_reserved_bits)
967 if (efer & EFER_FFXSR) {
968 struct kvm_cpuid_entry2 *feat;
970 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
971 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
975 if (efer & EFER_SVME) {
976 struct kvm_cpuid_entry2 *feat;
978 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
979 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
985 EXPORT_SYMBOL_GPL(kvm_valid_efer);
987 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
989 u64 old_efer = vcpu->arch.efer;
991 if (!kvm_valid_efer(vcpu, efer))
995 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
999 efer |= vcpu->arch.efer & EFER_LMA;
1001 kvm_x86_ops->set_efer(vcpu, efer);
1003 /* Update reserved bits */
1004 if ((efer ^ old_efer) & EFER_NX)
1005 kvm_mmu_reset_context(vcpu);
1010 void kvm_enable_efer_bits(u64 mask)
1012 efer_reserved_bits &= ~mask;
1014 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1017 * Writes msr value into into the appropriate "register".
1018 * Returns 0 on success, non-0 otherwise.
1019 * Assumes vcpu_load() was already called.
1021 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1023 switch (msr->index) {
1026 case MSR_KERNEL_GS_BASE:
1029 if (is_noncanonical_address(msr->data))
1032 case MSR_IA32_SYSENTER_EIP:
1033 case MSR_IA32_SYSENTER_ESP:
1035 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1036 * non-canonical address is written on Intel but not on
1037 * AMD (which ignores the top 32-bits, because it does
1038 * not implement 64-bit SYSENTER).
1040 * 64-bit code should hence be able to write a non-canonical
1041 * value on AMD. Making the address canonical ensures that
1042 * vmentry does not fail on Intel after writing a non-canonical
1043 * value, and that something deterministic happens if the guest
1044 * invokes 64-bit SYSENTER.
1046 msr->data = get_canonical(msr->data);
1048 return kvm_x86_ops->set_msr(vcpu, msr);
1050 EXPORT_SYMBOL_GPL(kvm_set_msr);
1053 * Adapt set_msr() to msr_io()'s calling convention
1055 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1057 struct msr_data msr;
1061 msr.host_initiated = true;
1062 r = kvm_get_msr(vcpu, &msr);
1070 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1072 struct msr_data msr;
1076 msr.host_initiated = true;
1077 return kvm_set_msr(vcpu, &msr);
1080 #ifdef CONFIG_X86_64
1081 struct pvclock_gtod_data {
1084 struct { /* extract of a clocksource struct */
1096 static struct pvclock_gtod_data pvclock_gtod_data;
1098 static void update_pvclock_gtod(struct timekeeper *tk)
1100 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1103 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1105 write_seqcount_begin(&vdata->seq);
1107 /* copy pvclock gtod data */
1108 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1109 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1110 vdata->clock.mask = tk->tkr_mono.mask;
1111 vdata->clock.mult = tk->tkr_mono.mult;
1112 vdata->clock.shift = tk->tkr_mono.shift;
1114 vdata->boot_ns = boot_ns;
1115 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1117 write_seqcount_end(&vdata->seq);
1121 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1124 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1125 * vcpu_enter_guest. This function is only called from
1126 * the physical CPU that is running vcpu.
1128 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1131 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1135 struct pvclock_wall_clock wc;
1136 struct timespec boot;
1141 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1146 ++version; /* first time write, random junk */
1150 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1153 * The guest calculates current wall clock time by adding
1154 * system time (updated by kvm_guest_time_update below) to the
1155 * wall clock specified here. guest system time equals host
1156 * system time for us, thus we must fill in host boot time here.
1160 if (kvm->arch.kvmclock_offset) {
1161 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1162 boot = timespec_sub(boot, ts);
1164 wc.sec = boot.tv_sec;
1165 wc.nsec = boot.tv_nsec;
1166 wc.version = version;
1168 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1171 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1174 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1176 uint32_t quotient, remainder;
1178 /* Don't try to replace with do_div(), this one calculates
1179 * "(dividend << 32) / divisor" */
1181 : "=a" (quotient), "=d" (remainder)
1182 : "0" (0), "1" (dividend), "r" (divisor) );
1186 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1187 s8 *pshift, u32 *pmultiplier)
1194 tps64 = base_khz * 1000LL;
1195 scaled64 = scaled_khz * 1000LL;
1196 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1201 tps32 = (uint32_t)tps64;
1202 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1203 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1211 *pmultiplier = div_frac(scaled64, tps32);
1213 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1214 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1217 static inline u64 get_kernel_ns(void)
1219 return ktime_get_boot_ns();
1222 #ifdef CONFIG_X86_64
1223 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1226 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1227 static unsigned long max_tsc_khz;
1229 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1231 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1232 vcpu->arch.virtual_tsc_shift);
1235 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1237 u64 v = (u64)khz * (1000000 + ppm);
1242 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1244 u32 thresh_lo, thresh_hi;
1245 int use_scaling = 0;
1247 /* tsc_khz can be zero if TSC calibration fails */
1248 if (this_tsc_khz == 0)
1251 /* Compute a scale to convert nanoseconds in TSC cycles */
1252 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1253 &vcpu->arch.virtual_tsc_shift,
1254 &vcpu->arch.virtual_tsc_mult);
1255 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1258 * Compute the variation in TSC rate which is acceptable
1259 * within the range of tolerance and decide if the
1260 * rate being applied is within that bounds of the hardware
1261 * rate. If so, no scaling or compensation need be done.
1263 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1264 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1265 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1266 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1269 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1272 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1274 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1275 vcpu->arch.virtual_tsc_mult,
1276 vcpu->arch.virtual_tsc_shift);
1277 tsc += vcpu->arch.this_tsc_write;
1281 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1283 #ifdef CONFIG_X86_64
1285 struct kvm_arch *ka = &vcpu->kvm->arch;
1286 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1288 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1289 atomic_read(&vcpu->kvm->online_vcpus));
1292 * Once the masterclock is enabled, always perform request in
1293 * order to update it.
1295 * In order to enable masterclock, the host clocksource must be TSC
1296 * and the vcpus need to have matched TSCs. When that happens,
1297 * perform request to enable masterclock.
1299 if (ka->use_master_clock ||
1300 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1301 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1303 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1304 atomic_read(&vcpu->kvm->online_vcpus),
1305 ka->use_master_clock, gtod->clock.vclock_mode);
1309 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1311 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1312 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1315 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1317 struct kvm *kvm = vcpu->kvm;
1318 u64 offset, ns, elapsed;
1319 unsigned long flags;
1322 bool already_matched;
1323 u64 data = msr->data;
1325 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1326 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1327 ns = get_kernel_ns();
1328 elapsed = ns - kvm->arch.last_tsc_nsec;
1330 if (vcpu->arch.virtual_tsc_khz) {
1333 /* n.b - signed multiplication and division required */
1334 usdiff = data - kvm->arch.last_tsc_write;
1335 #ifdef CONFIG_X86_64
1336 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1338 /* do_div() only does unsigned */
1339 asm("1: idivl %[divisor]\n"
1340 "2: xor %%edx, %%edx\n"
1341 " movl $0, %[faulted]\n"
1343 ".section .fixup,\"ax\"\n"
1344 "4: movl $1, %[faulted]\n"
1348 _ASM_EXTABLE(1b, 4b)
1350 : "=A"(usdiff), [faulted] "=r" (faulted)
1351 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1354 do_div(elapsed, 1000);
1359 /* idivl overflow => difference is larger than USEC_PER_SEC */
1361 usdiff = USEC_PER_SEC;
1363 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1366 * Special case: TSC write with a small delta (1 second) of virtual
1367 * cycle time against real time is interpreted as an attempt to
1368 * synchronize the CPU.
1370 * For a reliable TSC, we can match TSC offsets, and for an unstable
1371 * TSC, we add elapsed time in this computation. We could let the
1372 * compensation code attempt to catch up if we fall behind, but
1373 * it's better to try to match offsets from the beginning.
1375 if (usdiff < USEC_PER_SEC &&
1376 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1377 if (!check_tsc_unstable()) {
1378 offset = kvm->arch.cur_tsc_offset;
1379 pr_debug("kvm: matched tsc offset for %llu\n", data);
1381 u64 delta = nsec_to_cycles(vcpu, elapsed);
1383 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1384 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1387 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1390 * We split periods of matched TSC writes into generations.
1391 * For each generation, we track the original measured
1392 * nanosecond time, offset, and write, so if TSCs are in
1393 * sync, we can match exact offset, and if not, we can match
1394 * exact software computation in compute_guest_tsc()
1396 * These values are tracked in kvm->arch.cur_xxx variables.
1398 kvm->arch.cur_tsc_generation++;
1399 kvm->arch.cur_tsc_nsec = ns;
1400 kvm->arch.cur_tsc_write = data;
1401 kvm->arch.cur_tsc_offset = offset;
1403 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1404 kvm->arch.cur_tsc_generation, data);
1408 * We also track th most recent recorded KHZ, write and time to
1409 * allow the matching interval to be extended at each write.
1411 kvm->arch.last_tsc_nsec = ns;
1412 kvm->arch.last_tsc_write = data;
1413 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1415 vcpu->arch.last_guest_tsc = data;
1417 /* Keep track of which generation this VCPU has synchronized to */
1418 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1419 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1420 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1422 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1423 update_ia32_tsc_adjust_msr(vcpu, offset);
1424 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1425 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1427 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1429 kvm->arch.nr_vcpus_matched_tsc = 0;
1430 } else if (!already_matched) {
1431 kvm->arch.nr_vcpus_matched_tsc++;
1434 kvm_track_tsc_matching(vcpu);
1435 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1438 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1440 #ifdef CONFIG_X86_64
1442 static cycle_t read_tsc(void)
1448 * Empirically, a fence (of type that depends on the CPU)
1449 * before rdtsc is enough to ensure that rdtsc is ordered
1450 * with respect to loads. The various CPU manuals are unclear
1451 * as to whether rdtsc can be reordered with later loads,
1452 * but no one has ever seen it happen.
1455 ret = (cycle_t)vget_cycles();
1457 last = pvclock_gtod_data.clock.cycle_last;
1459 if (likely(ret >= last))
1463 * GCC likes to generate cmov here, but this branch is extremely
1464 * predictable (it's just a funciton of time and the likely is
1465 * very likely) and there's a data dependence, so force GCC
1466 * to generate a branch instead. I don't barrier() because
1467 * we don't actually need a barrier, and if this function
1468 * ever gets inlined it will generate worse code.
1474 static inline u64 vgettsc(cycle_t *cycle_now)
1477 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1479 *cycle_now = read_tsc();
1481 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1482 return v * gtod->clock.mult;
1485 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1487 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1493 seq = read_seqcount_begin(>od->seq);
1494 mode = gtod->clock.vclock_mode;
1495 ns = gtod->nsec_base;
1496 ns += vgettsc(cycle_now);
1497 ns >>= gtod->clock.shift;
1498 ns += gtod->boot_ns;
1499 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1505 /* returns true if host is using tsc clocksource */
1506 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1508 /* checked again under seqlock below */
1509 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1512 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1518 * Assuming a stable TSC across physical CPUS, and a stable TSC
1519 * across virtual CPUs, the following condition is possible.
1520 * Each numbered line represents an event visible to both
1521 * CPUs at the next numbered event.
1523 * "timespecX" represents host monotonic time. "tscX" represents
1526 * VCPU0 on CPU0 | VCPU1 on CPU1
1528 * 1. read timespec0,tsc0
1529 * 2. | timespec1 = timespec0 + N
1531 * 3. transition to guest | transition to guest
1532 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1533 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1534 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1536 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1539 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1541 * - 0 < N - M => M < N
1543 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1544 * always the case (the difference between two distinct xtime instances
1545 * might be smaller then the difference between corresponding TSC reads,
1546 * when updating guest vcpus pvclock areas).
1548 * To avoid that problem, do not allow visibility of distinct
1549 * system_timestamp/tsc_timestamp values simultaneously: use a master
1550 * copy of host monotonic time values. Update that master copy
1553 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1557 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1559 #ifdef CONFIG_X86_64
1560 struct kvm_arch *ka = &kvm->arch;
1562 bool host_tsc_clocksource, vcpus_matched;
1564 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1565 atomic_read(&kvm->online_vcpus));
1568 * If the host uses TSC clock, then passthrough TSC as stable
1571 host_tsc_clocksource = kvm_get_time_and_clockread(
1572 &ka->master_kernel_ns,
1573 &ka->master_cycle_now);
1575 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1576 && !backwards_tsc_observed
1577 && !ka->boot_vcpu_runs_old_kvmclock;
1579 if (ka->use_master_clock)
1580 atomic_set(&kvm_guest_has_master_clock, 1);
1582 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1583 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1588 static void kvm_gen_update_masterclock(struct kvm *kvm)
1590 #ifdef CONFIG_X86_64
1592 struct kvm_vcpu *vcpu;
1593 struct kvm_arch *ka = &kvm->arch;
1595 spin_lock(&ka->pvclock_gtod_sync_lock);
1596 kvm_make_mclock_inprogress_request(kvm);
1597 /* no guest entries from this point */
1598 pvclock_update_vm_gtod_copy(kvm);
1600 kvm_for_each_vcpu(i, vcpu, kvm)
1601 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1603 /* guest entries allowed */
1604 kvm_for_each_vcpu(i, vcpu, kvm)
1605 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1607 spin_unlock(&ka->pvclock_gtod_sync_lock);
1611 static int kvm_guest_time_update(struct kvm_vcpu *v)
1613 unsigned long flags, this_tsc_khz;
1614 struct kvm_vcpu_arch *vcpu = &v->arch;
1615 struct kvm_arch *ka = &v->kvm->arch;
1617 u64 tsc_timestamp, host_tsc;
1618 struct pvclock_vcpu_time_info guest_hv_clock;
1620 bool use_master_clock;
1626 * If the host uses TSC clock, then passthrough TSC as stable
1629 spin_lock(&ka->pvclock_gtod_sync_lock);
1630 use_master_clock = ka->use_master_clock;
1631 if (use_master_clock) {
1632 host_tsc = ka->master_cycle_now;
1633 kernel_ns = ka->master_kernel_ns;
1635 spin_unlock(&ka->pvclock_gtod_sync_lock);
1637 /* Keep irq disabled to prevent changes to the clock */
1638 local_irq_save(flags);
1639 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1640 if (unlikely(this_tsc_khz == 0)) {
1641 local_irq_restore(flags);
1642 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1645 if (!use_master_clock) {
1646 host_tsc = native_read_tsc();
1647 kernel_ns = get_kernel_ns();
1650 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1653 * We may have to catch up the TSC to match elapsed wall clock
1654 * time for two reasons, even if kvmclock is used.
1655 * 1) CPU could have been running below the maximum TSC rate
1656 * 2) Broken TSC compensation resets the base at each VCPU
1657 * entry to avoid unknown leaps of TSC even when running
1658 * again on the same CPU. This may cause apparent elapsed
1659 * time to disappear, and the guest to stand still or run
1662 if (vcpu->tsc_catchup) {
1663 u64 tsc = compute_guest_tsc(v, kernel_ns);
1664 if (tsc > tsc_timestamp) {
1665 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1666 tsc_timestamp = tsc;
1670 local_irq_restore(flags);
1672 if (!vcpu->pv_time_enabled)
1675 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1676 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1677 &vcpu->hv_clock.tsc_shift,
1678 &vcpu->hv_clock.tsc_to_system_mul);
1679 vcpu->hw_tsc_khz = this_tsc_khz;
1682 /* With all the info we got, fill in the values */
1683 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1684 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1685 vcpu->last_guest_tsc = tsc_timestamp;
1687 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1688 &guest_hv_clock, sizeof(guest_hv_clock))))
1691 /* This VCPU is paused, but it's legal for a guest to read another
1692 * VCPU's kvmclock, so we really have to follow the specification where
1693 * it says that version is odd if data is being modified, and even after
1696 * Version field updates must be kept separate. This is because
1697 * kvm_write_guest_cached might use a "rep movs" instruction, and
1698 * writes within a string instruction are weakly ordered. So there
1699 * are three writes overall.
1701 * As a small optimization, only write the version field in the first
1702 * and third write. The vcpu->pv_time cache is still valid, because the
1703 * version field is the first in the struct.
1705 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1707 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1708 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1710 sizeof(vcpu->hv_clock.version));
1714 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1715 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1717 if (vcpu->pvclock_set_guest_stopped_request) {
1718 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1719 vcpu->pvclock_set_guest_stopped_request = false;
1722 pvclock_flags |= PVCLOCK_COUNTS_FROM_ZERO;
1724 /* If the host uses TSC clocksource, then it is stable */
1725 if (use_master_clock)
1726 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1728 vcpu->hv_clock.flags = pvclock_flags;
1730 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1732 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1734 sizeof(vcpu->hv_clock));
1738 vcpu->hv_clock.version++;
1739 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1741 sizeof(vcpu->hv_clock.version));
1746 * kvmclock updates which are isolated to a given vcpu, such as
1747 * vcpu->cpu migration, should not allow system_timestamp from
1748 * the rest of the vcpus to remain static. Otherwise ntp frequency
1749 * correction applies to one vcpu's system_timestamp but not
1752 * So in those cases, request a kvmclock update for all vcpus.
1753 * We need to rate-limit these requests though, as they can
1754 * considerably slow guests that have a large number of vcpus.
1755 * The time for a remote vcpu to update its kvmclock is bound
1756 * by the delay we use to rate-limit the updates.
