2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
30 #include "assigned-dev.h"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/module.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <trace/events/kvm.h>
56 #define CREATE_TRACE_POINTS
59 #include <asm/debugreg.h>
63 #include <linux/kernel_stat.h>
64 #include <asm/fpu/internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
68 #define MAX_IO_MSRS 256
69 #define KVM_MAX_MCE_BANKS 32
70 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
72 #define emul_to_vcpu(ctxt) \
73 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
76 * - enable syscall per default because its emulated by KVM
77 * - enable LME and LMA per default on 64 bit KVM
81 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
83 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
86 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
87 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
89 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
90 static void process_nmi(struct kvm_vcpu *vcpu);
91 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
93 struct kvm_x86_ops *kvm_x86_ops;
94 EXPORT_SYMBOL_GPL(kvm_x86_ops);
96 static bool ignore_msrs = 0;
97 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
99 unsigned int min_timer_period_us = 500;
100 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
102 static bool __read_mostly kvmclock_periodic_sync = true;
103 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
105 bool kvm_has_tsc_control;
106 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
107 u32 kvm_max_guest_tsc_khz;
108 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
110 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
111 static u32 tsc_tolerance_ppm = 250;
112 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
114 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
115 unsigned int lapic_timer_advance_ns = 0;
116 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
118 static bool backwards_tsc_observed = false;
120 #define KVM_NR_SHARED_MSRS 16
122 struct kvm_shared_msrs_global {
124 u32 msrs[KVM_NR_SHARED_MSRS];
127 struct kvm_shared_msrs {
128 struct user_return_notifier urn;
130 struct kvm_shared_msr_values {
133 } values[KVM_NR_SHARED_MSRS];
136 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
137 static struct kvm_shared_msrs __percpu *shared_msrs;
139 struct kvm_stats_debugfs_item debugfs_entries[] = {
140 { "pf_fixed", VCPU_STAT(pf_fixed) },
141 { "pf_guest", VCPU_STAT(pf_guest) },
142 { "tlb_flush", VCPU_STAT(tlb_flush) },
143 { "invlpg", VCPU_STAT(invlpg) },
144 { "exits", VCPU_STAT(exits) },
145 { "io_exits", VCPU_STAT(io_exits) },
146 { "mmio_exits", VCPU_STAT(mmio_exits) },
147 { "signal_exits", VCPU_STAT(signal_exits) },
148 { "irq_window", VCPU_STAT(irq_window_exits) },
149 { "nmi_window", VCPU_STAT(nmi_window_exits) },
150 { "halt_exits", VCPU_STAT(halt_exits) },
151 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
152 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
153 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
154 { "hypercalls", VCPU_STAT(hypercalls) },
155 { "request_irq", VCPU_STAT(request_irq_exits) },
156 { "irq_exits", VCPU_STAT(irq_exits) },
157 { "host_state_reload", VCPU_STAT(host_state_reload) },
158 { "efer_reload", VCPU_STAT(efer_reload) },
159 { "fpu_reload", VCPU_STAT(fpu_reload) },
160 { "insn_emulation", VCPU_STAT(insn_emulation) },
161 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
162 { "irq_injections", VCPU_STAT(irq_injections) },
163 { "nmi_injections", VCPU_STAT(nmi_injections) },
164 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
165 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
166 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
167 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
168 { "mmu_flooded", VM_STAT(mmu_flooded) },
169 { "mmu_recycled", VM_STAT(mmu_recycled) },
170 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
171 { "mmu_unsync", VM_STAT(mmu_unsync) },
172 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
173 { "largepages", VM_STAT(lpages) },
177 u64 __read_mostly host_xcr0;
179 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
181 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
184 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
185 vcpu->arch.apf.gfns[i] = ~0;
188 static void kvm_on_user_return(struct user_return_notifier *urn)
191 struct kvm_shared_msrs *locals
192 = container_of(urn, struct kvm_shared_msrs, urn);
193 struct kvm_shared_msr_values *values;
195 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
196 values = &locals->values[slot];
197 if (values->host != values->curr) {
198 wrmsrl(shared_msrs_global.msrs[slot], values->host);
199 values->curr = values->host;
202 locals->registered = false;
203 user_return_notifier_unregister(urn);
206 static void shared_msr_update(unsigned slot, u32 msr)
209 unsigned int cpu = smp_processor_id();
210 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
212 /* only read, and nobody should modify it at this time,
213 * so don't need lock */
214 if (slot >= shared_msrs_global.nr) {
215 printk(KERN_ERR "kvm: invalid MSR slot!");
218 rdmsrl_safe(msr, &value);
219 smsr->values[slot].host = value;
220 smsr->values[slot].curr = value;
223 void kvm_define_shared_msr(unsigned slot, u32 msr)
225 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
226 shared_msrs_global.msrs[slot] = msr;
227 if (slot >= shared_msrs_global.nr)
228 shared_msrs_global.nr = slot + 1;
230 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
232 static void kvm_shared_msr_cpu_online(void)
236 for (i = 0; i < shared_msrs_global.nr; ++i)
237 shared_msr_update(i, shared_msrs_global.msrs[i]);
240 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
242 unsigned int cpu = smp_processor_id();
243 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
246 if (((value ^ smsr->values[slot].curr) & mask) == 0)
248 smsr->values[slot].curr = value;
249 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
253 if (!smsr->registered) {
254 smsr->urn.on_user_return = kvm_on_user_return;
255 user_return_notifier_register(&smsr->urn);
256 smsr->registered = true;
260 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
262 static void drop_user_return_notifiers(void)
264 unsigned int cpu = smp_processor_id();
265 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
267 if (smsr->registered)
268 kvm_on_user_return(&smsr->urn);
271 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
273 return vcpu->arch.apic_base;
275 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
277 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
279 u64 old_state = vcpu->arch.apic_base &
280 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
281 u64 new_state = msr_info->data &
282 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
283 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
284 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
286 if (!msr_info->host_initiated &&
287 ((msr_info->data & reserved_bits) != 0 ||
288 new_state == X2APIC_ENABLE ||
289 (new_state == MSR_IA32_APICBASE_ENABLE &&
290 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
291 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
295 kvm_lapic_set_base(vcpu, msr_info->data);
298 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
300 asmlinkage __visible void kvm_spurious_fault(void)
302 /* Fault while not rebooting. We want the trace. */
305 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
307 #define EXCPT_BENIGN 0
308 #define EXCPT_CONTRIBUTORY 1
311 static int exception_class(int vector)
321 return EXCPT_CONTRIBUTORY;
328 #define EXCPT_FAULT 0
330 #define EXCPT_ABORT 2
331 #define EXCPT_INTERRUPT 3
333 static int exception_type(int vector)
337 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
338 return EXCPT_INTERRUPT;
342 /* #DB is trap, as instruction watchpoints are handled elsewhere */
343 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
346 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
349 /* Reserved exceptions will result in fault */
353 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
354 unsigned nr, bool has_error, u32 error_code,
360 kvm_make_request(KVM_REQ_EVENT, vcpu);
362 if (!vcpu->arch.exception.pending) {
364 if (has_error && !is_protmode(vcpu))
366 vcpu->arch.exception.pending = true;
367 vcpu->arch.exception.has_error_code = has_error;
368 vcpu->arch.exception.nr = nr;
369 vcpu->arch.exception.error_code = error_code;
370 vcpu->arch.exception.reinject = reinject;
374 /* to check exception */
375 prev_nr = vcpu->arch.exception.nr;
376 if (prev_nr == DF_VECTOR) {
377 /* triple fault -> shutdown */
378 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
381 class1 = exception_class(prev_nr);
382 class2 = exception_class(nr);
383 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
384 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
385 /* generate double fault per SDM Table 5-5 */
386 vcpu->arch.exception.pending = true;
387 vcpu->arch.exception.has_error_code = true;
388 vcpu->arch.exception.nr = DF_VECTOR;
389 vcpu->arch.exception.error_code = 0;
391 /* replace previous exception with a new one in a hope
392 that instruction re-execution will regenerate lost
397 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
399 kvm_multiple_exception(vcpu, nr, false, 0, false);
401 EXPORT_SYMBOL_GPL(kvm_queue_exception);
403 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
405 kvm_multiple_exception(vcpu, nr, false, 0, true);
407 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
409 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
412 kvm_inject_gp(vcpu, 0);
414 kvm_x86_ops->skip_emulated_instruction(vcpu);
416 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
418 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
420 ++vcpu->stat.pf_guest;
421 vcpu->arch.cr2 = fault->address;
422 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
424 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
426 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
428 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
429 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
431 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
433 return fault->nested_page_fault;
436 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
438 atomic_inc(&vcpu->arch.nmi_queued);
439 kvm_make_request(KVM_REQ_NMI, vcpu);
441 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
443 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
445 kvm_multiple_exception(vcpu, nr, true, error_code, false);
447 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
449 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
451 kvm_multiple_exception(vcpu, nr, true, error_code, true);
453 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
456 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
457 * a #GP and return false.
459 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
461 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
463 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
466 EXPORT_SYMBOL_GPL(kvm_require_cpl);
468 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
470 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
473 kvm_queue_exception(vcpu, UD_VECTOR);
476 EXPORT_SYMBOL_GPL(kvm_require_dr);
479 * This function will be used to read from the physical memory of the currently
480 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
481 * can read from guest physical or from the guest's guest physical memory.
483 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
484 gfn_t ngfn, void *data, int offset, int len,
487 struct x86_exception exception;
491 ngpa = gfn_to_gpa(ngfn);
492 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
493 if (real_gfn == UNMAPPED_GVA)
496 real_gfn = gpa_to_gfn(real_gfn);
498 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
500 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
502 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
503 void *data, int offset, int len, u32 access)
505 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
506 data, offset, len, access);
510 * Load the pae pdptrs. Return true is they are all valid.
512 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
514 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
515 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
518 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
520 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
521 offset * sizeof(u64), sizeof(pdpte),
522 PFERR_USER_MASK|PFERR_WRITE_MASK);
527 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
528 if (is_present_gpte(pdpte[i]) &&
530 vcpu->arch.mmu.guest_rsvd_check.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);
625 if ((cr0 ^ old_cr0) & X86_CR0_CD)
626 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
630 EXPORT_SYMBOL_GPL(kvm_set_cr0);
632 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
634 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
636 EXPORT_SYMBOL_GPL(kvm_lmsw);
638 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
640 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
641 !vcpu->guest_xcr0_loaded) {
642 /* kvm_set_xcr() also depends on this */
643 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
644 vcpu->guest_xcr0_loaded = 1;
648 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
650 if (vcpu->guest_xcr0_loaded) {
651 if (vcpu->arch.xcr0 != host_xcr0)
652 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
653 vcpu->guest_xcr0_loaded = 0;
657 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
660 u64 old_xcr0 = vcpu->arch.xcr0;
663 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
664 if (index != XCR_XFEATURE_ENABLED_MASK)
666 if (!(xcr0 & XSTATE_FP))
668 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
672 * Do not allow the guest to set bits that we do not support
673 * saving. However, xcr0 bit 0 is always set, even if the
674 * emulated CPU does not support XSAVE (see fx_init).
676 valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
677 if (xcr0 & ~valid_bits)
680 if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
683 if (xcr0 & XSTATE_AVX512) {
684 if (!(xcr0 & XSTATE_YMM))
686 if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512)
689 kvm_put_guest_xcr0(vcpu);
690 vcpu->arch.xcr0 = xcr0;
692 if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
693 kvm_update_cpuid(vcpu);
697 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
699 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
700 __kvm_set_xcr(vcpu, index, xcr)) {
701 kvm_inject_gp(vcpu, 0);
706 EXPORT_SYMBOL_GPL(kvm_set_xcr);
708 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
710 unsigned long old_cr4 = kvm_read_cr4(vcpu);
711 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
712 X86_CR4_SMEP | X86_CR4_SMAP;
714 if (cr4 & CR4_RESERVED_BITS)
717 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
720 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
723 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
726 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
729 if (is_long_mode(vcpu)) {
730 if (!(cr4 & X86_CR4_PAE))
732 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
733 && ((cr4 ^ old_cr4) & pdptr_bits)
734 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
738 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
739 if (!guest_cpuid_has_pcid(vcpu))
742 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
743 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
747 if (kvm_x86_ops->set_cr4(vcpu, cr4))
750 if (((cr4 ^ old_cr4) & pdptr_bits) ||
751 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
752 kvm_mmu_reset_context(vcpu);
754 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
755 kvm_update_cpuid(vcpu);
759 EXPORT_SYMBOL_GPL(kvm_set_cr4);
761 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
764 cr3 &= ~CR3_PCID_INVD;
767 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
768 kvm_mmu_sync_roots(vcpu);
769 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
773 if (is_long_mode(vcpu)) {
774 if (cr3 & CR3_L_MODE_RESERVED_BITS)
776 } else if (is_pae(vcpu) && is_paging(vcpu) &&
777 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
780 vcpu->arch.cr3 = cr3;
781 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
782 kvm_mmu_new_cr3(vcpu);
785 EXPORT_SYMBOL_GPL(kvm_set_cr3);
787 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
789 if (cr8 & CR8_RESERVED_BITS)
791 if (irqchip_in_kernel(vcpu->kvm))
792 kvm_lapic_set_tpr(vcpu, cr8);
794 vcpu->arch.cr8 = cr8;
797 EXPORT_SYMBOL_GPL(kvm_set_cr8);
799 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
801 if (irqchip_in_kernel(vcpu->kvm))
802 return kvm_lapic_get_cr8(vcpu);
804 return vcpu->arch.cr8;
806 EXPORT_SYMBOL_GPL(kvm_get_cr8);
808 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
812 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
813 for (i = 0; i < KVM_NR_DB_REGS; i++)
814 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
815 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
819 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
821 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
822 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
825 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
829 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
830 dr7 = vcpu->arch.guest_debug_dr7;
832 dr7 = vcpu->arch.dr7;
833 kvm_x86_ops->set_dr7(vcpu, dr7);
834 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
835 if (dr7 & DR7_BP_EN_MASK)
836 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
839 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
841 u64 fixed = DR6_FIXED_1;
843 if (!guest_cpuid_has_rtm(vcpu))
848 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
852 vcpu->arch.db[dr] = val;
853 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
854 vcpu->arch.eff_db[dr] = val;
859 if (val & 0xffffffff00000000ULL)
861 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
862 kvm_update_dr6(vcpu);
867 if (val & 0xffffffff00000000ULL)
869 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
870 kvm_update_dr7(vcpu);
877 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
879 if (__kvm_set_dr(vcpu, dr, val)) {
880 kvm_inject_gp(vcpu, 0);
885 EXPORT_SYMBOL_GPL(kvm_set_dr);
887 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
891 *val = vcpu->arch.db[dr];
896 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
897 *val = vcpu->arch.dr6;
899 *val = kvm_x86_ops->get_dr6(vcpu);
904 *val = vcpu->arch.dr7;
909 EXPORT_SYMBOL_GPL(kvm_get_dr);
911 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
913 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
917 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
920 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
921 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
924 EXPORT_SYMBOL_GPL(kvm_rdpmc);
927 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
928 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
930 * This list is modified at module load time to reflect the
931 * capabilities of the host cpu. This capabilities test skips MSRs that are
932 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
933 * may depend on host virtualization features rather than host cpu features.
936 static u32 msrs_to_save[] = {
937 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
940 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
942 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
943 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
946 static unsigned num_msrs_to_save;
948 static u32 emulated_msrs[] = {
949 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
950 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
951 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
952 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
953 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
954 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
955 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
959 MSR_IA32_TSCDEADLINE,
960 MSR_IA32_MISC_ENABLE,
966 static unsigned num_emulated_msrs;
968 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
970 if (efer & efer_reserved_bits)
973 if (efer & EFER_FFXSR) {
974 struct kvm_cpuid_entry2 *feat;
976 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
977 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
981 if (efer & EFER_SVME) {
982 struct kvm_cpuid_entry2 *feat;
984 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
985 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
991 EXPORT_SYMBOL_GPL(kvm_valid_efer);
993 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
995 u64 old_efer = vcpu->arch.efer;
997 if (!kvm_valid_efer(vcpu, efer))
1001 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1005 efer |= vcpu->arch.efer & EFER_LMA;
1007 kvm_x86_ops->set_efer(vcpu, efer);
1009 /* Update reserved bits */
1010 if ((efer ^ old_efer) & EFER_NX)
1011 kvm_mmu_reset_context(vcpu);
1016 void kvm_enable_efer_bits(u64 mask)
1018 efer_reserved_bits &= ~mask;
1020 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1023 * Writes msr value into into the appropriate "register".
1024 * Returns 0 on success, non-0 otherwise.
1025 * Assumes vcpu_load() was already called.
1027 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1029 switch (msr->index) {
1032 case MSR_KERNEL_GS_BASE:
1035 if (is_noncanonical_address(msr->data))
1038 case MSR_IA32_SYSENTER_EIP:
1039 case MSR_IA32_SYSENTER_ESP:
1041 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1042 * non-canonical address is written on Intel but not on
1043 * AMD (which ignores the top 32-bits, because it does
1044 * not implement 64-bit SYSENTER).
1046 * 64-bit code should hence be able to write a non-canonical
1047 * value on AMD. Making the address canonical ensures that
1048 * vmentry does not fail on Intel after writing a non-canonical
1049 * value, and that something deterministic happens if the guest
1050 * invokes 64-bit SYSENTER.
1052 msr->data = get_canonical(msr->data);
1054 return kvm_x86_ops->set_msr(vcpu, msr);
1056 EXPORT_SYMBOL_GPL(kvm_set_msr);
1059 * Adapt set_msr() to msr_io()'s calling convention
1061 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1063 struct msr_data msr;
1067 msr.host_initiated = true;
1068 r = kvm_get_msr(vcpu, &msr);
1076 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1078 struct msr_data msr;
1082 msr.host_initiated = true;
1083 return kvm_set_msr(vcpu, &msr);
1086 #ifdef CONFIG_X86_64
1087 struct pvclock_gtod_data {
1090 struct { /* extract of a clocksource struct */
1102 static struct pvclock_gtod_data pvclock_gtod_data;
1104 static void update_pvclock_gtod(struct timekeeper *tk)
1106 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1109 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1111 write_seqcount_begin(&vdata->seq);
1113 /* copy pvclock gtod data */
1114 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1115 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1116 vdata->clock.mask = tk->tkr_mono.mask;
1117 vdata->clock.mult = tk->tkr_mono.mult;
1118 vdata->clock.shift = tk->tkr_mono.shift;
1120 vdata->boot_ns = boot_ns;
1121 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1123 write_seqcount_end(&vdata->seq);
1127 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1130 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1131 * vcpu_enter_guest. This function is only called from
1132 * the physical CPU that is running vcpu.
1134 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1137 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1141 struct pvclock_wall_clock wc;
1142 struct timespec boot;
1147 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1152 ++version; /* first time write, random junk */
1156 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1159 * The guest calculates current wall clock time by adding
1160 * system time (updated by kvm_guest_time_update below) to the
1161 * wall clock specified here. guest system time equals host
1162 * system time for us, thus we must fill in host boot time here.
1166 if (kvm->arch.kvmclock_offset) {
1167 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1168 boot = timespec_sub(boot, ts);
1170 wc.sec = boot.tv_sec;
1171 wc.nsec = boot.tv_nsec;
1172 wc.version = version;
1174 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1177 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1180 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1182 uint32_t quotient, remainder;
1184 /* Don't try to replace with do_div(), this one calculates
1185 * "(dividend << 32) / divisor" */
1187 : "=a" (quotient), "=d" (remainder)
1188 : "0" (0), "1" (dividend), "r" (divisor) );
1192 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1193 s8 *pshift, u32 *pmultiplier)
1200 tps64 = base_khz * 1000LL;
1201 scaled64 = scaled_khz * 1000LL;
1202 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1207 tps32 = (uint32_t)tps64;
1208 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1209 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1217 *pmultiplier = div_frac(scaled64, tps32);
1219 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1220 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1223 #ifdef CONFIG_X86_64
1224 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1227 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1228 static unsigned long max_tsc_khz;
1230 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1232 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1233 vcpu->arch.virtual_tsc_shift);
1236 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1238 u64 v = (u64)khz * (1000000 + ppm);
1243 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1245 u32 thresh_lo, thresh_hi;
1246 int use_scaling = 0;
1248 /* tsc_khz can be zero if TSC calibration fails */
1249 if (this_tsc_khz == 0)
1252 /* Compute a scale to convert nanoseconds in TSC cycles */
1253 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1254 &vcpu->arch.virtual_tsc_shift,
1255 &vcpu->arch.virtual_tsc_mult);
1256 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1259 * Compute the variation in TSC rate which is acceptable
1260 * within the range of tolerance and decide if the
1261 * rate being applied is within that bounds of the hardware
1262 * rate. If so, no scaling or compensation need be done.
