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 enter_smm(struct kvm_vcpu *vcpu);
95 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
97 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
98 EXPORT_SYMBOL_GPL(kvm_x86_ops);
100 static bool __read_mostly ignore_msrs = 0;
101 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
103 unsigned int min_timer_period_us = 500;
104 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
106 static bool __read_mostly kvmclock_periodic_sync = true;
107 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
109 bool __read_mostly kvm_has_tsc_control;
110 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
111 u32 __read_mostly kvm_max_guest_tsc_khz;
112 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
113 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
114 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
115 u64 __read_mostly kvm_max_tsc_scaling_ratio;
116 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
117 static u64 __read_mostly kvm_default_tsc_scaling_ratio;
119 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
120 static u32 __read_mostly tsc_tolerance_ppm = 250;
121 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
123 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
124 unsigned int __read_mostly lapic_timer_advance_ns = 0;
125 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
127 static bool __read_mostly vector_hashing = true;
128 module_param(vector_hashing, bool, S_IRUGO);
130 static bool __read_mostly backwards_tsc_observed = false;
132 #define KVM_NR_SHARED_MSRS 16
134 struct kvm_shared_msrs_global {
136 u32 msrs[KVM_NR_SHARED_MSRS];
139 struct kvm_shared_msrs {
140 struct user_return_notifier urn;
142 struct kvm_shared_msr_values {
145 } values[KVM_NR_SHARED_MSRS];
148 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
149 static struct kvm_shared_msrs __percpu *shared_msrs;
151 struct kvm_stats_debugfs_item debugfs_entries[] = {
152 { "pf_fixed", VCPU_STAT(pf_fixed) },
153 { "pf_guest", VCPU_STAT(pf_guest) },
154 { "tlb_flush", VCPU_STAT(tlb_flush) },
155 { "invlpg", VCPU_STAT(invlpg) },
156 { "exits", VCPU_STAT(exits) },
157 { "io_exits", VCPU_STAT(io_exits) },
158 { "mmio_exits", VCPU_STAT(mmio_exits) },
159 { "signal_exits", VCPU_STAT(signal_exits) },
160 { "irq_window", VCPU_STAT(irq_window_exits) },
161 { "nmi_window", VCPU_STAT(nmi_window_exits) },
162 { "halt_exits", VCPU_STAT(halt_exits) },
163 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
164 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
165 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
166 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
167 { "hypercalls", VCPU_STAT(hypercalls) },
168 { "request_irq", VCPU_STAT(request_irq_exits) },
169 { "irq_exits", VCPU_STAT(irq_exits) },
170 { "host_state_reload", VCPU_STAT(host_state_reload) },
171 { "efer_reload", VCPU_STAT(efer_reload) },
172 { "fpu_reload", VCPU_STAT(fpu_reload) },
173 { "insn_emulation", VCPU_STAT(insn_emulation) },
174 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
175 { "irq_injections", VCPU_STAT(irq_injections) },
176 { "nmi_injections", VCPU_STAT(nmi_injections) },
177 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
178 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
179 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
180 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
181 { "mmu_flooded", VM_STAT(mmu_flooded) },
182 { "mmu_recycled", VM_STAT(mmu_recycled) },
183 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
184 { "mmu_unsync", VM_STAT(mmu_unsync) },
185 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
186 { "largepages", VM_STAT(lpages) },
190 u64 __read_mostly host_xcr0;
192 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
194 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
197 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
198 vcpu->arch.apf.gfns[i] = ~0;
201 static void kvm_on_user_return(struct user_return_notifier *urn)
204 struct kvm_shared_msrs *locals
205 = container_of(urn, struct kvm_shared_msrs, urn);
206 struct kvm_shared_msr_values *values;
208 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
209 values = &locals->values[slot];
210 if (values->host != values->curr) {
211 wrmsrl(shared_msrs_global.msrs[slot], values->host);
212 values->curr = values->host;
215 locals->registered = false;
216 user_return_notifier_unregister(urn);
219 static void shared_msr_update(unsigned slot, u32 msr)
222 unsigned int cpu = smp_processor_id();
223 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
225 /* only read, and nobody should modify it at this time,
226 * so don't need lock */
227 if (slot >= shared_msrs_global.nr) {
228 printk(KERN_ERR "kvm: invalid MSR slot!");
231 rdmsrl_safe(msr, &value);
232 smsr->values[slot].host = value;
233 smsr->values[slot].curr = value;
236 void kvm_define_shared_msr(unsigned slot, u32 msr)
238 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
239 shared_msrs_global.msrs[slot] = msr;
240 if (slot >= shared_msrs_global.nr)
241 shared_msrs_global.nr = slot + 1;
243 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
245 static void kvm_shared_msr_cpu_online(void)
249 for (i = 0; i < shared_msrs_global.nr; ++i)
250 shared_msr_update(i, shared_msrs_global.msrs[i]);
253 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
255 unsigned int cpu = smp_processor_id();
256 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
259 if (((value ^ smsr->values[slot].curr) & mask) == 0)
261 smsr->values[slot].curr = value;
262 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
266 if (!smsr->registered) {
267 smsr->urn.on_user_return = kvm_on_user_return;
268 user_return_notifier_register(&smsr->urn);
269 smsr->registered = true;
273 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
275 static void drop_user_return_notifiers(void)
277 unsigned int cpu = smp_processor_id();
278 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
280 if (smsr->registered)
281 kvm_on_user_return(&smsr->urn);
284 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
286 return vcpu->arch.apic_base;
288 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
290 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
292 u64 old_state = vcpu->arch.apic_base &
293 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
294 u64 new_state = msr_info->data &
295 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
296 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
297 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
299 if (!msr_info->host_initiated &&
300 ((msr_info->data & reserved_bits) != 0 ||
301 new_state == X2APIC_ENABLE ||
302 (new_state == MSR_IA32_APICBASE_ENABLE &&
303 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
304 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
308 kvm_lapic_set_base(vcpu, msr_info->data);
311 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
313 asmlinkage __visible void kvm_spurious_fault(void)
315 /* Fault while not rebooting. We want the trace. */
318 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
320 #define EXCPT_BENIGN 0
321 #define EXCPT_CONTRIBUTORY 1
324 static int exception_class(int vector)
334 return EXCPT_CONTRIBUTORY;
341 #define EXCPT_FAULT 0
343 #define EXCPT_ABORT 2
344 #define EXCPT_INTERRUPT 3
346 static int exception_type(int vector)
350 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
351 return EXCPT_INTERRUPT;
355 /* #DB is trap, as instruction watchpoints are handled elsewhere */
356 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
359 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
362 /* Reserved exceptions will result in fault */
366 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
367 unsigned nr, bool has_error, u32 error_code,
373 kvm_make_request(KVM_REQ_EVENT, vcpu);
375 if (!vcpu->arch.exception.pending) {
377 if (has_error && !is_protmode(vcpu))
379 vcpu->arch.exception.pending = true;
380 vcpu->arch.exception.has_error_code = has_error;
381 vcpu->arch.exception.nr = nr;
382 vcpu->arch.exception.error_code = error_code;
383 vcpu->arch.exception.reinject = reinject;
387 /* to check exception */
388 prev_nr = vcpu->arch.exception.nr;
389 if (prev_nr == DF_VECTOR) {
390 /* triple fault -> shutdown */
391 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
394 class1 = exception_class(prev_nr);
395 class2 = exception_class(nr);
396 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
397 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
398 /* generate double fault per SDM Table 5-5 */
399 vcpu->arch.exception.pending = true;
400 vcpu->arch.exception.has_error_code = true;
401 vcpu->arch.exception.nr = DF_VECTOR;
402 vcpu->arch.exception.error_code = 0;
404 /* replace previous exception with a new one in a hope
405 that instruction re-execution will regenerate lost
410 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
412 kvm_multiple_exception(vcpu, nr, false, 0, false);
414 EXPORT_SYMBOL_GPL(kvm_queue_exception);
416 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
418 kvm_multiple_exception(vcpu, nr, false, 0, true);
420 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
422 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
425 kvm_inject_gp(vcpu, 0);
427 kvm_x86_ops->skip_emulated_instruction(vcpu);
429 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
431 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
433 ++vcpu->stat.pf_guest;
434 vcpu->arch.cr2 = fault->address;
435 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
437 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
439 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
441 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
442 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
444 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
446 return fault->nested_page_fault;
449 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
451 atomic_inc(&vcpu->arch.nmi_queued);
452 kvm_make_request(KVM_REQ_NMI, vcpu);
454 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
456 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
458 kvm_multiple_exception(vcpu, nr, true, error_code, false);
460 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
462 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
464 kvm_multiple_exception(vcpu, nr, true, error_code, true);
466 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
469 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
470 * a #GP and return false.
472 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
474 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
476 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
479 EXPORT_SYMBOL_GPL(kvm_require_cpl);
481 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
483 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
486 kvm_queue_exception(vcpu, UD_VECTOR);
489 EXPORT_SYMBOL_GPL(kvm_require_dr);
492 * This function will be used to read from the physical memory of the currently
493 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
494 * can read from guest physical or from the guest's guest physical memory.
496 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
497 gfn_t ngfn, void *data, int offset, int len,
500 struct x86_exception exception;
504 ngpa = gfn_to_gpa(ngfn);
505 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
506 if (real_gfn == UNMAPPED_GVA)
509 real_gfn = gpa_to_gfn(real_gfn);
511 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
513 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
515 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
516 void *data, int offset, int len, u32 access)
518 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
519 data, offset, len, access);
523 * Load the pae pdptrs. Return true is they are all valid.
525 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
527 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
528 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
531 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
533 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
534 offset * sizeof(u64), sizeof(pdpte),
535 PFERR_USER_MASK|PFERR_WRITE_MASK);
540 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
541 if (is_present_gpte(pdpte[i]) &&
543 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
550 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
551 __set_bit(VCPU_EXREG_PDPTR,
552 (unsigned long *)&vcpu->arch.regs_avail);
553 __set_bit(VCPU_EXREG_PDPTR,
554 (unsigned long *)&vcpu->arch.regs_dirty);
559 EXPORT_SYMBOL_GPL(load_pdptrs);
561 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
563 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
569 if (is_long_mode(vcpu) || !is_pae(vcpu))
572 if (!test_bit(VCPU_EXREG_PDPTR,
573 (unsigned long *)&vcpu->arch.regs_avail))
576 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
577 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
578 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
579 PFERR_USER_MASK | PFERR_WRITE_MASK);
582 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
588 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
590 unsigned long old_cr0 = kvm_read_cr0(vcpu);
591 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
596 if (cr0 & 0xffffffff00000000UL)
600 cr0 &= ~CR0_RESERVED_BITS;
602 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
605 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
608 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
610 if ((vcpu->arch.efer & EFER_LME)) {
615 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
620 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
625 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
628 kvm_x86_ops->set_cr0(vcpu, cr0);
630 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
631 kvm_clear_async_pf_completion_queue(vcpu);
632 kvm_async_pf_hash_reset(vcpu);
635 if ((cr0 ^ old_cr0) & update_bits)
636 kvm_mmu_reset_context(vcpu);
638 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
639 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
640 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
641 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
645 EXPORT_SYMBOL_GPL(kvm_set_cr0);
647 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
649 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
651 EXPORT_SYMBOL_GPL(kvm_lmsw);
653 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
655 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
656 !vcpu->guest_xcr0_loaded) {
657 /* kvm_set_xcr() also depends on this */
658 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
659 vcpu->guest_xcr0_loaded = 1;
663 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
665 if (vcpu->guest_xcr0_loaded) {
666 if (vcpu->arch.xcr0 != host_xcr0)
667 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
668 vcpu->guest_xcr0_loaded = 0;
672 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
675 u64 old_xcr0 = vcpu->arch.xcr0;
678 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
679 if (index != XCR_XFEATURE_ENABLED_MASK)
681 if (!(xcr0 & XFEATURE_MASK_FP))
683 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
687 * Do not allow the guest to set bits that we do not support
688 * saving. However, xcr0 bit 0 is always set, even if the
689 * emulated CPU does not support XSAVE (see fx_init).
691 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
692 if (xcr0 & ~valid_bits)
695 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
696 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
699 if (xcr0 & XFEATURE_MASK_AVX512) {
700 if (!(xcr0 & XFEATURE_MASK_YMM))
702 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
705 vcpu->arch.xcr0 = xcr0;
707 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
708 kvm_update_cpuid(vcpu);
712 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
714 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
715 __kvm_set_xcr(vcpu, index, xcr)) {
716 kvm_inject_gp(vcpu, 0);
721 EXPORT_SYMBOL_GPL(kvm_set_xcr);
723 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
725 unsigned long old_cr4 = kvm_read_cr4(vcpu);
726 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
727 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
729 if (cr4 & CR4_RESERVED_BITS)
732 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
735 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
738 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
741 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
744 if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
747 if (is_long_mode(vcpu)) {
748 if (!(cr4 & X86_CR4_PAE))
750 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
751 && ((cr4 ^ old_cr4) & pdptr_bits)
752 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
756 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
757 if (!guest_cpuid_has_pcid(vcpu))
760 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
761 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
765 if (kvm_x86_ops->set_cr4(vcpu, cr4))
768 if (((cr4 ^ old_cr4) & pdptr_bits) ||
769 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
770 kvm_mmu_reset_context(vcpu);
772 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
773 kvm_update_cpuid(vcpu);
777 EXPORT_SYMBOL_GPL(kvm_set_cr4);
779 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
782 cr3 &= ~CR3_PCID_INVD;
785 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
786 kvm_mmu_sync_roots(vcpu);
787 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
791 if (is_long_mode(vcpu)) {
792 if (cr3 & CR3_L_MODE_RESERVED_BITS)
794 } else if (is_pae(vcpu) && is_paging(vcpu) &&
795 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
798 vcpu->arch.cr3 = cr3;
799 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
800 kvm_mmu_new_cr3(vcpu);
803 EXPORT_SYMBOL_GPL(kvm_set_cr3);
805 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
807 if (cr8 & CR8_RESERVED_BITS)
809 if (lapic_in_kernel(vcpu))
810 kvm_lapic_set_tpr(vcpu, cr8);
812 vcpu->arch.cr8 = cr8;
815 EXPORT_SYMBOL_GPL(kvm_set_cr8);
817 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
819 if (lapic_in_kernel(vcpu))
820 return kvm_lapic_get_cr8(vcpu);
822 return vcpu->arch.cr8;
824 EXPORT_SYMBOL_GPL(kvm_get_cr8);
826 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
830 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
831 for (i = 0; i < KVM_NR_DB_REGS; i++)
832 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
833 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
837 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
839 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
840 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
843 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
847 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
848 dr7 = vcpu->arch.guest_debug_dr7;
850 dr7 = vcpu->arch.dr7;
851 kvm_x86_ops->set_dr7(vcpu, dr7);
852 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
853 if (dr7 & DR7_BP_EN_MASK)
854 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
857 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
859 u64 fixed = DR6_FIXED_1;
861 if (!guest_cpuid_has_rtm(vcpu))
866 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
870 vcpu->arch.db[dr] = val;
871 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
872 vcpu->arch.eff_db[dr] = val;
877 if (val & 0xffffffff00000000ULL)
879 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
880 kvm_update_dr6(vcpu);
885 if (val & 0xffffffff00000000ULL)
887 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
888 kvm_update_dr7(vcpu);
895 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
897 if (__kvm_set_dr(vcpu, dr, val)) {
898 kvm_inject_gp(vcpu, 0);
903 EXPORT_SYMBOL_GPL(kvm_set_dr);
905 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
909 *val = vcpu->arch.db[dr];
914 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
915 *val = vcpu->arch.dr6;
917 *val = kvm_x86_ops->get_dr6(vcpu);
922 *val = vcpu->arch.dr7;
927 EXPORT_SYMBOL_GPL(kvm_get_dr);
929 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
931 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
935 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
938 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
939 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
942 EXPORT_SYMBOL_GPL(kvm_rdpmc);
945 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
946 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
948 * This list is modified at module load time to reflect the
949 * capabilities of the host cpu. This capabilities test skips MSRs that are
950 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
951 * may depend on host virtualization features rather than host cpu features.
954 static u32 msrs_to_save[] = {
955 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
958 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
960 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
961 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
964 static unsigned num_msrs_to_save;
966 static u32 emulated_msrs[] = {
967 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
968 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
969 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
970 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
971 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
972 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
975 HV_X64_MSR_VP_RUNTIME,
977 HV_X64_MSR_STIMER0_CONFIG,
978 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
982 MSR_IA32_TSCDEADLINE,
983 MSR_IA32_MISC_ENABLE,
989 static unsigned num_emulated_msrs;
991 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
993 if (efer & efer_reserved_bits)
996 if (efer & EFER_FFXSR) {
997 struct kvm_cpuid_entry2 *feat;
999 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1000 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
1004 if (efer & EFER_SVME) {
1005 struct kvm_cpuid_entry2 *feat;
1007 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1008 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1014 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1016 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1018 u64 old_efer = vcpu->arch.efer;
1020 if (!kvm_valid_efer(vcpu, efer))
1024 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1028 efer |= vcpu->arch.efer & EFER_LMA;
1030 kvm_x86_ops->set_efer(vcpu, efer);
1032 /* Update reserved bits */
1033 if ((efer ^ old_efer) & EFER_NX)
1034 kvm_mmu_reset_context(vcpu);
1039 void kvm_enable_efer_bits(u64 mask)
1041 efer_reserved_bits &= ~mask;
1043 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1046 * Writes msr value into into the appropriate "register".
1047 * Returns 0 on success, non-0 otherwise.
1048 * Assumes vcpu_load() was already called.
1050 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1052 switch (msr->index) {
1055 case MSR_KERNEL_GS_BASE:
1058 if (is_noncanonical_address(msr->data))
1061 case MSR_IA32_SYSENTER_EIP:
1062 case MSR_IA32_SYSENTER_ESP:
1064 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1065 * non-canonical address is written on Intel but not on
1066 * AMD (which ignores the top 32-bits, because it does
1067 * not implement 64-bit SYSENTER).
1069 * 64-bit code should hence be able to write a non-canonical
1070 * value on AMD. Making the address canonical ensures that
1071 * vmentry does not fail on Intel after writing a non-canonical
1072 * value, and that something deterministic happens if the guest
1073 * invokes 64-bit SYSENTER.
1075 msr->data = get_canonical(msr->data);
1077 return kvm_x86_ops->set_msr(vcpu, msr);
1079 EXPORT_SYMBOL_GPL(kvm_set_msr);
1082 * Adapt set_msr() to msr_io()'s calling convention
1084 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1086 struct msr_data msr;
1090 msr.host_initiated = true;
1091 r = kvm_get_msr(vcpu, &msr);
1099 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1101 struct msr_data msr;
1105 msr.host_initiated = true;
1106 return kvm_set_msr(vcpu, &msr);
1109 #ifdef CONFIG_X86_64
1110 struct pvclock_gtod_data {
1113 struct { /* extract of a clocksource struct */
1125 static struct pvclock_gtod_data pvclock_gtod_data;
1127 static void update_pvclock_gtod(struct timekeeper *tk)
1129 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1132 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1134 write_seqcount_begin(&vdata->seq);
1136 /* copy pvclock gtod data */
1137 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1138 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1139 vdata->clock.mask = tk->tkr_mono.mask;
1140 vdata->clock.mult = tk->tkr_mono.mult;
1141 vdata->clock.shift = tk->tkr_mono.shift;
1143 vdata->boot_ns = boot_ns;
1144 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1146 write_seqcount_end(&vdata->seq);
1150 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1153 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1154 * vcpu_enter_guest. This function is only called from
1155 * the physical CPU that is running vcpu.
1157 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1160 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1164 struct pvclock_wall_clock wc;
1165 struct timespec boot;
1170 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1175 ++version; /* first time write, random junk */
1179 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1183 * The guest calculates current wall clock time by adding
1184 * system time (updated by kvm_guest_time_update below) to the
1185 * wall clock specified here. guest system time equals host
1186 * system time for us, thus we must fill in host boot time here.