1759 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1761 static void kvmclock_update_fn(struct work_struct *work)
1764 struct delayed_work *dwork = to_delayed_work(work);
1765 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1766 kvmclock_update_work);
1767 struct kvm *kvm = container_of(ka, struct kvm, arch);
1768 struct kvm_vcpu *vcpu;
1770 kvm_for_each_vcpu(i, vcpu, kvm) {
1771 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1772 kvm_vcpu_kick(vcpu);
1776 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1778 struct kvm *kvm = v->kvm;
1780 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1781 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1782 KVMCLOCK_UPDATE_DELAY);
1785 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1787 static void kvmclock_sync_fn(struct work_struct *work)
1789 struct delayed_work *dwork = to_delayed_work(work);
1790 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1791 kvmclock_sync_work);
1792 struct kvm *kvm = container_of(ka, struct kvm, arch);
1794 if (!kvmclock_periodic_sync)
1797 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1798 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1799 KVMCLOCK_SYNC_PERIOD);
1802 static bool msr_mtrr_valid(unsigned msr)
1805 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1806 case MSR_MTRRfix64K_00000:
1807 case MSR_MTRRfix16K_80000:
1808 case MSR_MTRRfix16K_A0000:
1809 case MSR_MTRRfix4K_C0000:
1810 case MSR_MTRRfix4K_C8000:
1811 case MSR_MTRRfix4K_D0000:
1812 case MSR_MTRRfix4K_D8000:
1813 case MSR_MTRRfix4K_E0000:
1814 case MSR_MTRRfix4K_E8000:
1815 case MSR_MTRRfix4K_F0000:
1816 case MSR_MTRRfix4K_F8000:
1817 case MSR_MTRRdefType:
1818 case MSR_IA32_CR_PAT:
1826 static bool valid_pat_type(unsigned t)
1828 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1831 static bool valid_mtrr_type(unsigned t)
1833 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1836 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1841 if (!msr_mtrr_valid(msr))
1844 if (msr == MSR_IA32_CR_PAT) {
1845 for (i = 0; i < 8; i++)
1846 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1849 } else if (msr == MSR_MTRRdefType) {
1852 return valid_mtrr_type(data & 0xff);
1853 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1854 for (i = 0; i < 8 ; i++)
1855 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1860 /* variable MTRRs */
1861 WARN_ON(!(msr >= 0x200 && msr < 0x200 + 2 * KVM_NR_VAR_MTRR));
1863 mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
1864 if ((msr & 1) == 0) {
1866 if (!valid_mtrr_type(data & 0xff))
1873 kvm_inject_gp(vcpu, 0);
1879 EXPORT_SYMBOL_GPL(kvm_mtrr_valid);
1881 static void update_mtrr(struct kvm_vcpu *vcpu, u32 msr)
1883 struct mtrr_state_type *mtrr_state = &vcpu->arch.mtrr_state;
1884 unsigned char mtrr_enabled = mtrr_state->enabled;
1885 gfn_t start, end, mask;
1887 bool is_fixed = true;
1889 if (msr == MSR_IA32_CR_PAT || !tdp_enabled ||
1890 !kvm_arch_has_noncoherent_dma(vcpu->kvm))
1893 if (!(mtrr_enabled & 0x2) && msr != MSR_MTRRdefType)
1897 case MSR_MTRRfix64K_00000:
1901 case MSR_MTRRfix16K_80000:
1905 case MSR_MTRRfix16K_A0000:
1909 case MSR_MTRRfix4K_C0000 ... MSR_MTRRfix4K_F8000:
1910 index = msr - MSR_MTRRfix4K_C0000;
1911 start = 0xc0000 + index * (32 << 10);
1912 end = start + (32 << 10);
1914 case MSR_MTRRdefType:
1920 /* variable range MTRRs. */
1922 index = (msr - 0x200) / 2;
1923 start = (((u64)mtrr_state->var_ranges[index].base_hi) << 32) +
1924 (mtrr_state->var_ranges[index].base_lo & PAGE_MASK);
1925 mask = (((u64)mtrr_state->var_ranges[index].mask_hi) << 32) +
1926 (mtrr_state->var_ranges[index].mask_lo & PAGE_MASK);
1927 mask |= ~0ULL << cpuid_maxphyaddr(vcpu);
1929 end = ((start & mask) | ~mask) + 1;
1932 if (is_fixed && !(mtrr_enabled & 0x1))
1935 kvm_zap_gfn_range(vcpu->kvm, gpa_to_gfn(start), gpa_to_gfn(end));
1938 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1940 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1942 if (!kvm_mtrr_valid(vcpu, msr, data))
1945 if (msr == MSR_MTRRdefType) {
1946 vcpu->arch.mtrr_state.def_type = data;
1947 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1948 } else if (msr == MSR_MTRRfix64K_00000)
1950 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1951 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1952 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1953 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1954 else if (msr == MSR_IA32_CR_PAT)
1955 vcpu->arch.pat = data;
1956 else { /* Variable MTRRs */
1957 int idx, is_mtrr_mask;
1960 idx = (msr - 0x200) / 2;
1961 is_mtrr_mask = msr - 0x200 - 2 * idx;
1964 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1967 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1971 update_mtrr(vcpu, msr);
1975 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1977 u64 mcg_cap = vcpu->arch.mcg_cap;
1978 unsigned bank_num = mcg_cap & 0xff;
1981 case MSR_IA32_MCG_STATUS:
1982 vcpu->arch.mcg_status = data;
1984 case MSR_IA32_MCG_CTL:
1985 if (!(mcg_cap & MCG_CTL_P))
1987 if (data != 0 && data != ~(u64)0)
1989 vcpu->arch.mcg_ctl = data;
1992 if (msr >= MSR_IA32_MC0_CTL &&
1993 msr < MSR_IA32_MCx_CTL(bank_num)) {
1994 u32 offset = msr - MSR_IA32_MC0_CTL;
1995 /* only 0 or all 1s can be written to IA32_MCi_CTL
1996 * some Linux kernels though clear bit 10 in bank 4 to
1997 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1998 * this to avoid an uncatched #GP in the guest
2000 if ((offset & 0x3) == 0 &&
2001 data != 0 && (data | (1 << 10)) != ~(u64)0)
2003 vcpu->arch.mce_banks[offset] = data;
2011 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2013 struct kvm *kvm = vcpu->kvm;
2014 int lm = is_long_mode(vcpu);
2015 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2016 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2017 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2018 : kvm->arch.xen_hvm_config.blob_size_32;
2019 u32 page_num = data & ~PAGE_MASK;
2020 u64 page_addr = data & PAGE_MASK;
2025 if (page_num >= blob_size)
2028 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2033 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2042 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
2044 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
2047 static bool kvm_hv_msr_partition_wide(u32 msr)
2051 case HV_X64_MSR_GUEST_OS_ID:
2052 case HV_X64_MSR_HYPERCALL:
2053 case HV_X64_MSR_REFERENCE_TSC:
2054 case HV_X64_MSR_TIME_REF_COUNT:
2062 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2064 struct kvm *kvm = vcpu->kvm;
2067 case HV_X64_MSR_GUEST_OS_ID:
2068 kvm->arch.hv_guest_os_id = data;
2069 /* setting guest os id to zero disables hypercall page */
2070 if (!kvm->arch.hv_guest_os_id)
2071 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
2073 case HV_X64_MSR_HYPERCALL: {
2078 /* if guest os id is not set hypercall should remain disabled */
2079 if (!kvm->arch.hv_guest_os_id)
2081 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
2082 kvm->arch.hv_hypercall = data;
2085 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
2086 addr = gfn_to_hva(kvm, gfn);
2087 if (kvm_is_error_hva(addr))
2089 kvm_x86_ops->patch_hypercall(vcpu, instructions);
2090 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
2091 if (__copy_to_user((void __user *)addr, instructions, 4))
2093 kvm->arch.hv_hypercall = data;
2094 mark_page_dirty(kvm, gfn);
2097 case HV_X64_MSR_REFERENCE_TSC: {
2099 HV_REFERENCE_TSC_PAGE tsc_ref;
2100 memset(&tsc_ref, 0, sizeof(tsc_ref));
2101 kvm->arch.hv_tsc_page = data;
2102 if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
2104 gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
2105 if (kvm_write_guest(kvm, gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
2106 &tsc_ref, sizeof(tsc_ref)))
2108 mark_page_dirty(kvm, gfn);
2112 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2113 "data 0x%llx\n", msr, data);
2119 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2122 case HV_X64_MSR_APIC_ASSIST_PAGE: {
2126 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
2127 vcpu->arch.hv_vapic = data;
2128 if (kvm_lapic_enable_pv_eoi(vcpu, 0))
2132 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
2133 addr = kvm_vcpu_gfn_to_hva(vcpu, gfn);
2134 if (kvm_is_error_hva(addr))
2136 if (__clear_user((void __user *)addr, PAGE_SIZE))
2138 vcpu->arch.hv_vapic = data;
2139 kvm_vcpu_mark_page_dirty(vcpu, gfn);
2140 if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
2144 case HV_X64_MSR_EOI:
2145 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
2146 case HV_X64_MSR_ICR:
2147 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
2148 case HV_X64_MSR_TPR:
2149 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
2151 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2152 "data 0x%llx\n", msr, data);
2159 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2161 gpa_t gpa = data & ~0x3f;
2163 /* Bits 2:5 are reserved, Should be zero */
2167 vcpu->arch.apf.msr_val = data;
2169 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2170 kvm_clear_async_pf_completion_queue(vcpu);
2171 kvm_async_pf_hash_reset(vcpu);
2175 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2179 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2180 kvm_async_pf_wakeup_all(vcpu);
2184 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2186 vcpu->arch.pv_time_enabled = false;
2189 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2193 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2196 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2197 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2198 vcpu->arch.st.accum_steal = delta;
2201 static void record_steal_time(struct kvm_vcpu *vcpu)
2203 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2206 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2207 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2210 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2211 vcpu->arch.st.steal.version += 2;
2212 vcpu->arch.st.accum_steal = 0;
2214 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2215 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2218 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2221 u32 msr = msr_info->index;
2222 u64 data = msr_info->data;
2225 case MSR_AMD64_NB_CFG:
2226 case MSR_IA32_UCODE_REV:
2227 case MSR_IA32_UCODE_WRITE:
2228 case MSR_VM_HSAVE_PA:
2229 case MSR_AMD64_PATCH_LOADER:
2230 case MSR_AMD64_BU_CFG2:
2234 return set_efer(vcpu, data);
2236 data &= ~(u64)0x40; /* ignore flush filter disable */
2237 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2238 data &= ~(u64)0x8; /* ignore TLB cache disable */
2239 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2241 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2246 case MSR_FAM10H_MMIO_CONF_BASE:
2248 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2253 case MSR_IA32_DEBUGCTLMSR:
2255 /* We support the non-activated case already */
2257 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2258 /* Values other than LBR and BTF are vendor-specific,
2259 thus reserved and should throw a #GP */
2262 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2265 case 0x200 ... 0x2ff:
2266 return set_msr_mtrr(vcpu, msr, data);
2267 case MSR_IA32_APICBASE:
2268 return kvm_set_apic_base(vcpu, msr_info);
2269 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2270 return kvm_x2apic_msr_write(vcpu, msr, data);
2271 case MSR_IA32_TSCDEADLINE:
2272 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2274 case MSR_IA32_TSC_ADJUST:
2275 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2276 if (!msr_info->host_initiated) {
2277 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2278 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2280 vcpu->arch.ia32_tsc_adjust_msr = data;
2283 case MSR_IA32_MISC_ENABLE:
2284 vcpu->arch.ia32_misc_enable_msr = data;
2286 case MSR_IA32_SMBASE:
2287 if (!msr_info->host_initiated)
2289 vcpu->arch.smbase = data;
2291 case MSR_KVM_WALL_CLOCK_NEW:
2292 case MSR_KVM_WALL_CLOCK:
2293 vcpu->kvm->arch.wall_clock = data;
2294 kvm_write_wall_clock(vcpu->kvm, data);
2296 case MSR_KVM_SYSTEM_TIME_NEW:
2297 case MSR_KVM_SYSTEM_TIME: {
2299 struct kvm_arch *ka = &vcpu->kvm->arch;
2301 kvmclock_reset(vcpu);
2303 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2304 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2306 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2307 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2310 ka->boot_vcpu_runs_old_kvmclock = tmp;
2312 ka->kvmclock_offset = -get_kernel_ns();
2315 vcpu->arch.time = data;
2316 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2318 /* we verify if the enable bit is set... */
2322 gpa_offset = data & ~(PAGE_MASK | 1);
2324 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2325 &vcpu->arch.pv_time, data & ~1ULL,
2326 sizeof(struct pvclock_vcpu_time_info)))
2327 vcpu->arch.pv_time_enabled = false;
2329 vcpu->arch.pv_time_enabled = true;
2333 case MSR_KVM_ASYNC_PF_EN:
2334 if (kvm_pv_enable_async_pf(vcpu, data))
2337 case MSR_KVM_STEAL_TIME:
2339 if (unlikely(!sched_info_on()))
2342 if (data & KVM_STEAL_RESERVED_MASK)
2345 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2346 data & KVM_STEAL_VALID_BITS,
2347 sizeof(struct kvm_steal_time)))
2350 vcpu->arch.st.msr_val = data;
2352 if (!(data & KVM_MSR_ENABLED))
2355 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2358 accumulate_steal_time(vcpu);
2361 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2364 case MSR_KVM_PV_EOI_EN:
2365 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2369 case MSR_IA32_MCG_CTL:
2370 case MSR_IA32_MCG_STATUS:
2371 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2372 return set_msr_mce(vcpu, msr, data);
2374 /* Performance counters are not protected by a CPUID bit,
2375 * so we should check all of them in the generic path for the sake of
2376 * cross vendor migration.
2377 * Writing a zero into the event select MSRs disables them,
2378 * which we perfectly emulate ;-). Any other value should be at least
2379 * reported, some guests depend on them.
2381 case MSR_K7_EVNTSEL0:
2382 case MSR_K7_EVNTSEL1:
2383 case MSR_K7_EVNTSEL2:
2384 case MSR_K7_EVNTSEL3:
2386 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2387 "0x%x data 0x%llx\n", msr, data);
2389 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2390 * so we ignore writes to make it happy.
2392 case MSR_K7_PERFCTR0:
2393 case MSR_K7_PERFCTR1:
2394 case MSR_K7_PERFCTR2:
2395 case MSR_K7_PERFCTR3:
2396 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2397 "0x%x data 0x%llx\n", msr, data);
2399 case MSR_P6_PERFCTR0:
2400 case MSR_P6_PERFCTR1:
2402 case MSR_P6_EVNTSEL0:
2403 case MSR_P6_EVNTSEL1:
2404 if (kvm_pmu_msr(vcpu, msr))
2405 return kvm_pmu_set_msr(vcpu, msr_info);
2407 if (pr || data != 0)
2408 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2409 "0x%x data 0x%llx\n", msr, data);
2411 case MSR_K7_CLK_CTL:
2413 * Ignore all writes to this no longer documented MSR.
2414 * Writes are only relevant for old K7 processors,
2415 * all pre-dating SVM, but a recommended workaround from
2416 * AMD for these chips. It is possible to specify the
2417 * affected processor models on the command line, hence
2418 * the need to ignore the workaround.
2421 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2422 if (kvm_hv_msr_partition_wide(msr)) {
2424 mutex_lock(&vcpu->kvm->lock);
2425 r = set_msr_hyperv_pw(vcpu, msr, data);
2426 mutex_unlock(&vcpu->kvm->lock);
2429 return set_msr_hyperv(vcpu, msr, data);
2431 case MSR_IA32_BBL_CR_CTL3:
2432 /* Drop writes to this legacy MSR -- see rdmsr
2433 * counterpart for further detail.
2435 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2437 case MSR_AMD64_OSVW_ID_LENGTH:
2438 if (!guest_cpuid_has_osvw(vcpu))
2440 vcpu->arch.osvw.length = data;
2442 case MSR_AMD64_OSVW_STATUS:
2443 if (!guest_cpuid_has_osvw(vcpu))
2445 vcpu->arch.osvw.status = data;
2448 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2449 return xen_hvm_config(vcpu, data);
2450 if (kvm_pmu_msr(vcpu, msr))
2451 return kvm_pmu_set_msr(vcpu, msr_info);
2453 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2457 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2464 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2468 * Reads an msr value (of 'msr_index') into 'pdata'.
2469 * Returns 0 on success, non-0 otherwise.
2470 * Assumes vcpu_load() was already called.
2472 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2474 return kvm_x86_ops->get_msr(vcpu, msr);
2476 EXPORT_SYMBOL_GPL(kvm_get_msr);
2478 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2480 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2482 if (!msr_mtrr_valid(msr))
2485 if (msr == MSR_MTRRdefType)
2486 *pdata = vcpu->arch.mtrr_state.def_type +
2487 (vcpu->arch.mtrr_state.enabled << 10);
2488 else if (msr == MSR_MTRRfix64K_00000)
2490 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2491 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2492 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2493 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2494 else if (msr == MSR_IA32_CR_PAT)
2495 *pdata = vcpu->arch.pat;
2496 else { /* Variable MTRRs */
2497 int idx, is_mtrr_mask;
2500 idx = (msr - 0x200) / 2;
2501 is_mtrr_mask = msr - 0x200 - 2 * idx;
2504 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2507 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2514 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2517 u64 mcg_cap = vcpu->arch.mcg_cap;
2518 unsigned bank_num = mcg_cap & 0xff;
2521 case MSR_IA32_P5_MC_ADDR:
2522 case MSR_IA32_P5_MC_TYPE:
2525 case MSR_IA32_MCG_CAP:
2526 data = vcpu->arch.mcg_cap;
2528 case MSR_IA32_MCG_CTL:
2529 if (!(mcg_cap & MCG_CTL_P))
2531 data = vcpu->arch.mcg_ctl;
2533 case MSR_IA32_MCG_STATUS:
2534 data = vcpu->arch.mcg_status;
2537 if (msr >= MSR_IA32_MC0_CTL &&
2538 msr < MSR_IA32_MCx_CTL(bank_num)) {
2539 u32 offset = msr - MSR_IA32_MC0_CTL;
2540 data = vcpu->arch.mce_banks[offset];
2549 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2552 struct kvm *kvm = vcpu->kvm;
2555 case HV_X64_MSR_GUEST_OS_ID:
2556 data = kvm->arch.hv_guest_os_id;
2558 case HV_X64_MSR_HYPERCALL:
2559 data = kvm->arch.hv_hypercall;
2561 case HV_X64_MSR_TIME_REF_COUNT: {
2563 div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2566 case HV_X64_MSR_REFERENCE_TSC:
2567 data = kvm->arch.hv_tsc_page;
2570 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2578 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2583 case HV_X64_MSR_VP_INDEX: {
2586 kvm_for_each_vcpu(r, v, vcpu->kvm) {
2594 case HV_X64_MSR_EOI:
2595 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2596 case HV_X64_MSR_ICR:
2597 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2598 case HV_X64_MSR_TPR:
2599 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2600 case HV_X64_MSR_APIC_ASSIST_PAGE:
2601 data = vcpu->arch.hv_vapic;
2604 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2611 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2615 switch (msr_info->index) {
2616 case MSR_IA32_PLATFORM_ID:
2617 case MSR_IA32_EBL_CR_POWERON:
2618 case MSR_IA32_DEBUGCTLMSR:
2619 case MSR_IA32_LASTBRANCHFROMIP:
2620 case MSR_IA32_LASTBRANCHTOIP:
2621 case MSR_IA32_LASTINTFROMIP:
2622 case MSR_IA32_LASTINTTOIP:
2625 case MSR_VM_HSAVE_PA:
2626 case MSR_K7_EVNTSEL0:
2627 case MSR_K7_EVNTSEL1:
2628 case MSR_K7_EVNTSEL2:
2629 case MSR_K7_EVNTSEL3:
2630 case MSR_K7_PERFCTR0:
2631 case MSR_K7_PERFCTR1:
2632 case MSR_K7_PERFCTR2:
2633 case MSR_K7_PERFCTR3:
2634 case MSR_K8_INT_PENDING_MSG:
2635 case MSR_AMD64_NB_CFG:
2636 case MSR_FAM10H_MMIO_CONF_BASE:
2637 case MSR_AMD64_BU_CFG2:
2640 case MSR_P6_PERFCTR0:
2641 case MSR_P6_PERFCTR1:
2642 case MSR_P6_EVNTSEL0:
2643 case MSR_P6_EVNTSEL1:
2644 if (kvm_pmu_msr(vcpu, msr_info->index))
2645 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2648 case MSR_IA32_UCODE_REV:
2649 msr_info->data = 0x100000000ULL;
2652 msr_info->data = 0x500 | KVM_NR_VAR_MTRR;
2654 case 0x200 ... 0x2ff:
2655 return get_msr_mtrr(vcpu, msr_info->index, &msr_info->data);
2656 case 0xcd: /* fsb frequency */
2660 * MSR_EBC_FREQUENCY_ID
2661 * Conservative value valid for even the basic CPU models.
2662 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2663 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2664 * and 266MHz for model 3, or 4. Set Core Clock
2665 * Frequency to System Bus Frequency Ratio to 1 (bits
2666 * 31:24) even though these are only valid for CPU
2667 * models > 2, however guests may end up dividing or
2668 * multiplying by zero otherwise.