1264 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1265 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1266 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1267 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1270 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1273 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1275 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1276 vcpu->arch.virtual_tsc_mult,
1277 vcpu->arch.virtual_tsc_shift);
1278 tsc += vcpu->arch.this_tsc_write;
1282 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1284 #ifdef CONFIG_X86_64
1286 struct kvm_arch *ka = &vcpu->kvm->arch;
1287 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1289 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1290 atomic_read(&vcpu->kvm->online_vcpus));
1293 * Once the masterclock is enabled, always perform request in
1294 * order to update it.
1296 * In order to enable masterclock, the host clocksource must be TSC
1297 * and the vcpus need to have matched TSCs. When that happens,
1298 * perform request to enable masterclock.
1300 if (ka->use_master_clock ||
1301 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1302 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1304 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1305 atomic_read(&vcpu->kvm->online_vcpus),
1306 ka->use_master_clock, gtod->clock.vclock_mode);
1310 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1312 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1313 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1316 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1318 struct kvm *kvm = vcpu->kvm;
1319 u64 offset, ns, elapsed;
1320 unsigned long flags;
1323 bool already_matched;
1324 u64 data = msr->data;
1326 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1327 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1328 ns = get_kernel_ns();
1329 elapsed = ns - kvm->arch.last_tsc_nsec;
1331 if (vcpu->arch.virtual_tsc_khz) {
1334 /* n.b - signed multiplication and division required */
1335 usdiff = data - kvm->arch.last_tsc_write;
1336 #ifdef CONFIG_X86_64
1337 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1339 /* do_div() only does unsigned */
1340 asm("1: idivl %[divisor]\n"
1341 "2: xor %%edx, %%edx\n"
1342 " movl $0, %[faulted]\n"
1344 ".section .fixup,\"ax\"\n"
1345 "4: movl $1, %[faulted]\n"
1349 _ASM_EXTABLE(1b, 4b)
1351 : "=A"(usdiff), [faulted] "=r" (faulted)
1352 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1355 do_div(elapsed, 1000);
1360 /* idivl overflow => difference is larger than USEC_PER_SEC */
1362 usdiff = USEC_PER_SEC;
1364 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1367 * Special case: TSC write with a small delta (1 second) of virtual
1368 * cycle time against real time is interpreted as an attempt to
1369 * synchronize the CPU.
1371 * For a reliable TSC, we can match TSC offsets, and for an unstable
1372 * TSC, we add elapsed time in this computation. We could let the
1373 * compensation code attempt to catch up if we fall behind, but
1374 * it's better to try to match offsets from the beginning.
1376 if (usdiff < USEC_PER_SEC &&
1377 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1378 if (!check_tsc_unstable()) {
1379 offset = kvm->arch.cur_tsc_offset;
1380 pr_debug("kvm: matched tsc offset for %llu\n", data);
1382 u64 delta = nsec_to_cycles(vcpu, elapsed);
1384 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1385 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1388 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1391 * We split periods of matched TSC writes into generations.
1392 * For each generation, we track the original measured
1393 * nanosecond time, offset, and write, so if TSCs are in
1394 * sync, we can match exact offset, and if not, we can match
1395 * exact software computation in compute_guest_tsc()
1397 * These values are tracked in kvm->arch.cur_xxx variables.
1399 kvm->arch.cur_tsc_generation++;
1400 kvm->arch.cur_tsc_nsec = ns;
1401 kvm->arch.cur_tsc_write = data;
1402 kvm->arch.cur_tsc_offset = offset;
1404 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1405 kvm->arch.cur_tsc_generation, data);
1409 * We also track th most recent recorded KHZ, write and time to
1410 * allow the matching interval to be extended at each write.
1412 kvm->arch.last_tsc_nsec = ns;
1413 kvm->arch.last_tsc_write = data;
1414 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1416 vcpu->arch.last_guest_tsc = data;
1418 /* Keep track of which generation this VCPU has synchronized to */
1419 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1420 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1421 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1423 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1424 update_ia32_tsc_adjust_msr(vcpu, offset);
1425 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1426 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1428 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1430 kvm->arch.nr_vcpus_matched_tsc = 0;
1431 } else if (!already_matched) {
1432 kvm->arch.nr_vcpus_matched_tsc++;
1435 kvm_track_tsc_matching(vcpu);
1436 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1439 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1441 #ifdef CONFIG_X86_64
1443 static cycle_t read_tsc(void)
1445 cycle_t ret = (cycle_t)rdtsc_ordered();
1446 u64 last = pvclock_gtod_data.clock.cycle_last;
1448 if (likely(ret >= last))
1452 * GCC likes to generate cmov here, but this branch is extremely
1453 * predictable (it's just a funciton of time and the likely is
1454 * very likely) and there's a data dependence, so force GCC
1455 * to generate a branch instead. I don't barrier() because
1456 * we don't actually need a barrier, and if this function
1457 * ever gets inlined it will generate worse code.
1463 static inline u64 vgettsc(cycle_t *cycle_now)
1466 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1468 *cycle_now = read_tsc();
1470 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1471 return v * gtod->clock.mult;
1474 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1476 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1482 seq = read_seqcount_begin(>od->seq);
1483 mode = gtod->clock.vclock_mode;
1484 ns = gtod->nsec_base;
1485 ns += vgettsc(cycle_now);
1486 ns >>= gtod->clock.shift;
1487 ns += gtod->boot_ns;
1488 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1494 /* returns true if host is using tsc clocksource */
1495 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1497 /* checked again under seqlock below */
1498 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1501 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1507 * Assuming a stable TSC across physical CPUS, and a stable TSC
1508 * across virtual CPUs, the following condition is possible.
1509 * Each numbered line represents an event visible to both
1510 * CPUs at the next numbered event.
1512 * "timespecX" represents host monotonic time. "tscX" represents
1515 * VCPU0 on CPU0 | VCPU1 on CPU1
1517 * 1. read timespec0,tsc0
1518 * 2. | timespec1 = timespec0 + N
1520 * 3. transition to guest | transition to guest
1521 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1522 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1523 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1525 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1528 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1530 * - 0 < N - M => M < N
1532 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1533 * always the case (the difference between two distinct xtime instances
1534 * might be smaller then the difference between corresponding TSC reads,
1535 * when updating guest vcpus pvclock areas).
1537 * To avoid that problem, do not allow visibility of distinct
1538 * system_timestamp/tsc_timestamp values simultaneously: use a master
1539 * copy of host monotonic time values. Update that master copy
1542 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1546 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1548 #ifdef CONFIG_X86_64
1549 struct kvm_arch *ka = &kvm->arch;
1551 bool host_tsc_clocksource, vcpus_matched;
1553 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1554 atomic_read(&kvm->online_vcpus));
1557 * If the host uses TSC clock, then passthrough TSC as stable
1560 host_tsc_clocksource = kvm_get_time_and_clockread(
1561 &ka->master_kernel_ns,
1562 &ka->master_cycle_now);
1564 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1565 && !backwards_tsc_observed
1566 && !ka->boot_vcpu_runs_old_kvmclock;
1568 if (ka->use_master_clock)
1569 atomic_set(&kvm_guest_has_master_clock, 1);
1571 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1572 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1577 static void kvm_gen_update_masterclock(struct kvm *kvm)
1579 #ifdef CONFIG_X86_64
1581 struct kvm_vcpu *vcpu;
1582 struct kvm_arch *ka = &kvm->arch;
1584 spin_lock(&ka->pvclock_gtod_sync_lock);
1585 kvm_make_mclock_inprogress_request(kvm);
1586 /* no guest entries from this point */
1587 pvclock_update_vm_gtod_copy(kvm);
1589 kvm_for_each_vcpu(i, vcpu, kvm)
1590 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1592 /* guest entries allowed */
1593 kvm_for_each_vcpu(i, vcpu, kvm)
1594 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1596 spin_unlock(&ka->pvclock_gtod_sync_lock);
1600 static int kvm_guest_time_update(struct kvm_vcpu *v)
1602 unsigned long flags, this_tsc_khz;
1603 struct kvm_vcpu_arch *vcpu = &v->arch;
1604 struct kvm_arch *ka = &v->kvm->arch;
1606 u64 tsc_timestamp, host_tsc;
1607 struct pvclock_vcpu_time_info guest_hv_clock;
1609 bool use_master_clock;
1615 * If the host uses TSC clock, then passthrough TSC as stable
1618 spin_lock(&ka->pvclock_gtod_sync_lock);
1619 use_master_clock = ka->use_master_clock;
1620 if (use_master_clock) {
1621 host_tsc = ka->master_cycle_now;
1622 kernel_ns = ka->master_kernel_ns;
1624 spin_unlock(&ka->pvclock_gtod_sync_lock);
1626 /* Keep irq disabled to prevent changes to the clock */
1627 local_irq_save(flags);
1628 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1629 if (unlikely(this_tsc_khz == 0)) {
1630 local_irq_restore(flags);
1631 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1634 if (!use_master_clock) {
1636 kernel_ns = get_kernel_ns();
1639 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1642 * We may have to catch up the TSC to match elapsed wall clock
1643 * time for two reasons, even if kvmclock is used.
1644 * 1) CPU could have been running below the maximum TSC rate
1645 * 2) Broken TSC compensation resets the base at each VCPU
1646 * entry to avoid unknown leaps of TSC even when running
1647 * again on the same CPU. This may cause apparent elapsed
1648 * time to disappear, and the guest to stand still or run
1651 if (vcpu->tsc_catchup) {
1652 u64 tsc = compute_guest_tsc(v, kernel_ns);
1653 if (tsc > tsc_timestamp) {
1654 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1655 tsc_timestamp = tsc;
1659 local_irq_restore(flags);
1661 if (!vcpu->pv_time_enabled)
1664 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1665 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1666 &vcpu->hv_clock.tsc_shift,
1667 &vcpu->hv_clock.tsc_to_system_mul);
1668 vcpu->hw_tsc_khz = this_tsc_khz;
1671 /* With all the info we got, fill in the values */
1672 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1673 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1674 vcpu->last_guest_tsc = tsc_timestamp;
1676 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1677 &guest_hv_clock, sizeof(guest_hv_clock))))
1680 /* This VCPU is paused, but it's legal for a guest to read another
1681 * VCPU's kvmclock, so we really have to follow the specification where
1682 * it says that version is odd if data is being modified, and even after
1685 * Version field updates must be kept separate. This is because
1686 * kvm_write_guest_cached might use a "rep movs" instruction, and
1687 * writes within a string instruction are weakly ordered. So there
1688 * are three writes overall.
1690 * As a small optimization, only write the version field in the first
1691 * and third write. The vcpu->pv_time cache is still valid, because the
1692 * version field is the first in the struct.
1694 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1696 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1697 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1699 sizeof(vcpu->hv_clock.version));
1703 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1704 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1706 if (vcpu->pvclock_set_guest_stopped_request) {
1707 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1708 vcpu->pvclock_set_guest_stopped_request = false;
1711 /* If the host uses TSC clocksource, then it is stable */
1712 if (use_master_clock)
1713 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1715 vcpu->hv_clock.flags = pvclock_flags;
1717 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1719 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1721 sizeof(vcpu->hv_clock));
1725 vcpu->hv_clock.version++;
1726 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1728 sizeof(vcpu->hv_clock.version));
1733 * kvmclock updates which are isolated to a given vcpu, such as
1734 * vcpu->cpu migration, should not allow system_timestamp from
1735 * the rest of the vcpus to remain static. Otherwise ntp frequency
1736 * correction applies to one vcpu's system_timestamp but not
1739 * So in those cases, request a kvmclock update for all vcpus.
1740 * We need to rate-limit these requests though, as they can
1741 * considerably slow guests that have a large number of vcpus.
1742 * The time for a remote vcpu to update its kvmclock is bound
1743 * by the delay we use to rate-limit the updates.
1746 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1748 static void kvmclock_update_fn(struct work_struct *work)
1751 struct delayed_work *dwork = to_delayed_work(work);
1752 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1753 kvmclock_update_work);
1754 struct kvm *kvm = container_of(ka, struct kvm, arch);
1755 struct kvm_vcpu *vcpu;
1757 kvm_for_each_vcpu(i, vcpu, kvm) {
1758 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1759 kvm_vcpu_kick(vcpu);
1763 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1765 struct kvm *kvm = v->kvm;
1767 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1768 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1769 KVMCLOCK_UPDATE_DELAY);
1772 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1774 static void kvmclock_sync_fn(struct work_struct *work)
1776 struct delayed_work *dwork = to_delayed_work(work);
1777 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1778 kvmclock_sync_work);
1779 struct kvm *kvm = container_of(ka, struct kvm, arch);
1781 if (!kvmclock_periodic_sync)
1784 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1785 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1786 KVMCLOCK_SYNC_PERIOD);
1789 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1791 u64 mcg_cap = vcpu->arch.mcg_cap;
1792 unsigned bank_num = mcg_cap & 0xff;
1795 case MSR_IA32_MCG_STATUS:
1796 vcpu->arch.mcg_status = data;
1798 case MSR_IA32_MCG_CTL:
1799 if (!(mcg_cap & MCG_CTL_P))
1801 if (data != 0 && data != ~(u64)0)
1803 vcpu->arch.mcg_ctl = data;
1806 if (msr >= MSR_IA32_MC0_CTL &&
1807 msr < MSR_IA32_MCx_CTL(bank_num)) {
1808 u32 offset = msr - MSR_IA32_MC0_CTL;
1809 /* only 0 or all 1s can be written to IA32_MCi_CTL
1810 * some Linux kernels though clear bit 10 in bank 4 to
1811 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1812 * this to avoid an uncatched #GP in the guest
1814 if ((offset & 0x3) == 0 &&
1815 data != 0 && (data | (1 << 10)) != ~(u64)0)
1817 vcpu->arch.mce_banks[offset] = data;
1825 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1827 struct kvm *kvm = vcpu->kvm;
1828 int lm = is_long_mode(vcpu);
1829 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1830 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1831 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1832 : kvm->arch.xen_hvm_config.blob_size_32;
1833 u32 page_num = data & ~PAGE_MASK;
1834 u64 page_addr = data & PAGE_MASK;
1839 if (page_num >= blob_size)
1842 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1847 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1856 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1858 gpa_t gpa = data & ~0x3f;
1860 /* Bits 2:5 are reserved, Should be zero */
1864 vcpu->arch.apf.msr_val = data;
1866 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1867 kvm_clear_async_pf_completion_queue(vcpu);
1868 kvm_async_pf_hash_reset(vcpu);
1872 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1876 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1877 kvm_async_pf_wakeup_all(vcpu);
1881 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1883 vcpu->arch.pv_time_enabled = false;
1886 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1890 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1893 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1894 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1895 vcpu->arch.st.accum_steal = delta;
1898 static void record_steal_time(struct kvm_vcpu *vcpu)
1900 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1903 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1904 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1907 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1908 vcpu->arch.st.steal.version += 2;
1909 vcpu->arch.st.accum_steal = 0;
1911 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1912 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1915 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1918 u32 msr = msr_info->index;
1919 u64 data = msr_info->data;
1922 case MSR_AMD64_NB_CFG:
1923 case MSR_IA32_UCODE_REV:
1924 case MSR_IA32_UCODE_WRITE:
1925 case MSR_VM_HSAVE_PA:
1926 case MSR_AMD64_PATCH_LOADER:
1927 case MSR_AMD64_BU_CFG2:
1931 return set_efer(vcpu, data);
1933 data &= ~(u64)0x40; /* ignore flush filter disable */
1934 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1935 data &= ~(u64)0x8; /* ignore TLB cache disable */
1936 data &= ~(u64)0x40000; /* ignore Mc status write enable */
1938 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1943 case MSR_FAM10H_MMIO_CONF_BASE:
1945 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1950 case MSR_IA32_DEBUGCTLMSR:
1952 /* We support the non-activated case already */
1954 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1955 /* Values other than LBR and BTF are vendor-specific,
1956 thus reserved and should throw a #GP */
1959 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1962 case 0x200 ... 0x2ff:
1963 return kvm_mtrr_set_msr(vcpu, msr, data);
1964 case MSR_IA32_APICBASE:
1965 return kvm_set_apic_base(vcpu, msr_info);
1966 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1967 return kvm_x2apic_msr_write(vcpu, msr, data);
1968 case MSR_IA32_TSCDEADLINE:
1969 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1971 case MSR_IA32_TSC_ADJUST:
1972 if (guest_cpuid_has_tsc_adjust(vcpu)) {
1973 if (!msr_info->host_initiated) {
1974 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
1975 adjust_tsc_offset_guest(vcpu, adj);
1977 vcpu->arch.ia32_tsc_adjust_msr = data;
1980 case MSR_IA32_MISC_ENABLE:
1981 vcpu->arch.ia32_misc_enable_msr = data;
1983 case MSR_IA32_SMBASE:
1984 if (!msr_info->host_initiated)
1986 vcpu->arch.smbase = data;
1988 case MSR_KVM_WALL_CLOCK_NEW:
1989 case MSR_KVM_WALL_CLOCK:
1990 vcpu->kvm->arch.wall_clock = data;
1991 kvm_write_wall_clock(vcpu->kvm, data);
1993 case MSR_KVM_SYSTEM_TIME_NEW:
1994 case MSR_KVM_SYSTEM_TIME: {
1996 struct kvm_arch *ka = &vcpu->kvm->arch;
1998 kvmclock_reset(vcpu);
2000 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2001 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2003 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2004 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2007 ka->boot_vcpu_runs_old_kvmclock = tmp;
2010 vcpu->arch.time = data;
2011 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2013 /* we verify if the enable bit is set... */
2017 gpa_offset = data & ~(PAGE_MASK | 1);
2019 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2020 &vcpu->arch.pv_time, data & ~1ULL,
2021 sizeof(struct pvclock_vcpu_time_info)))
2022 vcpu->arch.pv_time_enabled = false;
2024 vcpu->arch.pv_time_enabled = true;
2028 case MSR_KVM_ASYNC_PF_EN:
2029 if (kvm_pv_enable_async_pf(vcpu, data))
2032 case MSR_KVM_STEAL_TIME:
2034 if (unlikely(!sched_info_on()))
2037 if (data & KVM_STEAL_RESERVED_MASK)
2040 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2041 data & KVM_STEAL_VALID_BITS,
2042 sizeof(struct kvm_steal_time)))
2045 vcpu->arch.st.msr_val = data;
2047 if (!(data & KVM_MSR_ENABLED))
2050 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2053 accumulate_steal_time(vcpu);
2056 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2059 case MSR_KVM_PV_EOI_EN:
2060 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2064 case MSR_IA32_MCG_CTL:
2065 case MSR_IA32_MCG_STATUS:
2066 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2067 return set_msr_mce(vcpu, msr, data);
2069 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2070 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2071 pr = true; /* fall through */
2072 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2073 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2074 if (kvm_pmu_is_valid_msr(vcpu, msr))
2075 return kvm_pmu_set_msr(vcpu, msr_info);
2077 if (pr || data != 0)
2078 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2079 "0x%x data 0x%llx\n", msr, data);
2081 case MSR_K7_CLK_CTL:
2083 * Ignore all writes to this no longer documented MSR.