1190 if (kvm->arch.kvmclock_offset) {
1191 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1192 boot = timespec_sub(boot, ts);
1194 wc.sec = boot.tv_sec;
1195 wc.nsec = boot.tv_nsec;
1196 wc.version = version;
1198 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1201 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1204 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1206 do_shl32_div32(dividend, divisor);
1210 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1211 s8 *pshift, u32 *pmultiplier)
1219 scaled64 = scaled_hz;
1220 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1225 tps32 = (uint32_t)tps64;
1226 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1227 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1235 *pmultiplier = div_frac(scaled64, tps32);
1237 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1238 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1241 #ifdef CONFIG_X86_64
1242 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1245 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1246 static unsigned long max_tsc_khz;
1248 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1250 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1251 vcpu->arch.virtual_tsc_shift);
1254 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1256 u64 v = (u64)khz * (1000000 + ppm);
1261 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1265 /* Guest TSC same frequency as host TSC? */
1267 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1271 /* TSC scaling supported? */
1272 if (!kvm_has_tsc_control) {
1273 if (user_tsc_khz > tsc_khz) {
1274 vcpu->arch.tsc_catchup = 1;
1275 vcpu->arch.tsc_always_catchup = 1;
1278 WARN(1, "user requested TSC rate below hardware speed\n");
1283 /* TSC scaling required - calculate ratio */
1284 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1285 user_tsc_khz, tsc_khz);
1287 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1288 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1293 vcpu->arch.tsc_scaling_ratio = ratio;
1297 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1299 u32 thresh_lo, thresh_hi;
1300 int use_scaling = 0;
1302 /* tsc_khz can be zero if TSC calibration fails */
1303 if (user_tsc_khz == 0) {
1304 /* set tsc_scaling_ratio to a safe value */
1305 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1309 /* Compute a scale to convert nanoseconds in TSC cycles */
1310 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1311 &vcpu->arch.virtual_tsc_shift,
1312 &vcpu->arch.virtual_tsc_mult);
1313 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1316 * Compute the variation in TSC rate which is acceptable
1317 * within the range of tolerance and decide if the
1318 * rate being applied is within that bounds of the hardware
1319 * rate. If so, no scaling or compensation need be done.
1321 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1322 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1323 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1324 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1327 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1330 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1332 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1333 vcpu->arch.virtual_tsc_mult,
1334 vcpu->arch.virtual_tsc_shift);
1335 tsc += vcpu->arch.this_tsc_write;
1339 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1341 #ifdef CONFIG_X86_64
1343 struct kvm_arch *ka = &vcpu->kvm->arch;
1344 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1346 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1347 atomic_read(&vcpu->kvm->online_vcpus));
1350 * Once the masterclock is enabled, always perform request in
1351 * order to update it.
1353 * In order to enable masterclock, the host clocksource must be TSC
1354 * and the vcpus need to have matched TSCs. When that happens,
1355 * perform request to enable masterclock.
1357 if (ka->use_master_clock ||
1358 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1359 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1361 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1362 atomic_read(&vcpu->kvm->online_vcpus),
1363 ka->use_master_clock, gtod->clock.vclock_mode);
1367 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1369 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1370 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1374 * Multiply tsc by a fixed point number represented by ratio.
1376 * The most significant 64-N bits (mult) of ratio represent the
1377 * integral part of the fixed point number; the remaining N bits
1378 * (frac) represent the fractional part, ie. ratio represents a fixed
1379 * point number (mult + frac * 2^(-N)).
1381 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1383 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1385 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1388 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1391 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1393 if (ratio != kvm_default_tsc_scaling_ratio)
1394 _tsc = __scale_tsc(ratio, tsc);
1398 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1400 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1404 tsc = kvm_scale_tsc(vcpu, rdtsc());
1406 return target_tsc - tsc;
1409 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1411 return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
1413 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1415 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1417 struct kvm *kvm = vcpu->kvm;
1418 u64 offset, ns, elapsed;
1419 unsigned long flags;
1422 bool already_matched;
1423 u64 data = msr->data;
1425 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1426 offset = kvm_compute_tsc_offset(vcpu, data);
1427 ns = get_kernel_ns();
1428 elapsed = ns - kvm->arch.last_tsc_nsec;
1430 if (vcpu->arch.virtual_tsc_khz) {
1433 /* n.b - signed multiplication and division required */
1434 usdiff = data - kvm->arch.last_tsc_write;
1435 #ifdef CONFIG_X86_64
1436 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1438 /* do_div() only does unsigned */
1439 asm("1: idivl %[divisor]\n"
1440 "2: xor %%edx, %%edx\n"
1441 " movl $0, %[faulted]\n"
1443 ".section .fixup,\"ax\"\n"
1444 "4: movl $1, %[faulted]\n"
1448 _ASM_EXTABLE(1b, 4b)
1450 : "=A"(usdiff), [faulted] "=r" (faulted)
1451 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1454 do_div(elapsed, 1000);
1459 /* idivl overflow => difference is larger than USEC_PER_SEC */
1461 usdiff = USEC_PER_SEC;
1463 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1466 * Special case: TSC write with a small delta (1 second) of virtual
1467 * cycle time against real time is interpreted as an attempt to
1468 * synchronize the CPU.
1470 * For a reliable TSC, we can match TSC offsets, and for an unstable
1471 * TSC, we add elapsed time in this computation. We could let the
1472 * compensation code attempt to catch up if we fall behind, but
1473 * it's better to try to match offsets from the beginning.
1475 if (usdiff < USEC_PER_SEC &&
1476 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1477 if (!check_tsc_unstable()) {
1478 offset = kvm->arch.cur_tsc_offset;
1479 pr_debug("kvm: matched tsc offset for %llu\n", data);
1481 u64 delta = nsec_to_cycles(vcpu, elapsed);
1483 offset = kvm_compute_tsc_offset(vcpu, data);
1484 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1487 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1490 * We split periods of matched TSC writes into generations.
1491 * For each generation, we track the original measured
1492 * nanosecond time, offset, and write, so if TSCs are in
1493 * sync, we can match exact offset, and if not, we can match
1494 * exact software computation in compute_guest_tsc()
1496 * These values are tracked in kvm->arch.cur_xxx variables.
1498 kvm->arch.cur_tsc_generation++;
1499 kvm->arch.cur_tsc_nsec = ns;
1500 kvm->arch.cur_tsc_write = data;
1501 kvm->arch.cur_tsc_offset = offset;
1503 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1504 kvm->arch.cur_tsc_generation, data);
1508 * We also track th most recent recorded KHZ, write and time to
1509 * allow the matching interval to be extended at each write.
1511 kvm->arch.last_tsc_nsec = ns;
1512 kvm->arch.last_tsc_write = data;
1513 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1515 vcpu->arch.last_guest_tsc = data;
1517 /* Keep track of which generation this VCPU has synchronized to */
1518 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1519 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1520 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1522 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1523 update_ia32_tsc_adjust_msr(vcpu, offset);
1524 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1525 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1527 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1529 kvm->arch.nr_vcpus_matched_tsc = 0;
1530 } else if (!already_matched) {
1531 kvm->arch.nr_vcpus_matched_tsc++;
1534 kvm_track_tsc_matching(vcpu);
1535 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1538 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1540 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1543 kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1546 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1548 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1549 WARN_ON(adjustment < 0);
1550 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1551 kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1554 #ifdef CONFIG_X86_64
1556 static cycle_t read_tsc(void)
1558 cycle_t ret = (cycle_t)rdtsc_ordered();
1559 u64 last = pvclock_gtod_data.clock.cycle_last;
1561 if (likely(ret >= last))
1565 * GCC likes to generate cmov here, but this branch is extremely
1566 * predictable (it's just a function of time and the likely is
1567 * very likely) and there's a data dependence, so force GCC
1568 * to generate a branch instead. I don't barrier() because
1569 * we don't actually need a barrier, and if this function
1570 * ever gets inlined it will generate worse code.
1576 static inline u64 vgettsc(cycle_t *cycle_now)
1579 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1581 *cycle_now = read_tsc();
1583 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1584 return v * gtod->clock.mult;
1587 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1589 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1595 seq = read_seqcount_begin(>od->seq);
1596 mode = gtod->clock.vclock_mode;
1597 ns = gtod->nsec_base;
1598 ns += vgettsc(cycle_now);
1599 ns >>= gtod->clock.shift;
1600 ns += gtod->boot_ns;
1601 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1607 /* returns true if host is using tsc clocksource */
1608 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1610 /* checked again under seqlock below */
1611 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1614 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1620 * Assuming a stable TSC across physical CPUS, and a stable TSC
1621 * across virtual CPUs, the following condition is possible.
1622 * Each numbered line represents an event visible to both
1623 * CPUs at the next numbered event.
1625 * "timespecX" represents host monotonic time. "tscX" represents
1628 * VCPU0 on CPU0 | VCPU1 on CPU1
1630 * 1. read timespec0,tsc0
1631 * 2. | timespec1 = timespec0 + N
1633 * 3. transition to guest | transition to guest
1634 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1635 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1636 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1638 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1641 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1643 * - 0 < N - M => M < N
1645 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1646 * always the case (the difference between two distinct xtime instances
1647 * might be smaller then the difference between corresponding TSC reads,
1648 * when updating guest vcpus pvclock areas).
1650 * To avoid that problem, do not allow visibility of distinct
1651 * system_timestamp/tsc_timestamp values simultaneously: use a master
1652 * copy of host monotonic time values. Update that master copy
1655 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1659 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1661 #ifdef CONFIG_X86_64
1662 struct kvm_arch *ka = &kvm->arch;
1664 bool host_tsc_clocksource, vcpus_matched;
1666 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1667 atomic_read(&kvm->online_vcpus));
1670 * If the host uses TSC clock, then passthrough TSC as stable
1673 host_tsc_clocksource = kvm_get_time_and_clockread(
1674 &ka->master_kernel_ns,
1675 &ka->master_cycle_now);
1677 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1678 && !backwards_tsc_observed
1679 && !ka->boot_vcpu_runs_old_kvmclock;
1681 if (ka->use_master_clock)
1682 atomic_set(&kvm_guest_has_master_clock, 1);
1684 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1685 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1690 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1692 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1695 static void kvm_gen_update_masterclock(struct kvm *kvm)
1697 #ifdef CONFIG_X86_64
1699 struct kvm_vcpu *vcpu;
1700 struct kvm_arch *ka = &kvm->arch;
1702 spin_lock(&ka->pvclock_gtod_sync_lock);
1703 kvm_make_mclock_inprogress_request(kvm);
1704 /* no guest entries from this point */
1705 pvclock_update_vm_gtod_copy(kvm);
1707 kvm_for_each_vcpu(i, vcpu, kvm)
1708 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1710 /* guest entries allowed */
1711 kvm_for_each_vcpu(i, vcpu, kvm)
1712 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1714 spin_unlock(&ka->pvclock_gtod_sync_lock);
1718 static int kvm_guest_time_update(struct kvm_vcpu *v)
1720 unsigned long flags, tgt_tsc_khz;
1721 struct kvm_vcpu_arch *vcpu = &v->arch;
1722 struct kvm_arch *ka = &v->kvm->arch;
1724 u64 tsc_timestamp, host_tsc;
1725 struct pvclock_vcpu_time_info guest_hv_clock;
1727 bool use_master_clock;
1733 * If the host uses TSC clock, then passthrough TSC as stable
1736 spin_lock(&ka->pvclock_gtod_sync_lock);
1737 use_master_clock = ka->use_master_clock;
1738 if (use_master_clock) {
1739 host_tsc = ka->master_cycle_now;
1740 kernel_ns = ka->master_kernel_ns;
1742 spin_unlock(&ka->pvclock_gtod_sync_lock);
1744 /* Keep irq disabled to prevent changes to the clock */
1745 local_irq_save(flags);
1746 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1747 if (unlikely(tgt_tsc_khz == 0)) {
1748 local_irq_restore(flags);
1749 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1752 if (!use_master_clock) {
1754 kernel_ns = get_kernel_ns();
1757 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1760 * We may have to catch up the TSC to match elapsed wall clock
1761 * time for two reasons, even if kvmclock is used.
1762 * 1) CPU could have been running below the maximum TSC rate
1763 * 2) Broken TSC compensation resets the base at each VCPU
1764 * entry to avoid unknown leaps of TSC even when running
1765 * again on the same CPU. This may cause apparent elapsed
1766 * time to disappear, and the guest to stand still or run
1769 if (vcpu->tsc_catchup) {
1770 u64 tsc = compute_guest_tsc(v, kernel_ns);
1771 if (tsc > tsc_timestamp) {
1772 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1773 tsc_timestamp = tsc;
1777 local_irq_restore(flags);
1779 if (!vcpu->pv_time_enabled)
1782 if (kvm_has_tsc_control)
1783 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1785 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1786 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1787 &vcpu->hv_clock.tsc_shift,
1788 &vcpu->hv_clock.tsc_to_system_mul);
1789 vcpu->hw_tsc_khz = tgt_tsc_khz;
1792 /* With all the info we got, fill in the values */
1793 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1794 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1795 vcpu->last_guest_tsc = tsc_timestamp;
1797 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1798 &guest_hv_clock, sizeof(guest_hv_clock))))
1801 /* This VCPU is paused, but it's legal for a guest to read another
1802 * VCPU's kvmclock, so we really have to follow the specification where
1803 * it says that version is odd if data is being modified, and even after
1806 * Version field updates must be kept separate. This is because
1807 * kvm_write_guest_cached might use a "rep movs" instruction, and
1808 * writes within a string instruction are weakly ordered. So there
1809 * are three writes overall.
1811 * As a small optimization, only write the version field in the first
1812 * and third write. The vcpu->pv_time cache is still valid, because the
1813 * version field is the first in the struct.
1815 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1817 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1818 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1820 sizeof(vcpu->hv_clock.version));
1824 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1825 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1827 if (vcpu->pvclock_set_guest_stopped_request) {
1828 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1829 vcpu->pvclock_set_guest_stopped_request = false;
1832 /* If the host uses TSC clocksource, then it is stable */
1833 if (use_master_clock)
1834 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1836 vcpu->hv_clock.flags = pvclock_flags;
1838 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1840 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1842 sizeof(vcpu->hv_clock));
1846 vcpu->hv_clock.version++;
1847 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1849 sizeof(vcpu->hv_clock.version));
1854 * kvmclock updates which are isolated to a given vcpu, such as
1855 * vcpu->cpu migration, should not allow system_timestamp from
1856 * the rest of the vcpus to remain static. Otherwise ntp frequency
1857 * correction applies to one vcpu's system_timestamp but not
1860 * So in those cases, request a kvmclock update for all vcpus.
1861 * We need to rate-limit these requests though, as they can
1862 * considerably slow guests that have a large number of vcpus.
1863 * The time for a remote vcpu to update its kvmclock is bound
1864 * by the delay we use to rate-limit the updates.
1867 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1869 static void kvmclock_update_fn(struct work_struct *work)
1872 struct delayed_work *dwork = to_delayed_work(work);
1873 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1874 kvmclock_update_work);
1875 struct kvm *kvm = container_of(ka, struct kvm, arch);
1876 struct kvm_vcpu *vcpu;
1878 kvm_for_each_vcpu(i, vcpu, kvm) {
1879 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1880 kvm_vcpu_kick(vcpu);
1884 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1886 struct kvm *kvm = v->kvm;
1888 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1889 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1890 KVMCLOCK_UPDATE_DELAY);
1893 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1895 static void kvmclock_sync_fn(struct work_struct *work)
1897 struct delayed_work *dwork = to_delayed_work(work);
1898 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1899 kvmclock_sync_work);
1900 struct kvm *kvm = container_of(ka, struct kvm, arch);
1902 if (!kvmclock_periodic_sync)
1905 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1906 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1907 KVMCLOCK_SYNC_PERIOD);
1910 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1912 u64 mcg_cap = vcpu->arch.mcg_cap;
1913 unsigned bank_num = mcg_cap & 0xff;
1916 case MSR_IA32_MCG_STATUS:
1917 vcpu->arch.mcg_status = data;
1919 case MSR_IA32_MCG_CTL:
1920 if (!(mcg_cap & MCG_CTL_P))
1922 if (data != 0 && data != ~(u64)0)
1924 vcpu->arch.mcg_ctl = data;
1927 if (msr >= MSR_IA32_MC0_CTL &&
1928 msr < MSR_IA32_MCx_CTL(bank_num)) {
1929 u32 offset = msr - MSR_IA32_MC0_CTL;
1930 /* only 0 or all 1s can be written to IA32_MCi_CTL
1931 * some Linux kernels though clear bit 10 in bank 4 to
1932 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1933 * this to avoid an uncatched #GP in the guest
1935 if ((offset & 0x3) == 0 &&
1936 data != 0 && (data | (1 << 10)) != ~(u64)0)
1938 vcpu->arch.mce_banks[offset] = data;
1946 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1948 struct kvm *kvm = vcpu->kvm;
1949 int lm = is_long_mode(vcpu);
1950 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1951 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1952 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1953 : kvm->arch.xen_hvm_config.blob_size_32;
1954 u32 page_num = data & ~PAGE_MASK;
1955 u64 page_addr = data & PAGE_MASK;
1960 if (page_num >= blob_size)
1963 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1968 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1977 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1979 gpa_t gpa = data & ~0x3f;
1981 /* Bits 2:5 are reserved, Should be zero */
1985 vcpu->arch.apf.msr_val = data;
1987 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1988 kvm_clear_async_pf_completion_queue(vcpu);
1989 kvm_async_pf_hash_reset(vcpu);
1993 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1997 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1998 kvm_async_pf_wakeup_all(vcpu);
2002 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2004 vcpu->arch.pv_time_enabled = false;
2007 static void record_steal_time(struct kvm_vcpu *vcpu)
2009 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2012 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2013 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2016 if (vcpu->arch.st.steal.version & 1)
2017 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2019 vcpu->arch.st.steal.version += 1;
2021 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2022 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2026 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2027 vcpu->arch.st.last_steal;
2028 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2030 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2031 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2035 vcpu->arch.st.steal.version += 1;
2037 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2038 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2041 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2044 u32 msr = msr_info->index;
2045 u64 data = msr_info->data;
2048 case MSR_AMD64_NB_CFG:
2049 case MSR_IA32_UCODE_REV:
2050 case MSR_IA32_UCODE_WRITE:
2051 case MSR_VM_HSAVE_PA:
2052 case MSR_AMD64_PATCH_LOADER:
2053 case MSR_AMD64_BU_CFG2:
2057 return set_efer(vcpu, data);
2059 data &= ~(u64)0x40; /* ignore flush filter disable */
2060 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2061 data &= ~(u64)0x8; /* ignore TLB cache disable */
2062 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2064 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2069 case MSR_FAM10H_MMIO_CONF_BASE:
2071 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2076 case MSR_IA32_DEBUGCTLMSR:
2078 /* We support the non-activated case already */
2080 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2081 /* Values other than LBR and BTF are vendor-specific,
2082 thus reserved and should throw a #GP */
2085 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2088 case 0x200 ... 0x2ff:
2089 return kvm_mtrr_set_msr(vcpu, msr, data);
2090 case MSR_IA32_APICBASE:
2091 return kvm_set_apic_base(vcpu, msr_info);
2092 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2093 return kvm_x2apic_msr_write(vcpu, msr, data);
2094 case MSR_IA32_TSCDEADLINE:
2095 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2097 case MSR_IA32_TSC_ADJUST:
2098 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2099 if (!msr_info->host_initiated) {
2100 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2101 adjust_tsc_offset_guest(vcpu, adj);
2103 vcpu->arch.ia32_tsc_adjust_msr = data;
2106 case MSR_IA32_MISC_ENABLE:
2107 vcpu->arch.ia32_misc_enable_msr = data;
2109 case MSR_IA32_SMBASE:
2110 if (!msr_info->host_initiated)
2112 vcpu->arch.smbase = data;
2114 case MSR_KVM_WALL_CLOCK_NEW:
2115 case MSR_KVM_WALL_CLOCK:
2116 vcpu->kvm->arch.wall_clock = data;
2117 kvm_write_wall_clock(vcpu->kvm, data);
2119 case MSR_KVM_SYSTEM_TIME_NEW:
2120 case MSR_KVM_SYSTEM_TIME: {
2122 struct kvm_arch *ka = &vcpu->kvm->arch;
2124 kvmclock_reset(vcpu);
2126 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2127 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2129 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2130 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2133 ka->boot_vcpu_runs_old_kvmclock = tmp;
2136 vcpu->arch.time = data;
2137 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2139 /* we verify if the enable bit is set... */
2143 gpa_offset = data & ~(PAGE_MASK | 1);
2145 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2146 &vcpu->arch.pv_time, data & ~1ULL,
2147 sizeof(struct pvclock_vcpu_time_info)))
2148 vcpu->arch.pv_time_enabled = false;
2150 vcpu->arch.pv_time_enabled = true;
2154 case MSR_KVM_ASYNC_PF_EN:
2155 if (kvm_pv_enable_async_pf(vcpu, data))
2158 case MSR_KVM_STEAL_TIME:
2160 if (unlikely(!sched_info_on()))
2163 if (data & KVM_STEAL_RESERVED_MASK)
2166 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2167 data & KVM_STEAL_VALID_BITS,
2168 sizeof(struct kvm_steal_time)))
2171 vcpu->arch.st.msr_val = data;
2173 if (!(data & KVM_MSR_ENABLED))
2176 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2179 case MSR_KVM_PV_EOI_EN:
2180 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2184 case MSR_IA32_MCG_CTL:
2185 case MSR_IA32_MCG_STATUS:
2186 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2187 return set_msr_mce(vcpu, msr, data);
2189 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2190 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2191 pr = true; /* fall through */
2192 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2193 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2194 if (kvm_pmu_is_valid_msr(vcpu, msr))
2195 return kvm_pmu_set_msr(vcpu, msr_info);
2197 if (pr || data != 0)
2198 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2199 "0x%x data 0x%llx\n", msr, data);
2201 case MSR_K7_CLK_CTL:
2203 * Ignore all writes to this no longer documented MSR.
