2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
30 #include "assigned-dev.h"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/module.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <trace/events/kvm.h>
58 #define CREATE_TRACE_POINTS
61 #include <asm/debugreg.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
71 #define MAX_IO_MSRS 256
72 #define KVM_MAX_MCE_BANKS 32
73 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
75 #define emul_to_vcpu(ctxt) \
76 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
79 * - enable syscall per default because its emulated by KVM
80 * - enable LME and LMA per default on 64 bit KVM
84 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
86 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
89 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
90 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
92 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
93 static void process_nmi(struct kvm_vcpu *vcpu);
94 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
96 struct kvm_x86_ops *kvm_x86_ops;
97 EXPORT_SYMBOL_GPL(kvm_x86_ops);
99 static bool ignore_msrs = 0;
100 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
102 unsigned int min_timer_period_us = 500;
103 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
105 static bool __read_mostly kvmclock_periodic_sync = true;
106 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
108 bool kvm_has_tsc_control;
109 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
110 u32 kvm_max_guest_tsc_khz;
111 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
113 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
114 static u32 tsc_tolerance_ppm = 250;
115 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
117 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
118 unsigned int lapic_timer_advance_ns = 0;
119 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
121 static bool backwards_tsc_observed = false;
123 #define KVM_NR_SHARED_MSRS 16
125 struct kvm_shared_msrs_global {
127 u32 msrs[KVM_NR_SHARED_MSRS];
130 struct kvm_shared_msrs {
131 struct user_return_notifier urn;
133 struct kvm_shared_msr_values {
136 } values[KVM_NR_SHARED_MSRS];
139 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
140 static struct kvm_shared_msrs __percpu *shared_msrs;
142 struct kvm_stats_debugfs_item debugfs_entries[] = {
143 { "pf_fixed", VCPU_STAT(pf_fixed) },
144 { "pf_guest", VCPU_STAT(pf_guest) },
145 { "tlb_flush", VCPU_STAT(tlb_flush) },
146 { "invlpg", VCPU_STAT(invlpg) },
147 { "exits", VCPU_STAT(exits) },
148 { "io_exits", VCPU_STAT(io_exits) },
149 { "mmio_exits", VCPU_STAT(mmio_exits) },
150 { "signal_exits", VCPU_STAT(signal_exits) },
151 { "irq_window", VCPU_STAT(irq_window_exits) },
152 { "nmi_window", VCPU_STAT(nmi_window_exits) },
153 { "halt_exits", VCPU_STAT(halt_exits) },
154 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
155 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
156 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
157 { "hypercalls", VCPU_STAT(hypercalls) },
158 { "request_irq", VCPU_STAT(request_irq_exits) },
159 { "irq_exits", VCPU_STAT(irq_exits) },
160 { "host_state_reload", VCPU_STAT(host_state_reload) },
161 { "efer_reload", VCPU_STAT(efer_reload) },
162 { "fpu_reload", VCPU_STAT(fpu_reload) },
163 { "insn_emulation", VCPU_STAT(insn_emulation) },
164 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
165 { "irq_injections", VCPU_STAT(irq_injections) },
166 { "nmi_injections", VCPU_STAT(nmi_injections) },
167 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
168 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
169 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
170 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
171 { "mmu_flooded", VM_STAT(mmu_flooded) },
172 { "mmu_recycled", VM_STAT(mmu_recycled) },
173 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
174 { "mmu_unsync", VM_STAT(mmu_unsync) },
175 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
176 { "largepages", VM_STAT(lpages) },
180 u64 __read_mostly host_xcr0;
182 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
184 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
187 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
188 vcpu->arch.apf.gfns[i] = ~0;
191 static void kvm_on_user_return(struct user_return_notifier *urn)
194 struct kvm_shared_msrs *locals
195 = container_of(urn, struct kvm_shared_msrs, urn);
196 struct kvm_shared_msr_values *values;
198 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
199 values = &locals->values[slot];
200 if (values->host != values->curr) {
201 wrmsrl(shared_msrs_global.msrs[slot], values->host);
202 values->curr = values->host;
205 locals->registered = false;
206 user_return_notifier_unregister(urn);
209 static void shared_msr_update(unsigned slot, u32 msr)
212 unsigned int cpu = smp_processor_id();
213 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
215 /* only read, and nobody should modify it at this time,
216 * so don't need lock */
217 if (slot >= shared_msrs_global.nr) {
218 printk(KERN_ERR "kvm: invalid MSR slot!");
221 rdmsrl_safe(msr, &value);
222 smsr->values[slot].host = value;
223 smsr->values[slot].curr = value;
226 void kvm_define_shared_msr(unsigned slot, u32 msr)
228 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
229 shared_msrs_global.msrs[slot] = msr;
230 if (slot >= shared_msrs_global.nr)
231 shared_msrs_global.nr = slot + 1;
233 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
235 static void kvm_shared_msr_cpu_online(void)
239 for (i = 0; i < shared_msrs_global.nr; ++i)
240 shared_msr_update(i, shared_msrs_global.msrs[i]);
243 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
245 unsigned int cpu = smp_processor_id();
246 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
249 if (((value ^ smsr->values[slot].curr) & mask) == 0)
251 smsr->values[slot].curr = value;
252 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
256 if (!smsr->registered) {
257 smsr->urn.on_user_return = kvm_on_user_return;
258 user_return_notifier_register(&smsr->urn);
259 smsr->registered = true;
263 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
265 static void drop_user_return_notifiers(void)
267 unsigned int cpu = smp_processor_id();
268 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
270 if (smsr->registered)
271 kvm_on_user_return(&smsr->urn);
274 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
276 return vcpu->arch.apic_base;
278 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
280 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
282 u64 old_state = vcpu->arch.apic_base &
283 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
284 u64 new_state = msr_info->data &
285 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
286 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
287 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
289 if (!msr_info->host_initiated &&
290 ((msr_info->data & reserved_bits) != 0 ||
291 new_state == X2APIC_ENABLE ||
292 (new_state == MSR_IA32_APICBASE_ENABLE &&
293 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
294 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
298 kvm_lapic_set_base(vcpu, msr_info->data);
301 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
303 asmlinkage __visible void kvm_spurious_fault(void)
305 /* Fault while not rebooting. We want the trace. */
308 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
310 #define EXCPT_BENIGN 0
311 #define EXCPT_CONTRIBUTORY 1
314 static int exception_class(int vector)
324 return EXCPT_CONTRIBUTORY;
331 #define EXCPT_FAULT 0
333 #define EXCPT_ABORT 2
334 #define EXCPT_INTERRUPT 3
336 static int exception_type(int vector)
340 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
341 return EXCPT_INTERRUPT;
345 /* #DB is trap, as instruction watchpoints are handled elsewhere */
346 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
349 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
352 /* Reserved exceptions will result in fault */
356 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
357 unsigned nr, bool has_error, u32 error_code,
363 kvm_make_request(KVM_REQ_EVENT, vcpu);
365 if (!vcpu->arch.exception.pending) {
367 if (has_error && !is_protmode(vcpu))
369 vcpu->arch.exception.pending = true;
370 vcpu->arch.exception.has_error_code = has_error;
371 vcpu->arch.exception.nr = nr;
372 vcpu->arch.exception.error_code = error_code;
373 vcpu->arch.exception.reinject = reinject;
377 /* to check exception */
378 prev_nr = vcpu->arch.exception.nr;
379 if (prev_nr == DF_VECTOR) {
380 /* triple fault -> shutdown */
381 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
384 class1 = exception_class(prev_nr);
385 class2 = exception_class(nr);
386 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
387 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
388 /* generate double fault per SDM Table 5-5 */
389 vcpu->arch.exception.pending = true;
390 vcpu->arch.exception.has_error_code = true;
391 vcpu->arch.exception.nr = DF_VECTOR;
392 vcpu->arch.exception.error_code = 0;
394 /* replace previous exception with a new one in a hope
395 that instruction re-execution will regenerate lost
400 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
402 kvm_multiple_exception(vcpu, nr, false, 0, false);
404 EXPORT_SYMBOL_GPL(kvm_queue_exception);
406 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
408 kvm_multiple_exception(vcpu, nr, false, 0, true);
410 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
412 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
415 kvm_inject_gp(vcpu, 0);
417 kvm_x86_ops->skip_emulated_instruction(vcpu);
419 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
421 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
423 ++vcpu->stat.pf_guest;
424 vcpu->arch.cr2 = fault->address;
425 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
427 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
429 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
431 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
432 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
434 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
436 return fault->nested_page_fault;
439 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
441 atomic_inc(&vcpu->arch.nmi_queued);
442 kvm_make_request(KVM_REQ_NMI, vcpu);
444 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
446 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
448 kvm_multiple_exception(vcpu, nr, true, error_code, false);
450 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
452 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
454 kvm_multiple_exception(vcpu, nr, true, error_code, true);
456 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
459 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
460 * a #GP and return false.
462 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
464 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
466 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
469 EXPORT_SYMBOL_GPL(kvm_require_cpl);
471 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
473 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
476 kvm_queue_exception(vcpu, UD_VECTOR);
479 EXPORT_SYMBOL_GPL(kvm_require_dr);
482 * This function will be used to read from the physical memory of the currently
483 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
484 * can read from guest physical or from the guest's guest physical memory.
486 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
487 gfn_t ngfn, void *data, int offset, int len,
490 struct x86_exception exception;
494 ngpa = gfn_to_gpa(ngfn);
495 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
496 if (real_gfn == UNMAPPED_GVA)
499 real_gfn = gpa_to_gfn(real_gfn);
501 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
503 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
505 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
506 void *data, int offset, int len, u32 access)
508 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
509 data, offset, len, access);
513 * Load the pae pdptrs. Return true is they are all valid.
515 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
517 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
518 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
521 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
523 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
524 offset * sizeof(u64), sizeof(pdpte),
525 PFERR_USER_MASK|PFERR_WRITE_MASK);
530 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
531 if (is_present_gpte(pdpte[i]) &&
533 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
540 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
541 __set_bit(VCPU_EXREG_PDPTR,
542 (unsigned long *)&vcpu->arch.regs_avail);
543 __set_bit(VCPU_EXREG_PDPTR,
544 (unsigned long *)&vcpu->arch.regs_dirty);
549 EXPORT_SYMBOL_GPL(load_pdptrs);
551 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
553 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
559 if (is_long_mode(vcpu) || !is_pae(vcpu))
562 if (!test_bit(VCPU_EXREG_PDPTR,
563 (unsigned long *)&vcpu->arch.regs_avail))
566 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
567 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
568 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
569 PFERR_USER_MASK | PFERR_WRITE_MASK);
572 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
578 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
580 unsigned long old_cr0 = kvm_read_cr0(vcpu);
581 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
586 if (cr0 & 0xffffffff00000000UL)
590 cr0 &= ~CR0_RESERVED_BITS;
592 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
595 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
598 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
600 if ((vcpu->arch.efer & EFER_LME)) {
605 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
610 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
615 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
618 kvm_x86_ops->set_cr0(vcpu, cr0);
620 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
621 kvm_clear_async_pf_completion_queue(vcpu);
622 kvm_async_pf_hash_reset(vcpu);
625 if ((cr0 ^ old_cr0) & update_bits)
626 kvm_mmu_reset_context(vcpu);
628 if ((cr0 ^ old_cr0) & X86_CR0_CD)
629 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
633 EXPORT_SYMBOL_GPL(kvm_set_cr0);
635 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
637 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
639 EXPORT_SYMBOL_GPL(kvm_lmsw);
641 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
643 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
644 !vcpu->guest_xcr0_loaded) {
645 /* kvm_set_xcr() also depends on this */
646 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
647 vcpu->guest_xcr0_loaded = 1;
651 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
653 if (vcpu->guest_xcr0_loaded) {
654 if (vcpu->arch.xcr0 != host_xcr0)
655 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
656 vcpu->guest_xcr0_loaded = 0;
660 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
663 u64 old_xcr0 = vcpu->arch.xcr0;
666 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
667 if (index != XCR_XFEATURE_ENABLED_MASK)
669 if (!(xcr0 & XSTATE_FP))
671 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
675 * Do not allow the guest to set bits that we do not support
676 * saving. However, xcr0 bit 0 is always set, even if the
677 * emulated CPU does not support XSAVE (see fx_init).
679 valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
680 if (xcr0 & ~valid_bits)
683 if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
686 if (xcr0 & XSTATE_AVX512) {
687 if (!(xcr0 & XSTATE_YMM))
689 if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512)
692 kvm_put_guest_xcr0(vcpu);
693 vcpu->arch.xcr0 = xcr0;
695 if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
696 kvm_update_cpuid(vcpu);
700 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
702 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
703 __kvm_set_xcr(vcpu, index, xcr)) {
704 kvm_inject_gp(vcpu, 0);
709 EXPORT_SYMBOL_GPL(kvm_set_xcr);
711 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
713 unsigned long old_cr4 = kvm_read_cr4(vcpu);
714 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
715 X86_CR4_SMEP | X86_CR4_SMAP;
717 if (cr4 & CR4_RESERVED_BITS)
720 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
723 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
726 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
729 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
732 if (is_long_mode(vcpu)) {
733 if (!(cr4 & X86_CR4_PAE))
735 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
736 && ((cr4 ^ old_cr4) & pdptr_bits)
737 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
741 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
742 if (!guest_cpuid_has_pcid(vcpu))
745 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
746 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
750 if (kvm_x86_ops->set_cr4(vcpu, cr4))
753 if (((cr4 ^ old_cr4) & pdptr_bits) ||
754 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
755 kvm_mmu_reset_context(vcpu);
757 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
758 kvm_update_cpuid(vcpu);
762 EXPORT_SYMBOL_GPL(kvm_set_cr4);
764 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
767 cr3 &= ~CR3_PCID_INVD;
770 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
771 kvm_mmu_sync_roots(vcpu);
772 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
776 if (is_long_mode(vcpu)) {
777 if (cr3 & CR3_L_MODE_RESERVED_BITS)
779 } else if (is_pae(vcpu) && is_paging(vcpu) &&
780 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
783 vcpu->arch.cr3 = cr3;
784 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
785 kvm_mmu_new_cr3(vcpu);
788 EXPORT_SYMBOL_GPL(kvm_set_cr3);
790 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
792 if (cr8 & CR8_RESERVED_BITS)
794 if (lapic_in_kernel(vcpu))
795 kvm_lapic_set_tpr(vcpu, cr8);
797 vcpu->arch.cr8 = cr8;
800 EXPORT_SYMBOL_GPL(kvm_set_cr8);
802 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
804 if (lapic_in_kernel(vcpu))
805 return kvm_lapic_get_cr8(vcpu);
807 return vcpu->arch.cr8;
809 EXPORT_SYMBOL_GPL(kvm_get_cr8);
811 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
815 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
816 for (i = 0; i < KVM_NR_DB_REGS; i++)
817 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
818 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
822 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
824 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
825 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
828 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
832 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
833 dr7 = vcpu->arch.guest_debug_dr7;
835 dr7 = vcpu->arch.dr7;
836 kvm_x86_ops->set_dr7(vcpu, dr7);
837 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
838 if (dr7 & DR7_BP_EN_MASK)
839 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
842 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
844 u64 fixed = DR6_FIXED_1;
846 if (!guest_cpuid_has_rtm(vcpu))
851 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
855 vcpu->arch.db[dr] = val;
856 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
857 vcpu->arch.eff_db[dr] = val;
862 if (val & 0xffffffff00000000ULL)
864 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
865 kvm_update_dr6(vcpu);
870 if (val & 0xffffffff00000000ULL)
872 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
873 kvm_update_dr7(vcpu);
880 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
882 if (__kvm_set_dr(vcpu, dr, val)) {
883 kvm_inject_gp(vcpu, 0);
888 EXPORT_SYMBOL_GPL(kvm_set_dr);
890 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
894 *val = vcpu->arch.db[dr];
899 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
900 *val = vcpu->arch.dr6;
902 *val = kvm_x86_ops->get_dr6(vcpu);
907 *val = vcpu->arch.dr7;
912 EXPORT_SYMBOL_GPL(kvm_get_dr);
914 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
916 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
920 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
923 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
924 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
927 EXPORT_SYMBOL_GPL(kvm_rdpmc);
930 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
931 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
933 * This list is modified at module load time to reflect the
934 * capabilities of the host cpu. This capabilities test skips MSRs that are
935 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
936 * may depend on host virtualization features rather than host cpu features.
939 static u32 msrs_to_save[] = {
940 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
943 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
945 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
946 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
949 static unsigned num_msrs_to_save;
951 static u32 emulated_msrs[] = {
952 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
953 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
954 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
955 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
956 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
957 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
960 HV_X64_MSR_VP_RUNTIME,
961 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
965 MSR_IA32_TSCDEADLINE,
966 MSR_IA32_MISC_ENABLE,
972 static unsigned num_emulated_msrs;
974 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
976 if (efer & efer_reserved_bits)
979 if (efer & EFER_FFXSR) {
980 struct kvm_cpuid_entry2 *feat;
982 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
983 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
987 if (efer & EFER_SVME) {
988 struct kvm_cpuid_entry2 *feat;
990 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
991 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
997 EXPORT_SYMBOL_GPL(kvm_valid_efer);
999 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1001 u64 old_efer = vcpu->arch.efer;
1003 if (!kvm_valid_efer(vcpu, efer))
1007 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1011 efer |= vcpu->arch.efer & EFER_LMA;
1013 kvm_x86_ops->set_efer(vcpu, efer);
1015 /* Update reserved bits */
1016 if ((efer ^ old_efer) & EFER_NX)
1017 kvm_mmu_reset_context(vcpu);
1022 void kvm_enable_efer_bits(u64 mask)
1024 efer_reserved_bits &= ~mask;
1026 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1029 * Writes msr value into into the appropriate "register".
1030 * Returns 0 on success, non-0 otherwise.
1031 * Assumes vcpu_load() was already called.
1033 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1035 switch (msr->index) {
1038 case MSR_KERNEL_GS_BASE:
1041 if (is_noncanonical_address(msr->data))
1044 case MSR_IA32_SYSENTER_EIP:
1045 case MSR_IA32_SYSENTER_ESP:
1047 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1048 * non-canonical address is written on Intel but not on
1049 * AMD (which ignores the top 32-bits, because it does
1050 * not implement 64-bit SYSENTER).
1052 * 64-bit code should hence be able to write a non-canonical
1053 * value on AMD. Making the address canonical ensures that
1054 * vmentry does not fail on Intel after writing a non-canonical
1055 * value, and that something deterministic happens if the guest
1056 * invokes 64-bit SYSENTER.
1058 msr->data = get_canonical(msr->data);
1060 return kvm_x86_ops->set_msr(vcpu, msr);
1062 EXPORT_SYMBOL_GPL(kvm_set_msr);
1065 * Adapt set_msr() to msr_io()'s calling convention
1067 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1069 struct msr_data msr;
1073 msr.host_initiated = true;
1074 r = kvm_get_msr(vcpu, &msr);
1082 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1084 struct msr_data msr;
1088 msr.host_initiated = true;
1089 return kvm_set_msr(vcpu, &msr);
1092 #ifdef CONFIG_X86_64
1093 struct pvclock_gtod_data {
1096 struct { /* extract of a clocksource struct */
1108 static struct pvclock_gtod_data pvclock_gtod_data;
1110 static void update_pvclock_gtod(struct timekeeper *tk)
1112 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1115 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1117 write_seqcount_begin(&vdata->seq);
1119 /* copy pvclock gtod data */
1120 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1121 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1122 vdata->clock.mask = tk->tkr_mono.mask;
1123 vdata->clock.mult = tk->tkr_mono.mult;
1124 vdata->clock.shift = tk->tkr_mono.shift;
1126 vdata->boot_ns = boot_ns;
1127 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1129 write_seqcount_end(&vdata->seq);
1133 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1136 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1137 * vcpu_enter_guest. This function is only called from
1138 * the physical CPU that is running vcpu.
1140 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1143 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1147 struct pvclock_wall_clock wc;
1148 struct timespec boot;
1153 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1158 ++version; /* first time write, random junk */
1162 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1165 * The guest calculates current wall clock time by adding
1166 * system time (updated by kvm_guest_time_update below) to the
1167 * wall clock specified here. guest system time equals host
1168 * system time for us, thus we must fill in host boot time here.