2084 * Writes are only relevant for old K7 processors,
2085 * all pre-dating SVM, but a recommended workaround from
2086 * AMD for these chips. It is possible to specify the
2087 * affected processor models on the command line, hence
2088 * the need to ignore the workaround.
2091 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2092 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2093 case HV_X64_MSR_CRASH_CTL:
2094 return kvm_hv_set_msr_common(vcpu, msr, data,
2095 msr_info->host_initiated);
2096 case MSR_IA32_BBL_CR_CTL3:
2097 /* Drop writes to this legacy MSR -- see rdmsr
2098 * counterpart for further detail.
2100 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2102 case MSR_AMD64_OSVW_ID_LENGTH:
2103 if (!guest_cpuid_has_osvw(vcpu))
2105 vcpu->arch.osvw.length = data;
2107 case MSR_AMD64_OSVW_STATUS:
2108 if (!guest_cpuid_has_osvw(vcpu))
2110 vcpu->arch.osvw.status = data;
2113 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2114 return xen_hvm_config(vcpu, data);
2115 if (kvm_pmu_is_valid_msr(vcpu, msr))
2116 return kvm_pmu_set_msr(vcpu, msr_info);
2118 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2122 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2129 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2133 * Reads an msr value (of 'msr_index') into 'pdata'.
2134 * Returns 0 on success, non-0 otherwise.
2135 * Assumes vcpu_load() was already called.
2137 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2139 return kvm_x86_ops->get_msr(vcpu, msr);
2141 EXPORT_SYMBOL_GPL(kvm_get_msr);
2143 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2146 u64 mcg_cap = vcpu->arch.mcg_cap;
2147 unsigned bank_num = mcg_cap & 0xff;
2150 case MSR_IA32_P5_MC_ADDR:
2151 case MSR_IA32_P5_MC_TYPE:
2154 case MSR_IA32_MCG_CAP:
2155 data = vcpu->arch.mcg_cap;
2157 case MSR_IA32_MCG_CTL:
2158 if (!(mcg_cap & MCG_CTL_P))
2160 data = vcpu->arch.mcg_ctl;
2162 case MSR_IA32_MCG_STATUS:
2163 data = vcpu->arch.mcg_status;
2166 if (msr >= MSR_IA32_MC0_CTL &&
2167 msr < MSR_IA32_MCx_CTL(bank_num)) {
2168 u32 offset = msr - MSR_IA32_MC0_CTL;
2169 data = vcpu->arch.mce_banks[offset];
2178 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2180 switch (msr_info->index) {
2181 case MSR_IA32_PLATFORM_ID:
2182 case MSR_IA32_EBL_CR_POWERON:
2183 case MSR_IA32_DEBUGCTLMSR:
2184 case MSR_IA32_LASTBRANCHFROMIP:
2185 case MSR_IA32_LASTBRANCHTOIP:
2186 case MSR_IA32_LASTINTFROMIP:
2187 case MSR_IA32_LASTINTTOIP:
2189 case MSR_K8_TSEG_ADDR:
2190 case MSR_K8_TSEG_MASK:
2192 case MSR_VM_HSAVE_PA:
2193 case MSR_K8_INT_PENDING_MSG:
2194 case MSR_AMD64_NB_CFG:
2195 case MSR_FAM10H_MMIO_CONF_BASE:
2196 case MSR_AMD64_BU_CFG2:
2199 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2200 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2201 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2202 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2203 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2204 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2207 case MSR_IA32_UCODE_REV:
2208 msr_info->data = 0x100000000ULL;
2211 case 0x200 ... 0x2ff:
2212 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2213 case 0xcd: /* fsb frequency */
2217 * MSR_EBC_FREQUENCY_ID
2218 * Conservative value valid for even the basic CPU models.
2219 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2220 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2221 * and 266MHz for model 3, or 4. Set Core Clock
2222 * Frequency to System Bus Frequency Ratio to 1 (bits
2223 * 31:24) even though these are only valid for CPU
2224 * models > 2, however guests may end up dividing or
2225 * multiplying by zero otherwise.
2227 case MSR_EBC_FREQUENCY_ID:
2228 msr_info->data = 1 << 24;
2230 case MSR_IA32_APICBASE:
2231 msr_info->data = kvm_get_apic_base(vcpu);
2233 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2234 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2236 case MSR_IA32_TSCDEADLINE:
2237 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2239 case MSR_IA32_TSC_ADJUST:
2240 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2242 case MSR_IA32_MISC_ENABLE:
2243 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2245 case MSR_IA32_SMBASE:
2246 if (!msr_info->host_initiated)
2248 msr_info->data = vcpu->arch.smbase;
2250 case MSR_IA32_PERF_STATUS:
2251 /* TSC increment by tick */
2252 msr_info->data = 1000ULL;
2253 /* CPU multiplier */
2254 msr_info->data |= (((uint64_t)4ULL) << 40);
2257 msr_info->data = vcpu->arch.efer;
2259 case MSR_KVM_WALL_CLOCK:
2260 case MSR_KVM_WALL_CLOCK_NEW:
2261 msr_info->data = vcpu->kvm->arch.wall_clock;
2263 case MSR_KVM_SYSTEM_TIME:
2264 case MSR_KVM_SYSTEM_TIME_NEW:
2265 msr_info->data = vcpu->arch.time;
2267 case MSR_KVM_ASYNC_PF_EN:
2268 msr_info->data = vcpu->arch.apf.msr_val;
2270 case MSR_KVM_STEAL_TIME:
2271 msr_info->data = vcpu->arch.st.msr_val;
2273 case MSR_KVM_PV_EOI_EN:
2274 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2276 case MSR_IA32_P5_MC_ADDR:
2277 case MSR_IA32_P5_MC_TYPE:
2278 case MSR_IA32_MCG_CAP:
2279 case MSR_IA32_MCG_CTL:
2280 case MSR_IA32_MCG_STATUS:
2281 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2282 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2283 case MSR_K7_CLK_CTL:
2285 * Provide expected ramp-up count for K7. All other
2286 * are set to zero, indicating minimum divisors for
2289 * This prevents guest kernels on AMD host with CPU
2290 * type 6, model 8 and higher from exploding due to
2291 * the rdmsr failing.
2293 msr_info->data = 0x20000000;
2295 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2296 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2297 case HV_X64_MSR_CRASH_CTL:
2298 return kvm_hv_get_msr_common(vcpu,
2299 msr_info->index, &msr_info->data);
2301 case MSR_IA32_BBL_CR_CTL3:
2302 /* This legacy MSR exists but isn't fully documented in current
2303 * silicon. It is however accessed by winxp in very narrow
2304 * scenarios where it sets bit #19, itself documented as
2305 * a "reserved" bit. Best effort attempt to source coherent
2306 * read data here should the balance of the register be
2307 * interpreted by the guest:
2309 * L2 cache control register 3: 64GB range, 256KB size,
2310 * enabled, latency 0x1, configured
2312 msr_info->data = 0xbe702111;
2314 case MSR_AMD64_OSVW_ID_LENGTH:
2315 if (!guest_cpuid_has_osvw(vcpu))
2317 msr_info->data = vcpu->arch.osvw.length;
2319 case MSR_AMD64_OSVW_STATUS:
2320 if (!guest_cpuid_has_osvw(vcpu))
2322 msr_info->data = vcpu->arch.osvw.status;
2325 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2326 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2328 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2331 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2338 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2341 * Read or write a bunch of msrs. All parameters are kernel addresses.
2343 * @return number of msrs set successfully.
2345 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2346 struct kvm_msr_entry *entries,
2347 int (*do_msr)(struct kvm_vcpu *vcpu,
2348 unsigned index, u64 *data))
2352 idx = srcu_read_lock(&vcpu->kvm->srcu);
2353 for (i = 0; i < msrs->nmsrs; ++i)
2354 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2356 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2362 * Read or write a bunch of msrs. Parameters are user addresses.
2364 * @return number of msrs set successfully.
2366 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2367 int (*do_msr)(struct kvm_vcpu *vcpu,
2368 unsigned index, u64 *data),
2371 struct kvm_msrs msrs;
2372 struct kvm_msr_entry *entries;
2377 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2381 if (msrs.nmsrs >= MAX_IO_MSRS)
2384 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2385 entries = memdup_user(user_msrs->entries, size);
2386 if (IS_ERR(entries)) {
2387 r = PTR_ERR(entries);
2391 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2396 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2407 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2412 case KVM_CAP_IRQCHIP:
2414 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2415 case KVM_CAP_SET_TSS_ADDR:
2416 case KVM_CAP_EXT_CPUID:
2417 case KVM_CAP_EXT_EMUL_CPUID:
2418 case KVM_CAP_CLOCKSOURCE:
2420 case KVM_CAP_NOP_IO_DELAY:
2421 case KVM_CAP_MP_STATE:
2422 case KVM_CAP_SYNC_MMU:
2423 case KVM_CAP_USER_NMI:
2424 case KVM_CAP_REINJECT_CONTROL:
2425 case KVM_CAP_IRQ_INJECT_STATUS:
2426 case KVM_CAP_IOEVENTFD:
2427 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2429 case KVM_CAP_PIT_STATE2:
2430 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2431 case KVM_CAP_XEN_HVM:
2432 case KVM_CAP_ADJUST_CLOCK:
2433 case KVM_CAP_VCPU_EVENTS:
2434 case KVM_CAP_HYPERV:
2435 case KVM_CAP_HYPERV_VAPIC:
2436 case KVM_CAP_HYPERV_SPIN:
2437 case KVM_CAP_PCI_SEGMENT:
2438 case KVM_CAP_DEBUGREGS:
2439 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2441 case KVM_CAP_ASYNC_PF:
2442 case KVM_CAP_GET_TSC_KHZ:
2443 case KVM_CAP_KVMCLOCK_CTRL:
2444 case KVM_CAP_READONLY_MEM:
2445 case KVM_CAP_HYPERV_TIME:
2446 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2447 case KVM_CAP_TSC_DEADLINE_TIMER:
2448 case KVM_CAP_ENABLE_CAP_VM:
2449 case KVM_CAP_DISABLE_QUIRKS:
2450 case KVM_CAP_SET_BOOT_CPU_ID:
2451 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2452 case KVM_CAP_ASSIGN_DEV_IRQ:
2453 case KVM_CAP_PCI_2_3:
2457 case KVM_CAP_X86_SMM:
2458 /* SMBASE is usually relocated above 1M on modern chipsets,
2459 * and SMM handlers might indeed rely on 4G segment limits,
2460 * so do not report SMM to be available if real mode is
2461 * emulated via vm86 mode. Still, do not go to great lengths
2462 * to avoid userspace's usage of the feature, because it is a
2463 * fringe case that is not enabled except via specific settings
2464 * of the module parameters.
2466 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2468 case KVM_CAP_COALESCED_MMIO:
2469 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2472 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2474 case KVM_CAP_NR_VCPUS:
2475 r = KVM_SOFT_MAX_VCPUS;
2477 case KVM_CAP_MAX_VCPUS:
2480 case KVM_CAP_NR_MEMSLOTS:
2481 r = KVM_USER_MEM_SLOTS;
2483 case KVM_CAP_PV_MMU: /* obsolete */
2486 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2488 r = iommu_present(&pci_bus_type);
2492 r = KVM_MAX_MCE_BANKS;
2497 case KVM_CAP_TSC_CONTROL:
2498 r = kvm_has_tsc_control;
2508 long kvm_arch_dev_ioctl(struct file *filp,
2509 unsigned int ioctl, unsigned long arg)
2511 void __user *argp = (void __user *)arg;
2515 case KVM_GET_MSR_INDEX_LIST: {
2516 struct kvm_msr_list __user *user_msr_list = argp;
2517 struct kvm_msr_list msr_list;
2521 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2524 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2525 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2528 if (n < msr_list.nmsrs)
2531 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2532 num_msrs_to_save * sizeof(u32)))
2534 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2536 num_emulated_msrs * sizeof(u32)))
2541 case KVM_GET_SUPPORTED_CPUID:
2542 case KVM_GET_EMULATED_CPUID: {
2543 struct kvm_cpuid2 __user *cpuid_arg = argp;
2544 struct kvm_cpuid2 cpuid;
2547 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2550 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2556 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2561 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2564 mce_cap = KVM_MCE_CAP_SUPPORTED;
2566 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2578 static void wbinvd_ipi(void *garbage)
2583 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2585 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2588 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2590 /* Address WBINVD may be executed by guest */
2591 if (need_emulate_wbinvd(vcpu)) {
2592 if (kvm_x86_ops->has_wbinvd_exit())
2593 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2594 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2595 smp_call_function_single(vcpu->cpu,
2596 wbinvd_ipi, NULL, 1);
2599 kvm_x86_ops->vcpu_load(vcpu, cpu);
2601 /* Apply any externally detected TSC adjustments (due to suspend) */
2602 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2603 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2604 vcpu->arch.tsc_offset_adjustment = 0;
2605 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2608 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2609 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2610 rdtsc() - vcpu->arch.last_host_tsc;
2612 mark_tsc_unstable("KVM discovered backwards TSC");
2613 if (check_tsc_unstable()) {
2614 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2615 vcpu->arch.last_guest_tsc);
2616 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2617 vcpu->arch.tsc_catchup = 1;
2620 * On a host with synchronized TSC, there is no need to update
2621 * kvmclock on vcpu->cpu migration
2623 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2624 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2625 if (vcpu->cpu != cpu)
2626 kvm_migrate_timers(vcpu);
2630 accumulate_steal_time(vcpu);
2631 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2634 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2636 kvm_x86_ops->vcpu_put(vcpu);
2637 kvm_put_guest_fpu(vcpu);
2638 vcpu->arch.last_host_tsc = rdtsc();
2641 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2642 struct kvm_lapic_state *s)
2644 kvm_x86_ops->sync_pir_to_irr(vcpu);
2645 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2650 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2651 struct kvm_lapic_state *s)
2653 kvm_apic_post_state_restore(vcpu, s);
2654 update_cr8_intercept(vcpu);
2659 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2660 struct kvm_interrupt *irq)
2662 if (irq->irq >= KVM_NR_INTERRUPTS)
2664 if (irqchip_in_kernel(vcpu->kvm))
2667 kvm_queue_interrupt(vcpu, irq->irq, false);
2668 kvm_make_request(KVM_REQ_EVENT, vcpu);
2673 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2675 kvm_inject_nmi(vcpu);
2680 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2682 kvm_make_request(KVM_REQ_SMI, vcpu);
2687 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2688 struct kvm_tpr_access_ctl *tac)
2692 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2696 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2700 unsigned bank_num = mcg_cap & 0xff, bank;
2703 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2705 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2708 vcpu->arch.mcg_cap = mcg_cap;
2709 /* Init IA32_MCG_CTL to all 1s */
2710 if (mcg_cap & MCG_CTL_P)
2711 vcpu->arch.mcg_ctl = ~(u64)0;
2712 /* Init IA32_MCi_CTL to all 1s */
2713 for (bank = 0; bank < bank_num; bank++)
2714 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2719 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2720 struct kvm_x86_mce *mce)
2722 u64 mcg_cap = vcpu->arch.mcg_cap;
2723 unsigned bank_num = mcg_cap & 0xff;
2724 u64 *banks = vcpu->arch.mce_banks;
2726 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2729 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2730 * reporting is disabled
2732 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2733 vcpu->arch.mcg_ctl != ~(u64)0)
2735 banks += 4 * mce->bank;
2737 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2738 * reporting is disabled for the bank
2740 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2742 if (mce->status & MCI_STATUS_UC) {
2743 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2744 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2745 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2748 if (banks[1] & MCI_STATUS_VAL)
2749 mce->status |= MCI_STATUS_OVER;
2750 banks[2] = mce->addr;
2751 banks[3] = mce->misc;
2752 vcpu->arch.mcg_status = mce->mcg_status;
2753 banks[1] = mce->status;
2754 kvm_queue_exception(vcpu, MC_VECTOR);
2755 } else if (!(banks[1] & MCI_STATUS_VAL)
2756 || !(banks[1] & MCI_STATUS_UC)) {
2757 if (banks[1] & MCI_STATUS_VAL)
2758 mce->status |= MCI_STATUS_OVER;
2759 banks[2] = mce->addr;
2760 banks[3] = mce->misc;
2761 banks[1] = mce->status;
2763 banks[1] |= MCI_STATUS_OVER;
2767 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2768 struct kvm_vcpu_events *events)
2771 events->exception.injected =
2772 vcpu->arch.exception.pending &&
2773 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2774 events->exception.nr = vcpu->arch.exception.nr;
2775 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2776 events->exception.pad = 0;
2777 events->exception.error_code = vcpu->arch.exception.error_code;
2779 events->interrupt.injected =
2780 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2781 events->interrupt.nr = vcpu->arch.interrupt.nr;
2782 events->interrupt.soft = 0;
2783 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2785 events->nmi.injected = vcpu->arch.nmi_injected;
2786 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2787 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2788 events->nmi.pad = 0;
2790 events->sipi_vector = 0; /* never valid when reporting to user space */
2792 events->smi.smm = is_smm(vcpu);
2793 events->smi.pending = vcpu->arch.smi_pending;
2794 events->smi.smm_inside_nmi =
2795 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2796 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2798 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2799 | KVM_VCPUEVENT_VALID_SHADOW
2800 | KVM_VCPUEVENT_VALID_SMM);
2801 memset(&events->reserved, 0, sizeof(events->reserved));
2804 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2805 struct kvm_vcpu_events *events)
2807 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2808 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2809 | KVM_VCPUEVENT_VALID_SHADOW
2810 | KVM_VCPUEVENT_VALID_SMM))
2814 vcpu->arch.exception.pending = events->exception.injected;
2815 vcpu->arch.exception.nr = events->exception.nr;
2816 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2817 vcpu->arch.exception.error_code = events->exception.error_code;
2819 vcpu->arch.interrupt.pending = events->interrupt.injected;
2820 vcpu->arch.interrupt.nr = events->interrupt.nr;
2821 vcpu->arch.interrupt.soft = events->interrupt.soft;
2822 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2823 kvm_x86_ops->set_interrupt_shadow(vcpu,
2824 events->interrupt.shadow);
2826 vcpu->arch.nmi_injected = events->nmi.injected;
2827 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2828 vcpu->arch.nmi_pending = events->nmi.pending;
2829 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2831 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2832 kvm_vcpu_has_lapic(vcpu))
2833 vcpu->arch.apic->sipi_vector = events->sipi_vector;
2835 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
2836 if (events->smi.smm)
2837 vcpu->arch.hflags |= HF_SMM_MASK;
2839 vcpu->arch.hflags &= ~HF_SMM_MASK;
2840 vcpu->arch.smi_pending = events->smi.pending;
2841 if (events->smi.smm_inside_nmi)
2842 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
2844 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
2845 if (kvm_vcpu_has_lapic(vcpu)) {
2846 if (events->smi.latched_init)
2847 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2849 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2853 kvm_make_request(KVM_REQ_EVENT, vcpu);
2858 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2859 struct kvm_debugregs *dbgregs)
2863 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2864 kvm_get_dr(vcpu, 6, &val);
2866 dbgregs->dr7 = vcpu->arch.dr7;
2868 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2871 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2872 struct kvm_debugregs *dbgregs)
2877 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2878 kvm_update_dr0123(vcpu);
2879 vcpu->arch.dr6 = dbgregs->dr6;
2880 kvm_update_dr6(vcpu);
2881 vcpu->arch.dr7 = dbgregs->dr7;
2882 kvm_update_dr7(vcpu);
2887 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
2889 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
2891 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
2892 u64 xstate_bv = xsave->header.xfeatures;
2896 * Copy legacy XSAVE area, to avoid complications with CPUID
2897 * leaves 0 and 1 in the loop below.
2899 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
2902 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
2905 * Copy each region from the possibly compacted offset to the
2906 * non-compacted offset.