2204 * Writes are only relevant for old K7 processors,
2205 * all pre-dating SVM, but a recommended workaround from
2206 * AMD for these chips. It is possible to specify the
2207 * affected processor models on the command line, hence
2208 * the need to ignore the workaround.
2211 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2212 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2213 case HV_X64_MSR_CRASH_CTL:
2214 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2215 return kvm_hv_set_msr_common(vcpu, msr, data,
2216 msr_info->host_initiated);
2217 case MSR_IA32_BBL_CR_CTL3:
2218 /* Drop writes to this legacy MSR -- see rdmsr
2219 * counterpart for further detail.
2221 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2223 case MSR_AMD64_OSVW_ID_LENGTH:
2224 if (!guest_cpuid_has_osvw(vcpu))
2226 vcpu->arch.osvw.length = data;
2228 case MSR_AMD64_OSVW_STATUS:
2229 if (!guest_cpuid_has_osvw(vcpu))
2231 vcpu->arch.osvw.status = data;
2234 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2235 return xen_hvm_config(vcpu, data);
2236 if (kvm_pmu_is_valid_msr(vcpu, msr))
2237 return kvm_pmu_set_msr(vcpu, msr_info);
2239 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2243 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2250 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2254 * Reads an msr value (of 'msr_index') into 'pdata'.
2255 * Returns 0 on success, non-0 otherwise.
2256 * Assumes vcpu_load() was already called.
2258 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2260 return kvm_x86_ops->get_msr(vcpu, msr);
2262 EXPORT_SYMBOL_GPL(kvm_get_msr);
2264 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2267 u64 mcg_cap = vcpu->arch.mcg_cap;
2268 unsigned bank_num = mcg_cap & 0xff;
2271 case MSR_IA32_P5_MC_ADDR:
2272 case MSR_IA32_P5_MC_TYPE:
2275 case MSR_IA32_MCG_CAP:
2276 data = vcpu->arch.mcg_cap;
2278 case MSR_IA32_MCG_CTL:
2279 if (!(mcg_cap & MCG_CTL_P))
2281 data = vcpu->arch.mcg_ctl;
2283 case MSR_IA32_MCG_STATUS:
2284 data = vcpu->arch.mcg_status;
2287 if (msr >= MSR_IA32_MC0_CTL &&
2288 msr < MSR_IA32_MCx_CTL(bank_num)) {
2289 u32 offset = msr - MSR_IA32_MC0_CTL;
2290 data = vcpu->arch.mce_banks[offset];
2299 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2301 switch (msr_info->index) {
2302 case MSR_IA32_PLATFORM_ID:
2303 case MSR_IA32_EBL_CR_POWERON:
2304 case MSR_IA32_DEBUGCTLMSR:
2305 case MSR_IA32_LASTBRANCHFROMIP:
2306 case MSR_IA32_LASTBRANCHTOIP:
2307 case MSR_IA32_LASTINTFROMIP:
2308 case MSR_IA32_LASTINTTOIP:
2310 case MSR_K8_TSEG_ADDR:
2311 case MSR_K8_TSEG_MASK:
2313 case MSR_VM_HSAVE_PA:
2314 case MSR_K8_INT_PENDING_MSG:
2315 case MSR_AMD64_NB_CFG:
2316 case MSR_FAM10H_MMIO_CONF_BASE:
2317 case MSR_AMD64_BU_CFG2:
2318 case MSR_IA32_PERF_CTL:
2321 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2322 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2323 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2324 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2325 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2326 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2329 case MSR_IA32_UCODE_REV:
2330 msr_info->data = 0x100000000ULL;
2333 case 0x200 ... 0x2ff:
2334 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2335 case 0xcd: /* fsb frequency */
2339 * MSR_EBC_FREQUENCY_ID
2340 * Conservative value valid for even the basic CPU models.
2341 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2342 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2343 * and 266MHz for model 3, or 4. Set Core Clock
2344 * Frequency to System Bus Frequency Ratio to 1 (bits
2345 * 31:24) even though these are only valid for CPU
2346 * models > 2, however guests may end up dividing or
2347 * multiplying by zero otherwise.
2349 case MSR_EBC_FREQUENCY_ID:
2350 msr_info->data = 1 << 24;
2352 case MSR_IA32_APICBASE:
2353 msr_info->data = kvm_get_apic_base(vcpu);
2355 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2356 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2358 case MSR_IA32_TSCDEADLINE:
2359 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2361 case MSR_IA32_TSC_ADJUST:
2362 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2364 case MSR_IA32_MISC_ENABLE:
2365 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2367 case MSR_IA32_SMBASE:
2368 if (!msr_info->host_initiated)
2370 msr_info->data = vcpu->arch.smbase;
2372 case MSR_IA32_PERF_STATUS:
2373 /* TSC increment by tick */
2374 msr_info->data = 1000ULL;
2375 /* CPU multiplier */
2376 msr_info->data |= (((uint64_t)4ULL) << 40);
2379 msr_info->data = vcpu->arch.efer;
2381 case MSR_KVM_WALL_CLOCK:
2382 case MSR_KVM_WALL_CLOCK_NEW:
2383 msr_info->data = vcpu->kvm->arch.wall_clock;
2385 case MSR_KVM_SYSTEM_TIME:
2386 case MSR_KVM_SYSTEM_TIME_NEW:
2387 msr_info->data = vcpu->arch.time;
2389 case MSR_KVM_ASYNC_PF_EN:
2390 msr_info->data = vcpu->arch.apf.msr_val;
2392 case MSR_KVM_STEAL_TIME:
2393 msr_info->data = vcpu->arch.st.msr_val;
2395 case MSR_KVM_PV_EOI_EN:
2396 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2398 case MSR_IA32_P5_MC_ADDR:
2399 case MSR_IA32_P5_MC_TYPE:
2400 case MSR_IA32_MCG_CAP:
2401 case MSR_IA32_MCG_CTL:
2402 case MSR_IA32_MCG_STATUS:
2403 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2404 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2405 case MSR_K7_CLK_CTL:
2407 * Provide expected ramp-up count for K7. All other
2408 * are set to zero, indicating minimum divisors for
2411 * This prevents guest kernels on AMD host with CPU
2412 * type 6, model 8 and higher from exploding due to
2413 * the rdmsr failing.
2415 msr_info->data = 0x20000000;
2417 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2418 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2419 case HV_X64_MSR_CRASH_CTL:
2420 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2421 return kvm_hv_get_msr_common(vcpu,
2422 msr_info->index, &msr_info->data);
2424 case MSR_IA32_BBL_CR_CTL3:
2425 /* This legacy MSR exists but isn't fully documented in current
2426 * silicon. It is however accessed by winxp in very narrow
2427 * scenarios where it sets bit #19, itself documented as
2428 * a "reserved" bit. Best effort attempt to source coherent
2429 * read data here should the balance of the register be
2430 * interpreted by the guest:
2432 * L2 cache control register 3: 64GB range, 256KB size,
2433 * enabled, latency 0x1, configured
2435 msr_info->data = 0xbe702111;
2437 case MSR_AMD64_OSVW_ID_LENGTH:
2438 if (!guest_cpuid_has_osvw(vcpu))
2440 msr_info->data = vcpu->arch.osvw.length;
2442 case MSR_AMD64_OSVW_STATUS:
2443 if (!guest_cpuid_has_osvw(vcpu))
2445 msr_info->data = vcpu->arch.osvw.status;
2448 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2449 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2451 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2454 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2461 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2464 * Read or write a bunch of msrs. All parameters are kernel addresses.
2466 * @return number of msrs set successfully.
2468 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2469 struct kvm_msr_entry *entries,
2470 int (*do_msr)(struct kvm_vcpu *vcpu,
2471 unsigned index, u64 *data))
2475 idx = srcu_read_lock(&vcpu->kvm->srcu);
2476 for (i = 0; i < msrs->nmsrs; ++i)
2477 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2479 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2485 * Read or write a bunch of msrs. Parameters are user addresses.
2487 * @return number of msrs set successfully.
2489 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2490 int (*do_msr)(struct kvm_vcpu *vcpu,
2491 unsigned index, u64 *data),
2494 struct kvm_msrs msrs;
2495 struct kvm_msr_entry *entries;
2500 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2504 if (msrs.nmsrs >= MAX_IO_MSRS)
2507 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2508 entries = memdup_user(user_msrs->entries, size);
2509 if (IS_ERR(entries)) {
2510 r = PTR_ERR(entries);
2514 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2519 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2530 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2535 case KVM_CAP_IRQCHIP:
2537 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2538 case KVM_CAP_SET_TSS_ADDR:
2539 case KVM_CAP_EXT_CPUID:
2540 case KVM_CAP_EXT_EMUL_CPUID:
2541 case KVM_CAP_CLOCKSOURCE:
2543 case KVM_CAP_NOP_IO_DELAY:
2544 case KVM_CAP_MP_STATE:
2545 case KVM_CAP_SYNC_MMU:
2546 case KVM_CAP_USER_NMI:
2547 case KVM_CAP_REINJECT_CONTROL:
2548 case KVM_CAP_IRQ_INJECT_STATUS:
2549 case KVM_CAP_IOEVENTFD:
2550 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2552 case KVM_CAP_PIT_STATE2:
2553 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2554 case KVM_CAP_XEN_HVM:
2555 case KVM_CAP_ADJUST_CLOCK:
2556 case KVM_CAP_VCPU_EVENTS:
2557 case KVM_CAP_HYPERV:
2558 case KVM_CAP_HYPERV_VAPIC:
2559 case KVM_CAP_HYPERV_SPIN:
2560 case KVM_CAP_HYPERV_SYNIC:
2561 case KVM_CAP_PCI_SEGMENT:
2562 case KVM_CAP_DEBUGREGS:
2563 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2565 case KVM_CAP_ASYNC_PF:
2566 case KVM_CAP_GET_TSC_KHZ:
2567 case KVM_CAP_KVMCLOCK_CTRL:
2568 case KVM_CAP_READONLY_MEM:
2569 case KVM_CAP_HYPERV_TIME:
2570 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2571 case KVM_CAP_TSC_DEADLINE_TIMER:
2572 case KVM_CAP_ENABLE_CAP_VM:
2573 case KVM_CAP_DISABLE_QUIRKS:
2574 case KVM_CAP_SET_BOOT_CPU_ID:
2575 case KVM_CAP_SPLIT_IRQCHIP:
2576 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2577 case KVM_CAP_ASSIGN_DEV_IRQ:
2578 case KVM_CAP_PCI_2_3:
2582 case KVM_CAP_X86_SMM:
2583 /* SMBASE is usually relocated above 1M on modern chipsets,
2584 * and SMM handlers might indeed rely on 4G segment limits,
2585 * so do not report SMM to be available if real mode is
2586 * emulated via vm86 mode. Still, do not go to great lengths
2587 * to avoid userspace's usage of the feature, because it is a
2588 * fringe case that is not enabled except via specific settings
2589 * of the module parameters.
2591 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2593 case KVM_CAP_COALESCED_MMIO:
2594 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2597 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2599 case KVM_CAP_NR_VCPUS:
2600 r = KVM_SOFT_MAX_VCPUS;
2602 case KVM_CAP_MAX_VCPUS:
2605 case KVM_CAP_NR_MEMSLOTS:
2606 r = KVM_USER_MEM_SLOTS;
2608 case KVM_CAP_PV_MMU: /* obsolete */
2611 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2613 r = iommu_present(&pci_bus_type);
2617 r = KVM_MAX_MCE_BANKS;
2620 r = boot_cpu_has(X86_FEATURE_XSAVE);
2622 case KVM_CAP_TSC_CONTROL:
2623 r = kvm_has_tsc_control;
2633 long kvm_arch_dev_ioctl(struct file *filp,
2634 unsigned int ioctl, unsigned long arg)
2636 void __user *argp = (void __user *)arg;
2640 case KVM_GET_MSR_INDEX_LIST: {
2641 struct kvm_msr_list __user *user_msr_list = argp;
2642 struct kvm_msr_list msr_list;
2646 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2649 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2650 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2653 if (n < msr_list.nmsrs)
2656 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2657 num_msrs_to_save * sizeof(u32)))
2659 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2661 num_emulated_msrs * sizeof(u32)))
2666 case KVM_GET_SUPPORTED_CPUID:
2667 case KVM_GET_EMULATED_CPUID: {
2668 struct kvm_cpuid2 __user *cpuid_arg = argp;
2669 struct kvm_cpuid2 cpuid;
2672 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2675 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2681 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2686 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2689 mce_cap = KVM_MCE_CAP_SUPPORTED;
2691 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2703 static void wbinvd_ipi(void *garbage)
2708 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2710 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2713 static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
2715 set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
2718 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2720 /* Address WBINVD may be executed by guest */
2721 if (need_emulate_wbinvd(vcpu)) {
2722 if (kvm_x86_ops->has_wbinvd_exit())
2723 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2724 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2725 smp_call_function_single(vcpu->cpu,
2726 wbinvd_ipi, NULL, 1);
2729 kvm_x86_ops->vcpu_load(vcpu, cpu);
2731 /* Apply any externally detected TSC adjustments (due to suspend) */
2732 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2733 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2734 vcpu->arch.tsc_offset_adjustment = 0;
2735 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2738 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2739 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2740 rdtsc() - vcpu->arch.last_host_tsc;
2742 mark_tsc_unstable("KVM discovered backwards TSC");
2743 if (check_tsc_unstable()) {
2744 u64 offset = kvm_compute_tsc_offset(vcpu,
2745 vcpu->arch.last_guest_tsc);
2746 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2747 vcpu->arch.tsc_catchup = 1;
2750 * On a host with synchronized TSC, there is no need to update
2751 * kvmclock on vcpu->cpu migration
2753 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2754 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2755 if (vcpu->cpu != cpu)
2756 kvm_migrate_timers(vcpu);
2760 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2763 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2765 kvm_x86_ops->vcpu_put(vcpu);
2766 kvm_put_guest_fpu(vcpu);
2767 vcpu->arch.last_host_tsc = rdtsc();
2770 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2771 struct kvm_lapic_state *s)
2773 if (vcpu->arch.apicv_active)
2774 kvm_x86_ops->sync_pir_to_irr(vcpu);
2776 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2781 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2782 struct kvm_lapic_state *s)
2784 kvm_apic_post_state_restore(vcpu, s);
2785 update_cr8_intercept(vcpu);
2790 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2792 return (!lapic_in_kernel(vcpu) ||
2793 kvm_apic_accept_pic_intr(vcpu));
2797 * if userspace requested an interrupt window, check that the
2798 * interrupt window is open.
2800 * No need to exit to userspace if we already have an interrupt queued.
2802 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2804 return kvm_arch_interrupt_allowed(vcpu) &&
2805 !kvm_cpu_has_interrupt(vcpu) &&
2806 !kvm_event_needs_reinjection(vcpu) &&
2807 kvm_cpu_accept_dm_intr(vcpu);
2810 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2811 struct kvm_interrupt *irq)
2813 if (irq->irq >= KVM_NR_INTERRUPTS)
2816 if (!irqchip_in_kernel(vcpu->kvm)) {
2817 kvm_queue_interrupt(vcpu, irq->irq, false);
2818 kvm_make_request(KVM_REQ_EVENT, vcpu);
2823 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2824 * fail for in-kernel 8259.
2826 if (pic_in_kernel(vcpu->kvm))
2829 if (vcpu->arch.pending_external_vector != -1)
2832 vcpu->arch.pending_external_vector = irq->irq;
2833 kvm_make_request(KVM_REQ_EVENT, vcpu);
2837 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2839 kvm_inject_nmi(vcpu);
2844 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2846 kvm_make_request(KVM_REQ_SMI, vcpu);
2851 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2852 struct kvm_tpr_access_ctl *tac)
2856 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2860 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2864 unsigned bank_num = mcg_cap & 0xff, bank;
2867 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2869 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2872 vcpu->arch.mcg_cap = mcg_cap;
2873 /* Init IA32_MCG_CTL to all 1s */
2874 if (mcg_cap & MCG_CTL_P)
2875 vcpu->arch.mcg_ctl = ~(u64)0;
2876 /* Init IA32_MCi_CTL to all 1s */
2877 for (bank = 0; bank < bank_num; bank++)
2878 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2883 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2884 struct kvm_x86_mce *mce)
2886 u64 mcg_cap = vcpu->arch.mcg_cap;
2887 unsigned bank_num = mcg_cap & 0xff;
2888 u64 *banks = vcpu->arch.mce_banks;
2890 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2893 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2894 * reporting is disabled
2896 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2897 vcpu->arch.mcg_ctl != ~(u64)0)
2899 banks += 4 * mce->bank;
2901 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2902 * reporting is disabled for the bank
2904 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2906 if (mce->status & MCI_STATUS_UC) {
2907 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2908 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2909 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2912 if (banks[1] & MCI_STATUS_VAL)
2913 mce->status |= MCI_STATUS_OVER;
2914 banks[2] = mce->addr;
2915 banks[3] = mce->misc;
2916 vcpu->arch.mcg_status = mce->mcg_status;
2917 banks[1] = mce->status;
2918 kvm_queue_exception(vcpu, MC_VECTOR);
2919 } else if (!(banks[1] & MCI_STATUS_VAL)
2920 || !(banks[1] & MCI_STATUS_UC)) {
2921 if (banks[1] & MCI_STATUS_VAL)
2922 mce->status |= MCI_STATUS_OVER;
2923 banks[2] = mce->addr;
2924 banks[3] = mce->misc;
2925 banks[1] = mce->status;
2927 banks[1] |= MCI_STATUS_OVER;
2931 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2932 struct kvm_vcpu_events *events)
2935 events->exception.injected =
2936 vcpu->arch.exception.pending &&
2937 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2938 events->exception.nr = vcpu->arch.exception.nr;
2939 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2940 events->exception.pad = 0;
2941 events->exception.error_code = vcpu->arch.exception.error_code;
2943 events->interrupt.injected =
2944 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2945 events->interrupt.nr = vcpu->arch.interrupt.nr;
2946 events->interrupt.soft = 0;
2947 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2949 events->nmi.injected = vcpu->arch.nmi_injected;
2950 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2951 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2952 events->nmi.pad = 0;
2954 events->sipi_vector = 0; /* never valid when reporting to user space */
2956 events->smi.smm = is_smm(vcpu);
2957 events->smi.pending = vcpu->arch.smi_pending;
2958 events->smi.smm_inside_nmi =
2959 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2960 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2962 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2963 | KVM_VCPUEVENT_VALID_SHADOW
2964 | KVM_VCPUEVENT_VALID_SMM);
2965 memset(&events->reserved, 0, sizeof(events->reserved));
2968 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2969 struct kvm_vcpu_events *events)
2971 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2972 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2973 | KVM_VCPUEVENT_VALID_SHADOW
2974 | KVM_VCPUEVENT_VALID_SMM))
2977 if (events->exception.injected &&
2978 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
2982 vcpu->arch.exception.pending = events->exception.injected;
2983 vcpu->arch.exception.nr = events->exception.nr;
2984 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2985 vcpu->arch.exception.error_code = events->exception.error_code;
2987 vcpu->arch.interrupt.pending = events->interrupt.injected;
2988 vcpu->arch.interrupt.nr = events->interrupt.nr;
2989 vcpu->arch.interrupt.soft = events->interrupt.soft;
2990 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2991 kvm_x86_ops->set_interrupt_shadow(vcpu,
2992 events->interrupt.shadow);
2994 vcpu->arch.nmi_injected = events->nmi.injected;
2995 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2996 vcpu->arch.nmi_pending = events->nmi.pending;
2997 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2999 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3000 lapic_in_kernel(vcpu))
3001 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3003 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3004 if (events->smi.smm)
3005 vcpu->arch.hflags |= HF_SMM_MASK;
3007 vcpu->arch.hflags &= ~HF_SMM_MASK;
3008 vcpu->arch.smi_pending = events->smi.pending;
3009 if (events->smi.smm_inside_nmi)
3010 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3012 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3013 if (lapic_in_kernel(vcpu)) {
3014 if (events->smi.latched_init)
3015 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3017 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3021 kvm_make_request(KVM_REQ_EVENT, vcpu);
3026 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3027 struct kvm_debugregs *dbgregs)
3031 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3032 kvm_get_dr(vcpu, 6, &val);
3034 dbgregs->dr7 = vcpu->arch.dr7;
3036 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3039 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3040 struct kvm_debugregs *dbgregs)
3045 if (dbgregs->dr6 & ~0xffffffffull)
3047 if (dbgregs->dr7 & ~0xffffffffull)
3050 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3051 kvm_update_dr0123(vcpu);
3052 vcpu->arch.dr6 = dbgregs->dr6;
3053 kvm_update_dr6(vcpu);
3054 vcpu->arch.dr7 = dbgregs->dr7;
3055 kvm_update_dr7(vcpu);
3060 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3062 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3064 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3065 u64 xstate_bv = xsave->header.xfeatures;
3069 * Copy legacy XSAVE area, to avoid complications with CPUID
3070 * leaves 0 and 1 in the loop below.