1172 if (kvm->arch.kvmclock_offset) {
1173 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1174 boot = timespec_sub(boot, ts);
1176 wc.sec = boot.tv_sec;
1177 wc.nsec = boot.tv_nsec;
1178 wc.version = version;
1180 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1183 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1186 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1188 uint32_t quotient, remainder;
1190 /* Don't try to replace with do_div(), this one calculates
1191 * "(dividend << 32) / divisor" */
1193 : "=a" (quotient), "=d" (remainder)
1194 : "0" (0), "1" (dividend), "r" (divisor) );
1198 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1199 s8 *pshift, u32 *pmultiplier)
1206 tps64 = base_khz * 1000LL;
1207 scaled64 = scaled_khz * 1000LL;
1208 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1213 tps32 = (uint32_t)tps64;
1214 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1215 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1223 *pmultiplier = div_frac(scaled64, tps32);
1225 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1226 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1229 #ifdef CONFIG_X86_64
1230 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1233 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1234 static unsigned long max_tsc_khz;
1236 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1238 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1239 vcpu->arch.virtual_tsc_shift);
1242 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1244 u64 v = (u64)khz * (1000000 + ppm);
1249 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1251 u32 thresh_lo, thresh_hi;
1252 int use_scaling = 0;
1254 /* tsc_khz can be zero if TSC calibration fails */
1255 if (this_tsc_khz == 0)
1258 /* Compute a scale to convert nanoseconds in TSC cycles */
1259 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1260 &vcpu->arch.virtual_tsc_shift,
1261 &vcpu->arch.virtual_tsc_mult);
1262 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1265 * Compute the variation in TSC rate which is acceptable
1266 * within the range of tolerance and decide if the
1267 * rate being applied is within that bounds of the hardware
1268 * rate. If so, no scaling or compensation need be done.
1270 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1271 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1272 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1273 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1276 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1279 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1281 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1282 vcpu->arch.virtual_tsc_mult,
1283 vcpu->arch.virtual_tsc_shift);
1284 tsc += vcpu->arch.this_tsc_write;
1288 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1290 #ifdef CONFIG_X86_64
1292 struct kvm_arch *ka = &vcpu->kvm->arch;
1293 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1295 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1296 atomic_read(&vcpu->kvm->online_vcpus));
1299 * Once the masterclock is enabled, always perform request in
1300 * order to update it.
1302 * In order to enable masterclock, the host clocksource must be TSC
1303 * and the vcpus need to have matched TSCs. When that happens,
1304 * perform request to enable masterclock.
1306 if (ka->use_master_clock ||
1307 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1308 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1310 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1311 atomic_read(&vcpu->kvm->online_vcpus),
1312 ka->use_master_clock, gtod->clock.vclock_mode);
1316 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1318 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1319 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1322 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1324 struct kvm *kvm = vcpu->kvm;
1325 u64 offset, ns, elapsed;
1326 unsigned long flags;
1329 bool already_matched;
1330 u64 data = msr->data;
1332 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1333 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1334 ns = get_kernel_ns();
1335 elapsed = ns - kvm->arch.last_tsc_nsec;
1337 if (vcpu->arch.virtual_tsc_khz) {
1340 /* n.b - signed multiplication and division required */
1341 usdiff = data - kvm->arch.last_tsc_write;
1342 #ifdef CONFIG_X86_64
1343 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1345 /* do_div() only does unsigned */
1346 asm("1: idivl %[divisor]\n"
1347 "2: xor %%edx, %%edx\n"
1348 " movl $0, %[faulted]\n"
1350 ".section .fixup,\"ax\"\n"
1351 "4: movl $1, %[faulted]\n"
1355 _ASM_EXTABLE(1b, 4b)
1357 : "=A"(usdiff), [faulted] "=r" (faulted)
1358 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1361 do_div(elapsed, 1000);
1366 /* idivl overflow => difference is larger than USEC_PER_SEC */
1368 usdiff = USEC_PER_SEC;
1370 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1373 * Special case: TSC write with a small delta (1 second) of virtual
1374 * cycle time against real time is interpreted as an attempt to
1375 * synchronize the CPU.
1377 * For a reliable TSC, we can match TSC offsets, and for an unstable
1378 * TSC, we add elapsed time in this computation. We could let the
1379 * compensation code attempt to catch up if we fall behind, but
1380 * it's better to try to match offsets from the beginning.
1382 if (usdiff < USEC_PER_SEC &&
1383 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1384 if (!check_tsc_unstable()) {
1385 offset = kvm->arch.cur_tsc_offset;
1386 pr_debug("kvm: matched tsc offset for %llu\n", data);
1388 u64 delta = nsec_to_cycles(vcpu, elapsed);
1390 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1391 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1394 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1397 * We split periods of matched TSC writes into generations.
1398 * For each generation, we track the original measured
1399 * nanosecond time, offset, and write, so if TSCs are in
1400 * sync, we can match exact offset, and if not, we can match
1401 * exact software computation in compute_guest_tsc()
1403 * These values are tracked in kvm->arch.cur_xxx variables.
1405 kvm->arch.cur_tsc_generation++;
1406 kvm->arch.cur_tsc_nsec = ns;
1407 kvm->arch.cur_tsc_write = data;
1408 kvm->arch.cur_tsc_offset = offset;
1410 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1411 kvm->arch.cur_tsc_generation, data);
1415 * We also track th most recent recorded KHZ, write and time to
1416 * allow the matching interval to be extended at each write.
1418 kvm->arch.last_tsc_nsec = ns;
1419 kvm->arch.last_tsc_write = data;
1420 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1422 vcpu->arch.last_guest_tsc = data;
1424 /* Keep track of which generation this VCPU has synchronized to */
1425 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1426 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1427 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1429 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1430 update_ia32_tsc_adjust_msr(vcpu, offset);
1431 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1432 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1434 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1436 kvm->arch.nr_vcpus_matched_tsc = 0;
1437 } else if (!already_matched) {
1438 kvm->arch.nr_vcpus_matched_tsc++;
1441 kvm_track_tsc_matching(vcpu);
1442 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1445 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1447 #ifdef CONFIG_X86_64
1449 static cycle_t read_tsc(void)
1451 cycle_t ret = (cycle_t)rdtsc_ordered();
1452 u64 last = pvclock_gtod_data.clock.cycle_last;
1454 if (likely(ret >= last))
1458 * GCC likes to generate cmov here, but this branch is extremely
1459 * predictable (it's just a funciton of time and the likely is
1460 * very likely) and there's a data dependence, so force GCC
1461 * to generate a branch instead. I don't barrier() because
1462 * we don't actually need a barrier, and if this function
1463 * ever gets inlined it will generate worse code.
1469 static inline u64 vgettsc(cycle_t *cycle_now)
1472 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1474 *cycle_now = read_tsc();
1476 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1477 return v * gtod->clock.mult;
1480 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1482 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1488 seq = read_seqcount_begin(>od->seq);
1489 mode = gtod->clock.vclock_mode;
1490 ns = gtod->nsec_base;
1491 ns += vgettsc(cycle_now);
1492 ns >>= gtod->clock.shift;
1493 ns += gtod->boot_ns;
1494 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1500 /* returns true if host is using tsc clocksource */
1501 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1503 /* checked again under seqlock below */
1504 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1507 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1513 * Assuming a stable TSC across physical CPUS, and a stable TSC
1514 * across virtual CPUs, the following condition is possible.
1515 * Each numbered line represents an event visible to both
1516 * CPUs at the next numbered event.
1518 * "timespecX" represents host monotonic time. "tscX" represents
1521 * VCPU0 on CPU0 | VCPU1 on CPU1
1523 * 1. read timespec0,tsc0
1524 * 2. | timespec1 = timespec0 + N
1526 * 3. transition to guest | transition to guest
1527 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1528 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1529 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1531 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1534 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1536 * - 0 < N - M => M < N
1538 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1539 * always the case (the difference between two distinct xtime instances
1540 * might be smaller then the difference between corresponding TSC reads,
1541 * when updating guest vcpus pvclock areas).
1543 * To avoid that problem, do not allow visibility of distinct
1544 * system_timestamp/tsc_timestamp values simultaneously: use a master
1545 * copy of host monotonic time values. Update that master copy
1548 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1552 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1554 #ifdef CONFIG_X86_64
1555 struct kvm_arch *ka = &kvm->arch;
1557 bool host_tsc_clocksource, vcpus_matched;
1559 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1560 atomic_read(&kvm->online_vcpus));
1563 * If the host uses TSC clock, then passthrough TSC as stable
1566 host_tsc_clocksource = kvm_get_time_and_clockread(
1567 &ka->master_kernel_ns,
1568 &ka->master_cycle_now);
1570 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1571 && !backwards_tsc_observed
1572 && !ka->boot_vcpu_runs_old_kvmclock;
1574 if (ka->use_master_clock)
1575 atomic_set(&kvm_guest_has_master_clock, 1);
1577 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1578 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1583 static void kvm_gen_update_masterclock(struct kvm *kvm)
1585 #ifdef CONFIG_X86_64
1587 struct kvm_vcpu *vcpu;
1588 struct kvm_arch *ka = &kvm->arch;
1590 spin_lock(&ka->pvclock_gtod_sync_lock);
1591 kvm_make_mclock_inprogress_request(kvm);
1592 /* no guest entries from this point */
1593 pvclock_update_vm_gtod_copy(kvm);
1595 kvm_for_each_vcpu(i, vcpu, kvm)
1596 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1598 /* guest entries allowed */
1599 kvm_for_each_vcpu(i, vcpu, kvm)
1600 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1602 spin_unlock(&ka->pvclock_gtod_sync_lock);
1606 static int kvm_guest_time_update(struct kvm_vcpu *v)
1608 unsigned long flags, this_tsc_khz;
1609 struct kvm_vcpu_arch *vcpu = &v->arch;
1610 struct kvm_arch *ka = &v->kvm->arch;
1612 u64 tsc_timestamp, host_tsc;
1613 struct pvclock_vcpu_time_info guest_hv_clock;
1615 bool use_master_clock;
1621 * If the host uses TSC clock, then passthrough TSC as stable
1624 spin_lock(&ka->pvclock_gtod_sync_lock);
1625 use_master_clock = ka->use_master_clock;
1626 if (use_master_clock) {
1627 host_tsc = ka->master_cycle_now;
1628 kernel_ns = ka->master_kernel_ns;
1630 spin_unlock(&ka->pvclock_gtod_sync_lock);
1632 /* Keep irq disabled to prevent changes to the clock */
1633 local_irq_save(flags);
1634 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1635 if (unlikely(this_tsc_khz == 0)) {
1636 local_irq_restore(flags);
1637 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1640 if (!use_master_clock) {
1642 kernel_ns = get_kernel_ns();
1645 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1648 * We may have to catch up the TSC to match elapsed wall clock
1649 * time for two reasons, even if kvmclock is used.
1650 * 1) CPU could have been running below the maximum TSC rate
1651 * 2) Broken TSC compensation resets the base at each VCPU
1652 * entry to avoid unknown leaps of TSC even when running
1653 * again on the same CPU. This may cause apparent elapsed
1654 * time to disappear, and the guest to stand still or run
1657 if (vcpu->tsc_catchup) {
1658 u64 tsc = compute_guest_tsc(v, kernel_ns);
1659 if (tsc > tsc_timestamp) {
1660 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1661 tsc_timestamp = tsc;
1665 local_irq_restore(flags);
1667 if (!vcpu->pv_time_enabled)
1670 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1671 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1672 &vcpu->hv_clock.tsc_shift,
1673 &vcpu->hv_clock.tsc_to_system_mul);
1674 vcpu->hw_tsc_khz = this_tsc_khz;
1677 /* With all the info we got, fill in the values */
1678 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1679 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1680 vcpu->last_guest_tsc = tsc_timestamp;
1682 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1683 &guest_hv_clock, sizeof(guest_hv_clock))))
1686 /* This VCPU is paused, but it's legal for a guest to read another
1687 * VCPU's kvmclock, so we really have to follow the specification where
1688 * it says that version is odd if data is being modified, and even after
1691 * Version field updates must be kept separate. This is because
1692 * kvm_write_guest_cached might use a "rep movs" instruction, and
1693 * writes within a string instruction are weakly ordered. So there
1694 * are three writes overall.
1696 * As a small optimization, only write the version field in the first
1697 * and third write. The vcpu->pv_time cache is still valid, because the
1698 * version field is the first in the struct.
1700 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1702 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1703 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1705 sizeof(vcpu->hv_clock.version));
1709 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1710 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1712 if (vcpu->pvclock_set_guest_stopped_request) {
1713 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1714 vcpu->pvclock_set_guest_stopped_request = false;
1717 /* If the host uses TSC clocksource, then it is stable */
1718 if (use_master_clock)
1719 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1721 vcpu->hv_clock.flags = pvclock_flags;
1723 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1725 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1727 sizeof(vcpu->hv_clock));
1731 vcpu->hv_clock.version++;
1732 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1734 sizeof(vcpu->hv_clock.version));
1739 * kvmclock updates which are isolated to a given vcpu, such as
1740 * vcpu->cpu migration, should not allow system_timestamp from
1741 * the rest of the vcpus to remain static. Otherwise ntp frequency
1742 * correction applies to one vcpu's system_timestamp but not
1745 * So in those cases, request a kvmclock update for all vcpus.
1746 * We need to rate-limit these requests though, as they can
1747 * considerably slow guests that have a large number of vcpus.
1748 * The time for a remote vcpu to update its kvmclock is bound
1749 * by the delay we use to rate-limit the updates.
1752 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1754 static void kvmclock_update_fn(struct work_struct *work)
1757 struct delayed_work *dwork = to_delayed_work(work);
1758 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1759 kvmclock_update_work);
1760 struct kvm *kvm = container_of(ka, struct kvm, arch);
1761 struct kvm_vcpu *vcpu;
1763 kvm_for_each_vcpu(i, vcpu, kvm) {
1764 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1765 kvm_vcpu_kick(vcpu);
1769 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1771 struct kvm *kvm = v->kvm;
1773 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1774 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1775 KVMCLOCK_UPDATE_DELAY);
1778 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1780 static void kvmclock_sync_fn(struct work_struct *work)
1782 struct delayed_work *dwork = to_delayed_work(work);
1783 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1784 kvmclock_sync_work);
1785 struct kvm *kvm = container_of(ka, struct kvm, arch);
1787 if (!kvmclock_periodic_sync)
1790 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1791 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1792 KVMCLOCK_SYNC_PERIOD);
1795 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1797 u64 mcg_cap = vcpu->arch.mcg_cap;
1798 unsigned bank_num = mcg_cap & 0xff;
1801 case MSR_IA32_MCG_STATUS:
1802 vcpu->arch.mcg_status = data;
1804 case MSR_IA32_MCG_CTL:
1805 if (!(mcg_cap & MCG_CTL_P))
1807 if (data != 0 && data != ~(u64)0)
1809 vcpu->arch.mcg_ctl = data;
1812 if (msr >= MSR_IA32_MC0_CTL &&
1813 msr < MSR_IA32_MCx_CTL(bank_num)) {
1814 u32 offset = msr - MSR_IA32_MC0_CTL;
1815 /* only 0 or all 1s can be written to IA32_MCi_CTL
1816 * some Linux kernels though clear bit 10 in bank 4 to
1817 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1818 * this to avoid an uncatched #GP in the guest
1820 if ((offset & 0x3) == 0 &&
1821 data != 0 && (data | (1 << 10)) != ~(u64)0)
1823 vcpu->arch.mce_banks[offset] = data;
1831 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1833 struct kvm *kvm = vcpu->kvm;
1834 int lm = is_long_mode(vcpu);
1835 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1836 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1837 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1838 : kvm->arch.xen_hvm_config.blob_size_32;
1839 u32 page_num = data & ~PAGE_MASK;
1840 u64 page_addr = data & PAGE_MASK;
1845 if (page_num >= blob_size)
1848 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1853 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1862 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1864 gpa_t gpa = data & ~0x3f;
1866 /* Bits 2:5 are reserved, Should be zero */
1870 vcpu->arch.apf.msr_val = data;
1872 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1873 kvm_clear_async_pf_completion_queue(vcpu);
1874 kvm_async_pf_hash_reset(vcpu);
1878 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1882 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1883 kvm_async_pf_wakeup_all(vcpu);
1887 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1889 vcpu->arch.pv_time_enabled = false;
1892 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1896 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1899 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1900 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1901 vcpu->arch.st.accum_steal = delta;
1904 static void record_steal_time(struct kvm_vcpu *vcpu)
1906 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1909 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1910 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1913 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1914 vcpu->arch.st.steal.version += 2;
1915 vcpu->arch.st.accum_steal = 0;
1917 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1918 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1921 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1924 u32 msr = msr_info->index;
1925 u64 data = msr_info->data;
1928 case MSR_AMD64_NB_CFG:
1929 case MSR_IA32_UCODE_REV:
1930 case MSR_IA32_UCODE_WRITE:
1931 case MSR_VM_HSAVE_PA:
1932 case MSR_AMD64_PATCH_LOADER:
1933 case MSR_AMD64_BU_CFG2:
1937 return set_efer(vcpu, data);
1939 data &= ~(u64)0x40; /* ignore flush filter disable */
1940 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1941 data &= ~(u64)0x8; /* ignore TLB cache disable */
1942 data &= ~(u64)0x40000; /* ignore Mc status write enable */
1944 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1949 case MSR_FAM10H_MMIO_CONF_BASE:
1951 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1956 case MSR_IA32_DEBUGCTLMSR:
1958 /* We support the non-activated case already */
1960 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1961 /* Values other than LBR and BTF are vendor-specific,
1962 thus reserved and should throw a #GP */
1965 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1968 case 0x200 ... 0x2ff:
1969 return kvm_mtrr_set_msr(vcpu, msr, data);
1970 case MSR_IA32_APICBASE:
1971 return kvm_set_apic_base(vcpu, msr_info);
1972 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1973 return kvm_x2apic_msr_write(vcpu, msr, data);
1974 case MSR_IA32_TSCDEADLINE:
1975 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1977 case MSR_IA32_TSC_ADJUST:
1978 if (guest_cpuid_has_tsc_adjust(vcpu)) {
1979 if (!msr_info->host_initiated) {
1980 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
1981 adjust_tsc_offset_guest(vcpu, adj);
1983 vcpu->arch.ia32_tsc_adjust_msr = data;
1986 case MSR_IA32_MISC_ENABLE:
1987 vcpu->arch.ia32_misc_enable_msr = data;
1989 case MSR_IA32_SMBASE:
1990 if (!msr_info->host_initiated)
1992 vcpu->arch.smbase = data;
1994 case MSR_KVM_WALL_CLOCK_NEW:
1995 case MSR_KVM_WALL_CLOCK:
1996 vcpu->kvm->arch.wall_clock = data;
1997 kvm_write_wall_clock(vcpu->kvm, data);
1999 case MSR_KVM_SYSTEM_TIME_NEW:
2000 case MSR_KVM_SYSTEM_TIME: {
2002 struct kvm_arch *ka = &vcpu->kvm->arch;
2004 kvmclock_reset(vcpu);
2006 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2007 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2009 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2010 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2013 ka->boot_vcpu_runs_old_kvmclock = tmp;
2016 vcpu->arch.time = data;
2017 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2019 /* we verify if the enable bit is set... */
2023 gpa_offset = data & ~(PAGE_MASK | 1);
2025 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2026 &vcpu->arch.pv_time, data & ~1ULL,
2027 sizeof(struct pvclock_vcpu_time_info)))
2028 vcpu->arch.pv_time_enabled = false;
2030 vcpu->arch.pv_time_enabled = true;
2034 case MSR_KVM_ASYNC_PF_EN:
2035 if (kvm_pv_enable_async_pf(vcpu, data))
2038 case MSR_KVM_STEAL_TIME:
2040 if (unlikely(!sched_info_on()))
2043 if (data & KVM_STEAL_RESERVED_MASK)
2046 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2047 data & KVM_STEAL_VALID_BITS,
2048 sizeof(struct kvm_steal_time)))
2051 vcpu->arch.st.msr_val = data;
2053 if (!(data & KVM_MSR_ENABLED))
2056 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2059 accumulate_steal_time(vcpu);
2062 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2065 case MSR_KVM_PV_EOI_EN:
2066 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2070 case MSR_IA32_MCG_CTL:
2071 case MSR_IA32_MCG_STATUS:
2072 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2073 return set_msr_mce(vcpu, msr, data);
2075 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2076 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2077 pr = true; /* fall through */
2078 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2079 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2080 if (kvm_pmu_is_valid_msr(vcpu, msr))
2081 return kvm_pmu_set_msr(vcpu, msr_info);
2083 if (pr || data != 0)
2084 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2085 "0x%x data 0x%llx\n", msr, data);
2087 case MSR_K7_CLK_CTL:
2089 * Ignore all writes to this no longer documented MSR.