2908 valid = xstate_bv & ~XSTATE_FPSSE;
2910 u64 feature = valid & -valid;
2911 int index = fls64(feature) - 1;
2912 void *src = get_xsave_addr(xsave, feature);
2915 u32 size, offset, ecx, edx;
2916 cpuid_count(XSTATE_CPUID, index,
2917 &size, &offset, &ecx, &edx);
2918 memcpy(dest + offset, src, size);
2925 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
2927 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
2928 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
2932 * Copy legacy XSAVE area, to avoid complications with CPUID
2933 * leaves 0 and 1 in the loop below.
2935 memcpy(xsave, src, XSAVE_HDR_OFFSET);
2937 /* Set XSTATE_BV and possibly XCOMP_BV. */
2938 xsave->header.xfeatures = xstate_bv;
2940 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
2943 * Copy each region from the non-compacted offset to the
2944 * possibly compacted offset.
2946 valid = xstate_bv & ~XSTATE_FPSSE;
2948 u64 feature = valid & -valid;
2949 int index = fls64(feature) - 1;
2950 void *dest = get_xsave_addr(xsave, feature);
2953 u32 size, offset, ecx, edx;
2954 cpuid_count(XSTATE_CPUID, index,
2955 &size, &offset, &ecx, &edx);
2956 memcpy(dest, src + offset, size);
2963 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2964 struct kvm_xsave *guest_xsave)
2966 if (cpu_has_xsave) {
2967 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
2968 fill_xsave((u8 *) guest_xsave->region, vcpu);
2970 memcpy(guest_xsave->region,
2971 &vcpu->arch.guest_fpu.state.fxsave,
2972 sizeof(struct fxregs_state));
2973 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2978 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2979 struct kvm_xsave *guest_xsave)
2982 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2984 if (cpu_has_xsave) {
2986 * Here we allow setting states that are not present in
2987 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
2988 * with old userspace.
2990 if (xstate_bv & ~kvm_supported_xcr0())
2992 load_xsave(vcpu, (u8 *)guest_xsave->region);
2994 if (xstate_bv & ~XSTATE_FPSSE)
2996 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
2997 guest_xsave->region, sizeof(struct fxregs_state));
3002 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3003 struct kvm_xcrs *guest_xcrs)
3005 if (!cpu_has_xsave) {
3006 guest_xcrs->nr_xcrs = 0;
3010 guest_xcrs->nr_xcrs = 1;
3011 guest_xcrs->flags = 0;
3012 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3013 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3016 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3017 struct kvm_xcrs *guest_xcrs)
3024 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3027 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3028 /* Only support XCR0 currently */
3029 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3030 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3031 guest_xcrs->xcrs[i].value);
3040 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3041 * stopped by the hypervisor. This function will be called from the host only.
3042 * EINVAL is returned when the host attempts to set the flag for a guest that
3043 * does not support pv clocks.
3045 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3047 if (!vcpu->arch.pv_time_enabled)
3049 vcpu->arch.pvclock_set_guest_stopped_request = true;
3050 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3054 long kvm_arch_vcpu_ioctl(struct file *filp,
3055 unsigned int ioctl, unsigned long arg)
3057 struct kvm_vcpu *vcpu = filp->private_data;
3058 void __user *argp = (void __user *)arg;
3061 struct kvm_lapic_state *lapic;
3062 struct kvm_xsave *xsave;
3063 struct kvm_xcrs *xcrs;
3069 case KVM_GET_LAPIC: {
3071 if (!vcpu->arch.apic)
3073 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3078 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3082 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3087 case KVM_SET_LAPIC: {
3089 if (!vcpu->arch.apic)
3091 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3092 if (IS_ERR(u.lapic))
3093 return PTR_ERR(u.lapic);
3095 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3098 case KVM_INTERRUPT: {
3099 struct kvm_interrupt irq;
3102 if (copy_from_user(&irq, argp, sizeof irq))
3104 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3108 r = kvm_vcpu_ioctl_nmi(vcpu);
3112 r = kvm_vcpu_ioctl_smi(vcpu);
3115 case KVM_SET_CPUID: {
3116 struct kvm_cpuid __user *cpuid_arg = argp;
3117 struct kvm_cpuid cpuid;
3120 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3122 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3125 case KVM_SET_CPUID2: {
3126 struct kvm_cpuid2 __user *cpuid_arg = argp;
3127 struct kvm_cpuid2 cpuid;
3130 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3132 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3133 cpuid_arg->entries);
3136 case KVM_GET_CPUID2: {
3137 struct kvm_cpuid2 __user *cpuid_arg = argp;
3138 struct kvm_cpuid2 cpuid;
3141 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3143 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3144 cpuid_arg->entries);
3148 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3154 r = msr_io(vcpu, argp, do_get_msr, 1);
3157 r = msr_io(vcpu, argp, do_set_msr, 0);
3159 case KVM_TPR_ACCESS_REPORTING: {
3160 struct kvm_tpr_access_ctl tac;
3163 if (copy_from_user(&tac, argp, sizeof tac))
3165 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3169 if (copy_to_user(argp, &tac, sizeof tac))
3174 case KVM_SET_VAPIC_ADDR: {
3175 struct kvm_vapic_addr va;
3178 if (!irqchip_in_kernel(vcpu->kvm))
3181 if (copy_from_user(&va, argp, sizeof va))
3183 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3186 case KVM_X86_SETUP_MCE: {
3190 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3192 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3195 case KVM_X86_SET_MCE: {
3196 struct kvm_x86_mce mce;
3199 if (copy_from_user(&mce, argp, sizeof mce))
3201 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3204 case KVM_GET_VCPU_EVENTS: {
3205 struct kvm_vcpu_events events;
3207 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3210 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3215 case KVM_SET_VCPU_EVENTS: {
3216 struct kvm_vcpu_events events;
3219 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3222 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3225 case KVM_GET_DEBUGREGS: {
3226 struct kvm_debugregs dbgregs;
3228 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3231 if (copy_to_user(argp, &dbgregs,
3232 sizeof(struct kvm_debugregs)))
3237 case KVM_SET_DEBUGREGS: {
3238 struct kvm_debugregs dbgregs;
3241 if (copy_from_user(&dbgregs, argp,
3242 sizeof(struct kvm_debugregs)))
3245 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3248 case KVM_GET_XSAVE: {
3249 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3254 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3257 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3262 case KVM_SET_XSAVE: {
3263 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3264 if (IS_ERR(u.xsave))
3265 return PTR_ERR(u.xsave);
3267 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3270 case KVM_GET_XCRS: {
3271 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3276 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3279 if (copy_to_user(argp, u.xcrs,
3280 sizeof(struct kvm_xcrs)))
3285 case KVM_SET_XCRS: {
3286 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3288 return PTR_ERR(u.xcrs);
3290 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3293 case KVM_SET_TSC_KHZ: {
3297 user_tsc_khz = (u32)arg;
3299 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3302 if (user_tsc_khz == 0)
3303 user_tsc_khz = tsc_khz;
3305 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3310 case KVM_GET_TSC_KHZ: {
3311 r = vcpu->arch.virtual_tsc_khz;
3314 case KVM_KVMCLOCK_CTRL: {
3315 r = kvm_set_guest_paused(vcpu);
3326 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3328 return VM_FAULT_SIGBUS;
3331 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3335 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3337 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3341 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3344 kvm->arch.ept_identity_map_addr = ident_addr;
3348 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3349 u32 kvm_nr_mmu_pages)
3351 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3354 mutex_lock(&kvm->slots_lock);
3356 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3357 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3359 mutex_unlock(&kvm->slots_lock);
3363 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3365 return kvm->arch.n_max_mmu_pages;
3368 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3373 switch (chip->chip_id) {
3374 case KVM_IRQCHIP_PIC_MASTER:
3375 memcpy(&chip->chip.pic,
3376 &pic_irqchip(kvm)->pics[0],
3377 sizeof(struct kvm_pic_state));
3379 case KVM_IRQCHIP_PIC_SLAVE:
3380 memcpy(&chip->chip.pic,
3381 &pic_irqchip(kvm)->pics[1],
3382 sizeof(struct kvm_pic_state));
3384 case KVM_IRQCHIP_IOAPIC:
3385 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3394 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3399 switch (chip->chip_id) {
3400 case KVM_IRQCHIP_PIC_MASTER:
3401 spin_lock(&pic_irqchip(kvm)->lock);
3402 memcpy(&pic_irqchip(kvm)->pics[0],
3404 sizeof(struct kvm_pic_state));
3405 spin_unlock(&pic_irqchip(kvm)->lock);
3407 case KVM_IRQCHIP_PIC_SLAVE:
3408 spin_lock(&pic_irqchip(kvm)->lock);
3409 memcpy(&pic_irqchip(kvm)->pics[1],
3411 sizeof(struct kvm_pic_state));
3412 spin_unlock(&pic_irqchip(kvm)->lock);
3414 case KVM_IRQCHIP_IOAPIC:
3415 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3421 kvm_pic_update_irq(pic_irqchip(kvm));
3425 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3429 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3430 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3431 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3435 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3439 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3440 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3441 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3442 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3446 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3450 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3451 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3452 sizeof(ps->channels));
3453 ps->flags = kvm->arch.vpit->pit_state.flags;
3454 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3455 memset(&ps->reserved, 0, sizeof(ps->reserved));
3459 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3461 int r = 0, start = 0;
3462 u32 prev_legacy, cur_legacy;
3463 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3464 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3465 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3466 if (!prev_legacy && cur_legacy)
3468 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3469 sizeof(kvm->arch.vpit->pit_state.channels));
3470 kvm->arch.vpit->pit_state.flags = ps->flags;
3471 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3472 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3476 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3477 struct kvm_reinject_control *control)
3479 if (!kvm->arch.vpit)
3481 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3482 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3483 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3488 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3489 * @kvm: kvm instance
3490 * @log: slot id and address to which we copy the log
3492 * Steps 1-4 below provide general overview of dirty page logging. See
3493 * kvm_get_dirty_log_protect() function description for additional details.
3495 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3496 * always flush the TLB (step 4) even if previous step failed and the dirty
3497 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3498 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3499 * writes will be marked dirty for next log read.
3501 * 1. Take a snapshot of the bit and clear it if needed.
3502 * 2. Write protect the corresponding page.
3503 * 3. Copy the snapshot to the userspace.
3504 * 4. Flush TLB's if needed.
3506 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3508 bool is_dirty = false;
3511 mutex_lock(&kvm->slots_lock);
3514 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3516 if (kvm_x86_ops->flush_log_dirty)
3517 kvm_x86_ops->flush_log_dirty(kvm);
3519 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3522 * All the TLBs can be flushed out of mmu lock, see the comments in
3523 * kvm_mmu_slot_remove_write_access().
3525 lockdep_assert_held(&kvm->slots_lock);
3527 kvm_flush_remote_tlbs(kvm);
3529 mutex_unlock(&kvm->slots_lock);
3533 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3536 if (!irqchip_in_kernel(kvm))
3539 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3540 irq_event->irq, irq_event->level,
3545 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3546 struct kvm_enable_cap *cap)
3554 case KVM_CAP_DISABLE_QUIRKS:
3555 kvm->arch.disabled_quirks = cap->args[0];
3565 long kvm_arch_vm_ioctl(struct file *filp,
3566 unsigned int ioctl, unsigned long arg)
3568 struct kvm *kvm = filp->private_data;
3569 void __user *argp = (void __user *)arg;
3572 * This union makes it completely explicit to gcc-3.x
3573 * that these two variables' stack usage should be
3574 * combined, not added together.
3577 struct kvm_pit_state ps;
3578 struct kvm_pit_state2 ps2;
3579 struct kvm_pit_config pit_config;
3583 case KVM_SET_TSS_ADDR:
3584 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3586 case KVM_SET_IDENTITY_MAP_ADDR: {
3590 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3592 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3595 case KVM_SET_NR_MMU_PAGES:
3596 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3598 case KVM_GET_NR_MMU_PAGES:
3599 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3601 case KVM_CREATE_IRQCHIP: {
3602 struct kvm_pic *vpic;
3604 mutex_lock(&kvm->lock);
3607 goto create_irqchip_unlock;
3609 if (atomic_read(&kvm->online_vcpus))
3610 goto create_irqchip_unlock;
3612 vpic = kvm_create_pic(kvm);
3614 r = kvm_ioapic_init(kvm);
3616 mutex_lock(&kvm->slots_lock);
3617 kvm_destroy_pic(vpic);
3618 mutex_unlock(&kvm->slots_lock);
3619 goto create_irqchip_unlock;
3622 goto create_irqchip_unlock;
3623 r = kvm_setup_default_irq_routing(kvm);
3625 mutex_lock(&kvm->slots_lock);
3626 mutex_lock(&kvm->irq_lock);
3627 kvm_ioapic_destroy(kvm);
3628 kvm_destroy_pic(vpic);
3629 mutex_unlock(&kvm->irq_lock);
3630 mutex_unlock(&kvm->slots_lock);
3631 goto create_irqchip_unlock;
3633 /* Write kvm->irq_routing before kvm->arch.vpic. */
3635 kvm->arch.vpic = vpic;
3636 create_irqchip_unlock:
3637 mutex_unlock(&kvm->lock);
3640 case KVM_CREATE_PIT:
3641 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3643 case KVM_CREATE_PIT2:
3645 if (copy_from_user(&u.pit_config, argp,
3646 sizeof(struct kvm_pit_config)))
3649 mutex_lock(&kvm->slots_lock);
3652 goto create_pit_unlock;
3654 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3658 mutex_unlock(&kvm->slots_lock);
3660 case KVM_GET_IRQCHIP: {
3661 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3662 struct kvm_irqchip *chip;
3664 chip = memdup_user(argp, sizeof(*chip));
3671 if (!irqchip_in_kernel(kvm))
3672 goto get_irqchip_out;
3673 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3675 goto get_irqchip_out;
3677 if (copy_to_user(argp, chip, sizeof *chip))
3678 goto get_irqchip_out;
3684 case KVM_SET_IRQCHIP: {
3685 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3686 struct kvm_irqchip *chip;
3688 chip = memdup_user(argp, sizeof(*chip));
3695 if (!irqchip_in_kernel(kvm))
3696 goto set_irqchip_out;
3697 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3699 goto set_irqchip_out;
3707 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3710 if (!kvm->arch.vpit)
3712 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3716 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3723 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3726 if (!kvm->arch.vpit)
3728 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3731 case KVM_GET_PIT2: {
3733 if (!kvm->arch.vpit)
3735 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3739 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3744 case KVM_SET_PIT2: {
3746 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3749 if (!kvm->arch.vpit)
3751 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3754 case KVM_REINJECT_CONTROL: {
3755 struct kvm_reinject_control control;
3757 if (copy_from_user(&control, argp, sizeof(control)))
3759 r = kvm_vm_ioctl_reinject(kvm, &control);
3762 case KVM_SET_BOOT_CPU_ID:
3764 mutex_lock(&kvm->lock);
3765 if (atomic_read(&kvm->online_vcpus) != 0)
3768 kvm->arch.bsp_vcpu_id = arg;
3769 mutex_unlock(&kvm->lock);
3771 case KVM_XEN_HVM_CONFIG: {
3773 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3774 sizeof(struct kvm_xen_hvm_config)))
3777 if (kvm->arch.xen_hvm_config.flags)
3782 case KVM_SET_CLOCK: {
3783 struct kvm_clock_data user_ns;
3788 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3796 local_irq_disable();
3797 now_ns = get_kernel_ns();
3798 delta = user_ns.clock - now_ns;
3800 kvm->arch.kvmclock_offset = delta;
3801 kvm_gen_update_masterclock(kvm);
3804 case KVM_GET_CLOCK: {
3805 struct kvm_clock_data user_ns;
3808 local_irq_disable();
3809 now_ns = get_kernel_ns();
3810 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3813 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3816 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3821 case KVM_ENABLE_CAP: {
3822 struct kvm_enable_cap cap;
3825 if (copy_from_user(&cap, argp, sizeof(cap)))
3827 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
3831 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
3837 static void kvm_init_msr_list(void)
3842 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
3843 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3847 * Even MSRs that are valid in the host may not be exposed
3848 * to the guests in some cases. We could work around this
3849 * in VMX with the generic MSR save/load machinery, but it
3850 * is not really worthwhile since it will really only
3851 * happen with nested virtualization.
3853 switch (msrs_to_save[i]) {
3854 case MSR_IA32_BNDCFGS:
3855 if (!kvm_x86_ops->mpx_supported())
3863 msrs_to_save[j] = msrs_to_save[i];
3866 num_msrs_to_save = j;
3868 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
3869 switch (emulated_msrs[i]) {
3870 case MSR_IA32_SMBASE:
3871 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
3879 emulated_msrs[j] = emulated_msrs[i];
3882 num_emulated_msrs = j;
3885 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3893 if (!(vcpu->arch.apic &&
3894 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
3895 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
3906 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3913 if (!(vcpu->arch.apic &&
3914 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
3916 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
3918 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3928 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3929 struct kvm_segment *var, int seg)
3931 kvm_x86_ops->set_segment(vcpu, var, seg);
3934 void kvm_get_segment(struct kvm_vcpu *vcpu,
3935 struct kvm_segment *var, int seg)
3937 kvm_x86_ops->get_segment(vcpu, var, seg);
3940 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
3941 struct x86_exception *exception)
3945 BUG_ON(!mmu_is_nested(vcpu));
3947 /* NPT walks are always user-walks */
3948 access |= PFERR_USER_MASK;
3949 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
3954 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3955 struct x86_exception *exception)
3957 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3958 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3961 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3962 struct x86_exception *exception)
3964 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3965 access |= PFERR_FETCH_MASK;
3966 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3969 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3970 struct x86_exception *exception)
3972 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3973 access |= PFERR_WRITE_MASK;
3974 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3977 /* uses this to access any guest's mapped memory without checking CPL */
3978 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3979 struct x86_exception *exception)
3981 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3984 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3985 struct kvm_vcpu *vcpu, u32 access,
3986 struct x86_exception *exception)
3989 int r = X86EMUL_CONTINUE;
3992 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3994 unsigned offset = addr & (PAGE_SIZE-1);
3995 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3998 if (gpa == UNMAPPED_GVA)
3999 return X86EMUL_PROPAGATE_FAULT;
4000 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4003 r = X86EMUL_IO_NEEDED;
4015 /* used for instruction fetching */
4016 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4017 gva_t addr, void *val, unsigned int bytes,
4018 struct x86_exception *exception)
4020 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4021 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4025 /* Inline kvm_read_guest_virt_helper for speed. */
4026 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4028 if (unlikely(gpa == UNMAPPED_GVA))
4029 return X86EMUL_PROPAGATE_FAULT;
4031 offset = addr & (PAGE_SIZE-1);
4032 if (WARN_ON(offset + bytes > PAGE_SIZE))
4033 bytes = (unsigned)PAGE_SIZE - offset;
4034 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4036 if (unlikely(ret < 0))
4037 return X86EMUL_IO_NEEDED;
4039 return X86EMUL_CONTINUE;
4042 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4043 gva_t addr, void *val, unsigned int bytes,
4044 struct x86_exception *exception)
4046 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4047 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4049 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4052 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4054 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4055 gva_t addr, void *val, unsigned int bytes,
4056 struct x86_exception *exception)
4058 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4059 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4062 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4063 gva_t addr, void *val,
4065 struct x86_exception *exception)
4067 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4069 int r = X86EMUL_CONTINUE;
4072 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4075 unsigned offset = addr & (PAGE_SIZE-1);
4076 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4079 if (gpa == UNMAPPED_GVA)
4080 return X86EMUL_PROPAGATE_FAULT;
4081 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4083 r = X86EMUL_IO_NEEDED;
4094 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4096 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4097 gpa_t *gpa, struct x86_exception *exception,
4100 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4101 | (write ? PFERR_WRITE_MASK : 0);
4103 if (vcpu_match_mmio_gva(vcpu, gva)
4104 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4105 vcpu->arch.access, access)) {
4106 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4107 (gva & (PAGE_SIZE - 1));
4108 trace_vcpu_match_mmio(gva, *gpa, write, false);
4112 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4114 if (*gpa == UNMAPPED_GVA)
4117 /* For APIC access vmexit */
4118 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4121 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4122 trace_vcpu_match_mmio(gva, *gpa, write, true);
4129 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4130 const void *val, int bytes)
4134 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4137 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4141 struct read_write_emulator_ops {
4142 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4144 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4145 void *val, int bytes);
4146 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4147 int bytes, void *val);
4148 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4149 void *val, int bytes);
4153 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4155 if (vcpu->mmio_read_completed) {
4156 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4157 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4158 vcpu->mmio_read_completed = 0;
4165 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4166 void *val, int bytes)
4168 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4171 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4172 void *val, int bytes)
4174 return emulator_write_phys(vcpu, gpa, val, bytes);
4177 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4179 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4180 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4183 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4184 void *val, int bytes)
4186 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4187 return X86EMUL_IO_NEEDED;
4190 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4191 void *val, int bytes)
4193 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4195 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4196 return X86EMUL_CONTINUE;
4199 static const struct read_write_emulator_ops read_emultor = {
4200 .read_write_prepare = read_prepare,
4201 .read_write_emulate = read_emulate,
4202 .read_write_mmio = vcpu_mmio_read,
4203 .read_write_exit_mmio = read_exit_mmio,
4206 static const struct read_write_emulator_ops write_emultor = {
4207 .read_write_emulate = write_emulate,
4208 .read_write_mmio = write_mmio,
4209 .read_write_exit_mmio = write_exit_mmio,
4213 static int emulator_read_write_onepage(unsigned long addr, void *val,
4215 struct x86_exception *exception,
4216 struct kvm_vcpu *vcpu,
4217 const struct read_write_emulator_ops *ops)
4221 bool write = ops->write;
4222 struct kvm_mmio_fragment *frag;
4224 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4227 return X86EMUL_PROPAGATE_FAULT;
4229 /* For APIC access vmexit */
4233 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4234 return X86EMUL_CONTINUE;
4238 * Is this MMIO handled locally?