2670 case MSR_EBC_FREQUENCY_ID:
2671 msr_info->data = 1 << 24;
2673 case MSR_IA32_APICBASE:
2674 msr_info->data = kvm_get_apic_base(vcpu);
2676 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2677 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2679 case MSR_IA32_TSCDEADLINE:
2680 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2682 case MSR_IA32_TSC_ADJUST:
2683 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2685 case MSR_IA32_MISC_ENABLE:
2686 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2688 case MSR_IA32_SMBASE:
2689 if (!msr_info->host_initiated)
2691 msr_info->data = vcpu->arch.smbase;
2693 case MSR_IA32_PERF_STATUS:
2694 /* TSC increment by tick */
2695 msr_info->data = 1000ULL;
2696 /* CPU multiplier */
2697 data |= (((uint64_t)4ULL) << 40);
2700 msr_info->data = vcpu->arch.efer;
2702 case MSR_KVM_WALL_CLOCK:
2703 case MSR_KVM_WALL_CLOCK_NEW:
2704 msr_info->data = vcpu->kvm->arch.wall_clock;
2706 case MSR_KVM_SYSTEM_TIME:
2707 case MSR_KVM_SYSTEM_TIME_NEW:
2708 msr_info->data = vcpu->arch.time;
2710 case MSR_KVM_ASYNC_PF_EN:
2711 msr_info->data = vcpu->arch.apf.msr_val;
2713 case MSR_KVM_STEAL_TIME:
2714 msr_info->data = vcpu->arch.st.msr_val;
2716 case MSR_KVM_PV_EOI_EN:
2717 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2719 case MSR_IA32_P5_MC_ADDR:
2720 case MSR_IA32_P5_MC_TYPE:
2721 case MSR_IA32_MCG_CAP:
2722 case MSR_IA32_MCG_CTL:
2723 case MSR_IA32_MCG_STATUS:
2724 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2725 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2726 case MSR_K7_CLK_CTL:
2728 * Provide expected ramp-up count for K7. All other
2729 * are set to zero, indicating minimum divisors for
2732 * This prevents guest kernels on AMD host with CPU
2733 * type 6, model 8 and higher from exploding due to
2734 * the rdmsr failing.
2736 msr_info->data = 0x20000000;
2738 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2739 if (kvm_hv_msr_partition_wide(msr_info->index)) {
2741 mutex_lock(&vcpu->kvm->lock);
2742 r = get_msr_hyperv_pw(vcpu, msr_info->index, &msr_info->data);
2743 mutex_unlock(&vcpu->kvm->lock);
2746 return get_msr_hyperv(vcpu, msr_info->index, &msr_info->data);
2748 case MSR_IA32_BBL_CR_CTL3:
2749 /* This legacy MSR exists but isn't fully documented in current
2750 * silicon. It is however accessed by winxp in very narrow
2751 * scenarios where it sets bit #19, itself documented as
2752 * a "reserved" bit. Best effort attempt to source coherent
2753 * read data here should the balance of the register be
2754 * interpreted by the guest:
2756 * L2 cache control register 3: 64GB range, 256KB size,
2757 * enabled, latency 0x1, configured
2759 msr_info->data = 0xbe702111;
2761 case MSR_AMD64_OSVW_ID_LENGTH:
2762 if (!guest_cpuid_has_osvw(vcpu))
2764 msr_info->data = vcpu->arch.osvw.length;
2766 case MSR_AMD64_OSVW_STATUS:
2767 if (!guest_cpuid_has_osvw(vcpu))
2769 msr_info->data = vcpu->arch.osvw.status;
2772 if (kvm_pmu_msr(vcpu, msr_info->index))
2773 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2775 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2778 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2785 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2788 * Read or write a bunch of msrs. All parameters are kernel addresses.
2790 * @return number of msrs set successfully.
2792 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2793 struct kvm_msr_entry *entries,
2794 int (*do_msr)(struct kvm_vcpu *vcpu,
2795 unsigned index, u64 *data))
2799 idx = srcu_read_lock(&vcpu->kvm->srcu);
2800 for (i = 0; i < msrs->nmsrs; ++i)
2801 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2803 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2809 * Read or write a bunch of msrs. Parameters are user addresses.
2811 * @return number of msrs set successfully.
2813 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2814 int (*do_msr)(struct kvm_vcpu *vcpu,
2815 unsigned index, u64 *data),
2818 struct kvm_msrs msrs;
2819 struct kvm_msr_entry *entries;
2824 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2828 if (msrs.nmsrs >= MAX_IO_MSRS)
2831 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2832 entries = memdup_user(user_msrs->entries, size);
2833 if (IS_ERR(entries)) {
2834 r = PTR_ERR(entries);
2838 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2843 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2854 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2859 case KVM_CAP_IRQCHIP:
2861 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2862 case KVM_CAP_SET_TSS_ADDR:
2863 case KVM_CAP_EXT_CPUID:
2864 case KVM_CAP_EXT_EMUL_CPUID:
2865 case KVM_CAP_CLOCKSOURCE:
2867 case KVM_CAP_NOP_IO_DELAY:
2868 case KVM_CAP_MP_STATE:
2869 case KVM_CAP_SYNC_MMU:
2870 case KVM_CAP_USER_NMI:
2871 case KVM_CAP_REINJECT_CONTROL:
2872 case KVM_CAP_IRQ_INJECT_STATUS:
2873 case KVM_CAP_IOEVENTFD:
2874 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2876 case KVM_CAP_PIT_STATE2:
2877 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2878 case KVM_CAP_XEN_HVM:
2879 case KVM_CAP_ADJUST_CLOCK:
2880 case KVM_CAP_VCPU_EVENTS:
2881 case KVM_CAP_HYPERV:
2882 case KVM_CAP_HYPERV_VAPIC:
2883 case KVM_CAP_HYPERV_SPIN:
2884 case KVM_CAP_PCI_SEGMENT:
2885 case KVM_CAP_DEBUGREGS:
2886 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2888 case KVM_CAP_ASYNC_PF:
2889 case KVM_CAP_GET_TSC_KHZ:
2890 case KVM_CAP_KVMCLOCK_CTRL:
2891 case KVM_CAP_READONLY_MEM:
2892 case KVM_CAP_HYPERV_TIME:
2893 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2894 case KVM_CAP_TSC_DEADLINE_TIMER:
2895 case KVM_CAP_ENABLE_CAP_VM:
2896 case KVM_CAP_DISABLE_QUIRKS:
2897 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2898 case KVM_CAP_ASSIGN_DEV_IRQ:
2899 case KVM_CAP_PCI_2_3:
2903 case KVM_CAP_COALESCED_MMIO:
2904 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2907 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2909 case KVM_CAP_NR_VCPUS:
2910 r = KVM_SOFT_MAX_VCPUS;
2912 case KVM_CAP_MAX_VCPUS:
2915 case KVM_CAP_NR_MEMSLOTS:
2916 r = KVM_USER_MEM_SLOTS;
2918 case KVM_CAP_PV_MMU: /* obsolete */
2921 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2923 r = iommu_present(&pci_bus_type);
2927 r = KVM_MAX_MCE_BANKS;
2932 case KVM_CAP_TSC_CONTROL:
2933 r = kvm_has_tsc_control;
2943 long kvm_arch_dev_ioctl(struct file *filp,
2944 unsigned int ioctl, unsigned long arg)
2946 void __user *argp = (void __user *)arg;
2950 case KVM_GET_MSR_INDEX_LIST: {
2951 struct kvm_msr_list __user *user_msr_list = argp;
2952 struct kvm_msr_list msr_list;
2956 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2959 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2960 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2963 if (n < msr_list.nmsrs)
2966 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2967 num_msrs_to_save * sizeof(u32)))
2969 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2971 num_emulated_msrs * sizeof(u32)))
2976 case KVM_GET_SUPPORTED_CPUID:
2977 case KVM_GET_EMULATED_CPUID: {
2978 struct kvm_cpuid2 __user *cpuid_arg = argp;
2979 struct kvm_cpuid2 cpuid;
2982 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2985 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2991 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2996 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2999 mce_cap = KVM_MCE_CAP_SUPPORTED;
3001 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
3013 static void wbinvd_ipi(void *garbage)
3018 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3020 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3023 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3025 /* Address WBINVD may be executed by guest */
3026 if (need_emulate_wbinvd(vcpu)) {
3027 if (kvm_x86_ops->has_wbinvd_exit())
3028 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3029 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3030 smp_call_function_single(vcpu->cpu,
3031 wbinvd_ipi, NULL, 1);
3034 kvm_x86_ops->vcpu_load(vcpu, cpu);
3036 /* Apply any externally detected TSC adjustments (due to suspend) */
3037 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3038 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3039 vcpu->arch.tsc_offset_adjustment = 0;
3040 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3043 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
3044 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3045 native_read_tsc() - vcpu->arch.last_host_tsc;
3047 mark_tsc_unstable("KVM discovered backwards TSC");
3048 if (check_tsc_unstable()) {
3049 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
3050 vcpu->arch.last_guest_tsc);
3051 kvm_x86_ops->write_tsc_offset(vcpu, offset);
3052 vcpu->arch.tsc_catchup = 1;
3055 * On a host with synchronized TSC, there is no need to update
3056 * kvmclock on vcpu->cpu migration
3058 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3059 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3060 if (vcpu->cpu != cpu)
3061 kvm_migrate_timers(vcpu);
3065 accumulate_steal_time(vcpu);
3066 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3069 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3071 kvm_x86_ops->vcpu_put(vcpu);
3072 kvm_put_guest_fpu(vcpu);
3073 vcpu->arch.last_host_tsc = native_read_tsc();
3076 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3077 struct kvm_lapic_state *s)
3079 kvm_x86_ops->sync_pir_to_irr(vcpu);
3080 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
3085 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3086 struct kvm_lapic_state *s)
3088 kvm_apic_post_state_restore(vcpu, s);
3089 update_cr8_intercept(vcpu);
3094 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3095 struct kvm_interrupt *irq)
3097 if (irq->irq >= KVM_NR_INTERRUPTS)
3099 if (irqchip_in_kernel(vcpu->kvm))
3102 kvm_queue_interrupt(vcpu, irq->irq, false);
3103 kvm_make_request(KVM_REQ_EVENT, vcpu);
3108 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3110 kvm_inject_nmi(vcpu);
3115 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3117 kvm_make_request(KVM_REQ_SMI, vcpu);
3122 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3123 struct kvm_tpr_access_ctl *tac)
3127 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3131 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3135 unsigned bank_num = mcg_cap & 0xff, bank;
3138 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3140 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
3143 vcpu->arch.mcg_cap = mcg_cap;
3144 /* Init IA32_MCG_CTL to all 1s */
3145 if (mcg_cap & MCG_CTL_P)
3146 vcpu->arch.mcg_ctl = ~(u64)0;
3147 /* Init IA32_MCi_CTL to all 1s */
3148 for (bank = 0; bank < bank_num; bank++)
3149 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3154 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3155 struct kvm_x86_mce *mce)
3157 u64 mcg_cap = vcpu->arch.mcg_cap;
3158 unsigned bank_num = mcg_cap & 0xff;
3159 u64 *banks = vcpu->arch.mce_banks;
3161 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3164 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3165 * reporting is disabled
3167 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3168 vcpu->arch.mcg_ctl != ~(u64)0)
3170 banks += 4 * mce->bank;
3172 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3173 * reporting is disabled for the bank
3175 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3177 if (mce->status & MCI_STATUS_UC) {
3178 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3179 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3180 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3183 if (banks[1] & MCI_STATUS_VAL)
3184 mce->status |= MCI_STATUS_OVER;
3185 banks[2] = mce->addr;
3186 banks[3] = mce->misc;
3187 vcpu->arch.mcg_status = mce->mcg_status;
3188 banks[1] = mce->status;
3189 kvm_queue_exception(vcpu, MC_VECTOR);
3190 } else if (!(banks[1] & MCI_STATUS_VAL)
3191 || !(banks[1] & MCI_STATUS_UC)) {
3192 if (banks[1] & MCI_STATUS_VAL)
3193 mce->status |= MCI_STATUS_OVER;
3194 banks[2] = mce->addr;
3195 banks[3] = mce->misc;
3196 banks[1] = mce->status;
3198 banks[1] |= MCI_STATUS_OVER;
3202 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3203 struct kvm_vcpu_events *events)
3206 events->exception.injected =
3207 vcpu->arch.exception.pending &&
3208 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3209 events->exception.nr = vcpu->arch.exception.nr;
3210 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3211 events->exception.pad = 0;
3212 events->exception.error_code = vcpu->arch.exception.error_code;
3214 events->interrupt.injected =
3215 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3216 events->interrupt.nr = vcpu->arch.interrupt.nr;
3217 events->interrupt.soft = 0;
3218 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3220 events->nmi.injected = vcpu->arch.nmi_injected;
3221 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3222 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3223 events->nmi.pad = 0;
3225 events->sipi_vector = 0; /* never valid when reporting to user space */
3227 events->smi.smm = is_smm(vcpu);
3228 events->smi.pending = vcpu->arch.smi_pending;
3229 events->smi.smm_inside_nmi =
3230 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3231 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3233 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3234 | KVM_VCPUEVENT_VALID_SHADOW
3235 | KVM_VCPUEVENT_VALID_SMM);
3236 memset(&events->reserved, 0, sizeof(events->reserved));
3239 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3240 struct kvm_vcpu_events *events)
3242 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3243 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3244 | KVM_VCPUEVENT_VALID_SHADOW
3245 | KVM_VCPUEVENT_VALID_SMM))
3249 vcpu->arch.exception.pending = events->exception.injected;
3250 vcpu->arch.exception.nr = events->exception.nr;
3251 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3252 vcpu->arch.exception.error_code = events->exception.error_code;
3254 vcpu->arch.interrupt.pending = events->interrupt.injected;
3255 vcpu->arch.interrupt.nr = events->interrupt.nr;
3256 vcpu->arch.interrupt.soft = events->interrupt.soft;
3257 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3258 kvm_x86_ops->set_interrupt_shadow(vcpu,
3259 events->interrupt.shadow);
3261 vcpu->arch.nmi_injected = events->nmi.injected;
3262 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3263 vcpu->arch.nmi_pending = events->nmi.pending;
3264 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3266 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3267 kvm_vcpu_has_lapic(vcpu))
3268 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3270 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3271 if (events->smi.smm)
3272 vcpu->arch.hflags |= HF_SMM_MASK;
3274 vcpu->arch.hflags &= ~HF_SMM_MASK;
3275 vcpu->arch.smi_pending = events->smi.pending;
3276 if (events->smi.smm_inside_nmi)
3277 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3279 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3280 if (kvm_vcpu_has_lapic(vcpu)) {
3281 if (events->smi.latched_init)
3282 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3284 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3288 kvm_make_request(KVM_REQ_EVENT, vcpu);
3293 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3294 struct kvm_debugregs *dbgregs)
3298 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3299 kvm_get_dr(vcpu, 6, &val);
3301 dbgregs->dr7 = vcpu->arch.dr7;
3303 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3306 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3307 struct kvm_debugregs *dbgregs)
3312 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3313 kvm_update_dr0123(vcpu);
3314 vcpu->arch.dr6 = dbgregs->dr6;
3315 kvm_update_dr6(vcpu);
3316 vcpu->arch.dr7 = dbgregs->dr7;
3317 kvm_update_dr7(vcpu);
3322 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3324 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3326 struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3327 u64 xstate_bv = xsave->xsave_hdr.xstate_bv;
3331 * Copy legacy XSAVE area, to avoid complications with CPUID
3332 * leaves 0 and 1 in the loop below.
3334 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3337 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3340 * Copy each region from the possibly compacted offset to the
3341 * non-compacted offset.
3343 valid = xstate_bv & ~XSTATE_FPSSE;
3345 u64 feature = valid & -valid;
3346 int index = fls64(feature) - 1;
3347 void *src = get_xsave_addr(xsave, feature);
3350 u32 size, offset, ecx, edx;
3351 cpuid_count(XSTATE_CPUID, index,
3352 &size, &offset, &ecx, &edx);
3353 memcpy(dest + offset, src, size);
3360 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3362 struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3363 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3367 * Copy legacy XSAVE area, to avoid complications with CPUID
3368 * leaves 0 and 1 in the loop below.
3370 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3372 /* Set XSTATE_BV and possibly XCOMP_BV. */
3373 xsave->xsave_hdr.xstate_bv = xstate_bv;
3375 xsave->xsave_hdr.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3378 * Copy each region from the non-compacted offset to the
3379 * possibly compacted offset.
3381 valid = xstate_bv & ~XSTATE_FPSSE;
3383 u64 feature = valid & -valid;
3384 int index = fls64(feature) - 1;
3385 void *dest = get_xsave_addr(xsave, feature);
3388 u32 size, offset, ecx, edx;
3389 cpuid_count(XSTATE_CPUID, index,
3390 &size, &offset, &ecx, &edx);
3391 memcpy(dest, src + offset, size);
3399 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3400 struct kvm_xsave *guest_xsave)
3402 if (cpu_has_xsave) {
3403 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3404 fill_xsave((u8 *) guest_xsave->region, vcpu);
3406 memcpy(guest_xsave->region,
3407 &vcpu->arch.guest_fpu.state->fxsave,
3408 sizeof(struct i387_fxsave_struct));
3409 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3414 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3415 struct kvm_xsave *guest_xsave)
3418 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3420 if (cpu_has_xsave) {
3422 * Here we allow setting states that are not present in
3423 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3424 * with old userspace.
3426 if (xstate_bv & ~kvm_supported_xcr0())
3428 load_xsave(vcpu, (u8 *)guest_xsave->region);
3430 if (xstate_bv & ~XSTATE_FPSSE)
3432 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3433 guest_xsave->region, sizeof(struct i387_fxsave_struct));
3438 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3439 struct kvm_xcrs *guest_xcrs)
3441 if (!cpu_has_xsave) {
3442 guest_xcrs->nr_xcrs = 0;
3446 guest_xcrs->nr_xcrs = 1;
3447 guest_xcrs->flags = 0;
3448 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3449 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3452 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3453 struct kvm_xcrs *guest_xcrs)
3460 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3463 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3464 /* Only support XCR0 currently */
3465 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3466 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3467 guest_xcrs->xcrs[i].value);
3476 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3477 * stopped by the hypervisor. This function will be called from the host only.
3478 * EINVAL is returned when the host attempts to set the flag for a guest that
3479 * does not support pv clocks.