2090 * Writes are only relevant for old K7 processors,
2091 * all pre-dating SVM, but a recommended workaround from
2092 * AMD for these chips. It is possible to specify the
2093 * affected processor models on the command line, hence
2094 * the need to ignore the workaround.
2097 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2098 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2099 case HV_X64_MSR_CRASH_CTL:
2100 return kvm_hv_set_msr_common(vcpu, msr, data,
2101 msr_info->host_initiated);
2102 case MSR_IA32_BBL_CR_CTL3:
2103 /* Drop writes to this legacy MSR -- see rdmsr
2104 * counterpart for further detail.
2106 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2108 case MSR_AMD64_OSVW_ID_LENGTH:
2109 if (!guest_cpuid_has_osvw(vcpu))
2111 vcpu->arch.osvw.length = data;
2113 case MSR_AMD64_OSVW_STATUS:
2114 if (!guest_cpuid_has_osvw(vcpu))
2116 vcpu->arch.osvw.status = data;
2119 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2120 return xen_hvm_config(vcpu, data);
2121 if (kvm_pmu_is_valid_msr(vcpu, msr))
2122 return kvm_pmu_set_msr(vcpu, msr_info);
2124 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2128 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2135 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2139 * Reads an msr value (of 'msr_index') into 'pdata'.
2140 * Returns 0 on success, non-0 otherwise.
2141 * Assumes vcpu_load() was already called.
2143 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2145 return kvm_x86_ops->get_msr(vcpu, msr);
2147 EXPORT_SYMBOL_GPL(kvm_get_msr);
2149 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2152 u64 mcg_cap = vcpu->arch.mcg_cap;
2153 unsigned bank_num = mcg_cap & 0xff;
2156 case MSR_IA32_P5_MC_ADDR:
2157 case MSR_IA32_P5_MC_TYPE:
2160 case MSR_IA32_MCG_CAP:
2161 data = vcpu->arch.mcg_cap;
2163 case MSR_IA32_MCG_CTL:
2164 if (!(mcg_cap & MCG_CTL_P))
2166 data = vcpu->arch.mcg_ctl;
2168 case MSR_IA32_MCG_STATUS:
2169 data = vcpu->arch.mcg_status;
2172 if (msr >= MSR_IA32_MC0_CTL &&
2173 msr < MSR_IA32_MCx_CTL(bank_num)) {
2174 u32 offset = msr - MSR_IA32_MC0_CTL;
2175 data = vcpu->arch.mce_banks[offset];
2184 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2186 switch (msr_info->index) {
2187 case MSR_IA32_PLATFORM_ID:
2188 case MSR_IA32_EBL_CR_POWERON:
2189 case MSR_IA32_DEBUGCTLMSR:
2190 case MSR_IA32_LASTBRANCHFROMIP:
2191 case MSR_IA32_LASTBRANCHTOIP:
2192 case MSR_IA32_LASTINTFROMIP:
2193 case MSR_IA32_LASTINTTOIP:
2195 case MSR_K8_TSEG_ADDR:
2196 case MSR_K8_TSEG_MASK:
2198 case MSR_VM_HSAVE_PA:
2199 case MSR_K8_INT_PENDING_MSG:
2200 case MSR_AMD64_NB_CFG:
2201 case MSR_FAM10H_MMIO_CONF_BASE:
2202 case MSR_AMD64_BU_CFG2:
2205 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2206 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2207 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2208 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2209 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2210 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2213 case MSR_IA32_UCODE_REV:
2214 msr_info->data = 0x100000000ULL;
2217 case 0x200 ... 0x2ff:
2218 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2219 case 0xcd: /* fsb frequency */
2223 * MSR_EBC_FREQUENCY_ID
2224 * Conservative value valid for even the basic CPU models.
2225 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2226 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2227 * and 266MHz for model 3, or 4. Set Core Clock
2228 * Frequency to System Bus Frequency Ratio to 1 (bits
2229 * 31:24) even though these are only valid for CPU
2230 * models > 2, however guests may end up dividing or
2231 * multiplying by zero otherwise.
2233 case MSR_EBC_FREQUENCY_ID:
2234 msr_info->data = 1 << 24;
2236 case MSR_IA32_APICBASE:
2237 msr_info->data = kvm_get_apic_base(vcpu);
2239 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2240 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2242 case MSR_IA32_TSCDEADLINE:
2243 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2245 case MSR_IA32_TSC_ADJUST:
2246 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2248 case MSR_IA32_MISC_ENABLE:
2249 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2251 case MSR_IA32_SMBASE:
2252 if (!msr_info->host_initiated)
2254 msr_info->data = vcpu->arch.smbase;
2256 case MSR_IA32_PERF_STATUS:
2257 /* TSC increment by tick */
2258 msr_info->data = 1000ULL;
2259 /* CPU multiplier */
2260 msr_info->data |= (((uint64_t)4ULL) << 40);
2263 msr_info->data = vcpu->arch.efer;
2265 case MSR_KVM_WALL_CLOCK:
2266 case MSR_KVM_WALL_CLOCK_NEW:
2267 msr_info->data = vcpu->kvm->arch.wall_clock;
2269 case MSR_KVM_SYSTEM_TIME:
2270 case MSR_KVM_SYSTEM_TIME_NEW:
2271 msr_info->data = vcpu->arch.time;
2273 case MSR_KVM_ASYNC_PF_EN:
2274 msr_info->data = vcpu->arch.apf.msr_val;
2276 case MSR_KVM_STEAL_TIME:
2277 msr_info->data = vcpu->arch.st.msr_val;
2279 case MSR_KVM_PV_EOI_EN:
2280 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2282 case MSR_IA32_P5_MC_ADDR:
2283 case MSR_IA32_P5_MC_TYPE:
2284 case MSR_IA32_MCG_CAP:
2285 case MSR_IA32_MCG_CTL:
2286 case MSR_IA32_MCG_STATUS:
2287 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2288 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2289 case MSR_K7_CLK_CTL:
2291 * Provide expected ramp-up count for K7. All other
2292 * are set to zero, indicating minimum divisors for
2295 * This prevents guest kernels on AMD host with CPU
2296 * type 6, model 8 and higher from exploding due to
2297 * the rdmsr failing.
2299 msr_info->data = 0x20000000;
2301 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2302 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2303 case HV_X64_MSR_CRASH_CTL:
2304 return kvm_hv_get_msr_common(vcpu,
2305 msr_info->index, &msr_info->data);
2307 case MSR_IA32_BBL_CR_CTL3:
2308 /* This legacy MSR exists but isn't fully documented in current
2309 * silicon. It is however accessed by winxp in very narrow
2310 * scenarios where it sets bit #19, itself documented as
2311 * a "reserved" bit. Best effort attempt to source coherent
2312 * read data here should the balance of the register be
2313 * interpreted by the guest:
2315 * L2 cache control register 3: 64GB range, 256KB size,
2316 * enabled, latency 0x1, configured
2318 msr_info->data = 0xbe702111;
2320 case MSR_AMD64_OSVW_ID_LENGTH:
2321 if (!guest_cpuid_has_osvw(vcpu))
2323 msr_info->data = vcpu->arch.osvw.length;
2325 case MSR_AMD64_OSVW_STATUS:
2326 if (!guest_cpuid_has_osvw(vcpu))
2328 msr_info->data = vcpu->arch.osvw.status;
2331 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2332 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2334 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2337 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2344 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2347 * Read or write a bunch of msrs. All parameters are kernel addresses.
2349 * @return number of msrs set successfully.
2351 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2352 struct kvm_msr_entry *entries,
2353 int (*do_msr)(struct kvm_vcpu *vcpu,
2354 unsigned index, u64 *data))
2358 idx = srcu_read_lock(&vcpu->kvm->srcu);
2359 for (i = 0; i < msrs->nmsrs; ++i)
2360 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2362 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2368 * Read or write a bunch of msrs. Parameters are user addresses.
2370 * @return number of msrs set successfully.
2372 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2373 int (*do_msr)(struct kvm_vcpu *vcpu,
2374 unsigned index, u64 *data),
2377 struct kvm_msrs msrs;
2378 struct kvm_msr_entry *entries;
2383 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2387 if (msrs.nmsrs >= MAX_IO_MSRS)
2390 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2391 entries = memdup_user(user_msrs->entries, size);
2392 if (IS_ERR(entries)) {
2393 r = PTR_ERR(entries);
2397 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2402 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2413 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2418 case KVM_CAP_IRQCHIP:
2420 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2421 case KVM_CAP_SET_TSS_ADDR:
2422 case KVM_CAP_EXT_CPUID:
2423 case KVM_CAP_EXT_EMUL_CPUID:
2424 case KVM_CAP_CLOCKSOURCE:
2426 case KVM_CAP_NOP_IO_DELAY:
2427 case KVM_CAP_MP_STATE:
2428 case KVM_CAP_SYNC_MMU:
2429 case KVM_CAP_USER_NMI:
2430 case KVM_CAP_REINJECT_CONTROL:
2431 case KVM_CAP_IRQ_INJECT_STATUS:
2432 case KVM_CAP_IOEVENTFD:
2433 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2435 case KVM_CAP_PIT_STATE2:
2436 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2437 case KVM_CAP_XEN_HVM:
2438 case KVM_CAP_ADJUST_CLOCK:
2439 case KVM_CAP_VCPU_EVENTS:
2440 case KVM_CAP_HYPERV:
2441 case KVM_CAP_HYPERV_VAPIC:
2442 case KVM_CAP_HYPERV_SPIN:
2443 case KVM_CAP_PCI_SEGMENT:
2444 case KVM_CAP_DEBUGREGS:
2445 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2447 case KVM_CAP_ASYNC_PF:
2448 case KVM_CAP_GET_TSC_KHZ:
2449 case KVM_CAP_KVMCLOCK_CTRL:
2450 case KVM_CAP_READONLY_MEM:
2451 case KVM_CAP_HYPERV_TIME:
2452 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2453 case KVM_CAP_TSC_DEADLINE_TIMER:
2454 case KVM_CAP_ENABLE_CAP_VM:
2455 case KVM_CAP_DISABLE_QUIRKS:
2456 case KVM_CAP_SET_BOOT_CPU_ID:
2457 case KVM_CAP_SPLIT_IRQCHIP:
2458 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2459 case KVM_CAP_ASSIGN_DEV_IRQ:
2460 case KVM_CAP_PCI_2_3:
2464 case KVM_CAP_X86_SMM:
2465 /* SMBASE is usually relocated above 1M on modern chipsets,
2466 * and SMM handlers might indeed rely on 4G segment limits,
2467 * so do not report SMM to be available if real mode is
2468 * emulated via vm86 mode. Still, do not go to great lengths
2469 * to avoid userspace's usage of the feature, because it is a
2470 * fringe case that is not enabled except via specific settings
2471 * of the module parameters.
2473 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2475 case KVM_CAP_COALESCED_MMIO:
2476 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2479 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2481 case KVM_CAP_NR_VCPUS:
2482 r = KVM_SOFT_MAX_VCPUS;
2484 case KVM_CAP_MAX_VCPUS:
2487 case KVM_CAP_NR_MEMSLOTS:
2488 r = KVM_USER_MEM_SLOTS;
2490 case KVM_CAP_PV_MMU: /* obsolete */
2493 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2495 r = iommu_present(&pci_bus_type);
2499 r = KVM_MAX_MCE_BANKS;
2504 case KVM_CAP_TSC_CONTROL:
2505 r = kvm_has_tsc_control;
2515 long kvm_arch_dev_ioctl(struct file *filp,
2516 unsigned int ioctl, unsigned long arg)
2518 void __user *argp = (void __user *)arg;
2522 case KVM_GET_MSR_INDEX_LIST: {
2523 struct kvm_msr_list __user *user_msr_list = argp;
2524 struct kvm_msr_list msr_list;
2528 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2531 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2532 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2535 if (n < msr_list.nmsrs)
2538 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2539 num_msrs_to_save * sizeof(u32)))
2541 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2543 num_emulated_msrs * sizeof(u32)))
2548 case KVM_GET_SUPPORTED_CPUID:
2549 case KVM_GET_EMULATED_CPUID: {
2550 struct kvm_cpuid2 __user *cpuid_arg = argp;
2551 struct kvm_cpuid2 cpuid;
2554 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2557 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2563 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2568 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2571 mce_cap = KVM_MCE_CAP_SUPPORTED;
2573 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2585 static void wbinvd_ipi(void *garbage)
2590 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2592 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2595 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2597 /* Address WBINVD may be executed by guest */
2598 if (need_emulate_wbinvd(vcpu)) {
2599 if (kvm_x86_ops->has_wbinvd_exit())
2600 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2601 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2602 smp_call_function_single(vcpu->cpu,
2603 wbinvd_ipi, NULL, 1);
2606 kvm_x86_ops->vcpu_load(vcpu, cpu);
2608 /* Apply any externally detected TSC adjustments (due to suspend) */
2609 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2610 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2611 vcpu->arch.tsc_offset_adjustment = 0;
2612 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2615 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2616 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2617 rdtsc() - vcpu->arch.last_host_tsc;
2619 mark_tsc_unstable("KVM discovered backwards TSC");
2620 if (check_tsc_unstable()) {
2621 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2622 vcpu->arch.last_guest_tsc);
2623 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2624 vcpu->arch.tsc_catchup = 1;
2627 * On a host with synchronized TSC, there is no need to update
2628 * kvmclock on vcpu->cpu migration
2630 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2631 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2632 if (vcpu->cpu != cpu)
2633 kvm_migrate_timers(vcpu);
2637 accumulate_steal_time(vcpu);
2638 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2641 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2643 kvm_x86_ops->vcpu_put(vcpu);
2644 kvm_put_guest_fpu(vcpu);
2645 vcpu->arch.last_host_tsc = rdtsc();
2648 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2649 struct kvm_lapic_state *s)
2651 kvm_x86_ops->sync_pir_to_irr(vcpu);
2652 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2657 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2658 struct kvm_lapic_state *s)
2660 kvm_apic_post_state_restore(vcpu, s);
2661 update_cr8_intercept(vcpu);
2666 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2667 struct kvm_interrupt *irq)
2669 if (irq->irq >= KVM_NR_INTERRUPTS)
2672 if (!irqchip_in_kernel(vcpu->kvm)) {
2673 kvm_queue_interrupt(vcpu, irq->irq, false);
2674 kvm_make_request(KVM_REQ_EVENT, vcpu);
2679 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2680 * fail for in-kernel 8259.
2682 if (pic_in_kernel(vcpu->kvm))
2685 if (vcpu->arch.pending_external_vector != -1)
2688 vcpu->arch.pending_external_vector = irq->irq;
2692 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2694 kvm_inject_nmi(vcpu);
2699 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2701 kvm_make_request(KVM_REQ_SMI, vcpu);
2706 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2707 struct kvm_tpr_access_ctl *tac)
2711 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2715 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2719 unsigned bank_num = mcg_cap & 0xff, bank;
2722 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2724 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2727 vcpu->arch.mcg_cap = mcg_cap;
2728 /* Init IA32_MCG_CTL to all 1s */
2729 if (mcg_cap & MCG_CTL_P)
2730 vcpu->arch.mcg_ctl = ~(u64)0;
2731 /* Init IA32_MCi_CTL to all 1s */
2732 for (bank = 0; bank < bank_num; bank++)
2733 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2738 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2739 struct kvm_x86_mce *mce)
2741 u64 mcg_cap = vcpu->arch.mcg_cap;
2742 unsigned bank_num = mcg_cap & 0xff;
2743 u64 *banks = vcpu->arch.mce_banks;
2745 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2748 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2749 * reporting is disabled
2751 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2752 vcpu->arch.mcg_ctl != ~(u64)0)
2754 banks += 4 * mce->bank;
2756 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2757 * reporting is disabled for the bank
2759 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2761 if (mce->status & MCI_STATUS_UC) {
2762 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2763 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2764 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2767 if (banks[1] & MCI_STATUS_VAL)
2768 mce->status |= MCI_STATUS_OVER;
2769 banks[2] = mce->addr;
2770 banks[3] = mce->misc;
2771 vcpu->arch.mcg_status = mce->mcg_status;
2772 banks[1] = mce->status;
2773 kvm_queue_exception(vcpu, MC_VECTOR);
2774 } else if (!(banks[1] & MCI_STATUS_VAL)
2775 || !(banks[1] & MCI_STATUS_UC)) {
2776 if (banks[1] & MCI_STATUS_VAL)
2777 mce->status |= MCI_STATUS_OVER;
2778 banks[2] = mce->addr;
2779 banks[3] = mce->misc;
2780 banks[1] = mce->status;
2782 banks[1] |= MCI_STATUS_OVER;
2786 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2787 struct kvm_vcpu_events *events)
2790 events->exception.injected =
2791 vcpu->arch.exception.pending &&
2792 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2793 events->exception.nr = vcpu->arch.exception.nr;
2794 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2795 events->exception.pad = 0;
2796 events->exception.error_code = vcpu->arch.exception.error_code;
2798 events->interrupt.injected =
2799 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2800 events->interrupt.nr = vcpu->arch.interrupt.nr;
2801 events->interrupt.soft = 0;
2802 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2804 events->nmi.injected = vcpu->arch.nmi_injected;
2805 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2806 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2807 events->nmi.pad = 0;
2809 events->sipi_vector = 0; /* never valid when reporting to user space */
2811 events->smi.smm = is_smm(vcpu);
2812 events->smi.pending = vcpu->arch.smi_pending;
2813 events->smi.smm_inside_nmi =
2814 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2815 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2817 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2818 | KVM_VCPUEVENT_VALID_SHADOW
2819 | KVM_VCPUEVENT_VALID_SMM);
2820 memset(&events->reserved, 0, sizeof(events->reserved));
2823 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2824 struct kvm_vcpu_events *events)
2826 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2827 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2828 | KVM_VCPUEVENT_VALID_SHADOW
2829 | KVM_VCPUEVENT_VALID_SMM))
2833 vcpu->arch.exception.pending = events->exception.injected;
2834 vcpu->arch.exception.nr = events->exception.nr;
2835 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2836 vcpu->arch.exception.error_code = events->exception.error_code;
2838 vcpu->arch.interrupt.pending = events->interrupt.injected;
2839 vcpu->arch.interrupt.nr = events->interrupt.nr;
2840 vcpu->arch.interrupt.soft = events->interrupt.soft;
2841 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2842 kvm_x86_ops->set_interrupt_shadow(vcpu,
2843 events->interrupt.shadow);
2845 vcpu->arch.nmi_injected = events->nmi.injected;
2846 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2847 vcpu->arch.nmi_pending = events->nmi.pending;
2848 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2850 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2851 kvm_vcpu_has_lapic(vcpu))
2852 vcpu->arch.apic->sipi_vector = events->sipi_vector;
2854 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
2855 if (events->smi.smm)
2856 vcpu->arch.hflags |= HF_SMM_MASK;
2858 vcpu->arch.hflags &= ~HF_SMM_MASK;
2859 vcpu->arch.smi_pending = events->smi.pending;
2860 if (events->smi.smm_inside_nmi)
2861 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
2863 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
2864 if (kvm_vcpu_has_lapic(vcpu)) {
2865 if (events->smi.latched_init)
2866 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2868 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
2872 kvm_make_request(KVM_REQ_EVENT, vcpu);
2877 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2878 struct kvm_debugregs *dbgregs)
2882 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2883 kvm_get_dr(vcpu, 6, &val);
2885 dbgregs->dr7 = vcpu->arch.dr7;
2887 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2890 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2891 struct kvm_debugregs *dbgregs)
2896 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2897 kvm_update_dr0123(vcpu);
2898 vcpu->arch.dr6 = dbgregs->dr6;
2899 kvm_update_dr6(vcpu);
2900 vcpu->arch.dr7 = dbgregs->dr7;
2901 kvm_update_dr7(vcpu);
2906 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
2908 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
2910 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
2911 u64 xstate_bv = xsave->header.xfeatures;
2915 * Copy legacy XSAVE area, to avoid complications with CPUID
2916 * leaves 0 and 1 in the loop below.
2918 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
2921 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
2924 * Copy each region from the possibly compacted offset to the
2925 * non-compacted offset.
2927 valid = xstate_bv & ~XSTATE_FPSSE;
2929 u64 feature = valid & -valid;
2930 int index = fls64(feature) - 1;
2931 void *src = get_xsave_addr(xsave, feature);
2934 u32 size, offset, ecx, edx;
2935 cpuid_count(XSTATE_CPUID, index,
2936 &size, &offset, &ecx, &edx);
2937 memcpy(dest + offset, src, size);
2944 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
2946 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
2947 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
2951 * Copy legacy XSAVE area, to avoid complications with CPUID
2952 * leaves 0 and 1 in the loop below.