4240 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4241 if (handled == bytes)
4242 return X86EMUL_CONTINUE;
4248 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4249 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4253 return X86EMUL_CONTINUE;
4256 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4258 void *val, unsigned int bytes,
4259 struct x86_exception *exception,
4260 const struct read_write_emulator_ops *ops)
4262 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4266 if (ops->read_write_prepare &&
4267 ops->read_write_prepare(vcpu, val, bytes))
4268 return X86EMUL_CONTINUE;
4270 vcpu->mmio_nr_fragments = 0;
4272 /* Crossing a page boundary? */
4273 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4276 now = -addr & ~PAGE_MASK;
4277 rc = emulator_read_write_onepage(addr, val, now, exception,
4280 if (rc != X86EMUL_CONTINUE)
4283 if (ctxt->mode != X86EMUL_MODE_PROT64)
4289 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4291 if (rc != X86EMUL_CONTINUE)
4294 if (!vcpu->mmio_nr_fragments)
4297 gpa = vcpu->mmio_fragments[0].gpa;
4299 vcpu->mmio_needed = 1;
4300 vcpu->mmio_cur_fragment = 0;
4302 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4303 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4304 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4305 vcpu->run->mmio.phys_addr = gpa;
4307 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4310 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4314 struct x86_exception *exception)
4316 return emulator_read_write(ctxt, addr, val, bytes,
4317 exception, &read_emultor);
4320 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4324 struct x86_exception *exception)
4326 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4327 exception, &write_emultor);
4330 #define CMPXCHG_TYPE(t, ptr, old, new) \
4331 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4333 #ifdef CONFIG_X86_64
4334 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4336 # define CMPXCHG64(ptr, old, new) \
4337 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4340 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4345 struct x86_exception *exception)
4347 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4353 /* guests cmpxchg8b have to be emulated atomically */
4354 if (bytes > 8 || (bytes & (bytes - 1)))
4357 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4359 if (gpa == UNMAPPED_GVA ||
4360 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4363 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4366 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4367 if (is_error_page(page))
4370 kaddr = kmap_atomic(page);
4371 kaddr += offset_in_page(gpa);
4374 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4377 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4380 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4383 exchanged = CMPXCHG64(kaddr, old, new);
4388 kunmap_atomic(kaddr);
4389 kvm_release_page_dirty(page);
4392 return X86EMUL_CMPXCHG_FAILED;
4394 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4395 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4397 return X86EMUL_CONTINUE;
4400 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4402 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4405 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4407 /* TODO: String I/O for in kernel device */
4410 if (vcpu->arch.pio.in)
4411 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4412 vcpu->arch.pio.size, pd);
4414 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4415 vcpu->arch.pio.port, vcpu->arch.pio.size,
4420 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4421 unsigned short port, void *val,
4422 unsigned int count, bool in)
4424 vcpu->arch.pio.port = port;
4425 vcpu->arch.pio.in = in;
4426 vcpu->arch.pio.count = count;
4427 vcpu->arch.pio.size = size;
4429 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4430 vcpu->arch.pio.count = 0;
4434 vcpu->run->exit_reason = KVM_EXIT_IO;
4435 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4436 vcpu->run->io.size = size;
4437 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4438 vcpu->run->io.count = count;
4439 vcpu->run->io.port = port;
4444 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4445 int size, unsigned short port, void *val,
4448 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4451 if (vcpu->arch.pio.count)
4454 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4457 memcpy(val, vcpu->arch.pio_data, size * count);
4458 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4459 vcpu->arch.pio.count = 0;
4466 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4467 int size, unsigned short port,
4468 const void *val, unsigned int count)
4470 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4472 memcpy(vcpu->arch.pio_data, val, size * count);
4473 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4474 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4477 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4479 return kvm_x86_ops->get_segment_base(vcpu, seg);
4482 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4484 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4487 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4489 if (!need_emulate_wbinvd(vcpu))
4490 return X86EMUL_CONTINUE;
4492 if (kvm_x86_ops->has_wbinvd_exit()) {
4493 int cpu = get_cpu();
4495 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4496 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4497 wbinvd_ipi, NULL, 1);
4499 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4502 return X86EMUL_CONTINUE;
4505 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4507 kvm_x86_ops->skip_emulated_instruction(vcpu);
4508 return kvm_emulate_wbinvd_noskip(vcpu);
4510 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4514 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4516 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4519 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4520 unsigned long *dest)
4522 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4525 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4526 unsigned long value)
4529 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4532 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4534 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4537 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4539 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4540 unsigned long value;
4544 value = kvm_read_cr0(vcpu);
4547 value = vcpu->arch.cr2;
4550 value = kvm_read_cr3(vcpu);
4553 value = kvm_read_cr4(vcpu);
4556 value = kvm_get_cr8(vcpu);
4559 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4566 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4568 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4573 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4576 vcpu->arch.cr2 = val;
4579 res = kvm_set_cr3(vcpu, val);
4582 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4585 res = kvm_set_cr8(vcpu, val);
4588 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4595 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4597 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4600 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4602 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4605 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4607 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4610 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4612 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4615 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4617 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4620 static unsigned long emulator_get_cached_segment_base(
4621 struct x86_emulate_ctxt *ctxt, int seg)
4623 return get_segment_base(emul_to_vcpu(ctxt), seg);
4626 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4627 struct desc_struct *desc, u32 *base3,
4630 struct kvm_segment var;
4632 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4633 *selector = var.selector;
4636 memset(desc, 0, sizeof(*desc));
4642 set_desc_limit(desc, var.limit);
4643 set_desc_base(desc, (unsigned long)var.base);
4644 #ifdef CONFIG_X86_64
4646 *base3 = var.base >> 32;
4648 desc->type = var.type;
4650 desc->dpl = var.dpl;
4651 desc->p = var.present;
4652 desc->avl = var.avl;
4660 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4661 struct desc_struct *desc, u32 base3,
4664 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4665 struct kvm_segment var;
4667 var.selector = selector;
4668 var.base = get_desc_base(desc);
4669 #ifdef CONFIG_X86_64
4670 var.base |= ((u64)base3) << 32;
4672 var.limit = get_desc_limit(desc);
4674 var.limit = (var.limit << 12) | 0xfff;
4675 var.type = desc->type;
4676 var.dpl = desc->dpl;
4681 var.avl = desc->avl;
4682 var.present = desc->p;
4683 var.unusable = !var.present;
4686 kvm_set_segment(vcpu, &var, seg);
4690 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4691 u32 msr_index, u64 *pdata)
4693 struct msr_data msr;
4696 msr.index = msr_index;
4697 msr.host_initiated = false;
4698 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4706 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4707 u32 msr_index, u64 data)
4709 struct msr_data msr;
4712 msr.index = msr_index;
4713 msr.host_initiated = false;
4714 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4717 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4719 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4721 return vcpu->arch.smbase;
4724 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4726 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4728 vcpu->arch.smbase = smbase;
4731 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4734 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4737 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4738 u32 pmc, u64 *pdata)
4740 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4743 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4745 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4748 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4751 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4753 * CR0.TS may reference the host fpu state, not the guest fpu state,
4754 * so it may be clear at this point.
4759 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4764 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4765 struct x86_instruction_info *info,
4766 enum x86_intercept_stage stage)
4768 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4771 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4772 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4774 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4777 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4779 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4782 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4784 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4787 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
4789 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
4792 static const struct x86_emulate_ops emulate_ops = {
4793 .read_gpr = emulator_read_gpr,
4794 .write_gpr = emulator_write_gpr,
4795 .read_std = kvm_read_guest_virt_system,
4796 .write_std = kvm_write_guest_virt_system,
4797 .fetch = kvm_fetch_guest_virt,
4798 .read_emulated = emulator_read_emulated,
4799 .write_emulated = emulator_write_emulated,
4800 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4801 .invlpg = emulator_invlpg,
4802 .pio_in_emulated = emulator_pio_in_emulated,
4803 .pio_out_emulated = emulator_pio_out_emulated,
4804 .get_segment = emulator_get_segment,
4805 .set_segment = emulator_set_segment,
4806 .get_cached_segment_base = emulator_get_cached_segment_base,
4807 .get_gdt = emulator_get_gdt,
4808 .get_idt = emulator_get_idt,
4809 .set_gdt = emulator_set_gdt,
4810 .set_idt = emulator_set_idt,
4811 .get_cr = emulator_get_cr,
4812 .set_cr = emulator_set_cr,
4813 .cpl = emulator_get_cpl,
4814 .get_dr = emulator_get_dr,
4815 .set_dr = emulator_set_dr,
4816 .get_smbase = emulator_get_smbase,
4817 .set_smbase = emulator_set_smbase,
4818 .set_msr = emulator_set_msr,
4819 .get_msr = emulator_get_msr,
4820 .check_pmc = emulator_check_pmc,
4821 .read_pmc = emulator_read_pmc,
4822 .halt = emulator_halt,
4823 .wbinvd = emulator_wbinvd,
4824 .fix_hypercall = emulator_fix_hypercall,
4825 .get_fpu = emulator_get_fpu,
4826 .put_fpu = emulator_put_fpu,
4827 .intercept = emulator_intercept,
4828 .get_cpuid = emulator_get_cpuid,
4829 .set_nmi_mask = emulator_set_nmi_mask,
4832 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4834 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
4836 * an sti; sti; sequence only disable interrupts for the first
4837 * instruction. So, if the last instruction, be it emulated or
4838 * not, left the system with the INT_STI flag enabled, it
4839 * means that the last instruction is an sti. We should not
4840 * leave the flag on in this case. The same goes for mov ss
4842 if (int_shadow & mask)
4844 if (unlikely(int_shadow || mask)) {
4845 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4847 kvm_make_request(KVM_REQ_EVENT, vcpu);
4851 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
4853 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4854 if (ctxt->exception.vector == PF_VECTOR)
4855 return kvm_propagate_fault(vcpu, &ctxt->exception);
4857 if (ctxt->exception.error_code_valid)
4858 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4859 ctxt->exception.error_code);
4861 kvm_queue_exception(vcpu, ctxt->exception.vector);
4865 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4867 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4870 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4872 ctxt->eflags = kvm_get_rflags(vcpu);
4873 ctxt->eip = kvm_rip_read(vcpu);
4874 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4875 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4876 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
4877 cs_db ? X86EMUL_MODE_PROT32 :
4878 X86EMUL_MODE_PROT16;
4879 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
4880 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
4881 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
4882 ctxt->emul_flags = vcpu->arch.hflags;
4884 init_decode_cache(ctxt);
4885 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4888 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4890 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4893 init_emulate_ctxt(vcpu);
4897 ctxt->_eip = ctxt->eip + inc_eip;
4898 ret = emulate_int_real(ctxt, irq);
4900 if (ret != X86EMUL_CONTINUE)
4901 return EMULATE_FAIL;
4903 ctxt->eip = ctxt->_eip;
4904 kvm_rip_write(vcpu, ctxt->eip);
4905 kvm_set_rflags(vcpu, ctxt->eflags);
4907 if (irq == NMI_VECTOR)
4908 vcpu->arch.nmi_pending = 0;
4910 vcpu->arch.interrupt.pending = false;
4912 return EMULATE_DONE;
4914 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4916 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4918 int r = EMULATE_DONE;
4920 ++vcpu->stat.insn_emulation_fail;
4921 trace_kvm_emulate_insn_failed(vcpu);
4922 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
4923 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4924 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4925 vcpu->run->internal.ndata = 0;
4928 kvm_queue_exception(vcpu, UD_VECTOR);
4933 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4934 bool write_fault_to_shadow_pgtable,
4940 if (emulation_type & EMULTYPE_NO_REEXECUTE)
4943 if (!vcpu->arch.mmu.direct_map) {
4945 * Write permission should be allowed since only
4946 * write access need to be emulated.
4948 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4951 * If the mapping is invalid in guest, let cpu retry
4952 * it to generate fault.
4954 if (gpa == UNMAPPED_GVA)
4959 * Do not retry the unhandleable instruction if it faults on the
4960 * readonly host memory, otherwise it will goto a infinite loop:
4961 * retry instruction -> write #PF -> emulation fail -> retry
4962 * instruction -> ...
4964 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4967 * If the instruction failed on the error pfn, it can not be fixed,
4968 * report the error to userspace.
4970 if (is_error_noslot_pfn(pfn))
4973 kvm_release_pfn_clean(pfn);
4975 /* The instructions are well-emulated on direct mmu. */
4976 if (vcpu->arch.mmu.direct_map) {
4977 unsigned int indirect_shadow_pages;
4979 spin_lock(&vcpu->kvm->mmu_lock);
4980 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
4981 spin_unlock(&vcpu->kvm->mmu_lock);
4983 if (indirect_shadow_pages)
4984 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4990 * if emulation was due to access to shadowed page table
4991 * and it failed try to unshadow page and re-enter the
4992 * guest to let CPU execute the instruction.
4994 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4997 * If the access faults on its page table, it can not
4998 * be fixed by unprotecting shadow page and it should
4999 * be reported to userspace.
5001 return !write_fault_to_shadow_pgtable;
5004 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5005 unsigned long cr2, int emulation_type)
5007 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5008 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5010 last_retry_eip = vcpu->arch.last_retry_eip;
5011 last_retry_addr = vcpu->arch.last_retry_addr;
5014 * If the emulation is caused by #PF and it is non-page_table
5015 * writing instruction, it means the VM-EXIT is caused by shadow
5016 * page protected, we can zap the shadow page and retry this
5017 * instruction directly.
5019 * Note: if the guest uses a non-page-table modifying instruction
5020 * on the PDE that points to the instruction, then we will unmap
5021 * the instruction and go to an infinite loop. So, we cache the
5022 * last retried eip and the last fault address, if we meet the eip
5023 * and the address again, we can break out of the potential infinite
5026 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5028 if (!(emulation_type & EMULTYPE_RETRY))
5031 if (x86_page_table_writing_insn(ctxt))
5034 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5037 vcpu->arch.last_retry_eip = ctxt->eip;
5038 vcpu->arch.last_retry_addr = cr2;
5040 if (!vcpu->arch.mmu.direct_map)
5041 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5043 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5048 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5049 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5051 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5053 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5054 /* This is a good place to trace that we are exiting SMM. */
5055 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5057 if (unlikely(vcpu->arch.smi_pending)) {
5058 kvm_make_request(KVM_REQ_SMI, vcpu);
5059 vcpu->arch.smi_pending = 0;
5061 /* Process a latched INIT, if any. */
5062 kvm_make_request(KVM_REQ_EVENT, vcpu);
5066 kvm_mmu_reset_context(vcpu);
5069 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5071 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5073 vcpu->arch.hflags = emul_flags;
5075 if (changed & HF_SMM_MASK)
5076 kvm_smm_changed(vcpu);
5079 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5088 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5089 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5094 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5096 struct kvm_run *kvm_run = vcpu->run;
5099 * rflags is the old, "raw" value of the flags. The new value has
5100 * not been saved yet.
5102 * This is correct even for TF set by the guest, because "the
5103 * processor will not generate this exception after the instruction
5104 * that sets the TF flag".
5106 if (unlikely(rflags & X86_EFLAGS_TF)) {
5107 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5108 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5110 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5111 kvm_run->debug.arch.exception = DB_VECTOR;
5112 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5113 *r = EMULATE_USER_EXIT;
5115 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5117 * "Certain debug exceptions may clear bit 0-3. The
5118 * remaining contents of the DR6 register are never
5119 * cleared by the processor".
5121 vcpu->arch.dr6 &= ~15;
5122 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5123 kvm_queue_exception(vcpu, DB_VECTOR);
5128 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5130 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5131 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5132 struct kvm_run *kvm_run = vcpu->run;
5133 unsigned long eip = kvm_get_linear_rip(vcpu);
5134 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5135 vcpu->arch.guest_debug_dr7,
5139 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5140 kvm_run->debug.arch.pc = eip;
5141 kvm_run->debug.arch.exception = DB_VECTOR;
5142 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5143 *r = EMULATE_USER_EXIT;
5148 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5149 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5150 unsigned long eip = kvm_get_linear_rip(vcpu);
5151 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5156 vcpu->arch.dr6 &= ~15;
5157 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5158 kvm_queue_exception(vcpu, DB_VECTOR);
5167 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5174 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5175 bool writeback = true;
5176 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5179 * Clear write_fault_to_shadow_pgtable here to ensure it is
5182 vcpu->arch.write_fault_to_shadow_pgtable = false;
5183 kvm_clear_exception_queue(vcpu);
5185 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5186 init_emulate_ctxt(vcpu);
5189 * We will reenter on the same instruction since
5190 * we do not set complete_userspace_io. This does not
5191 * handle watchpoints yet, those would be handled in
5194 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5197 ctxt->interruptibility = 0;
5198 ctxt->have_exception = false;
5199 ctxt->exception.vector = -1;
5200 ctxt->perm_ok = false;
5202 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5204 r = x86_decode_insn(ctxt, insn, insn_len);
5206 trace_kvm_emulate_insn_start(vcpu);
5207 ++vcpu->stat.insn_emulation;
5208 if (r != EMULATION_OK) {
5209 if (emulation_type & EMULTYPE_TRAP_UD)
5210 return EMULATE_FAIL;
5211 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5213 return EMULATE_DONE;
5214 if (emulation_type & EMULTYPE_SKIP)
5215 return EMULATE_FAIL;
5216 return handle_emulation_failure(vcpu);
5220 if (emulation_type & EMULTYPE_SKIP) {
5221 kvm_rip_write(vcpu, ctxt->_eip);
5222 if (ctxt->eflags & X86_EFLAGS_RF)
5223 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5224 return EMULATE_DONE;
5227 if (retry_instruction(ctxt, cr2, emulation_type))
5228 return EMULATE_DONE;
5230 /* this is needed for vmware backdoor interface to work since it
5231 changes registers values during IO operation */
5232 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5233 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5234 emulator_invalidate_register_cache(ctxt);
5238 r = x86_emulate_insn(ctxt);
5240 if (r == EMULATION_INTERCEPTED)
5241 return EMULATE_DONE;
5243 if (r == EMULATION_FAILED) {
5244 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5246 return EMULATE_DONE;
5248 return handle_emulation_failure(vcpu);
5251 if (ctxt->have_exception) {
5253 if (inject_emulated_exception(vcpu))
5255 } else if (vcpu->arch.pio.count) {
5256 if (!vcpu->arch.pio.in) {
5257 /* FIXME: return into emulator if single-stepping. */
5258 vcpu->arch.pio.count = 0;
5261 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5263 r = EMULATE_USER_EXIT;
5264 } else if (vcpu->mmio_needed) {
5265 if (!vcpu->mmio_is_write)
5267 r = EMULATE_USER_EXIT;
5268 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5269 } else if (r == EMULATION_RESTART)
5275 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5276 toggle_interruptibility(vcpu, ctxt->interruptibility);
5277 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5278 if (vcpu->arch.hflags != ctxt->emul_flags)
5279 kvm_set_hflags(vcpu, ctxt->emul_flags);
5280 kvm_rip_write(vcpu, ctxt->eip);
5281 if (r == EMULATE_DONE)
5282 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5283 if (!ctxt->have_exception ||
5284 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5285 __kvm_set_rflags(vcpu, ctxt->eflags);
5288 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5289 * do nothing, and it will be requested again as soon as
5290 * the shadow expires. But we still need to check here,
5291 * because POPF has no interrupt shadow.