3481 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3483 if (!vcpu->arch.pv_time_enabled)
3485 vcpu->arch.pvclock_set_guest_stopped_request = true;
3486 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3490 long kvm_arch_vcpu_ioctl(struct file *filp,
3491 unsigned int ioctl, unsigned long arg)
3493 struct kvm_vcpu *vcpu = filp->private_data;
3494 void __user *argp = (void __user *)arg;
3497 struct kvm_lapic_state *lapic;
3498 struct kvm_xsave *xsave;
3499 struct kvm_xcrs *xcrs;
3505 case KVM_GET_LAPIC: {
3507 if (!vcpu->arch.apic)
3509 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3514 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3518 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3523 case KVM_SET_LAPIC: {
3525 if (!vcpu->arch.apic)
3527 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3528 if (IS_ERR(u.lapic))
3529 return PTR_ERR(u.lapic);
3531 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3534 case KVM_INTERRUPT: {
3535 struct kvm_interrupt irq;
3538 if (copy_from_user(&irq, argp, sizeof irq))
3540 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3544 r = kvm_vcpu_ioctl_nmi(vcpu);
3548 r = kvm_vcpu_ioctl_smi(vcpu);
3551 case KVM_SET_CPUID: {
3552 struct kvm_cpuid __user *cpuid_arg = argp;
3553 struct kvm_cpuid cpuid;
3556 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3558 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3561 case KVM_SET_CPUID2: {
3562 struct kvm_cpuid2 __user *cpuid_arg = argp;
3563 struct kvm_cpuid2 cpuid;
3566 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3568 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3569 cpuid_arg->entries);
3572 case KVM_GET_CPUID2: {
3573 struct kvm_cpuid2 __user *cpuid_arg = argp;
3574 struct kvm_cpuid2 cpuid;
3577 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3579 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3580 cpuid_arg->entries);
3584 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3590 r = msr_io(vcpu, argp, do_get_msr, 1);
3593 r = msr_io(vcpu, argp, do_set_msr, 0);
3595 case KVM_TPR_ACCESS_REPORTING: {
3596 struct kvm_tpr_access_ctl tac;
3599 if (copy_from_user(&tac, argp, sizeof tac))
3601 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3605 if (copy_to_user(argp, &tac, sizeof tac))
3610 case KVM_SET_VAPIC_ADDR: {
3611 struct kvm_vapic_addr va;
3614 if (!irqchip_in_kernel(vcpu->kvm))
3617 if (copy_from_user(&va, argp, sizeof va))
3619 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3622 case KVM_X86_SETUP_MCE: {
3626 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3628 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3631 case KVM_X86_SET_MCE: {
3632 struct kvm_x86_mce mce;
3635 if (copy_from_user(&mce, argp, sizeof mce))
3637 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3640 case KVM_GET_VCPU_EVENTS: {
3641 struct kvm_vcpu_events events;
3643 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3646 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3651 case KVM_SET_VCPU_EVENTS: {
3652 struct kvm_vcpu_events events;
3655 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3658 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3661 case KVM_GET_DEBUGREGS: {
3662 struct kvm_debugregs dbgregs;
3664 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3667 if (copy_to_user(argp, &dbgregs,
3668 sizeof(struct kvm_debugregs)))
3673 case KVM_SET_DEBUGREGS: {
3674 struct kvm_debugregs dbgregs;
3677 if (copy_from_user(&dbgregs, argp,
3678 sizeof(struct kvm_debugregs)))
3681 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3684 case KVM_GET_XSAVE: {
3685 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3690 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3693 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3698 case KVM_SET_XSAVE: {
3699 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3700 if (IS_ERR(u.xsave))
3701 return PTR_ERR(u.xsave);
3703 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3706 case KVM_GET_XCRS: {
3707 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3712 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3715 if (copy_to_user(argp, u.xcrs,
3716 sizeof(struct kvm_xcrs)))
3721 case KVM_SET_XCRS: {
3722 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3724 return PTR_ERR(u.xcrs);
3726 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3729 case KVM_SET_TSC_KHZ: {
3733 user_tsc_khz = (u32)arg;
3735 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3738 if (user_tsc_khz == 0)
3739 user_tsc_khz = tsc_khz;
3741 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3746 case KVM_GET_TSC_KHZ: {
3747 r = vcpu->arch.virtual_tsc_khz;
3750 case KVM_KVMCLOCK_CTRL: {
3751 r = kvm_set_guest_paused(vcpu);
3762 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3764 return VM_FAULT_SIGBUS;
3767 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3771 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3773 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3777 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3780 kvm->arch.ept_identity_map_addr = ident_addr;
3784 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3785 u32 kvm_nr_mmu_pages)
3787 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3790 mutex_lock(&kvm->slots_lock);
3792 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3793 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3795 mutex_unlock(&kvm->slots_lock);
3799 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3801 return kvm->arch.n_max_mmu_pages;
3804 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3809 switch (chip->chip_id) {
3810 case KVM_IRQCHIP_PIC_MASTER:
3811 memcpy(&chip->chip.pic,
3812 &pic_irqchip(kvm)->pics[0],
3813 sizeof(struct kvm_pic_state));
3815 case KVM_IRQCHIP_PIC_SLAVE:
3816 memcpy(&chip->chip.pic,
3817 &pic_irqchip(kvm)->pics[1],
3818 sizeof(struct kvm_pic_state));
3820 case KVM_IRQCHIP_IOAPIC:
3821 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3830 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3835 switch (chip->chip_id) {
3836 case KVM_IRQCHIP_PIC_MASTER:
3837 spin_lock(&pic_irqchip(kvm)->lock);
3838 memcpy(&pic_irqchip(kvm)->pics[0],
3840 sizeof(struct kvm_pic_state));
3841 spin_unlock(&pic_irqchip(kvm)->lock);
3843 case KVM_IRQCHIP_PIC_SLAVE:
3844 spin_lock(&pic_irqchip(kvm)->lock);
3845 memcpy(&pic_irqchip(kvm)->pics[1],
3847 sizeof(struct kvm_pic_state));
3848 spin_unlock(&pic_irqchip(kvm)->lock);
3850 case KVM_IRQCHIP_IOAPIC:
3851 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3857 kvm_pic_update_irq(pic_irqchip(kvm));
3861 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3865 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3866 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3867 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3871 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3875 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3876 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3877 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3878 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3882 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3886 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3887 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3888 sizeof(ps->channels));
3889 ps->flags = kvm->arch.vpit->pit_state.flags;
3890 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3891 memset(&ps->reserved, 0, sizeof(ps->reserved));
3895 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3897 int r = 0, start = 0;
3898 u32 prev_legacy, cur_legacy;
3899 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3900 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3901 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3902 if (!prev_legacy && cur_legacy)
3904 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3905 sizeof(kvm->arch.vpit->pit_state.channels));
3906 kvm->arch.vpit->pit_state.flags = ps->flags;
3907 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3908 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3912 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3913 struct kvm_reinject_control *control)
3915 if (!kvm->arch.vpit)
3917 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3918 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3919 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3924 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3925 * @kvm: kvm instance
3926 * @log: slot id and address to which we copy the log
3928 * Steps 1-4 below provide general overview of dirty page logging. See
3929 * kvm_get_dirty_log_protect() function description for additional details.
3931 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3932 * always flush the TLB (step 4) even if previous step failed and the dirty
3933 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3934 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3935 * writes will be marked dirty for next log read.
3937 * 1. Take a snapshot of the bit and clear it if needed.
3938 * 2. Write protect the corresponding page.
3939 * 3. Copy the snapshot to the userspace.
3940 * 4. Flush TLB's if needed.
3942 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3944 bool is_dirty = false;
3947 mutex_lock(&kvm->slots_lock);
3950 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3952 if (kvm_x86_ops->flush_log_dirty)
3953 kvm_x86_ops->flush_log_dirty(kvm);
3955 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3958 * All the TLBs can be flushed out of mmu lock, see the comments in
3959 * kvm_mmu_slot_remove_write_access().
3961 lockdep_assert_held(&kvm->slots_lock);
3963 kvm_flush_remote_tlbs(kvm);
3965 mutex_unlock(&kvm->slots_lock);
3969 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3972 if (!irqchip_in_kernel(kvm))
3975 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3976 irq_event->irq, irq_event->level,
3981 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3982 struct kvm_enable_cap *cap)
3990 case KVM_CAP_DISABLE_QUIRKS:
3991 kvm->arch.disabled_quirks = cap->args[0];
4001 long kvm_arch_vm_ioctl(struct file *filp,
4002 unsigned int ioctl, unsigned long arg)
4004 struct kvm *kvm = filp->private_data;
4005 void __user *argp = (void __user *)arg;
4008 * This union makes it completely explicit to gcc-3.x
4009 * that these two variables' stack usage should be
4010 * combined, not added together.
4013 struct kvm_pit_state ps;
4014 struct kvm_pit_state2 ps2;
4015 struct kvm_pit_config pit_config;
4019 case KVM_SET_TSS_ADDR:
4020 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4022 case KVM_SET_IDENTITY_MAP_ADDR: {
4026 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4028 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4031 case KVM_SET_NR_MMU_PAGES:
4032 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4034 case KVM_GET_NR_MMU_PAGES:
4035 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4037 case KVM_CREATE_IRQCHIP: {
4038 struct kvm_pic *vpic;
4040 mutex_lock(&kvm->lock);
4043 goto create_irqchip_unlock;
4045 if (atomic_read(&kvm->online_vcpus))
4046 goto create_irqchip_unlock;
4048 vpic = kvm_create_pic(kvm);
4050 r = kvm_ioapic_init(kvm);
4052 mutex_lock(&kvm->slots_lock);
4053 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
4055 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
4057 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
4059 mutex_unlock(&kvm->slots_lock);
4061 goto create_irqchip_unlock;
4064 goto create_irqchip_unlock;
4066 kvm->arch.vpic = vpic;
4068 r = kvm_setup_default_irq_routing(kvm);
4070 mutex_lock(&kvm->slots_lock);
4071 mutex_lock(&kvm->irq_lock);
4072 kvm_ioapic_destroy(kvm);
4073 kvm_destroy_pic(kvm);
4074 mutex_unlock(&kvm->irq_lock);
4075 mutex_unlock(&kvm->slots_lock);
4077 create_irqchip_unlock:
4078 mutex_unlock(&kvm->lock);
4081 case KVM_CREATE_PIT:
4082 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4084 case KVM_CREATE_PIT2:
4086 if (copy_from_user(&u.pit_config, argp,
4087 sizeof(struct kvm_pit_config)))
4090 mutex_lock(&kvm->slots_lock);
4093 goto create_pit_unlock;
4095 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4099 mutex_unlock(&kvm->slots_lock);
4101 case KVM_GET_IRQCHIP: {
4102 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4103 struct kvm_irqchip *chip;
4105 chip = memdup_user(argp, sizeof(*chip));
4112 if (!irqchip_in_kernel(kvm))
4113 goto get_irqchip_out;
4114 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4116 goto get_irqchip_out;
4118 if (copy_to_user(argp, chip, sizeof *chip))
4119 goto get_irqchip_out;
4125 case KVM_SET_IRQCHIP: {
4126 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4127 struct kvm_irqchip *chip;
4129 chip = memdup_user(argp, sizeof(*chip));
4136 if (!irqchip_in_kernel(kvm))
4137 goto set_irqchip_out;
4138 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4140 goto set_irqchip_out;
4148 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4151 if (!kvm->arch.vpit)
4153 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4157 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4164 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4167 if (!kvm->arch.vpit)
4169 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4172 case KVM_GET_PIT2: {
4174 if (!kvm->arch.vpit)
4176 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4180 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4185 case KVM_SET_PIT2: {
4187 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4190 if (!kvm->arch.vpit)
4192 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4195 case KVM_REINJECT_CONTROL: {
4196 struct kvm_reinject_control control;
4198 if (copy_from_user(&control, argp, sizeof(control)))
4200 r = kvm_vm_ioctl_reinject(kvm, &control);
4203 case KVM_XEN_HVM_CONFIG: {
4205 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4206 sizeof(struct kvm_xen_hvm_config)))
4209 if (kvm->arch.xen_hvm_config.flags)
4214 case KVM_SET_CLOCK: {
4215 struct kvm_clock_data user_ns;
4220 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4228 local_irq_disable();
4229 now_ns = get_kernel_ns();
4230 delta = user_ns.clock - now_ns;
4232 kvm->arch.kvmclock_offset = delta;
4233 kvm_gen_update_masterclock(kvm);
4236 case KVM_GET_CLOCK: {
4237 struct kvm_clock_data user_ns;
4240 local_irq_disable();
4241 now_ns = get_kernel_ns();
4242 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4245 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4248 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4253 case KVM_ENABLE_CAP: {
4254 struct kvm_enable_cap cap;
4257 if (copy_from_user(&cap, argp, sizeof(cap)))
4259 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4263 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4269 static void kvm_init_msr_list(void)
4274 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4275 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4279 * Even MSRs that are valid in the host may not be exposed
4280 * to the guests in some cases. We could work around this
4281 * in VMX with the generic MSR save/load machinery, but it
4282 * is not really worthwhile since it will really only
4283 * happen with nested virtualization.
4285 switch (msrs_to_save[i]) {
4286 case MSR_IA32_BNDCFGS:
4287 if (!kvm_x86_ops->mpx_supported())
4295 msrs_to_save[j] = msrs_to_save[i];
4298 num_msrs_to_save = j;
4300 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4301 switch (emulated_msrs[i]) {
4307 emulated_msrs[j] = emulated_msrs[i];
4310 num_emulated_msrs = j;
4313 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4321 if (!(vcpu->arch.apic &&
4322 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4323 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4334 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4341 if (!(vcpu->arch.apic &&
4342 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4344 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4346 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4356 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4357 struct kvm_segment *var, int seg)
4359 kvm_x86_ops->set_segment(vcpu, var, seg);
4362 void kvm_get_segment(struct kvm_vcpu *vcpu,
4363 struct kvm_segment *var, int seg)
4365 kvm_x86_ops->get_segment(vcpu, var, seg);
4368 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4369 struct x86_exception *exception)
4373 BUG_ON(!mmu_is_nested(vcpu));
4375 /* NPT walks are always user-walks */
4376 access |= PFERR_USER_MASK;
4377 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4382 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4383 struct x86_exception *exception)
4385 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4386 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4389 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4390 struct x86_exception *exception)
4392 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4393 access |= PFERR_FETCH_MASK;
4394 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4397 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4398 struct x86_exception *exception)
4400 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4401 access |= PFERR_WRITE_MASK;
4402 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4405 /* uses this to access any guest's mapped memory without checking CPL */
4406 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4407 struct x86_exception *exception)
4409 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4412 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4413 struct kvm_vcpu *vcpu, u32 access,
4414 struct x86_exception *exception)
4417 int r = X86EMUL_CONTINUE;
4420 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4422 unsigned offset = addr & (PAGE_SIZE-1);
4423 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4426 if (gpa == UNMAPPED_GVA)
4427 return X86EMUL_PROPAGATE_FAULT;
4428 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4431 r = X86EMUL_IO_NEEDED;
4443 /* used for instruction fetching */
4444 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4445 gva_t addr, void *val, unsigned int bytes,
4446 struct x86_exception *exception)
4448 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4449 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4453 /* Inline kvm_read_guest_virt_helper for speed. */
4454 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4456 if (unlikely(gpa == UNMAPPED_GVA))
4457 return X86EMUL_PROPAGATE_FAULT;
4459 offset = addr & (PAGE_SIZE-1);
4460 if (WARN_ON(offset + bytes > PAGE_SIZE))
4461 bytes = (unsigned)PAGE_SIZE - offset;
4462 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4464 if (unlikely(ret < 0))
4465 return X86EMUL_IO_NEEDED;
4467 return X86EMUL_CONTINUE;
4470 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4471 gva_t addr, void *val, unsigned int bytes,
4472 struct x86_exception *exception)
4474 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4475 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4477 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4480 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4482 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4483 gva_t addr, void *val, unsigned int bytes,
4484 struct x86_exception *exception)
4486 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4487 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4490 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4491 gva_t addr, void *val,
4493 struct x86_exception *exception)
4495 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4497 int r = X86EMUL_CONTINUE;
4500 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4503 unsigned offset = addr & (PAGE_SIZE-1);
4504 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4507 if (gpa == UNMAPPED_GVA)
4508 return X86EMUL_PROPAGATE_FAULT;
4509 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4511 r = X86EMUL_IO_NEEDED;
4522 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4524 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4525 gpa_t *gpa, struct x86_exception *exception,
4528 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4529 | (write ? PFERR_WRITE_MASK : 0);
4531 if (vcpu_match_mmio_gva(vcpu, gva)
4532 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4533 vcpu->arch.access, access)) {
4534 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4535 (gva & (PAGE_SIZE - 1));
4536 trace_vcpu_match_mmio(gva, *gpa, write, false);
4540 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4542 if (*gpa == UNMAPPED_GVA)
4545 /* For APIC access vmexit */
4546 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4549 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4550 trace_vcpu_match_mmio(gva, *gpa, write, true);
4557 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4558 const void *val, int bytes)
4562 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4565 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4569 struct read_write_emulator_ops {
4570 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4572 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4573 void *val, int bytes);
4574 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4575 int bytes, void *val);
4576 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4577 void *val, int bytes);
4581 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4583 if (vcpu->mmio_read_completed) {
4584 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4585 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4586 vcpu->mmio_read_completed = 0;
4593 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4594 void *val, int bytes)
4596 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4599 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4600 void *val, int bytes)
4602 return emulator_write_phys(vcpu, gpa, val, bytes);
4605 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4607 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4608 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4611 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4612 void *val, int bytes)
4614 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4615 return X86EMUL_IO_NEEDED;
4618 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4619 void *val, int bytes)
4621 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4623 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4624 return X86EMUL_CONTINUE;
4627 static const struct read_write_emulator_ops read_emultor = {
4628 .read_write_prepare = read_prepare,
4629 .read_write_emulate = read_emulate,
4630 .read_write_mmio = vcpu_mmio_read,
4631 .read_write_exit_mmio = read_exit_mmio,
4634 static const struct read_write_emulator_ops write_emultor = {
4635 .read_write_emulate = write_emulate,
4636 .read_write_mmio = write_mmio,
4637 .read_write_exit_mmio = write_exit_mmio,
4641 static int emulator_read_write_onepage(unsigned long addr, void *val,
4643 struct x86_exception *exception,
4644 struct kvm_vcpu *vcpu,
4645 const struct read_write_emulator_ops *ops)
4649 bool write = ops->write;
4650 struct kvm_mmio_fragment *frag;
4652 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4655 return X86EMUL_PROPAGATE_FAULT;
4657 /* For APIC access vmexit */
4661 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4662 return X86EMUL_CONTINUE;
4666 * Is this MMIO handled locally?