2954 memcpy(xsave, src, XSAVE_HDR_OFFSET);
2956 /* Set XSTATE_BV and possibly XCOMP_BV. */
2957 xsave->header.xfeatures = xstate_bv;
2959 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
2962 * Copy each region from the non-compacted offset to the
2963 * possibly compacted offset.
2965 valid = xstate_bv & ~XSTATE_FPSSE;
2967 u64 feature = valid & -valid;
2968 int index = fls64(feature) - 1;
2969 void *dest = get_xsave_addr(xsave, feature);
2972 u32 size, offset, ecx, edx;
2973 cpuid_count(XSTATE_CPUID, index,
2974 &size, &offset, &ecx, &edx);
2975 memcpy(dest, src + offset, size);
2982 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2983 struct kvm_xsave *guest_xsave)
2985 if (cpu_has_xsave) {
2986 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
2987 fill_xsave((u8 *) guest_xsave->region, vcpu);
2989 memcpy(guest_xsave->region,
2990 &vcpu->arch.guest_fpu.state.fxsave,
2991 sizeof(struct fxregs_state));
2992 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2997 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2998 struct kvm_xsave *guest_xsave)
3001 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3003 if (cpu_has_xsave) {
3005 * Here we allow setting states that are not present in
3006 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3007 * with old userspace.
3009 if (xstate_bv & ~kvm_supported_xcr0())
3011 load_xsave(vcpu, (u8 *)guest_xsave->region);
3013 if (xstate_bv & ~XSTATE_FPSSE)
3015 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3016 guest_xsave->region, sizeof(struct fxregs_state));
3021 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3022 struct kvm_xcrs *guest_xcrs)
3024 if (!cpu_has_xsave) {
3025 guest_xcrs->nr_xcrs = 0;
3029 guest_xcrs->nr_xcrs = 1;
3030 guest_xcrs->flags = 0;
3031 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3032 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3035 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3036 struct kvm_xcrs *guest_xcrs)
3043 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3046 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3047 /* Only support XCR0 currently */
3048 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3049 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3050 guest_xcrs->xcrs[i].value);
3059 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3060 * stopped by the hypervisor. This function will be called from the host only.
3061 * EINVAL is returned when the host attempts to set the flag for a guest that
3062 * does not support pv clocks.
3064 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3066 if (!vcpu->arch.pv_time_enabled)
3068 vcpu->arch.pvclock_set_guest_stopped_request = true;
3069 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3073 long kvm_arch_vcpu_ioctl(struct file *filp,
3074 unsigned int ioctl, unsigned long arg)
3076 struct kvm_vcpu *vcpu = filp->private_data;
3077 void __user *argp = (void __user *)arg;
3080 struct kvm_lapic_state *lapic;
3081 struct kvm_xsave *xsave;
3082 struct kvm_xcrs *xcrs;
3088 case KVM_GET_LAPIC: {
3090 if (!vcpu->arch.apic)
3092 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3097 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3101 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3106 case KVM_SET_LAPIC: {
3108 if (!vcpu->arch.apic)
3110 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3111 if (IS_ERR(u.lapic))
3112 return PTR_ERR(u.lapic);
3114 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3117 case KVM_INTERRUPT: {
3118 struct kvm_interrupt irq;
3121 if (copy_from_user(&irq, argp, sizeof irq))
3123 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3127 r = kvm_vcpu_ioctl_nmi(vcpu);
3131 r = kvm_vcpu_ioctl_smi(vcpu);
3134 case KVM_SET_CPUID: {
3135 struct kvm_cpuid __user *cpuid_arg = argp;
3136 struct kvm_cpuid cpuid;
3139 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3141 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3144 case KVM_SET_CPUID2: {
3145 struct kvm_cpuid2 __user *cpuid_arg = argp;
3146 struct kvm_cpuid2 cpuid;
3149 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3151 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3152 cpuid_arg->entries);
3155 case KVM_GET_CPUID2: {
3156 struct kvm_cpuid2 __user *cpuid_arg = argp;
3157 struct kvm_cpuid2 cpuid;
3160 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3162 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3163 cpuid_arg->entries);
3167 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3173 r = msr_io(vcpu, argp, do_get_msr, 1);
3176 r = msr_io(vcpu, argp, do_set_msr, 0);
3178 case KVM_TPR_ACCESS_REPORTING: {
3179 struct kvm_tpr_access_ctl tac;
3182 if (copy_from_user(&tac, argp, sizeof tac))
3184 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3188 if (copy_to_user(argp, &tac, sizeof tac))
3193 case KVM_SET_VAPIC_ADDR: {
3194 struct kvm_vapic_addr va;
3197 if (!lapic_in_kernel(vcpu))
3200 if (copy_from_user(&va, argp, sizeof va))
3202 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3205 case KVM_X86_SETUP_MCE: {
3209 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3211 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3214 case KVM_X86_SET_MCE: {
3215 struct kvm_x86_mce mce;
3218 if (copy_from_user(&mce, argp, sizeof mce))
3220 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3223 case KVM_GET_VCPU_EVENTS: {
3224 struct kvm_vcpu_events events;
3226 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3229 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3234 case KVM_SET_VCPU_EVENTS: {
3235 struct kvm_vcpu_events events;
3238 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3241 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3244 case KVM_GET_DEBUGREGS: {
3245 struct kvm_debugregs dbgregs;
3247 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3250 if (copy_to_user(argp, &dbgregs,
3251 sizeof(struct kvm_debugregs)))
3256 case KVM_SET_DEBUGREGS: {
3257 struct kvm_debugregs dbgregs;
3260 if (copy_from_user(&dbgregs, argp,
3261 sizeof(struct kvm_debugregs)))
3264 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3267 case KVM_GET_XSAVE: {
3268 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3273 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3276 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3281 case KVM_SET_XSAVE: {
3282 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3283 if (IS_ERR(u.xsave))
3284 return PTR_ERR(u.xsave);
3286 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3289 case KVM_GET_XCRS: {
3290 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3295 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3298 if (copy_to_user(argp, u.xcrs,
3299 sizeof(struct kvm_xcrs)))
3304 case KVM_SET_XCRS: {
3305 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3307 return PTR_ERR(u.xcrs);
3309 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3312 case KVM_SET_TSC_KHZ: {
3316 user_tsc_khz = (u32)arg;
3318 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3321 if (user_tsc_khz == 0)
3322 user_tsc_khz = tsc_khz;
3324 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3329 case KVM_GET_TSC_KHZ: {
3330 r = vcpu->arch.virtual_tsc_khz;
3333 case KVM_KVMCLOCK_CTRL: {
3334 r = kvm_set_guest_paused(vcpu);
3345 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3347 return VM_FAULT_SIGBUS;
3350 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3354 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3356 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3360 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3363 kvm->arch.ept_identity_map_addr = ident_addr;
3367 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3368 u32 kvm_nr_mmu_pages)
3370 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3373 mutex_lock(&kvm->slots_lock);
3375 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3376 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3378 mutex_unlock(&kvm->slots_lock);
3382 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3384 return kvm->arch.n_max_mmu_pages;
3387 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3392 switch (chip->chip_id) {
3393 case KVM_IRQCHIP_PIC_MASTER:
3394 memcpy(&chip->chip.pic,
3395 &pic_irqchip(kvm)->pics[0],
3396 sizeof(struct kvm_pic_state));
3398 case KVM_IRQCHIP_PIC_SLAVE:
3399 memcpy(&chip->chip.pic,
3400 &pic_irqchip(kvm)->pics[1],
3401 sizeof(struct kvm_pic_state));
3403 case KVM_IRQCHIP_IOAPIC:
3404 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3413 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3418 switch (chip->chip_id) {
3419 case KVM_IRQCHIP_PIC_MASTER:
3420 spin_lock(&pic_irqchip(kvm)->lock);
3421 memcpy(&pic_irqchip(kvm)->pics[0],
3423 sizeof(struct kvm_pic_state));
3424 spin_unlock(&pic_irqchip(kvm)->lock);
3426 case KVM_IRQCHIP_PIC_SLAVE:
3427 spin_lock(&pic_irqchip(kvm)->lock);
3428 memcpy(&pic_irqchip(kvm)->pics[1],
3430 sizeof(struct kvm_pic_state));
3431 spin_unlock(&pic_irqchip(kvm)->lock);
3433 case KVM_IRQCHIP_IOAPIC:
3434 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3440 kvm_pic_update_irq(pic_irqchip(kvm));
3444 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3448 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3449 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3450 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3454 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3458 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3459 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3460 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3461 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3465 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3469 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3470 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3471 sizeof(ps->channels));
3472 ps->flags = kvm->arch.vpit->pit_state.flags;
3473 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3474 memset(&ps->reserved, 0, sizeof(ps->reserved));
3478 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3480 int r = 0, start = 0;
3481 u32 prev_legacy, cur_legacy;
3482 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3483 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3484 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3485 if (!prev_legacy && cur_legacy)
3487 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3488 sizeof(kvm->arch.vpit->pit_state.channels));
3489 kvm->arch.vpit->pit_state.flags = ps->flags;
3490 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3491 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3495 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3496 struct kvm_reinject_control *control)
3498 if (!kvm->arch.vpit)
3500 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3501 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3502 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3507 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3508 * @kvm: kvm instance
3509 * @log: slot id and address to which we copy the log
3511 * Steps 1-4 below provide general overview of dirty page logging. See
3512 * kvm_get_dirty_log_protect() function description for additional details.
3514 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3515 * always flush the TLB (step 4) even if previous step failed and the dirty
3516 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3517 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3518 * writes will be marked dirty for next log read.
3520 * 1. Take a snapshot of the bit and clear it if needed.
3521 * 2. Write protect the corresponding page.
3522 * 3. Copy the snapshot to the userspace.
3523 * 4. Flush TLB's if needed.
3525 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3527 bool is_dirty = false;
3530 mutex_lock(&kvm->slots_lock);
3533 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3535 if (kvm_x86_ops->flush_log_dirty)
3536 kvm_x86_ops->flush_log_dirty(kvm);
3538 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3541 * All the TLBs can be flushed out of mmu lock, see the comments in
3542 * kvm_mmu_slot_remove_write_access().
3544 lockdep_assert_held(&kvm->slots_lock);
3546 kvm_flush_remote_tlbs(kvm);
3548 mutex_unlock(&kvm->slots_lock);
3552 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3555 if (!irqchip_in_kernel(kvm))
3558 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3559 irq_event->irq, irq_event->level,
3564 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3565 struct kvm_enable_cap *cap)
3573 case KVM_CAP_DISABLE_QUIRKS:
3574 kvm->arch.disabled_quirks = cap->args[0];
3577 case KVM_CAP_SPLIT_IRQCHIP: {
3578 mutex_lock(&kvm->lock);
3580 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3581 goto split_irqchip_unlock;
3583 if (irqchip_in_kernel(kvm))
3584 goto split_irqchip_unlock;
3585 if (atomic_read(&kvm->online_vcpus))
3586 goto split_irqchip_unlock;
3587 r = kvm_setup_empty_irq_routing(kvm);
3589 goto split_irqchip_unlock;
3590 /* Pairs with irqchip_in_kernel. */
3592 kvm->arch.irqchip_split = true;
3593 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3595 split_irqchip_unlock:
3596 mutex_unlock(&kvm->lock);
3606 long kvm_arch_vm_ioctl(struct file *filp,
3607 unsigned int ioctl, unsigned long arg)
3609 struct kvm *kvm = filp->private_data;
3610 void __user *argp = (void __user *)arg;
3613 * This union makes it completely explicit to gcc-3.x
3614 * that these two variables' stack usage should be
3615 * combined, not added together.
3618 struct kvm_pit_state ps;
3619 struct kvm_pit_state2 ps2;
3620 struct kvm_pit_config pit_config;
3624 case KVM_SET_TSS_ADDR:
3625 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3627 case KVM_SET_IDENTITY_MAP_ADDR: {
3631 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3633 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3636 case KVM_SET_NR_MMU_PAGES:
3637 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3639 case KVM_GET_NR_MMU_PAGES:
3640 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3642 case KVM_CREATE_IRQCHIP: {
3643 struct kvm_pic *vpic;
3645 mutex_lock(&kvm->lock);
3648 goto create_irqchip_unlock;
3650 if (atomic_read(&kvm->online_vcpus))
3651 goto create_irqchip_unlock;
3653 vpic = kvm_create_pic(kvm);
3655 r = kvm_ioapic_init(kvm);
3657 mutex_lock(&kvm->slots_lock);
3658 kvm_destroy_pic(vpic);
3659 mutex_unlock(&kvm->slots_lock);
3660 goto create_irqchip_unlock;
3663 goto create_irqchip_unlock;
3664 r = kvm_setup_default_irq_routing(kvm);
3666 mutex_lock(&kvm->slots_lock);
3667 mutex_lock(&kvm->irq_lock);
3668 kvm_ioapic_destroy(kvm);
3669 kvm_destroy_pic(vpic);
3670 mutex_unlock(&kvm->irq_lock);
3671 mutex_unlock(&kvm->slots_lock);
3672 goto create_irqchip_unlock;
3674 /* Write kvm->irq_routing before kvm->arch.vpic. */
3676 kvm->arch.vpic = vpic;
3677 create_irqchip_unlock:
3678 mutex_unlock(&kvm->lock);
3681 case KVM_CREATE_PIT:
3682 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3684 case KVM_CREATE_PIT2:
3686 if (copy_from_user(&u.pit_config, argp,
3687 sizeof(struct kvm_pit_config)))
3690 mutex_lock(&kvm->slots_lock);
3693 goto create_pit_unlock;
3695 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3699 mutex_unlock(&kvm->slots_lock);
3701 case KVM_GET_IRQCHIP: {
3702 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3703 struct kvm_irqchip *chip;
3705 chip = memdup_user(argp, sizeof(*chip));
3712 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3713 goto get_irqchip_out;
3714 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3716 goto get_irqchip_out;
3718 if (copy_to_user(argp, chip, sizeof *chip))
3719 goto get_irqchip_out;
3725 case KVM_SET_IRQCHIP: {
3726 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3727 struct kvm_irqchip *chip;
3729 chip = memdup_user(argp, sizeof(*chip));
3736 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3737 goto set_irqchip_out;
3738 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3740 goto set_irqchip_out;
3748 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3751 if (!kvm->arch.vpit)
3753 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3757 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3764 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3767 if (!kvm->arch.vpit)
3769 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3772 case KVM_GET_PIT2: {
3774 if (!kvm->arch.vpit)
3776 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3780 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3785 case KVM_SET_PIT2: {
3787 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3790 if (!kvm->arch.vpit)
3792 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3795 case KVM_REINJECT_CONTROL: {
3796 struct kvm_reinject_control control;
3798 if (copy_from_user(&control, argp, sizeof(control)))
3800 r = kvm_vm_ioctl_reinject(kvm, &control);
3803 case KVM_SET_BOOT_CPU_ID:
3805 mutex_lock(&kvm->lock);
3806 if (atomic_read(&kvm->online_vcpus) != 0)
3809 kvm->arch.bsp_vcpu_id = arg;
3810 mutex_unlock(&kvm->lock);
3812 case KVM_XEN_HVM_CONFIG: {
3814 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3815 sizeof(struct kvm_xen_hvm_config)))
3818 if (kvm->arch.xen_hvm_config.flags)
3823 case KVM_SET_CLOCK: {
3824 struct kvm_clock_data user_ns;
3829 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3837 local_irq_disable();
3838 now_ns = get_kernel_ns();
3839 delta = user_ns.clock - now_ns;
3841 kvm->arch.kvmclock_offset = delta;
3842 kvm_gen_update_masterclock(kvm);
3845 case KVM_GET_CLOCK: {
3846 struct kvm_clock_data user_ns;
3849 local_irq_disable();
3850 now_ns = get_kernel_ns();
3851 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3854 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3857 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3862 case KVM_ENABLE_CAP: {
3863 struct kvm_enable_cap cap;
3866 if (copy_from_user(&cap, argp, sizeof(cap)))
3868 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
3872 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
3878 static void kvm_init_msr_list(void)
3883 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
3884 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3888 * Even MSRs that are valid in the host may not be exposed
3889 * to the guests in some cases. We could work around this
3890 * in VMX with the generic MSR save/load machinery, but it
3891 * is not really worthwhile since it will really only
3892 * happen with nested virtualization.
3894 switch (msrs_to_save[i]) {
3895 case MSR_IA32_BNDCFGS:
3896 if (!kvm_x86_ops->mpx_supported())
3904 msrs_to_save[j] = msrs_to_save[i];
3907 num_msrs_to_save = j;
3909 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
3910 switch (emulated_msrs[i]) {
3911 case MSR_IA32_SMBASE:
3912 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
3920 emulated_msrs[j] = emulated_msrs[i];
3923 num_emulated_msrs = j;
3926 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3934 if (!(vcpu->arch.apic &&
3935 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
3936 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
3947 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3954 if (!(vcpu->arch.apic &&
3955 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
3957 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
3959 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3969 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3970 struct kvm_segment *var, int seg)
3972 kvm_x86_ops->set_segment(vcpu, var, seg);
3975 void kvm_get_segment(struct kvm_vcpu *vcpu,
3976 struct kvm_segment *var, int seg)
3978 kvm_x86_ops->get_segment(vcpu, var, seg);
3981 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
3982 struct x86_exception *exception)
3986 BUG_ON(!mmu_is_nested(vcpu));
3988 /* NPT walks are always user-walks */
3989 access |= PFERR_USER_MASK;
3990 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
3995 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3996 struct x86_exception *exception)
3998 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3999 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4002 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4003 struct x86_exception *exception)
4005 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4006 access |= PFERR_FETCH_MASK;
4007 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4010 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4011 struct x86_exception *exception)
4013 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4014 access |= PFERR_WRITE_MASK;
4015 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4018 /* uses this to access any guest's mapped memory without checking CPL */
4019 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4020 struct x86_exception *exception)
4022 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4025 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4026 struct kvm_vcpu *vcpu, u32 access,
4027 struct x86_exception *exception)
4030 int r = X86EMUL_CONTINUE;
4033 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4035 unsigned offset = addr & (PAGE_SIZE-1);
4036 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4039 if (gpa == UNMAPPED_GVA)
4040 return X86EMUL_PROPAGATE_FAULT;
4041 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4044 r = X86EMUL_IO_NEEDED;
4056 /* used for instruction fetching */
4057 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4058 gva_t addr, void *val, unsigned int bytes,
4059 struct x86_exception *exception)
4061 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4062 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4066 /* Inline kvm_read_guest_virt_helper for speed. */
4067 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4069 if (unlikely(gpa == UNMAPPED_GVA))
4070 return X86EMUL_PROPAGATE_FAULT;
4072 offset = addr & (PAGE_SIZE-1);
4073 if (WARN_ON(offset + bytes > PAGE_SIZE))
4074 bytes = (unsigned)PAGE_SIZE - offset;
4075 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4077 if (unlikely(ret < 0))
4078 return X86EMUL_IO_NEEDED;
4080 return X86EMUL_CONTINUE;
4083 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4084 gva_t addr, void *val, unsigned int bytes,
4085 struct x86_exception *exception)
4087 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4088 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4090 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4093 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4095 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4096 gva_t addr, void *val, unsigned int bytes,
4097 struct x86_exception *exception)
4099 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4100 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4103 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4104 gva_t addr, void *val,
4106 struct x86_exception *exception)
4108 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4110 int r = X86EMUL_CONTINUE;
4113 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4116 unsigned offset = addr & (PAGE_SIZE-1);
4117 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4120 if (gpa == UNMAPPED_GVA)
4121 return X86EMUL_PROPAGATE_FAULT;
4122 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4124 r = X86EMUL_IO_NEEDED;
4135 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4137 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4138 gpa_t *gpa, struct x86_exception *exception,
4141 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4142 | (write ? PFERR_WRITE_MASK : 0);
4144 if (vcpu_match_mmio_gva(vcpu, gva)
4145 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4146 vcpu->arch.access, access)) {
4147 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4148 (gva & (PAGE_SIZE - 1));
4149 trace_vcpu_match_mmio(gva, *gpa, write, false);
4153 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4155 if (*gpa == UNMAPPED_GVA)
4158 /* For APIC access vmexit */
4159 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4162 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4163 trace_vcpu_match_mmio(gva, *gpa, write, true);
4170 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4171 const void *val, int bytes)
4175 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4178 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4182 struct read_write_emulator_ops {
4183 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4185 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4186 void *val, int bytes);
4187 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4188 int bytes, void *val);
4189 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4190 void *val, int bytes);
4194 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4196 if (vcpu->mmio_read_completed) {
4197 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4198 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4199 vcpu->mmio_read_completed = 0;
4206 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4207 void *val, int bytes)
4209 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4212 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4213 void *val, int bytes)
4215 return emulator_write_phys(vcpu, gpa, val, bytes);
4218 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4220 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4221 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4224 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4225 void *val, int bytes)
4227 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4228 return X86EMUL_IO_NEEDED;
4231 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4232 void *val, int bytes)
4234 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4236 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4237 return X86EMUL_CONTINUE;
4240 static const struct read_write_emulator_ops read_emultor = {
4241 .read_write_prepare = read_prepare,
4242 .read_write_emulate = read_emulate,
4243 .read_write_mmio = vcpu_mmio_read,
4244 .read_write_exit_mmio = read_exit_mmio,
4247 static const struct read_write_emulator_ops write_emultor = {
4248 .read_write_emulate = write_emulate,
4249 .read_write_mmio = write_mmio,
4250 .read_write_exit_mmio = write_exit_mmio,
4254 static int emulator_read_write_onepage(unsigned long addr, void *val,
4256 struct x86_exception *exception,
4257 struct kvm_vcpu *vcpu,
4258 const struct read_write_emulator_ops *ops)
4262 bool write = ops->write;
4263 struct kvm_mmio_fragment *frag;
4265 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4268 return X86EMUL_PROPAGATE_FAULT;
4270 /* For APIC access vmexit */
4274 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4275 return X86EMUL_CONTINUE;
4279 * Is this MMIO handled locally?