3072 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3075 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3078 * Copy each region from the possibly compacted offset to the
3079 * non-compacted offset.
3081 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3083 u64 feature = valid & -valid;
3084 int index = fls64(feature) - 1;
3085 void *src = get_xsave_addr(xsave, feature);
3088 u32 size, offset, ecx, edx;
3089 cpuid_count(XSTATE_CPUID, index,
3090 &size, &offset, &ecx, &edx);
3091 memcpy(dest + offset, src, size);
3098 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3100 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3101 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3105 * Copy legacy XSAVE area, to avoid complications with CPUID
3106 * leaves 0 and 1 in the loop below.
3108 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3110 /* Set XSTATE_BV and possibly XCOMP_BV. */
3111 xsave->header.xfeatures = xstate_bv;
3112 if (boot_cpu_has(X86_FEATURE_XSAVES))
3113 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3116 * Copy each region from the non-compacted offset to the
3117 * possibly compacted offset.
3119 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3121 u64 feature = valid & -valid;
3122 int index = fls64(feature) - 1;
3123 void *dest = get_xsave_addr(xsave, feature);
3126 u32 size, offset, ecx, edx;
3127 cpuid_count(XSTATE_CPUID, index,
3128 &size, &offset, &ecx, &edx);
3129 memcpy(dest, src + offset, size);
3136 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3137 struct kvm_xsave *guest_xsave)
3139 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3140 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3141 fill_xsave((u8 *) guest_xsave->region, vcpu);
3143 memcpy(guest_xsave->region,
3144 &vcpu->arch.guest_fpu.state.fxsave,
3145 sizeof(struct fxregs_state));
3146 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3147 XFEATURE_MASK_FPSSE;
3151 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3152 struct kvm_xsave *guest_xsave)
3155 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3157 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3159 * Here we allow setting states that are not present in
3160 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3161 * with old userspace.
3163 if (xstate_bv & ~kvm_supported_xcr0())
3165 load_xsave(vcpu, (u8 *)guest_xsave->region);
3167 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3169 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3170 guest_xsave->region, sizeof(struct fxregs_state));
3175 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3176 struct kvm_xcrs *guest_xcrs)
3178 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3179 guest_xcrs->nr_xcrs = 0;
3183 guest_xcrs->nr_xcrs = 1;
3184 guest_xcrs->flags = 0;
3185 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3186 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3189 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3190 struct kvm_xcrs *guest_xcrs)
3194 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3197 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3200 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3201 /* Only support XCR0 currently */
3202 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3203 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3204 guest_xcrs->xcrs[i].value);
3213 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3214 * stopped by the hypervisor. This function will be called from the host only.
3215 * EINVAL is returned when the host attempts to set the flag for a guest that
3216 * does not support pv clocks.
3218 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3220 if (!vcpu->arch.pv_time_enabled)
3222 vcpu->arch.pvclock_set_guest_stopped_request = true;
3223 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3227 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3228 struct kvm_enable_cap *cap)
3234 case KVM_CAP_HYPERV_SYNIC:
3235 return kvm_hv_activate_synic(vcpu);
3241 long kvm_arch_vcpu_ioctl(struct file *filp,
3242 unsigned int ioctl, unsigned long arg)
3244 struct kvm_vcpu *vcpu = filp->private_data;
3245 void __user *argp = (void __user *)arg;
3248 struct kvm_lapic_state *lapic;
3249 struct kvm_xsave *xsave;
3250 struct kvm_xcrs *xcrs;
3256 case KVM_GET_LAPIC: {
3258 if (!lapic_in_kernel(vcpu))
3260 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3265 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3269 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3274 case KVM_SET_LAPIC: {
3276 if (!lapic_in_kernel(vcpu))
3278 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3279 if (IS_ERR(u.lapic))
3280 return PTR_ERR(u.lapic);
3282 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3285 case KVM_INTERRUPT: {
3286 struct kvm_interrupt irq;
3289 if (copy_from_user(&irq, argp, sizeof irq))
3291 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3295 r = kvm_vcpu_ioctl_nmi(vcpu);
3299 r = kvm_vcpu_ioctl_smi(vcpu);
3302 case KVM_SET_CPUID: {
3303 struct kvm_cpuid __user *cpuid_arg = argp;
3304 struct kvm_cpuid cpuid;
3307 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3309 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3312 case KVM_SET_CPUID2: {
3313 struct kvm_cpuid2 __user *cpuid_arg = argp;
3314 struct kvm_cpuid2 cpuid;
3317 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3319 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3320 cpuid_arg->entries);
3323 case KVM_GET_CPUID2: {
3324 struct kvm_cpuid2 __user *cpuid_arg = argp;
3325 struct kvm_cpuid2 cpuid;
3328 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3330 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3331 cpuid_arg->entries);
3335 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3341 r = msr_io(vcpu, argp, do_get_msr, 1);
3344 r = msr_io(vcpu, argp, do_set_msr, 0);
3346 case KVM_TPR_ACCESS_REPORTING: {
3347 struct kvm_tpr_access_ctl tac;
3350 if (copy_from_user(&tac, argp, sizeof tac))
3352 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3356 if (copy_to_user(argp, &tac, sizeof tac))
3361 case KVM_SET_VAPIC_ADDR: {
3362 struct kvm_vapic_addr va;
3365 if (!lapic_in_kernel(vcpu))
3368 if (copy_from_user(&va, argp, sizeof va))
3370 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3373 case KVM_X86_SETUP_MCE: {
3377 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3379 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3382 case KVM_X86_SET_MCE: {
3383 struct kvm_x86_mce mce;
3386 if (copy_from_user(&mce, argp, sizeof mce))
3388 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3391 case KVM_GET_VCPU_EVENTS: {
3392 struct kvm_vcpu_events events;
3394 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3397 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3402 case KVM_SET_VCPU_EVENTS: {
3403 struct kvm_vcpu_events events;
3406 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3409 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3412 case KVM_GET_DEBUGREGS: {
3413 struct kvm_debugregs dbgregs;
3415 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3418 if (copy_to_user(argp, &dbgregs,
3419 sizeof(struct kvm_debugregs)))
3424 case KVM_SET_DEBUGREGS: {
3425 struct kvm_debugregs dbgregs;
3428 if (copy_from_user(&dbgregs, argp,
3429 sizeof(struct kvm_debugregs)))
3432 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3435 case KVM_GET_XSAVE: {
3436 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3441 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3444 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3449 case KVM_SET_XSAVE: {
3450 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3451 if (IS_ERR(u.xsave))
3452 return PTR_ERR(u.xsave);
3454 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3457 case KVM_GET_XCRS: {
3458 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3463 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3466 if (copy_to_user(argp, u.xcrs,
3467 sizeof(struct kvm_xcrs)))
3472 case KVM_SET_XCRS: {
3473 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3475 return PTR_ERR(u.xcrs);
3477 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3480 case KVM_SET_TSC_KHZ: {
3484 user_tsc_khz = (u32)arg;
3486 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3489 if (user_tsc_khz == 0)
3490 user_tsc_khz = tsc_khz;
3492 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3497 case KVM_GET_TSC_KHZ: {
3498 r = vcpu->arch.virtual_tsc_khz;
3501 case KVM_KVMCLOCK_CTRL: {
3502 r = kvm_set_guest_paused(vcpu);
3505 case KVM_ENABLE_CAP: {
3506 struct kvm_enable_cap cap;
3509 if (copy_from_user(&cap, argp, sizeof(cap)))
3511 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3522 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3524 return VM_FAULT_SIGBUS;
3527 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3531 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3533 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3537 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3540 kvm->arch.ept_identity_map_addr = ident_addr;
3544 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3545 u32 kvm_nr_mmu_pages)
3547 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3550 mutex_lock(&kvm->slots_lock);
3552 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3553 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3555 mutex_unlock(&kvm->slots_lock);
3559 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3561 return kvm->arch.n_max_mmu_pages;
3564 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3569 switch (chip->chip_id) {
3570 case KVM_IRQCHIP_PIC_MASTER:
3571 memcpy(&chip->chip.pic,
3572 &pic_irqchip(kvm)->pics[0],
3573 sizeof(struct kvm_pic_state));
3575 case KVM_IRQCHIP_PIC_SLAVE:
3576 memcpy(&chip->chip.pic,
3577 &pic_irqchip(kvm)->pics[1],
3578 sizeof(struct kvm_pic_state));
3580 case KVM_IRQCHIP_IOAPIC:
3581 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3590 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3595 switch (chip->chip_id) {
3596 case KVM_IRQCHIP_PIC_MASTER:
3597 spin_lock(&pic_irqchip(kvm)->lock);
3598 memcpy(&pic_irqchip(kvm)->pics[0],
3600 sizeof(struct kvm_pic_state));
3601 spin_unlock(&pic_irqchip(kvm)->lock);
3603 case KVM_IRQCHIP_PIC_SLAVE:
3604 spin_lock(&pic_irqchip(kvm)->lock);
3605 memcpy(&pic_irqchip(kvm)->pics[1],
3607 sizeof(struct kvm_pic_state));
3608 spin_unlock(&pic_irqchip(kvm)->lock);
3610 case KVM_IRQCHIP_IOAPIC:
3611 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3617 kvm_pic_update_irq(pic_irqchip(kvm));
3621 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3623 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3625 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3627 mutex_lock(&kps->lock);
3628 memcpy(ps, &kps->channels, sizeof(*ps));
3629 mutex_unlock(&kps->lock);
3633 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3636 struct kvm_pit *pit = kvm->arch.vpit;
3638 mutex_lock(&pit->pit_state.lock);
3639 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3640 for (i = 0; i < 3; i++)
3641 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3642 mutex_unlock(&pit->pit_state.lock);
3646 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3648 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3649 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3650 sizeof(ps->channels));
3651 ps->flags = kvm->arch.vpit->pit_state.flags;
3652 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3653 memset(&ps->reserved, 0, sizeof(ps->reserved));
3657 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3661 u32 prev_legacy, cur_legacy;
3662 struct kvm_pit *pit = kvm->arch.vpit;
3664 mutex_lock(&pit->pit_state.lock);
3665 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3666 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3667 if (!prev_legacy && cur_legacy)
3669 memcpy(&pit->pit_state.channels, &ps->channels,
3670 sizeof(pit->pit_state.channels));
3671 pit->pit_state.flags = ps->flags;
3672 for (i = 0; i < 3; i++)
3673 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3675 mutex_unlock(&pit->pit_state.lock);
3679 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3680 struct kvm_reinject_control *control)
3682 struct kvm_pit *pit = kvm->arch.vpit;
3687 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3688 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3689 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3691 mutex_lock(&pit->pit_state.lock);
3692 kvm_pit_set_reinject(pit, control->pit_reinject);
3693 mutex_unlock(&pit->pit_state.lock);
3699 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3700 * @kvm: kvm instance
3701 * @log: slot id and address to which we copy the log
3703 * Steps 1-4 below provide general overview of dirty page logging. See
3704 * kvm_get_dirty_log_protect() function description for additional details.
3706 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3707 * always flush the TLB (step 4) even if previous step failed and the dirty
3708 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3709 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3710 * writes will be marked dirty for next log read.
3712 * 1. Take a snapshot of the bit and clear it if needed.
3713 * 2. Write protect the corresponding page.
3714 * 3. Copy the snapshot to the userspace.
3715 * 4. Flush TLB's if needed.
3717 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3719 bool is_dirty = false;
3722 mutex_lock(&kvm->slots_lock);
3725 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3727 if (kvm_x86_ops->flush_log_dirty)
3728 kvm_x86_ops->flush_log_dirty(kvm);
3730 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3733 * All the TLBs can be flushed out of mmu lock, see the comments in
3734 * kvm_mmu_slot_remove_write_access().
3736 lockdep_assert_held(&kvm->slots_lock);
3738 kvm_flush_remote_tlbs(kvm);
3740 mutex_unlock(&kvm->slots_lock);
3744 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3747 if (!irqchip_in_kernel(kvm))
3750 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3751 irq_event->irq, irq_event->level,
3756 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3757 struct kvm_enable_cap *cap)
3765 case KVM_CAP_DISABLE_QUIRKS:
3766 kvm->arch.disabled_quirks = cap->args[0];
3769 case KVM_CAP_SPLIT_IRQCHIP: {
3770 mutex_lock(&kvm->lock);
3772 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3773 goto split_irqchip_unlock;
3775 if (irqchip_in_kernel(kvm))
3776 goto split_irqchip_unlock;
3777 if (atomic_read(&kvm->online_vcpus))
3778 goto split_irqchip_unlock;
3779 r = kvm_setup_empty_irq_routing(kvm);
3781 goto split_irqchip_unlock;
3782 /* Pairs with irqchip_in_kernel. */
3784 kvm->arch.irqchip_split = true;
3785 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3787 split_irqchip_unlock:
3788 mutex_unlock(&kvm->lock);
3798 long kvm_arch_vm_ioctl(struct file *filp,
3799 unsigned int ioctl, unsigned long arg)
3801 struct kvm *kvm = filp->private_data;
3802 void __user *argp = (void __user *)arg;
3805 * This union makes it completely explicit to gcc-3.x
3806 * that these two variables' stack usage should be
3807 * combined, not added together.
3810 struct kvm_pit_state ps;
3811 struct kvm_pit_state2 ps2;
3812 struct kvm_pit_config pit_config;
3816 case KVM_SET_TSS_ADDR:
3817 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3819 case KVM_SET_IDENTITY_MAP_ADDR: {
3823 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3825 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3828 case KVM_SET_NR_MMU_PAGES:
3829 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3831 case KVM_GET_NR_MMU_PAGES:
3832 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3834 case KVM_CREATE_IRQCHIP: {
3835 struct kvm_pic *vpic;
3837 mutex_lock(&kvm->lock);
3840 goto create_irqchip_unlock;
3842 if (atomic_read(&kvm->online_vcpus))
3843 goto create_irqchip_unlock;
3845 vpic = kvm_create_pic(kvm);
3847 r = kvm_ioapic_init(kvm);
3849 mutex_lock(&kvm->slots_lock);
3850 kvm_destroy_pic(vpic);
3851 mutex_unlock(&kvm->slots_lock);
3852 goto create_irqchip_unlock;
3855 goto create_irqchip_unlock;
3856 r = kvm_setup_default_irq_routing(kvm);
3858 mutex_lock(&kvm->slots_lock);
3859 mutex_lock(&kvm->irq_lock);
3860 kvm_ioapic_destroy(kvm);
3861 kvm_destroy_pic(vpic);
3862 mutex_unlock(&kvm->irq_lock);
3863 mutex_unlock(&kvm->slots_lock);
3864 goto create_irqchip_unlock;
3866 /* Write kvm->irq_routing before kvm->arch.vpic. */
3868 kvm->arch.vpic = vpic;
3869 create_irqchip_unlock:
3870 mutex_unlock(&kvm->lock);
3873 case KVM_CREATE_PIT:
3874 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3876 case KVM_CREATE_PIT2:
3878 if (copy_from_user(&u.pit_config, argp,
3879 sizeof(struct kvm_pit_config)))
3882 mutex_lock(&kvm->lock);
3885 goto create_pit_unlock;
3887 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3891 mutex_unlock(&kvm->lock);
3893 case KVM_GET_IRQCHIP: {
3894 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3895 struct kvm_irqchip *chip;
3897 chip = memdup_user(argp, sizeof(*chip));
3904 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3905 goto get_irqchip_out;
3906 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3908 goto get_irqchip_out;
3910 if (copy_to_user(argp, chip, sizeof *chip))
3911 goto get_irqchip_out;
3917 case KVM_SET_IRQCHIP: {
3918 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3919 struct kvm_irqchip *chip;
3921 chip = memdup_user(argp, sizeof(*chip));
3928 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3929 goto set_irqchip_out;
3930 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3932 goto set_irqchip_out;
3940 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3943 if (!kvm->arch.vpit)
3945 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3949 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3956 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3959 if (!kvm->arch.vpit)
3961 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3964 case KVM_GET_PIT2: {
3966 if (!kvm->arch.vpit)
3968 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3972 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3977 case KVM_SET_PIT2: {
3979 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3982 if (!kvm->arch.vpit)
3984 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3987 case KVM_REINJECT_CONTROL: {
3988 struct kvm_reinject_control control;
3990 if (copy_from_user(&control, argp, sizeof(control)))
3992 r = kvm_vm_ioctl_reinject(kvm, &control);
3995 case KVM_SET_BOOT_CPU_ID:
3997 mutex_lock(&kvm->lock);
3998 if (atomic_read(&kvm->online_vcpus) != 0)
4001 kvm->arch.bsp_vcpu_id = arg;
4002 mutex_unlock(&kvm->lock);
4004 case KVM_XEN_HVM_CONFIG: {
4006 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4007 sizeof(struct kvm_xen_hvm_config)))
4010 if (kvm->arch.xen_hvm_config.flags)
4015 case KVM_SET_CLOCK: {
4016 struct kvm_clock_data user_ns;
4021 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4029 local_irq_disable();
4030 now_ns = get_kernel_ns();
4031 delta = user_ns.clock - now_ns;
4033 kvm->arch.kvmclock_offset = delta;
4034 kvm_gen_update_masterclock(kvm);
4037 case KVM_GET_CLOCK: {
4038 struct kvm_clock_data user_ns;
4041 local_irq_disable();
4042 now_ns = get_kernel_ns();
4043 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4046 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4049 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4054 case KVM_ENABLE_CAP: {
4055 struct kvm_enable_cap cap;
4058 if (copy_from_user(&cap, argp, sizeof(cap)))
4060 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4064 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4070 static void kvm_init_msr_list(void)
4075 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4076 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4080 * Even MSRs that are valid in the host may not be exposed
4081 * to the guests in some cases.
4083 switch (msrs_to_save[i]) {
4084 case MSR_IA32_BNDCFGS:
4085 if (!kvm_x86_ops->mpx_supported())
4089 if (!kvm_x86_ops->rdtscp_supported())
4097 msrs_to_save[j] = msrs_to_save[i];
4100 num_msrs_to_save = j;
4102 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4103 switch (emulated_msrs[i]) {
4104 case MSR_IA32_SMBASE:
4105 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4113 emulated_msrs[j] = emulated_msrs[i];
4116 num_emulated_msrs = j;
4119 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4127 if (!(lapic_in_kernel(vcpu) &&
4128 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4129 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4140 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4147 if (!(lapic_in_kernel(vcpu) &&
4148 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4150 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4152 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4162 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4163 struct kvm_segment *var, int seg)
4165 kvm_x86_ops->set_segment(vcpu, var, seg);
4168 void kvm_get_segment(struct kvm_vcpu *vcpu,
4169 struct kvm_segment *var, int seg)
4171 kvm_x86_ops->get_segment(vcpu, var, seg);
4174 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4175 struct x86_exception *exception)
4179 BUG_ON(!mmu_is_nested(vcpu));
4181 /* NPT walks are always user-walks */
4182 access |= PFERR_USER_MASK;
4183 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4188 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4189 struct x86_exception *exception)
4191 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4192 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4195 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4196 struct x86_exception *exception)
4198 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4199 access |= PFERR_FETCH_MASK;
4200 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4203 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4204 struct x86_exception *exception)
4206 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4207 access |= PFERR_WRITE_MASK;
4208 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4211 /* uses this to access any guest's mapped memory without checking CPL */
4212 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4213 struct x86_exception *exception)
4215 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4218 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4219 struct kvm_vcpu *vcpu, u32 access,
4220 struct x86_exception *exception)
4223 int r = X86EMUL_CONTINUE;
4226 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4228 unsigned offset = addr & (PAGE_SIZE-1);
4229 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4232 if (gpa == UNMAPPED_GVA)
4233 return X86EMUL_PROPAGATE_FAULT;
4234 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4237 r = X86EMUL_IO_NEEDED;
4249 /* used for instruction fetching */
4250 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4251 gva_t addr, void *val, unsigned int bytes,
4252 struct x86_exception *exception)
4254 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4255 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4259 /* Inline kvm_read_guest_virt_helper for speed. */
4260 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4262 if (unlikely(gpa == UNMAPPED_GVA))
4263 return X86EMUL_PROPAGATE_FAULT;
4265 offset = addr & (PAGE_SIZE-1);
4266 if (WARN_ON(offset + bytes > PAGE_SIZE))
4267 bytes = (unsigned)PAGE_SIZE - offset;
4268 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4270 if (unlikely(ret < 0))
4271 return X86EMUL_IO_NEEDED;
4273 return X86EMUL_CONTINUE;
4276 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4277 gva_t addr, void *val, unsigned int bytes,
4278 struct x86_exception *exception)
4280 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4281 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4283 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4286 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4288 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4289 gva_t addr, void *val, unsigned int bytes,
4290 struct x86_exception *exception)
4292 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4293 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4296 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4297 unsigned long addr, void *val, unsigned int bytes)
4299 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4300 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4302 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4305 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4306 gva_t addr, void *val,
4308 struct x86_exception *exception)
4310 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4312 int r = X86EMUL_CONTINUE;
4315 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4318 unsigned offset = addr & (PAGE_SIZE-1);
4319 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4322 if (gpa == UNMAPPED_GVA)
4323 return X86EMUL_PROPAGATE_FAULT;
4324 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4326 r = X86EMUL_IO_NEEDED;
4337 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4339 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4340 gpa_t *gpa, struct x86_exception *exception,
4343 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4344 | (write ? PFERR_WRITE_MASK : 0);
4347 * currently PKRU is only applied to ept enabled guest so
4348 * there is no pkey in EPT page table for L1 guest or EPT
4349 * shadow page table for L2 guest.
