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"
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.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 <linux/sched/stat.h>
58 #include <trace/events/kvm.h>
60 #include <asm/debugreg.h>
64 #include <linux/kernel_stat.h>
65 #include <asm/fpu/internal.h> /* Ugh! */
66 #include <asm/pvclock.h>
67 #include <asm/div64.h>
68 #include <asm/irq_remapping.h>
70 #define CREATE_TRACE_POINTS
73 #define MAX_IO_MSRS 256
74 #define KVM_MAX_MCE_BANKS 32
75 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
76 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
78 #define emul_to_vcpu(ctxt) \
79 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
82 * - enable syscall per default because its emulated by KVM
83 * - enable LME and LMA per default on 64 bit KVM
87 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
89 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
92 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
93 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
95 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
96 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
98 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
99 static void process_nmi(struct kvm_vcpu *vcpu);
100 static void enter_smm(struct kvm_vcpu *vcpu);
101 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
103 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
104 EXPORT_SYMBOL_GPL(kvm_x86_ops);
106 static bool __read_mostly ignore_msrs = 0;
107 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
109 unsigned int min_timer_period_us = 500;
110 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
112 static bool __read_mostly kvmclock_periodic_sync = true;
113 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
115 bool __read_mostly kvm_has_tsc_control;
116 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
117 u32 __read_mostly kvm_max_guest_tsc_khz;
118 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
119 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
120 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
121 u64 __read_mostly kvm_max_tsc_scaling_ratio;
122 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
123 u64 __read_mostly kvm_default_tsc_scaling_ratio;
124 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
126 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
127 static u32 __read_mostly tsc_tolerance_ppm = 250;
128 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
130 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
131 unsigned int __read_mostly lapic_timer_advance_ns = 0;
132 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
134 static bool __read_mostly vector_hashing = true;
135 module_param(vector_hashing, bool, S_IRUGO);
137 #define KVM_NR_SHARED_MSRS 16
139 struct kvm_shared_msrs_global {
141 u32 msrs[KVM_NR_SHARED_MSRS];
144 struct kvm_shared_msrs {
145 struct user_return_notifier urn;
147 struct kvm_shared_msr_values {
150 } values[KVM_NR_SHARED_MSRS];
153 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
154 static struct kvm_shared_msrs __percpu *shared_msrs;
156 struct kvm_stats_debugfs_item debugfs_entries[] = {
157 { "pf_fixed", VCPU_STAT(pf_fixed) },
158 { "pf_guest", VCPU_STAT(pf_guest) },
159 { "tlb_flush", VCPU_STAT(tlb_flush) },
160 { "invlpg", VCPU_STAT(invlpg) },
161 { "exits", VCPU_STAT(exits) },
162 { "io_exits", VCPU_STAT(io_exits) },
163 { "mmio_exits", VCPU_STAT(mmio_exits) },
164 { "signal_exits", VCPU_STAT(signal_exits) },
165 { "irq_window", VCPU_STAT(irq_window_exits) },
166 { "nmi_window", VCPU_STAT(nmi_window_exits) },
167 { "halt_exits", VCPU_STAT(halt_exits) },
168 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
169 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
170 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
171 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
172 { "hypercalls", VCPU_STAT(hypercalls) },
173 { "request_irq", VCPU_STAT(request_irq_exits) },
174 { "irq_exits", VCPU_STAT(irq_exits) },
175 { "host_state_reload", VCPU_STAT(host_state_reload) },
176 { "efer_reload", VCPU_STAT(efer_reload) },
177 { "fpu_reload", VCPU_STAT(fpu_reload) },
178 { "insn_emulation", VCPU_STAT(insn_emulation) },
179 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
180 { "irq_injections", VCPU_STAT(irq_injections) },
181 { "nmi_injections", VCPU_STAT(nmi_injections) },
182 { "req_event", VCPU_STAT(req_event) },
183 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
184 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
185 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
186 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
187 { "mmu_flooded", VM_STAT(mmu_flooded) },
188 { "mmu_recycled", VM_STAT(mmu_recycled) },
189 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
190 { "mmu_unsync", VM_STAT(mmu_unsync) },
191 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
192 { "largepages", VM_STAT(lpages) },
193 { "max_mmu_page_hash_collisions",
194 VM_STAT(max_mmu_page_hash_collisions) },
198 u64 __read_mostly host_xcr0;
200 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
202 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
205 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
206 vcpu->arch.apf.gfns[i] = ~0;
209 static void kvm_on_user_return(struct user_return_notifier *urn)
212 struct kvm_shared_msrs *locals
213 = container_of(urn, struct kvm_shared_msrs, urn);
214 struct kvm_shared_msr_values *values;
218 * Disabling irqs at this point since the following code could be
219 * interrupted and executed through kvm_arch_hardware_disable()
221 local_irq_save(flags);
222 if (locals->registered) {
223 locals->registered = false;
224 user_return_notifier_unregister(urn);
226 local_irq_restore(flags);
227 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
228 values = &locals->values[slot];
229 if (values->host != values->curr) {
230 wrmsrl(shared_msrs_global.msrs[slot], values->host);
231 values->curr = values->host;
236 static void shared_msr_update(unsigned slot, u32 msr)
239 unsigned int cpu = smp_processor_id();
240 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
242 /* only read, and nobody should modify it at this time,
243 * so don't need lock */
244 if (slot >= shared_msrs_global.nr) {
245 printk(KERN_ERR "kvm: invalid MSR slot!");
248 rdmsrl_safe(msr, &value);
249 smsr->values[slot].host = value;
250 smsr->values[slot].curr = value;
253 void kvm_define_shared_msr(unsigned slot, u32 msr)
255 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
256 shared_msrs_global.msrs[slot] = msr;
257 if (slot >= shared_msrs_global.nr)
258 shared_msrs_global.nr = slot + 1;
260 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
262 static void kvm_shared_msr_cpu_online(void)
266 for (i = 0; i < shared_msrs_global.nr; ++i)
267 shared_msr_update(i, shared_msrs_global.msrs[i]);
270 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
272 unsigned int cpu = smp_processor_id();
273 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
276 if (((value ^ smsr->values[slot].curr) & mask) == 0)
278 smsr->values[slot].curr = value;
279 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
283 if (!smsr->registered) {
284 smsr->urn.on_user_return = kvm_on_user_return;
285 user_return_notifier_register(&smsr->urn);
286 smsr->registered = true;
290 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
292 static void drop_user_return_notifiers(void)
294 unsigned int cpu = smp_processor_id();
295 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
297 if (smsr->registered)
298 kvm_on_user_return(&smsr->urn);
301 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
303 return vcpu->arch.apic_base;
305 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
307 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
309 u64 old_state = vcpu->arch.apic_base &
310 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
311 u64 new_state = msr_info->data &
312 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
313 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
314 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
316 if (!msr_info->host_initiated &&
317 ((msr_info->data & reserved_bits) != 0 ||
318 new_state == X2APIC_ENABLE ||
319 (new_state == MSR_IA32_APICBASE_ENABLE &&
320 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
321 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
325 kvm_lapic_set_base(vcpu, msr_info->data);
328 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
330 asmlinkage __visible void kvm_spurious_fault(void)
332 /* Fault while not rebooting. We want the trace. */
335 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
337 #define EXCPT_BENIGN 0
338 #define EXCPT_CONTRIBUTORY 1
341 static int exception_class(int vector)
351 return EXCPT_CONTRIBUTORY;
358 #define EXCPT_FAULT 0
360 #define EXCPT_ABORT 2
361 #define EXCPT_INTERRUPT 3
363 static int exception_type(int vector)
367 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
368 return EXCPT_INTERRUPT;
372 /* #DB is trap, as instruction watchpoints are handled elsewhere */
373 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
376 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
379 /* Reserved exceptions will result in fault */
383 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
384 unsigned nr, bool has_error, u32 error_code,
390 kvm_make_request(KVM_REQ_EVENT, vcpu);
392 if (!vcpu->arch.exception.pending) {
394 if (has_error && !is_protmode(vcpu))
396 vcpu->arch.exception.pending = true;
397 vcpu->arch.exception.has_error_code = has_error;
398 vcpu->arch.exception.nr = nr;
399 vcpu->arch.exception.error_code = error_code;
400 vcpu->arch.exception.reinject = reinject;
404 /* to check exception */
405 prev_nr = vcpu->arch.exception.nr;
406 if (prev_nr == DF_VECTOR) {
407 /* triple fault -> shutdown */
408 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
411 class1 = exception_class(prev_nr);
412 class2 = exception_class(nr);
413 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
414 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
415 /* generate double fault per SDM Table 5-5 */
416 vcpu->arch.exception.pending = true;
417 vcpu->arch.exception.has_error_code = true;
418 vcpu->arch.exception.nr = DF_VECTOR;
419 vcpu->arch.exception.error_code = 0;
421 /* replace previous exception with a new one in a hope
422 that instruction re-execution will regenerate lost
427 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
429 kvm_multiple_exception(vcpu, nr, false, 0, false);
431 EXPORT_SYMBOL_GPL(kvm_queue_exception);
433 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
435 kvm_multiple_exception(vcpu, nr, false, 0, true);
437 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
439 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
442 kvm_inject_gp(vcpu, 0);
444 return kvm_skip_emulated_instruction(vcpu);
448 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
450 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
452 ++vcpu->stat.pf_guest;
453 vcpu->arch.exception.nested_apf =
454 is_guest_mode(vcpu) && fault->async_page_fault;
455 if (vcpu->arch.exception.nested_apf)
456 vcpu->arch.apf.nested_apf_token = fault->address;
458 vcpu->arch.cr2 = fault->address;
459 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
461 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
463 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
465 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
466 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
468 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
470 return fault->nested_page_fault;
473 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
475 atomic_inc(&vcpu->arch.nmi_queued);
476 kvm_make_request(KVM_REQ_NMI, vcpu);
478 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
480 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
482 kvm_multiple_exception(vcpu, nr, true, error_code, false);
484 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
486 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
488 kvm_multiple_exception(vcpu, nr, true, error_code, true);
490 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
493 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
494 * a #GP and return false.
496 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
498 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
500 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
503 EXPORT_SYMBOL_GPL(kvm_require_cpl);
505 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
507 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
510 kvm_queue_exception(vcpu, UD_VECTOR);
513 EXPORT_SYMBOL_GPL(kvm_require_dr);
516 * This function will be used to read from the physical memory of the currently
517 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
518 * can read from guest physical or from the guest's guest physical memory.
520 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
521 gfn_t ngfn, void *data, int offset, int len,
524 struct x86_exception exception;
528 ngpa = gfn_to_gpa(ngfn);
529 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
530 if (real_gfn == UNMAPPED_GVA)
533 real_gfn = gpa_to_gfn(real_gfn);
535 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
537 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
539 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
540 void *data, int offset, int len, u32 access)
542 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
543 data, offset, len, access);
547 * Load the pae pdptrs. Return true is they are all valid.
549 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
551 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
552 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
555 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
557 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
558 offset * sizeof(u64), sizeof(pdpte),
559 PFERR_USER_MASK|PFERR_WRITE_MASK);
564 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
565 if ((pdpte[i] & PT_PRESENT_MASK) &&
567 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
574 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
575 __set_bit(VCPU_EXREG_PDPTR,
576 (unsigned long *)&vcpu->arch.regs_avail);
577 __set_bit(VCPU_EXREG_PDPTR,
578 (unsigned long *)&vcpu->arch.regs_dirty);
583 EXPORT_SYMBOL_GPL(load_pdptrs);
585 bool pdptrs_changed(struct kvm_vcpu *vcpu)
587 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
593 if (is_long_mode(vcpu) || !is_pae(vcpu))
596 if (!test_bit(VCPU_EXREG_PDPTR,
597 (unsigned long *)&vcpu->arch.regs_avail))
600 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
601 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
602 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
603 PFERR_USER_MASK | PFERR_WRITE_MASK);
606 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
611 EXPORT_SYMBOL_GPL(pdptrs_changed);
613 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
615 unsigned long old_cr0 = kvm_read_cr0(vcpu);
616 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
621 if (cr0 & 0xffffffff00000000UL)
625 cr0 &= ~CR0_RESERVED_BITS;
627 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
630 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
633 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
635 if ((vcpu->arch.efer & EFER_LME)) {
640 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
645 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
650 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
653 kvm_x86_ops->set_cr0(vcpu, cr0);
655 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
656 kvm_clear_async_pf_completion_queue(vcpu);
657 kvm_async_pf_hash_reset(vcpu);
660 if ((cr0 ^ old_cr0) & update_bits)
661 kvm_mmu_reset_context(vcpu);
663 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
664 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
665 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
666 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
670 EXPORT_SYMBOL_GPL(kvm_set_cr0);
672 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
674 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
676 EXPORT_SYMBOL_GPL(kvm_lmsw);
678 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
680 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
681 !vcpu->guest_xcr0_loaded) {
682 /* kvm_set_xcr() also depends on this */
683 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
684 vcpu->guest_xcr0_loaded = 1;
688 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
690 if (vcpu->guest_xcr0_loaded) {
691 if (vcpu->arch.xcr0 != host_xcr0)
692 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
693 vcpu->guest_xcr0_loaded = 0;
697 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
700 u64 old_xcr0 = vcpu->arch.xcr0;
703 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
704 if (index != XCR_XFEATURE_ENABLED_MASK)
706 if (!(xcr0 & XFEATURE_MASK_FP))
708 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
712 * Do not allow the guest to set bits that we do not support
713 * saving. However, xcr0 bit 0 is always set, even if the
714 * emulated CPU does not support XSAVE (see fx_init).