4281 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4282 if (handled == bytes)
4283 return X86EMUL_CONTINUE;
4289 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4290 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4294 return X86EMUL_CONTINUE;
4297 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4299 void *val, unsigned int bytes,
4300 struct x86_exception *exception,
4301 const struct read_write_emulator_ops *ops)
4303 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4307 if (ops->read_write_prepare &&
4308 ops->read_write_prepare(vcpu, val, bytes))
4309 return X86EMUL_CONTINUE;
4311 vcpu->mmio_nr_fragments = 0;
4313 /* Crossing a page boundary? */
4314 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4317 now = -addr & ~PAGE_MASK;
4318 rc = emulator_read_write_onepage(addr, val, now, exception,
4321 if (rc != X86EMUL_CONTINUE)
4324 if (ctxt->mode != X86EMUL_MODE_PROT64)
4330 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4332 if (rc != X86EMUL_CONTINUE)
4335 if (!vcpu->mmio_nr_fragments)
4338 gpa = vcpu->mmio_fragments[0].gpa;
4340 vcpu->mmio_needed = 1;
4341 vcpu->mmio_cur_fragment = 0;
4343 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4344 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4345 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4346 vcpu->run->mmio.phys_addr = gpa;
4348 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4351 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4355 struct x86_exception *exception)
4357 return emulator_read_write(ctxt, addr, val, bytes,
4358 exception, &read_emultor);
4361 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4365 struct x86_exception *exception)
4367 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4368 exception, &write_emultor);
4371 #define CMPXCHG_TYPE(t, ptr, old, new) \
4372 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4374 #ifdef CONFIG_X86_64
4375 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4377 # define CMPXCHG64(ptr, old, new) \
4378 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4381 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4386 struct x86_exception *exception)
4388 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4394 /* guests cmpxchg8b have to be emulated atomically */
4395 if (bytes > 8 || (bytes & (bytes - 1)))
4398 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4400 if (gpa == UNMAPPED_GVA ||
4401 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4404 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4407 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4408 if (is_error_page(page))
4411 kaddr = kmap_atomic(page);
4412 kaddr += offset_in_page(gpa);
4415 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4418 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4421 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4424 exchanged = CMPXCHG64(kaddr, old, new);
4429 kunmap_atomic(kaddr);
4430 kvm_release_page_dirty(page);
4433 return X86EMUL_CMPXCHG_FAILED;
4435 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4436 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4438 return X86EMUL_CONTINUE;
4441 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4443 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4446 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4448 /* TODO: String I/O for in kernel device */
4451 if (vcpu->arch.pio.in)
4452 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4453 vcpu->arch.pio.size, pd);
4455 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4456 vcpu->arch.pio.port, vcpu->arch.pio.size,
4461 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4462 unsigned short port, void *val,
4463 unsigned int count, bool in)
4465 vcpu->arch.pio.port = port;
4466 vcpu->arch.pio.in = in;
4467 vcpu->arch.pio.count = count;
4468 vcpu->arch.pio.size = size;
4470 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4471 vcpu->arch.pio.count = 0;
4475 vcpu->run->exit_reason = KVM_EXIT_IO;
4476 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4477 vcpu->run->io.size = size;
4478 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4479 vcpu->run->io.count = count;
4480 vcpu->run->io.port = port;
4485 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4486 int size, unsigned short port, void *val,
4489 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4492 if (vcpu->arch.pio.count)
4495 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4498 memcpy(val, vcpu->arch.pio_data, size * count);
4499 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4500 vcpu->arch.pio.count = 0;
4507 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4508 int size, unsigned short port,
4509 const void *val, unsigned int count)
4511 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4513 memcpy(vcpu->arch.pio_data, val, size * count);
4514 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4515 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4518 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4520 return kvm_x86_ops->get_segment_base(vcpu, seg);
4523 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4525 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4528 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4530 if (!need_emulate_wbinvd(vcpu))
4531 return X86EMUL_CONTINUE;
4533 if (kvm_x86_ops->has_wbinvd_exit()) {
4534 int cpu = get_cpu();
4536 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4537 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4538 wbinvd_ipi, NULL, 1);
4540 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4543 return X86EMUL_CONTINUE;
4546 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4548 kvm_x86_ops->skip_emulated_instruction(vcpu);
4549 return kvm_emulate_wbinvd_noskip(vcpu);
4551 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4555 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4557 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4560 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4561 unsigned long *dest)
4563 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4566 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4567 unsigned long value)
4570 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4573 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4575 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4578 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4580 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4581 unsigned long value;
4585 value = kvm_read_cr0(vcpu);
4588 value = vcpu->arch.cr2;
4591 value = kvm_read_cr3(vcpu);
4594 value = kvm_read_cr4(vcpu);
4597 value = kvm_get_cr8(vcpu);
4600 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4607 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4609 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4614 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4617 vcpu->arch.cr2 = val;
4620 res = kvm_set_cr3(vcpu, val);
4623 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4626 res = kvm_set_cr8(vcpu, val);
4629 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4636 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4638 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4641 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4643 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4646 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4648 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4651 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4653 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4656 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4658 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4661 static unsigned long emulator_get_cached_segment_base(
4662 struct x86_emulate_ctxt *ctxt, int seg)
4664 return get_segment_base(emul_to_vcpu(ctxt), seg);
4667 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4668 struct desc_struct *desc, u32 *base3,
4671 struct kvm_segment var;
4673 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4674 *selector = var.selector;
4677 memset(desc, 0, sizeof(*desc));
4683 set_desc_limit(desc, var.limit);
4684 set_desc_base(desc, (unsigned long)var.base);
4685 #ifdef CONFIG_X86_64
4687 *base3 = var.base >> 32;
4689 desc->type = var.type;
4691 desc->dpl = var.dpl;
4692 desc->p = var.present;
4693 desc->avl = var.avl;
4701 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4702 struct desc_struct *desc, u32 base3,
4705 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4706 struct kvm_segment var;
4708 var.selector = selector;
4709 var.base = get_desc_base(desc);
4710 #ifdef CONFIG_X86_64
4711 var.base |= ((u64)base3) << 32;
4713 var.limit = get_desc_limit(desc);
4715 var.limit = (var.limit << 12) | 0xfff;
4716 var.type = desc->type;
4717 var.dpl = desc->dpl;
4722 var.avl = desc->avl;
4723 var.present = desc->p;
4724 var.unusable = !var.present;
4727 kvm_set_segment(vcpu, &var, seg);
4731 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4732 u32 msr_index, u64 *pdata)
4734 struct msr_data msr;
4737 msr.index = msr_index;
4738 msr.host_initiated = false;
4739 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4747 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4748 u32 msr_index, u64 data)
4750 struct msr_data msr;
4753 msr.index = msr_index;
4754 msr.host_initiated = false;
4755 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4758 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4760 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4762 return vcpu->arch.smbase;
4765 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4767 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4769 vcpu->arch.smbase = smbase;
4772 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4775 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4778 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4779 u32 pmc, u64 *pdata)
4781 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4784 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4786 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4789 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4792 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4794 * CR0.TS may reference the host fpu state, not the guest fpu state,
4795 * so it may be clear at this point.
4800 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4805 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4806 struct x86_instruction_info *info,
4807 enum x86_intercept_stage stage)
4809 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4812 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4813 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4815 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4818 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4820 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4823 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4825 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4828 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
4830 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
4833 static const struct x86_emulate_ops emulate_ops = {
4834 .read_gpr = emulator_read_gpr,
4835 .write_gpr = emulator_write_gpr,
4836 .read_std = kvm_read_guest_virt_system,
4837 .write_std = kvm_write_guest_virt_system,
4838 .fetch = kvm_fetch_guest_virt,
4839 .read_emulated = emulator_read_emulated,
4840 .write_emulated = emulator_write_emulated,
4841 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4842 .invlpg = emulator_invlpg,
4843 .pio_in_emulated = emulator_pio_in_emulated,
4844 .pio_out_emulated = emulator_pio_out_emulated,
4845 .get_segment = emulator_get_segment,
4846 .set_segment = emulator_set_segment,
4847 .get_cached_segment_base = emulator_get_cached_segment_base,
4848 .get_gdt = emulator_get_gdt,
4849 .get_idt = emulator_get_idt,
4850 .set_gdt = emulator_set_gdt,
4851 .set_idt = emulator_set_idt,
4852 .get_cr = emulator_get_cr,
4853 .set_cr = emulator_set_cr,
4854 .cpl = emulator_get_cpl,
4855 .get_dr = emulator_get_dr,
4856 .set_dr = emulator_set_dr,
4857 .get_smbase = emulator_get_smbase,
4858 .set_smbase = emulator_set_smbase,
4859 .set_msr = emulator_set_msr,
4860 .get_msr = emulator_get_msr,
4861 .check_pmc = emulator_check_pmc,
4862 .read_pmc = emulator_read_pmc,
4863 .halt = emulator_halt,
4864 .wbinvd = emulator_wbinvd,
4865 .fix_hypercall = emulator_fix_hypercall,
4866 .get_fpu = emulator_get_fpu,
4867 .put_fpu = emulator_put_fpu,
4868 .intercept = emulator_intercept,
4869 .get_cpuid = emulator_get_cpuid,
4870 .set_nmi_mask = emulator_set_nmi_mask,
4873 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4875 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
4877 * an sti; sti; sequence only disable interrupts for the first
4878 * instruction. So, if the last instruction, be it emulated or
4879 * not, left the system with the INT_STI flag enabled, it
4880 * means that the last instruction is an sti. We should not
4881 * leave the flag on in this case. The same goes for mov ss
4883 if (int_shadow & mask)
4885 if (unlikely(int_shadow || mask)) {
4886 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4888 kvm_make_request(KVM_REQ_EVENT, vcpu);
4892 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
4894 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4895 if (ctxt->exception.vector == PF_VECTOR)
4896 return kvm_propagate_fault(vcpu, &ctxt->exception);
4898 if (ctxt->exception.error_code_valid)
4899 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4900 ctxt->exception.error_code);
4902 kvm_queue_exception(vcpu, ctxt->exception.vector);
4906 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4908 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4911 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4913 ctxt->eflags = kvm_get_rflags(vcpu);
4914 ctxt->eip = kvm_rip_read(vcpu);
4915 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4916 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4917 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
4918 cs_db ? X86EMUL_MODE_PROT32 :
4919 X86EMUL_MODE_PROT16;
4920 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
4921 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
4922 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
4923 ctxt->emul_flags = vcpu->arch.hflags;
4925 init_decode_cache(ctxt);
4926 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4929 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4931 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4934 init_emulate_ctxt(vcpu);
4938 ctxt->_eip = ctxt->eip + inc_eip;
4939 ret = emulate_int_real(ctxt, irq);
4941 if (ret != X86EMUL_CONTINUE)
4942 return EMULATE_FAIL;
4944 ctxt->eip = ctxt->_eip;
4945 kvm_rip_write(vcpu, ctxt->eip);
4946 kvm_set_rflags(vcpu, ctxt->eflags);
4948 if (irq == NMI_VECTOR)
4949 vcpu->arch.nmi_pending = 0;
4951 vcpu->arch.interrupt.pending = false;
4953 return EMULATE_DONE;
4955 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4957 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4959 int r = EMULATE_DONE;
4961 ++vcpu->stat.insn_emulation_fail;
4962 trace_kvm_emulate_insn_failed(vcpu);
4963 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
4964 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4965 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4966 vcpu->run->internal.ndata = 0;
4969 kvm_queue_exception(vcpu, UD_VECTOR);
4974 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4975 bool write_fault_to_shadow_pgtable,
4981 if (emulation_type & EMULTYPE_NO_REEXECUTE)
4984 if (!vcpu->arch.mmu.direct_map) {
4986 * Write permission should be allowed since only
4987 * write access need to be emulated.
4989 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4992 * If the mapping is invalid in guest, let cpu retry
4993 * it to generate fault.
4995 if (gpa == UNMAPPED_GVA)
5000 * Do not retry the unhandleable instruction if it faults on the
5001 * readonly host memory, otherwise it will goto a infinite loop:
5002 * retry instruction -> write #PF -> emulation fail -> retry
5003 * instruction -> ...
5005 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5008 * If the instruction failed on the error pfn, it can not be fixed,
5009 * report the error to userspace.
5011 if (is_error_noslot_pfn(pfn))
5014 kvm_release_pfn_clean(pfn);
5016 /* The instructions are well-emulated on direct mmu. */
5017 if (vcpu->arch.mmu.direct_map) {
5018 unsigned int indirect_shadow_pages;
5020 spin_lock(&vcpu->kvm->mmu_lock);
5021 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5022 spin_unlock(&vcpu->kvm->mmu_lock);
5024 if (indirect_shadow_pages)
5025 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5031 * if emulation was due to access to shadowed page table
5032 * and it failed try to unshadow page and re-enter the
5033 * guest to let CPU execute the instruction.
5035 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5038 * If the access faults on its page table, it can not
5039 * be fixed by unprotecting shadow page and it should
5040 * be reported to userspace.
5042 return !write_fault_to_shadow_pgtable;
5045 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5046 unsigned long cr2, int emulation_type)
5048 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5049 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5051 last_retry_eip = vcpu->arch.last_retry_eip;
5052 last_retry_addr = vcpu->arch.last_retry_addr;
5055 * If the emulation is caused by #PF and it is non-page_table
5056 * writing instruction, it means the VM-EXIT is caused by shadow
5057 * page protected, we can zap the shadow page and retry this
5058 * instruction directly.
5060 * Note: if the guest uses a non-page-table modifying instruction
5061 * on the PDE that points to the instruction, then we will unmap
5062 * the instruction and go to an infinite loop. So, we cache the
5063 * last retried eip and the last fault address, if we meet the eip
5064 * and the address again, we can break out of the potential infinite
5067 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5069 if (!(emulation_type & EMULTYPE_RETRY))
5072 if (x86_page_table_writing_insn(ctxt))
5075 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5078 vcpu->arch.last_retry_eip = ctxt->eip;
5079 vcpu->arch.last_retry_addr = cr2;
5081 if (!vcpu->arch.mmu.direct_map)
5082 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5084 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5089 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5090 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5092 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5094 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5095 /* This is a good place to trace that we are exiting SMM. */
5096 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5098 if (unlikely(vcpu->arch.smi_pending)) {
5099 kvm_make_request(KVM_REQ_SMI, vcpu);
5100 vcpu->arch.smi_pending = 0;
5102 /* Process a latched INIT, if any. */
5103 kvm_make_request(KVM_REQ_EVENT, vcpu);
5107 kvm_mmu_reset_context(vcpu);
5110 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5112 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5114 vcpu->arch.hflags = emul_flags;
5116 if (changed & HF_SMM_MASK)
5117 kvm_smm_changed(vcpu);
5120 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5129 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5130 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5135 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5137 struct kvm_run *kvm_run = vcpu->run;
5140 * rflags is the old, "raw" value of the flags. The new value has
5141 * not been saved yet.
5143 * This is correct even for TF set by the guest, because "the
5144 * processor will not generate this exception after the instruction
5145 * that sets the TF flag".
5147 if (unlikely(rflags & X86_EFLAGS_TF)) {
5148 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5149 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5151 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5152 kvm_run->debug.arch.exception = DB_VECTOR;
5153 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5154 *r = EMULATE_USER_EXIT;
5156 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5158 * "Certain debug exceptions may clear bit 0-3. The
5159 * remaining contents of the DR6 register are never
5160 * cleared by the processor".
5162 vcpu->arch.dr6 &= ~15;
5163 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5164 kvm_queue_exception(vcpu, DB_VECTOR);
5169 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5171 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5172 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5173 struct kvm_run *kvm_run = vcpu->run;
5174 unsigned long eip = kvm_get_linear_rip(vcpu);
5175 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5176 vcpu->arch.guest_debug_dr7,
5180 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5181 kvm_run->debug.arch.pc = eip;
5182 kvm_run->debug.arch.exception = DB_VECTOR;
5183 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5184 *r = EMULATE_USER_EXIT;
5189 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5190 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5191 unsigned long eip = kvm_get_linear_rip(vcpu);
5192 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5197 vcpu->arch.dr6 &= ~15;
5198 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5199 kvm_queue_exception(vcpu, DB_VECTOR);
5208 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5215 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5216 bool writeback = true;
5217 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5220 * Clear write_fault_to_shadow_pgtable here to ensure it is
5223 vcpu->arch.write_fault_to_shadow_pgtable = false;
5224 kvm_clear_exception_queue(vcpu);
5226 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5227 init_emulate_ctxt(vcpu);
5230 * We will reenter on the same instruction since
5231 * we do not set complete_userspace_io. This does not
5232 * handle watchpoints yet, those would be handled in
5235 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5238 ctxt->interruptibility = 0;
5239 ctxt->have_exception = false;
5240 ctxt->exception.vector = -1;
5241 ctxt->perm_ok = false;
5243 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5245 r = x86_decode_insn(ctxt, insn, insn_len);
5247 trace_kvm_emulate_insn_start(vcpu);
5248 ++vcpu->stat.insn_emulation;
5249 if (r != EMULATION_OK) {
5250 if (emulation_type & EMULTYPE_TRAP_UD)
5251 return EMULATE_FAIL;
5252 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5254 return EMULATE_DONE;
5255 if (emulation_type & EMULTYPE_SKIP)
5256 return EMULATE_FAIL;
5257 return handle_emulation_failure(vcpu);
5261 if (emulation_type & EMULTYPE_SKIP) {
5262 kvm_rip_write(vcpu, ctxt->_eip);
5263 if (ctxt->eflags & X86_EFLAGS_RF)
5264 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5265 return EMULATE_DONE;
5268 if (retry_instruction(ctxt, cr2, emulation_type))
5269 return EMULATE_DONE;
5271 /* this is needed for vmware backdoor interface to work since it
5272 changes registers values during IO operation */
5273 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5274 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5275 emulator_invalidate_register_cache(ctxt);
5279 r = x86_emulate_insn(ctxt);
5281 if (r == EMULATION_INTERCEPTED)
5282 return EMULATE_DONE;
5284 if (r == EMULATION_FAILED) {
5285 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5287 return EMULATE_DONE;
5289 return handle_emulation_failure(vcpu);
5292 if (ctxt->have_exception) {
5294 if (inject_emulated_exception(vcpu))
5296 } else if (vcpu->arch.pio.count) {
5297 if (!vcpu->arch.pio.in) {
5298 /* FIXME: return into emulator if single-stepping. */
5299 vcpu->arch.pio.count = 0;
5302 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5304 r = EMULATE_USER_EXIT;
5305 } else if (vcpu->mmio_needed) {
5306 if (!vcpu->mmio_is_write)
5308 r = EMULATE_USER_EXIT;
5309 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5310 } else if (r == EMULATION_RESTART)
5316 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5317 toggle_interruptibility(vcpu, ctxt->interruptibility);
5318 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5319 if (vcpu->arch.hflags != ctxt->emul_flags)
5320 kvm_set_hflags(vcpu, ctxt->emul_flags);
5321 kvm_rip_write(vcpu, ctxt->eip);
5322 if (r == EMULATE_DONE)
5323 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5324 if (!ctxt->have_exception ||
5325 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5326 __kvm_set_rflags(vcpu, ctxt->eflags);
5329 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5330 * do nothing, and it will be requested again as soon as
5331 * the shadow expires. But we still need to check here,
5332 * because POPF has no interrupt shadow.