4351 if (vcpu_match_mmio_gva(vcpu, gva)
4352 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4353 vcpu->arch.access, 0, access)) {
4354 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4355 (gva & (PAGE_SIZE - 1));
4356 trace_vcpu_match_mmio(gva, *gpa, write, false);
4360 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4362 if (*gpa == UNMAPPED_GVA)
4365 /* For APIC access vmexit */
4366 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4369 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4370 trace_vcpu_match_mmio(gva, *gpa, write, true);
4377 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4378 const void *val, int bytes)
4382 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4385 kvm_page_track_write(vcpu, gpa, val, bytes);
4389 struct read_write_emulator_ops {
4390 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4392 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4393 void *val, int bytes);
4394 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4395 int bytes, void *val);
4396 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4397 void *val, int bytes);
4401 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4403 if (vcpu->mmio_read_completed) {
4404 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4405 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4406 vcpu->mmio_read_completed = 0;
4413 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4414 void *val, int bytes)
4416 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4419 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4420 void *val, int bytes)
4422 return emulator_write_phys(vcpu, gpa, val, bytes);
4425 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4427 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4428 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4431 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4432 void *val, int bytes)
4434 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4435 return X86EMUL_IO_NEEDED;
4438 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4439 void *val, int bytes)
4441 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4443 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4444 return X86EMUL_CONTINUE;
4447 static const struct read_write_emulator_ops read_emultor = {
4448 .read_write_prepare = read_prepare,
4449 .read_write_emulate = read_emulate,
4450 .read_write_mmio = vcpu_mmio_read,
4451 .read_write_exit_mmio = read_exit_mmio,
4454 static const struct read_write_emulator_ops write_emultor = {
4455 .read_write_emulate = write_emulate,
4456 .read_write_mmio = write_mmio,
4457 .read_write_exit_mmio = write_exit_mmio,
4461 static int emulator_read_write_onepage(unsigned long addr, void *val,
4463 struct x86_exception *exception,
4464 struct kvm_vcpu *vcpu,
4465 const struct read_write_emulator_ops *ops)
4469 bool write = ops->write;
4470 struct kvm_mmio_fragment *frag;
4472 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4475 return X86EMUL_PROPAGATE_FAULT;
4477 /* For APIC access vmexit */
4481 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4482 return X86EMUL_CONTINUE;
4486 * Is this MMIO handled locally?
4488 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4489 if (handled == bytes)
4490 return X86EMUL_CONTINUE;
4496 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4497 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4501 return X86EMUL_CONTINUE;
4504 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4506 void *val, unsigned int bytes,
4507 struct x86_exception *exception,
4508 const struct read_write_emulator_ops *ops)
4510 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4514 if (ops->read_write_prepare &&
4515 ops->read_write_prepare(vcpu, val, bytes))
4516 return X86EMUL_CONTINUE;
4518 vcpu->mmio_nr_fragments = 0;
4520 /* Crossing a page boundary? */
4521 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4524 now = -addr & ~PAGE_MASK;
4525 rc = emulator_read_write_onepage(addr, val, now, exception,
4528 if (rc != X86EMUL_CONTINUE)
4531 if (ctxt->mode != X86EMUL_MODE_PROT64)
4537 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4539 if (rc != X86EMUL_CONTINUE)
4542 if (!vcpu->mmio_nr_fragments)
4545 gpa = vcpu->mmio_fragments[0].gpa;
4547 vcpu->mmio_needed = 1;
4548 vcpu->mmio_cur_fragment = 0;
4550 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4551 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4552 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4553 vcpu->run->mmio.phys_addr = gpa;
4555 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4558 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4562 struct x86_exception *exception)
4564 return emulator_read_write(ctxt, addr, val, bytes,
4565 exception, &read_emultor);
4568 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4572 struct x86_exception *exception)
4574 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4575 exception, &write_emultor);
4578 #define CMPXCHG_TYPE(t, ptr, old, new) \
4579 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4581 #ifdef CONFIG_X86_64
4582 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4584 # define CMPXCHG64(ptr, old, new) \
4585 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4588 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4593 struct x86_exception *exception)
4595 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4601 /* guests cmpxchg8b have to be emulated atomically */
4602 if (bytes > 8 || (bytes & (bytes - 1)))
4605 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4607 if (gpa == UNMAPPED_GVA ||
4608 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4611 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4614 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4615 if (is_error_page(page))
4618 kaddr = kmap_atomic(page);
4619 kaddr += offset_in_page(gpa);
4622 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4625 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4628 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4631 exchanged = CMPXCHG64(kaddr, old, new);
4636 kunmap_atomic(kaddr);
4637 kvm_release_page_dirty(page);
4640 return X86EMUL_CMPXCHG_FAILED;
4642 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4643 kvm_page_track_write(vcpu, gpa, new, bytes);
4645 return X86EMUL_CONTINUE;
4648 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4650 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4653 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4655 /* TODO: String I/O for in kernel device */
4658 if (vcpu->arch.pio.in)
4659 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4660 vcpu->arch.pio.size, pd);
4662 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4663 vcpu->arch.pio.port, vcpu->arch.pio.size,
4668 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4669 unsigned short port, void *val,
4670 unsigned int count, bool in)
4672 vcpu->arch.pio.port = port;
4673 vcpu->arch.pio.in = in;
4674 vcpu->arch.pio.count = count;
4675 vcpu->arch.pio.size = size;
4677 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4678 vcpu->arch.pio.count = 0;
4682 vcpu->run->exit_reason = KVM_EXIT_IO;
4683 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4684 vcpu->run->io.size = size;
4685 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4686 vcpu->run->io.count = count;
4687 vcpu->run->io.port = port;
4692 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4693 int size, unsigned short port, void *val,
4696 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4699 if (vcpu->arch.pio.count)
4702 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4705 memcpy(val, vcpu->arch.pio_data, size * count);
4706 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4707 vcpu->arch.pio.count = 0;
4714 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4715 int size, unsigned short port,
4716 const void *val, unsigned int count)
4718 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4720 memcpy(vcpu->arch.pio_data, val, size * count);
4721 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4722 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4725 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4727 return kvm_x86_ops->get_segment_base(vcpu, seg);
4730 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4732 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4735 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4737 if (!need_emulate_wbinvd(vcpu))
4738 return X86EMUL_CONTINUE;
4740 if (kvm_x86_ops->has_wbinvd_exit()) {
4741 int cpu = get_cpu();
4743 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4744 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4745 wbinvd_ipi, NULL, 1);
4747 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4750 return X86EMUL_CONTINUE;
4753 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4755 kvm_x86_ops->skip_emulated_instruction(vcpu);
4756 return kvm_emulate_wbinvd_noskip(vcpu);
4758 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4762 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4764 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4767 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4768 unsigned long *dest)
4770 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4773 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4774 unsigned long value)
4777 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4780 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4782 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4785 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4787 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4788 unsigned long value;
4792 value = kvm_read_cr0(vcpu);
4795 value = vcpu->arch.cr2;
4798 value = kvm_read_cr3(vcpu);
4801 value = kvm_read_cr4(vcpu);
4804 value = kvm_get_cr8(vcpu);
4807 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4814 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4816 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4821 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4824 vcpu->arch.cr2 = val;
4827 res = kvm_set_cr3(vcpu, val);
4830 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4833 res = kvm_set_cr8(vcpu, val);
4836 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4843 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4845 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4848 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4850 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4853 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4855 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4858 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4860 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4863 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4865 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4868 static unsigned long emulator_get_cached_segment_base(
4869 struct x86_emulate_ctxt *ctxt, int seg)
4871 return get_segment_base(emul_to_vcpu(ctxt), seg);
4874 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4875 struct desc_struct *desc, u32 *base3,
4878 struct kvm_segment var;
4880 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4881 *selector = var.selector;
4884 memset(desc, 0, sizeof(*desc));
4890 set_desc_limit(desc, var.limit);
4891 set_desc_base(desc, (unsigned long)var.base);
4892 #ifdef CONFIG_X86_64
4894 *base3 = var.base >> 32;
4896 desc->type = var.type;
4898 desc->dpl = var.dpl;
4899 desc->p = var.present;
4900 desc->avl = var.avl;
4908 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4909 struct desc_struct *desc, u32 base3,
4912 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4913 struct kvm_segment var;
4915 var.selector = selector;
4916 var.base = get_desc_base(desc);
4917 #ifdef CONFIG_X86_64
4918 var.base |= ((u64)base3) << 32;
4920 var.limit = get_desc_limit(desc);
4922 var.limit = (var.limit << 12) | 0xfff;
4923 var.type = desc->type;
4924 var.dpl = desc->dpl;
4929 var.avl = desc->avl;
4930 var.present = desc->p;
4931 var.unusable = !var.present;
4934 kvm_set_segment(vcpu, &var, seg);
4938 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4939 u32 msr_index, u64 *pdata)
4941 struct msr_data msr;
4944 msr.index = msr_index;
4945 msr.host_initiated = false;
4946 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4954 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4955 u32 msr_index, u64 data)
4957 struct msr_data msr;
4960 msr.index = msr_index;
4961 msr.host_initiated = false;
4962 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4965 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4967 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4969 return vcpu->arch.smbase;
4972 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
4974 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4976 vcpu->arch.smbase = smbase;
4979 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4982 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
4985 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4986 u32 pmc, u64 *pdata)
4988 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
4991 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4993 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4996 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4999 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5001 * CR0.TS may reference the host fpu state, not the guest fpu state,
5002 * so it may be clear at this point.
5007 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5012 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5013 struct x86_instruction_info *info,
5014 enum x86_intercept_stage stage)
5016 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5019 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5020 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5022 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5025 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5027 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5030 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5032 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5035 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5037 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5040 static const struct x86_emulate_ops emulate_ops = {
5041 .read_gpr = emulator_read_gpr,
5042 .write_gpr = emulator_write_gpr,
5043 .read_std = kvm_read_guest_virt_system,
5044 .write_std = kvm_write_guest_virt_system,
5045 .read_phys = kvm_read_guest_phys_system,
5046 .fetch = kvm_fetch_guest_virt,
5047 .read_emulated = emulator_read_emulated,
5048 .write_emulated = emulator_write_emulated,
5049 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5050 .invlpg = emulator_invlpg,
5051 .pio_in_emulated = emulator_pio_in_emulated,
5052 .pio_out_emulated = emulator_pio_out_emulated,
5053 .get_segment = emulator_get_segment,
5054 .set_segment = emulator_set_segment,
5055 .get_cached_segment_base = emulator_get_cached_segment_base,
5056 .get_gdt = emulator_get_gdt,
5057 .get_idt = emulator_get_idt,
5058 .set_gdt = emulator_set_gdt,
5059 .set_idt = emulator_set_idt,
5060 .get_cr = emulator_get_cr,
5061 .set_cr = emulator_set_cr,
5062 .cpl = emulator_get_cpl,
5063 .get_dr = emulator_get_dr,
5064 .set_dr = emulator_set_dr,
5065 .get_smbase = emulator_get_smbase,
5066 .set_smbase = emulator_set_smbase,
5067 .set_msr = emulator_set_msr,
5068 .get_msr = emulator_get_msr,
5069 .check_pmc = emulator_check_pmc,
5070 .read_pmc = emulator_read_pmc,
5071 .halt = emulator_halt,
5072 .wbinvd = emulator_wbinvd,
5073 .fix_hypercall = emulator_fix_hypercall,
5074 .get_fpu = emulator_get_fpu,
5075 .put_fpu = emulator_put_fpu,
5076 .intercept = emulator_intercept,
5077 .get_cpuid = emulator_get_cpuid,
5078 .set_nmi_mask = emulator_set_nmi_mask,
5081 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5083 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5085 * an sti; sti; sequence only disable interrupts for the first
5086 * instruction. So, if the last instruction, be it emulated or
5087 * not, left the system with the INT_STI flag enabled, it
5088 * means that the last instruction is an sti. We should not
5089 * leave the flag on in this case. The same goes for mov ss
5091 if (int_shadow & mask)
5093 if (unlikely(int_shadow || mask)) {
5094 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5096 kvm_make_request(KVM_REQ_EVENT, vcpu);
5100 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5102 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5103 if (ctxt->exception.vector == PF_VECTOR)
5104 return kvm_propagate_fault(vcpu, &ctxt->exception);
5106 if (ctxt->exception.error_code_valid)
5107 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5108 ctxt->exception.error_code);
5110 kvm_queue_exception(vcpu, ctxt->exception.vector);
5114 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5116 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5119 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5121 ctxt->eflags = kvm_get_rflags(vcpu);
5122 ctxt->eip = kvm_rip_read(vcpu);
5123 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5124 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5125 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5126 cs_db ? X86EMUL_MODE_PROT32 :
5127 X86EMUL_MODE_PROT16;
5128 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5129 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5130 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5131 ctxt->emul_flags = vcpu->arch.hflags;
5133 init_decode_cache(ctxt);
5134 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5137 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5139 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5142 init_emulate_ctxt(vcpu);
5146 ctxt->_eip = ctxt->eip + inc_eip;
5147 ret = emulate_int_real(ctxt, irq);
5149 if (ret != X86EMUL_CONTINUE)
5150 return EMULATE_FAIL;
5152 ctxt->eip = ctxt->_eip;
5153 kvm_rip_write(vcpu, ctxt->eip);
5154 kvm_set_rflags(vcpu, ctxt->eflags);
5156 if (irq == NMI_VECTOR)
5157 vcpu->arch.nmi_pending = 0;
5159 vcpu->arch.interrupt.pending = false;
5161 return EMULATE_DONE;
5163 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5165 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5167 int r = EMULATE_DONE;
5169 ++vcpu->stat.insn_emulation_fail;
5170 trace_kvm_emulate_insn_failed(vcpu);
5171 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5172 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5173 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5174 vcpu->run->internal.ndata = 0;
5177 kvm_queue_exception(vcpu, UD_VECTOR);
5182 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5183 bool write_fault_to_shadow_pgtable,
5189 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5192 if (!vcpu->arch.mmu.direct_map) {
5194 * Write permission should be allowed since only
5195 * write access need to be emulated.
5197 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5200 * If the mapping is invalid in guest, let cpu retry
5201 * it to generate fault.
5203 if (gpa == UNMAPPED_GVA)
5208 * Do not retry the unhandleable instruction if it faults on the
5209 * readonly host memory, otherwise it will goto a infinite loop:
5210 * retry instruction -> write #PF -> emulation fail -> retry
5211 * instruction -> ...
5213 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5216 * If the instruction failed on the error pfn, it can not be fixed,
5217 * report the error to userspace.
5219 if (is_error_noslot_pfn(pfn))
5222 kvm_release_pfn_clean(pfn);
5224 /* The instructions are well-emulated on direct mmu. */
5225 if (vcpu->arch.mmu.direct_map) {
5226 unsigned int indirect_shadow_pages;
5228 spin_lock(&vcpu->kvm->mmu_lock);
5229 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5230 spin_unlock(&vcpu->kvm->mmu_lock);
5232 if (indirect_shadow_pages)
5233 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5239 * if emulation was due to access to shadowed page table
5240 * and it failed try to unshadow page and re-enter the
5241 * guest to let CPU execute the instruction.
5243 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5246 * If the access faults on its page table, it can not
5247 * be fixed by unprotecting shadow page and it should
5248 * be reported to userspace.
5250 return !write_fault_to_shadow_pgtable;
5253 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5254 unsigned long cr2, int emulation_type)
5256 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5257 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5259 last_retry_eip = vcpu->arch.last_retry_eip;
5260 last_retry_addr = vcpu->arch.last_retry_addr;
5263 * If the emulation is caused by #PF and it is non-page_table
5264 * writing instruction, it means the VM-EXIT is caused by shadow
5265 * page protected, we can zap the shadow page and retry this
5266 * instruction directly.
5268 * Note: if the guest uses a non-page-table modifying instruction
5269 * on the PDE that points to the instruction, then we will unmap
5270 * the instruction and go to an infinite loop. So, we cache the
5271 * last retried eip and the last fault address, if we meet the eip
5272 * and the address again, we can break out of the potential infinite
5275 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5277 if (!(emulation_type & EMULTYPE_RETRY))
5280 if (x86_page_table_writing_insn(ctxt))
5283 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5286 vcpu->arch.last_retry_eip = ctxt->eip;
5287 vcpu->arch.last_retry_addr = cr2;
5289 if (!vcpu->arch.mmu.direct_map)
5290 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5292 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5297 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5298 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5300 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5302 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5303 /* This is a good place to trace that we are exiting SMM. */
5304 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5306 /* Process a latched INIT or SMI, if any. */
5307 kvm_make_request(KVM_REQ_EVENT, vcpu);
5310 kvm_mmu_reset_context(vcpu);
5313 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5315 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5317 vcpu->arch.hflags = emul_flags;
5319 if (changed & HF_SMM_MASK)
5320 kvm_smm_changed(vcpu);
5323 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5332 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5333 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5338 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5340 struct kvm_run *kvm_run = vcpu->run;
5343 * rflags is the old, "raw" value of the flags. The new value has
5344 * not been saved yet.
5346 * This is correct even for TF set by the guest, because "the
5347 * processor will not generate this exception after the instruction
5348 * that sets the TF flag".
5350 if (unlikely(rflags & X86_EFLAGS_TF)) {
5351 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5352 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5354 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5355 kvm_run->debug.arch.exception = DB_VECTOR;
5356 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5357 *r = EMULATE_USER_EXIT;
5359 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5361 * "Certain debug exceptions may clear bit 0-3. The
5362 * remaining contents of the DR6 register are never
5363 * cleared by the processor".
5365 vcpu->arch.dr6 &= ~15;
5366 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5367 kvm_queue_exception(vcpu, DB_VECTOR);
5372 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5374 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5375 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5376 struct kvm_run *kvm_run = vcpu->run;
5377 unsigned long eip = kvm_get_linear_rip(vcpu);
5378 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5379 vcpu->arch.guest_debug_dr7,
5383 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5384 kvm_run->debug.arch.pc = eip;
5385 kvm_run->debug.arch.exception = DB_VECTOR;
5386 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5387 *r = EMULATE_USER_EXIT;
5392 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5393 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5394 unsigned long eip = kvm_get_linear_rip(vcpu);
5395 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5400 vcpu->arch.dr6 &= ~15;
5401 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5402 kvm_queue_exception(vcpu, DB_VECTOR);
5411 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5418 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5419 bool writeback = true;
5420 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5423 * Clear write_fault_to_shadow_pgtable here to ensure it is
5426 vcpu->arch.write_fault_to_shadow_pgtable = false;
5427 kvm_clear_exception_queue(vcpu);
5429 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5430 init_emulate_ctxt(vcpu);
5433 * We will reenter on the same instruction since
5434 * we do not set complete_userspace_io. This does not
5435 * handle watchpoints yet, those would be handled in
5438 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5441 ctxt->interruptibility = 0;
5442 ctxt->have_exception = false;
5443 ctxt->exception.vector = -1;
5444 ctxt->perm_ok = false;
5446 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5448 r = x86_decode_insn(ctxt, insn, insn_len);
5450 trace_kvm_emulate_insn_start(vcpu);
5451 ++vcpu->stat.insn_emulation;
5452 if (r != EMULATION_OK) {
5453 if (emulation_type & EMULTYPE_TRAP_UD)
5454 return EMULATE_FAIL;
5455 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5457 return EMULATE_DONE;
5458 if (emulation_type & EMULTYPE_SKIP)
5459 return EMULATE_FAIL;
5460 return handle_emulation_failure(vcpu);
5464 if (emulation_type & EMULTYPE_SKIP) {
5465 kvm_rip_write(vcpu, ctxt->_eip);
5466 if (ctxt->eflags & X86_EFLAGS_RF)
5467 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5468 return EMULATE_DONE;
5471 if (retry_instruction(ctxt, cr2, emulation_type))
5472 return EMULATE_DONE;
5474 /* this is needed for vmware backdoor interface to work since it
5475 changes registers values during IO operation */
5476 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5477 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5478 emulator_invalidate_register_cache(ctxt);
5482 r = x86_emulate_insn(ctxt);
5484 if (r == EMULATION_INTERCEPTED)
5485 return EMULATE_DONE;
5487 if (r == EMULATION_FAILED) {
5488 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5490 return EMULATE_DONE;
5492 return handle_emulation_failure(vcpu);
5495 if (ctxt->have_exception) {
5497 if (inject_emulated_exception(vcpu))
5499 } else if (vcpu->arch.pio.count) {
5500 if (!vcpu->arch.pio.in) {
5501 /* FIXME: return into emulator if single-stepping. */
5502 vcpu->arch.pio.count = 0;
5505 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5507 r = EMULATE_USER_EXIT;
5508 } else if (vcpu->mmio_needed) {
5509 if (!vcpu->mmio_is_write)
5511 r = EMULATE_USER_EXIT;
5512 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5513 } else if (r == EMULATION_RESTART)
5519 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5520 toggle_interruptibility(vcpu, ctxt->interruptibility);
5521 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5522 if (vcpu->arch.hflags != ctxt->emul_flags)
5523 kvm_set_hflags(vcpu, ctxt->emul_flags);
5524 kvm_rip_write(vcpu, ctxt->eip);
5525 if (r == EMULATE_DONE)
5526 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5527 if (!ctxt->have_exception ||
5528 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5529 __kvm_set_rflags(vcpu, ctxt->eflags);
5532 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5533 * do nothing, and it will be requested again as soon as
5534 * the shadow expires. But we still need to check here,
5535 * because POPF has no interrupt shadow.