716 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
717 if (xcr0 & ~valid_bits)
720 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
721 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
724 if (xcr0 & XFEATURE_MASK_AVX512) {
725 if (!(xcr0 & XFEATURE_MASK_YMM))
727 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
730 vcpu->arch.xcr0 = xcr0;
732 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
733 kvm_update_cpuid(vcpu);
737 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
739 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
740 __kvm_set_xcr(vcpu, index, xcr)) {
741 kvm_inject_gp(vcpu, 0);
746 EXPORT_SYMBOL_GPL(kvm_set_xcr);
748 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
750 unsigned long old_cr4 = kvm_read_cr4(vcpu);
751 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
752 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
754 if (cr4 & CR4_RESERVED_BITS)
757 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
760 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
763 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
766 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
769 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
772 if (is_long_mode(vcpu)) {
773 if (!(cr4 & X86_CR4_PAE))
775 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
776 && ((cr4 ^ old_cr4) & pdptr_bits)
777 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
781 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
782 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
785 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
786 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
790 if (kvm_x86_ops->set_cr4(vcpu, cr4))
793 if (((cr4 ^ old_cr4) & pdptr_bits) ||
794 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
795 kvm_mmu_reset_context(vcpu);
797 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
798 kvm_update_cpuid(vcpu);
802 EXPORT_SYMBOL_GPL(kvm_set_cr4);
804 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
807 cr3 &= ~CR3_PCID_INVD;
810 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
811 kvm_mmu_sync_roots(vcpu);
812 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
816 if (is_long_mode(vcpu)) {
817 if (cr3 & CR3_L_MODE_RESERVED_BITS)
819 } else if (is_pae(vcpu) && is_paging(vcpu) &&
820 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
823 vcpu->arch.cr3 = cr3;
824 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
825 kvm_mmu_new_cr3(vcpu);
828 EXPORT_SYMBOL_GPL(kvm_set_cr3);
830 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
832 if (cr8 & CR8_RESERVED_BITS)
834 if (lapic_in_kernel(vcpu))
835 kvm_lapic_set_tpr(vcpu, cr8);
837 vcpu->arch.cr8 = cr8;
840 EXPORT_SYMBOL_GPL(kvm_set_cr8);
842 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
844 if (lapic_in_kernel(vcpu))
845 return kvm_lapic_get_cr8(vcpu);
847 return vcpu->arch.cr8;
849 EXPORT_SYMBOL_GPL(kvm_get_cr8);
851 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
855 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
856 for (i = 0; i < KVM_NR_DB_REGS; i++)
857 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
858 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
862 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
864 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
865 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
868 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
872 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
873 dr7 = vcpu->arch.guest_debug_dr7;
875 dr7 = vcpu->arch.dr7;
876 kvm_x86_ops->set_dr7(vcpu, dr7);
877 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
878 if (dr7 & DR7_BP_EN_MASK)
879 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
882 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
884 u64 fixed = DR6_FIXED_1;
886 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
891 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
895 vcpu->arch.db[dr] = val;
896 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
897 vcpu->arch.eff_db[dr] = val;
902 if (val & 0xffffffff00000000ULL)
904 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
905 kvm_update_dr6(vcpu);
910 if (val & 0xffffffff00000000ULL)
912 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
913 kvm_update_dr7(vcpu);
920 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
922 if (__kvm_set_dr(vcpu, dr, val)) {
923 kvm_inject_gp(vcpu, 0);
928 EXPORT_SYMBOL_GPL(kvm_set_dr);
930 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
934 *val = vcpu->arch.db[dr];
939 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
940 *val = vcpu->arch.dr6;
942 *val = kvm_x86_ops->get_dr6(vcpu);
947 *val = vcpu->arch.dr7;
952 EXPORT_SYMBOL_GPL(kvm_get_dr);
954 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
956 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
960 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
963 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
964 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
967 EXPORT_SYMBOL_GPL(kvm_rdpmc);
970 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
971 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
973 * This list is modified at module load time to reflect the
974 * capabilities of the host cpu. This capabilities test skips MSRs that are
975 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
976 * may depend on host virtualization features rather than host cpu features.
979 static u32 msrs_to_save[] = {
980 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
983 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
985 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
986 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
989 static unsigned num_msrs_to_save;
991 static u32 emulated_msrs[] = {
992 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
993 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
994 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
995 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
996 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
997 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
998 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1000 HV_X64_MSR_VP_INDEX,
1001 HV_X64_MSR_VP_RUNTIME,
1002 HV_X64_MSR_SCONTROL,
1003 HV_X64_MSR_STIMER0_CONFIG,
1004 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1007 MSR_IA32_TSC_ADJUST,
1008 MSR_IA32_TSCDEADLINE,
1009 MSR_IA32_MISC_ENABLE,
1010 MSR_IA32_MCG_STATUS,
1012 MSR_IA32_MCG_EXT_CTL,
1015 MSR_MISC_FEATURES_ENABLES,
1018 static unsigned num_emulated_msrs;
1020 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1022 if (efer & efer_reserved_bits)
1025 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1028 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1033 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1035 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1037 u64 old_efer = vcpu->arch.efer;
1039 if (!kvm_valid_efer(vcpu, efer))
1043 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1047 efer |= vcpu->arch.efer & EFER_LMA;
1049 kvm_x86_ops->set_efer(vcpu, efer);
1051 /* Update reserved bits */
1052 if ((efer ^ old_efer) & EFER_NX)
1053 kvm_mmu_reset_context(vcpu);
1058 void kvm_enable_efer_bits(u64 mask)
1060 efer_reserved_bits &= ~mask;
1062 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1065 * Writes msr value into into the appropriate "register".
1066 * Returns 0 on success, non-0 otherwise.
1067 * Assumes vcpu_load() was already called.
1069 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1071 switch (msr->index) {
1074 case MSR_KERNEL_GS_BASE:
1077 if (is_noncanonical_address(msr->data))
1080 case MSR_IA32_SYSENTER_EIP:
1081 case MSR_IA32_SYSENTER_ESP:
1083 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1084 * non-canonical address is written on Intel but not on
1085 * AMD (which ignores the top 32-bits, because it does
1086 * not implement 64-bit SYSENTER).
1088 * 64-bit code should hence be able to write a non-canonical
1089 * value on AMD. Making the address canonical ensures that
1090 * vmentry does not fail on Intel after writing a non-canonical
1091 * value, and that something deterministic happens if the guest
1092 * invokes 64-bit SYSENTER.
1094 msr->data = get_canonical(msr->data);
1096 return kvm_x86_ops->set_msr(vcpu, msr);
1098 EXPORT_SYMBOL_GPL(kvm_set_msr);
1101 * Adapt set_msr() to msr_io()'s calling convention
1103 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1105 struct msr_data msr;
1109 msr.host_initiated = true;
1110 r = kvm_get_msr(vcpu, &msr);
1118 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1120 struct msr_data msr;
1124 msr.host_initiated = true;
1125 return kvm_set_msr(vcpu, &msr);
1128 #ifdef CONFIG_X86_64
1129 struct pvclock_gtod_data {
1132 struct { /* extract of a clocksource struct */
1145 static struct pvclock_gtod_data pvclock_gtod_data;
1147 static void update_pvclock_gtod(struct timekeeper *tk)
1149 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1152 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1154 write_seqcount_begin(&vdata->seq);
1156 /* copy pvclock gtod data */
1157 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1158 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1159 vdata->clock.mask = tk->tkr_mono.mask;
1160 vdata->clock.mult = tk->tkr_mono.mult;
1161 vdata->clock.shift = tk->tkr_mono.shift;
1163 vdata->boot_ns = boot_ns;
1164 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1166 vdata->wall_time_sec = tk->xtime_sec;
1168 write_seqcount_end(&vdata->seq);
1172 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1175 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1176 * vcpu_enter_guest. This function is only called from
1177 * the physical CPU that is running vcpu.
1179 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1182 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1186 struct pvclock_wall_clock wc;
1187 struct timespec64 boot;
1192 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1197 ++version; /* first time write, random junk */
1201 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1205 * The guest calculates current wall clock time by adding
1206 * system time (updated by kvm_guest_time_update below) to the
1207 * wall clock specified here. guest system time equals host
1208 * system time for us, thus we must fill in host boot time here.
1210 getboottime64(&boot);
1212 if (kvm->arch.kvmclock_offset) {
1213 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1214 boot = timespec64_sub(boot, ts);
1216 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1217 wc.nsec = boot.tv_nsec;
1218 wc.version = version;
1220 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1223 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1226 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1228 do_shl32_div32(dividend, divisor);
1232 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1233 s8 *pshift, u32 *pmultiplier)
1241 scaled64 = scaled_hz;
1242 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1247 tps32 = (uint32_t)tps64;
1248 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1249 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1257 *pmultiplier = div_frac(scaled64, tps32);
1259 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1260 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1263 #ifdef CONFIG_X86_64
1264 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1267 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1268 static unsigned long max_tsc_khz;
1270 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1272 u64 v = (u64)khz * (1000000 + ppm);
1277 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1281 /* Guest TSC same frequency as host TSC? */
1283 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1287 /* TSC scaling supported? */
1288 if (!kvm_has_tsc_control) {
1289 if (user_tsc_khz > tsc_khz) {
1290 vcpu->arch.tsc_catchup = 1;
1291 vcpu->arch.tsc_always_catchup = 1;
1294 WARN(1, "user requested TSC rate below hardware speed\n");
1299 /* TSC scaling required - calculate ratio */
1300 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1301 user_tsc_khz, tsc_khz);
1303 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1304 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1309 vcpu->arch.tsc_scaling_ratio = ratio;
1313 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1315 u32 thresh_lo, thresh_hi;
1316 int use_scaling = 0;
1318 /* tsc_khz can be zero if TSC calibration fails */
1319 if (user_tsc_khz == 0) {
1320 /* set tsc_scaling_ratio to a safe value */
1321 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1325 /* Compute a scale to convert nanoseconds in TSC cycles */
1326 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1327 &vcpu->arch.virtual_tsc_shift,
1328 &vcpu->arch.virtual_tsc_mult);
1329 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1332 * Compute the variation in TSC rate which is acceptable
1333 * within the range of tolerance and decide if the
1334 * rate being applied is within that bounds of the hardware
1335 * rate. If so, no scaling or compensation need be done.
1337 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1338 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1339 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1340 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1343 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1346 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1348 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1349 vcpu->arch.virtual_tsc_mult,
1350 vcpu->arch.virtual_tsc_shift);
1351 tsc += vcpu->arch.this_tsc_write;
1355 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1357 #ifdef CONFIG_X86_64
1359 struct kvm_arch *ka = &vcpu->kvm->arch;
1360 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1362 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1363 atomic_read(&vcpu->kvm->online_vcpus));
1366 * Once the masterclock is enabled, always perform request in
1367 * order to update it.
1369 * In order to enable masterclock, the host clocksource must be TSC
1370 * and the vcpus need to have matched TSCs. When that happens,
1371 * perform request to enable masterclock.
1373 if (ka->use_master_clock ||
1374 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1375 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1377 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1378 atomic_read(&vcpu->kvm->online_vcpus),
1379 ka->use_master_clock, gtod->clock.vclock_mode);
1383 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1385 u64 curr_offset = vcpu->arch.tsc_offset;
1386 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1390 * Multiply tsc by a fixed point number represented by ratio.
1392 * The most significant 64-N bits (mult) of ratio represent the
1393 * integral part of the fixed point number; the remaining N bits
1394 * (frac) represent the fractional part, ie. ratio represents a fixed
1395 * point number (mult + frac * 2^(-N)).
1397 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1399 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1401 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1404 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1407 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1409 if (ratio != kvm_default_tsc_scaling_ratio)
1410 _tsc = __scale_tsc(ratio, tsc);
1414 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1416 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1420 tsc = kvm_scale_tsc(vcpu, rdtsc());
1422 return target_tsc - tsc;
1425 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1427 return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1429 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1431 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1433 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1434 vcpu->arch.tsc_offset = offset;
1437 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1439 struct kvm *kvm = vcpu->kvm;
1440 u64 offset, ns, elapsed;
1441 unsigned long flags;
1443 bool already_matched;
1444 u64 data = msr->data;
1445 bool synchronizing = false;
1447 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1448 offset = kvm_compute_tsc_offset(vcpu, data);
1449 ns = ktime_get_boot_ns();
1450 elapsed = ns - kvm->arch.last_tsc_nsec;
1452 if (vcpu->arch.virtual_tsc_khz) {
1453 if (data == 0 && msr->host_initiated) {
1455 * detection of vcpu initialization -- need to sync
1456 * with other vCPUs. This particularly helps to keep
1457 * kvm_clock stable after CPU hotplug
1459 synchronizing = true;
1461 u64 tsc_exp = kvm->arch.last_tsc_write +
1462 nsec_to_cycles(vcpu, elapsed);
1463 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1465 * Special case: TSC write with a small delta (1 second)
1466 * of virtual cycle time against real time is
1467 * interpreted as an attempt to synchronize the CPU.
1469 synchronizing = data < tsc_exp + tsc_hz &&
1470 data + tsc_hz > tsc_exp;
1475 * For a reliable TSC, we can match TSC offsets, and for an unstable
1476 * TSC, we add elapsed time in this computation. We could let the
1477 * compensation code attempt to catch up if we fall behind, but
1478 * it's better to try to match offsets from the beginning.