5334 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5335 kvm_make_request(KVM_REQ_EVENT, vcpu);
5337 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5341 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5343 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5345 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5346 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5347 size, port, &val, 1);
5348 /* do not return to emulator after return from userspace */
5349 vcpu->arch.pio.count = 0;
5352 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5354 static void tsc_bad(void *info)
5356 __this_cpu_write(cpu_tsc_khz, 0);
5359 static void tsc_khz_changed(void *data)
5361 struct cpufreq_freqs *freq = data;
5362 unsigned long khz = 0;
5366 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5367 khz = cpufreq_quick_get(raw_smp_processor_id());
5370 __this_cpu_write(cpu_tsc_khz, khz);
5373 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5376 struct cpufreq_freqs *freq = data;
5378 struct kvm_vcpu *vcpu;
5379 int i, send_ipi = 0;
5382 * We allow guests to temporarily run on slowing clocks,
5383 * provided we notify them after, or to run on accelerating
5384 * clocks, provided we notify them before. Thus time never
5387 * However, we have a problem. We can't atomically update
5388 * the frequency of a given CPU from this function; it is
5389 * merely a notifier, which can be called from any CPU.
5390 * Changing the TSC frequency at arbitrary points in time
5391 * requires a recomputation of local variables related to
5392 * the TSC for each VCPU. We must flag these local variables
5393 * to be updated and be sure the update takes place with the
5394 * new frequency before any guests proceed.
5396 * Unfortunately, the combination of hotplug CPU and frequency
5397 * change creates an intractable locking scenario; the order
5398 * of when these callouts happen is undefined with respect to
5399 * CPU hotplug, and they can race with each other. As such,
5400 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5401 * undefined; you can actually have a CPU frequency change take
5402 * place in between the computation of X and the setting of the
5403 * variable. To protect against this problem, all updates of
5404 * the per_cpu tsc_khz variable are done in an interrupt
5405 * protected IPI, and all callers wishing to update the value
5406 * must wait for a synchronous IPI to complete (which is trivial
5407 * if the caller is on the CPU already). This establishes the
5408 * necessary total order on variable updates.
5410 * Note that because a guest time update may take place
5411 * anytime after the setting of the VCPU's request bit, the
5412 * correct TSC value must be set before the request. However,
5413 * to ensure the update actually makes it to any guest which
5414 * starts running in hardware virtualization between the set
5415 * and the acquisition of the spinlock, we must also ping the
5416 * CPU after setting the request bit.
5420 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5422 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5425 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5427 spin_lock(&kvm_lock);
5428 list_for_each_entry(kvm, &vm_list, vm_list) {
5429 kvm_for_each_vcpu(i, vcpu, kvm) {
5430 if (vcpu->cpu != freq->cpu)
5432 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5433 if (vcpu->cpu != smp_processor_id())
5437 spin_unlock(&kvm_lock);
5439 if (freq->old < freq->new && send_ipi) {
5441 * We upscale the frequency. Must make the guest
5442 * doesn't see old kvmclock values while running with
5443 * the new frequency, otherwise we risk the guest sees
5444 * time go backwards.
5446 * In case we update the frequency for another cpu
5447 * (which might be in guest context) send an interrupt
5448 * to kick the cpu out of guest context. Next time
5449 * guest context is entered kvmclock will be updated,
5450 * so the guest will not see stale values.
5452 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5457 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5458 .notifier_call = kvmclock_cpufreq_notifier
5461 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5462 unsigned long action, void *hcpu)
5464 unsigned int cpu = (unsigned long)hcpu;
5468 case CPU_DOWN_FAILED:
5469 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5471 case CPU_DOWN_PREPARE:
5472 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5478 static struct notifier_block kvmclock_cpu_notifier_block = {
5479 .notifier_call = kvmclock_cpu_notifier,
5480 .priority = -INT_MAX
5483 static void kvm_timer_init(void)
5487 max_tsc_khz = tsc_khz;
5489 cpu_notifier_register_begin();
5490 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5491 #ifdef CONFIG_CPU_FREQ
5492 struct cpufreq_policy policy;
5493 memset(&policy, 0, sizeof(policy));
5495 cpufreq_get_policy(&policy, cpu);
5496 if (policy.cpuinfo.max_freq)
5497 max_tsc_khz = policy.cpuinfo.max_freq;
5500 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5501 CPUFREQ_TRANSITION_NOTIFIER);
5503 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5504 for_each_online_cpu(cpu)
5505 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5507 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5508 cpu_notifier_register_done();
5512 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5514 int kvm_is_in_guest(void)
5516 return __this_cpu_read(current_vcpu) != NULL;
5519 static int kvm_is_user_mode(void)
5523 if (__this_cpu_read(current_vcpu))
5524 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5526 return user_mode != 0;
5529 static unsigned long kvm_get_guest_ip(void)
5531 unsigned long ip = 0;
5533 if (__this_cpu_read(current_vcpu))
5534 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5539 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5540 .is_in_guest = kvm_is_in_guest,
5541 .is_user_mode = kvm_is_user_mode,
5542 .get_guest_ip = kvm_get_guest_ip,
5545 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5547 __this_cpu_write(current_vcpu, vcpu);
5549 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5551 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5553 __this_cpu_write(current_vcpu, NULL);
5555 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5557 static void kvm_set_mmio_spte_mask(void)
5560 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5563 * Set the reserved bits and the present bit of an paging-structure
5564 * entry to generate page fault with PFER.RSV = 1.
5566 /* Mask the reserved physical address bits. */
5567 mask = rsvd_bits(maxphyaddr, 51);
5569 /* Bit 62 is always reserved for 32bit host. */
5570 mask |= 0x3ull << 62;
5572 /* Set the present bit. */
5575 #ifdef CONFIG_X86_64
5577 * If reserved bit is not supported, clear the present bit to disable
5580 if (maxphyaddr == 52)
5584 kvm_mmu_set_mmio_spte_mask(mask);
5587 #ifdef CONFIG_X86_64
5588 static void pvclock_gtod_update_fn(struct work_struct *work)
5592 struct kvm_vcpu *vcpu;
5595 spin_lock(&kvm_lock);
5596 list_for_each_entry(kvm, &vm_list, vm_list)
5597 kvm_for_each_vcpu(i, vcpu, kvm)
5598 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5599 atomic_set(&kvm_guest_has_master_clock, 0);
5600 spin_unlock(&kvm_lock);
5603 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5606 * Notification about pvclock gtod data update.
5608 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5611 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5612 struct timekeeper *tk = priv;
5614 update_pvclock_gtod(tk);
5616 /* disable master clock if host does not trust, or does not
5617 * use, TSC clocksource
5619 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5620 atomic_read(&kvm_guest_has_master_clock) != 0)
5621 queue_work(system_long_wq, &pvclock_gtod_work);
5626 static struct notifier_block pvclock_gtod_notifier = {
5627 .notifier_call = pvclock_gtod_notify,
5631 int kvm_arch_init(void *opaque)
5634 struct kvm_x86_ops *ops = opaque;
5637 printk(KERN_ERR "kvm: already loaded the other module\n");
5642 if (!ops->cpu_has_kvm_support()) {
5643 printk(KERN_ERR "kvm: no hardware support\n");
5647 if (ops->disabled_by_bios()) {
5648 printk(KERN_ERR "kvm: disabled by bios\n");
5654 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5656 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5660 r = kvm_mmu_module_init();
5662 goto out_free_percpu;
5664 kvm_set_mmio_spte_mask();
5668 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5669 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5673 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5676 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5679 #ifdef CONFIG_X86_64
5680 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5686 free_percpu(shared_msrs);
5691 void kvm_arch_exit(void)
5693 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5695 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5696 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5697 CPUFREQ_TRANSITION_NOTIFIER);
5698 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5699 #ifdef CONFIG_X86_64
5700 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5703 kvm_mmu_module_exit();
5704 free_percpu(shared_msrs);
5707 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5709 ++vcpu->stat.halt_exits;
5710 if (lapic_in_kernel(vcpu)) {
5711 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5714 vcpu->run->exit_reason = KVM_EXIT_HLT;
5718 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5720 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5722 kvm_x86_ops->skip_emulated_instruction(vcpu);
5723 return kvm_vcpu_halt(vcpu);
5725 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5728 * kvm_pv_kick_cpu_op: Kick a vcpu.
5730 * @apicid - apicid of vcpu to be kicked.
5732 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5734 struct kvm_lapic_irq lapic_irq;
5736 lapic_irq.shorthand = 0;
5737 lapic_irq.dest_mode = 0;
5738 lapic_irq.dest_id = apicid;
5739 lapic_irq.msi_redir_hint = false;
5741 lapic_irq.delivery_mode = APIC_DM_REMRD;
5742 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5745 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5747 unsigned long nr, a0, a1, a2, a3, ret;
5748 int op_64_bit, r = 1;
5750 kvm_x86_ops->skip_emulated_instruction(vcpu);
5752 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5753 return kvm_hv_hypercall(vcpu);
5755 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5756 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5757 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5758 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5759 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5761 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5763 op_64_bit = is_64_bit_mode(vcpu);
5772 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5778 case KVM_HC_VAPIC_POLL_IRQ:
5781 case KVM_HC_KICK_CPU:
5782 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5792 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5793 ++vcpu->stat.hypercalls;
5796 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5798 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5800 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5801 char instruction[3];
5802 unsigned long rip = kvm_rip_read(vcpu);
5804 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5806 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5810 * Check if userspace requested an interrupt window, and that the
5811 * interrupt window is open.
5813 * No need to exit to userspace if we already have an interrupt queued.
5815 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5817 if (!vcpu->run->request_interrupt_window || pic_in_kernel(vcpu->kvm))
5820 if (kvm_cpu_has_interrupt(vcpu))
5823 return (irqchip_split(vcpu->kvm)
5824 ? kvm_apic_accept_pic_intr(vcpu)
5825 : kvm_arch_interrupt_allowed(vcpu));
5828 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5830 struct kvm_run *kvm_run = vcpu->run;
5832 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5833 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
5834 kvm_run->cr8 = kvm_get_cr8(vcpu);
5835 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5836 if (!irqchip_in_kernel(vcpu->kvm))
5837 kvm_run->ready_for_interrupt_injection =
5838 kvm_arch_interrupt_allowed(vcpu) &&
5839 !kvm_cpu_has_interrupt(vcpu) &&
5840 !kvm_event_needs_reinjection(vcpu);
5841 else if (!pic_in_kernel(vcpu->kvm))
5842 kvm_run->ready_for_interrupt_injection =
5843 kvm_apic_accept_pic_intr(vcpu) &&
5844 !kvm_cpu_has_interrupt(vcpu);
5846 kvm_run->ready_for_interrupt_injection = 1;
5849 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5853 if (!kvm_x86_ops->update_cr8_intercept)
5856 if (!vcpu->arch.apic)
5859 if (!vcpu->arch.apic->vapic_addr)
5860 max_irr = kvm_lapic_find_highest_irr(vcpu);
5867 tpr = kvm_lapic_get_cr8(vcpu);
5869 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5872 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
5876 /* try to reinject previous events if any */
5877 if (vcpu->arch.exception.pending) {
5878 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5879 vcpu->arch.exception.has_error_code,
5880 vcpu->arch.exception.error_code);
5882 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
5883 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
5886 if (vcpu->arch.exception.nr == DB_VECTOR &&
5887 (vcpu->arch.dr7 & DR7_GD)) {
5888 vcpu->arch.dr7 &= ~DR7_GD;
5889 kvm_update_dr7(vcpu);
5892 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5893 vcpu->arch.exception.has_error_code,
5894 vcpu->arch.exception.error_code,
5895 vcpu->arch.exception.reinject);
5899 if (vcpu->arch.nmi_injected) {
5900 kvm_x86_ops->set_nmi(vcpu);
5904 if (vcpu->arch.interrupt.pending) {
5905 kvm_x86_ops->set_irq(vcpu);
5909 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5910 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5915 /* try to inject new event if pending */
5916 if (vcpu->arch.nmi_pending) {
5917 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5918 --vcpu->arch.nmi_pending;
5919 vcpu->arch.nmi_injected = true;
5920 kvm_x86_ops->set_nmi(vcpu);
5922 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
5924 * Because interrupts can be injected asynchronously, we are
5925 * calling check_nested_events again here to avoid a race condition.
5926 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
5927 * proposal and current concerns. Perhaps we should be setting
5928 * KVM_REQ_EVENT only on certain events and not unconditionally?
5930 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5931 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5935 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5936 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5938 kvm_x86_ops->set_irq(vcpu);
5944 static void process_nmi(struct kvm_vcpu *vcpu)
5949 * x86 is limited to one NMI running, and one NMI pending after it.
5950 * If an NMI is already in progress, limit further NMIs to just one.
5951 * Otherwise, allow two (and we'll inject the first one immediately).
5953 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5956 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5957 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5958 kvm_make_request(KVM_REQ_EVENT, vcpu);
5961 #define put_smstate(type, buf, offset, val) \
5962 *(type *)((buf) + (offset) - 0x7e00) = val
5964 static u32 process_smi_get_segment_flags(struct kvm_segment *seg)
5967 flags |= seg->g << 23;
5968 flags |= seg->db << 22;
5969 flags |= seg->l << 21;
5970 flags |= seg->avl << 20;
5971 flags |= seg->present << 15;
5972 flags |= seg->dpl << 13;
5973 flags |= seg->s << 12;
5974 flags |= seg->type << 8;
5978 static void process_smi_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
5980 struct kvm_segment seg;
5983 kvm_get_segment(vcpu, &seg, n);
5984 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
5987 offset = 0x7f84 + n * 12;
5989 offset = 0x7f2c + (n - 3) * 12;
5991 put_smstate(u32, buf, offset + 8, seg.base);
5992 put_smstate(u32, buf, offset + 4, seg.limit);
5993 put_smstate(u32, buf, offset, process_smi_get_segment_flags(&seg));
5996 #ifdef CONFIG_X86_64
5997 static void process_smi_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
5999 struct kvm_segment seg;
6003 kvm_get_segment(vcpu, &seg, n);
6004 offset = 0x7e00 + n * 16;
6006 flags = process_smi_get_segment_flags(&seg) >> 8;
6007 put_smstate(u16, buf, offset, seg.selector);
6008 put_smstate(u16, buf, offset + 2, flags);
6009 put_smstate(u32, buf, offset + 4, seg.limit);
6010 put_smstate(u64, buf, offset + 8, seg.base);
6014 static void process_smi_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6017 struct kvm_segment seg;
6021 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6022 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6023 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6024 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6026 for (i = 0; i < 8; i++)
6027 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6029 kvm_get_dr(vcpu, 6, &val);
6030 put_smstate(u32, buf, 0x7fcc, (u32)val);
6031 kvm_get_dr(vcpu, 7, &val);
6032 put_smstate(u32, buf, 0x7fc8, (u32)val);
6034 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6035 put_smstate(u32, buf, 0x7fc4, seg.selector);
6036 put_smstate(u32, buf, 0x7f64, seg.base);
6037 put_smstate(u32, buf, 0x7f60, seg.limit);
6038 put_smstate(u32, buf, 0x7f5c, process_smi_get_segment_flags(&seg));
6040 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6041 put_smstate(u32, buf, 0x7fc0, seg.selector);
6042 put_smstate(u32, buf, 0x7f80, seg.base);
6043 put_smstate(u32, buf, 0x7f7c, seg.limit);
6044 put_smstate(u32, buf, 0x7f78, process_smi_get_segment_flags(&seg));
6046 kvm_x86_ops->get_gdt(vcpu, &dt);
6047 put_smstate(u32, buf, 0x7f74, dt.address);
6048 put_smstate(u32, buf, 0x7f70, dt.size);
6050 kvm_x86_ops->get_idt(vcpu, &dt);
6051 put_smstate(u32, buf, 0x7f58, dt.address);
6052 put_smstate(u32, buf, 0x7f54, dt.size);
6054 for (i = 0; i < 6; i++)
6055 process_smi_save_seg_32(vcpu, buf, i);
6057 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6060 put_smstate(u32, buf, 0x7efc, 0x00020000);
6061 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6064 static void process_smi_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6066 #ifdef CONFIG_X86_64
6068 struct kvm_segment seg;
6072 for (i = 0; i < 16; i++)
6073 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6075 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6076 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6078 kvm_get_dr(vcpu, 6, &val);
6079 put_smstate(u64, buf, 0x7f68, val);
6080 kvm_get_dr(vcpu, 7, &val);
6081 put_smstate(u64, buf, 0x7f60, val);
6083 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6084 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6085 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6087 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6090 put_smstate(u32, buf, 0x7efc, 0x00020064);
6092 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6094 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6095 put_smstate(u16, buf, 0x7e90, seg.selector);
6096 put_smstate(u16, buf, 0x7e92, process_smi_get_segment_flags(&seg) >> 8);
6097 put_smstate(u32, buf, 0x7e94, seg.limit);
6098 put_smstate(u64, buf, 0x7e98, seg.base);
6100 kvm_x86_ops->get_idt(vcpu, &dt);
6101 put_smstate(u32, buf, 0x7e84, dt.size);
6102 put_smstate(u64, buf, 0x7e88, dt.address);
6104 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6105 put_smstate(u16, buf, 0x7e70, seg.selector);
6106 put_smstate(u16, buf, 0x7e72, process_smi_get_segment_flags(&seg) >> 8);
6107 put_smstate(u32, buf, 0x7e74, seg.limit);
6108 put_smstate(u64, buf, 0x7e78, seg.base);
6110 kvm_x86_ops->get_gdt(vcpu, &dt);
6111 put_smstate(u32, buf, 0x7e64, dt.size);
6112 put_smstate(u64, buf, 0x7e68, dt.address);
6114 for (i = 0; i < 6; i++)
6115 process_smi_save_seg_64(vcpu, buf, i);
6121 static void process_smi(struct kvm_vcpu *vcpu)
6123 struct kvm_segment cs, ds;
6129 vcpu->arch.smi_pending = true;
6133 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6134 vcpu->arch.hflags |= HF_SMM_MASK;
6135 memset(buf, 0, 512);
6136 if (guest_cpuid_has_longmode(vcpu))
6137 process_smi_save_state_64(vcpu, buf);
6139 process_smi_save_state_32(vcpu, buf);
6141 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6143 if (kvm_x86_ops->get_nmi_mask(vcpu))
6144 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6146 kvm_x86_ops->set_nmi_mask(vcpu, true);
6148 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6149 kvm_rip_write(vcpu, 0x8000);
6151 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6152 kvm_x86_ops->set_cr0(vcpu, cr0);
6153 vcpu->arch.cr0 = cr0;
6155 kvm_x86_ops->set_cr4(vcpu, 0);
6157 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6158 dt.address = dt.size = 0;
6159 kvm_x86_ops->set_idt(vcpu, &dt);
6161 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6163 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6164 cs.base = vcpu->arch.smbase;
6169 cs.limit = ds.limit = 0xffffffff;
6170 cs.type = ds.type = 0x3;
6171 cs.dpl = ds.dpl = 0;
6176 cs.avl = ds.avl = 0;
6177 cs.present = ds.present = 1;
6178 cs.unusable = ds.unusable = 0;
6179 cs.padding = ds.padding = 0;
6181 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6182 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6183 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6184 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6185 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6186 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6188 if (guest_cpuid_has_longmode(vcpu))
6189 kvm_x86_ops->set_efer(vcpu, 0);
6191 kvm_update_cpuid(vcpu);
6192 kvm_mmu_reset_context(vcpu);
6195 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6197 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6200 memset(vcpu->arch.eoi_exit_bitmap, 0, 256 / 8);
6202 if (irqchip_split(vcpu->kvm))
6203 kvm_scan_ioapic_routes(vcpu, vcpu->arch.eoi_exit_bitmap);
6205 kvm_ioapic_scan_entry(vcpu, vcpu->arch.eoi_exit_bitmap);
6206 kvm_x86_ops->load_eoi_exitmap(vcpu);
6209 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6211 ++vcpu->stat.tlb_flush;
6212 kvm_x86_ops->tlb_flush(vcpu);
6215 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6217 struct page *page = NULL;
6219 if (!lapic_in_kernel(vcpu))
6222 if (!kvm_x86_ops->set_apic_access_page_addr)
6225 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6226 if (is_error_page(page))
6228 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6231 * Do not pin apic access page in memory, the MMU notifier
6232 * will call us again if it is migrated or swapped out.