4668 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4669 if (handled == bytes)
4670 return X86EMUL_CONTINUE;
4676 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4677 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4681 return X86EMUL_CONTINUE;
4684 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4686 void *val, unsigned int bytes,
4687 struct x86_exception *exception,
4688 const struct read_write_emulator_ops *ops)
4690 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4694 if (ops->read_write_prepare &&
4695 ops->read_write_prepare(vcpu, val, bytes))
4696 return X86EMUL_CONTINUE;
4698 vcpu->mmio_nr_fragments = 0;
4700 /* Crossing a page boundary? */
4701 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4704 now = -addr & ~PAGE_MASK;
4705 rc = emulator_read_write_onepage(addr, val, now, exception,
4708 if (rc != X86EMUL_CONTINUE)
4711 if (ctxt->mode != X86EMUL_MODE_PROT64)
4717 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4719 if (rc != X86EMUL_CONTINUE)
4722 if (!vcpu->mmio_nr_fragments)
4725 gpa = vcpu->mmio_fragments[0].gpa;
4727 vcpu->mmio_needed = 1;
4728 vcpu->mmio_cur_fragment = 0;
4730 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4731 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4732 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4733 vcpu->run->mmio.phys_addr = gpa;
4735 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4738 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4742 struct x86_exception *exception)
4744 return emulator_read_write(ctxt, addr, val, bytes,
4745 exception, &read_emultor);
4748 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4752 struct x86_exception *exception)
4754 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4755 exception, &write_emultor);
4758 #define CMPXCHG_TYPE(t, ptr, old, new) \
4759 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4761 #ifdef CONFIG_X86_64
4762 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4764 # define CMPXCHG64(ptr, old, new) \
4765 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4768 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4773 struct x86_exception *exception)
4775 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4781 /* guests cmpxchg8b have to be emulated atomically */
4782 if (bytes > 8 || (bytes & (bytes - 1)))
4785 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4787 if (gpa == UNMAPPED_GVA ||
4788 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4791 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4794 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4795 if (is_error_page(page))
4798 kaddr = kmap_atomic(page);
4799 kaddr += offset_in_page(gpa);
4802 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4805 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4808 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4811 exchanged = CMPXCHG64(kaddr, old, new);
4816 kunmap_atomic(kaddr);
4817 kvm_release_page_dirty(page);
4820 return X86EMUL_CMPXCHG_FAILED;
4822 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4823 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4825 return X86EMUL_CONTINUE;
4828 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4830 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4833 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4835 /* TODO: String I/O for in kernel device */
4838 if (vcpu->arch.pio.in)
4839 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4840 vcpu->arch.pio.size, pd);
4842 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4843 vcpu->arch.pio.port, vcpu->arch.pio.size,
4848 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4849 unsigned short port, void *val,
4850 unsigned int count, bool in)
4852 vcpu->arch.pio.port = port;
4853 vcpu->arch.pio.in = in;
4854 vcpu->arch.pio.count = count;
4855 vcpu->arch.pio.size = size;
4857 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4858 vcpu->arch.pio.count = 0;
4862 vcpu->run->exit_reason = KVM_EXIT_IO;
4863 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4864 vcpu->run->io.size = size;
4865 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4866 vcpu->run->io.count = count;
4867 vcpu->run->io.port = port;
4872 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4873 int size, unsigned short port, void *val,
4876 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4879 if (vcpu->arch.pio.count)
4882 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4885 memcpy(val, vcpu->arch.pio_data, size * count);
4886 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4887 vcpu->arch.pio.count = 0;
4894 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4895 int size, unsigned short port,
4896 const void *val, unsigned int count)
4898 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4900 memcpy(vcpu->arch.pio_data, val, size * count);
4901 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4902 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4905 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4907 return kvm_x86_ops->get_segment_base(vcpu, seg);
4910 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4912 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4915 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4917 if (!need_emulate_wbinvd(vcpu))
4918 return X86EMUL_CONTINUE;
4920 if (kvm_x86_ops->has_wbinvd_exit()) {
4921 int cpu = get_cpu();
4923 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4924 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4925 wbinvd_ipi, NULL, 1);
4927 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4930 return X86EMUL_CONTINUE;
4933 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4935 kvm_x86_ops->skip_emulated_instruction(vcpu);
4936 return kvm_emulate_wbinvd_noskip(vcpu);
4938 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4942 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4944 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4947 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4948 unsigned long *dest)
4950 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4953 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4954 unsigned long value)
4957 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4960 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4962 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4965 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4967 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4968 unsigned long value;
4972 value = kvm_read_cr0(vcpu);
4975 value = vcpu->arch.cr2;
4978 value = kvm_read_cr3(vcpu);
4981 value = kvm_read_cr4(vcpu);
4984 value = kvm_get_cr8(vcpu);
4987 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4994 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4996 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5001 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5004 vcpu->arch.cr2 = val;
5007 res = kvm_set_cr3(vcpu, val);
5010 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5013 res = kvm_set_cr8(vcpu, val);
5016 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5023 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5025 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5028 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5030 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5033 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5035 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5038 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5040 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5043 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5045 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5048 static unsigned long emulator_get_cached_segment_base(
5049 struct x86_emulate_ctxt *ctxt, int seg)
5051 return get_segment_base(emul_to_vcpu(ctxt), seg);
5054 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5055 struct desc_struct *desc, u32 *base3,
5058 struct kvm_segment var;
5060 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5061 *selector = var.selector;
5064 memset(desc, 0, sizeof(*desc));
5070 set_desc_limit(desc, var.limit);
5071 set_desc_base(desc, (unsigned long)var.base);
5072 #ifdef CONFIG_X86_64
5074 *base3 = var.base >> 32;
5076 desc->type = var.type;
5078 desc->dpl = var.dpl;
5079 desc->p = var.present;
5080 desc->avl = var.avl;
5088 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5089 struct desc_struct *desc, u32 base3,
5092 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5093 struct kvm_segment var;
5095 var.selector = selector;
5096 var.base = get_desc_base(desc);
5097 #ifdef CONFIG_X86_64
5098 var.base |= ((u64)base3) << 32;
5100 var.limit = get_desc_limit(desc);
5102 var.limit = (var.limit << 12) | 0xfff;
5103 var.type = desc->type;
5104 var.dpl = desc->dpl;
5109 var.avl = desc->avl;
5110 var.present = desc->p;
5111 var.unusable = !var.present;
5114 kvm_set_segment(vcpu, &var, seg);
5118 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5119 u32 msr_index, u64 *pdata)
5121 struct msr_data msr;
5124 msr.index = msr_index;
5125 msr.host_initiated = false;
5126 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5134 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5135 u32 msr_index, u64 data)
5137 struct msr_data msr;
5140 msr.index = msr_index;
5141 msr.host_initiated = false;
5142 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5145 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5147 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5149 return vcpu->arch.smbase;
5152 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5154 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5156 vcpu->arch.smbase = smbase;
5159 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5162 return kvm_pmu_check_pmc(emul_to_vcpu(ctxt), pmc);
5165 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5166 u32 pmc, u64 *pdata)
5168 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
5171 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5173 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5176 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5179 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5181 * CR0.TS may reference the host fpu state, not the guest fpu state,
5182 * so it may be clear at this point.
5187 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5192 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5193 struct x86_instruction_info *info,
5194 enum x86_intercept_stage stage)
5196 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5199 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5200 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5202 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5205 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5207 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5210 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5212 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5215 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5217 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5220 static const struct x86_emulate_ops emulate_ops = {
5221 .read_gpr = emulator_read_gpr,
5222 .write_gpr = emulator_write_gpr,
5223 .read_std = kvm_read_guest_virt_system,
5224 .write_std = kvm_write_guest_virt_system,
5225 .fetch = kvm_fetch_guest_virt,
5226 .read_emulated = emulator_read_emulated,
5227 .write_emulated = emulator_write_emulated,
5228 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5229 .invlpg = emulator_invlpg,
5230 .pio_in_emulated = emulator_pio_in_emulated,
5231 .pio_out_emulated = emulator_pio_out_emulated,
5232 .get_segment = emulator_get_segment,
5233 .set_segment = emulator_set_segment,
5234 .get_cached_segment_base = emulator_get_cached_segment_base,
5235 .get_gdt = emulator_get_gdt,
5236 .get_idt = emulator_get_idt,
5237 .set_gdt = emulator_set_gdt,
5238 .set_idt = emulator_set_idt,
5239 .get_cr = emulator_get_cr,
5240 .set_cr = emulator_set_cr,
5241 .cpl = emulator_get_cpl,
5242 .get_dr = emulator_get_dr,
5243 .set_dr = emulator_set_dr,
5244 .get_smbase = emulator_get_smbase,
5245 .set_smbase = emulator_set_smbase,
5246 .set_msr = emulator_set_msr,
5247 .get_msr = emulator_get_msr,
5248 .check_pmc = emulator_check_pmc,
5249 .read_pmc = emulator_read_pmc,
5250 .halt = emulator_halt,
5251 .wbinvd = emulator_wbinvd,
5252 .fix_hypercall = emulator_fix_hypercall,
5253 .get_fpu = emulator_get_fpu,
5254 .put_fpu = emulator_put_fpu,
5255 .intercept = emulator_intercept,
5256 .get_cpuid = emulator_get_cpuid,
5257 .set_nmi_mask = emulator_set_nmi_mask,
5260 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5262 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5264 * an sti; sti; sequence only disable interrupts for the first
5265 * instruction. So, if the last instruction, be it emulated or
5266 * not, left the system with the INT_STI flag enabled, it
5267 * means that the last instruction is an sti. We should not
5268 * leave the flag on in this case. The same goes for mov ss
5270 if (int_shadow & mask)
5272 if (unlikely(int_shadow || mask)) {
5273 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5275 kvm_make_request(KVM_REQ_EVENT, vcpu);
5279 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5281 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5282 if (ctxt->exception.vector == PF_VECTOR)
5283 return kvm_propagate_fault(vcpu, &ctxt->exception);
5285 if (ctxt->exception.error_code_valid)
5286 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5287 ctxt->exception.error_code);
5289 kvm_queue_exception(vcpu, ctxt->exception.vector);
5293 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5295 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5298 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5300 ctxt->eflags = kvm_get_rflags(vcpu);
5301 ctxt->eip = kvm_rip_read(vcpu);
5302 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5303 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5304 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5305 cs_db ? X86EMUL_MODE_PROT32 :
5306 X86EMUL_MODE_PROT16;
5307 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5308 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5309 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5310 ctxt->emul_flags = vcpu->arch.hflags;
5312 init_decode_cache(ctxt);
5313 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5316 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5318 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5321 init_emulate_ctxt(vcpu);
5325 ctxt->_eip = ctxt->eip + inc_eip;
5326 ret = emulate_int_real(ctxt, irq);
5328 if (ret != X86EMUL_CONTINUE)
5329 return EMULATE_FAIL;
5331 ctxt->eip = ctxt->_eip;
5332 kvm_rip_write(vcpu, ctxt->eip);
5333 kvm_set_rflags(vcpu, ctxt->eflags);
5335 if (irq == NMI_VECTOR)
5336 vcpu->arch.nmi_pending = 0;
5338 vcpu->arch.interrupt.pending = false;
5340 return EMULATE_DONE;
5342 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5344 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5346 int r = EMULATE_DONE;
5348 ++vcpu->stat.insn_emulation_fail;
5349 trace_kvm_emulate_insn_failed(vcpu);
5350 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5351 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5352 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5353 vcpu->run->internal.ndata = 0;
5356 kvm_queue_exception(vcpu, UD_VECTOR);
5361 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5362 bool write_fault_to_shadow_pgtable,
5368 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5371 if (!vcpu->arch.mmu.direct_map) {
5373 * Write permission should be allowed since only
5374 * write access need to be emulated.
5376 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5379 * If the mapping is invalid in guest, let cpu retry
5380 * it to generate fault.
5382 if (gpa == UNMAPPED_GVA)
5387 * Do not retry the unhandleable instruction if it faults on the
5388 * readonly host memory, otherwise it will goto a infinite loop:
5389 * retry instruction -> write #PF -> emulation fail -> retry
5390 * instruction -> ...
5392 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5395 * If the instruction failed on the error pfn, it can not be fixed,
5396 * report the error to userspace.
5398 if (is_error_noslot_pfn(pfn))
5401 kvm_release_pfn_clean(pfn);
5403 /* The instructions are well-emulated on direct mmu. */
5404 if (vcpu->arch.mmu.direct_map) {
5405 unsigned int indirect_shadow_pages;
5407 spin_lock(&vcpu->kvm->mmu_lock);
5408 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5409 spin_unlock(&vcpu->kvm->mmu_lock);
5411 if (indirect_shadow_pages)
5412 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5418 * if emulation was due to access to shadowed page table
5419 * and it failed try to unshadow page and re-enter the
5420 * guest to let CPU execute the instruction.
5422 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5425 * If the access faults on its page table, it can not
5426 * be fixed by unprotecting shadow page and it should
5427 * be reported to userspace.
5429 return !write_fault_to_shadow_pgtable;
5432 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5433 unsigned long cr2, int emulation_type)
5435 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5436 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5438 last_retry_eip = vcpu->arch.last_retry_eip;
5439 last_retry_addr = vcpu->arch.last_retry_addr;
5442 * If the emulation is caused by #PF and it is non-page_table
5443 * writing instruction, it means the VM-EXIT is caused by shadow
5444 * page protected, we can zap the shadow page and retry this
5445 * instruction directly.
5447 * Note: if the guest uses a non-page-table modifying instruction
5448 * on the PDE that points to the instruction, then we will unmap
5449 * the instruction and go to an infinite loop. So, we cache the
5450 * last retried eip and the last fault address, if we meet the eip
5451 * and the address again, we can break out of the potential infinite
5454 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5456 if (!(emulation_type & EMULTYPE_RETRY))
5459 if (x86_page_table_writing_insn(ctxt))
5462 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5465 vcpu->arch.last_retry_eip = ctxt->eip;
5466 vcpu->arch.last_retry_addr = cr2;
5468 if (!vcpu->arch.mmu.direct_map)
5469 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5471 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5476 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5477 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5479 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5481 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5482 /* This is a good place to trace that we are exiting SMM. */
5483 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5485 if (unlikely(vcpu->arch.smi_pending)) {
5486 kvm_make_request(KVM_REQ_SMI, vcpu);
5487 vcpu->arch.smi_pending = 0;
5489 /* Process a latched INIT, if any. */
5490 kvm_make_request(KVM_REQ_EVENT, vcpu);
5494 kvm_mmu_reset_context(vcpu);
5497 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5499 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5501 vcpu->arch.hflags = emul_flags;
5503 if (changed & HF_SMM_MASK)
5504 kvm_smm_changed(vcpu);
5507 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5516 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5517 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5522 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5524 struct kvm_run *kvm_run = vcpu->run;
5527 * rflags is the old, "raw" value of the flags. The new value has
5528 * not been saved yet.
5530 * This is correct even for TF set by the guest, because "the
5531 * processor will not generate this exception after the instruction
5532 * that sets the TF flag".
5534 if (unlikely(rflags & X86_EFLAGS_TF)) {
5535 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5536 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5538 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5539 kvm_run->debug.arch.exception = DB_VECTOR;
5540 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5541 *r = EMULATE_USER_EXIT;
5543 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5545 * "Certain debug exceptions may clear bit 0-3. The
5546 * remaining contents of the DR6 register are never
5547 * cleared by the processor".
5549 vcpu->arch.dr6 &= ~15;
5550 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5551 kvm_queue_exception(vcpu, DB_VECTOR);
5556 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5558 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5559 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5560 struct kvm_run *kvm_run = vcpu->run;
5561 unsigned long eip = kvm_get_linear_rip(vcpu);
5562 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5563 vcpu->arch.guest_debug_dr7,
5567 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5568 kvm_run->debug.arch.pc = eip;
5569 kvm_run->debug.arch.exception = DB_VECTOR;
5570 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5571 *r = EMULATE_USER_EXIT;
5576 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5577 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5578 unsigned long eip = kvm_get_linear_rip(vcpu);
5579 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5584 vcpu->arch.dr6 &= ~15;
5585 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5586 kvm_queue_exception(vcpu, DB_VECTOR);
5595 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5602 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5603 bool writeback = true;
5604 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5607 * Clear write_fault_to_shadow_pgtable here to ensure it is
5610 vcpu->arch.write_fault_to_shadow_pgtable = false;
5611 kvm_clear_exception_queue(vcpu);
5613 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5614 init_emulate_ctxt(vcpu);
5617 * We will reenter on the same instruction since
5618 * we do not set complete_userspace_io. This does not
5619 * handle watchpoints yet, those would be handled in
5622 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5625 ctxt->interruptibility = 0;
5626 ctxt->have_exception = false;
5627 ctxt->exception.vector = -1;
5628 ctxt->perm_ok = false;
5630 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5632 r = x86_decode_insn(ctxt, insn, insn_len);
5634 trace_kvm_emulate_insn_start(vcpu);
5635 ++vcpu->stat.insn_emulation;
5636 if (r != EMULATION_OK) {
5637 if (emulation_type & EMULTYPE_TRAP_UD)
5638 return EMULATE_FAIL;
5639 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5641 return EMULATE_DONE;
5642 if (emulation_type & EMULTYPE_SKIP)
5643 return EMULATE_FAIL;
5644 return handle_emulation_failure(vcpu);
5648 if (emulation_type & EMULTYPE_SKIP) {
5649 kvm_rip_write(vcpu, ctxt->_eip);
5650 if (ctxt->eflags & X86_EFLAGS_RF)
5651 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5652 return EMULATE_DONE;
5655 if (retry_instruction(ctxt, cr2, emulation_type))
5656 return EMULATE_DONE;
5658 /* this is needed for vmware backdoor interface to work since it
5659 changes registers values during IO operation */
5660 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5661 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5662 emulator_invalidate_register_cache(ctxt);
5666 r = x86_emulate_insn(ctxt);
5668 if (r == EMULATION_INTERCEPTED)
5669 return EMULATE_DONE;
5671 if (r == EMULATION_FAILED) {
5672 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5674 return EMULATE_DONE;
5676 return handle_emulation_failure(vcpu);
5679 if (ctxt->have_exception) {
5681 if (inject_emulated_exception(vcpu))
5683 } else if (vcpu->arch.pio.count) {
5684 if (!vcpu->arch.pio.in) {
5685 /* FIXME: return into emulator if single-stepping. */
5686 vcpu->arch.pio.count = 0;
5689 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5691 r = EMULATE_USER_EXIT;
5692 } else if (vcpu->mmio_needed) {
5693 if (!vcpu->mmio_is_write)
5695 r = EMULATE_USER_EXIT;
5696 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5697 } else if (r == EMULATION_RESTART)
5703 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5704 toggle_interruptibility(vcpu, ctxt->interruptibility);
5705 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5706 if (vcpu->arch.hflags != ctxt->emul_flags)
5707 kvm_set_hflags(vcpu, ctxt->emul_flags);
5708 kvm_rip_write(vcpu, ctxt->eip);
5709 if (r == EMULATE_DONE)
5710 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5711 if (!ctxt->have_exception ||
5712 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5713 __kvm_set_rflags(vcpu, ctxt->eflags);
5716 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5717 * do nothing, and it will be requested again as soon as
5718 * the shadow expires. But we still need to check here,
5719 * because POPF has no interrupt shadow.
5721 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5722 kvm_make_request(KVM_REQ_EVENT, vcpu);
5724 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5728 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5730 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5732 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5733 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5734 size, port, &val, 1);
5735 /* do not return to emulator after return from userspace */
5736 vcpu->arch.pio.count = 0;
5739 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5741 static void tsc_bad(void *info)
5743 __this_cpu_write(cpu_tsc_khz, 0);
5746 static void tsc_khz_changed(void *data)
5748 struct cpufreq_freqs *freq = data;
5749 unsigned long khz = 0;
5753 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5754 khz = cpufreq_quick_get(raw_smp_processor_id());
5757 __this_cpu_write(cpu_tsc_khz, khz);
5760 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5763 struct cpufreq_freqs *freq = data;
5765 struct kvm_vcpu *vcpu;
5766 int i, send_ipi = 0;
5769 * We allow guests to temporarily run on slowing clocks,
5770 * provided we notify them after, or to run on accelerating
5771 * clocks, provided we notify them before. Thus time never
5774 * However, we have a problem. We can't atomically update
5775 * the frequency of a given CPU from this function; it is
5776 * merely a notifier, which can be called from any CPU.
5777 * Changing the TSC frequency at arbitrary points in time
5778 * requires a recomputation of local variables related to
5779 * the TSC for each VCPU. We must flag these local variables
5780 * to be updated and be sure the update takes place with the
5781 * new frequency before any guests proceed.