1480 if (synchronizing &&
1481 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1482 if (!check_tsc_unstable()) {
1483 offset = kvm->arch.cur_tsc_offset;
1484 pr_debug("kvm: matched tsc offset for %llu\n", data);
1486 u64 delta = nsec_to_cycles(vcpu, elapsed);
1488 offset = kvm_compute_tsc_offset(vcpu, data);
1489 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1492 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1495 * We split periods of matched TSC writes into generations.
1496 * For each generation, we track the original measured
1497 * nanosecond time, offset, and write, so if TSCs are in
1498 * sync, we can match exact offset, and if not, we can match
1499 * exact software computation in compute_guest_tsc()
1501 * These values are tracked in kvm->arch.cur_xxx variables.
1503 kvm->arch.cur_tsc_generation++;
1504 kvm->arch.cur_tsc_nsec = ns;
1505 kvm->arch.cur_tsc_write = data;
1506 kvm->arch.cur_tsc_offset = offset;
1508 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1509 kvm->arch.cur_tsc_generation, data);
1513 * We also track th most recent recorded KHZ, write and time to
1514 * allow the matching interval to be extended at each write.
1516 kvm->arch.last_tsc_nsec = ns;
1517 kvm->arch.last_tsc_write = data;
1518 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1520 vcpu->arch.last_guest_tsc = data;
1522 /* Keep track of which generation this VCPU has synchronized to */
1523 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1524 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1525 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1527 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1528 update_ia32_tsc_adjust_msr(vcpu, offset);
1530 kvm_vcpu_write_tsc_offset(vcpu, offset);
1531 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1533 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1535 kvm->arch.nr_vcpus_matched_tsc = 0;
1536 } else if (!already_matched) {
1537 kvm->arch.nr_vcpus_matched_tsc++;
1540 kvm_track_tsc_matching(vcpu);
1541 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1544 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1546 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1549 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1552 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1554 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1555 WARN_ON(adjustment < 0);
1556 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1557 adjust_tsc_offset_guest(vcpu, adjustment);
1560 #ifdef CONFIG_X86_64
1562 static u64 read_tsc(void)
1564 u64 ret = (u64)rdtsc_ordered();
1565 u64 last = pvclock_gtod_data.clock.cycle_last;
1567 if (likely(ret >= last))
1571 * GCC likes to generate cmov here, but this branch is extremely
1572 * predictable (it's just a function of time and the likely is
1573 * very likely) and there's a data dependence, so force GCC
1574 * to generate a branch instead. I don't barrier() because
1575 * we don't actually need a barrier, and if this function
1576 * ever gets inlined it will generate worse code.
1582 static inline u64 vgettsc(u64 *cycle_now)
1585 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1587 *cycle_now = read_tsc();
1589 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1590 return v * gtod->clock.mult;
1593 static int do_monotonic_boot(s64 *t, u64 *cycle_now)
1595 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1601 seq = read_seqcount_begin(>od->seq);
1602 mode = gtod->clock.vclock_mode;
1603 ns = gtod->nsec_base;
1604 ns += vgettsc(cycle_now);
1605 ns >>= gtod->clock.shift;
1606 ns += gtod->boot_ns;
1607 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1613 static int do_realtime(struct timespec *ts, u64 *cycle_now)
1615 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1621 seq = read_seqcount_begin(>od->seq);
1622 mode = gtod->clock.vclock_mode;
1623 ts->tv_sec = gtod->wall_time_sec;
1624 ns = gtod->nsec_base;
1625 ns += vgettsc(cycle_now);
1626 ns >>= gtod->clock.shift;
1627 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1629 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1635 /* returns true if host is using tsc clocksource */
1636 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
1638 /* checked again under seqlock below */
1639 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1642 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1645 /* returns true if host is using tsc clocksource */
1646 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1649 /* checked again under seqlock below */
1650 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1653 return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1659 * Assuming a stable TSC across physical CPUS, and a stable TSC
1660 * across virtual CPUs, the following condition is possible.
1661 * Each numbered line represents an event visible to both
1662 * CPUs at the next numbered event.
1664 * "timespecX" represents host monotonic time. "tscX" represents
1667 * VCPU0 on CPU0 | VCPU1 on CPU1
1669 * 1. read timespec0,tsc0
1670 * 2. | timespec1 = timespec0 + N
1672 * 3. transition to guest | transition to guest
1673 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1674 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1675 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1677 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1680 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1682 * - 0 < N - M => M < N
1684 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1685 * always the case (the difference between two distinct xtime instances
1686 * might be smaller then the difference between corresponding TSC reads,
1687 * when updating guest vcpus pvclock areas).
1689 * To avoid that problem, do not allow visibility of distinct
1690 * system_timestamp/tsc_timestamp values simultaneously: use a master
1691 * copy of host monotonic time values. Update that master copy
1694 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1698 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1700 #ifdef CONFIG_X86_64
1701 struct kvm_arch *ka = &kvm->arch;
1703 bool host_tsc_clocksource, vcpus_matched;
1705 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1706 atomic_read(&kvm->online_vcpus));
1709 * If the host uses TSC clock, then passthrough TSC as stable
1712 host_tsc_clocksource = kvm_get_time_and_clockread(
1713 &ka->master_kernel_ns,
1714 &ka->master_cycle_now);
1716 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1717 && !ka->backwards_tsc_observed
1718 && !ka->boot_vcpu_runs_old_kvmclock;
1720 if (ka->use_master_clock)
1721 atomic_set(&kvm_guest_has_master_clock, 1);
1723 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1724 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1729 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1731 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1734 static void kvm_gen_update_masterclock(struct kvm *kvm)
1736 #ifdef CONFIG_X86_64
1738 struct kvm_vcpu *vcpu;
1739 struct kvm_arch *ka = &kvm->arch;
1741 spin_lock(&ka->pvclock_gtod_sync_lock);
1742 kvm_make_mclock_inprogress_request(kvm);
1743 /* no guest entries from this point */
1744 pvclock_update_vm_gtod_copy(kvm);
1746 kvm_for_each_vcpu(i, vcpu, kvm)
1747 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1749 /* guest entries allowed */
1750 kvm_for_each_vcpu(i, vcpu, kvm)
1751 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1753 spin_unlock(&ka->pvclock_gtod_sync_lock);
1757 u64 get_kvmclock_ns(struct kvm *kvm)
1759 struct kvm_arch *ka = &kvm->arch;
1760 struct pvclock_vcpu_time_info hv_clock;
1763 spin_lock(&ka->pvclock_gtod_sync_lock);
1764 if (!ka->use_master_clock) {
1765 spin_unlock(&ka->pvclock_gtod_sync_lock);
1766 return ktime_get_boot_ns() + ka->kvmclock_offset;
1769 hv_clock.tsc_timestamp = ka->master_cycle_now;
1770 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1771 spin_unlock(&ka->pvclock_gtod_sync_lock);
1773 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1776 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1777 &hv_clock.tsc_shift,
1778 &hv_clock.tsc_to_system_mul);
1779 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1786 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1788 struct kvm_vcpu_arch *vcpu = &v->arch;
1789 struct pvclock_vcpu_time_info guest_hv_clock;
1791 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1792 &guest_hv_clock, sizeof(guest_hv_clock))))
1795 /* This VCPU is paused, but it's legal for a guest to read another
1796 * VCPU's kvmclock, so we really have to follow the specification where
1797 * it says that version is odd if data is being modified, and even after
1800 * Version field updates must be kept separate. This is because
1801 * kvm_write_guest_cached might use a "rep movs" instruction, and
1802 * writes within a string instruction are weakly ordered. So there
1803 * are three writes overall.
1805 * As a small optimization, only write the version field in the first
1806 * and third write. The vcpu->pv_time cache is still valid, because the
1807 * version field is the first in the struct.
1809 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1811 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1812 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1814 sizeof(vcpu->hv_clock.version));
1818 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1819 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1821 if (vcpu->pvclock_set_guest_stopped_request) {
1822 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1823 vcpu->pvclock_set_guest_stopped_request = false;
1826 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1828 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1830 sizeof(vcpu->hv_clock));
1834 vcpu->hv_clock.version++;
1835 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1837 sizeof(vcpu->hv_clock.version));
1840 static int kvm_guest_time_update(struct kvm_vcpu *v)
1842 unsigned long flags, tgt_tsc_khz;
1843 struct kvm_vcpu_arch *vcpu = &v->arch;
1844 struct kvm_arch *ka = &v->kvm->arch;
1846 u64 tsc_timestamp, host_tsc;
1848 bool use_master_clock;
1854 * If the host uses TSC clock, then passthrough TSC as stable
1857 spin_lock(&ka->pvclock_gtod_sync_lock);
1858 use_master_clock = ka->use_master_clock;
1859 if (use_master_clock) {
1860 host_tsc = ka->master_cycle_now;
1861 kernel_ns = ka->master_kernel_ns;
1863 spin_unlock(&ka->pvclock_gtod_sync_lock);
1865 /* Keep irq disabled to prevent changes to the clock */
1866 local_irq_save(flags);
1867 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1868 if (unlikely(tgt_tsc_khz == 0)) {
1869 local_irq_restore(flags);
1870 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1873 if (!use_master_clock) {
1875 kernel_ns = ktime_get_boot_ns();
1878 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1881 * We may have to catch up the TSC to match elapsed wall clock
1882 * time for two reasons, even if kvmclock is used.
1883 * 1) CPU could have been running below the maximum TSC rate
1884 * 2) Broken TSC compensation resets the base at each VCPU
1885 * entry to avoid unknown leaps of TSC even when running
1886 * again on the same CPU. This may cause apparent elapsed
1887 * time to disappear, and the guest to stand still or run
1890 if (vcpu->tsc_catchup) {
1891 u64 tsc = compute_guest_tsc(v, kernel_ns);
1892 if (tsc > tsc_timestamp) {
1893 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1894 tsc_timestamp = tsc;
1898 local_irq_restore(flags);
1900 /* With all the info we got, fill in the values */
1902 if (kvm_has_tsc_control)
1903 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1905 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1906 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1907 &vcpu->hv_clock.tsc_shift,
1908 &vcpu->hv_clock.tsc_to_system_mul);
1909 vcpu->hw_tsc_khz = tgt_tsc_khz;
1912 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1913 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1914 vcpu->last_guest_tsc = tsc_timestamp;
1916 /* If the host uses TSC clocksource, then it is stable */
1918 if (use_master_clock)
1919 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1921 vcpu->hv_clock.flags = pvclock_flags;
1923 if (vcpu->pv_time_enabled)
1924 kvm_setup_pvclock_page(v);
1925 if (v == kvm_get_vcpu(v->kvm, 0))
1926 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
1931 * kvmclock updates which are isolated to a given vcpu, such as
1932 * vcpu->cpu migration, should not allow system_timestamp from
1933 * the rest of the vcpus to remain static. Otherwise ntp frequency
1934 * correction applies to one vcpu's system_timestamp but not
1937 * So in those cases, request a kvmclock update for all vcpus.
1938 * We need to rate-limit these requests though, as they can
1939 * considerably slow guests that have a large number of vcpus.
1940 * The time for a remote vcpu to update its kvmclock is bound
1941 * by the delay we use to rate-limit the updates.