5537 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5538 kvm_make_request(KVM_REQ_EVENT, vcpu);
5540 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5544 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5546 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5548 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5549 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5550 size, port, &val, 1);
5551 /* do not return to emulator after return from userspace */
5552 vcpu->arch.pio.count = 0;
5555 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5557 static void tsc_bad(void *info)
5559 __this_cpu_write(cpu_tsc_khz, 0);
5562 static void tsc_khz_changed(void *data)
5564 struct cpufreq_freqs *freq = data;
5565 unsigned long khz = 0;
5569 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5570 khz = cpufreq_quick_get(raw_smp_processor_id());
5573 __this_cpu_write(cpu_tsc_khz, khz);
5576 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5579 struct cpufreq_freqs *freq = data;
5581 struct kvm_vcpu *vcpu;
5582 int i, send_ipi = 0;
5585 * We allow guests to temporarily run on slowing clocks,
5586 * provided we notify them after, or to run on accelerating
5587 * clocks, provided we notify them before. Thus time never
5590 * However, we have a problem. We can't atomically update
5591 * the frequency of a given CPU from this function; it is
5592 * merely a notifier, which can be called from any CPU.
5593 * Changing the TSC frequency at arbitrary points in time
5594 * requires a recomputation of local variables related to
5595 * the TSC for each VCPU. We must flag these local variables
5596 * to be updated and be sure the update takes place with the
5597 * new frequency before any guests proceed.
5599 * Unfortunately, the combination of hotplug CPU and frequency
5600 * change creates an intractable locking scenario; the order
5601 * of when these callouts happen is undefined with respect to
5602 * CPU hotplug, and they can race with each other. As such,
5603 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5604 * undefined; you can actually have a CPU frequency change take
5605 * place in between the computation of X and the setting of the
5606 * variable. To protect against this problem, all updates of
5607 * the per_cpu tsc_khz variable are done in an interrupt
5608 * protected IPI, and all callers wishing to update the value
5609 * must wait for a synchronous IPI to complete (which is trivial
5610 * if the caller is on the CPU already). This establishes the
5611 * necessary total order on variable updates.
5613 * Note that because a guest time update may take place
5614 * anytime after the setting of the VCPU's request bit, the
5615 * correct TSC value must be set before the request. However,
5616 * to ensure the update actually makes it to any guest which
5617 * starts running in hardware virtualization between the set
5618 * and the acquisition of the spinlock, we must also ping the
5619 * CPU after setting the request bit.
5623 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5625 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5628 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5630 spin_lock(&kvm_lock);
5631 list_for_each_entry(kvm, &vm_list, vm_list) {
5632 kvm_for_each_vcpu(i, vcpu, kvm) {
5633 if (vcpu->cpu != freq->cpu)
5635 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5636 if (vcpu->cpu != smp_processor_id())
5640 spin_unlock(&kvm_lock);
5642 if (freq->old < freq->new && send_ipi) {
5644 * We upscale the frequency. Must make the guest
5645 * doesn't see old kvmclock values while running with
5646 * the new frequency, otherwise we risk the guest sees
5647 * time go backwards.
5649 * In case we update the frequency for another cpu
5650 * (which might be in guest context) send an interrupt
5651 * to kick the cpu out of guest context. Next time
5652 * guest context is entered kvmclock will be updated,
5653 * so the guest will not see stale values.
5655 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5660 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5661 .notifier_call = kvmclock_cpufreq_notifier
5664 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5665 unsigned long action, void *hcpu)
5667 unsigned int cpu = (unsigned long)hcpu;
5671 case CPU_DOWN_FAILED:
5672 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5674 case CPU_DOWN_PREPARE:
5675 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5681 static struct notifier_block kvmclock_cpu_notifier_block = {
5682 .notifier_call = kvmclock_cpu_notifier,
5683 .priority = -INT_MAX
5686 static void kvm_timer_init(void)
5690 max_tsc_khz = tsc_khz;
5692 cpu_notifier_register_begin();
5693 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5694 #ifdef CONFIG_CPU_FREQ
5695 struct cpufreq_policy policy;
5696 memset(&policy, 0, sizeof(policy));
5698 cpufreq_get_policy(&policy, cpu);
5699 if (policy.cpuinfo.max_freq)
5700 max_tsc_khz = policy.cpuinfo.max_freq;
5703 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5704 CPUFREQ_TRANSITION_NOTIFIER);
5706 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5707 for_each_online_cpu(cpu)
5708 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5710 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5711 cpu_notifier_register_done();
5715 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5717 int kvm_is_in_guest(void)
5719 return __this_cpu_read(current_vcpu) != NULL;
5722 static int kvm_is_user_mode(void)
5726 if (__this_cpu_read(current_vcpu))
5727 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5729 return user_mode != 0;
5732 static unsigned long kvm_get_guest_ip(void)
5734 unsigned long ip = 0;
5736 if (__this_cpu_read(current_vcpu))
5737 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5742 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5743 .is_in_guest = kvm_is_in_guest,
5744 .is_user_mode = kvm_is_user_mode,
5745 .get_guest_ip = kvm_get_guest_ip,
5748 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5750 __this_cpu_write(current_vcpu, vcpu);
5752 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5754 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5756 __this_cpu_write(current_vcpu, NULL);
5758 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5760 static void kvm_set_mmio_spte_mask(void)
5763 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5766 * Set the reserved bits and the present bit of an paging-structure
5767 * entry to generate page fault with PFER.RSV = 1.
5769 /* Mask the reserved physical address bits. */
5770 mask = rsvd_bits(maxphyaddr, 51);
5772 /* Bit 62 is always reserved for 32bit host. */
5773 mask |= 0x3ull << 62;
5775 /* Set the present bit. */
5778 #ifdef CONFIG_X86_64
5780 * If reserved bit is not supported, clear the present bit to disable
5783 if (maxphyaddr == 52)
5787 kvm_mmu_set_mmio_spte_mask(mask);
5790 #ifdef CONFIG_X86_64
5791 static void pvclock_gtod_update_fn(struct work_struct *work)
5795 struct kvm_vcpu *vcpu;
5798 spin_lock(&kvm_lock);
5799 list_for_each_entry(kvm, &vm_list, vm_list)
5800 kvm_for_each_vcpu(i, vcpu, kvm)
5801 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5802 atomic_set(&kvm_guest_has_master_clock, 0);
5803 spin_unlock(&kvm_lock);
5806 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5809 * Notification about pvclock gtod data update.
5811 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5814 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5815 struct timekeeper *tk = priv;
5817 update_pvclock_gtod(tk);
5819 /* disable master clock if host does not trust, or does not
5820 * use, TSC clocksource
5822 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5823 atomic_read(&kvm_guest_has_master_clock) != 0)
5824 queue_work(system_long_wq, &pvclock_gtod_work);
5829 static struct notifier_block pvclock_gtod_notifier = {
5830 .notifier_call = pvclock_gtod_notify,
5834 int kvm_arch_init(void *opaque)
5837 struct kvm_x86_ops *ops = opaque;
5840 printk(KERN_ERR "kvm: already loaded the other module\n");
5845 if (!ops->cpu_has_kvm_support()) {
5846 printk(KERN_ERR "kvm: no hardware support\n");
5850 if (ops->disabled_by_bios()) {
5851 printk(KERN_ERR "kvm: disabled by bios\n");
5857 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5859 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5863 r = kvm_mmu_module_init();
5865 goto out_free_percpu;
5867 kvm_set_mmio_spte_mask();
5871 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5872 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5876 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5878 if (boot_cpu_has(X86_FEATURE_XSAVE))
5879 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5882 #ifdef CONFIG_X86_64
5883 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5889 free_percpu(shared_msrs);
5894 void kvm_arch_exit(void)
5896 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5898 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5899 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5900 CPUFREQ_TRANSITION_NOTIFIER);
5901 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5902 #ifdef CONFIG_X86_64
5903 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5906 kvm_mmu_module_exit();
5907 free_percpu(shared_msrs);
5910 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5912 ++vcpu->stat.halt_exits;
5913 if (lapic_in_kernel(vcpu)) {
5914 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5917 vcpu->run->exit_reason = KVM_EXIT_HLT;
5921 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5923 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5925 kvm_x86_ops->skip_emulated_instruction(vcpu);
5926 return kvm_vcpu_halt(vcpu);
5928 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5931 * kvm_pv_kick_cpu_op: Kick a vcpu.
5933 * @apicid - apicid of vcpu to be kicked.
5935 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5937 struct kvm_lapic_irq lapic_irq;
5939 lapic_irq.shorthand = 0;
5940 lapic_irq.dest_mode = 0;
5941 lapic_irq.dest_id = apicid;
5942 lapic_irq.msi_redir_hint = false;
5944 lapic_irq.delivery_mode = APIC_DM_REMRD;
5945 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5948 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
5950 vcpu->arch.apicv_active = false;
5951 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
5954 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5956 unsigned long nr, a0, a1, a2, a3, ret;
5957 int op_64_bit, r = 1;
5959 kvm_x86_ops->skip_emulated_instruction(vcpu);
5961 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5962 return kvm_hv_hypercall(vcpu);
5964 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5965 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5966 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5967 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5968 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5970 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5972 op_64_bit = is_64_bit_mode(vcpu);
5981 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5987 case KVM_HC_VAPIC_POLL_IRQ:
5990 case KVM_HC_KICK_CPU:
5991 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6001 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6002 ++vcpu->stat.hypercalls;
6005 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6007 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6009 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6010 char instruction[3];
6011 unsigned long rip = kvm_rip_read(vcpu);
6013 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6015 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
6018 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6020 return vcpu->run->request_interrupt_window &&
6021 likely(!pic_in_kernel(vcpu->kvm));
6024 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6026 struct kvm_run *kvm_run = vcpu->run;
6028 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6029 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6030 kvm_run->cr8 = kvm_get_cr8(vcpu);
6031 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6032 kvm_run->ready_for_interrupt_injection =
6033 pic_in_kernel(vcpu->kvm) ||
6034 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6037 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6041 if (!kvm_x86_ops->update_cr8_intercept)
6044 if (!lapic_in_kernel(vcpu))
6047 if (vcpu->arch.apicv_active)
6050 if (!vcpu->arch.apic->vapic_addr)
6051 max_irr = kvm_lapic_find_highest_irr(vcpu);
6058 tpr = kvm_lapic_get_cr8(vcpu);
6060 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6063 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6067 /* try to reinject previous events if any */
6068 if (vcpu->arch.exception.pending) {
6069 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6070 vcpu->arch.exception.has_error_code,
6071 vcpu->arch.exception.error_code);
6073 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6074 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6077 if (vcpu->arch.exception.nr == DB_VECTOR &&
6078 (vcpu->arch.dr7 & DR7_GD)) {
6079 vcpu->arch.dr7 &= ~DR7_GD;
6080 kvm_update_dr7(vcpu);
6083 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6084 vcpu->arch.exception.has_error_code,
6085 vcpu->arch.exception.error_code,
6086 vcpu->arch.exception.reinject);
6090 if (vcpu->arch.nmi_injected) {
6091 kvm_x86_ops->set_nmi(vcpu);
6095 if (vcpu->arch.interrupt.pending) {
6096 kvm_x86_ops->set_irq(vcpu);
6100 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6101 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6106 /* try to inject new event if pending */
6107 if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6108 vcpu->arch.smi_pending = false;
6110 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6111 --vcpu->arch.nmi_pending;
6112 vcpu->arch.nmi_injected = true;
6113 kvm_x86_ops->set_nmi(vcpu);
6114 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6116 * Because interrupts can be injected asynchronously, we are
6117 * calling check_nested_events again here to avoid a race condition.
6118 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6119 * proposal and current concerns. Perhaps we should be setting
6120 * KVM_REQ_EVENT only on certain events and not unconditionally?
6122 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6123 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6127 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6128 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6130 kvm_x86_ops->set_irq(vcpu);
6137 static void process_nmi(struct kvm_vcpu *vcpu)
6142 * x86 is limited to one NMI running, and one NMI pending after it.
6143 * If an NMI is already in progress, limit further NMIs to just one.
6144 * Otherwise, allow two (and we'll inject the first one immediately).
6146 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6149 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6150 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6151 kvm_make_request(KVM_REQ_EVENT, vcpu);
6154 #define put_smstate(type, buf, offset, val) \
6155 *(type *)((buf) + (offset) - 0x7e00) = val
6157 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6160 flags |= seg->g << 23;
6161 flags |= seg->db << 22;
6162 flags |= seg->l << 21;
6163 flags |= seg->avl << 20;
6164 flags |= seg->present << 15;
6165 flags |= seg->dpl << 13;
6166 flags |= seg->s << 12;
6167 flags |= seg->type << 8;
6171 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6173 struct kvm_segment seg;
6176 kvm_get_segment(vcpu, &seg, n);
6177 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6180 offset = 0x7f84 + n * 12;
6182 offset = 0x7f2c + (n - 3) * 12;
6184 put_smstate(u32, buf, offset + 8, seg.base);
6185 put_smstate(u32, buf, offset + 4, seg.limit);
6186 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6189 #ifdef CONFIG_X86_64
6190 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6192 struct kvm_segment seg;
6196 kvm_get_segment(vcpu, &seg, n);
6197 offset = 0x7e00 + n * 16;
6199 flags = enter_smm_get_segment_flags(&seg) >> 8;
6200 put_smstate(u16, buf, offset, seg.selector);
6201 put_smstate(u16, buf, offset + 2, flags);
6202 put_smstate(u32, buf, offset + 4, seg.limit);
6203 put_smstate(u64, buf, offset + 8, seg.base);
6207 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6210 struct kvm_segment seg;
6214 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6215 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6216 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6217 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6219 for (i = 0; i < 8; i++)
6220 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6222 kvm_get_dr(vcpu, 6, &val);
6223 put_smstate(u32, buf, 0x7fcc, (u32)val);
6224 kvm_get_dr(vcpu, 7, &val);
6225 put_smstate(u32, buf, 0x7fc8, (u32)val);
6227 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6228 put_smstate(u32, buf, 0x7fc4, seg.selector);
6229 put_smstate(u32, buf, 0x7f64, seg.base);
6230 put_smstate(u32, buf, 0x7f60, seg.limit);
6231 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6233 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6234 put_smstate(u32, buf, 0x7fc0, seg.selector);
6235 put_smstate(u32, buf, 0x7f80, seg.base);
6236 put_smstate(u32, buf, 0x7f7c, seg.limit);
6237 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6239 kvm_x86_ops->get_gdt(vcpu, &dt);
6240 put_smstate(u32, buf, 0x7f74, dt.address);
6241 put_smstate(u32, buf, 0x7f70, dt.size);
6243 kvm_x86_ops->get_idt(vcpu, &dt);
6244 put_smstate(u32, buf, 0x7f58, dt.address);
6245 put_smstate(u32, buf, 0x7f54, dt.size);
6247 for (i = 0; i < 6; i++)
6248 enter_smm_save_seg_32(vcpu, buf, i);
6250 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6253 put_smstate(u32, buf, 0x7efc, 0x00020000);
6254 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6257 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6259 #ifdef CONFIG_X86_64
6261 struct kvm_segment seg;
6265 for (i = 0; i < 16; i++)
6266 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6268 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6269 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6271 kvm_get_dr(vcpu, 6, &val);
6272 put_smstate(u64, buf, 0x7f68, val);
6273 kvm_get_dr(vcpu, 7, &val);
6274 put_smstate(u64, buf, 0x7f60, val);
6276 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6277 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6278 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6280 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6283 put_smstate(u32, buf, 0x7efc, 0x00020064);
6285 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6287 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6288 put_smstate(u16, buf, 0x7e90, seg.selector);
6289 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6290 put_smstate(u32, buf, 0x7e94, seg.limit);
6291 put_smstate(u64, buf, 0x7e98, seg.base);
6293 kvm_x86_ops->get_idt(vcpu, &dt);
6294 put_smstate(u32, buf, 0x7e84, dt.size);
6295 put_smstate(u64, buf, 0x7e88, dt.address);
6297 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6298 put_smstate(u16, buf, 0x7e70, seg.selector);
6299 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6300 put_smstate(u32, buf, 0x7e74, seg.limit);
6301 put_smstate(u64, buf, 0x7e78, seg.base);
6303 kvm_x86_ops->get_gdt(vcpu, &dt);
6304 put_smstate(u32, buf, 0x7e64, dt.size);
6305 put_smstate(u64, buf, 0x7e68, dt.address);
6307 for (i = 0; i < 6; i++)
6308 enter_smm_save_seg_64(vcpu, buf, i);
6314 static void enter_smm(struct kvm_vcpu *vcpu)
6316 struct kvm_segment cs, ds;
6321 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6322 vcpu->arch.hflags |= HF_SMM_MASK;
6323 memset(buf, 0, 512);
6324 if (guest_cpuid_has_longmode(vcpu))
6325 enter_smm_save_state_64(vcpu, buf);
6327 enter_smm_save_state_32(vcpu, buf);
6329 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6331 if (kvm_x86_ops->get_nmi_mask(vcpu))
6332 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6334 kvm_x86_ops->set_nmi_mask(vcpu, true);
6336 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6337 kvm_rip_write(vcpu, 0x8000);
6339 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6340 kvm_x86_ops->set_cr0(vcpu, cr0);
6341 vcpu->arch.cr0 = cr0;
6343 kvm_x86_ops->set_cr4(vcpu, 0);
6345 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6346 dt.address = dt.size = 0;
6347 kvm_x86_ops->set_idt(vcpu, &dt);
6349 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6351 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6352 cs.base = vcpu->arch.smbase;
6357 cs.limit = ds.limit = 0xffffffff;
6358 cs.type = ds.type = 0x3;
6359 cs.dpl = ds.dpl = 0;
6364 cs.avl = ds.avl = 0;
6365 cs.present = ds.present = 1;
6366 cs.unusable = ds.unusable = 0;
6367 cs.padding = ds.padding = 0;
6369 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6370 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6371 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6372 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6373 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6374 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6376 if (guest_cpuid_has_longmode(vcpu))
6377 kvm_x86_ops->set_efer(vcpu, 0);
6379 kvm_update_cpuid(vcpu);
6380 kvm_mmu_reset_context(vcpu);
6383 static void process_smi(struct kvm_vcpu *vcpu)
6385 vcpu->arch.smi_pending = true;
6386 kvm_make_request(KVM_REQ_EVENT, vcpu);
6389 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6391 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6394 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6396 u64 eoi_exit_bitmap[4];
6398 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6401 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6403 if (irqchip_split(vcpu->kvm))
6404 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6406 if (vcpu->arch.apicv_active)
6407 kvm_x86_ops->sync_pir_to_irr(vcpu);
6408 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6410 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6411 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6412 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6415 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6417 ++vcpu->stat.tlb_flush;
6418 kvm_x86_ops->tlb_flush(vcpu);
6421 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6423 struct page *page = NULL;
6425 if (!lapic_in_kernel(vcpu))
6428 if (!kvm_x86_ops->set_apic_access_page_addr)
6431 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6432 if (is_error_page(page))
6434 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6437 * Do not pin apic access page in memory, the MMU notifier
6438 * will call us again if it is migrated or swapped out.
6442 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6444 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6445 unsigned long address)
6448 * The physical address of apic access page is stored in the VMCS.
6449 * Update it when it becomes invalid.