5783 * Unfortunately, the combination of hotplug CPU and frequency
5784 * change creates an intractable locking scenario; the order
5785 * of when these callouts happen is undefined with respect to
5786 * CPU hotplug, and they can race with each other. As such,
5787 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5788 * undefined; you can actually have a CPU frequency change take
5789 * place in between the computation of X and the setting of the
5790 * variable. To protect against this problem, all updates of
5791 * the per_cpu tsc_khz variable are done in an interrupt
5792 * protected IPI, and all callers wishing to update the value
5793 * must wait for a synchronous IPI to complete (which is trivial
5794 * if the caller is on the CPU already). This establishes the
5795 * necessary total order on variable updates.
5797 * Note that because a guest time update may take place
5798 * anytime after the setting of the VCPU's request bit, the
5799 * correct TSC value must be set before the request. However,
5800 * to ensure the update actually makes it to any guest which
5801 * starts running in hardware virtualization between the set
5802 * and the acquisition of the spinlock, we must also ping the
5803 * CPU after setting the request bit.
5807 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5809 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5812 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5814 spin_lock(&kvm_lock);
5815 list_for_each_entry(kvm, &vm_list, vm_list) {
5816 kvm_for_each_vcpu(i, vcpu, kvm) {
5817 if (vcpu->cpu != freq->cpu)
5819 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5820 if (vcpu->cpu != smp_processor_id())
5824 spin_unlock(&kvm_lock);
5826 if (freq->old < freq->new && send_ipi) {
5828 * We upscale the frequency. Must make the guest
5829 * doesn't see old kvmclock values while running with
5830 * the new frequency, otherwise we risk the guest sees
5831 * time go backwards.
5833 * In case we update the frequency for another cpu
5834 * (which might be in guest context) send an interrupt
5835 * to kick the cpu out of guest context. Next time
5836 * guest context is entered kvmclock will be updated,
5837 * so the guest will not see stale values.
5839 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5844 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5845 .notifier_call = kvmclock_cpufreq_notifier
5848 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5849 unsigned long action, void *hcpu)
5851 unsigned int cpu = (unsigned long)hcpu;
5855 case CPU_DOWN_FAILED:
5856 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5858 case CPU_DOWN_PREPARE:
5859 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5865 static struct notifier_block kvmclock_cpu_notifier_block = {
5866 .notifier_call = kvmclock_cpu_notifier,
5867 .priority = -INT_MAX
5870 static void kvm_timer_init(void)
5874 max_tsc_khz = tsc_khz;
5876 cpu_notifier_register_begin();
5877 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5878 #ifdef CONFIG_CPU_FREQ
5879 struct cpufreq_policy policy;
5880 memset(&policy, 0, sizeof(policy));
5882 cpufreq_get_policy(&policy, cpu);
5883 if (policy.cpuinfo.max_freq)
5884 max_tsc_khz = policy.cpuinfo.max_freq;
5887 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5888 CPUFREQ_TRANSITION_NOTIFIER);
5890 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5891 for_each_online_cpu(cpu)
5892 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5894 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5895 cpu_notifier_register_done();
5899 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5901 int kvm_is_in_guest(void)
5903 return __this_cpu_read(current_vcpu) != NULL;
5906 static int kvm_is_user_mode(void)
5910 if (__this_cpu_read(current_vcpu))
5911 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5913 return user_mode != 0;
5916 static unsigned long kvm_get_guest_ip(void)
5918 unsigned long ip = 0;
5920 if (__this_cpu_read(current_vcpu))
5921 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5926 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5927 .is_in_guest = kvm_is_in_guest,
5928 .is_user_mode = kvm_is_user_mode,
5929 .get_guest_ip = kvm_get_guest_ip,
5932 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5934 __this_cpu_write(current_vcpu, vcpu);
5936 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5938 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5940 __this_cpu_write(current_vcpu, NULL);
5942 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5944 static void kvm_set_mmio_spte_mask(void)
5947 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5950 * Set the reserved bits and the present bit of an paging-structure
5951 * entry to generate page fault with PFER.RSV = 1.
5953 /* Mask the reserved physical address bits. */
5954 mask = rsvd_bits(maxphyaddr, 51);
5956 /* Bit 62 is always reserved for 32bit host. */
5957 mask |= 0x3ull << 62;
5959 /* Set the present bit. */
5962 #ifdef CONFIG_X86_64
5964 * If reserved bit is not supported, clear the present bit to disable
5967 if (maxphyaddr == 52)
5971 kvm_mmu_set_mmio_spte_mask(mask);
5974 #ifdef CONFIG_X86_64
5975 static void pvclock_gtod_update_fn(struct work_struct *work)
5979 struct kvm_vcpu *vcpu;
5982 spin_lock(&kvm_lock);
5983 list_for_each_entry(kvm, &vm_list, vm_list)
5984 kvm_for_each_vcpu(i, vcpu, kvm)
5985 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5986 atomic_set(&kvm_guest_has_master_clock, 0);
5987 spin_unlock(&kvm_lock);
5990 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5993 * Notification about pvclock gtod data update.
5995 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5998 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5999 struct timekeeper *tk = priv;
6001 update_pvclock_gtod(tk);
6003 /* disable master clock if host does not trust, or does not
6004 * use, TSC clocksource
6006 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6007 atomic_read(&kvm_guest_has_master_clock) != 0)
6008 queue_work(system_long_wq, &pvclock_gtod_work);
6013 static struct notifier_block pvclock_gtod_notifier = {
6014 .notifier_call = pvclock_gtod_notify,
6018 int kvm_arch_init(void *opaque)
6021 struct kvm_x86_ops *ops = opaque;
6024 printk(KERN_ERR "kvm: already loaded the other module\n");
6029 if (!ops->cpu_has_kvm_support()) {
6030 printk(KERN_ERR "kvm: no hardware support\n");
6034 if (ops->disabled_by_bios()) {
6035 printk(KERN_ERR "kvm: disabled by bios\n");
6041 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6043 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6047 r = kvm_mmu_module_init();
6049 goto out_free_percpu;
6051 kvm_set_mmio_spte_mask();
6055 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6056 PT_DIRTY_MASK, PT64_NX_MASK, 0);
6060 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6063 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6066 #ifdef CONFIG_X86_64
6067 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6073 free_percpu(shared_msrs);
6078 void kvm_arch_exit(void)
6080 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6082 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6083 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6084 CPUFREQ_TRANSITION_NOTIFIER);
6085 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
6086 #ifdef CONFIG_X86_64
6087 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6090 kvm_mmu_module_exit();
6091 free_percpu(shared_msrs);
6094 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6096 ++vcpu->stat.halt_exits;
6097 if (irqchip_in_kernel(vcpu->kvm)) {
6098 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6101 vcpu->run->exit_reason = KVM_EXIT_HLT;
6105 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6107 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6109 kvm_x86_ops->skip_emulated_instruction(vcpu);
6110 return kvm_vcpu_halt(vcpu);
6112 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6114 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
6116 u64 param, ingpa, outgpa, ret;
6117 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
6118 bool fast, longmode;
6121 * hypercall generates UD from non zero cpl and real mode
6124 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
6125 kvm_queue_exception(vcpu, UD_VECTOR);
6129 longmode = is_64_bit_mode(vcpu);
6132 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
6133 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
6134 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
6135 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
6136 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
6137 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
6139 #ifdef CONFIG_X86_64
6141 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
6142 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
6143 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
6147 code = param & 0xffff;
6148 fast = (param >> 16) & 0x1;
6149 rep_cnt = (param >> 32) & 0xfff;
6150 rep_idx = (param >> 48) & 0xfff;
6152 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
6155 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
6156 kvm_vcpu_on_spin(vcpu);
6159 res = HV_STATUS_INVALID_HYPERCALL_CODE;
6163 ret = res | (((u64)rep_done & 0xfff) << 32);
6165 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6167 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
6168 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
6175 * kvm_pv_kick_cpu_op: Kick a vcpu.
6177 * @apicid - apicid of vcpu to be kicked.
6179 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6181 struct kvm_lapic_irq lapic_irq;
6183 lapic_irq.shorthand = 0;
6184 lapic_irq.dest_mode = 0;
6185 lapic_irq.dest_id = apicid;
6186 lapic_irq.msi_redir_hint = false;
6188 lapic_irq.delivery_mode = APIC_DM_REMRD;
6189 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6192 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6194 unsigned long nr, a0, a1, a2, a3, ret;
6195 int op_64_bit, r = 1;
6197 kvm_x86_ops->skip_emulated_instruction(vcpu);
6199 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6200 return kvm_hv_hypercall(vcpu);
6202 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6203 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6204 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6205 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6206 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6208 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6210 op_64_bit = is_64_bit_mode(vcpu);
6219 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6225 case KVM_HC_VAPIC_POLL_IRQ:
6228 case KVM_HC_KICK_CPU:
6229 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6239 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6240 ++vcpu->stat.hypercalls;
6243 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6245 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6247 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6248 char instruction[3];
6249 unsigned long rip = kvm_rip_read(vcpu);
6251 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6253 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
6257 * Check if userspace requested an interrupt window, and that the
6258 * interrupt window is open.
6260 * No need to exit to userspace if we already have an interrupt queued.
6262 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6264 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
6265 vcpu->run->request_interrupt_window &&
6266 kvm_arch_interrupt_allowed(vcpu));
6269 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6271 struct kvm_run *kvm_run = vcpu->run;
6273 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6274 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6275 kvm_run->cr8 = kvm_get_cr8(vcpu);
6276 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6277 if (irqchip_in_kernel(vcpu->kvm))
6278 kvm_run->ready_for_interrupt_injection = 1;
6280 kvm_run->ready_for_interrupt_injection =
6281 kvm_arch_interrupt_allowed(vcpu) &&
6282 !kvm_cpu_has_interrupt(vcpu) &&
6283 !kvm_event_needs_reinjection(vcpu);
6286 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6290 if (!kvm_x86_ops->update_cr8_intercept)
6293 if (!vcpu->arch.apic)
6296 if (!vcpu->arch.apic->vapic_addr)
6297 max_irr = kvm_lapic_find_highest_irr(vcpu);
6304 tpr = kvm_lapic_get_cr8(vcpu);
6306 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6309 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6313 /* try to reinject previous events if any */
6314 if (vcpu->arch.exception.pending) {
6315 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6316 vcpu->arch.exception.has_error_code,
6317 vcpu->arch.exception.error_code);
6319 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6320 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6323 if (vcpu->arch.exception.nr == DB_VECTOR &&
6324 (vcpu->arch.dr7 & DR7_GD)) {
6325 vcpu->arch.dr7 &= ~DR7_GD;
6326 kvm_update_dr7(vcpu);
6329 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6330 vcpu->arch.exception.has_error_code,
6331 vcpu->arch.exception.error_code,
6332 vcpu->arch.exception.reinject);
6336 if (vcpu->arch.nmi_injected) {
6337 kvm_x86_ops->set_nmi(vcpu);
6341 if (vcpu->arch.interrupt.pending) {
6342 kvm_x86_ops->set_irq(vcpu);
6346 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6347 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6352 /* try to inject new event if pending */
6353 if (vcpu->arch.nmi_pending) {
6354 if (kvm_x86_ops->nmi_allowed(vcpu)) {
6355 --vcpu->arch.nmi_pending;
6356 vcpu->arch.nmi_injected = true;
6357 kvm_x86_ops->set_nmi(vcpu);
6359 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6361 * Because interrupts can be injected asynchronously, we are
6362 * calling check_nested_events again here to avoid a race condition.
6363 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6364 * proposal and current concerns. Perhaps we should be setting
6365 * KVM_REQ_EVENT only on certain events and not unconditionally?
6367 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6368 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6372 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6373 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6375 kvm_x86_ops->set_irq(vcpu);
6381 static void process_nmi(struct kvm_vcpu *vcpu)
6386 * x86 is limited to one NMI running, and one NMI pending after it.
6387 * If an NMI is already in progress, limit further NMIs to just one.
6388 * Otherwise, allow two (and we'll inject the first one immediately).
6390 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6393 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6394 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6395 kvm_make_request(KVM_REQ_EVENT, vcpu);
6398 #define put_smstate(type, buf, offset, val) \
6399 *(type *)((buf) + (offset) - 0x7e00) = val
6401 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
6404 flags |= seg->g << 23;
6405 flags |= seg->db << 22;
6406 flags |= seg->l << 21;
6407 flags |= seg->avl << 20;
6408 flags |= seg->present << 15;
6409 flags |= seg->dpl << 13;
6410 flags |= seg->s << 12;
6411 flags |= seg->type << 8;
6415 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6417 struct kvm_segment seg;
6420 kvm_get_segment(vcpu, &seg, n);
6421 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6424 offset = 0x7f84 + n * 12;
6426 offset = 0x7f2c + (n - 3) * 12;
6428 put_smstate(u32, buf, offset + 8, seg.base);
6429 put_smstate(u32, buf, offset + 4, seg.limit);
6430 put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
6433 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6435 struct kvm_segment seg;
6439 kvm_get_segment(vcpu, &seg, n);
6440 offset = 0x7e00 + n * 16;
6442 flags = process_smi_get_segment_flags(&seg) >> 8;
6443 put_smstate(u16, buf, offset, seg.selector);
6444 put_smstate(u16, buf, offset + 2, flags);
6445 put_smstate(u32, buf, offset + 4, seg.limit);
6446 put_smstate(u64, buf, offset + 8, seg.base);
6449 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6452 struct kvm_segment seg;
6456 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6457 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6458 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6459 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6461 for (i = 0; i < 8; i++)
6462 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6464 kvm_get_dr(vcpu, 6, &val);
6465 put_smstate(u32, buf, 0x7fcc, (u32)val);
6466 kvm_get_dr(vcpu, 7, &val);
6467 put_smstate(u32, buf, 0x7fc8, (u32)val);
6469 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6470 put_smstate(u32, buf, 0x7fc4, seg.selector);
6471 put_smstate(u32, buf, 0x7f64, seg.base);
6472 put_smstate(u32, buf, 0x7f60, seg.limit);
6473 put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
6475 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6476 put_smstate(u32, buf, 0x7fc0, seg.selector);
6477 put_smstate(u32, buf, 0x7f80, seg.base);
6478 put_smstate(u32, buf, 0x7f7c, seg.limit);
6479 put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6481 kvm_x86_ops->get_gdt(vcpu, &dt);
6482 put_smstate(u32, buf, 0x7f74, dt.address);
6483 put_smstate(u32, buf, 0x7f70, dt.size);
6485 kvm_x86_ops->get_idt(vcpu, &dt);
6486 put_smstate(u32, buf, 0x7f58, dt.address);
6487 put_smstate(u32, buf, 0x7f54, dt.size);
6489 for (i = 0; i < 6; i++)
6490 process_smi_save_seg_32(vcpu, buf, i);
6492 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6495 put_smstate(u32, buf, 0x7efc, 0x00020000);
6496 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6499 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6501 #ifdef CONFIG_X86_64
6503 struct kvm_segment seg;
6507 for (i = 0; i < 16; i++)
6508 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6510 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6511 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6513 kvm_get_dr(vcpu, 6, &val);
6514 put_smstate(u64, buf, 0x7f68, val);
6515 kvm_get_dr(vcpu, 7, &val);
6516 put_smstate(u64, buf, 0x7f60, val);
6518 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6519 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6520 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6522 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6525 put_smstate(u32, buf, 0x7efc, 0x00020064);
6527 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6529 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6530 put_smstate(u16, buf, 0x7e90, seg.selector);
6531 put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6532 put_smstate(u32, buf, 0x7e94, seg.limit);
6533 put_smstate(u64, buf, 0x7e98, seg.base);
6535 kvm_x86_ops->get_idt(vcpu, &dt);
6536 put_smstate(u32, buf, 0x7e84, dt.size);
6537 put_smstate(u64, buf, 0x7e88, dt.address);
6539 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6540 put_smstate(u16, buf, 0x7e70, seg.selector);
6541 put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6542 put_smstate(u32, buf, 0x7e74, seg.limit);
6543 put_smstate(u64, buf, 0x7e78, seg.base);
6545 kvm_x86_ops->get_gdt(vcpu, &dt);
6546 put_smstate(u32, buf, 0x7e64, dt.size);
6547 put_smstate(u64, buf, 0x7e68, dt.address);
6549 for (i = 0; i < 6; i++)
6550 process_smi_save_seg_64(vcpu, buf, i);
6556 static void process_smi(struct kvm_vcpu *vcpu)
6558 struct kvm_segment cs, ds;
6563 vcpu->arch.smi_pending = true;
6567 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6568 vcpu->arch.hflags |= HF_SMM_MASK;
6569 memset(buf, 0, 512);
6570 if (guest_cpuid_has_longmode(vcpu))
6571 process_smi_save_state_64(vcpu, buf);
6573 process_smi_save_state_32(vcpu, buf);
6575 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6577 if (kvm_x86_ops->get_nmi_mask(vcpu))
6578 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6580 kvm_x86_ops->set_nmi_mask(vcpu, true);
6582 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6583 kvm_rip_write(vcpu, 0x8000);
6585 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6586 kvm_x86_ops->set_cr0(vcpu, cr0);
6587 vcpu->arch.cr0 = cr0;
6589 kvm_x86_ops->set_cr4(vcpu, 0);
6591 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6593 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6594 cs.base = vcpu->arch.smbase;
6599 cs.limit = ds.limit = 0xffffffff;
6600 cs.type = ds.type = 0x3;
6601 cs.dpl = ds.dpl = 0;
6606 cs.avl = ds.avl = 0;
6607 cs.present = ds.present = 1;
6608 cs.unusable = ds.unusable = 0;
6609 cs.padding = ds.padding = 0;
6611 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6612 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6613 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6614 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6615 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6616 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6618 if (guest_cpuid_has_longmode(vcpu))
6619 kvm_x86_ops->set_efer(vcpu, 0);
6621 kvm_update_cpuid(vcpu);
6622 kvm_mmu_reset_context(vcpu);
6625 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6627 u64 eoi_exit_bitmap[4];
6630 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6633 memset(eoi_exit_bitmap, 0, 32);
6636 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6637 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6638 kvm_apic_update_tmr(vcpu, tmr);
6641 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6643 ++vcpu->stat.tlb_flush;
6644 kvm_x86_ops->tlb_flush(vcpu);
6647 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6649 struct page *page = NULL;
6651 if (!irqchip_in_kernel(vcpu->kvm))
6654 if (!kvm_x86_ops->set_apic_access_page_addr)
6657 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6658 if (is_error_page(page))
6660 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6663 * Do not pin apic access page in memory, the MMU notifier
6664 * will call us again if it is migrated or swapped out.
6668 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6670 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6671 unsigned long address)
6674 * The physical address of apic access page is stored in the VMCS.
6675 * Update it when it becomes invalid.