5293 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5294 kvm_make_request(KVM_REQ_EVENT, vcpu);
5296 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5300 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5302 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5304 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5305 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5306 size, port, &val, 1);
5307 /* do not return to emulator after return from userspace */
5308 vcpu->arch.pio.count = 0;
5311 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5313 static void tsc_bad(void *info)
5315 __this_cpu_write(cpu_tsc_khz, 0);
5318 static void tsc_khz_changed(void *data)
5320 struct cpufreq_freqs *freq = data;
5321 unsigned long khz = 0;
5325 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5326 khz = cpufreq_quick_get(raw_smp_processor_id());
5329 __this_cpu_write(cpu_tsc_khz, khz);
5332 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5335 struct cpufreq_freqs *freq = data;
5337 struct kvm_vcpu *vcpu;
5338 int i, send_ipi = 0;
5341 * We allow guests to temporarily run on slowing clocks,
5342 * provided we notify them after, or to run on accelerating
5343 * clocks, provided we notify them before. Thus time never
5346 * However, we have a problem. We can't atomically update
5347 * the frequency of a given CPU from this function; it is
5348 * merely a notifier, which can be called from any CPU.
5349 * Changing the TSC frequency at arbitrary points in time
5350 * requires a recomputation of local variables related to
5351 * the TSC for each VCPU. We must flag these local variables
5352 * to be updated and be sure the update takes place with the
5353 * new frequency before any guests proceed.
5355 * Unfortunately, the combination of hotplug CPU and frequency
5356 * change creates an intractable locking scenario; the order
5357 * of when these callouts happen is undefined with respect to
5358 * CPU hotplug, and they can race with each other. As such,
5359 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5360 * undefined; you can actually have a CPU frequency change take
5361 * place in between the computation of X and the setting of the
5362 * variable. To protect against this problem, all updates of
5363 * the per_cpu tsc_khz variable are done in an interrupt
5364 * protected IPI, and all callers wishing to update the value
5365 * must wait for a synchronous IPI to complete (which is trivial
5366 * if the caller is on the CPU already). This establishes the
5367 * necessary total order on variable updates.
5369 * Note that because a guest time update may take place
5370 * anytime after the setting of the VCPU's request bit, the
5371 * correct TSC value must be set before the request. However,
5372 * to ensure the update actually makes it to any guest which
5373 * starts running in hardware virtualization between the set
5374 * and the acquisition of the spinlock, we must also ping the
5375 * CPU after setting the request bit.
5379 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5381 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5384 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5386 spin_lock(&kvm_lock);
5387 list_for_each_entry(kvm, &vm_list, vm_list) {
5388 kvm_for_each_vcpu(i, vcpu, kvm) {
5389 if (vcpu->cpu != freq->cpu)
5391 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5392 if (vcpu->cpu != smp_processor_id())
5396 spin_unlock(&kvm_lock);
5398 if (freq->old < freq->new && send_ipi) {
5400 * We upscale the frequency. Must make the guest
5401 * doesn't see old kvmclock values while running with
5402 * the new frequency, otherwise we risk the guest sees
5403 * time go backwards.
5405 * In case we update the frequency for another cpu
5406 * (which might be in guest context) send an interrupt
5407 * to kick the cpu out of guest context. Next time
5408 * guest context is entered kvmclock will be updated,
5409 * so the guest will not see stale values.
5411 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5416 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5417 .notifier_call = kvmclock_cpufreq_notifier
5420 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5421 unsigned long action, void *hcpu)
5423 unsigned int cpu = (unsigned long)hcpu;
5427 case CPU_DOWN_FAILED:
5428 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5430 case CPU_DOWN_PREPARE:
5431 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5437 static struct notifier_block kvmclock_cpu_notifier_block = {
5438 .notifier_call = kvmclock_cpu_notifier,
5439 .priority = -INT_MAX
5442 static void kvm_timer_init(void)
5446 max_tsc_khz = tsc_khz;
5448 cpu_notifier_register_begin();
5449 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5450 #ifdef CONFIG_CPU_FREQ
5451 struct cpufreq_policy policy;
5452 memset(&policy, 0, sizeof(policy));
5454 cpufreq_get_policy(&policy, cpu);
5455 if (policy.cpuinfo.max_freq)
5456 max_tsc_khz = policy.cpuinfo.max_freq;
5459 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5460 CPUFREQ_TRANSITION_NOTIFIER);
5462 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5463 for_each_online_cpu(cpu)
5464 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5466 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5467 cpu_notifier_register_done();
5471 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5473 int kvm_is_in_guest(void)
5475 return __this_cpu_read(current_vcpu) != NULL;
5478 static int kvm_is_user_mode(void)
5482 if (__this_cpu_read(current_vcpu))
5483 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5485 return user_mode != 0;
5488 static unsigned long kvm_get_guest_ip(void)
5490 unsigned long ip = 0;
5492 if (__this_cpu_read(current_vcpu))
5493 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5498 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5499 .is_in_guest = kvm_is_in_guest,
5500 .is_user_mode = kvm_is_user_mode,
5501 .get_guest_ip = kvm_get_guest_ip,
5504 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5506 __this_cpu_write(current_vcpu, vcpu);
5508 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5510 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5512 __this_cpu_write(current_vcpu, NULL);
5514 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5516 static void kvm_set_mmio_spte_mask(void)
5519 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5522 * Set the reserved bits and the present bit of an paging-structure
5523 * entry to generate page fault with PFER.RSV = 1.
5525 /* Mask the reserved physical address bits. */
5526 mask = rsvd_bits(maxphyaddr, 51);
5528 /* Bit 62 is always reserved for 32bit host. */
5529 mask |= 0x3ull << 62;
5531 /* Set the present bit. */
5534 #ifdef CONFIG_X86_64
5536 * If reserved bit is not supported, clear the present bit to disable
5539 if (maxphyaddr == 52)
5543 kvm_mmu_set_mmio_spte_mask(mask);
5546 #ifdef CONFIG_X86_64
5547 static void pvclock_gtod_update_fn(struct work_struct *work)
5551 struct kvm_vcpu *vcpu;
5554 spin_lock(&kvm_lock);
5555 list_for_each_entry(kvm, &vm_list, vm_list)
5556 kvm_for_each_vcpu(i, vcpu, kvm)
5557 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5558 atomic_set(&kvm_guest_has_master_clock, 0);
5559 spin_unlock(&kvm_lock);
5562 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5565 * Notification about pvclock gtod data update.
5567 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5570 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5571 struct timekeeper *tk = priv;
5573 update_pvclock_gtod(tk);
5575 /* disable master clock if host does not trust, or does not
5576 * use, TSC clocksource
5578 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5579 atomic_read(&kvm_guest_has_master_clock) != 0)
5580 queue_work(system_long_wq, &pvclock_gtod_work);
5585 static struct notifier_block pvclock_gtod_notifier = {
5586 .notifier_call = pvclock_gtod_notify,
5590 int kvm_arch_init(void *opaque)
5593 struct kvm_x86_ops *ops = opaque;
5596 printk(KERN_ERR "kvm: already loaded the other module\n");
5601 if (!ops->cpu_has_kvm_support()) {
5602 printk(KERN_ERR "kvm: no hardware support\n");
5606 if (ops->disabled_by_bios()) {
5607 printk(KERN_ERR "kvm: disabled by bios\n");
5613 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5615 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5619 r = kvm_mmu_module_init();
5621 goto out_free_percpu;
5623 kvm_set_mmio_spte_mask();
5627 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5628 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5632 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5635 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5638 #ifdef CONFIG_X86_64
5639 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5645 free_percpu(shared_msrs);
5650 void kvm_arch_exit(void)
5652 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5654 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5655 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5656 CPUFREQ_TRANSITION_NOTIFIER);
5657 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5658 #ifdef CONFIG_X86_64
5659 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5662 kvm_mmu_module_exit();
5663 free_percpu(shared_msrs);
5666 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5668 ++vcpu->stat.halt_exits;
5669 if (irqchip_in_kernel(vcpu->kvm)) {
5670 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5673 vcpu->run->exit_reason = KVM_EXIT_HLT;
5677 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5679 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5681 kvm_x86_ops->skip_emulated_instruction(vcpu);
5682 return kvm_vcpu_halt(vcpu);
5684 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5687 * kvm_pv_kick_cpu_op: Kick a vcpu.
5689 * @apicid - apicid of vcpu to be kicked.
5691 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5693 struct kvm_lapic_irq lapic_irq;
5695 lapic_irq.shorthand = 0;
5696 lapic_irq.dest_mode = 0;
5697 lapic_irq.dest_id = apicid;
5698 lapic_irq.msi_redir_hint = false;
5700 lapic_irq.delivery_mode = APIC_DM_REMRD;
5701 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5704 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5706 unsigned long nr, a0, a1, a2, a3, ret;
5707 int op_64_bit, r = 1;
5709 kvm_x86_ops->skip_emulated_instruction(vcpu);
5711 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5712 return kvm_hv_hypercall(vcpu);
5714 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5715 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5716 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5717 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5718 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5720 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5722 op_64_bit = is_64_bit_mode(vcpu);
5731 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5737 case KVM_HC_VAPIC_POLL_IRQ:
5740 case KVM_HC_KICK_CPU:
5741 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5751 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5752 ++vcpu->stat.hypercalls;
5755 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5757 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5759 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5760 char instruction[3];
5761 unsigned long rip = kvm_rip_read(vcpu);
5763 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5765 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5769 * Check if userspace requested an interrupt window, and that the
5770 * interrupt window is open.
5772 * No need to exit to userspace if we already have an interrupt queued.
5774 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5776 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5777 vcpu->run->request_interrupt_window &&
5778 kvm_arch_interrupt_allowed(vcpu));
5781 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5783 struct kvm_run *kvm_run = vcpu->run;
5785 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5786 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
5787 kvm_run->cr8 = kvm_get_cr8(vcpu);
5788 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5789 if (irqchip_in_kernel(vcpu->kvm))
5790 kvm_run->ready_for_interrupt_injection = 1;
5792 kvm_run->ready_for_interrupt_injection =
5793 kvm_arch_interrupt_allowed(vcpu) &&
5794 !kvm_cpu_has_interrupt(vcpu) &&
5795 !kvm_event_needs_reinjection(vcpu);
5798 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5802 if (!kvm_x86_ops->update_cr8_intercept)
5805 if (!vcpu->arch.apic)
5808 if (!vcpu->arch.apic->vapic_addr)
5809 max_irr = kvm_lapic_find_highest_irr(vcpu);
5816 tpr = kvm_lapic_get_cr8(vcpu);
5818 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5821 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
5825 /* try to reinject previous events if any */
5826 if (vcpu->arch.exception.pending) {
5827 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5828 vcpu->arch.exception.has_error_code,
5829 vcpu->arch.exception.error_code);
5831 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
5832 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
5835 if (vcpu->arch.exception.nr == DB_VECTOR &&
5836 (vcpu->arch.dr7 & DR7_GD)) {
5837 vcpu->arch.dr7 &= ~DR7_GD;
5838 kvm_update_dr7(vcpu);
5841 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5842 vcpu->arch.exception.has_error_code,
5843 vcpu->arch.exception.error_code,
5844 vcpu->arch.exception.reinject);
5848 if (vcpu->arch.nmi_injected) {
5849 kvm_x86_ops->set_nmi(vcpu);
5853 if (vcpu->arch.interrupt.pending) {
5854 kvm_x86_ops->set_irq(vcpu);
5858 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5859 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5864 /* try to inject new event if pending */
5865 if (vcpu->arch.nmi_pending) {
5866 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5867 --vcpu->arch.nmi_pending;
5868 vcpu->arch.nmi_injected = true;
5869 kvm_x86_ops->set_nmi(vcpu);
5871 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
5873 * Because interrupts can be injected asynchronously, we are
5874 * calling check_nested_events again here to avoid a race condition.
5875 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
5876 * proposal and current concerns. Perhaps we should be setting
5877 * KVM_REQ_EVENT only on certain events and not unconditionally?
5879 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5880 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5884 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5885 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5887 kvm_x86_ops->set_irq(vcpu);
5893 static void process_nmi(struct kvm_vcpu *vcpu)
5898 * x86 is limited to one NMI running, and one NMI pending after it.
5899 * If an NMI is already in progress, limit further NMIs to just one.
5900 * Otherwise, allow two (and we'll inject the first one immediately).
5902 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5905 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5906 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5907 kvm_make_request(KVM_REQ_EVENT, vcpu);
5910 #define put_smstate(type, buf, offset, val) \
5911 *(type *)((buf) + (offset) - 0x7e00) = val
5913 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
5916 flags |= seg->g << 23;
5917 flags |= seg->db << 22;
5918 flags |= seg->l << 21;
5919 flags |= seg->avl << 20;
5920 flags |= seg->present << 15;
5921 flags |= seg->dpl << 13;
5922 flags |= seg->s << 12;
5923 flags |= seg->type << 8;
5927 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
5929 struct kvm_segment seg;
5932 kvm_get_segment(vcpu, &seg, n);
5933 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
5936 offset = 0x7f84 + n * 12;
5938 offset = 0x7f2c + (n - 3) * 12;
5940 put_smstate(u32, buf, offset + 8, seg.base);
5941 put_smstate(u32, buf, offset + 4, seg.limit);
5942 put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
5945 #ifdef CONFIG_X86_64
5946 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
5948 struct kvm_segment seg;
5952 kvm_get_segment(vcpu, &seg, n);
5953 offset = 0x7e00 + n * 16;
5955 flags = process_smi_get_segment_flags(&seg) >> 8;
5956 put_smstate(u16, buf, offset, seg.selector);
5957 put_smstate(u16, buf, offset + 2, flags);
5958 put_smstate(u32, buf, offset + 4, seg.limit);
5959 put_smstate(u64, buf, offset + 8, seg.base);
5963 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
5966 struct kvm_segment seg;
5970 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
5971 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
5972 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
5973 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
5975 for (i = 0; i < 8; i++)
5976 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
5978 kvm_get_dr(vcpu, 6, &val);
5979 put_smstate(u32, buf, 0x7fcc, (u32)val);
5980 kvm_get_dr(vcpu, 7, &val);
5981 put_smstate(u32, buf, 0x7fc8, (u32)val);
5983 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
5984 put_smstate(u32, buf, 0x7fc4, seg.selector);
5985 put_smstate(u32, buf, 0x7f64, seg.base);
5986 put_smstate(u32, buf, 0x7f60, seg.limit);
5987 put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
5989 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
5990 put_smstate(u32, buf, 0x7fc0, seg.selector);
5991 put_smstate(u32, buf, 0x7f80, seg.base);
5992 put_smstate(u32, buf, 0x7f7c, seg.limit);
5993 put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
5995 kvm_x86_ops->get_gdt(vcpu, &dt);
5996 put_smstate(u32, buf, 0x7f74, dt.address);
5997 put_smstate(u32, buf, 0x7f70, dt.size);
5999 kvm_x86_ops->get_idt(vcpu, &dt);
6000 put_smstate(u32, buf, 0x7f58, dt.address);
6001 put_smstate(u32, buf, 0x7f54, dt.size);
6003 for (i = 0; i < 6; i++)
6004 process_smi_save_seg_32(vcpu, buf, i);
6006 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6009 put_smstate(u32, buf, 0x7efc, 0x00020000);
6010 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6013 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6015 #ifdef CONFIG_X86_64
6017 struct kvm_segment seg;
6021 for (i = 0; i < 16; i++)
6022 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6024 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6025 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6027 kvm_get_dr(vcpu, 6, &val);
6028 put_smstate(u64, buf, 0x7f68, val);
6029 kvm_get_dr(vcpu, 7, &val);
6030 put_smstate(u64, buf, 0x7f60, val);
6032 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6033 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6034 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6036 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6039 put_smstate(u32, buf, 0x7efc, 0x00020064);
6041 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6043 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6044 put_smstate(u16, buf, 0x7e90, seg.selector);
6045 put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6046 put_smstate(u32, buf, 0x7e94, seg.limit);
6047 put_smstate(u64, buf, 0x7e98, seg.base);
6049 kvm_x86_ops->get_idt(vcpu, &dt);
6050 put_smstate(u32, buf, 0x7e84, dt.size);
6051 put_smstate(u64, buf, 0x7e88, dt.address);
6053 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6054 put_smstate(u16, buf, 0x7e70, seg.selector);
6055 put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6056 put_smstate(u32, buf, 0x7e74, seg.limit);
6057 put_smstate(u64, buf, 0x7e78, seg.base);
6059 kvm_x86_ops->get_gdt(vcpu, &dt);
6060 put_smstate(u32, buf, 0x7e64, dt.size);
6061 put_smstate(u64, buf, 0x7e68, dt.address);
6063 for (i = 0; i < 6; i++)
6064 process_smi_save_seg_64(vcpu, buf, i);
6070 static void process_smi(struct kvm_vcpu *vcpu)
6072 struct kvm_segment cs, ds;
6078 vcpu->arch.smi_pending = true;
6082 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6083 vcpu->arch.hflags |= HF_SMM_MASK;
6084 memset(buf, 0, 512);
6085 if (guest_cpuid_has_longmode(vcpu))
6086 process_smi_save_state_64(vcpu, buf);
6088 process_smi_save_state_32(vcpu, buf);
6090 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6092 if (kvm_x86_ops->get_nmi_mask(vcpu))
6093 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6095 kvm_x86_ops->set_nmi_mask(vcpu, true);
6097 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6098 kvm_rip_write(vcpu, 0x8000);
6100 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6101 kvm_x86_ops->set_cr0(vcpu, cr0);
6102 vcpu->arch.cr0 = cr0;
6104 kvm_x86_ops->set_cr4(vcpu, 0);
6106 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6107 dt.address = dt.size = 0;
6108 kvm_x86_ops->set_idt(vcpu, &dt);
6110 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6112 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6113 cs.base = vcpu->arch.smbase;
6118 cs.limit = ds.limit = 0xffffffff;
6119 cs.type = ds.type = 0x3;
6120 cs.dpl = ds.dpl = 0;
6125 cs.avl = ds.avl = 0;
6126 cs.present = ds.present = 1;
6127 cs.unusable = ds.unusable = 0;
6128 cs.padding = ds.padding = 0;
6130 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6131 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6132 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6133 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6134 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6135 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6137 if (guest_cpuid_has_longmode(vcpu))
6138 kvm_x86_ops->set_efer(vcpu, 0);
6140 kvm_update_cpuid(vcpu);
6141 kvm_mmu_reset_context(vcpu);
6144 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6146 u64 eoi_exit_bitmap[4];
6149 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6152 memset(eoi_exit_bitmap, 0, 32);
6155 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6156 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6157 kvm_apic_update_tmr(vcpu, tmr);
6160 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6162 ++vcpu->stat.tlb_flush;
6163 kvm_x86_ops->tlb_flush(vcpu);
6166 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6168 struct page *page = NULL;
6170 if (!irqchip_in_kernel(vcpu->kvm))
6173 if (!kvm_x86_ops->set_apic_access_page_addr)
6176 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6177 if (is_error_page(page))
6179 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6182 * Do not pin apic access page in memory, the MMU notifier
6183 * will call us again if it is migrated or swapped out.