1944 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1946 static void kvmclock_update_fn(struct work_struct *work)
1949 struct delayed_work *dwork = to_delayed_work(work);
1950 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1951 kvmclock_update_work);
1952 struct kvm *kvm = container_of(ka, struct kvm, arch);
1953 struct kvm_vcpu *vcpu;
1955 kvm_for_each_vcpu(i, vcpu, kvm) {
1956 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1957 kvm_vcpu_kick(vcpu);
1961 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1963 struct kvm *kvm = v->kvm;
1965 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1966 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1967 KVMCLOCK_UPDATE_DELAY);
1970 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1972 static void kvmclock_sync_fn(struct work_struct *work)
1974 struct delayed_work *dwork = to_delayed_work(work);
1975 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1976 kvmclock_sync_work);
1977 struct kvm *kvm = container_of(ka, struct kvm, arch);
1979 if (!kvmclock_periodic_sync)
1982 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1983 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1984 KVMCLOCK_SYNC_PERIOD);
1987 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1989 u64 mcg_cap = vcpu->arch.mcg_cap;
1990 unsigned bank_num = mcg_cap & 0xff;
1993 case MSR_IA32_MCG_STATUS:
1994 vcpu->arch.mcg_status = data;
1996 case MSR_IA32_MCG_CTL:
1997 if (!(mcg_cap & MCG_CTL_P))
1999 if (data != 0 && data != ~(u64)0)
2001 vcpu->arch.mcg_ctl = data;
2004 if (msr >= MSR_IA32_MC0_CTL &&
2005 msr < MSR_IA32_MCx_CTL(bank_num)) {
2006 u32 offset = msr - MSR_IA32_MC0_CTL;
2007 /* only 0 or all 1s can be written to IA32_MCi_CTL
2008 * some Linux kernels though clear bit 10 in bank 4 to
2009 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2010 * this to avoid an uncatched #GP in the guest
2012 if ((offset & 0x3) == 0 &&
2013 data != 0 && (data | (1 << 10)) != ~(u64)0)
2015 vcpu->arch.mce_banks[offset] = data;
2023 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2025 struct kvm *kvm = vcpu->kvm;
2026 int lm = is_long_mode(vcpu);
2027 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2028 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2029 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2030 : kvm->arch.xen_hvm_config.blob_size_32;
2031 u32 page_num = data & ~PAGE_MASK;
2032 u64 page_addr = data & PAGE_MASK;
2037 if (page_num >= blob_size)
2040 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2045 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2054 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2056 gpa_t gpa = data & ~0x3f;
2058 /* Bits 3:5 are reserved, Should be zero */
2062 vcpu->arch.apf.msr_val = data;
2064 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2065 kvm_clear_async_pf_completion_queue(vcpu);
2066 kvm_async_pf_hash_reset(vcpu);
2070 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2074 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2075 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2076 kvm_async_pf_wakeup_all(vcpu);
2080 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2082 vcpu->arch.pv_time_enabled = false;
2085 static void record_steal_time(struct kvm_vcpu *vcpu)
2087 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2090 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2091 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2094 vcpu->arch.st.steal.preempted = 0;
2096 if (vcpu->arch.st.steal.version & 1)
2097 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2099 vcpu->arch.st.steal.version += 1;
2101 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2102 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2106 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2107 vcpu->arch.st.last_steal;
2108 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2110 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2111 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2115 vcpu->arch.st.steal.version += 1;
2117 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2118 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2121 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2124 u32 msr = msr_info->index;
2125 u64 data = msr_info->data;
2128 case MSR_AMD64_NB_CFG:
2129 case MSR_IA32_UCODE_REV:
2130 case MSR_IA32_UCODE_WRITE:
2131 case MSR_VM_HSAVE_PA:
2132 case MSR_AMD64_PATCH_LOADER:
2133 case MSR_AMD64_BU_CFG2:
2134 case MSR_AMD64_DC_CFG:
2138 return set_efer(vcpu, data);
2140 data &= ~(u64)0x40; /* ignore flush filter disable */
2141 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2142 data &= ~(u64)0x8; /* ignore TLB cache disable */
2143 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2145 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2150 case MSR_FAM10H_MMIO_CONF_BASE:
2152 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2157 case MSR_IA32_DEBUGCTLMSR:
2159 /* We support the non-activated case already */
2161 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2162 /* Values other than LBR and BTF are vendor-specific,
2163 thus reserved and should throw a #GP */
2166 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2169 case 0x200 ... 0x2ff:
2170 return kvm_mtrr_set_msr(vcpu, msr, data);
2171 case MSR_IA32_APICBASE:
2172 return kvm_set_apic_base(vcpu, msr_info);
2173 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2174 return kvm_x2apic_msr_write(vcpu, msr, data);
2175 case MSR_IA32_TSCDEADLINE:
2176 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2178 case MSR_IA32_TSC_ADJUST:
2179 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2180 if (!msr_info->host_initiated) {
2181 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2182 adjust_tsc_offset_guest(vcpu, adj);
2184 vcpu->arch.ia32_tsc_adjust_msr = data;
2187 case MSR_IA32_MISC_ENABLE:
2188 vcpu->arch.ia32_misc_enable_msr = data;
2190 case MSR_IA32_SMBASE:
2191 if (!msr_info->host_initiated)
2193 vcpu->arch.smbase = data;
2195 case MSR_KVM_WALL_CLOCK_NEW:
2196 case MSR_KVM_WALL_CLOCK:
2197 vcpu->kvm->arch.wall_clock = data;
2198 kvm_write_wall_clock(vcpu->kvm, data);
2200 case MSR_KVM_SYSTEM_TIME_NEW:
2201 case MSR_KVM_SYSTEM_TIME: {
2202 struct kvm_arch *ka = &vcpu->kvm->arch;
2204 kvmclock_reset(vcpu);
2206 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2207 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2209 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2210 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2212 ka->boot_vcpu_runs_old_kvmclock = tmp;
2215 vcpu->arch.time = data;
2216 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2218 /* we verify if the enable bit is set... */
2222 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2223 &vcpu->arch.pv_time, data & ~1ULL,
2224 sizeof(struct pvclock_vcpu_time_info)))
2225 vcpu->arch.pv_time_enabled = false;
2227 vcpu->arch.pv_time_enabled = true;
2231 case MSR_KVM_ASYNC_PF_EN:
2232 if (kvm_pv_enable_async_pf(vcpu, data))
2235 case MSR_KVM_STEAL_TIME:
2237 if (unlikely(!sched_info_on()))
2240 if (data & KVM_STEAL_RESERVED_MASK)
2243 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2244 data & KVM_STEAL_VALID_BITS,
2245 sizeof(struct kvm_steal_time)))
2248 vcpu->arch.st.msr_val = data;
2250 if (!(data & KVM_MSR_ENABLED))
2253 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2256 case MSR_KVM_PV_EOI_EN:
2257 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2261 case MSR_IA32_MCG_CTL:
2262 case MSR_IA32_MCG_STATUS:
2263 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2264 return set_msr_mce(vcpu, msr, data);
2266 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2267 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2268 pr = true; /* fall through */
2269 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2270 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2271 if (kvm_pmu_is_valid_msr(vcpu, msr))
2272 return kvm_pmu_set_msr(vcpu, msr_info);
2274 if (pr || data != 0)
2275 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2276 "0x%x data 0x%llx\n", msr, data);
2278 case MSR_K7_CLK_CTL:
2280 * Ignore all writes to this no longer documented MSR.
2281 * Writes are only relevant for old K7 processors,
2282 * all pre-dating SVM, but a recommended workaround from
2283 * AMD for these chips. It is possible to specify the
2284 * affected processor models on the command line, hence
2285 * the need to ignore the workaround.
2288 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2289 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2290 case HV_X64_MSR_CRASH_CTL:
2291 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2292 return kvm_hv_set_msr_common(vcpu, msr, data,
2293 msr_info->host_initiated);
2294 case MSR_IA32_BBL_CR_CTL3:
2295 /* Drop writes to this legacy MSR -- see rdmsr
2296 * counterpart for further detail.
2298 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", msr, data);
2300 case MSR_AMD64_OSVW_ID_LENGTH:
2301 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2303 vcpu->arch.osvw.length = data;
2305 case MSR_AMD64_OSVW_STATUS:
2306 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2308 vcpu->arch.osvw.status = data;
2310 case MSR_PLATFORM_INFO:
2311 if (!msr_info->host_initiated ||
2312 data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2313 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2314 cpuid_fault_enabled(vcpu)))
2316 vcpu->arch.msr_platform_info = data;
2318 case MSR_MISC_FEATURES_ENABLES:
2319 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2320 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2321 !supports_cpuid_fault(vcpu)))
2323 vcpu->arch.msr_misc_features_enables = data;
2326 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2327 return xen_hvm_config(vcpu, data);
2328 if (kvm_pmu_is_valid_msr(vcpu, msr))
2329 return kvm_pmu_set_msr(vcpu, msr_info);
2331 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2335 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2342 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2346 * Reads an msr value (of 'msr_index') into 'pdata'.
2347 * Returns 0 on success, non-0 otherwise.
2348 * Assumes vcpu_load() was already called.
2350 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2352 return kvm_x86_ops->get_msr(vcpu, msr);
2354 EXPORT_SYMBOL_GPL(kvm_get_msr);
2356 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2359 u64 mcg_cap = vcpu->arch.mcg_cap;
2360 unsigned bank_num = mcg_cap & 0xff;
2363 case MSR_IA32_P5_MC_ADDR:
2364 case MSR_IA32_P5_MC_TYPE:
2367 case MSR_IA32_MCG_CAP:
2368 data = vcpu->arch.mcg_cap;
2370 case MSR_IA32_MCG_CTL:
2371 if (!(mcg_cap & MCG_CTL_P))
2373 data = vcpu->arch.mcg_ctl;
2375 case MSR_IA32_MCG_STATUS:
2376 data = vcpu->arch.mcg_status;
2379 if (msr >= MSR_IA32_MC0_CTL &&
2380 msr < MSR_IA32_MCx_CTL(bank_num)) {
2381 u32 offset = msr - MSR_IA32_MC0_CTL;
2382 data = vcpu->arch.mce_banks[offset];
2391 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2393 switch (msr_info->index) {
2394 case MSR_IA32_PLATFORM_ID:
2395 case MSR_IA32_EBL_CR_POWERON:
2396 case MSR_IA32_DEBUGCTLMSR:
2397 case MSR_IA32_LASTBRANCHFROMIP:
2398 case MSR_IA32_LASTBRANCHTOIP:
2399 case MSR_IA32_LASTINTFROMIP:
2400 case MSR_IA32_LASTINTTOIP:
2402 case MSR_K8_TSEG_ADDR:
2403 case MSR_K8_TSEG_MASK:
2405 case MSR_VM_HSAVE_PA:
2406 case MSR_K8_INT_PENDING_MSG:
2407 case MSR_AMD64_NB_CFG:
2408 case MSR_FAM10H_MMIO_CONF_BASE:
2409 case MSR_AMD64_BU_CFG2:
2410 case MSR_IA32_PERF_CTL:
2411 case MSR_AMD64_DC_CFG:
2414 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2415 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2416 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2417 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2418 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2419 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2422 case MSR_IA32_UCODE_REV:
2423 msr_info->data = 0x100000000ULL;
2426 case 0x200 ... 0x2ff:
2427 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2428 case 0xcd: /* fsb frequency */
2432 * MSR_EBC_FREQUENCY_ID
2433 * Conservative value valid for even the basic CPU models.
2434 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2435 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2436 * and 266MHz for model 3, or 4. Set Core Clock
2437 * Frequency to System Bus Frequency Ratio to 1 (bits
2438 * 31:24) even though these are only valid for CPU
2439 * models > 2, however guests may end up dividing or
2440 * multiplying by zero otherwise.
2442 case MSR_EBC_FREQUENCY_ID:
2443 msr_info->data = 1 << 24;
2445 case MSR_IA32_APICBASE:
2446 msr_info->data = kvm_get_apic_base(vcpu);
2448 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2449 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2451 case MSR_IA32_TSCDEADLINE:
2452 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2454 case MSR_IA32_TSC_ADJUST:
2455 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2457 case MSR_IA32_MISC_ENABLE:
2458 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2460 case MSR_IA32_SMBASE:
2461 if (!msr_info->host_initiated)
2463 msr_info->data = vcpu->arch.smbase;
2465 case MSR_IA32_PERF_STATUS:
2466 /* TSC increment by tick */
2467 msr_info->data = 1000ULL;
2468 /* CPU multiplier */
2469 msr_info->data |= (((uint64_t)4ULL) << 40);
2472 msr_info->data = vcpu->arch.efer;
2474 case MSR_KVM_WALL_CLOCK:
2475 case MSR_KVM_WALL_CLOCK_NEW:
2476 msr_info->data = vcpu->kvm->arch.wall_clock;
2478 case MSR_KVM_SYSTEM_TIME:
2479 case MSR_KVM_SYSTEM_TIME_NEW:
2480 msr_info->data = vcpu->arch.time;
2482 case MSR_KVM_ASYNC_PF_EN:
2483 msr_info->data = vcpu->arch.apf.msr_val;
2485 case MSR_KVM_STEAL_TIME:
2486 msr_info->data = vcpu->arch.st.msr_val;
2488 case MSR_KVM_PV_EOI_EN:
2489 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2491 case MSR_IA32_P5_MC_ADDR:
2492 case MSR_IA32_P5_MC_TYPE:
2493 case MSR_IA32_MCG_CAP:
2494 case MSR_IA32_MCG_CTL:
2495 case MSR_IA32_MCG_STATUS:
2496 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2497 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2498 case MSR_K7_CLK_CTL:
2500 * Provide expected ramp-up count for K7. All other
2501 * are set to zero, indicating minimum divisors for
2504 * This prevents guest kernels on AMD host with CPU
2505 * type 6, model 8 and higher from exploding due to
2506 * the rdmsr failing.
2508 msr_info->data = 0x20000000;
2510 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2511 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2512 case HV_X64_MSR_CRASH_CTL:
2513 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2514 return kvm_hv_get_msr_common(vcpu,
2515 msr_info->index, &msr_info->data);
2517 case MSR_IA32_BBL_CR_CTL3:
2518 /* This legacy MSR exists but isn't fully documented in current
2519 * silicon. It is however accessed by winxp in very narrow
2520 * scenarios where it sets bit #19, itself documented as
2521 * a "reserved" bit. Best effort attempt to source coherent
2522 * read data here should the balance of the register be
2523 * interpreted by the guest:
2525 * L2 cache control register 3: 64GB range, 256KB size,
2526 * enabled, latency 0x1, configured
2528 msr_info->data = 0xbe702111;
2530 case MSR_AMD64_OSVW_ID_LENGTH:
2531 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2533 msr_info->data = vcpu->arch.osvw.length;
2535 case MSR_AMD64_OSVW_STATUS:
2536 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2538 msr_info->data = vcpu->arch.osvw.status;
2540 case MSR_PLATFORM_INFO:
2541 msr_info->data = vcpu->arch.msr_platform_info;
2543 case MSR_MISC_FEATURES_ENABLES:
2544 msr_info->data = vcpu->arch.msr_misc_features_enables;
2547 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2548 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2550 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2554 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2561 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2564 * Read or write a bunch of msrs. All parameters are kernel addresses.
2566 * @return number of msrs set successfully.