6677 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6678 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6682 * Returns 1 to let vcpu_run() continue the guest execution loop without
6683 * exiting to the userspace. Otherwise, the value will be returned to the
6686 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6689 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6690 vcpu->run->request_interrupt_window;
6691 bool req_immediate_exit = false;
6693 if (vcpu->requests) {
6694 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6695 kvm_mmu_unload(vcpu);
6696 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6697 __kvm_migrate_timers(vcpu);
6698 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6699 kvm_gen_update_masterclock(vcpu->kvm);
6700 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6701 kvm_gen_kvmclock_update(vcpu);
6702 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6703 r = kvm_guest_time_update(vcpu);
6707 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6708 kvm_mmu_sync_roots(vcpu);
6709 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6710 kvm_vcpu_flush_tlb(vcpu);
6711 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6712 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6716 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6717 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6721 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6722 vcpu->fpu_active = 0;
6723 kvm_x86_ops->fpu_deactivate(vcpu);
6725 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6726 /* Page is swapped out. Do synthetic halt */
6727 vcpu->arch.apf.halted = true;
6731 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6732 record_steal_time(vcpu);
6733 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6735 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6737 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6738 kvm_handle_pmu_event(vcpu);
6739 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6740 kvm_deliver_pmi(vcpu);
6741 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6742 vcpu_scan_ioapic(vcpu);
6743 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6744 kvm_vcpu_reload_apic_access_page(vcpu);
6747 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6748 kvm_apic_accept_events(vcpu);
6749 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6754 if (inject_pending_event(vcpu, req_int_win) != 0)
6755 req_immediate_exit = true;
6756 /* enable NMI/IRQ window open exits if needed */
6757 else if (vcpu->arch.nmi_pending)
6758 kvm_x86_ops->enable_nmi_window(vcpu);
6759 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6760 kvm_x86_ops->enable_irq_window(vcpu);
6762 if (kvm_lapic_enabled(vcpu)) {
6764 * Update architecture specific hints for APIC
6765 * virtual interrupt delivery.
6767 if (kvm_x86_ops->hwapic_irr_update)
6768 kvm_x86_ops->hwapic_irr_update(vcpu,
6769 kvm_lapic_find_highest_irr(vcpu));
6770 update_cr8_intercept(vcpu);
6771 kvm_lapic_sync_to_vapic(vcpu);
6775 r = kvm_mmu_reload(vcpu);
6777 goto cancel_injection;
6782 kvm_x86_ops->prepare_guest_switch(vcpu);
6783 if (vcpu->fpu_active)
6784 kvm_load_guest_fpu(vcpu);
6785 kvm_load_guest_xcr0(vcpu);
6787 vcpu->mode = IN_GUEST_MODE;
6789 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6791 /* We should set ->mode before check ->requests,
6792 * see the comment in make_all_cpus_request.
6794 smp_mb__after_srcu_read_unlock();
6796 local_irq_disable();
6798 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6799 || need_resched() || signal_pending(current)) {
6800 vcpu->mode = OUTSIDE_GUEST_MODE;
6804 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6806 goto cancel_injection;
6809 if (req_immediate_exit)
6810 smp_send_reschedule(vcpu->cpu);
6812 __kvm_guest_enter();
6814 if (unlikely(vcpu->arch.switch_db_regs)) {
6816 set_debugreg(vcpu->arch.eff_db[0], 0);
6817 set_debugreg(vcpu->arch.eff_db[1], 1);
6818 set_debugreg(vcpu->arch.eff_db[2], 2);
6819 set_debugreg(vcpu->arch.eff_db[3], 3);
6820 set_debugreg(vcpu->arch.dr6, 6);
6821 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6824 trace_kvm_entry(vcpu->vcpu_id);
6825 wait_lapic_expire(vcpu);
6826 kvm_x86_ops->run(vcpu);
6829 * Do this here before restoring debug registers on the host. And
6830 * since we do this before handling the vmexit, a DR access vmexit
6831 * can (a) read the correct value of the debug registers, (b) set
6832 * KVM_DEBUGREG_WONT_EXIT again.
6834 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6837 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6838 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6839 for (i = 0; i < KVM_NR_DB_REGS; i++)
6840 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6844 * If the guest has used debug registers, at least dr7
6845 * will be disabled while returning to the host.
6846 * If we don't have active breakpoints in the host, we don't
6847 * care about the messed up debug address registers. But if
6848 * we have some of them active, restore the old state.
6850 if (hw_breakpoint_active())
6851 hw_breakpoint_restore();
6853 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6856 vcpu->mode = OUTSIDE_GUEST_MODE;
6859 /* Interrupt is enabled by handle_external_intr() */
6860 kvm_x86_ops->handle_external_intr(vcpu);
6865 * We must have an instruction between local_irq_enable() and
6866 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6867 * the interrupt shadow. The stat.exits increment will do nicely.
6868 * But we need to prevent reordering, hence this barrier():
6876 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6879 * Profile KVM exit RIPs:
6881 if (unlikely(prof_on == KVM_PROFILING)) {
6882 unsigned long rip = kvm_rip_read(vcpu);
6883 profile_hit(KVM_PROFILING, (void *)rip);
6886 if (unlikely(vcpu->arch.tsc_always_catchup))
6887 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6889 if (vcpu->arch.apic_attention)
6890 kvm_lapic_sync_from_vapic(vcpu);
6892 r = kvm_x86_ops->handle_exit(vcpu);
6896 kvm_x86_ops->cancel_injection(vcpu);
6897 if (unlikely(vcpu->arch.apic_attention))
6898 kvm_lapic_sync_from_vapic(vcpu);
6903 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6905 if (!kvm_arch_vcpu_runnable(vcpu)) {
6906 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6907 kvm_vcpu_block(vcpu);
6908 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6909 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6913 kvm_apic_accept_events(vcpu);
6914 switch(vcpu->arch.mp_state) {
6915 case KVM_MP_STATE_HALTED:
6916 vcpu->arch.pv.pv_unhalted = false;
6917 vcpu->arch.mp_state =
6918 KVM_MP_STATE_RUNNABLE;
6919 case KVM_MP_STATE_RUNNABLE:
6920 vcpu->arch.apf.halted = false;
6922 case KVM_MP_STATE_INIT_RECEIVED:
6931 static int vcpu_run(struct kvm_vcpu *vcpu)
6934 struct kvm *kvm = vcpu->kvm;
6936 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6939 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6940 !vcpu->arch.apf.halted)
6941 r = vcpu_enter_guest(vcpu);
6943 r = vcpu_block(kvm, vcpu);
6947 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6948 if (kvm_cpu_has_pending_timer(vcpu))
6949 kvm_inject_pending_timer_irqs(vcpu);
6951 if (dm_request_for_irq_injection(vcpu)) {
6953 vcpu->run->exit_reason = KVM_EXIT_INTR;
6954 ++vcpu->stat.request_irq_exits;
6958 kvm_check_async_pf_completion(vcpu);
6960 if (signal_pending(current)) {
6962 vcpu->run->exit_reason = KVM_EXIT_INTR;
6963 ++vcpu->stat.signal_exits;
6966 if (need_resched()) {
6967 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6969 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6973 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6978 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6981 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6982 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6983 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6984 if (r != EMULATE_DONE)
6989 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6991 BUG_ON(!vcpu->arch.pio.count);
6993 return complete_emulated_io(vcpu);
6997 * Implements the following, as a state machine:
7001 * for each mmio piece in the fragment
7009 * for each mmio piece in the fragment
7014 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7016 struct kvm_run *run = vcpu->run;
7017 struct kvm_mmio_fragment *frag;
7020 BUG_ON(!vcpu->mmio_needed);
7022 /* Complete previous fragment */
7023 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7024 len = min(8u, frag->len);
7025 if (!vcpu->mmio_is_write)
7026 memcpy(frag->data, run->mmio.data, len);
7028 if (frag->len <= 8) {
7029 /* Switch to the next fragment. */
7031 vcpu->mmio_cur_fragment++;
7033 /* Go forward to the next mmio piece. */
7039 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7040 vcpu->mmio_needed = 0;
7042 /* FIXME: return into emulator if single-stepping. */
7043 if (vcpu->mmio_is_write)
7045 vcpu->mmio_read_completed = 1;
7046 return complete_emulated_io(vcpu);
7049 run->exit_reason = KVM_EXIT_MMIO;
7050 run->mmio.phys_addr = frag->gpa;
7051 if (vcpu->mmio_is_write)
7052 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7053 run->mmio.len = min(8u, frag->len);
7054 run->mmio.is_write = vcpu->mmio_is_write;
7055 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7060 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7065 if (!tsk_used_math(current) && init_fpu(current))
7068 if (vcpu->sigset_active)
7069 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
7071 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7072 kvm_vcpu_block(vcpu);
7073 kvm_apic_accept_events(vcpu);
7074 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
7079 /* re-sync apic's tpr */
7080 if (!irqchip_in_kernel(vcpu->kvm)) {
7081 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7087 if (unlikely(vcpu->arch.complete_userspace_io)) {
7088 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7089 vcpu->arch.complete_userspace_io = NULL;
7094 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7099 post_kvm_run_save(vcpu);
7100 if (vcpu->sigset_active)
7101 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
7106 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7108 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7110 * We are here if userspace calls get_regs() in the middle of
7111 * instruction emulation. Registers state needs to be copied
7112 * back from emulation context to vcpu. Userspace shouldn't do
7113 * that usually, but some bad designed PV devices (vmware
7114 * backdoor interface) need this to work
7116 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7117 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7119 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7120 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7121 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7122 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7123 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7124 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7125 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7126 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7127 #ifdef CONFIG_X86_64
7128 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7129 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7130 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7131 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7132 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7133 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7134 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7135 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7138 regs->rip = kvm_rip_read(vcpu);
7139 regs->rflags = kvm_get_rflags(vcpu);
7144 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7146 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7147 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7149 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7150 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7151 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7152 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7153 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7154 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7155 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7156 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7157 #ifdef CONFIG_X86_64
7158 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7159 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7160 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7161 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7162 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7163 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7164 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7165 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7168 kvm_rip_write(vcpu, regs->rip);
7169 kvm_set_rflags(vcpu, regs->rflags);
7171 vcpu->arch.exception.pending = false;
7173 kvm_make_request(KVM_REQ_EVENT, vcpu);
7178 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7180 struct kvm_segment cs;
7182 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7186 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7188 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7189 struct kvm_sregs *sregs)
7193 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7194 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7195 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7196 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7197 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7198 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7200 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7201 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7203 kvm_x86_ops->get_idt(vcpu, &dt);
7204 sregs->idt.limit = dt.size;
7205 sregs->idt.base = dt.address;
7206 kvm_x86_ops->get_gdt(vcpu, &dt);
7207 sregs->gdt.limit = dt.size;
7208 sregs->gdt.base = dt.address;
7210 sregs->cr0 = kvm_read_cr0(vcpu);
7211 sregs->cr2 = vcpu->arch.cr2;
7212 sregs->cr3 = kvm_read_cr3(vcpu);
7213 sregs->cr4 = kvm_read_cr4(vcpu);
7214 sregs->cr8 = kvm_get_cr8(vcpu);
7215 sregs->efer = vcpu->arch.efer;
7216 sregs->apic_base = kvm_get_apic_base(vcpu);
7218 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7220 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7221 set_bit(vcpu->arch.interrupt.nr,
7222 (unsigned long *)sregs->interrupt_bitmap);
7227 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7228 struct kvm_mp_state *mp_state)
7230 kvm_apic_accept_events(vcpu);
7231 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7232 vcpu->arch.pv.pv_unhalted)
7233 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7235 mp_state->mp_state = vcpu->arch.mp_state;
7240 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7241 struct kvm_mp_state *mp_state)
7243 if (!kvm_vcpu_has_lapic(vcpu) &&
7244 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7247 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7248 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7249 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7251 vcpu->arch.mp_state = mp_state->mp_state;
7252 kvm_make_request(KVM_REQ_EVENT, vcpu);
7256 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7257 int reason, bool has_error_code, u32 error_code)
7259 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7262 init_emulate_ctxt(vcpu);
7264 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7265 has_error_code, error_code);
7268 return EMULATE_FAIL;
7270 kvm_rip_write(vcpu, ctxt->eip);
7271 kvm_set_rflags(vcpu, ctxt->eflags);
7272 kvm_make_request(KVM_REQ_EVENT, vcpu);
7273 return EMULATE_DONE;
7275 EXPORT_SYMBOL_GPL(kvm_task_switch);
7277 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7278 struct kvm_sregs *sregs)
7280 struct msr_data apic_base_msr;
7281 int mmu_reset_needed = 0;
7282 int pending_vec, max_bits, idx;
7285 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7288 dt.size = sregs->idt.limit;
7289 dt.address = sregs->idt.base;
7290 kvm_x86_ops->set_idt(vcpu, &dt);
7291 dt.size = sregs->gdt.limit;
7292 dt.address = sregs->gdt.base;
7293 kvm_x86_ops->set_gdt(vcpu, &dt);
7295 vcpu->arch.cr2 = sregs->cr2;
7296 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7297 vcpu->arch.cr3 = sregs->cr3;
7298 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7300 kvm_set_cr8(vcpu, sregs->cr8);
7302 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7303 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7304 apic_base_msr.data = sregs->apic_base;
7305 apic_base_msr.host_initiated = true;
7306 kvm_set_apic_base(vcpu, &apic_base_msr);
7308 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7309 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7310 vcpu->arch.cr0 = sregs->cr0;
7312 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7313 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7314 if (sregs->cr4 & X86_CR4_OSXSAVE)
7315 kvm_update_cpuid(vcpu);
7317 idx = srcu_read_lock(&vcpu->kvm->srcu);
7318 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7319 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7320 mmu_reset_needed = 1;
7322 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7324 if (mmu_reset_needed)
7325 kvm_mmu_reset_context(vcpu);
7327 max_bits = KVM_NR_INTERRUPTS;
7328 pending_vec = find_first_bit(
7329 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7330 if (pending_vec < max_bits) {
7331 kvm_queue_interrupt(vcpu, pending_vec, false);
7332 pr_debug("Set back pending irq %d\n", pending_vec);
7335 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7336 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7337 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7338 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7339 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7340 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7342 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7343 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7345 update_cr8_intercept(vcpu);
7347 /* Older userspace won't unhalt the vcpu on reset. */
7348 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7349 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7351 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7353 kvm_make_request(KVM_REQ_EVENT, vcpu);
7358 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7359 struct kvm_guest_debug *dbg)
7361 unsigned long rflags;
7364 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7366 if (vcpu->arch.exception.pending)
7368 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7369 kvm_queue_exception(vcpu, DB_VECTOR);
7371 kvm_queue_exception(vcpu, BP_VECTOR);
7375 * Read rflags as long as potentially injected trace flags are still
7378 rflags = kvm_get_rflags(vcpu);
7380 vcpu->guest_debug = dbg->control;
7381 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7382 vcpu->guest_debug = 0;
7384 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7385 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7386 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7387 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7389 for (i = 0; i < KVM_NR_DB_REGS; i++)
7390 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7392 kvm_update_dr7(vcpu);
7394 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7395 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7396 get_segment_base(vcpu, VCPU_SREG_CS);
7399 * Trigger an rflags update that will inject or remove the trace
7402 kvm_set_rflags(vcpu, rflags);
7404 kvm_x86_ops->update_db_bp_intercept(vcpu);
7414 * Translate a guest virtual address to a guest physical address.
7416 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7417 struct kvm_translation *tr)
7419 unsigned long vaddr = tr->linear_address;
7423 idx = srcu_read_lock(&vcpu->kvm->srcu);
7424 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7425 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7426 tr->physical_address = gpa;
7427 tr->valid = gpa != UNMAPPED_GVA;
7434 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7436 struct i387_fxsave_struct *fxsave =
7437 &vcpu->arch.guest_fpu.state->fxsave;
7439 memcpy(fpu->fpr, fxsave->st_space, 128);
7440 fpu->fcw = fxsave->cwd;
7441 fpu->fsw = fxsave->swd;
7442 fpu->ftwx = fxsave->twd;
7443 fpu->last_opcode = fxsave->fop;
7444 fpu->last_ip = fxsave->rip;
7445 fpu->last_dp = fxsave->rdp;
7446 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7451 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7453 struct i387_fxsave_struct *fxsave =
7454 &vcpu->arch.guest_fpu.state->fxsave;
7456 memcpy(fxsave->st_space, fpu->fpr, 128);
7457 fxsave->cwd = fpu->fcw;
7458 fxsave->swd = fpu->fsw;
7459 fxsave->twd = fpu->ftwx;
7460 fxsave->fop = fpu->last_opcode;
7461 fxsave->rip = fpu->last_ip;
7462 fxsave->rdp = fpu->last_dp;
7463 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7468 int fx_init(struct kvm_vcpu *vcpu, bool init_event)
7472 err = fpu_alloc(&vcpu->arch.guest_fpu);
7477 fpu_finit(&vcpu->arch.guest_fpu);
7480 vcpu->arch.guest_fpu.state->xsave.xsave_hdr.xcomp_bv =
7481 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7484 * Ensure guest xcr0 is valid for loading
7486 vcpu->arch.xcr0 = XSTATE_FP;
7488 vcpu->arch.cr0 |= X86_CR0_ET;
7492 EXPORT_SYMBOL_GPL(fx_init);
7494 static void fx_free(struct kvm_vcpu *vcpu)
7496 fpu_free(&vcpu->arch.guest_fpu);
7499 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7501 if (vcpu->guest_fpu_loaded)
7505 * Restore all possible states in the guest,
7506 * and assume host would use all available bits.
7507 * Guest xcr0 would be loaded later.
7509 kvm_put_guest_xcr0(vcpu);
7510 vcpu->guest_fpu_loaded = 1;
7511 __kernel_fpu_begin();
7512 fpu_restore_checking(&vcpu->arch.guest_fpu);
7516 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7518 kvm_put_guest_xcr0(vcpu);
7520 if (!vcpu->guest_fpu_loaded) {
7521 vcpu->fpu_counter = 0;
7525 vcpu->guest_fpu_loaded = 0;
7526 fpu_save_init(&vcpu->arch.guest_fpu);
7528 ++vcpu->stat.fpu_reload;
7530 * If using eager FPU mode, or if the guest is a frequent user
7531 * of the FPU, just leave the FPU active for next time.
7532 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7533 * the FPU in bursts will revert to loading it on demand.
7535 if (!vcpu->arch.eager_fpu) {
7536 if (++vcpu->fpu_counter < 5)
7537 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7542 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7544 kvmclock_reset(vcpu);
7546 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7548 kvm_x86_ops->vcpu_free(vcpu);
7551 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7554 struct kvm_vcpu *vcpu;
7556 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7557 printk_once(KERN_WARNING
7558 "kvm: SMP vm created on host with unstable TSC; "
7559 "guest TSC will not be reliable\n");
7561 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7564 * Activate fpu unconditionally in case the guest needs eager FPU. It will be
7565 * deactivated soon if it doesn't.