6451 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6452 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6456 * Returns 1 to let vcpu_run() continue the guest execution loop without
6457 * exiting to the userspace. Otherwise, the value will be returned to the
6460 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6464 dm_request_for_irq_injection(vcpu) &&
6465 kvm_cpu_accept_dm_intr(vcpu);
6467 bool req_immediate_exit = false;
6469 if (vcpu->requests) {
6470 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6471 kvm_mmu_unload(vcpu);
6472 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6473 __kvm_migrate_timers(vcpu);
6474 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6475 kvm_gen_update_masterclock(vcpu->kvm);
6476 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6477 kvm_gen_kvmclock_update(vcpu);
6478 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6479 r = kvm_guest_time_update(vcpu);
6483 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6484 kvm_mmu_sync_roots(vcpu);
6485 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6486 kvm_vcpu_flush_tlb(vcpu);
6487 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6488 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6492 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6493 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6497 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6498 vcpu->fpu_active = 0;
6499 kvm_x86_ops->fpu_deactivate(vcpu);
6501 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6502 /* Page is swapped out. Do synthetic halt */
6503 vcpu->arch.apf.halted = true;
6507 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6508 record_steal_time(vcpu);
6509 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6511 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6513 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6514 kvm_pmu_handle_event(vcpu);
6515 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6516 kvm_pmu_deliver_pmi(vcpu);
6517 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6518 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6519 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6520 vcpu->arch.ioapic_handled_vectors)) {
6521 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6522 vcpu->run->eoi.vector =
6523 vcpu->arch.pending_ioapic_eoi;
6528 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6529 vcpu_scan_ioapic(vcpu);
6530 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6531 kvm_vcpu_reload_apic_access_page(vcpu);
6532 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6533 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6534 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6538 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6539 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6540 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6544 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6545 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6546 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6552 * KVM_REQ_HV_STIMER has to be processed after
6553 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6554 * depend on the guest clock being up-to-date
6556 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6557 kvm_hv_process_stimers(vcpu);
6561 * KVM_REQ_EVENT is not set when posted interrupts are set by
6562 * VT-d hardware, so we have to update RVI unconditionally.
6564 if (kvm_lapic_enabled(vcpu)) {
6566 * Update architecture specific hints for APIC
6567 * virtual interrupt delivery.
6569 if (vcpu->arch.apicv_active)
6570 kvm_x86_ops->hwapic_irr_update(vcpu,
6571 kvm_lapic_find_highest_irr(vcpu));
6574 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6575 kvm_apic_accept_events(vcpu);
6576 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6581 if (inject_pending_event(vcpu, req_int_win) != 0)
6582 req_immediate_exit = true;
6584 /* Enable NMI/IRQ window open exits if needed.
6586 * SMIs have two cases: 1) they can be nested, and
6587 * then there is nothing to do here because RSM will
6588 * cause a vmexit anyway; 2) or the SMI can be pending
6589 * because inject_pending_event has completed the
6590 * injection of an IRQ or NMI from the previous vmexit,
6591 * and then we request an immediate exit to inject the SMI.
6593 if (vcpu->arch.smi_pending && !is_smm(vcpu))
6594 req_immediate_exit = true;
6595 if (vcpu->arch.nmi_pending)
6596 kvm_x86_ops->enable_nmi_window(vcpu);
6597 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6598 kvm_x86_ops->enable_irq_window(vcpu);
6601 if (kvm_lapic_enabled(vcpu)) {
6602 update_cr8_intercept(vcpu);
6603 kvm_lapic_sync_to_vapic(vcpu);
6607 r = kvm_mmu_reload(vcpu);
6609 goto cancel_injection;
6614 kvm_x86_ops->prepare_guest_switch(vcpu);
6615 if (vcpu->fpu_active)
6616 kvm_load_guest_fpu(vcpu);
6617 vcpu->mode = IN_GUEST_MODE;
6619 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6622 * We should set ->mode before check ->requests,
6623 * Please see the comment in kvm_make_all_cpus_request.
6624 * This also orders the write to mode from any reads
6625 * to the page tables done while the VCPU is running.
6626 * Please see the comment in kvm_flush_remote_tlbs.
6628 smp_mb__after_srcu_read_unlock();
6630 local_irq_disable();
6632 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6633 || need_resched() || signal_pending(current)) {
6634 vcpu->mode = OUTSIDE_GUEST_MODE;
6638 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6640 goto cancel_injection;
6643 kvm_load_guest_xcr0(vcpu);
6645 if (req_immediate_exit) {
6646 kvm_make_request(KVM_REQ_EVENT, vcpu);
6647 smp_send_reschedule(vcpu->cpu);
6650 trace_kvm_entry(vcpu->vcpu_id);
6651 wait_lapic_expire(vcpu);
6652 __kvm_guest_enter();
6654 if (unlikely(vcpu->arch.switch_db_regs)) {
6656 set_debugreg(vcpu->arch.eff_db[0], 0);
6657 set_debugreg(vcpu->arch.eff_db[1], 1);
6658 set_debugreg(vcpu->arch.eff_db[2], 2);
6659 set_debugreg(vcpu->arch.eff_db[3], 3);
6660 set_debugreg(vcpu->arch.dr6, 6);
6661 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6664 kvm_x86_ops->run(vcpu);
6667 * Do this here before restoring debug registers on the host. And
6668 * since we do this before handling the vmexit, a DR access vmexit
6669 * can (a) read the correct value of the debug registers, (b) set
6670 * KVM_DEBUGREG_WONT_EXIT again.
6672 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6673 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6674 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6675 kvm_update_dr0123(vcpu);
6676 kvm_update_dr6(vcpu);
6677 kvm_update_dr7(vcpu);
6678 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6682 * If the guest has used debug registers, at least dr7
6683 * will be disabled while returning to the host.
6684 * If we don't have active breakpoints in the host, we don't
6685 * care about the messed up debug address registers. But if
6686 * we have some of them active, restore the old state.
6688 if (hw_breakpoint_active())
6689 hw_breakpoint_restore();
6691 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6693 vcpu->mode = OUTSIDE_GUEST_MODE;
6696 kvm_put_guest_xcr0(vcpu);
6698 /* Interrupt is enabled by handle_external_intr() */
6699 kvm_x86_ops->handle_external_intr(vcpu);
6704 * We must have an instruction between local_irq_enable() and
6705 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6706 * the interrupt shadow. The stat.exits increment will do nicely.
6707 * But we need to prevent reordering, hence this barrier():
6715 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6718 * Profile KVM exit RIPs:
6720 if (unlikely(prof_on == KVM_PROFILING)) {
6721 unsigned long rip = kvm_rip_read(vcpu);
6722 profile_hit(KVM_PROFILING, (void *)rip);
6725 if (unlikely(vcpu->arch.tsc_always_catchup))
6726 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6728 if (vcpu->arch.apic_attention)
6729 kvm_lapic_sync_from_vapic(vcpu);
6731 r = kvm_x86_ops->handle_exit(vcpu);
6735 kvm_x86_ops->cancel_injection(vcpu);
6736 if (unlikely(vcpu->arch.apic_attention))
6737 kvm_lapic_sync_from_vapic(vcpu);
6742 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6744 if (!kvm_arch_vcpu_runnable(vcpu) &&
6745 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6746 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6747 kvm_vcpu_block(vcpu);
6748 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6750 if (kvm_x86_ops->post_block)
6751 kvm_x86_ops->post_block(vcpu);
6753 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6757 kvm_apic_accept_events(vcpu);
6758 switch(vcpu->arch.mp_state) {
6759 case KVM_MP_STATE_HALTED:
6760 vcpu->arch.pv.pv_unhalted = false;
6761 vcpu->arch.mp_state =
6762 KVM_MP_STATE_RUNNABLE;
6763 case KVM_MP_STATE_RUNNABLE:
6764 vcpu->arch.apf.halted = false;
6766 case KVM_MP_STATE_INIT_RECEIVED:
6775 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6777 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6778 !vcpu->arch.apf.halted);
6781 static int vcpu_run(struct kvm_vcpu *vcpu)
6784 struct kvm *kvm = vcpu->kvm;
6786 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6789 if (kvm_vcpu_running(vcpu)) {
6790 r = vcpu_enter_guest(vcpu);
6792 r = vcpu_block(kvm, vcpu);
6798 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6799 if (kvm_cpu_has_pending_timer(vcpu))
6800 kvm_inject_pending_timer_irqs(vcpu);
6802 if (dm_request_for_irq_injection(vcpu) &&
6803 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
6805 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6806 ++vcpu->stat.request_irq_exits;
6810 kvm_check_async_pf_completion(vcpu);
6812 if (signal_pending(current)) {
6814 vcpu->run->exit_reason = KVM_EXIT_INTR;
6815 ++vcpu->stat.signal_exits;
6818 if (need_resched()) {
6819 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6821 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6825 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6830 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6833 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6834 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6835 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6836 if (r != EMULATE_DONE)
6841 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6843 BUG_ON(!vcpu->arch.pio.count);
6845 return complete_emulated_io(vcpu);
6849 * Implements the following, as a state machine:
6853 * for each mmio piece in the fragment
6861 * for each mmio piece in the fragment
6866 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6868 struct kvm_run *run = vcpu->run;
6869 struct kvm_mmio_fragment *frag;
6872 BUG_ON(!vcpu->mmio_needed);
6874 /* Complete previous fragment */
6875 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6876 len = min(8u, frag->len);
6877 if (!vcpu->mmio_is_write)
6878 memcpy(frag->data, run->mmio.data, len);
6880 if (frag->len <= 8) {
6881 /* Switch to the next fragment. */
6883 vcpu->mmio_cur_fragment++;
6885 /* Go forward to the next mmio piece. */
6891 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6892 vcpu->mmio_needed = 0;
6894 /* FIXME: return into emulator if single-stepping. */
6895 if (vcpu->mmio_is_write)
6897 vcpu->mmio_read_completed = 1;
6898 return complete_emulated_io(vcpu);
6901 run->exit_reason = KVM_EXIT_MMIO;
6902 run->mmio.phys_addr = frag->gpa;
6903 if (vcpu->mmio_is_write)
6904 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6905 run->mmio.len = min(8u, frag->len);
6906 run->mmio.is_write = vcpu->mmio_is_write;
6907 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6912 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6914 struct fpu *fpu = ¤t->thread.fpu;
6918 fpu__activate_curr(fpu);
6920 if (vcpu->sigset_active)
6921 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6923 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6924 kvm_vcpu_block(vcpu);
6925 kvm_apic_accept_events(vcpu);
6926 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6931 /* re-sync apic's tpr */
6932 if (!lapic_in_kernel(vcpu)) {
6933 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6939 if (unlikely(vcpu->arch.complete_userspace_io)) {
6940 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6941 vcpu->arch.complete_userspace_io = NULL;
6946 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6951 post_kvm_run_save(vcpu);
6952 if (vcpu->sigset_active)
6953 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6958 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6960 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6962 * We are here if userspace calls get_regs() in the middle of
6963 * instruction emulation. Registers state needs to be copied
6964 * back from emulation context to vcpu. Userspace shouldn't do
6965 * that usually, but some bad designed PV devices (vmware
6966 * backdoor interface) need this to work
6968 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6969 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6971 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6972 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6973 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6974 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6975 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6976 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6977 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6978 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6979 #ifdef CONFIG_X86_64
6980 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6981 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6982 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6983 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6984 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6985 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6986 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6987 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6990 regs->rip = kvm_rip_read(vcpu);
6991 regs->rflags = kvm_get_rflags(vcpu);
6996 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6998 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6999 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7001 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7002 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7003 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7004 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7005 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7006 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7007 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7008 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7009 #ifdef CONFIG_X86_64
7010 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7011 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7012 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7013 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7014 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7015 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7016 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7017 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7020 kvm_rip_write(vcpu, regs->rip);
7021 kvm_set_rflags(vcpu, regs->rflags);
7023 vcpu->arch.exception.pending = false;
7025 kvm_make_request(KVM_REQ_EVENT, vcpu);
7030 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7032 struct kvm_segment cs;
7034 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7038 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7040 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7041 struct kvm_sregs *sregs)
7045 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7046 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7047 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7048 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7049 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7050 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7052 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7053 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7055 kvm_x86_ops->get_idt(vcpu, &dt);
7056 sregs->idt.limit = dt.size;
7057 sregs->idt.base = dt.address;
7058 kvm_x86_ops->get_gdt(vcpu, &dt);
7059 sregs->gdt.limit = dt.size;
7060 sregs->gdt.base = dt.address;
7062 sregs->cr0 = kvm_read_cr0(vcpu);
7063 sregs->cr2 = vcpu->arch.cr2;
7064 sregs->cr3 = kvm_read_cr3(vcpu);
7065 sregs->cr4 = kvm_read_cr4(vcpu);
7066 sregs->cr8 = kvm_get_cr8(vcpu);
7067 sregs->efer = vcpu->arch.efer;
7068 sregs->apic_base = kvm_get_apic_base(vcpu);
7070 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7072 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7073 set_bit(vcpu->arch.interrupt.nr,
7074 (unsigned long *)sregs->interrupt_bitmap);
7079 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7080 struct kvm_mp_state *mp_state)
7082 kvm_apic_accept_events(vcpu);
7083 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7084 vcpu->arch.pv.pv_unhalted)
7085 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7087 mp_state->mp_state = vcpu->arch.mp_state;
7092 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7093 struct kvm_mp_state *mp_state)
7095 if (!lapic_in_kernel(vcpu) &&
7096 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7099 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7100 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7101 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7103 vcpu->arch.mp_state = mp_state->mp_state;
7104 kvm_make_request(KVM_REQ_EVENT, vcpu);
7108 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7109 int reason, bool has_error_code, u32 error_code)
7111 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7114 init_emulate_ctxt(vcpu);
7116 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7117 has_error_code, error_code);
7120 return EMULATE_FAIL;
7122 kvm_rip_write(vcpu, ctxt->eip);
7123 kvm_set_rflags(vcpu, ctxt->eflags);
7124 kvm_make_request(KVM_REQ_EVENT, vcpu);
7125 return EMULATE_DONE;
7127 EXPORT_SYMBOL_GPL(kvm_task_switch);
7129 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7130 struct kvm_sregs *sregs)
7132 struct msr_data apic_base_msr;
7133 int mmu_reset_needed = 0;
7134 int pending_vec, max_bits, idx;
7137 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7140 dt.size = sregs->idt.limit;
7141 dt.address = sregs->idt.base;
7142 kvm_x86_ops->set_idt(vcpu, &dt);
7143 dt.size = sregs->gdt.limit;
7144 dt.address = sregs->gdt.base;
7145 kvm_x86_ops->set_gdt(vcpu, &dt);
7147 vcpu->arch.cr2 = sregs->cr2;
7148 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7149 vcpu->arch.cr3 = sregs->cr3;
7150 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7152 kvm_set_cr8(vcpu, sregs->cr8);
7154 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7155 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7156 apic_base_msr.data = sregs->apic_base;
7157 apic_base_msr.host_initiated = true;
7158 kvm_set_apic_base(vcpu, &apic_base_msr);
7160 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7161 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7162 vcpu->arch.cr0 = sregs->cr0;
7164 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7165 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7166 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7167 kvm_update_cpuid(vcpu);
7169 idx = srcu_read_lock(&vcpu->kvm->srcu);
7170 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7171 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7172 mmu_reset_needed = 1;
7174 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7176 if (mmu_reset_needed)
7177 kvm_mmu_reset_context(vcpu);
7179 max_bits = KVM_NR_INTERRUPTS;
7180 pending_vec = find_first_bit(
7181 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7182 if (pending_vec < max_bits) {
7183 kvm_queue_interrupt(vcpu, pending_vec, false);
7184 pr_debug("Set back pending irq %d\n", pending_vec);
7187 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7188 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7189 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7190 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7191 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7192 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7194 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7195 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7197 update_cr8_intercept(vcpu);
7199 /* Older userspace won't unhalt the vcpu on reset. */
7200 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7201 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7203 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7205 kvm_make_request(KVM_REQ_EVENT, vcpu);
7210 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7211 struct kvm_guest_debug *dbg)
7213 unsigned long rflags;
7216 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7218 if (vcpu->arch.exception.pending)
7220 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7221 kvm_queue_exception(vcpu, DB_VECTOR);
7223 kvm_queue_exception(vcpu, BP_VECTOR);
7227 * Read rflags as long as potentially injected trace flags are still
7230 rflags = kvm_get_rflags(vcpu);
7232 vcpu->guest_debug = dbg->control;
7233 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7234 vcpu->guest_debug = 0;
7236 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7237 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7238 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7239 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7241 for (i = 0; i < KVM_NR_DB_REGS; i++)
7242 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7244 kvm_update_dr7(vcpu);
7246 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7247 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7248 get_segment_base(vcpu, VCPU_SREG_CS);
7251 * Trigger an rflags update that will inject or remove the trace
7254 kvm_set_rflags(vcpu, rflags);
7256 kvm_x86_ops->update_bp_intercept(vcpu);
7266 * Translate a guest virtual address to a guest physical address.
7268 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7269 struct kvm_translation *tr)
7271 unsigned long vaddr = tr->linear_address;
7275 idx = srcu_read_lock(&vcpu->kvm->srcu);
7276 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7277 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7278 tr->physical_address = gpa;
7279 tr->valid = gpa != UNMAPPED_GVA;
7286 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7288 struct fxregs_state *fxsave =
7289 &vcpu->arch.guest_fpu.state.fxsave;
7291 memcpy(fpu->fpr, fxsave->st_space, 128);
7292 fpu->fcw = fxsave->cwd;
7293 fpu->fsw = fxsave->swd;
7294 fpu->ftwx = fxsave->twd;
7295 fpu->last_opcode = fxsave->fop;
7296 fpu->last_ip = fxsave->rip;
7297 fpu->last_dp = fxsave->rdp;
7298 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7303 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7305 struct fxregs_state *fxsave =
7306 &vcpu->arch.guest_fpu.state.fxsave;
7308 memcpy(fxsave->st_space, fpu->fpr, 128);
7309 fxsave->cwd = fpu->fcw;
7310 fxsave->swd = fpu->fsw;
7311 fxsave->twd = fpu->ftwx;
7312 fxsave->fop = fpu->last_opcode;
7313 fxsave->rip = fpu->last_ip;
7314 fxsave->rdp = fpu->last_dp;
7315 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7320 static void fx_init(struct kvm_vcpu *vcpu)
7322 fpstate_init(&vcpu->arch.guest_fpu.state);
7323 if (boot_cpu_has(X86_FEATURE_XSAVES))
7324 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7325 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7328 * Ensure guest xcr0 is valid for loading
7330 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7332 vcpu->arch.cr0 |= X86_CR0_ET;
7335 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7337 if (vcpu->guest_fpu_loaded)
7341 * Restore all possible states in the guest,
7342 * and assume host would use all available bits.
7343 * Guest xcr0 would be loaded later.
7345 vcpu->guest_fpu_loaded = 1;
7346 __kernel_fpu_begin();
7347 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7351 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7353 if (!vcpu->guest_fpu_loaded) {
7354 vcpu->fpu_counter = 0;
7358 vcpu->guest_fpu_loaded = 0;
7359 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7361 ++vcpu->stat.fpu_reload;
7363 * If using eager FPU mode, or if the guest is a frequent user
7364 * of the FPU, just leave the FPU active for next time.
7365 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7366 * the FPU in bursts will revert to loading it on demand.