2568 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2569 struct kvm_msr_entry *entries,
2570 int (*do_msr)(struct kvm_vcpu *vcpu,
2571 unsigned index, u64 *data))
2575 idx = srcu_read_lock(&vcpu->kvm->srcu);
2576 for (i = 0; i < msrs->nmsrs; ++i)
2577 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2579 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2585 * Read or write a bunch of msrs. Parameters are user addresses.
2587 * @return number of msrs set successfully.
2589 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2590 int (*do_msr)(struct kvm_vcpu *vcpu,
2591 unsigned index, u64 *data),
2594 struct kvm_msrs msrs;
2595 struct kvm_msr_entry *entries;
2600 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2604 if (msrs.nmsrs >= MAX_IO_MSRS)
2607 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2608 entries = memdup_user(user_msrs->entries, size);
2609 if (IS_ERR(entries)) {
2610 r = PTR_ERR(entries);
2614 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2619 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2630 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2635 case KVM_CAP_IRQCHIP:
2637 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2638 case KVM_CAP_SET_TSS_ADDR:
2639 case KVM_CAP_EXT_CPUID:
2640 case KVM_CAP_EXT_EMUL_CPUID:
2641 case KVM_CAP_CLOCKSOURCE:
2643 case KVM_CAP_NOP_IO_DELAY:
2644 case KVM_CAP_MP_STATE:
2645 case KVM_CAP_SYNC_MMU:
2646 case KVM_CAP_USER_NMI:
2647 case KVM_CAP_REINJECT_CONTROL:
2648 case KVM_CAP_IRQ_INJECT_STATUS:
2649 case KVM_CAP_IOEVENTFD:
2650 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2652 case KVM_CAP_PIT_STATE2:
2653 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2654 case KVM_CAP_XEN_HVM:
2655 case KVM_CAP_VCPU_EVENTS:
2656 case KVM_CAP_HYPERV:
2657 case KVM_CAP_HYPERV_VAPIC:
2658 case KVM_CAP_HYPERV_SPIN:
2659 case KVM_CAP_HYPERV_SYNIC:
2660 case KVM_CAP_HYPERV_SYNIC2:
2661 case KVM_CAP_HYPERV_VP_INDEX:
2662 case KVM_CAP_PCI_SEGMENT:
2663 case KVM_CAP_DEBUGREGS:
2664 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2666 case KVM_CAP_ASYNC_PF:
2667 case KVM_CAP_GET_TSC_KHZ:
2668 case KVM_CAP_KVMCLOCK_CTRL:
2669 case KVM_CAP_READONLY_MEM:
2670 case KVM_CAP_HYPERV_TIME:
2671 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2672 case KVM_CAP_TSC_DEADLINE_TIMER:
2673 case KVM_CAP_ENABLE_CAP_VM:
2674 case KVM_CAP_DISABLE_QUIRKS:
2675 case KVM_CAP_SET_BOOT_CPU_ID:
2676 case KVM_CAP_SPLIT_IRQCHIP:
2677 case KVM_CAP_IMMEDIATE_EXIT:
2680 case KVM_CAP_ADJUST_CLOCK:
2681 r = KVM_CLOCK_TSC_STABLE;
2683 case KVM_CAP_X86_GUEST_MWAIT:
2684 r = kvm_mwait_in_guest();
2686 case KVM_CAP_X86_SMM:
2687 /* SMBASE is usually relocated above 1M on modern chipsets,
2688 * and SMM handlers might indeed rely on 4G segment limits,
2689 * so do not report SMM to be available if real mode is
2690 * emulated via vm86 mode. Still, do not go to great lengths
2691 * to avoid userspace's usage of the feature, because it is a
2692 * fringe case that is not enabled except via specific settings
2693 * of the module parameters.
2695 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2698 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2700 case KVM_CAP_NR_VCPUS:
2701 r = KVM_SOFT_MAX_VCPUS;
2703 case KVM_CAP_MAX_VCPUS:
2706 case KVM_CAP_NR_MEMSLOTS:
2707 r = KVM_USER_MEM_SLOTS;
2709 case KVM_CAP_PV_MMU: /* obsolete */
2713 r = KVM_MAX_MCE_BANKS;
2716 r = boot_cpu_has(X86_FEATURE_XSAVE);
2718 case KVM_CAP_TSC_CONTROL:
2719 r = kvm_has_tsc_control;
2721 case KVM_CAP_X2APIC_API:
2722 r = KVM_X2APIC_API_VALID_FLAGS;
2732 long kvm_arch_dev_ioctl(struct file *filp,
2733 unsigned int ioctl, unsigned long arg)
2735 void __user *argp = (void __user *)arg;
2739 case KVM_GET_MSR_INDEX_LIST: {
2740 struct kvm_msr_list __user *user_msr_list = argp;
2741 struct kvm_msr_list msr_list;
2745 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2748 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2749 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2752 if (n < msr_list.nmsrs)
2755 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2756 num_msrs_to_save * sizeof(u32)))
2758 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2760 num_emulated_msrs * sizeof(u32)))
2765 case KVM_GET_SUPPORTED_CPUID:
2766 case KVM_GET_EMULATED_CPUID: {
2767 struct kvm_cpuid2 __user *cpuid_arg = argp;
2768 struct kvm_cpuid2 cpuid;
2771 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2774 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2780 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2785 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2787 if (copy_to_user(argp, &kvm_mce_cap_supported,
2788 sizeof(kvm_mce_cap_supported)))
2800 static void wbinvd_ipi(void *garbage)
2805 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2807 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2810 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2812 /* Address WBINVD may be executed by guest */
2813 if (need_emulate_wbinvd(vcpu)) {
2814 if (kvm_x86_ops->has_wbinvd_exit())
2815 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2816 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2817 smp_call_function_single(vcpu->cpu,
2818 wbinvd_ipi, NULL, 1);
2821 kvm_x86_ops->vcpu_load(vcpu, cpu);
2823 /* Apply any externally detected TSC adjustments (due to suspend) */
2824 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2825 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2826 vcpu->arch.tsc_offset_adjustment = 0;
2827 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2830 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2831 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2832 rdtsc() - vcpu->arch.last_host_tsc;
2834 mark_tsc_unstable("KVM discovered backwards TSC");
2836 if (check_tsc_unstable()) {
2837 u64 offset = kvm_compute_tsc_offset(vcpu,
2838 vcpu->arch.last_guest_tsc);
2839 kvm_vcpu_write_tsc_offset(vcpu, offset);
2840 vcpu->arch.tsc_catchup = 1;
2843 if (kvm_lapic_hv_timer_in_use(vcpu))
2844 kvm_lapic_restart_hv_timer(vcpu);
2847 * On a host with synchronized TSC, there is no need to update
2848 * kvmclock on vcpu->cpu migration
2850 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2851 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2852 if (vcpu->cpu != cpu)
2853 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
2857 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2860 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2862 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2865 vcpu->arch.st.steal.preempted = 1;
2867 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
2868 &vcpu->arch.st.steal.preempted,
2869 offsetof(struct kvm_steal_time, preempted),
2870 sizeof(vcpu->arch.st.steal.preempted));
2873 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2877 if (vcpu->preempted)
2878 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
2881 * Disable page faults because we're in atomic context here.
2882 * kvm_write_guest_offset_cached() would call might_fault()
2883 * that relies on pagefault_disable() to tell if there's a
2884 * bug. NOTE: the write to guest memory may not go through if
2885 * during postcopy live migration or if there's heavy guest
2888 pagefault_disable();
2890 * kvm_memslots() will be called by
2891 * kvm_write_guest_offset_cached() so take the srcu lock.
2893 idx = srcu_read_lock(&vcpu->kvm->srcu);
2894 kvm_steal_time_set_preempted(vcpu);
2895 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2897 kvm_x86_ops->vcpu_put(vcpu);
2898 kvm_put_guest_fpu(vcpu);
2899 vcpu->arch.last_host_tsc = rdtsc();
2902 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2903 struct kvm_lapic_state *s)
2905 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
2906 kvm_x86_ops->sync_pir_to_irr(vcpu);
2908 return kvm_apic_get_state(vcpu, s);
2911 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2912 struct kvm_lapic_state *s)
2916 r = kvm_apic_set_state(vcpu, s);
2919 update_cr8_intercept(vcpu);
2924 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2926 return (!lapic_in_kernel(vcpu) ||
2927 kvm_apic_accept_pic_intr(vcpu));
2931 * if userspace requested an interrupt window, check that the
2932 * interrupt window is open.
2934 * No need to exit to userspace if we already have an interrupt queued.
2936 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2938 return kvm_arch_interrupt_allowed(vcpu) &&
2939 !kvm_cpu_has_interrupt(vcpu) &&
2940 !kvm_event_needs_reinjection(vcpu) &&
2941 kvm_cpu_accept_dm_intr(vcpu);
2944 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2945 struct kvm_interrupt *irq)
2947 if (irq->irq >= KVM_NR_INTERRUPTS)
2950 if (!irqchip_in_kernel(vcpu->kvm)) {
2951 kvm_queue_interrupt(vcpu, irq->irq, false);
2952 kvm_make_request(KVM_REQ_EVENT, vcpu);
2957 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2958 * fail for in-kernel 8259.
2960 if (pic_in_kernel(vcpu->kvm))
2963 if (vcpu->arch.pending_external_vector != -1)
2966 vcpu->arch.pending_external_vector = irq->irq;
2967 kvm_make_request(KVM_REQ_EVENT, vcpu);
2971 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2973 kvm_inject_nmi(vcpu);
2978 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2980 kvm_make_request(KVM_REQ_SMI, vcpu);
2985 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2986 struct kvm_tpr_access_ctl *tac)
2990 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2994 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2998 unsigned bank_num = mcg_cap & 0xff, bank;
3001 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3003 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3006 vcpu->arch.mcg_cap = mcg_cap;
3007 /* Init IA32_MCG_CTL to all 1s */
3008 if (mcg_cap & MCG_CTL_P)
3009 vcpu->arch.mcg_ctl = ~(u64)0;
3010 /* Init IA32_MCi_CTL to all 1s */
3011 for (bank = 0; bank < bank_num; bank++)
3012 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3014 if (kvm_x86_ops->setup_mce)
3015 kvm_x86_ops->setup_mce(vcpu);
3020 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3021 struct kvm_x86_mce *mce)
3023 u64 mcg_cap = vcpu->arch.mcg_cap;
3024 unsigned bank_num = mcg_cap & 0xff;
3025 u64 *banks = vcpu->arch.mce_banks;
3027 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3030 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3031 * reporting is disabled
3033 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3034 vcpu->arch.mcg_ctl != ~(u64)0)
3036 banks += 4 * mce->bank;
3038 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3039 * reporting is disabled for the bank
3041 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3043 if (mce->status & MCI_STATUS_UC) {
3044 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3045 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3046 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3049 if (banks[1] & MCI_STATUS_VAL)
3050 mce->status |= MCI_STATUS_OVER;
3051 banks[2] = mce->addr;
3052 banks[3] = mce->misc;
3053 vcpu->arch.mcg_status = mce->mcg_status;
3054 banks[1] = mce->status;
3055 kvm_queue_exception(vcpu, MC_VECTOR);
3056 } else if (!(banks[1] & MCI_STATUS_VAL)
3057 || !(banks[1] & MCI_STATUS_UC)) {
3058 if (banks[1] & MCI_STATUS_VAL)
3059 mce->status |= MCI_STATUS_OVER;
3060 banks[2] = mce->addr;
3061 banks[3] = mce->misc;
3062 banks[1] = mce->status;
3064 banks[1] |= MCI_STATUS_OVER;
3068 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3069 struct kvm_vcpu_events *events)
3072 events->exception.injected =
3073 vcpu->arch.exception.pending &&
3074 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3075 events->exception.nr = vcpu->arch.exception.nr;
3076 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3077 events->exception.pad = 0;
3078 events->exception.error_code = vcpu->arch.exception.error_code;
3080 events->interrupt.injected =
3081 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3082 events->interrupt.nr = vcpu->arch.interrupt.nr;
3083 events->interrupt.soft = 0;
3084 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3086 events->nmi.injected = vcpu->arch.nmi_injected;
3087 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3088 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3089 events->nmi.pad = 0;
3091 events->sipi_vector = 0; /* never valid when reporting to user space */
3093 events->smi.smm = is_smm(vcpu);
3094 events->smi.pending = vcpu->arch.smi_pending;
3095 events->smi.smm_inside_nmi =
3096 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3097 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3099 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3100 | KVM_VCPUEVENT_VALID_SHADOW
3101 | KVM_VCPUEVENT_VALID_SMM);
3102 memset(&events->reserved, 0, sizeof(events->reserved));
3105 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3107 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3108 struct kvm_vcpu_events *events)
3110 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3111 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3112 | KVM_VCPUEVENT_VALID_SHADOW
3113 | KVM_VCPUEVENT_VALID_SMM))
3116 if (events->exception.injected &&
3117 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3118 is_guest_mode(vcpu)))
3121 /* INITs are latched while in SMM */
3122 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3123 (events->smi.smm || events->smi.pending) &&
3124 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3128 vcpu->arch.exception.pending = events->exception.injected;
3129 vcpu->arch.exception.nr = events->exception.nr;
3130 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3131 vcpu->arch.exception.error_code = events->exception.error_code;
3133 vcpu->arch.interrupt.pending = events->interrupt.injected;
3134 vcpu->arch.interrupt.nr = events->interrupt.nr;
3135 vcpu->arch.interrupt.soft = events->interrupt.soft;
3136 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3137 kvm_x86_ops->set_interrupt_shadow(vcpu,
3138 events->interrupt.shadow);
3140 vcpu->arch.nmi_injected = events->nmi.injected;
3141 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3142 vcpu->arch.nmi_pending = events->nmi.pending;
3143 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3145 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3146 lapic_in_kernel(vcpu))
3147 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3149 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3150 u32 hflags = vcpu->arch.hflags;
3151 if (events->smi.smm)
3152 hflags |= HF_SMM_MASK;
3154 hflags &= ~HF_SMM_MASK;
3155 kvm_set_hflags(vcpu, hflags);
3157 vcpu->arch.smi_pending = events->smi.pending;
3159 if (events->smi.smm) {
3160 if (events->smi.smm_inside_nmi)
3161 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3163 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3164 if (lapic_in_kernel(vcpu)) {
3165 if (events->smi.latched_init)
3166 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3168 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3173 kvm_make_request(KVM_REQ_EVENT, vcpu);
3178 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3179 struct kvm_debugregs *dbgregs)
3183 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3184 kvm_get_dr(vcpu, 6, &val);
3186 dbgregs->dr7 = vcpu->arch.dr7;
3188 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3191 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3192 struct kvm_debugregs *dbgregs)
3197 if (dbgregs->dr6 & ~0xffffffffull)
3199 if (dbgregs->dr7 & ~0xffffffffull)
3202 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3203 kvm_update_dr0123(vcpu);
3204 vcpu->arch.dr6 = dbgregs->dr6;
3205 kvm_update_dr6(vcpu);
3206 vcpu->arch.dr7 = dbgregs->dr7;
3207 kvm_update_dr7(vcpu);
3212 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3214 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3216 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3217 u64 xstate_bv = xsave->header.xfeatures;
3221 * Copy legacy XSAVE area, to avoid complications with CPUID
3222 * leaves 0 and 1 in the loop below.