6187 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6189 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6190 unsigned long address)
6193 * The physical address of apic access page is stored in the VMCS.
6194 * Update it when it becomes invalid.
6196 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6197 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6201 * Returns 1 to let vcpu_run() continue the guest execution loop without
6202 * exiting to the userspace. Otherwise, the value will be returned to the
6205 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6208 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6209 vcpu->run->request_interrupt_window;
6210 bool req_immediate_exit = false;
6212 if (vcpu->requests) {
6213 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6214 kvm_mmu_unload(vcpu);
6215 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6216 __kvm_migrate_timers(vcpu);
6217 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6218 kvm_gen_update_masterclock(vcpu->kvm);
6219 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6220 kvm_gen_kvmclock_update(vcpu);
6221 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6222 r = kvm_guest_time_update(vcpu);
6226 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6227 kvm_mmu_sync_roots(vcpu);
6228 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6229 kvm_vcpu_flush_tlb(vcpu);
6230 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6231 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6235 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6236 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6240 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6241 vcpu->fpu_active = 0;
6242 kvm_x86_ops->fpu_deactivate(vcpu);
6244 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6245 /* Page is swapped out. Do synthetic halt */
6246 vcpu->arch.apf.halted = true;
6250 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6251 record_steal_time(vcpu);
6252 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6254 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6256 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6257 kvm_pmu_handle_event(vcpu);
6258 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6259 kvm_pmu_deliver_pmi(vcpu);
6260 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6261 vcpu_scan_ioapic(vcpu);
6262 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6263 kvm_vcpu_reload_apic_access_page(vcpu);
6264 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6265 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6266 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6272 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6273 kvm_apic_accept_events(vcpu);
6274 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6279 if (inject_pending_event(vcpu, req_int_win) != 0)
6280 req_immediate_exit = true;
6281 /* enable NMI/IRQ window open exits if needed */
6282 else if (vcpu->arch.nmi_pending)
6283 kvm_x86_ops->enable_nmi_window(vcpu);
6284 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6285 kvm_x86_ops->enable_irq_window(vcpu);
6287 if (kvm_lapic_enabled(vcpu)) {
6289 * Update architecture specific hints for APIC
6290 * virtual interrupt delivery.
6292 if (kvm_x86_ops->hwapic_irr_update)
6293 kvm_x86_ops->hwapic_irr_update(vcpu,
6294 kvm_lapic_find_highest_irr(vcpu));
6295 update_cr8_intercept(vcpu);
6296 kvm_lapic_sync_to_vapic(vcpu);
6300 r = kvm_mmu_reload(vcpu);
6302 goto cancel_injection;
6307 kvm_x86_ops->prepare_guest_switch(vcpu);
6308 if (vcpu->fpu_active)
6309 kvm_load_guest_fpu(vcpu);
6310 kvm_load_guest_xcr0(vcpu);
6312 vcpu->mode = IN_GUEST_MODE;
6314 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6316 /* We should set ->mode before check ->requests,
6317 * see the comment in make_all_cpus_request.
6319 smp_mb__after_srcu_read_unlock();
6321 local_irq_disable();
6323 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6324 || need_resched() || signal_pending(current)) {
6325 vcpu->mode = OUTSIDE_GUEST_MODE;
6329 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6331 goto cancel_injection;
6334 if (req_immediate_exit)
6335 smp_send_reschedule(vcpu->cpu);
6337 __kvm_guest_enter();
6339 if (unlikely(vcpu->arch.switch_db_regs)) {
6341 set_debugreg(vcpu->arch.eff_db[0], 0);
6342 set_debugreg(vcpu->arch.eff_db[1], 1);
6343 set_debugreg(vcpu->arch.eff_db[2], 2);
6344 set_debugreg(vcpu->arch.eff_db[3], 3);
6345 set_debugreg(vcpu->arch.dr6, 6);
6346 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6349 trace_kvm_entry(vcpu->vcpu_id);
6350 wait_lapic_expire(vcpu);
6351 kvm_x86_ops->run(vcpu);
6354 * Do this here before restoring debug registers on the host. And
6355 * since we do this before handling the vmexit, a DR access vmexit
6356 * can (a) read the correct value of the debug registers, (b) set
6357 * KVM_DEBUGREG_WONT_EXIT again.
6359 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6362 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6363 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6364 for (i = 0; i < KVM_NR_DB_REGS; i++)
6365 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6369 * If the guest has used debug registers, at least dr7
6370 * will be disabled while returning to the host.
6371 * If we don't have active breakpoints in the host, we don't
6372 * care about the messed up debug address registers. But if
6373 * we have some of them active, restore the old state.
6375 if (hw_breakpoint_active())
6376 hw_breakpoint_restore();
6378 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6381 vcpu->mode = OUTSIDE_GUEST_MODE;
6384 /* Interrupt is enabled by handle_external_intr() */
6385 kvm_x86_ops->handle_external_intr(vcpu);
6390 * We must have an instruction between local_irq_enable() and
6391 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6392 * the interrupt shadow. The stat.exits increment will do nicely.
6393 * But we need to prevent reordering, hence this barrier():
6401 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6404 * Profile KVM exit RIPs:
6406 if (unlikely(prof_on == KVM_PROFILING)) {
6407 unsigned long rip = kvm_rip_read(vcpu);
6408 profile_hit(KVM_PROFILING, (void *)rip);
6411 if (unlikely(vcpu->arch.tsc_always_catchup))
6412 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6414 if (vcpu->arch.apic_attention)
6415 kvm_lapic_sync_from_vapic(vcpu);
6417 r = kvm_x86_ops->handle_exit(vcpu);
6421 kvm_x86_ops->cancel_injection(vcpu);
6422 if (unlikely(vcpu->arch.apic_attention))
6423 kvm_lapic_sync_from_vapic(vcpu);
6428 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6430 if (!kvm_arch_vcpu_runnable(vcpu)) {
6431 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6432 kvm_vcpu_block(vcpu);
6433 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6434 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6438 kvm_apic_accept_events(vcpu);
6439 switch(vcpu->arch.mp_state) {
6440 case KVM_MP_STATE_HALTED:
6441 vcpu->arch.pv.pv_unhalted = false;
6442 vcpu->arch.mp_state =
6443 KVM_MP_STATE_RUNNABLE;
6444 case KVM_MP_STATE_RUNNABLE:
6445 vcpu->arch.apf.halted = false;
6447 case KVM_MP_STATE_INIT_RECEIVED:
6456 static int vcpu_run(struct kvm_vcpu *vcpu)
6459 struct kvm *kvm = vcpu->kvm;
6461 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6464 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6465 !vcpu->arch.apf.halted)
6466 r = vcpu_enter_guest(vcpu);
6468 r = vcpu_block(kvm, vcpu);
6472 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6473 if (kvm_cpu_has_pending_timer(vcpu))
6474 kvm_inject_pending_timer_irqs(vcpu);
6476 if (dm_request_for_irq_injection(vcpu)) {
6478 vcpu->run->exit_reason = KVM_EXIT_INTR;
6479 ++vcpu->stat.request_irq_exits;
6483 kvm_check_async_pf_completion(vcpu);
6485 if (signal_pending(current)) {
6487 vcpu->run->exit_reason = KVM_EXIT_INTR;
6488 ++vcpu->stat.signal_exits;
6491 if (need_resched()) {
6492 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6494 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6498 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6503 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6506 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6507 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6508 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6509 if (r != EMULATE_DONE)
6514 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6516 BUG_ON(!vcpu->arch.pio.count);
6518 return complete_emulated_io(vcpu);
6522 * Implements the following, as a state machine:
6526 * for each mmio piece in the fragment
6534 * for each mmio piece in the fragment
6539 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6541 struct kvm_run *run = vcpu->run;
6542 struct kvm_mmio_fragment *frag;
6545 BUG_ON(!vcpu->mmio_needed);
6547 /* Complete previous fragment */
6548 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6549 len = min(8u, frag->len);
6550 if (!vcpu->mmio_is_write)
6551 memcpy(frag->data, run->mmio.data, len);
6553 if (frag->len <= 8) {
6554 /* Switch to the next fragment. */
6556 vcpu->mmio_cur_fragment++;
6558 /* Go forward to the next mmio piece. */
6564 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6565 vcpu->mmio_needed = 0;
6567 /* FIXME: return into emulator if single-stepping. */
6568 if (vcpu->mmio_is_write)
6570 vcpu->mmio_read_completed = 1;
6571 return complete_emulated_io(vcpu);
6574 run->exit_reason = KVM_EXIT_MMIO;
6575 run->mmio.phys_addr = frag->gpa;
6576 if (vcpu->mmio_is_write)
6577 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6578 run->mmio.len = min(8u, frag->len);
6579 run->mmio.is_write = vcpu->mmio_is_write;
6580 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6585 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6587 struct fpu *fpu = ¤t->thread.fpu;
6591 fpu__activate_curr(fpu);
6593 if (vcpu->sigset_active)
6594 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6596 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6597 kvm_vcpu_block(vcpu);
6598 kvm_apic_accept_events(vcpu);
6599 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6604 /* re-sync apic's tpr */
6605 if (!irqchip_in_kernel(vcpu->kvm)) {
6606 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6612 if (unlikely(vcpu->arch.complete_userspace_io)) {
6613 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6614 vcpu->arch.complete_userspace_io = NULL;
6619 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6624 post_kvm_run_save(vcpu);
6625 if (vcpu->sigset_active)
6626 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6631 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6633 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6635 * We are here if userspace calls get_regs() in the middle of
6636 * instruction emulation. Registers state needs to be copied
6637 * back from emulation context to vcpu. Userspace shouldn't do
6638 * that usually, but some bad designed PV devices (vmware
6639 * backdoor interface) need this to work
6641 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6642 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6644 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6645 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6646 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6647 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6648 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6649 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6650 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6651 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6652 #ifdef CONFIG_X86_64
6653 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6654 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6655 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6656 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6657 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6658 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6659 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6660 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6663 regs->rip = kvm_rip_read(vcpu);
6664 regs->rflags = kvm_get_rflags(vcpu);
6669 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6671 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6672 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6674 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6675 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6676 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6677 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6678 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6679 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6680 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6681 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6682 #ifdef CONFIG_X86_64
6683 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6684 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6685 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6686 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6687 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6688 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6689 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6690 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6693 kvm_rip_write(vcpu, regs->rip);
6694 kvm_set_rflags(vcpu, regs->rflags);
6696 vcpu->arch.exception.pending = false;
6698 kvm_make_request(KVM_REQ_EVENT, vcpu);
6703 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6705 struct kvm_segment cs;
6707 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6711 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6713 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6714 struct kvm_sregs *sregs)
6718 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6719 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6720 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6721 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6722 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6723 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6725 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6726 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6728 kvm_x86_ops->get_idt(vcpu, &dt);
6729 sregs->idt.limit = dt.size;
6730 sregs->idt.base = dt.address;
6731 kvm_x86_ops->get_gdt(vcpu, &dt);
6732 sregs->gdt.limit = dt.size;
6733 sregs->gdt.base = dt.address;
6735 sregs->cr0 = kvm_read_cr0(vcpu);
6736 sregs->cr2 = vcpu->arch.cr2;
6737 sregs->cr3 = kvm_read_cr3(vcpu);
6738 sregs->cr4 = kvm_read_cr4(vcpu);
6739 sregs->cr8 = kvm_get_cr8(vcpu);
6740 sregs->efer = vcpu->arch.efer;
6741 sregs->apic_base = kvm_get_apic_base(vcpu);
6743 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6745 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6746 set_bit(vcpu->arch.interrupt.nr,
6747 (unsigned long *)sregs->interrupt_bitmap);
6752 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6753 struct kvm_mp_state *mp_state)
6755 kvm_apic_accept_events(vcpu);
6756 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6757 vcpu->arch.pv.pv_unhalted)
6758 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6760 mp_state->mp_state = vcpu->arch.mp_state;
6765 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6766 struct kvm_mp_state *mp_state)
6768 if (!kvm_vcpu_has_lapic(vcpu) &&
6769 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6772 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6773 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6774 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6776 vcpu->arch.mp_state = mp_state->mp_state;
6777 kvm_make_request(KVM_REQ_EVENT, vcpu);
6781 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6782 int reason, bool has_error_code, u32 error_code)
6784 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6787 init_emulate_ctxt(vcpu);
6789 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6790 has_error_code, error_code);
6793 return EMULATE_FAIL;
6795 kvm_rip_write(vcpu, ctxt->eip);
6796 kvm_set_rflags(vcpu, ctxt->eflags);
6797 kvm_make_request(KVM_REQ_EVENT, vcpu);
6798 return EMULATE_DONE;
6800 EXPORT_SYMBOL_GPL(kvm_task_switch);
6802 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6803 struct kvm_sregs *sregs)
6805 struct msr_data apic_base_msr;
6806 int mmu_reset_needed = 0;
6807 int pending_vec, max_bits, idx;
6810 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6813 dt.size = sregs->idt.limit;
6814 dt.address = sregs->idt.base;
6815 kvm_x86_ops->set_idt(vcpu, &dt);
6816 dt.size = sregs->gdt.limit;
6817 dt.address = sregs->gdt.base;
6818 kvm_x86_ops->set_gdt(vcpu, &dt);
6820 vcpu->arch.cr2 = sregs->cr2;
6821 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6822 vcpu->arch.cr3 = sregs->cr3;
6823 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6825 kvm_set_cr8(vcpu, sregs->cr8);
6827 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6828 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6829 apic_base_msr.data = sregs->apic_base;
6830 apic_base_msr.host_initiated = true;
6831 kvm_set_apic_base(vcpu, &apic_base_msr);
6833 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6834 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6835 vcpu->arch.cr0 = sregs->cr0;
6837 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6838 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6839 if (sregs->cr4 & X86_CR4_OSXSAVE)
6840 kvm_update_cpuid(vcpu);
6842 idx = srcu_read_lock(&vcpu->kvm->srcu);
6843 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6844 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6845 mmu_reset_needed = 1;
6847 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6849 if (mmu_reset_needed)
6850 kvm_mmu_reset_context(vcpu);
6852 max_bits = KVM_NR_INTERRUPTS;
6853 pending_vec = find_first_bit(
6854 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6855 if (pending_vec < max_bits) {
6856 kvm_queue_interrupt(vcpu, pending_vec, false);
6857 pr_debug("Set back pending irq %d\n", pending_vec);
6860 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6861 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6862 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6863 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6864 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6865 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6867 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6868 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6870 update_cr8_intercept(vcpu);
6872 /* Older userspace won't unhalt the vcpu on reset. */
6873 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6874 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6876 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6878 kvm_make_request(KVM_REQ_EVENT, vcpu);
6883 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6884 struct kvm_guest_debug *dbg)
6886 unsigned long rflags;
6889 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6891 if (vcpu->arch.exception.pending)
6893 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6894 kvm_queue_exception(vcpu, DB_VECTOR);
6896 kvm_queue_exception(vcpu, BP_VECTOR);
6900 * Read rflags as long as potentially injected trace flags are still
6903 rflags = kvm_get_rflags(vcpu);
6905 vcpu->guest_debug = dbg->control;
6906 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6907 vcpu->guest_debug = 0;
6909 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6910 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6911 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6912 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6914 for (i = 0; i < KVM_NR_DB_REGS; i++)
6915 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6917 kvm_update_dr7(vcpu);
6919 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6920 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6921 get_segment_base(vcpu, VCPU_SREG_CS);
6924 * Trigger an rflags update that will inject or remove the trace
6927 kvm_set_rflags(vcpu, rflags);
6929 kvm_x86_ops->update_db_bp_intercept(vcpu);
6939 * Translate a guest virtual address to a guest physical address.
6941 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6942 struct kvm_translation *tr)
6944 unsigned long vaddr = tr->linear_address;
6948 idx = srcu_read_lock(&vcpu->kvm->srcu);
6949 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6950 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6951 tr->physical_address = gpa;
6952 tr->valid = gpa != UNMAPPED_GVA;
6959 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6961 struct fxregs_state *fxsave =
6962 &vcpu->arch.guest_fpu.state.fxsave;
6964 memcpy(fpu->fpr, fxsave->st_space, 128);
6965 fpu->fcw = fxsave->cwd;
6966 fpu->fsw = fxsave->swd;
6967 fpu->ftwx = fxsave->twd;
6968 fpu->last_opcode = fxsave->fop;
6969 fpu->last_ip = fxsave->rip;
6970 fpu->last_dp = fxsave->rdp;
6971 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6976 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6978 struct fxregs_state *fxsave =
6979 &vcpu->arch.guest_fpu.state.fxsave;
6981 memcpy(fxsave->st_space, fpu->fpr, 128);
6982 fxsave->cwd = fpu->fcw;
6983 fxsave->swd = fpu->fsw;
6984 fxsave->twd = fpu->ftwx;
6985 fxsave->fop = fpu->last_opcode;
6986 fxsave->rip = fpu->last_ip;
6987 fxsave->rdp = fpu->last_dp;
6988 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6993 static void fx_init(struct kvm_vcpu *vcpu)
6995 fpstate_init(&vcpu->arch.guest_fpu.state);
6997 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
6998 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7001 * Ensure guest xcr0 is valid for loading
7003 vcpu->arch.xcr0 = XSTATE_FP;
7005 vcpu->arch.cr0 |= X86_CR0_ET;
7008 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7010 if (vcpu->guest_fpu_loaded)
7014 * Restore all possible states in the guest,
7015 * and assume host would use all available bits.
7016 * Guest xcr0 would be loaded later.
7018 kvm_put_guest_xcr0(vcpu);
7019 vcpu->guest_fpu_loaded = 1;
7020 __kernel_fpu_begin();
7021 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7025 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7027 kvm_put_guest_xcr0(vcpu);
7029 if (!vcpu->guest_fpu_loaded) {
7030 vcpu->fpu_counter = 0;
7034 vcpu->guest_fpu_loaded = 0;
7035 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7037 ++vcpu->stat.fpu_reload;
7039 * If using eager FPU mode, or if the guest is a frequent user
7040 * of the FPU, just leave the FPU active for next time.
7041 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7042 * the FPU in bursts will revert to loading it on demand.