6236 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6238 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6239 unsigned long address)
6242 * The physical address of apic access page is stored in the VMCS.
6243 * Update it when it becomes invalid.
6245 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6246 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6250 * Returns 1 to let vcpu_run() continue the guest execution loop without
6251 * exiting to the userspace. Otherwise, the value will be returned to the
6254 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6257 bool req_int_win = !lapic_in_kernel(vcpu) &&
6258 vcpu->run->request_interrupt_window;
6259 bool req_immediate_exit = false;
6261 if (vcpu->requests) {
6262 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6263 kvm_mmu_unload(vcpu);
6264 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6265 __kvm_migrate_timers(vcpu);
6266 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6267 kvm_gen_update_masterclock(vcpu->kvm);
6268 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6269 kvm_gen_kvmclock_update(vcpu);
6270 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6271 r = kvm_guest_time_update(vcpu);
6275 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6276 kvm_mmu_sync_roots(vcpu);
6277 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6278 kvm_vcpu_flush_tlb(vcpu);
6279 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6280 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6284 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6285 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6289 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6290 vcpu->fpu_active = 0;
6291 kvm_x86_ops->fpu_deactivate(vcpu);
6293 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6294 /* Page is swapped out. Do synthetic halt */
6295 vcpu->arch.apf.halted = true;
6299 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6300 record_steal_time(vcpu);
6301 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6303 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6305 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6306 kvm_pmu_handle_event(vcpu);
6307 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6308 kvm_pmu_deliver_pmi(vcpu);
6309 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6310 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6311 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6312 (void *) vcpu->arch.eoi_exit_bitmap)) {
6313 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6314 vcpu->run->eoi.vector =
6315 vcpu->arch.pending_ioapic_eoi;
6320 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6321 vcpu_scan_ioapic(vcpu);
6322 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6323 kvm_vcpu_reload_apic_access_page(vcpu);
6324 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6325 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6326 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6330 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6331 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6332 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6339 * KVM_REQ_EVENT is not set when posted interrupts are set by
6340 * VT-d hardware, so we have to update RVI unconditionally.
6342 if (kvm_lapic_enabled(vcpu)) {
6344 * Update architecture specific hints for APIC
6345 * virtual interrupt delivery.
6347 if (kvm_x86_ops->hwapic_irr_update)
6348 kvm_x86_ops->hwapic_irr_update(vcpu,
6349 kvm_lapic_find_highest_irr(vcpu));
6352 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6353 kvm_apic_accept_events(vcpu);
6354 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6359 if (inject_pending_event(vcpu, req_int_win) != 0)
6360 req_immediate_exit = true;
6361 /* enable NMI/IRQ window open exits if needed */
6362 else if (vcpu->arch.nmi_pending)
6363 kvm_x86_ops->enable_nmi_window(vcpu);
6364 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6365 kvm_x86_ops->enable_irq_window(vcpu);
6367 if (kvm_lapic_enabled(vcpu)) {
6368 update_cr8_intercept(vcpu);
6369 kvm_lapic_sync_to_vapic(vcpu);
6373 r = kvm_mmu_reload(vcpu);
6375 goto cancel_injection;
6380 kvm_x86_ops->prepare_guest_switch(vcpu);
6381 if (vcpu->fpu_active)
6382 kvm_load_guest_fpu(vcpu);
6383 kvm_load_guest_xcr0(vcpu);
6385 vcpu->mode = IN_GUEST_MODE;
6387 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6389 /* We should set ->mode before check ->requests,
6390 * see the comment in make_all_cpus_request.
6392 smp_mb__after_srcu_read_unlock();
6394 local_irq_disable();
6396 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6397 || need_resched() || signal_pending(current)) {
6398 vcpu->mode = OUTSIDE_GUEST_MODE;
6402 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6404 goto cancel_injection;
6407 if (req_immediate_exit)
6408 smp_send_reschedule(vcpu->cpu);
6410 __kvm_guest_enter();
6412 if (unlikely(vcpu->arch.switch_db_regs)) {
6414 set_debugreg(vcpu->arch.eff_db[0], 0);
6415 set_debugreg(vcpu->arch.eff_db[1], 1);
6416 set_debugreg(vcpu->arch.eff_db[2], 2);
6417 set_debugreg(vcpu->arch.eff_db[3], 3);
6418 set_debugreg(vcpu->arch.dr6, 6);
6419 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6422 trace_kvm_entry(vcpu->vcpu_id);
6423 wait_lapic_expire(vcpu);
6424 kvm_x86_ops->run(vcpu);
6427 * Do this here before restoring debug registers on the host. And
6428 * since we do this before handling the vmexit, a DR access vmexit
6429 * can (a) read the correct value of the debug registers, (b) set
6430 * KVM_DEBUGREG_WONT_EXIT again.
6432 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6435 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6436 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6437 for (i = 0; i < KVM_NR_DB_REGS; i++)
6438 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6442 * If the guest has used debug registers, at least dr7
6443 * will be disabled while returning to the host.
6444 * If we don't have active breakpoints in the host, we don't
6445 * care about the messed up debug address registers. But if
6446 * we have some of them active, restore the old state.
6448 if (hw_breakpoint_active())
6449 hw_breakpoint_restore();
6451 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6454 vcpu->mode = OUTSIDE_GUEST_MODE;
6457 /* Interrupt is enabled by handle_external_intr() */
6458 kvm_x86_ops->handle_external_intr(vcpu);
6463 * We must have an instruction between local_irq_enable() and
6464 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6465 * the interrupt shadow. The stat.exits increment will do nicely.
6466 * But we need to prevent reordering, hence this barrier():
6474 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6477 * Profile KVM exit RIPs:
6479 if (unlikely(prof_on == KVM_PROFILING)) {
6480 unsigned long rip = kvm_rip_read(vcpu);
6481 profile_hit(KVM_PROFILING, (void *)rip);
6484 if (unlikely(vcpu->arch.tsc_always_catchup))
6485 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6487 if (vcpu->arch.apic_attention)
6488 kvm_lapic_sync_from_vapic(vcpu);
6490 r = kvm_x86_ops->handle_exit(vcpu);
6494 kvm_x86_ops->cancel_injection(vcpu);
6495 if (unlikely(vcpu->arch.apic_attention))
6496 kvm_lapic_sync_from_vapic(vcpu);
6501 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6503 if (!kvm_arch_vcpu_runnable(vcpu) &&
6504 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6505 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6506 kvm_vcpu_block(vcpu);
6507 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6509 if (kvm_x86_ops->post_block)
6510 kvm_x86_ops->post_block(vcpu);
6512 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6516 kvm_apic_accept_events(vcpu);
6517 switch(vcpu->arch.mp_state) {
6518 case KVM_MP_STATE_HALTED:
6519 vcpu->arch.pv.pv_unhalted = false;
6520 vcpu->arch.mp_state =
6521 KVM_MP_STATE_RUNNABLE;
6522 case KVM_MP_STATE_RUNNABLE:
6523 vcpu->arch.apf.halted = false;
6525 case KVM_MP_STATE_INIT_RECEIVED:
6534 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6536 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6537 !vcpu->arch.apf.halted);
6540 static int vcpu_run(struct kvm_vcpu *vcpu)
6543 struct kvm *kvm = vcpu->kvm;
6545 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6548 if (kvm_vcpu_running(vcpu)) {
6549 r = vcpu_enter_guest(vcpu);
6551 r = vcpu_block(kvm, vcpu);
6557 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6558 if (kvm_cpu_has_pending_timer(vcpu))
6559 kvm_inject_pending_timer_irqs(vcpu);
6561 if (dm_request_for_irq_injection(vcpu)) {
6563 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6564 ++vcpu->stat.request_irq_exits;
6568 kvm_check_async_pf_completion(vcpu);
6570 if (signal_pending(current)) {
6572 vcpu->run->exit_reason = KVM_EXIT_INTR;
6573 ++vcpu->stat.signal_exits;
6576 if (need_resched()) {
6577 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6579 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6583 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6588 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6591 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6592 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6593 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6594 if (r != EMULATE_DONE)
6599 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6601 BUG_ON(!vcpu->arch.pio.count);
6603 return complete_emulated_io(vcpu);
6607 * Implements the following, as a state machine:
6611 * for each mmio piece in the fragment
6619 * for each mmio piece in the fragment
6624 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6626 struct kvm_run *run = vcpu->run;
6627 struct kvm_mmio_fragment *frag;
6630 BUG_ON(!vcpu->mmio_needed);
6632 /* Complete previous fragment */
6633 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6634 len = min(8u, frag->len);
6635 if (!vcpu->mmio_is_write)
6636 memcpy(frag->data, run->mmio.data, len);
6638 if (frag->len <= 8) {
6639 /* Switch to the next fragment. */
6641 vcpu->mmio_cur_fragment++;
6643 /* Go forward to the next mmio piece. */
6649 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6650 vcpu->mmio_needed = 0;
6652 /* FIXME: return into emulator if single-stepping. */
6653 if (vcpu->mmio_is_write)
6655 vcpu->mmio_read_completed = 1;
6656 return complete_emulated_io(vcpu);
6659 run->exit_reason = KVM_EXIT_MMIO;
6660 run->mmio.phys_addr = frag->gpa;
6661 if (vcpu->mmio_is_write)
6662 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6663 run->mmio.len = min(8u, frag->len);
6664 run->mmio.is_write = vcpu->mmio_is_write;
6665 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6670 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6672 struct fpu *fpu = ¤t->thread.fpu;
6676 fpu__activate_curr(fpu);
6678 if (vcpu->sigset_active)
6679 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6681 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6682 kvm_vcpu_block(vcpu);
6683 kvm_apic_accept_events(vcpu);
6684 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6689 /* re-sync apic's tpr */
6690 if (!lapic_in_kernel(vcpu)) {
6691 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6697 if (unlikely(vcpu->arch.complete_userspace_io)) {
6698 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6699 vcpu->arch.complete_userspace_io = NULL;
6704 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6709 post_kvm_run_save(vcpu);
6710 if (vcpu->sigset_active)
6711 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6716 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6718 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6720 * We are here if userspace calls get_regs() in the middle of
6721 * instruction emulation. Registers state needs to be copied
6722 * back from emulation context to vcpu. Userspace shouldn't do
6723 * that usually, but some bad designed PV devices (vmware
6724 * backdoor interface) need this to work
6726 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6727 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6729 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6730 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6731 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6732 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6733 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6734 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6735 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6736 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6737 #ifdef CONFIG_X86_64
6738 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6739 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6740 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6741 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6742 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6743 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6744 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6745 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6748 regs->rip = kvm_rip_read(vcpu);
6749 regs->rflags = kvm_get_rflags(vcpu);
6754 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6756 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6757 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6759 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6760 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6761 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6762 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6763 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6764 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6765 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6766 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6767 #ifdef CONFIG_X86_64
6768 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6769 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6770 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6771 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6772 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6773 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6774 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6775 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6778 kvm_rip_write(vcpu, regs->rip);
6779 kvm_set_rflags(vcpu, regs->rflags);
6781 vcpu->arch.exception.pending = false;
6783 kvm_make_request(KVM_REQ_EVENT, vcpu);
6788 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6790 struct kvm_segment cs;
6792 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6796 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6798 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6799 struct kvm_sregs *sregs)
6803 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6804 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6805 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6806 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6807 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6808 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6810 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6811 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6813 kvm_x86_ops->get_idt(vcpu, &dt);
6814 sregs->idt.limit = dt.size;
6815 sregs->idt.base = dt.address;
6816 kvm_x86_ops->get_gdt(vcpu, &dt);
6817 sregs->gdt.limit = dt.size;
6818 sregs->gdt.base = dt.address;
6820 sregs->cr0 = kvm_read_cr0(vcpu);
6821 sregs->cr2 = vcpu->arch.cr2;
6822 sregs->cr3 = kvm_read_cr3(vcpu);
6823 sregs->cr4 = kvm_read_cr4(vcpu);
6824 sregs->cr8 = kvm_get_cr8(vcpu);
6825 sregs->efer = vcpu->arch.efer;
6826 sregs->apic_base = kvm_get_apic_base(vcpu);
6828 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6830 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6831 set_bit(vcpu->arch.interrupt.nr,
6832 (unsigned long *)sregs->interrupt_bitmap);
6837 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6838 struct kvm_mp_state *mp_state)
6840 kvm_apic_accept_events(vcpu);
6841 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6842 vcpu->arch.pv.pv_unhalted)
6843 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6845 mp_state->mp_state = vcpu->arch.mp_state;
6850 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6851 struct kvm_mp_state *mp_state)
6853 if (!kvm_vcpu_has_lapic(vcpu) &&
6854 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6857 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6858 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6859 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6861 vcpu->arch.mp_state = mp_state->mp_state;
6862 kvm_make_request(KVM_REQ_EVENT, vcpu);
6866 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6867 int reason, bool has_error_code, u32 error_code)
6869 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6872 init_emulate_ctxt(vcpu);
6874 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6875 has_error_code, error_code);
6878 return EMULATE_FAIL;
6880 kvm_rip_write(vcpu, ctxt->eip);
6881 kvm_set_rflags(vcpu, ctxt->eflags);
6882 kvm_make_request(KVM_REQ_EVENT, vcpu);
6883 return EMULATE_DONE;
6885 EXPORT_SYMBOL_GPL(kvm_task_switch);
6887 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6888 struct kvm_sregs *sregs)
6890 struct msr_data apic_base_msr;
6891 int mmu_reset_needed = 0;
6892 int pending_vec, max_bits, idx;
6895 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6898 dt.size = sregs->idt.limit;
6899 dt.address = sregs->idt.base;
6900 kvm_x86_ops->set_idt(vcpu, &dt);
6901 dt.size = sregs->gdt.limit;
6902 dt.address = sregs->gdt.base;
6903 kvm_x86_ops->set_gdt(vcpu, &dt);
6905 vcpu->arch.cr2 = sregs->cr2;
6906 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6907 vcpu->arch.cr3 = sregs->cr3;
6908 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6910 kvm_set_cr8(vcpu, sregs->cr8);
6912 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6913 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6914 apic_base_msr.data = sregs->apic_base;
6915 apic_base_msr.host_initiated = true;
6916 kvm_set_apic_base(vcpu, &apic_base_msr);
6918 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6919 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6920 vcpu->arch.cr0 = sregs->cr0;
6922 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6923 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6924 if (sregs->cr4 & X86_CR4_OSXSAVE)
6925 kvm_update_cpuid(vcpu);
6927 idx = srcu_read_lock(&vcpu->kvm->srcu);
6928 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6929 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6930 mmu_reset_needed = 1;
6932 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6934 if (mmu_reset_needed)
6935 kvm_mmu_reset_context(vcpu);
6937 max_bits = KVM_NR_INTERRUPTS;
6938 pending_vec = find_first_bit(
6939 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6940 if (pending_vec < max_bits) {
6941 kvm_queue_interrupt(vcpu, pending_vec, false);
6942 pr_debug("Set back pending irq %d\n", pending_vec);
6945 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6946 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6947 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6948 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6949 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6950 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6952 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6953 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6955 update_cr8_intercept(vcpu);
6957 /* Older userspace won't unhalt the vcpu on reset. */
6958 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6959 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6961 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6963 kvm_make_request(KVM_REQ_EVENT, vcpu);
6968 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6969 struct kvm_guest_debug *dbg)
6971 unsigned long rflags;
6974 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6976 if (vcpu->arch.exception.pending)
6978 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6979 kvm_queue_exception(vcpu, DB_VECTOR);
6981 kvm_queue_exception(vcpu, BP_VECTOR);
6985 * Read rflags as long as potentially injected trace flags are still
6988 rflags = kvm_get_rflags(vcpu);
6990 vcpu->guest_debug = dbg->control;
6991 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6992 vcpu->guest_debug = 0;
6994 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6995 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6996 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6997 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6999 for (i = 0; i < KVM_NR_DB_REGS; i++)
7000 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7002 kvm_update_dr7(vcpu);
7004 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7005 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7006 get_segment_base(vcpu, VCPU_SREG_CS);
7009 * Trigger an rflags update that will inject or remove the trace
7012 kvm_set_rflags(vcpu, rflags);
7014 kvm_x86_ops->update_db_bp_intercept(vcpu);
7024 * Translate a guest virtual address to a guest physical address.
7026 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7027 struct kvm_translation *tr)
7029 unsigned long vaddr = tr->linear_address;
7033 idx = srcu_read_lock(&vcpu->kvm->srcu);
7034 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7035 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7036 tr->physical_address = gpa;
7037 tr->valid = gpa != UNMAPPED_GVA;
7044 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7046 struct fxregs_state *fxsave =
7047 &vcpu->arch.guest_fpu.state.fxsave;
7049 memcpy(fpu->fpr, fxsave->st_space, 128);
7050 fpu->fcw = fxsave->cwd;
7051 fpu->fsw = fxsave->swd;
7052 fpu->ftwx = fxsave->twd;
7053 fpu->last_opcode = fxsave->fop;
7054 fpu->last_ip = fxsave->rip;
7055 fpu->last_dp = fxsave->rdp;
7056 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7061 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7063 struct fxregs_state *fxsave =
7064 &vcpu->arch.guest_fpu.state.fxsave;
7066 memcpy(fxsave->st_space, fpu->fpr, 128);
7067 fxsave->cwd = fpu->fcw;
7068 fxsave->swd = fpu->fsw;
7069 fxsave->twd = fpu->ftwx;
7070 fxsave->fop = fpu->last_opcode;
7071 fxsave->rip = fpu->last_ip;
7072 fxsave->rdp = fpu->last_dp;
7073 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7078 static void fx_init(struct kvm_vcpu *vcpu)
7080 fpstate_init(&vcpu->arch.guest_fpu.state);
7082 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7083 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7086 * Ensure guest xcr0 is valid for loading
7088 vcpu->arch.xcr0 = XSTATE_FP;
7090 vcpu->arch.cr0 |= X86_CR0_ET;
7093 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7095 if (vcpu->guest_fpu_loaded)
7099 * Restore all possible states in the guest,
7100 * and assume host would use all available bits.
7101 * Guest xcr0 would be loaded later.
7103 kvm_put_guest_xcr0(vcpu);
7104 vcpu->guest_fpu_loaded = 1;
7105 __kernel_fpu_begin();
7106 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7110 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7112 kvm_put_guest_xcr0(vcpu);
7114 if (!vcpu->guest_fpu_loaded) {
7115 vcpu->fpu_counter = 0;
7119 vcpu->guest_fpu_loaded = 0;
7120 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7122 ++vcpu->stat.fpu_reload;
7124 * If using eager FPU mode, or if the guest is a frequent user
7125 * of the FPU, just leave the FPU active for next time.
7126 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7127 * the FPU in bursts will revert to loading it on demand.