7368 if (!use_eager_fpu()) {
7369 if (++vcpu->fpu_counter < 5)
7370 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7375 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7377 kvmclock_reset(vcpu);
7379 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7380 kvm_x86_ops->vcpu_free(vcpu);
7383 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7386 struct kvm_vcpu *vcpu;
7388 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7389 printk_once(KERN_WARNING
7390 "kvm: SMP vm created on host with unstable TSC; "
7391 "guest TSC will not be reliable\n");
7393 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7398 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7402 kvm_vcpu_mtrr_init(vcpu);
7403 r = vcpu_load(vcpu);
7406 kvm_vcpu_reset(vcpu, false);
7407 kvm_mmu_setup(vcpu);
7412 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7414 struct msr_data msr;
7415 struct kvm *kvm = vcpu->kvm;
7417 if (vcpu_load(vcpu))
7420 msr.index = MSR_IA32_TSC;
7421 msr.host_initiated = true;
7422 kvm_write_tsc(vcpu, &msr);
7425 if (!kvmclock_periodic_sync)
7428 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7429 KVMCLOCK_SYNC_PERIOD);
7432 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7435 vcpu->arch.apf.msr_val = 0;
7437 r = vcpu_load(vcpu);
7439 kvm_mmu_unload(vcpu);
7442 kvm_x86_ops->vcpu_free(vcpu);
7445 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7447 vcpu->arch.hflags = 0;
7449 vcpu->arch.smi_pending = 0;
7450 atomic_set(&vcpu->arch.nmi_queued, 0);
7451 vcpu->arch.nmi_pending = 0;
7452 vcpu->arch.nmi_injected = false;
7453 kvm_clear_interrupt_queue(vcpu);
7454 kvm_clear_exception_queue(vcpu);
7456 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7457 kvm_update_dr0123(vcpu);
7458 vcpu->arch.dr6 = DR6_INIT;
7459 kvm_update_dr6(vcpu);
7460 vcpu->arch.dr7 = DR7_FIXED_1;
7461 kvm_update_dr7(vcpu);
7465 kvm_make_request(KVM_REQ_EVENT, vcpu);
7466 vcpu->arch.apf.msr_val = 0;
7467 vcpu->arch.st.msr_val = 0;
7469 kvmclock_reset(vcpu);
7471 kvm_clear_async_pf_completion_queue(vcpu);
7472 kvm_async_pf_hash_reset(vcpu);
7473 vcpu->arch.apf.halted = false;
7476 kvm_pmu_reset(vcpu);
7477 vcpu->arch.smbase = 0x30000;
7480 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7481 vcpu->arch.regs_avail = ~0;
7482 vcpu->arch.regs_dirty = ~0;
7484 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7487 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7489 struct kvm_segment cs;
7491 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7492 cs.selector = vector << 8;
7493 cs.base = vector << 12;
7494 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7495 kvm_rip_write(vcpu, 0);
7498 int kvm_arch_hardware_enable(void)
7501 struct kvm_vcpu *vcpu;
7506 bool stable, backwards_tsc = false;
7508 kvm_shared_msr_cpu_online();
7509 ret = kvm_x86_ops->hardware_enable();
7513 local_tsc = rdtsc();
7514 stable = !check_tsc_unstable();
7515 list_for_each_entry(kvm, &vm_list, vm_list) {
7516 kvm_for_each_vcpu(i, vcpu, kvm) {
7517 if (!stable && vcpu->cpu == smp_processor_id())
7518 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7519 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7520 backwards_tsc = true;
7521 if (vcpu->arch.last_host_tsc > max_tsc)
7522 max_tsc = vcpu->arch.last_host_tsc;
7528 * Sometimes, even reliable TSCs go backwards. This happens on
7529 * platforms that reset TSC during suspend or hibernate actions, but
7530 * maintain synchronization. We must compensate. Fortunately, we can
7531 * detect that condition here, which happens early in CPU bringup,
7532 * before any KVM threads can be running. Unfortunately, we can't
7533 * bring the TSCs fully up to date with real time, as we aren't yet far
7534 * enough into CPU bringup that we know how much real time has actually
7535 * elapsed; our helper function, get_kernel_ns() will be using boot
7536 * variables that haven't been updated yet.
7538 * So we simply find the maximum observed TSC above, then record the
7539 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7540 * the adjustment will be applied. Note that we accumulate
7541 * adjustments, in case multiple suspend cycles happen before some VCPU
7542 * gets a chance to run again. In the event that no KVM threads get a
7543 * chance to run, we will miss the entire elapsed period, as we'll have
7544 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7545 * loose cycle time. This isn't too big a deal, since the loss will be
7546 * uniform across all VCPUs (not to mention the scenario is extremely
7547 * unlikely). It is possible that a second hibernate recovery happens
7548 * much faster than a first, causing the observed TSC here to be
7549 * smaller; this would require additional padding adjustment, which is
7550 * why we set last_host_tsc to the local tsc observed here.
7552 * N.B. - this code below runs only on platforms with reliable TSC,
7553 * as that is the only way backwards_tsc is set above. Also note
7554 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7555 * have the same delta_cyc adjustment applied if backwards_tsc
7556 * is detected. Note further, this adjustment is only done once,
7557 * as we reset last_host_tsc on all VCPUs to stop this from being
7558 * called multiple times (one for each physical CPU bringup).
7560 * Platforms with unreliable TSCs don't have to deal with this, they
7561 * will be compensated by the logic in vcpu_load, which sets the TSC to
7562 * catchup mode. This will catchup all VCPUs to real time, but cannot
7563 * guarantee that they stay in perfect synchronization.
7565 if (backwards_tsc) {
7566 u64 delta_cyc = max_tsc - local_tsc;
7567 backwards_tsc_observed = true;
7568 list_for_each_entry(kvm, &vm_list, vm_list) {
7569 kvm_for_each_vcpu(i, vcpu, kvm) {
7570 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7571 vcpu->arch.last_host_tsc = local_tsc;
7572 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7576 * We have to disable TSC offset matching.. if you were
7577 * booting a VM while issuing an S4 host suspend....
7578 * you may have some problem. Solving this issue is
7579 * left as an exercise to the reader.
7581 kvm->arch.last_tsc_nsec = 0;
7582 kvm->arch.last_tsc_write = 0;
7589 void kvm_arch_hardware_disable(void)
7591 kvm_x86_ops->hardware_disable();
7592 drop_user_return_notifiers();
7595 int kvm_arch_hardware_setup(void)
7599 r = kvm_x86_ops->hardware_setup();
7603 if (kvm_has_tsc_control) {
7605 * Make sure the user can only configure tsc_khz values that
7606 * fit into a signed integer.
7607 * A min value is not calculated needed because it will always
7608 * be 1 on all machines.
7610 u64 max = min(0x7fffffffULL,
7611 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7612 kvm_max_guest_tsc_khz = max;
7614 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7617 kvm_init_msr_list();
7621 void kvm_arch_hardware_unsetup(void)
7623 kvm_x86_ops->hardware_unsetup();
7626 void kvm_arch_check_processor_compat(void *rtn)
7628 kvm_x86_ops->check_processor_compatibility(rtn);
7631 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7633 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7635 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7637 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7639 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7642 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7644 return irqchip_in_kernel(vcpu->kvm) == lapic_in_kernel(vcpu);
7647 struct static_key kvm_no_apic_vcpu __read_mostly;
7648 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7650 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7656 BUG_ON(vcpu->kvm == NULL);
7659 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7660 vcpu->arch.pv.pv_unhalted = false;
7661 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7662 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7663 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7665 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7667 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7672 vcpu->arch.pio_data = page_address(page);
7674 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7676 r = kvm_mmu_create(vcpu);
7678 goto fail_free_pio_data;
7680 if (irqchip_in_kernel(kvm)) {
7681 r = kvm_create_lapic(vcpu);
7683 goto fail_mmu_destroy;
7685 static_key_slow_inc(&kvm_no_apic_vcpu);
7687 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7689 if (!vcpu->arch.mce_banks) {
7691 goto fail_free_lapic;
7693 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7695 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7697 goto fail_free_mce_banks;
7702 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7703 vcpu->arch.pv_time_enabled = false;
7705 vcpu->arch.guest_supported_xcr0 = 0;
7706 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7708 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7710 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7712 kvm_async_pf_hash_reset(vcpu);
7715 vcpu->arch.pending_external_vector = -1;
7717 kvm_hv_vcpu_init(vcpu);
7721 fail_free_mce_banks:
7722 kfree(vcpu->arch.mce_banks);
7724 kvm_free_lapic(vcpu);
7726 kvm_mmu_destroy(vcpu);
7728 free_page((unsigned long)vcpu->arch.pio_data);
7733 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7737 kvm_hv_vcpu_uninit(vcpu);
7738 kvm_pmu_destroy(vcpu);
7739 kfree(vcpu->arch.mce_banks);
7740 kvm_free_lapic(vcpu);
7741 idx = srcu_read_lock(&vcpu->kvm->srcu);
7742 kvm_mmu_destroy(vcpu);
7743 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7744 free_page((unsigned long)vcpu->arch.pio_data);
7745 if (!lapic_in_kernel(vcpu))
7746 static_key_slow_dec(&kvm_no_apic_vcpu);
7749 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7751 kvm_x86_ops->sched_in(vcpu, cpu);
7754 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7759 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7760 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7761 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7762 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7763 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7765 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7766 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7767 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7768 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7769 &kvm->arch.irq_sources_bitmap);
7771 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7772 mutex_init(&kvm->arch.apic_map_lock);
7773 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7775 pvclock_update_vm_gtod_copy(kvm);
7777 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7778 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7780 kvm_page_track_init(kvm);
7781 kvm_mmu_init_vm(kvm);
7783 if (kvm_x86_ops->vm_init)
7784 return kvm_x86_ops->vm_init(kvm);
7789 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7792 r = vcpu_load(vcpu);
7794 kvm_mmu_unload(vcpu);
7798 static void kvm_free_vcpus(struct kvm *kvm)
7801 struct kvm_vcpu *vcpu;
7804 * Unpin any mmu pages first.
7806 kvm_for_each_vcpu(i, vcpu, kvm) {
7807 kvm_clear_async_pf_completion_queue(vcpu);
7808 kvm_unload_vcpu_mmu(vcpu);
7810 kvm_for_each_vcpu(i, vcpu, kvm)
7811 kvm_arch_vcpu_free(vcpu);
7813 mutex_lock(&kvm->lock);
7814 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7815 kvm->vcpus[i] = NULL;
7817 atomic_set(&kvm->online_vcpus, 0);
7818 mutex_unlock(&kvm->lock);
7821 void kvm_arch_sync_events(struct kvm *kvm)
7823 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7824 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7825 kvm_free_all_assigned_devices(kvm);
7829 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7833 struct kvm_memslots *slots = kvm_memslots(kvm);
7834 struct kvm_memory_slot *slot, old;
7836 /* Called with kvm->slots_lock held. */
7837 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7840 slot = id_to_memslot(slots, id);
7846 * MAP_SHARED to prevent internal slot pages from being moved
7849 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7850 MAP_SHARED | MAP_ANONYMOUS, 0);
7851 if (IS_ERR((void *)hva))
7852 return PTR_ERR((void *)hva);
7861 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7862 struct kvm_userspace_memory_region m;
7864 m.slot = id | (i << 16);
7866 m.guest_phys_addr = gpa;
7867 m.userspace_addr = hva;
7868 m.memory_size = size;
7869 r = __kvm_set_memory_region(kvm, &m);
7875 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7881 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7883 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7887 mutex_lock(&kvm->slots_lock);
7888 r = __x86_set_memory_region(kvm, id, gpa, size);
7889 mutex_unlock(&kvm->slots_lock);
7893 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7895 void kvm_arch_destroy_vm(struct kvm *kvm)
7897 if (current->mm == kvm->mm) {
7899 * Free memory regions allocated on behalf of userspace,
7900 * unless the the memory map has changed due to process exit
7903 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7904 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7905 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7907 if (kvm_x86_ops->vm_destroy)
7908 kvm_x86_ops->vm_destroy(kvm);
7909 kvm_iommu_unmap_guest(kvm);
7910 kfree(kvm->arch.vpic);
7911 kfree(kvm->arch.vioapic);
7912 kvm_free_vcpus(kvm);
7913 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7914 kvm_mmu_uninit_vm(kvm);
7917 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7918 struct kvm_memory_slot *dont)
7922 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7923 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7924 kvfree(free->arch.rmap[i]);
7925 free->arch.rmap[i] = NULL;
7930 if (!dont || free->arch.lpage_info[i - 1] !=
7931 dont->arch.lpage_info[i - 1]) {
7932 kvfree(free->arch.lpage_info[i - 1]);
7933 free->arch.lpage_info[i - 1] = NULL;
7937 kvm_page_track_free_memslot(free, dont);
7940 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7941 unsigned long npages)
7945 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7946 struct kvm_lpage_info *linfo;
7951 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7952 slot->base_gfn, level) + 1;
7954 slot->arch.rmap[i] =
7955 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7956 if (!slot->arch.rmap[i])
7961 linfo = kvm_kvzalloc(lpages * sizeof(*linfo));
7965 slot->arch.lpage_info[i - 1] = linfo;
7967 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7968 linfo[0].disallow_lpage = 1;
7969 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7970 linfo[lpages - 1].disallow_lpage = 1;
7971 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7973 * If the gfn and userspace address are not aligned wrt each
7974 * other, or if explicitly asked to, disable large page
7975 * support for this slot
7977 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7978 !kvm_largepages_enabled()) {
7981 for (j = 0; j < lpages; ++j)
7982 linfo[j].disallow_lpage = 1;
7986 if (kvm_page_track_create_memslot(slot, npages))
7992 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7993 kvfree(slot->arch.rmap[i]);
7994 slot->arch.rmap[i] = NULL;
7998 kvfree(slot->arch.lpage_info[i - 1]);
7999 slot->arch.lpage_info[i - 1] = NULL;
8004 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8007 * memslots->generation has been incremented.
8008 * mmio generation may have reached its maximum value.
8010 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8013 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8014 struct kvm_memory_slot *memslot,
8015 const struct kvm_userspace_memory_region *mem,
8016 enum kvm_mr_change change)
8021 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8022 struct kvm_memory_slot *new)
8024 /* Still write protect RO slot */
8025 if (new->flags & KVM_MEM_READONLY) {
8026 kvm_mmu_slot_remove_write_access(kvm, new);
8031 * Call kvm_x86_ops dirty logging hooks when they are valid.
8033 * kvm_x86_ops->slot_disable_log_dirty is called when:
8035 * - KVM_MR_CREATE with dirty logging is disabled
8036 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8038 * The reason is, in case of PML, we need to set D-bit for any slots
8039 * with dirty logging disabled in order to eliminate unnecessary GPA
8040 * logging in PML buffer (and potential PML buffer full VMEXT). This
8041 * guarantees leaving PML enabled during guest's lifetime won't have
8042 * any additonal overhead from PML when guest is running with dirty
8043 * logging disabled for memory slots.
8045 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8046 * to dirty logging mode.
8048 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8050 * In case of write protect:
8052 * Write protect all pages for dirty logging.
8054 * All the sptes including the large sptes which point to this
8055 * slot are set to readonly. We can not create any new large
8056 * spte on this slot until the end of the logging.
8058 * See the comments in fast_page_fault().
8060 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8061 if (kvm_x86_ops->slot_enable_log_dirty)
8062 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8064 kvm_mmu_slot_remove_write_access(kvm, new);
8066 if (kvm_x86_ops->slot_disable_log_dirty)
8067 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8071 void kvm_arch_commit_memory_region(struct kvm *kvm,
8072 const struct kvm_userspace_memory_region *mem,
8073 const struct kvm_memory_slot *old,
8074 const struct kvm_memory_slot *new,
8075 enum kvm_mr_change change)
8077 int nr_mmu_pages = 0;
8079 if (!kvm->arch.n_requested_mmu_pages)
8080 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8083 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8086 * Dirty logging tracks sptes in 4k granularity, meaning that large
8087 * sptes have to be split. If live migration is successful, the guest
8088 * in the source machine will be destroyed and large sptes will be
8089 * created in the destination. However, if the guest continues to run
8090 * in the source machine (for example if live migration fails), small
8091 * sptes will remain around and cause bad performance.
8093 * Scan sptes if dirty logging has been stopped, dropping those
8094 * which can be collapsed into a single large-page spte. Later
8095 * page faults will create the large-page sptes.
8097 if ((change != KVM_MR_DELETE) &&
8098 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8099 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8100 kvm_mmu_zap_collapsible_sptes(kvm, new);
8103 * Set up write protection and/or dirty logging for the new slot.
8105 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8106 * been zapped so no dirty logging staff is needed for old slot. For
8107 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8108 * new and it's also covered when dealing with the new slot.
8110 * FIXME: const-ify all uses of struct kvm_memory_slot.
8112 if (change != KVM_MR_DELETE)
8113 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8116 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8118 kvm_mmu_invalidate_zap_all_pages(kvm);
8121 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8122 struct kvm_memory_slot *slot)
8124 kvm_mmu_invalidate_zap_all_pages(kvm);
8127 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8129 if (!list_empty_careful(&vcpu->async_pf.done))
8132 if (kvm_apic_has_events(vcpu))
8135 if (vcpu->arch.pv.pv_unhalted)
8138 if (atomic_read(&vcpu->arch.nmi_queued))
8141 if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8144 if (kvm_arch_interrupt_allowed(vcpu) &&
8145 kvm_cpu_has_interrupt(vcpu))
8148 if (kvm_hv_has_stimer_pending(vcpu))
8154 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8156 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8157 kvm_x86_ops->check_nested_events(vcpu, false);
8159 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8162 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8164 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8167 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8169 return kvm_x86_ops->interrupt_allowed(vcpu);
8172 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8174 if (is_64_bit_mode(vcpu))
8175 return kvm_rip_read(vcpu);
8176 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8177 kvm_rip_read(vcpu));
8179 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8181 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8183 return kvm_get_linear_rip(vcpu) == linear_rip;
8185 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8187 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8189 unsigned long rflags;
8191 rflags = kvm_x86_ops->get_rflags(vcpu);
8192 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8193 rflags &= ~X86_EFLAGS_TF;
8196 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8198 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8200 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8201 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8202 rflags |= X86_EFLAGS_TF;
8203 kvm_x86_ops->set_rflags(vcpu, rflags);
8206 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8208 __kvm_set_rflags(vcpu, rflags);
8209 kvm_make_request(KVM_REQ_EVENT, vcpu);
8211 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8213 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8217 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8221 r = kvm_mmu_reload(vcpu);
8225 if (!vcpu->arch.mmu.direct_map &&
8226 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8229 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8232 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8234 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8237 static inline u32 kvm_async_pf_next_probe(u32 key)
8239 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8242 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8244 u32 key = kvm_async_pf_hash_fn(gfn);
8246 while (vcpu->arch.apf.gfns[key] != ~0)
8247 key = kvm_async_pf_next_probe(key);
8249 vcpu->arch.apf.gfns[key] = gfn;
8252 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8255 u32 key = kvm_async_pf_hash_fn(gfn);
8257 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8258 (vcpu->arch.apf.gfns[key] != gfn &&
8259 vcpu->arch.apf.gfns[key] != ~0); i++)
8260 key = kvm_async_pf_next_probe(key);
8265 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8267 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8270 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8274 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8276 vcpu->arch.apf.gfns[i] = ~0;
8278 j = kvm_async_pf_next_probe(j);
8279 if (vcpu->arch.apf.gfns[j] == ~0)
8281 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8283 * k lies cyclically in ]i,j]
8285 * |....j i.k.| or |.k..j i...|
8287 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8288 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8293 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8296 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8300 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8301 struct kvm_async_pf *work)
8303 struct x86_exception fault;
8305 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8306 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8308 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8309 (vcpu->arch.apf.send_user_only &&
8310 kvm_x86_ops->get_cpl(vcpu) == 0))
8311 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8312 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8313 fault.vector = PF_VECTOR;
8314 fault.error_code_valid = true;
8315 fault.error_code = 0;
8316 fault.nested_page_fault = false;
8317 fault.address = work->arch.token;
8318 kvm_inject_page_fault(vcpu, &fault);
8322 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8323 struct kvm_async_pf *work)
8325 struct x86_exception fault;
8327 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8328 if (work->wakeup_all)
8329 work->arch.token = ~0; /* broadcast wakeup */
8331 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8333 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8334 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8335 fault.vector = PF_VECTOR;
8336 fault.error_code_valid = true;
8337 fault.error_code = 0;
8338 fault.nested_page_fault = false;
8339 fault.address = work->arch.token;
8340 kvm_inject_page_fault(vcpu, &fault);
8342 vcpu->arch.apf.halted = false;
8343 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8346 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8348 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8351 return !kvm_event_needs_reinjection(vcpu) &&
8352 kvm_x86_ops->interrupt_allowed(vcpu);
8355 void kvm_arch_start_assignment(struct kvm *kvm)
8357 atomic_inc(&kvm->arch.assigned_device_count);
8359 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8361 void kvm_arch_end_assignment(struct kvm *kvm)
8363 atomic_dec(&kvm->arch.assigned_device_count);
8365 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8367 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8369 return atomic_read(&kvm->arch.assigned_device_count);
8371 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8373 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8375 atomic_inc(&kvm->arch.noncoherent_dma_count);
8377 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8379 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8381 atomic_dec(&kvm->arch.noncoherent_dma_count);
8383 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8385 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8387 return atomic_read(&kvm->arch.noncoherent_dma_count);
8389 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8391 bool kvm_arch_has_irq_bypass(void)
8393 return kvm_x86_ops->update_pi_irte != NULL;
8396 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8397 struct irq_bypass_producer *prod)
8399 struct kvm_kernel_irqfd *irqfd =
8400 container_of(cons, struct kvm_kernel_irqfd, consumer);
8402 irqfd->producer = prod;
8404 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8405 prod->irq, irqfd->gsi, 1);
8408 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8409 struct irq_bypass_producer *prod)
8412 struct kvm_kernel_irqfd *irqfd =
8413 container_of(cons, struct kvm_kernel_irqfd, consumer);
8415 WARN_ON(irqfd->producer != prod);
8416 irqfd->producer = NULL;
8419 * When producer of consumer is unregistered, we change back to
8420 * remapped mode, so we can re-use the current implementation
8421 * when the irq is masked/disabled or the consumer side (KVM
8422 * int this case doesn't want to receive the interrupts.
8424 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8426 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8427 " fails: %d\n", irqfd->consumer.token, ret);
8430 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8431 uint32_t guest_irq, bool set)
8433 if (!kvm_x86_ops->update_pi_irte)
8436 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8439 bool kvm_vector_hashing_enabled(void)
8441 return vector_hashing;
8443 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8445 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8446 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8447 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8448 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8449 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8450 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8451 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8452 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8453 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8454 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8455 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8456 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8457 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8458 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8459 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8460 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8461 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8462 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8463 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);