7044 if (!vcpu->arch.eager_fpu) {
7045 if (++vcpu->fpu_counter < 5)
7046 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7051 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7053 kvmclock_reset(vcpu);
7055 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7056 kvm_x86_ops->vcpu_free(vcpu);
7059 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7062 struct kvm_vcpu *vcpu;
7064 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7065 printk_once(KERN_WARNING
7066 "kvm: SMP vm created on host with unstable TSC; "
7067 "guest TSC will not be reliable\n");
7069 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7074 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7078 kvm_vcpu_mtrr_init(vcpu);
7079 r = vcpu_load(vcpu);
7082 kvm_vcpu_reset(vcpu, false);
7083 kvm_mmu_setup(vcpu);
7088 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7090 struct msr_data msr;
7091 struct kvm *kvm = vcpu->kvm;
7093 if (vcpu_load(vcpu))
7096 msr.index = MSR_IA32_TSC;
7097 msr.host_initiated = true;
7098 kvm_write_tsc(vcpu, &msr);
7101 if (!kvmclock_periodic_sync)
7104 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7105 KVMCLOCK_SYNC_PERIOD);
7108 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7111 vcpu->arch.apf.msr_val = 0;
7113 r = vcpu_load(vcpu);
7115 kvm_mmu_unload(vcpu);
7118 kvm_x86_ops->vcpu_free(vcpu);
7121 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7123 vcpu->arch.hflags = 0;
7125 atomic_set(&vcpu->arch.nmi_queued, 0);
7126 vcpu->arch.nmi_pending = 0;
7127 vcpu->arch.nmi_injected = false;
7128 kvm_clear_interrupt_queue(vcpu);
7129 kvm_clear_exception_queue(vcpu);
7131 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7132 kvm_update_dr0123(vcpu);
7133 vcpu->arch.dr6 = DR6_INIT;
7134 kvm_update_dr6(vcpu);
7135 vcpu->arch.dr7 = DR7_FIXED_1;
7136 kvm_update_dr7(vcpu);
7140 kvm_make_request(KVM_REQ_EVENT, vcpu);
7141 vcpu->arch.apf.msr_val = 0;
7142 vcpu->arch.st.msr_val = 0;
7144 kvmclock_reset(vcpu);
7146 kvm_clear_async_pf_completion_queue(vcpu);
7147 kvm_async_pf_hash_reset(vcpu);
7148 vcpu->arch.apf.halted = false;
7151 kvm_pmu_reset(vcpu);
7152 vcpu->arch.smbase = 0x30000;
7155 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7156 vcpu->arch.regs_avail = ~0;
7157 vcpu->arch.regs_dirty = ~0;
7159 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7162 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7164 struct kvm_segment cs;
7166 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7167 cs.selector = vector << 8;
7168 cs.base = vector << 12;
7169 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7170 kvm_rip_write(vcpu, 0);
7173 int kvm_arch_hardware_enable(void)
7176 struct kvm_vcpu *vcpu;
7181 bool stable, backwards_tsc = false;
7183 kvm_shared_msr_cpu_online();
7184 ret = kvm_x86_ops->hardware_enable();
7188 local_tsc = rdtsc();
7189 stable = !check_tsc_unstable();
7190 list_for_each_entry(kvm, &vm_list, vm_list) {
7191 kvm_for_each_vcpu(i, vcpu, kvm) {
7192 if (!stable && vcpu->cpu == smp_processor_id())
7193 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7194 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7195 backwards_tsc = true;
7196 if (vcpu->arch.last_host_tsc > max_tsc)
7197 max_tsc = vcpu->arch.last_host_tsc;
7203 * Sometimes, even reliable TSCs go backwards. This happens on
7204 * platforms that reset TSC during suspend or hibernate actions, but
7205 * maintain synchronization. We must compensate. Fortunately, we can
7206 * detect that condition here, which happens early in CPU bringup,
7207 * before any KVM threads can be running. Unfortunately, we can't
7208 * bring the TSCs fully up to date with real time, as we aren't yet far
7209 * enough into CPU bringup that we know how much real time has actually
7210 * elapsed; our helper function, get_kernel_ns() will be using boot
7211 * variables that haven't been updated yet.
7213 * So we simply find the maximum observed TSC above, then record the
7214 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7215 * the adjustment will be applied. Note that we accumulate
7216 * adjustments, in case multiple suspend cycles happen before some VCPU
7217 * gets a chance to run again. In the event that no KVM threads get a
7218 * chance to run, we will miss the entire elapsed period, as we'll have
7219 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7220 * loose cycle time. This isn't too big a deal, since the loss will be
7221 * uniform across all VCPUs (not to mention the scenario is extremely
7222 * unlikely). It is possible that a second hibernate recovery happens
7223 * much faster than a first, causing the observed TSC here to be
7224 * smaller; this would require additional padding adjustment, which is
7225 * why we set last_host_tsc to the local tsc observed here.
7227 * N.B. - this code below runs only on platforms with reliable TSC,
7228 * as that is the only way backwards_tsc is set above. Also note
7229 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7230 * have the same delta_cyc adjustment applied if backwards_tsc
7231 * is detected. Note further, this adjustment is only done once,
7232 * as we reset last_host_tsc on all VCPUs to stop this from being
7233 * called multiple times (one for each physical CPU bringup).
7235 * Platforms with unreliable TSCs don't have to deal with this, they
7236 * will be compensated by the logic in vcpu_load, which sets the TSC to
7237 * catchup mode. This will catchup all VCPUs to real time, but cannot
7238 * guarantee that they stay in perfect synchronization.
7240 if (backwards_tsc) {
7241 u64 delta_cyc = max_tsc - local_tsc;
7242 backwards_tsc_observed = true;
7243 list_for_each_entry(kvm, &vm_list, vm_list) {
7244 kvm_for_each_vcpu(i, vcpu, kvm) {
7245 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7246 vcpu->arch.last_host_tsc = local_tsc;
7247 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7251 * We have to disable TSC offset matching.. if you were
7252 * booting a VM while issuing an S4 host suspend....
7253 * you may have some problem. Solving this issue is
7254 * left as an exercise to the reader.
7256 kvm->arch.last_tsc_nsec = 0;
7257 kvm->arch.last_tsc_write = 0;
7264 void kvm_arch_hardware_disable(void)
7266 kvm_x86_ops->hardware_disable();
7267 drop_user_return_notifiers();
7270 int kvm_arch_hardware_setup(void)
7274 r = kvm_x86_ops->hardware_setup();
7278 kvm_init_msr_list();
7282 void kvm_arch_hardware_unsetup(void)
7284 kvm_x86_ops->hardware_unsetup();
7287 void kvm_arch_check_processor_compat(void *rtn)
7289 kvm_x86_ops->check_processor_compatibility(rtn);
7292 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7294 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7296 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7298 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7300 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7303 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7305 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7308 struct static_key kvm_no_apic_vcpu __read_mostly;
7310 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7316 BUG_ON(vcpu->kvm == NULL);
7319 vcpu->arch.pv.pv_unhalted = false;
7320 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7321 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7322 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7324 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7326 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7331 vcpu->arch.pio_data = page_address(page);
7333 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7335 r = kvm_mmu_create(vcpu);
7337 goto fail_free_pio_data;
7339 if (irqchip_in_kernel(kvm)) {
7340 r = kvm_create_lapic(vcpu);
7342 goto fail_mmu_destroy;
7344 static_key_slow_inc(&kvm_no_apic_vcpu);
7346 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7348 if (!vcpu->arch.mce_banks) {
7350 goto fail_free_lapic;
7352 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7354 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7356 goto fail_free_mce_banks;
7361 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7362 vcpu->arch.pv_time_enabled = false;
7364 vcpu->arch.guest_supported_xcr0 = 0;
7365 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7367 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7369 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7371 kvm_async_pf_hash_reset(vcpu);
7376 fail_free_mce_banks:
7377 kfree(vcpu->arch.mce_banks);
7379 kvm_free_lapic(vcpu);
7381 kvm_mmu_destroy(vcpu);
7383 free_page((unsigned long)vcpu->arch.pio_data);
7388 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7392 kvm_pmu_destroy(vcpu);
7393 kfree(vcpu->arch.mce_banks);
7394 kvm_free_lapic(vcpu);
7395 idx = srcu_read_lock(&vcpu->kvm->srcu);
7396 kvm_mmu_destroy(vcpu);
7397 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7398 free_page((unsigned long)vcpu->arch.pio_data);
7399 if (!irqchip_in_kernel(vcpu->kvm))
7400 static_key_slow_dec(&kvm_no_apic_vcpu);
7403 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7405 kvm_x86_ops->sched_in(vcpu, cpu);
7408 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7413 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7414 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7415 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7416 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7417 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7419 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7420 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7421 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7422 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7423 &kvm->arch.irq_sources_bitmap);
7425 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7426 mutex_init(&kvm->arch.apic_map_lock);
7427 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7429 pvclock_update_vm_gtod_copy(kvm);
7431 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7432 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7437 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7440 r = vcpu_load(vcpu);
7442 kvm_mmu_unload(vcpu);
7446 static void kvm_free_vcpus(struct kvm *kvm)
7449 struct kvm_vcpu *vcpu;
7452 * Unpin any mmu pages first.
7454 kvm_for_each_vcpu(i, vcpu, kvm) {
7455 kvm_clear_async_pf_completion_queue(vcpu);
7456 kvm_unload_vcpu_mmu(vcpu);
7458 kvm_for_each_vcpu(i, vcpu, kvm)
7459 kvm_arch_vcpu_free(vcpu);
7461 mutex_lock(&kvm->lock);
7462 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7463 kvm->vcpus[i] = NULL;
7465 atomic_set(&kvm->online_vcpus, 0);
7466 mutex_unlock(&kvm->lock);
7469 void kvm_arch_sync_events(struct kvm *kvm)
7471 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7472 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7473 kvm_free_all_assigned_devices(kvm);
7477 int __x86_set_memory_region(struct kvm *kvm,
7478 const struct kvm_userspace_memory_region *mem)
7482 /* Called with kvm->slots_lock held. */
7483 BUG_ON(mem->slot >= KVM_MEM_SLOTS_NUM);
7485 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7486 struct kvm_userspace_memory_region m = *mem;
7489 r = __kvm_set_memory_region(kvm, &m);
7496 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7498 int x86_set_memory_region(struct kvm *kvm,
7499 const struct kvm_userspace_memory_region *mem)
7503 mutex_lock(&kvm->slots_lock);
7504 r = __x86_set_memory_region(kvm, mem);
7505 mutex_unlock(&kvm->slots_lock);
7509 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7511 void kvm_arch_destroy_vm(struct kvm *kvm)
7513 if (current->mm == kvm->mm) {
7515 * Free memory regions allocated on behalf of userspace,
7516 * unless the the memory map has changed due to process exit
7519 struct kvm_userspace_memory_region mem;
7520 memset(&mem, 0, sizeof(mem));
7521 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7522 x86_set_memory_region(kvm, &mem);
7524 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7525 x86_set_memory_region(kvm, &mem);
7527 mem.slot = TSS_PRIVATE_MEMSLOT;
7528 x86_set_memory_region(kvm, &mem);
7530 kvm_iommu_unmap_guest(kvm);
7531 kfree(kvm->arch.vpic);
7532 kfree(kvm->arch.vioapic);
7533 kvm_free_vcpus(kvm);
7534 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7537 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7538 struct kvm_memory_slot *dont)
7542 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7543 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7544 kvfree(free->arch.rmap[i]);
7545 free->arch.rmap[i] = NULL;
7550 if (!dont || free->arch.lpage_info[i - 1] !=
7551 dont->arch.lpage_info[i - 1]) {
7552 kvfree(free->arch.lpage_info[i - 1]);
7553 free->arch.lpage_info[i - 1] = NULL;
7558 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7559 unsigned long npages)
7563 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7568 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7569 slot->base_gfn, level) + 1;
7571 slot->arch.rmap[i] =
7572 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7573 if (!slot->arch.rmap[i])
7578 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7579 sizeof(*slot->arch.lpage_info[i - 1]));
7580 if (!slot->arch.lpage_info[i - 1])
7583 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7584 slot->arch.lpage_info[i - 1][0].write_count = 1;
7585 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7586 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7587 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7589 * If the gfn and userspace address are not aligned wrt each
7590 * other, or if explicitly asked to, disable large page
7591 * support for this slot
7593 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7594 !kvm_largepages_enabled()) {
7597 for (j = 0; j < lpages; ++j)
7598 slot->arch.lpage_info[i - 1][j].write_count = 1;
7605 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7606 kvfree(slot->arch.rmap[i]);
7607 slot->arch.rmap[i] = NULL;
7611 kvfree(slot->arch.lpage_info[i - 1]);
7612 slot->arch.lpage_info[i - 1] = NULL;
7617 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7620 * memslots->generation has been incremented.
7621 * mmio generation may have reached its maximum value.
7623 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7626 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7627 struct kvm_memory_slot *memslot,
7628 const struct kvm_userspace_memory_region *mem,
7629 enum kvm_mr_change change)
7632 * Only private memory slots need to be mapped here since
7633 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7635 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7636 unsigned long userspace_addr;
7639 * MAP_SHARED to prevent internal slot pages from being moved
7642 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7643 PROT_READ | PROT_WRITE,
7644 MAP_SHARED | MAP_ANONYMOUS, 0);
7646 if (IS_ERR((void *)userspace_addr))
7647 return PTR_ERR((void *)userspace_addr);
7649 memslot->userspace_addr = userspace_addr;
7655 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7656 struct kvm_memory_slot *new)
7658 /* Still write protect RO slot */
7659 if (new->flags & KVM_MEM_READONLY) {
7660 kvm_mmu_slot_remove_write_access(kvm, new);
7665 * Call kvm_x86_ops dirty logging hooks when they are valid.
7667 * kvm_x86_ops->slot_disable_log_dirty is called when:
7669 * - KVM_MR_CREATE with dirty logging is disabled
7670 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7672 * The reason is, in case of PML, we need to set D-bit for any slots
7673 * with dirty logging disabled in order to eliminate unnecessary GPA
7674 * logging in PML buffer (and potential PML buffer full VMEXT). This
7675 * guarantees leaving PML enabled during guest's lifetime won't have
7676 * any additonal overhead from PML when guest is running with dirty
7677 * logging disabled for memory slots.
7679 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7680 * to dirty logging mode.
7682 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7684 * In case of write protect:
7686 * Write protect all pages for dirty logging.
7688 * All the sptes including the large sptes which point to this
7689 * slot are set to readonly. We can not create any new large
7690 * spte on this slot until the end of the logging.
7692 * See the comments in fast_page_fault().
7694 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7695 if (kvm_x86_ops->slot_enable_log_dirty)
7696 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7698 kvm_mmu_slot_remove_write_access(kvm, new);
7700 if (kvm_x86_ops->slot_disable_log_dirty)
7701 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7705 void kvm_arch_commit_memory_region(struct kvm *kvm,
7706 const struct kvm_userspace_memory_region *mem,
7707 const struct kvm_memory_slot *old,
7708 const struct kvm_memory_slot *new,
7709 enum kvm_mr_change change)
7711 int nr_mmu_pages = 0;
7713 if (change == KVM_MR_DELETE && old->id >= KVM_USER_MEM_SLOTS) {
7716 ret = vm_munmap(old->userspace_addr,
7717 old->npages * PAGE_SIZE);
7720 "kvm_vm_ioctl_set_memory_region: "
7721 "failed to munmap memory\n");
7724 if (!kvm->arch.n_requested_mmu_pages)
7725 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7728 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7731 * Dirty logging tracks sptes in 4k granularity, meaning that large
7732 * sptes have to be split. If live migration is successful, the guest
7733 * in the source machine will be destroyed and large sptes will be
7734 * created in the destination. However, if the guest continues to run
7735 * in the source machine (for example if live migration fails), small
7736 * sptes will remain around and cause bad performance.
7738 * Scan sptes if dirty logging has been stopped, dropping those
7739 * which can be collapsed into a single large-page spte. Later
7740 * page faults will create the large-page sptes.
7742 if ((change != KVM_MR_DELETE) &&
7743 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7744 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7745 kvm_mmu_zap_collapsible_sptes(kvm, new);
7748 * Set up write protection and/or dirty logging for the new slot.
7750 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7751 * been zapped so no dirty logging staff is needed for old slot. For
7752 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7753 * new and it's also covered when dealing with the new slot.
7755 * FIXME: const-ify all uses of struct kvm_memory_slot.
7757 if (change != KVM_MR_DELETE)
7758 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
7761 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7763 kvm_mmu_invalidate_zap_all_pages(kvm);
7766 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7767 struct kvm_memory_slot *slot)
7769 kvm_mmu_invalidate_zap_all_pages(kvm);
7772 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7774 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7775 kvm_x86_ops->check_nested_events(vcpu, false);
7777 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7778 !vcpu->arch.apf.halted)
7779 || !list_empty_careful(&vcpu->async_pf.done)
7780 || kvm_apic_has_events(vcpu)
7781 || vcpu->arch.pv.pv_unhalted
7782 || atomic_read(&vcpu->arch.nmi_queued) ||
7783 (kvm_arch_interrupt_allowed(vcpu) &&
7784 kvm_cpu_has_interrupt(vcpu));
7787 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7789 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7792 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7794 return kvm_x86_ops->interrupt_allowed(vcpu);
7797 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
7799 if (is_64_bit_mode(vcpu))
7800 return kvm_rip_read(vcpu);
7801 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
7802 kvm_rip_read(vcpu));
7804 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
7806 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7808 return kvm_get_linear_rip(vcpu) == linear_rip;
7810 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7812 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7814 unsigned long rflags;
7816 rflags = kvm_x86_ops->get_rflags(vcpu);
7817 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7818 rflags &= ~X86_EFLAGS_TF;
7821 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7823 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7825 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7826 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7827 rflags |= X86_EFLAGS_TF;
7828 kvm_x86_ops->set_rflags(vcpu, rflags);
7831 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7833 __kvm_set_rflags(vcpu, rflags);
7834 kvm_make_request(KVM_REQ_EVENT, vcpu);
7836 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7838 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7842 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7846 r = kvm_mmu_reload(vcpu);
7850 if (!vcpu->arch.mmu.direct_map &&
7851 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7854 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7857 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7859 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7862 static inline u32 kvm_async_pf_next_probe(u32 key)
7864 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7867 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7869 u32 key = kvm_async_pf_hash_fn(gfn);
7871 while (vcpu->arch.apf.gfns[key] != ~0)
7872 key = kvm_async_pf_next_probe(key);
7874 vcpu->arch.apf.gfns[key] = gfn;
7877 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7880 u32 key = kvm_async_pf_hash_fn(gfn);
7882 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7883 (vcpu->arch.apf.gfns[key] != gfn &&
7884 vcpu->arch.apf.gfns[key] != ~0); i++)
7885 key = kvm_async_pf_next_probe(key);
7890 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7892 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7895 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7899 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7901 vcpu->arch.apf.gfns[i] = ~0;
7903 j = kvm_async_pf_next_probe(j);
7904 if (vcpu->arch.apf.gfns[j] == ~0)
7906 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7908 * k lies cyclically in ]i,j]
7910 * |....j i.k.| or |.k..j i...|
7912 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7913 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7918 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7921 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7925 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7926 struct kvm_async_pf *work)
7928 struct x86_exception fault;
7930 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7931 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7933 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7934 (vcpu->arch.apf.send_user_only &&
7935 kvm_x86_ops->get_cpl(vcpu) == 0))
7936 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7937 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7938 fault.vector = PF_VECTOR;
7939 fault.error_code_valid = true;
7940 fault.error_code = 0;
7941 fault.nested_page_fault = false;
7942 fault.address = work->arch.token;
7943 kvm_inject_page_fault(vcpu, &fault);
7947 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7948 struct kvm_async_pf *work)
7950 struct x86_exception fault;
7952 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7953 if (work->wakeup_all)
7954 work->arch.token = ~0; /* broadcast wakeup */
7956 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7958 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7959 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7960 fault.vector = PF_VECTOR;
7961 fault.error_code_valid = true;
7962 fault.error_code = 0;
7963 fault.nested_page_fault = false;
7964 fault.address = work->arch.token;
7965 kvm_inject_page_fault(vcpu, &fault);
7967 vcpu->arch.apf.halted = false;
7968 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7971 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7973 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7976 return !kvm_event_needs_reinjection(vcpu) &&
7977 kvm_x86_ops->interrupt_allowed(vcpu);
7980 void kvm_arch_start_assignment(struct kvm *kvm)
7982 atomic_inc(&kvm->arch.assigned_device_count);
7984 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
7986 void kvm_arch_end_assignment(struct kvm *kvm)
7988 atomic_dec(&kvm->arch.assigned_device_count);
7990 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
7992 bool kvm_arch_has_assigned_device(struct kvm *kvm)
7994 return atomic_read(&kvm->arch.assigned_device_count);
7996 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
7998 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8000 atomic_inc(&kvm->arch.noncoherent_dma_count);
8002 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8004 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8006 atomic_dec(&kvm->arch.noncoherent_dma_count);
8008 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8010 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8012 return atomic_read(&kvm->arch.noncoherent_dma_count);
8014 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8016 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8017 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8018 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8019 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8020 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8021 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8022 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8023 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8024 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8025 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8026 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8027 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8028 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8029 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8030 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);