7129 if (!vcpu->arch.eager_fpu) {
7130 if (++vcpu->fpu_counter < 5)
7131 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7136 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7138 kvmclock_reset(vcpu);
7140 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7141 kvm_x86_ops->vcpu_free(vcpu);
7144 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7147 struct kvm_vcpu *vcpu;
7149 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7150 printk_once(KERN_WARNING
7151 "kvm: SMP vm created on host with unstable TSC; "
7152 "guest TSC will not be reliable\n");
7154 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7159 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7163 kvm_vcpu_mtrr_init(vcpu);
7164 r = vcpu_load(vcpu);
7167 kvm_vcpu_reset(vcpu, false);
7168 kvm_mmu_setup(vcpu);
7173 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7175 struct msr_data msr;
7176 struct kvm *kvm = vcpu->kvm;
7178 if (vcpu_load(vcpu))
7181 msr.index = MSR_IA32_TSC;
7182 msr.host_initiated = true;
7183 kvm_write_tsc(vcpu, &msr);
7186 if (!kvmclock_periodic_sync)
7189 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7190 KVMCLOCK_SYNC_PERIOD);
7193 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7196 vcpu->arch.apf.msr_val = 0;
7198 r = vcpu_load(vcpu);
7200 kvm_mmu_unload(vcpu);
7203 kvm_x86_ops->vcpu_free(vcpu);
7206 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7208 vcpu->arch.hflags = 0;
7210 atomic_set(&vcpu->arch.nmi_queued, 0);
7211 vcpu->arch.nmi_pending = 0;
7212 vcpu->arch.nmi_injected = false;
7213 kvm_clear_interrupt_queue(vcpu);
7214 kvm_clear_exception_queue(vcpu);
7216 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7217 kvm_update_dr0123(vcpu);
7218 vcpu->arch.dr6 = DR6_INIT;
7219 kvm_update_dr6(vcpu);
7220 vcpu->arch.dr7 = DR7_FIXED_1;
7221 kvm_update_dr7(vcpu);
7225 kvm_make_request(KVM_REQ_EVENT, vcpu);
7226 vcpu->arch.apf.msr_val = 0;
7227 vcpu->arch.st.msr_val = 0;
7229 kvmclock_reset(vcpu);
7231 kvm_clear_async_pf_completion_queue(vcpu);
7232 kvm_async_pf_hash_reset(vcpu);
7233 vcpu->arch.apf.halted = false;
7236 kvm_pmu_reset(vcpu);
7237 vcpu->arch.smbase = 0x30000;
7240 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7241 vcpu->arch.regs_avail = ~0;
7242 vcpu->arch.regs_dirty = ~0;
7244 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7247 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7249 struct kvm_segment cs;
7251 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7252 cs.selector = vector << 8;
7253 cs.base = vector << 12;
7254 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7255 kvm_rip_write(vcpu, 0);
7258 int kvm_arch_hardware_enable(void)
7261 struct kvm_vcpu *vcpu;
7266 bool stable, backwards_tsc = false;
7268 kvm_shared_msr_cpu_online();
7269 ret = kvm_x86_ops->hardware_enable();
7273 local_tsc = rdtsc();
7274 stable = !check_tsc_unstable();
7275 list_for_each_entry(kvm, &vm_list, vm_list) {
7276 kvm_for_each_vcpu(i, vcpu, kvm) {
7277 if (!stable && vcpu->cpu == smp_processor_id())
7278 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7279 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7280 backwards_tsc = true;
7281 if (vcpu->arch.last_host_tsc > max_tsc)
7282 max_tsc = vcpu->arch.last_host_tsc;
7288 * Sometimes, even reliable TSCs go backwards. This happens on
7289 * platforms that reset TSC during suspend or hibernate actions, but
7290 * maintain synchronization. We must compensate. Fortunately, we can
7291 * detect that condition here, which happens early in CPU bringup,
7292 * before any KVM threads can be running. Unfortunately, we can't
7293 * bring the TSCs fully up to date with real time, as we aren't yet far
7294 * enough into CPU bringup that we know how much real time has actually
7295 * elapsed; our helper function, get_kernel_ns() will be using boot
7296 * variables that haven't been updated yet.
7298 * So we simply find the maximum observed TSC above, then record the
7299 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7300 * the adjustment will be applied. Note that we accumulate
7301 * adjustments, in case multiple suspend cycles happen before some VCPU
7302 * gets a chance to run again. In the event that no KVM threads get a
7303 * chance to run, we will miss the entire elapsed period, as we'll have
7304 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7305 * loose cycle time. This isn't too big a deal, since the loss will be
7306 * uniform across all VCPUs (not to mention the scenario is extremely
7307 * unlikely). It is possible that a second hibernate recovery happens
7308 * much faster than a first, causing the observed TSC here to be
7309 * smaller; this would require additional padding adjustment, which is
7310 * why we set last_host_tsc to the local tsc observed here.
7312 * N.B. - this code below runs only on platforms with reliable TSC,
7313 * as that is the only way backwards_tsc is set above. Also note
7314 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7315 * have the same delta_cyc adjustment applied if backwards_tsc
7316 * is detected. Note further, this adjustment is only done once,
7317 * as we reset last_host_tsc on all VCPUs to stop this from being
7318 * called multiple times (one for each physical CPU bringup).
7320 * Platforms with unreliable TSCs don't have to deal with this, they
7321 * will be compensated by the logic in vcpu_load, which sets the TSC to
7322 * catchup mode. This will catchup all VCPUs to real time, but cannot
7323 * guarantee that they stay in perfect synchronization.
7325 if (backwards_tsc) {
7326 u64 delta_cyc = max_tsc - local_tsc;
7327 backwards_tsc_observed = true;
7328 list_for_each_entry(kvm, &vm_list, vm_list) {
7329 kvm_for_each_vcpu(i, vcpu, kvm) {
7330 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7331 vcpu->arch.last_host_tsc = local_tsc;
7332 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7336 * We have to disable TSC offset matching.. if you were
7337 * booting a VM while issuing an S4 host suspend....
7338 * you may have some problem. Solving this issue is
7339 * left as an exercise to the reader.
7341 kvm->arch.last_tsc_nsec = 0;
7342 kvm->arch.last_tsc_write = 0;
7349 void kvm_arch_hardware_disable(void)
7351 kvm_x86_ops->hardware_disable();
7352 drop_user_return_notifiers();
7355 int kvm_arch_hardware_setup(void)
7359 r = kvm_x86_ops->hardware_setup();
7363 kvm_init_msr_list();
7367 void kvm_arch_hardware_unsetup(void)
7369 kvm_x86_ops->hardware_unsetup();
7372 void kvm_arch_check_processor_compat(void *rtn)
7374 kvm_x86_ops->check_processor_compatibility(rtn);
7377 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7379 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7381 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7383 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7385 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7388 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7390 return irqchip_in_kernel(vcpu->kvm) == lapic_in_kernel(vcpu);
7393 struct static_key kvm_no_apic_vcpu __read_mostly;
7395 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7401 BUG_ON(vcpu->kvm == NULL);
7404 vcpu->arch.pv.pv_unhalted = false;
7405 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7406 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7407 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7409 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7411 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7416 vcpu->arch.pio_data = page_address(page);
7418 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7420 r = kvm_mmu_create(vcpu);
7422 goto fail_free_pio_data;
7424 if (irqchip_in_kernel(kvm)) {
7425 r = kvm_create_lapic(vcpu);
7427 goto fail_mmu_destroy;
7429 static_key_slow_inc(&kvm_no_apic_vcpu);
7431 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7433 if (!vcpu->arch.mce_banks) {
7435 goto fail_free_lapic;
7437 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7439 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7441 goto fail_free_mce_banks;
7446 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7447 vcpu->arch.pv_time_enabled = false;
7449 vcpu->arch.guest_supported_xcr0 = 0;
7450 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7452 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7454 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7456 kvm_async_pf_hash_reset(vcpu);
7459 vcpu->arch.pending_external_vector = -1;
7463 fail_free_mce_banks:
7464 kfree(vcpu->arch.mce_banks);
7466 kvm_free_lapic(vcpu);
7468 kvm_mmu_destroy(vcpu);
7470 free_page((unsigned long)vcpu->arch.pio_data);
7475 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7479 kvm_pmu_destroy(vcpu);
7480 kfree(vcpu->arch.mce_banks);
7481 kvm_free_lapic(vcpu);
7482 idx = srcu_read_lock(&vcpu->kvm->srcu);
7483 kvm_mmu_destroy(vcpu);
7484 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7485 free_page((unsigned long)vcpu->arch.pio_data);
7486 if (!lapic_in_kernel(vcpu))
7487 static_key_slow_dec(&kvm_no_apic_vcpu);
7490 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7492 kvm_x86_ops->sched_in(vcpu, cpu);
7495 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7500 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7501 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7502 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7503 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7504 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7506 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7507 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7508 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7509 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7510 &kvm->arch.irq_sources_bitmap);
7512 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7513 mutex_init(&kvm->arch.apic_map_lock);
7514 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7516 pvclock_update_vm_gtod_copy(kvm);
7518 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7519 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7524 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7527 r = vcpu_load(vcpu);
7529 kvm_mmu_unload(vcpu);
7533 static void kvm_free_vcpus(struct kvm *kvm)
7536 struct kvm_vcpu *vcpu;
7539 * Unpin any mmu pages first.
7541 kvm_for_each_vcpu(i, vcpu, kvm) {
7542 kvm_clear_async_pf_completion_queue(vcpu);
7543 kvm_unload_vcpu_mmu(vcpu);
7545 kvm_for_each_vcpu(i, vcpu, kvm)
7546 kvm_arch_vcpu_free(vcpu);
7548 mutex_lock(&kvm->lock);
7549 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7550 kvm->vcpus[i] = NULL;
7552 atomic_set(&kvm->online_vcpus, 0);
7553 mutex_unlock(&kvm->lock);
7556 void kvm_arch_sync_events(struct kvm *kvm)
7558 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7559 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7560 kvm_free_all_assigned_devices(kvm);
7564 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7568 struct kvm_memslots *slots = kvm_memslots(kvm);
7569 struct kvm_memory_slot *slot, old;
7571 /* Called with kvm->slots_lock held. */
7572 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7575 slot = id_to_memslot(slots, id);
7577 if (WARN_ON(slot->npages))
7581 * MAP_SHARED to prevent internal slot pages from being moved
7584 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7585 MAP_SHARED | MAP_ANONYMOUS, 0);
7586 if (IS_ERR((void *)hva))
7587 return PTR_ERR((void *)hva);
7596 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7597 struct kvm_userspace_memory_region m;
7599 m.slot = id | (i << 16);
7601 m.guest_phys_addr = gpa;
7602 m.userspace_addr = hva;
7603 m.memory_size = size;
7604 r = __kvm_set_memory_region(kvm, &m);
7610 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7616 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7618 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7622 mutex_lock(&kvm->slots_lock);
7623 r = __x86_set_memory_region(kvm, id, gpa, size);
7624 mutex_unlock(&kvm->slots_lock);
7628 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7630 void kvm_arch_destroy_vm(struct kvm *kvm)
7632 if (current->mm == kvm->mm) {
7634 * Free memory regions allocated on behalf of userspace,
7635 * unless the the memory map has changed due to process exit
7638 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7639 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7640 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7642 kvm_iommu_unmap_guest(kvm);
7643 kfree(kvm->arch.vpic);
7644 kfree(kvm->arch.vioapic);
7645 kvm_free_vcpus(kvm);
7646 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7649 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7650 struct kvm_memory_slot *dont)
7654 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7655 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7656 kvfree(free->arch.rmap[i]);
7657 free->arch.rmap[i] = NULL;
7662 if (!dont || free->arch.lpage_info[i - 1] !=
7663 dont->arch.lpage_info[i - 1]) {
7664 kvfree(free->arch.lpage_info[i - 1]);
7665 free->arch.lpage_info[i - 1] = NULL;
7670 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7671 unsigned long npages)
7675 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7680 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7681 slot->base_gfn, level) + 1;
7683 slot->arch.rmap[i] =
7684 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7685 if (!slot->arch.rmap[i])
7690 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7691 sizeof(*slot->arch.lpage_info[i - 1]));
7692 if (!slot->arch.lpage_info[i - 1])
7695 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7696 slot->arch.lpage_info[i - 1][0].write_count = 1;
7697 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7698 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7699 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7701 * If the gfn and userspace address are not aligned wrt each
7702 * other, or if explicitly asked to, disable large page
7703 * support for this slot
7705 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7706 !kvm_largepages_enabled()) {
7709 for (j = 0; j < lpages; ++j)
7710 slot->arch.lpage_info[i - 1][j].write_count = 1;
7717 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7718 kvfree(slot->arch.rmap[i]);
7719 slot->arch.rmap[i] = NULL;
7723 kvfree(slot->arch.lpage_info[i - 1]);
7724 slot->arch.lpage_info[i - 1] = NULL;
7729 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
7732 * memslots->generation has been incremented.
7733 * mmio generation may have reached its maximum value.
7735 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
7738 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7739 struct kvm_memory_slot *memslot,
7740 const struct kvm_userspace_memory_region *mem,
7741 enum kvm_mr_change change)
7746 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7747 struct kvm_memory_slot *new)
7749 /* Still write protect RO slot */
7750 if (new->flags & KVM_MEM_READONLY) {
7751 kvm_mmu_slot_remove_write_access(kvm, new);
7756 * Call kvm_x86_ops dirty logging hooks when they are valid.
7758 * kvm_x86_ops->slot_disable_log_dirty is called when:
7760 * - KVM_MR_CREATE with dirty logging is disabled
7761 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7763 * The reason is, in case of PML, we need to set D-bit for any slots
7764 * with dirty logging disabled in order to eliminate unnecessary GPA
7765 * logging in PML buffer (and potential PML buffer full VMEXT). This
7766 * guarantees leaving PML enabled during guest's lifetime won't have
7767 * any additonal overhead from PML when guest is running with dirty
7768 * logging disabled for memory slots.
7770 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7771 * to dirty logging mode.
7773 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7775 * In case of write protect:
7777 * Write protect all pages for dirty logging.
7779 * All the sptes including the large sptes which point to this
7780 * slot are set to readonly. We can not create any new large
7781 * spte on this slot until the end of the logging.
7783 * See the comments in fast_page_fault().
7785 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7786 if (kvm_x86_ops->slot_enable_log_dirty)
7787 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7789 kvm_mmu_slot_remove_write_access(kvm, new);
7791 if (kvm_x86_ops->slot_disable_log_dirty)
7792 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7796 void kvm_arch_commit_memory_region(struct kvm *kvm,
7797 const struct kvm_userspace_memory_region *mem,
7798 const struct kvm_memory_slot *old,
7799 const struct kvm_memory_slot *new,
7800 enum kvm_mr_change change)
7802 int nr_mmu_pages = 0;
7804 if (!kvm->arch.n_requested_mmu_pages)
7805 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7808 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7811 * Dirty logging tracks sptes in 4k granularity, meaning that large
7812 * sptes have to be split. If live migration is successful, the guest
7813 * in the source machine will be destroyed and large sptes will be
7814 * created in the destination. However, if the guest continues to run
7815 * in the source machine (for example if live migration fails), small
7816 * sptes will remain around and cause bad performance.
7818 * Scan sptes if dirty logging has been stopped, dropping those
7819 * which can be collapsed into a single large-page spte. Later
7820 * page faults will create the large-page sptes.
7822 if ((change != KVM_MR_DELETE) &&
7823 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
7824 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
7825 kvm_mmu_zap_collapsible_sptes(kvm, new);
7828 * Set up write protection and/or dirty logging for the new slot.
7830 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7831 * been zapped so no dirty logging staff is needed for old slot. For
7832 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7833 * new and it's also covered when dealing with the new slot.
7835 * FIXME: const-ify all uses of struct kvm_memory_slot.
7837 if (change != KVM_MR_DELETE)
7838 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
7841 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7843 kvm_mmu_invalidate_zap_all_pages(kvm);
7846 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7847 struct kvm_memory_slot *slot)
7849 kvm_mmu_invalidate_zap_all_pages(kvm);
7852 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
7854 if (!list_empty_careful(&vcpu->async_pf.done))
7857 if (kvm_apic_has_events(vcpu))
7860 if (vcpu->arch.pv.pv_unhalted)
7863 if (atomic_read(&vcpu->arch.nmi_queued))
7866 if (test_bit(KVM_REQ_SMI, &vcpu->requests))
7869 if (kvm_arch_interrupt_allowed(vcpu) &&
7870 kvm_cpu_has_interrupt(vcpu))
7876 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7878 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7879 kvm_x86_ops->check_nested_events(vcpu, false);
7881 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
7884 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7886 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7889 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7891 return kvm_x86_ops->interrupt_allowed(vcpu);
7894 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
7896 if (is_64_bit_mode(vcpu))
7897 return kvm_rip_read(vcpu);
7898 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
7899 kvm_rip_read(vcpu));
7901 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
7903 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7905 return kvm_get_linear_rip(vcpu) == linear_rip;
7907 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7909 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7911 unsigned long rflags;
7913 rflags = kvm_x86_ops->get_rflags(vcpu);
7914 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7915 rflags &= ~X86_EFLAGS_TF;
7918 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7920 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7922 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7923 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7924 rflags |= X86_EFLAGS_TF;
7925 kvm_x86_ops->set_rflags(vcpu, rflags);
7928 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7930 __kvm_set_rflags(vcpu, rflags);
7931 kvm_make_request(KVM_REQ_EVENT, vcpu);
7933 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7935 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7939 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7943 r = kvm_mmu_reload(vcpu);
7947 if (!vcpu->arch.mmu.direct_map &&
7948 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7951 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7954 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7956 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7959 static inline u32 kvm_async_pf_next_probe(u32 key)
7961 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7964 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7966 u32 key = kvm_async_pf_hash_fn(gfn);
7968 while (vcpu->arch.apf.gfns[key] != ~0)
7969 key = kvm_async_pf_next_probe(key);
7971 vcpu->arch.apf.gfns[key] = gfn;
7974 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7977 u32 key = kvm_async_pf_hash_fn(gfn);
7979 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7980 (vcpu->arch.apf.gfns[key] != gfn &&
7981 vcpu->arch.apf.gfns[key] != ~0); i++)
7982 key = kvm_async_pf_next_probe(key);
7987 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7989 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7992 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7996 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7998 vcpu->arch.apf.gfns[i] = ~0;
8000 j = kvm_async_pf_next_probe(j);
8001 if (vcpu->arch.apf.gfns[j] == ~0)
8003 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8005 * k lies cyclically in ]i,j]
8007 * |....j i.k.| or |.k..j i...|
8009 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8010 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8015 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8018 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8022 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8023 struct kvm_async_pf *work)
8025 struct x86_exception fault;
8027 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8028 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8030 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8031 (vcpu->arch.apf.send_user_only &&
8032 kvm_x86_ops->get_cpl(vcpu) == 0))
8033 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8034 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8035 fault.vector = PF_VECTOR;
8036 fault.error_code_valid = true;
8037 fault.error_code = 0;
8038 fault.nested_page_fault = false;
8039 fault.address = work->arch.token;
8040 kvm_inject_page_fault(vcpu, &fault);
8044 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8045 struct kvm_async_pf *work)
8047 struct x86_exception fault;
8049 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8050 if (work->wakeup_all)
8051 work->arch.token = ~0; /* broadcast wakeup */
8053 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8055 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8056 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8057 fault.vector = PF_VECTOR;
8058 fault.error_code_valid = true;
8059 fault.error_code = 0;
8060 fault.nested_page_fault = false;
8061 fault.address = work->arch.token;
8062 kvm_inject_page_fault(vcpu, &fault);
8064 vcpu->arch.apf.halted = false;
8065 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8068 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8070 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8073 return !kvm_event_needs_reinjection(vcpu) &&
8074 kvm_x86_ops->interrupt_allowed(vcpu);
8077 void kvm_arch_start_assignment(struct kvm *kvm)
8079 atomic_inc(&kvm->arch.assigned_device_count);
8081 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8083 void kvm_arch_end_assignment(struct kvm *kvm)
8085 atomic_dec(&kvm->arch.assigned_device_count);
8087 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8089 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8091 return atomic_read(&kvm->arch.assigned_device_count);
8093 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8095 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8097 atomic_inc(&kvm->arch.noncoherent_dma_count);
8099 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8101 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8103 atomic_dec(&kvm->arch.noncoherent_dma_count);
8105 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8107 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8109 return atomic_read(&kvm->arch.noncoherent_dma_count);
8111 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8113 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8114 struct irq_bypass_producer *prod)
8116 struct kvm_kernel_irqfd *irqfd =
8117 container_of(cons, struct kvm_kernel_irqfd, consumer);
8119 if (kvm_x86_ops->update_pi_irte) {
8120 irqfd->producer = prod;
8121 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8122 prod->irq, irqfd->gsi, 1);
8128 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8129 struct irq_bypass_producer *prod)
8132 struct kvm_kernel_irqfd *irqfd =
8133 container_of(cons, struct kvm_kernel_irqfd, consumer);
8135 if (!kvm_x86_ops->update_pi_irte) {
8136 WARN_ON(irqfd->producer != NULL);
8140 WARN_ON(irqfd->producer != prod);
8141 irqfd->producer = NULL;
8144 * When producer of consumer is unregistered, we change back to
8145 * remapped mode, so we can re-use the current implementation
8146 * when the irq is masked/disabed or the consumer side (KVM
8147 * int this case doesn't want to receive the interrupts.
8149 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8151 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8152 " fails: %d\n", irqfd->consumer.token, ret);
8155 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8156 uint32_t guest_irq, bool set)
8158 if (!kvm_x86_ops->update_pi_irte)
8161 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8164 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8165 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8166 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8167 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8168 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8169 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8170 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8171 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8172 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8173 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8174 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8175 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8176 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8177 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8178 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8179 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8180 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);