3224 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3227 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3228 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3231 * Copy each region from the possibly compacted offset to the
3232 * non-compacted offset.
3234 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3236 u64 feature = valid & -valid;
3237 int index = fls64(feature) - 1;
3238 void *src = get_xsave_addr(xsave, feature);
3241 u32 size, offset, ecx, edx;
3242 cpuid_count(XSTATE_CPUID, index,
3243 &size, &offset, &ecx, &edx);
3244 memcpy(dest + offset, src, size);
3251 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3253 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3254 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3258 * Copy legacy XSAVE area, to avoid complications with CPUID
3259 * leaves 0 and 1 in the loop below.
3261 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3263 /* Set XSTATE_BV and possibly XCOMP_BV. */
3264 xsave->header.xfeatures = xstate_bv;
3265 if (boot_cpu_has(X86_FEATURE_XSAVES))
3266 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3269 * Copy each region from the non-compacted offset to the
3270 * possibly compacted offset.
3272 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3274 u64 feature = valid & -valid;
3275 int index = fls64(feature) - 1;
3276 void *dest = get_xsave_addr(xsave, feature);
3279 u32 size, offset, ecx, edx;
3280 cpuid_count(XSTATE_CPUID, index,
3281 &size, &offset, &ecx, &edx);
3282 memcpy(dest, src + offset, size);
3289 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3290 struct kvm_xsave *guest_xsave)
3292 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3293 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3294 fill_xsave((u8 *) guest_xsave->region, vcpu);
3296 memcpy(guest_xsave->region,
3297 &vcpu->arch.guest_fpu.state.fxsave,
3298 sizeof(struct fxregs_state));
3299 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3300 XFEATURE_MASK_FPSSE;
3304 #define XSAVE_MXCSR_OFFSET 24
3306 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3307 struct kvm_xsave *guest_xsave)
3310 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3311 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3313 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3315 * Here we allow setting states that are not present in
3316 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3317 * with old userspace.
3319 if (xstate_bv & ~kvm_supported_xcr0() ||
3320 mxcsr & ~mxcsr_feature_mask)
3322 load_xsave(vcpu, (u8 *)guest_xsave->region);
3324 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3325 mxcsr & ~mxcsr_feature_mask)
3327 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3328 guest_xsave->region, sizeof(struct fxregs_state));
3333 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3334 struct kvm_xcrs *guest_xcrs)
3336 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3337 guest_xcrs->nr_xcrs = 0;
3341 guest_xcrs->nr_xcrs = 1;
3342 guest_xcrs->flags = 0;
3343 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3344 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3347 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3348 struct kvm_xcrs *guest_xcrs)
3352 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3355 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3358 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3359 /* Only support XCR0 currently */
3360 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3361 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3362 guest_xcrs->xcrs[i].value);
3371 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3372 * stopped by the hypervisor. This function will be called from the host only.
3373 * EINVAL is returned when the host attempts to set the flag for a guest that
3374 * does not support pv clocks.
3376 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3378 if (!vcpu->arch.pv_time_enabled)
3380 vcpu->arch.pvclock_set_guest_stopped_request = true;
3381 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3385 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3386 struct kvm_enable_cap *cap)
3392 case KVM_CAP_HYPERV_SYNIC2:
3395 case KVM_CAP_HYPERV_SYNIC:
3396 if (!irqchip_in_kernel(vcpu->kvm))
3398 return kvm_hv_activate_synic(vcpu, cap->cap ==
3399 KVM_CAP_HYPERV_SYNIC2);
3405 long kvm_arch_vcpu_ioctl(struct file *filp,
3406 unsigned int ioctl, unsigned long arg)
3408 struct kvm_vcpu *vcpu = filp->private_data;
3409 void __user *argp = (void __user *)arg;
3412 struct kvm_lapic_state *lapic;
3413 struct kvm_xsave *xsave;
3414 struct kvm_xcrs *xcrs;
3420 case KVM_GET_LAPIC: {
3422 if (!lapic_in_kernel(vcpu))
3424 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3429 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3433 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3438 case KVM_SET_LAPIC: {
3440 if (!lapic_in_kernel(vcpu))
3442 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3443 if (IS_ERR(u.lapic))
3444 return PTR_ERR(u.lapic);
3446 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3449 case KVM_INTERRUPT: {
3450 struct kvm_interrupt irq;
3453 if (copy_from_user(&irq, argp, sizeof irq))
3455 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3459 r = kvm_vcpu_ioctl_nmi(vcpu);
3463 r = kvm_vcpu_ioctl_smi(vcpu);
3466 case KVM_SET_CPUID: {
3467 struct kvm_cpuid __user *cpuid_arg = argp;
3468 struct kvm_cpuid cpuid;
3471 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3473 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3476 case KVM_SET_CPUID2: {
3477 struct kvm_cpuid2 __user *cpuid_arg = argp;
3478 struct kvm_cpuid2 cpuid;
3481 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3483 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3484 cpuid_arg->entries);
3487 case KVM_GET_CPUID2: {
3488 struct kvm_cpuid2 __user *cpuid_arg = argp;
3489 struct kvm_cpuid2 cpuid;
3492 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3494 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3495 cpuid_arg->entries);
3499 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3505 r = msr_io(vcpu, argp, do_get_msr, 1);
3508 r = msr_io(vcpu, argp, do_set_msr, 0);
3510 case KVM_TPR_ACCESS_REPORTING: {
3511 struct kvm_tpr_access_ctl tac;
3514 if (copy_from_user(&tac, argp, sizeof tac))
3516 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3520 if (copy_to_user(argp, &tac, sizeof tac))
3525 case KVM_SET_VAPIC_ADDR: {
3526 struct kvm_vapic_addr va;
3530 if (!lapic_in_kernel(vcpu))
3533 if (copy_from_user(&va, argp, sizeof va))
3535 idx = srcu_read_lock(&vcpu->kvm->srcu);
3536 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3537 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3540 case KVM_X86_SETUP_MCE: {
3544 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3546 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3549 case KVM_X86_SET_MCE: {
3550 struct kvm_x86_mce mce;
3553 if (copy_from_user(&mce, argp, sizeof mce))
3555 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3558 case KVM_GET_VCPU_EVENTS: {
3559 struct kvm_vcpu_events events;
3561 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3564 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3569 case KVM_SET_VCPU_EVENTS: {
3570 struct kvm_vcpu_events events;
3573 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3576 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3579 case KVM_GET_DEBUGREGS: {
3580 struct kvm_debugregs dbgregs;
3582 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3585 if (copy_to_user(argp, &dbgregs,
3586 sizeof(struct kvm_debugregs)))
3591 case KVM_SET_DEBUGREGS: {
3592 struct kvm_debugregs dbgregs;
3595 if (copy_from_user(&dbgregs, argp,
3596 sizeof(struct kvm_debugregs)))
3599 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3602 case KVM_GET_XSAVE: {
3603 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3608 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3611 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3616 case KVM_SET_XSAVE: {
3617 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3618 if (IS_ERR(u.xsave))
3619 return PTR_ERR(u.xsave);
3621 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3624 case KVM_GET_XCRS: {
3625 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3630 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3633 if (copy_to_user(argp, u.xcrs,
3634 sizeof(struct kvm_xcrs)))
3639 case KVM_SET_XCRS: {
3640 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3642 return PTR_ERR(u.xcrs);
3644 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3647 case KVM_SET_TSC_KHZ: {
3651 user_tsc_khz = (u32)arg;
3653 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3656 if (user_tsc_khz == 0)
3657 user_tsc_khz = tsc_khz;
3659 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3664 case KVM_GET_TSC_KHZ: {
3665 r = vcpu->arch.virtual_tsc_khz;
3668 case KVM_KVMCLOCK_CTRL: {
3669 r = kvm_set_guest_paused(vcpu);
3672 case KVM_ENABLE_CAP: {
3673 struct kvm_enable_cap cap;
3676 if (copy_from_user(&cap, argp, sizeof(cap)))
3678 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3689 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3691 return VM_FAULT_SIGBUS;
3694 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3698 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3700 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3704 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3707 kvm->arch.ept_identity_map_addr = ident_addr;
3711 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3712 u32 kvm_nr_mmu_pages)
3714 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3717 mutex_lock(&kvm->slots_lock);
3719 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3720 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3722 mutex_unlock(&kvm->slots_lock);
3726 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3728 return kvm->arch.n_max_mmu_pages;
3731 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3733 struct kvm_pic *pic = kvm->arch.vpic;
3737 switch (chip->chip_id) {
3738 case KVM_IRQCHIP_PIC_MASTER:
3739 memcpy(&chip->chip.pic, &pic->pics[0],
3740 sizeof(struct kvm_pic_state));
3742 case KVM_IRQCHIP_PIC_SLAVE:
3743 memcpy(&chip->chip.pic, &pic->pics[1],
3744 sizeof(struct kvm_pic_state));
3746 case KVM_IRQCHIP_IOAPIC:
3747 kvm_get_ioapic(kvm, &chip->chip.ioapic);
3756 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3758 struct kvm_pic *pic = kvm->arch.vpic;
3762 switch (chip->chip_id) {
3763 case KVM_IRQCHIP_PIC_MASTER:
3764 spin_lock(&pic->lock);
3765 memcpy(&pic->pics[0], &chip->chip.pic,
3766 sizeof(struct kvm_pic_state));
3767 spin_unlock(&pic->lock);
3769 case KVM_IRQCHIP_PIC_SLAVE:
3770 spin_lock(&pic->lock);
3771 memcpy(&pic->pics[1], &chip->chip.pic,
3772 sizeof(struct kvm_pic_state));
3773 spin_unlock(&pic->lock);
3775 case KVM_IRQCHIP_IOAPIC:
3776 kvm_set_ioapic(kvm, &chip->chip.ioapic);
3782 kvm_pic_update_irq(pic);
3786 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3788 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3790 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3792 mutex_lock(&kps->lock);
3793 memcpy(ps, &kps->channels, sizeof(*ps));
3794 mutex_unlock(&kps->lock);
3798 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3801 struct kvm_pit *pit = kvm->arch.vpit;
3803 mutex_lock(&pit->pit_state.lock);
3804 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3805 for (i = 0; i < 3; i++)
3806 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3807 mutex_unlock(&pit->pit_state.lock);
3811 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3813 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3814 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3815 sizeof(ps->channels));
3816 ps->flags = kvm->arch.vpit->pit_state.flags;
3817 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3818 memset(&ps->reserved, 0, sizeof(ps->reserved));
3822 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3826 u32 prev_legacy, cur_legacy;
3827 struct kvm_pit *pit = kvm->arch.vpit;
3829 mutex_lock(&pit->pit_state.lock);
3830 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3831 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3832 if (!prev_legacy && cur_legacy)
3834 memcpy(&pit->pit_state.channels, &ps->channels,
3835 sizeof(pit->pit_state.channels));
3836 pit->pit_state.flags = ps->flags;
3837 for (i = 0; i < 3; i++)
3838 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3840 mutex_unlock(&pit->pit_state.lock);
3844 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3845 struct kvm_reinject_control *control)
3847 struct kvm_pit *pit = kvm->arch.vpit;
3852 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3853 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3854 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3856 mutex_lock(&pit->pit_state.lock);
3857 kvm_pit_set_reinject(pit, control->pit_reinject);
3858 mutex_unlock(&pit->pit_state.lock);
3864 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3865 * @kvm: kvm instance
3866 * @log: slot id and address to which we copy the log
3868 * Steps 1-4 below provide general overview of dirty page logging. See
3869 * kvm_get_dirty_log_protect() function description for additional details.