7567 kvm_x86_ops->fpu_activate(vcpu);
7571 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7575 vcpu->arch.mtrr_state.have_fixed = 1;
7576 r = vcpu_load(vcpu);
7579 kvm_vcpu_reset(vcpu, false);
7580 kvm_mmu_setup(vcpu);
7586 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7588 struct msr_data msr;
7589 struct kvm *kvm = vcpu->kvm;
7591 if (vcpu_load(vcpu))
7594 msr.index = MSR_IA32_TSC;
7595 msr.host_initiated = true;
7596 kvm_write_tsc(vcpu, &msr);
7599 if (!kvmclock_periodic_sync)
7602 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7603 KVMCLOCK_SYNC_PERIOD);
7606 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7609 vcpu->arch.apf.msr_val = 0;
7611 r = vcpu_load(vcpu);
7613 kvm_mmu_unload(vcpu);
7617 kvm_x86_ops->vcpu_free(vcpu);
7620 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7622 vcpu->arch.hflags = 0;
7624 atomic_set(&vcpu->arch.nmi_queued, 0);
7625 vcpu->arch.nmi_pending = 0;
7626 vcpu->arch.nmi_injected = false;
7627 kvm_clear_interrupt_queue(vcpu);
7628 kvm_clear_exception_queue(vcpu);
7630 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7631 kvm_update_dr0123(vcpu);
7632 vcpu->arch.dr6 = DR6_INIT;
7633 kvm_update_dr6(vcpu);
7634 vcpu->arch.dr7 = DR7_FIXED_1;
7635 kvm_update_dr7(vcpu);
7639 kvm_make_request(KVM_REQ_EVENT, vcpu);
7640 vcpu->arch.apf.msr_val = 0;
7641 vcpu->arch.st.msr_val = 0;
7643 kvmclock_reset(vcpu);
7645 kvm_clear_async_pf_completion_queue(vcpu);
7646 kvm_async_pf_hash_reset(vcpu);
7647 vcpu->arch.apf.halted = false;
7650 kvm_pmu_reset(vcpu);
7651 vcpu->arch.smbase = 0x30000;
7654 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7655 vcpu->arch.regs_avail = ~0;
7656 vcpu->arch.regs_dirty = ~0;
7658 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7661 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7663 struct kvm_segment cs;
7665 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7666 cs.selector = vector << 8;
7667 cs.base = vector << 12;
7668 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7669 kvm_rip_write(vcpu, 0);
7672 int kvm_arch_hardware_enable(void)
7675 struct kvm_vcpu *vcpu;
7680 bool stable, backwards_tsc = false;
7682 kvm_shared_msr_cpu_online();
7683 ret = kvm_x86_ops->hardware_enable();
7687 local_tsc = native_read_tsc();
7688 stable = !check_tsc_unstable();
7689 list_for_each_entry(kvm, &vm_list, vm_list) {
7690 kvm_for_each_vcpu(i, vcpu, kvm) {
7691 if (!stable && vcpu->cpu == smp_processor_id())
7692 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7693 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7694 backwards_tsc = true;
7695 if (vcpu->arch.last_host_tsc > max_tsc)
7696 max_tsc = vcpu->arch.last_host_tsc;
7702 * Sometimes, even reliable TSCs go backwards. This happens on
7703 * platforms that reset TSC during suspend or hibernate actions, but
7704 * maintain synchronization. We must compensate. Fortunately, we can
7705 * detect that condition here, which happens early in CPU bringup,
7706 * before any KVM threads can be running. Unfortunately, we can't
7707 * bring the TSCs fully up to date with real time, as we aren't yet far
7708 * enough into CPU bringup that we know how much real time has actually
7709 * elapsed; our helper function, get_kernel_ns() will be using boot
7710 * variables that haven't been updated yet.
7712 * So we simply find the maximum observed TSC above, then record the
7713 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7714 * the adjustment will be applied. Note that we accumulate
7715 * adjustments, in case multiple suspend cycles happen before some VCPU
7716 * gets a chance to run again. In the event that no KVM threads get a
7717 * chance to run, we will miss the entire elapsed period, as we'll have
7718 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7719 * loose cycle time. This isn't too big a deal, since the loss will be
7720 * uniform across all VCPUs (not to mention the scenario is extremely
7721 * unlikely). It is possible that a second hibernate recovery happens
7722 * much faster than a first, causing the observed TSC here to be
7723 * smaller; this would require additional padding adjustment, which is
7724 * why we set last_host_tsc to the local tsc observed here.
7726 * N.B. - this code below runs only on platforms with reliable TSC,
7727 * as that is the only way backwards_tsc is set above. Also note
7728 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7729 * have the same delta_cyc adjustment applied if backwards_tsc
7730 * is detected. Note further, this adjustment is only done once,
7731 * as we reset last_host_tsc on all VCPUs to stop this from being
7732 * called multiple times (one for each physical CPU bringup).
7734 * Platforms with unreliable TSCs don't have to deal with this, they
7735 * will be compensated by the logic in vcpu_load, which sets the TSC to
7736 * catchup mode. This will catchup all VCPUs to real time, but cannot
7737 * guarantee that they stay in perfect synchronization.
7739 if (backwards_tsc) {
7740 u64 delta_cyc = max_tsc - local_tsc;
7741 backwards_tsc_observed = true;
7742 list_for_each_entry(kvm, &vm_list, vm_list) {
7743 kvm_for_each_vcpu(i, vcpu, kvm) {
7744 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7745 vcpu->arch.last_host_tsc = local_tsc;
7746 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7750 * We have to disable TSC offset matching.. if you were
7751 * booting a VM while issuing an S4 host suspend....
7752 * you may have some problem. Solving this issue is
7753 * left as an exercise to the reader.
7755 kvm->arch.last_tsc_nsec = 0;
7756 kvm->arch.last_tsc_write = 0;
7763 void kvm_arch_hardware_disable(void)
7765 kvm_x86_ops->hardware_disable();
7766 drop_user_return_notifiers();
7769 int kvm_arch_hardware_setup(void)
7773 r = kvm_x86_ops->hardware_setup();
7777 kvm_init_msr_list();
7781 void kvm_arch_hardware_unsetup(void)
7783 kvm_x86_ops->hardware_unsetup();
7786 void kvm_arch_check_processor_compat(void *rtn)
7788 kvm_x86_ops->check_processor_compatibility(rtn);
7791 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7793 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7796 struct static_key kvm_no_apic_vcpu __read_mostly;
7798 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7804 BUG_ON(vcpu->kvm == NULL);
7807 vcpu->arch.pv.pv_unhalted = false;
7808 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7809 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7810 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7812 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7814 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7819 vcpu->arch.pio_data = page_address(page);
7821 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7823 r = kvm_mmu_create(vcpu);
7825 goto fail_free_pio_data;
7827 if (irqchip_in_kernel(kvm)) {
7828 r = kvm_create_lapic(vcpu);
7830 goto fail_mmu_destroy;
7832 static_key_slow_inc(&kvm_no_apic_vcpu);
7834 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7836 if (!vcpu->arch.mce_banks) {
7838 goto fail_free_lapic;
7840 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7842 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7844 goto fail_free_mce_banks;
7847 r = fx_init(vcpu, false);
7849 goto fail_free_wbinvd_dirty_mask;
7851 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7852 vcpu->arch.pv_time_enabled = false;
7854 vcpu->arch.guest_supported_xcr0 = 0;
7855 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7857 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7859 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7861 kvm_async_pf_hash_reset(vcpu);
7865 fail_free_wbinvd_dirty_mask:
7866 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7867 fail_free_mce_banks:
7868 kfree(vcpu->arch.mce_banks);
7870 kvm_free_lapic(vcpu);
7872 kvm_mmu_destroy(vcpu);
7874 free_page((unsigned long)vcpu->arch.pio_data);
7879 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7883 kvm_pmu_destroy(vcpu);
7884 kfree(vcpu->arch.mce_banks);
7885 kvm_free_lapic(vcpu);
7886 idx = srcu_read_lock(&vcpu->kvm->srcu);
7887 kvm_mmu_destroy(vcpu);
7888 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7889 free_page((unsigned long)vcpu->arch.pio_data);
7890 if (!irqchip_in_kernel(vcpu->kvm))
7891 static_key_slow_dec(&kvm_no_apic_vcpu);
7894 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7896 kvm_x86_ops->sched_in(vcpu, cpu);
7899 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7904 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7905 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7906 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7907 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7908 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7910 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7911 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7912 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7913 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7914 &kvm->arch.irq_sources_bitmap);
7916 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7917 mutex_init(&kvm->arch.apic_map_lock);
7918 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7920 pvclock_update_vm_gtod_copy(kvm);
7922 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7923 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7928 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7931 r = vcpu_load(vcpu);
7933 kvm_mmu_unload(vcpu);
7937 static void kvm_free_vcpus(struct kvm *kvm)
7940 struct kvm_vcpu *vcpu;
7943 * Unpin any mmu pages first.
7945 kvm_for_each_vcpu(i, vcpu, kvm) {
7946 kvm_clear_async_pf_completion_queue(vcpu);
7947 kvm_unload_vcpu_mmu(vcpu);
7949 kvm_for_each_vcpu(i, vcpu, kvm)
7950 kvm_arch_vcpu_free(vcpu);
7952 mutex_lock(&kvm->lock);
7953 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7954 kvm->vcpus[i] = NULL;
7956 atomic_set(&kvm->online_vcpus, 0);
7957 mutex_unlock(&kvm->lock);
7960 void kvm_arch_sync_events(struct kvm *kvm)
7962 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7963 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7964 kvm_free_all_assigned_devices(kvm);
7968 int __x86_set_memory_region(struct kvm *kvm,
7969 const struct kvm_userspace_memory_region *mem)
7973 /* Called with kvm->slots_lock held. */
7974 BUG_ON(mem->slot >= KVM_MEM_SLOTS_NUM);
7976 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7977 struct kvm_userspace_memory_region m = *mem;
7980 r = __kvm_set_memory_region(kvm, &m);
7987 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7989 int x86_set_memory_region(struct kvm *kvm,
7990 const struct kvm_userspace_memory_region *mem)
7994 mutex_lock(&kvm->slots_lock);
7995 r = __x86_set_memory_region(kvm, mem);
7996 mutex_unlock(&kvm->slots_lock);
8000 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8002 void kvm_arch_destroy_vm(struct kvm *kvm)
8004 if (current->mm == kvm->mm) {
8006 * Free memory regions allocated on behalf of userspace,
8007 * unless the the memory map has changed due to process exit
8010 struct kvm_userspace_memory_region mem;
8011 memset(&mem, 0, sizeof(mem));
8012 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
8013 x86_set_memory_region(kvm, &mem);
8015 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
8016 x86_set_memory_region(kvm, &mem);
8018 mem.slot = TSS_PRIVATE_MEMSLOT;
8019 x86_set_memory_region(kvm, &mem);
8021 kvm_iommu_unmap_guest(kvm);
8022 kfree(kvm->arch.vpic);
8023 kfree(kvm->arch.vioapic);
8024 kvm_free_vcpus(kvm);
8025 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8028 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8029 struct kvm_memory_slot *dont)
8033 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8034 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8035 kvfree(free->arch.rmap[i]);
8036 free->arch.rmap[i] = NULL;
8041 if (!dont || free->arch.lpage_info[i - 1] !=
8042 dont->arch.lpage_info[i - 1]) {
8043 kvfree(free->arch.lpage_info[i - 1]);
8044 free->arch.lpage_info[i - 1] = NULL;
8049 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8050 unsigned long npages)
8054 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8059 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8060 slot->base_gfn, level) + 1;
8062 slot->arch.rmap[i] =
8063 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
8064 if (!slot->arch.rmap[i])
8069 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
8070 sizeof(*slot->arch.lpage_info[i - 1]));
8071 if (!slot->arch.lpage_info[i - 1])
8074 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8075 slot->arch.lpage_info[i - 1][0].write_count = 1;
8076 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8077 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
8078 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8080 * If the gfn and userspace address are not aligned wrt each
8081 * other, or if explicitly asked to, disable large page
8082 * support for this slot
8084 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8085 !kvm_largepages_enabled()) {
8088 for (j = 0; j < lpages; ++j)
8089 slot->arch.lpage_info[i - 1][j].write_count = 1;
8096 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8097 kvfree(slot->arch.rmap[i]);
8098 slot->arch.rmap[i] = NULL;
8102 kvfree(slot->arch.lpage_info[i - 1]);
8103 slot->arch.lpage_info[i - 1] = NULL;
8108 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8111 * memslots->generation has been incremented.
8112 * mmio generation may have reached its maximum value.
8114 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8117 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8118 struct kvm_memory_slot *memslot,
8119 const struct kvm_userspace_memory_region *mem,
8120 enum kvm_mr_change change)
8123 * Only private memory slots need to be mapped here since
8124 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
8126 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
8127 unsigned long userspace_addr;
8130 * MAP_SHARED to prevent internal slot pages from being moved
8133 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
8134 PROT_READ | PROT_WRITE,
8135 MAP_SHARED | MAP_ANONYMOUS, 0);
8137 if (IS_ERR((void *)userspace_addr))
8138 return PTR_ERR((void *)userspace_addr);
8140 memslot->userspace_addr = userspace_addr;
8146 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8147 struct kvm_memory_slot *new)
8149 /* Still write protect RO slot */
8150 if (new->flags & KVM_MEM_READONLY) {
8151 kvm_mmu_slot_remove_write_access(kvm, new);
8156 * Call kvm_x86_ops dirty logging hooks when they are valid.
8158 * kvm_x86_ops->slot_disable_log_dirty is called when:
8160 * - KVM_MR_CREATE with dirty logging is disabled
8161 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8163 * The reason is, in case of PML, we need to set D-bit for any slots
8164 * with dirty logging disabled in order to eliminate unnecessary GPA
8165 * logging in PML buffer (and potential PML buffer full VMEXT). This
8166 * guarantees leaving PML enabled during guest's lifetime won't have
8167 * any additonal overhead from PML when guest is running with dirty
8168 * logging disabled for memory slots.
8170 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8171 * to dirty logging mode.
8173 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8175 * In case of write protect:
8177 * Write protect all pages for dirty logging.
8179 * All the sptes including the large sptes which point to this
8180 * slot are set to readonly. We can not create any new large
8181 * spte on this slot until the end of the logging.
8183 * See the comments in fast_page_fault().
8185 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8186 if (kvm_x86_ops->slot_enable_log_dirty)
8187 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8189 kvm_mmu_slot_remove_write_access(kvm, new);
8191 if (kvm_x86_ops->slot_disable_log_dirty)
8192 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8196 void kvm_arch_commit_memory_region(struct kvm *kvm,
8197 const struct kvm_userspace_memory_region *mem,
8198 const struct kvm_memory_slot *old,
8199 const struct kvm_memory_slot *new,
8200 enum kvm_mr_change change)
8202 int nr_mmu_pages = 0;
8204 if (change == KVM_MR_DELETE && old->id >= KVM_USER_MEM_SLOTS) {
8207 ret = vm_munmap(old->userspace_addr,
8208 old->npages * PAGE_SIZE);
8211 "kvm_vm_ioctl_set_memory_region: "
8212 "failed to munmap memory\n");
8215 if (!kvm->arch.n_requested_mmu_pages)
8216 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8219 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8222 * Dirty logging tracks sptes in 4k granularity, meaning that large
8223 * sptes have to be split. If live migration is successful, the guest
8224 * in the source machine will be destroyed and large sptes will be
8225 * created in the destination. However, if the guest continues to run
8226 * in the source machine (for example if live migration fails), small
8227 * sptes will remain around and cause bad performance.
8229 * Scan sptes if dirty logging has been stopped, dropping those
8230 * which can be collapsed into a single large-page spte. Later
8231 * page faults will create the large-page sptes.
8233 if ((change != KVM_MR_DELETE) &&
8234 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8235 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8236 kvm_mmu_zap_collapsible_sptes(kvm, new);
8239 * Set up write protection and/or dirty logging for the new slot.
8241 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8242 * been zapped so no dirty logging staff is needed for old slot. For
8243 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8244 * new and it's also covered when dealing with the new slot.
8246 * FIXME: const-ify all uses of struct kvm_memory_slot.
8248 if (change != KVM_MR_DELETE)
8249 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8252 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8254 kvm_mmu_invalidate_zap_all_pages(kvm);
8257 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8258 struct kvm_memory_slot *slot)
8260 kvm_mmu_invalidate_zap_all_pages(kvm);
8263 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8265 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8266 kvm_x86_ops->check_nested_events(vcpu, false);
8268 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8269 !vcpu->arch.apf.halted)
8270 || !list_empty_careful(&vcpu->async_pf.done)
8271 || kvm_apic_has_events(vcpu)
8272 || vcpu->arch.pv.pv_unhalted
8273 || atomic_read(&vcpu->arch.nmi_queued) ||
8274 (kvm_arch_interrupt_allowed(vcpu) &&
8275 kvm_cpu_has_interrupt(vcpu));
8278 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8280 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8283 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8285 return kvm_x86_ops->interrupt_allowed(vcpu);
8288 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8290 if (is_64_bit_mode(vcpu))
8291 return kvm_rip_read(vcpu);
8292 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8293 kvm_rip_read(vcpu));
8295 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8297 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8299 return kvm_get_linear_rip(vcpu) == linear_rip;
8301 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8303 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8305 unsigned long rflags;
8307 rflags = kvm_x86_ops->get_rflags(vcpu);
8308 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8309 rflags &= ~X86_EFLAGS_TF;
8312 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8314 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8316 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8317 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8318 rflags |= X86_EFLAGS_TF;
8319 kvm_x86_ops->set_rflags(vcpu, rflags);
8322 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8324 __kvm_set_rflags(vcpu, rflags);
8325 kvm_make_request(KVM_REQ_EVENT, vcpu);
8327 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8329 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8333 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8337 r = kvm_mmu_reload(vcpu);
8341 if (!vcpu->arch.mmu.direct_map &&
8342 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8345 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8348 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8350 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8353 static inline u32 kvm_async_pf_next_probe(u32 key)
8355 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8358 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8360 u32 key = kvm_async_pf_hash_fn(gfn);
8362 while (vcpu->arch.apf.gfns[key] != ~0)
8363 key = kvm_async_pf_next_probe(key);
8365 vcpu->arch.apf.gfns[key] = gfn;
8368 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8371 u32 key = kvm_async_pf_hash_fn(gfn);
8373 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8374 (vcpu->arch.apf.gfns[key] != gfn &&
8375 vcpu->arch.apf.gfns[key] != ~0); i++)
8376 key = kvm_async_pf_next_probe(key);
8381 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8383 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8386 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8390 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8392 vcpu->arch.apf.gfns[i] = ~0;
8394 j = kvm_async_pf_next_probe(j);
8395 if (vcpu->arch.apf.gfns[j] == ~0)
8397 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8399 * k lies cyclically in ]i,j]
8401 * |....j i.k.| or |.k..j i...|
8403 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8404 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8409 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8412 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8416 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8417 struct kvm_async_pf *work)
8419 struct x86_exception fault;
8421 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8422 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8424 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8425 (vcpu->arch.apf.send_user_only &&
8426 kvm_x86_ops->get_cpl(vcpu) == 0))
8427 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8428 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8429 fault.vector = PF_VECTOR;
8430 fault.error_code_valid = true;
8431 fault.error_code = 0;
8432 fault.nested_page_fault = false;
8433 fault.address = work->arch.token;
8434 kvm_inject_page_fault(vcpu, &fault);
8438 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8439 struct kvm_async_pf *work)
8441 struct x86_exception fault;
8443 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8444 if (work->wakeup_all)
8445 work->arch.token = ~0; /* broadcast wakeup */
8447 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8449 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8450 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8451 fault.vector = PF_VECTOR;
8452 fault.error_code_valid = true;
8453 fault.error_code = 0;
8454 fault.nested_page_fault = false;
8455 fault.address = work->arch.token;
8456 kvm_inject_page_fault(vcpu, &fault);
8458 vcpu->arch.apf.halted = false;
8459 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8462 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8464 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8467 return !kvm_event_needs_reinjection(vcpu) &&
8468 kvm_x86_ops->interrupt_allowed(vcpu);
8471 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8473 atomic_inc(&kvm->arch.noncoherent_dma_count);
8475 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8477 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8479 atomic_dec(&kvm->arch.noncoherent_dma_count);
8481 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8483 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8485 return atomic_read(&kvm->arch.noncoherent_dma_count);
8487 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8489 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8490 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8491 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8492 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8493 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8494 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8495 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8496 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8497 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8498 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8499 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8500 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8501 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8502 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8503 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);