3871 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3872 * always flush the TLB (step 4) even if previous step failed and the dirty
3873 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3874 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3875 * writes will be marked dirty for next log read.
3877 * 1. Take a snapshot of the bit and clear it if needed.
3878 * 2. Write protect the corresponding page.
3879 * 3. Copy the snapshot to the userspace.
3880 * 4. Flush TLB's if needed.
3882 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3884 bool is_dirty = false;
3887 mutex_lock(&kvm->slots_lock);
3890 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3892 if (kvm_x86_ops->flush_log_dirty)
3893 kvm_x86_ops->flush_log_dirty(kvm);
3895 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3898 * All the TLBs can be flushed out of mmu lock, see the comments in
3899 * kvm_mmu_slot_remove_write_access().
3901 lockdep_assert_held(&kvm->slots_lock);
3903 kvm_flush_remote_tlbs(kvm);
3905 mutex_unlock(&kvm->slots_lock);
3909 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3912 if (!irqchip_in_kernel(kvm))
3915 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3916 irq_event->irq, irq_event->level,
3921 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3922 struct kvm_enable_cap *cap)
3930 case KVM_CAP_DISABLE_QUIRKS:
3931 kvm->arch.disabled_quirks = cap->args[0];
3934 case KVM_CAP_SPLIT_IRQCHIP: {
3935 mutex_lock(&kvm->lock);
3937 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3938 goto split_irqchip_unlock;
3940 if (irqchip_in_kernel(kvm))
3941 goto split_irqchip_unlock;
3942 if (kvm->created_vcpus)
3943 goto split_irqchip_unlock;
3944 r = kvm_setup_empty_irq_routing(kvm);
3946 goto split_irqchip_unlock;
3947 /* Pairs with irqchip_in_kernel. */
3949 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
3950 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3952 split_irqchip_unlock:
3953 mutex_unlock(&kvm->lock);
3956 case KVM_CAP_X2APIC_API:
3958 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3961 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3962 kvm->arch.x2apic_format = true;
3963 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3964 kvm->arch.x2apic_broadcast_quirk_disabled = true;
3975 long kvm_arch_vm_ioctl(struct file *filp,
3976 unsigned int ioctl, unsigned long arg)
3978 struct kvm *kvm = filp->private_data;
3979 void __user *argp = (void __user *)arg;
3982 * This union makes it completely explicit to gcc-3.x
3983 * that these two variables' stack usage should be
3984 * combined, not added together.
3987 struct kvm_pit_state ps;
3988 struct kvm_pit_state2 ps2;
3989 struct kvm_pit_config pit_config;
3993 case KVM_SET_TSS_ADDR:
3994 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3996 case KVM_SET_IDENTITY_MAP_ADDR: {
4000 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4002 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4005 case KVM_SET_NR_MMU_PAGES:
4006 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4008 case KVM_GET_NR_MMU_PAGES:
4009 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4011 case KVM_CREATE_IRQCHIP: {
4012 mutex_lock(&kvm->lock);
4015 if (irqchip_in_kernel(kvm))
4016 goto create_irqchip_unlock;
4019 if (kvm->created_vcpus)
4020 goto create_irqchip_unlock;
4022 r = kvm_pic_init(kvm);
4024 goto create_irqchip_unlock;
4026 r = kvm_ioapic_init(kvm);
4028 kvm_pic_destroy(kvm);
4029 goto create_irqchip_unlock;
4032 r = kvm_setup_default_irq_routing(kvm);
4034 kvm_ioapic_destroy(kvm);
4035 kvm_pic_destroy(kvm);
4036 goto create_irqchip_unlock;
4038 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4040 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4041 create_irqchip_unlock:
4042 mutex_unlock(&kvm->lock);
4045 case KVM_CREATE_PIT:
4046 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4048 case KVM_CREATE_PIT2:
4050 if (copy_from_user(&u.pit_config, argp,
4051 sizeof(struct kvm_pit_config)))
4054 mutex_lock(&kvm->lock);
4057 goto create_pit_unlock;
4059 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4063 mutex_unlock(&kvm->lock);
4065 case KVM_GET_IRQCHIP: {
4066 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4067 struct kvm_irqchip *chip;
4069 chip = memdup_user(argp, sizeof(*chip));
4076 if (!irqchip_kernel(kvm))
4077 goto get_irqchip_out;
4078 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4080 goto get_irqchip_out;
4082 if (copy_to_user(argp, chip, sizeof *chip))
4083 goto get_irqchip_out;
4089 case KVM_SET_IRQCHIP: {
4090 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4091 struct kvm_irqchip *chip;
4093 chip = memdup_user(argp, sizeof(*chip));
4100 if (!irqchip_kernel(kvm))
4101 goto set_irqchip_out;
4102 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4104 goto set_irqchip_out;
4112 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4115 if (!kvm->arch.vpit)
4117 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4121 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4128 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4131 if (!kvm->arch.vpit)
4133 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4136 case KVM_GET_PIT2: {
4138 if (!kvm->arch.vpit)
4140 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4144 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4149 case KVM_SET_PIT2: {
4151 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4154 if (!kvm->arch.vpit)
4156 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4159 case KVM_REINJECT_CONTROL: {
4160 struct kvm_reinject_control control;
4162 if (copy_from_user(&control, argp, sizeof(control)))
4164 r = kvm_vm_ioctl_reinject(kvm, &control);
4167 case KVM_SET_BOOT_CPU_ID:
4169 mutex_lock(&kvm->lock);
4170 if (kvm->created_vcpus)
4173 kvm->arch.bsp_vcpu_id = arg;
4174 mutex_unlock(&kvm->lock);
4176 case KVM_XEN_HVM_CONFIG: {
4178 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4179 sizeof(struct kvm_xen_hvm_config)))
4182 if (kvm->arch.xen_hvm_config.flags)
4187 case KVM_SET_CLOCK: {
4188 struct kvm_clock_data user_ns;
4192 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4201 * TODO: userspace has to take care of races with VCPU_RUN, so
4202 * kvm_gen_update_masterclock() can be cut down to locked
4203 * pvclock_update_vm_gtod_copy().
4205 kvm_gen_update_masterclock(kvm);
4206 now_ns = get_kvmclock_ns(kvm);
4207 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4208 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4211 case KVM_GET_CLOCK: {
4212 struct kvm_clock_data user_ns;
4215 now_ns = get_kvmclock_ns(kvm);
4216 user_ns.clock = now_ns;
4217 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4218 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4221 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4226 case KVM_ENABLE_CAP: {
4227 struct kvm_enable_cap cap;
4230 if (copy_from_user(&cap, argp, sizeof(cap)))
4232 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4242 static void kvm_init_msr_list(void)
4247 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4248 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4252 * Even MSRs that are valid in the host may not be exposed
4253 * to the guests in some cases.
4255 switch (msrs_to_save[i]) {
4256 case MSR_IA32_BNDCFGS:
4257 if (!kvm_x86_ops->mpx_supported())
4261 if (!kvm_x86_ops->rdtscp_supported())
4269 msrs_to_save[j] = msrs_to_save[i];
4272 num_msrs_to_save = j;
4274 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4275 switch (emulated_msrs[i]) {
4276 case MSR_IA32_SMBASE:
4277 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4285 emulated_msrs[j] = emulated_msrs[i];
4288 num_emulated_msrs = j;
4291 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4299 if (!(lapic_in_kernel(vcpu) &&
4300 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4301 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4312 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4319 if (!(lapic_in_kernel(vcpu) &&
4320 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4322 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4324 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4334 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4335 struct kvm_segment *var, int seg)
4337 kvm_x86_ops->set_segment(vcpu, var, seg);
4340 void kvm_get_segment(struct kvm_vcpu *vcpu,
4341 struct kvm_segment *var, int seg)
4343 kvm_x86_ops->get_segment(vcpu, var, seg);
4346 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4347 struct x86_exception *exception)
4351 BUG_ON(!mmu_is_nested(vcpu));
4353 /* NPT walks are always user-walks */
4354 access |= PFERR_USER_MASK;
4355 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4360 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4361 struct x86_exception *exception)
4363 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4364 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4367 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4368 struct x86_exception *exception)
4370 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4371 access |= PFERR_FETCH_MASK;
4372 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4375 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4376 struct x86_exception *exception)
4378 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4379 access |= PFERR_WRITE_MASK;
4380 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4383 /* uses this to access any guest's mapped memory without checking CPL */
4384 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4385 struct x86_exception *exception)
4387 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4390 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4391 struct kvm_vcpu *vcpu, u32 access,
4392 struct x86_exception *exception)
4395 int r = X86EMUL_CONTINUE;
4398 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4400 unsigned offset = addr & (PAGE_SIZE-1);
4401 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4404 if (gpa == UNMAPPED_GVA)
4405 return X86EMUL_PROPAGATE_FAULT;
4406 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4409 r = X86EMUL_IO_NEEDED;
4421 /* used for instruction fetching */
4422 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4423 gva_t addr, void *val, unsigned int bytes,
4424 struct x86_exception *exception)
4426 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4427 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4431 /* Inline kvm_read_guest_virt_helper for speed. */
4432 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4434 if (unlikely(gpa == UNMAPPED_GVA))
4435 return X86EMUL_PROPAGATE_FAULT;
4437 offset = addr & (PAGE_SIZE-1);
4438 if (WARN_ON(offset + bytes > PAGE_SIZE))
4439 bytes = (unsigned)PAGE_SIZE - offset;
4440 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4442 if (unlikely(ret < 0))
4443 return X86EMUL_IO_NEEDED;
4445 return X86EMUL_CONTINUE;
4448 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4449 gva_t addr, void *val, unsigned int bytes,
4450 struct x86_exception *exception)
4452 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4453 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4455 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4458 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4460 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4461 gva_t addr, void *val, unsigned int bytes,
4462 struct x86_exception *exception)
4464 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4465 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4468 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4469 unsigned long addr, void *val, unsigned int bytes)
4471 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4472 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4474 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4477 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4478 gva_t addr, void *val,
4480 struct x86_exception *exception)
4482 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4484 int r = X86EMUL_CONTINUE;
4487 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4490 unsigned offset = addr & (PAGE_SIZE-1);
4491 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4494 if (gpa == UNMAPPED_GVA)
4495 return X86EMUL_PROPAGATE_FAULT;
4496 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4498 r = X86EMUL_IO_NEEDED;
4509 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4511 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4512 gpa_t gpa, bool write)
4514 /* For APIC access vmexit */
4515 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4518 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4519 trace_vcpu_match_mmio(gva, gpa, write, true);
4526 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4527 gpa_t *gpa, struct x86_exception *exception,
4530 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4531 | (write ? PFERR_WRITE_MASK : 0);
4534 * currently PKRU is only applied to ept enabled guest so
4535 * there is no pkey in EPT page table for L1 guest or EPT
4536 * shadow page table for L2 guest.
4538 if (vcpu_match_mmio_gva(vcpu, gva)
4539 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4540 vcpu->arch.access, 0, access)) {
4541 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4542 (gva & (PAGE_SIZE - 1));
4543 trace_vcpu_match_mmio(gva, *gpa, write, false);
4547 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4549 if (*gpa == UNMAPPED_GVA)
4552 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4555 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4556 const void *val, int bytes)
4560 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4563 kvm_page_track_write(vcpu, gpa, val, bytes);
4567 struct read_write_emulator_ops {
4568 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4570 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4571 void *val, int bytes);
4572 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4573 int bytes, void *val);
4574 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4575 void *val, int bytes);
4579 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4581 if (vcpu->mmio_read_completed) {
4582 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4583 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4584 vcpu->mmio_read_completed = 0;
4591 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4592 void *val, int bytes)
4594 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4597 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4598 void *val, int bytes)
4600 return emulator_write_phys(vcpu, gpa, val, bytes);
4603 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4605 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4606 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4609 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4610 void *val, int bytes)
4612 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4613 return X86EMUL_IO_NEEDED;
4616 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4617 void *val, int bytes)
4619 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4621 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4622 return X86EMUL_CONTINUE;
4625 static const struct read_write_emulator_ops read_emultor = {
4626 .read_write_prepare = read_prepare,
4627 .read_write_emulate = read_emulate,
4628 .read_write_mmio = vcpu_mmio_read,
4629 .read_write_exit_mmio = read_exit_mmio,
4632 static const struct read_write_emulator_ops write_emultor = {
4633 .read_write_emulate = write_emulate,
4634 .read_write_mmio = write_mmio,
4635 .read_write_exit_mmio = write_exit_mmio,
4639 static int emulator_read_write_onepage(unsigned long addr, void *val,
4641 struct x86_exception *exception,
4642 struct kvm_vcpu *vcpu,
4643 const struct read_write_emulator_ops *ops)
4647 bool write = ops->write;
4648 struct kvm_mmio_fragment *frag;
4649 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4652 * If the exit was due to a NPF we may already have a GPA.
4653 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4654 * Note, this cannot be used on string operations since string
4655 * operation using rep will only have the initial GPA from the NPF
4658 if (vcpu->arch.gpa_available &&
4659 emulator_can_use_gpa(ctxt) &&
4660 vcpu_is_mmio_gpa(vcpu, addr, exception->address, write) &&
4661 (addr & ~PAGE_MASK) == (exception->address & ~PAGE_MASK)) {
4662 gpa = exception->address;
4666 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4669 return X86EMUL_PROPAGATE_FAULT;
4671 /* For APIC access vmexit */
4675 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4676 return X86EMUL_CONTINUE;
4680 * Is this MMIO handled locally?
4682 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4683 if (handled == bytes)
4684 return X86EMUL_CONTINUE;
4690 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4691 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4695 return X86EMUL_CONTINUE;
4698 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4700 void *val, unsigned int bytes,
4701 struct x86_exception *exception,
4702 const struct read_write_emulator_ops *ops)
4704 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4708 if (ops->read_write_prepare &&
4709 ops->read_write_prepare(vcpu, val, bytes))
4710 return X86EMUL_CONTINUE;
4712 vcpu->mmio_nr_fragments = 0;
4714 /* Crossing a page boundary? */
4715 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4718 now = -addr & ~PAGE_MASK;
4719 rc = emulator_read_write_onepage(addr, val, now, exception,
4722 if (rc != X86EMUL_CONTINUE)
4725 if (ctxt->mode != X86EMUL_MODE_PROT64)
4731 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4733 if (rc != X86EMUL_CONTINUE)
4736 if (!vcpu->mmio_nr_fragments)
4739 gpa = vcpu->mmio_fragments[0].gpa;
4741 vcpu->mmio_needed = 1;
4742 vcpu->mmio_cur_fragment = 0;
4744 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4745 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4746 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4747 vcpu->run->mmio.phys_addr = gpa;
4749 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4752 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4756 struct x86_exception *exception)
4758 return emulator_read_write(ctxt, addr, val, bytes,
4759 exception, &read_emultor);
4762 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4766 struct x86_exception *exception)
4768 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4769 exception, &write_emultor);
4772 #define CMPXCHG_TYPE(t, ptr, old, new) \
4773 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4775 #ifdef CONFIG_X86_64
4776 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4778 # define CMPXCHG64(ptr, old, new) \
4779 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4782 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4787 struct x86_exception *exception)
4789 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4795 /* guests cmpxchg8b have to be emulated atomically */
4796 if (bytes > 8 || (bytes & (bytes - 1)))
4799 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4801 if (gpa == UNMAPPED_GVA ||
4802 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4805 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4808 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4809 if (is_error_page(page))
4812 kaddr = kmap_atomic(page);
4813 kaddr += offset_in_page(gpa);
4816 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4819 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4822 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4825 exchanged = CMPXCHG64(kaddr, old, new);
4830 kunmap_atomic(kaddr);
4831 kvm_release_page_dirty(page);
4834 return X86EMUL_CMPXCHG_FAILED;
4836 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4837 kvm_page_track_write(vcpu, gpa, new, bytes);
4839 return X86EMUL_CONTINUE;
4842 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4844 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4847 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4851 for (i = 0; i < vcpu->arch.pio.count; i++) {
4852 if (vcpu->arch.pio.in)
4853 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4854 vcpu->arch.pio.size, pd);
4856 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4857 vcpu->arch.pio.port, vcpu->arch.pio.size,
4861 pd += vcpu->arch.pio.size;
4866 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4867 unsigned short port, void *val,
4868 unsigned int count, bool in)
4870 vcpu->arch.pio.port = port;
4871 vcpu->arch.pio.in = in;
4872 vcpu->arch.pio.count = count;
4873 vcpu->arch.pio.size = size;
4875 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4876 vcpu->arch.pio.count = 0;
4880 vcpu->run->exit_reason = KVM_EXIT_IO;
4881 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4882 vcpu->run->io.size = size;
4883 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4884 vcpu->run->io.count = count;
4885 vcpu->run->io.port = port;
4890 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4891 int size, unsigned short port, void *val,
4894 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4897 if (vcpu->arch.pio.count)
4900 memset(vcpu->arch.pio_data, 0, size * count);
4902 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4905 memcpy(val, vcpu->arch.pio_data, size * count);
4906 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4907 vcpu->arch.pio.count = 0;
4914 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4915 int size, unsigned short port,
4916 const void *val, unsigned int count)
4918 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4920 memcpy(vcpu->arch.pio_data, val, size * count);
4921 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4922 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4925 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4927 return kvm_x86_ops->get_segment_base(vcpu, seg);
4930 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4932 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4935 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4937 if (!need_emulate_wbinvd(vcpu))
4938 return X86EMUL_CONTINUE;
4940 if (kvm_x86_ops->has_wbinvd_exit()) {
4941 int cpu = get_cpu();
4943 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4944 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4945 wbinvd_ipi, NULL, 1);
4947 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4950 return X86EMUL_CONTINUE;
4953 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4955 kvm_emulate_wbinvd_noskip(vcpu);
4956 return kvm_skip_emulated_instruction(vcpu);
4958 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4962 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4964 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4967 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4968 unsigned long *dest)
4970 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4973 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4974 unsigned long value)
4977 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4980 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4982 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4985 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4987 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4988 unsigned long value;
4992 value = kvm_read_cr0(vcpu);
4995 value = vcpu->arch.cr2;
4998 value = kvm_read_cr3(vcpu);
5001 value = kvm_read_cr4(vcpu);
5004 value = kvm_get_cr8(vcpu);
5007 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5014 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5016 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5021 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5024 vcpu->arch.cr2 = val;
5027 res = kvm_set_cr3(vcpu, val);
5030 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5033 res = kvm_set_cr8(vcpu, val);
5036 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5043 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5045 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5048 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5050 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5053 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5055 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5058 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5060 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5063 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5065 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5068 static unsigned long emulator_get_cached_segment_base(
5069 struct x86_emulate_ctxt *ctxt, int seg)
5071 return get_segment_base(emul_to_vcpu(ctxt), seg);
5074 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5075 struct desc_struct *desc, u32 *base3,
5078 struct kvm_segment var;
5080 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5081 *selector = var.selector;
5084 memset(desc, 0, sizeof(*desc));
5092 set_desc_limit(desc, var.limit);
5093 set_desc_base(desc, (unsigned long)var.base);
5094 #ifdef CONFIG_X86_64
5096 *base3 = var.base >> 32;
5098 desc->type = var.type;
5100 desc->dpl = var.dpl;
5101 desc->p = var.present;
5102 desc->avl = var.avl;
5110 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5111 struct desc_struct *desc, u32 base3,
5114 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5115 struct kvm_segment var;
5117 var.selector = selector;
5118 var.base = get_desc_base(desc);
5119 #ifdef CONFIG_X86_64
5120 var.base |= ((u64)base3) << 32;
5122 var.limit = get_desc_limit(desc);
5124 var.limit = (var.limit << 12) | 0xfff;
5125 var.type = desc->type;
5126 var.dpl = desc->dpl;
5131 var.avl = desc->avl;
5132 var.present = desc->p;
5133 var.unusable = !var.present;
5136 kvm_set_segment(vcpu, &var, seg);
5140 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5141 u32 msr_index, u64 *pdata)
5143 struct msr_data msr;
5146 msr.index = msr_index;
5147 msr.host_initiated = false;
5148 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5156 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5157 u32 msr_index, u64 data)
5159 struct msr_data msr;
5162 msr.index = msr_index;
5163 msr.host_initiated = false;
5164 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5167 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5169 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5171 return vcpu->arch.smbase;
5174 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5176 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5178 vcpu->arch.smbase = smbase;
5181 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5184 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5187 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5188 u32 pmc, u64 *pdata)
5190 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5193 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5195 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5198 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5201 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5204 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5209 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5210 struct x86_instruction_info *info,
5211 enum x86_intercept_stage stage)
5213 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5216 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5217 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5219 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5222 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5224 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5227 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5229 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5232 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5234 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5237 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5239 return emul_to_vcpu(ctxt)->arch.hflags;
5242 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5244 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5247 static const struct x86_emulate_ops emulate_ops = {
5248 .read_gpr = emulator_read_gpr,
5249 .write_gpr = emulator_write_gpr,
5250 .read_std = kvm_read_guest_virt_system,
5251 .write_std = kvm_write_guest_virt_system,
5252 .read_phys = kvm_read_guest_phys_system,
5253 .fetch = kvm_fetch_guest_virt,
5254 .read_emulated = emulator_read_emulated,
5255 .write_emulated = emulator_write_emulated,
5256 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5257 .invlpg = emulator_invlpg,
5258 .pio_in_emulated = emulator_pio_in_emulated,
5259 .pio_out_emulated = emulator_pio_out_emulated,
5260 .get_segment = emulator_get_segment,
5261 .set_segment = emulator_set_segment,
5262 .get_cached_segment_base = emulator_get_cached_segment_base,
5263 .get_gdt = emulator_get_gdt,
5264 .get_idt = emulator_get_idt,
5265 .set_gdt = emulator_set_gdt,
5266 .set_idt = emulator_set_idt,
5267 .get_cr = emulator_get_cr,
5268 .set_cr = emulator_set_cr,
5269 .cpl = emulator_get_cpl,
5270 .get_dr = emulator_get_dr,
5271 .set_dr = emulator_set_dr,
5272 .get_smbase = emulator_get_smbase,
5273 .set_smbase = emulator_set_smbase,
5274 .set_msr = emulator_set_msr,
5275 .get_msr = emulator_get_msr,
5276 .check_pmc = emulator_check_pmc,
5277 .read_pmc = emulator_read_pmc,
5278 .halt = emulator_halt,
5279 .wbinvd = emulator_wbinvd,
5280 .fix_hypercall = emulator_fix_hypercall,
5281 .get_fpu = emulator_get_fpu,
5282 .put_fpu = emulator_put_fpu,
5283 .intercept = emulator_intercept,
5284 .get_cpuid = emulator_get_cpuid,
5285 .set_nmi_mask = emulator_set_nmi_mask,
5286 .get_hflags = emulator_get_hflags,
5287 .set_hflags = emulator_set_hflags,
5290 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5292 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5294 * an sti; sti; sequence only disable interrupts for the first
5295 * instruction. So, if the last instruction, be it emulated or
5296 * not, left the system with the INT_STI flag enabled, it
5297 * means that the last instruction is an sti. We should not
5298 * leave the flag on in this case. The same goes for mov ss
5300 if (int_shadow & mask)
5302 if (unlikely(int_shadow || mask)) {
5303 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5305 kvm_make_request(KVM_REQ_EVENT, vcpu);
5309 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5311 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5312 if (ctxt->exception.vector == PF_VECTOR)
5313 return kvm_propagate_fault(vcpu, &ctxt->exception);
5315 if (ctxt->exception.error_code_valid)
5316 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5317 ctxt->exception.error_code);
5319 kvm_queue_exception(vcpu, ctxt->exception.vector);
5323 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5325 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5328 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5330 ctxt->eflags = kvm_get_rflags(vcpu);
5331 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5333 ctxt->eip = kvm_rip_read(vcpu);
5334 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5335 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5336 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5337 cs_db ? X86EMUL_MODE_PROT32 :
5338 X86EMUL_MODE_PROT16;
5339 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5340 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5341 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5343 init_decode_cache(ctxt);
5344 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5347 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5349 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5352 init_emulate_ctxt(vcpu);
5356 ctxt->_eip = ctxt->eip + inc_eip;
5357 ret = emulate_int_real(ctxt, irq);
5359 if (ret != X86EMUL_CONTINUE)
5360 return EMULATE_FAIL;
5362 ctxt->eip = ctxt->_eip;
5363 kvm_rip_write(vcpu, ctxt->eip);
5364 kvm_set_rflags(vcpu, ctxt->eflags);
5366 if (irq == NMI_VECTOR)
5367 vcpu->arch.nmi_pending = 0;
5369 vcpu->arch.interrupt.pending = false;
5371 return EMULATE_DONE;
5373 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5375 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5377 int r = EMULATE_DONE;
5379 ++vcpu->stat.insn_emulation_fail;
5380 trace_kvm_emulate_insn_failed(vcpu);
5381 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5382 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5383 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5384 vcpu->run->internal.ndata = 0;
5387 kvm_queue_exception(vcpu, UD_VECTOR);
5392 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5393 bool write_fault_to_shadow_pgtable,
5399 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5402 if (!vcpu->arch.mmu.direct_map) {
5404 * Write permission should be allowed since only
5405 * write access need to be emulated.
5407 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5410 * If the mapping is invalid in guest, let cpu retry
5411 * it to generate fault.
5413 if (gpa == UNMAPPED_GVA)
5418 * Do not retry the unhandleable instruction if it faults on the
5419 * readonly host memory, otherwise it will goto a infinite loop:
5420 * retry instruction -> write #PF -> emulation fail -> retry
5421 * instruction -> ...
5423 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5426 * If the instruction failed on the error pfn, it can not be fixed,
5427 * report the error to userspace.
5429 if (is_error_noslot_pfn(pfn))
5432 kvm_release_pfn_clean(pfn);
5434 /* The instructions are well-emulated on direct mmu. */
5435 if (vcpu->arch.mmu.direct_map) {
5436 unsigned int indirect_shadow_pages;
5438 spin_lock(&vcpu->kvm->mmu_lock);
5439 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5440 spin_unlock(&vcpu->kvm->mmu_lock);
5442 if (indirect_shadow_pages)
5443 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5449 * if emulation was due to access to shadowed page table
5450 * and it failed try to unshadow page and re-enter the
5451 * guest to let CPU execute the instruction.
5453 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5456 * If the access faults on its page table, it can not
5457 * be fixed by unprotecting shadow page and it should
5458 * be reported to userspace.
5460 return !write_fault_to_shadow_pgtable;
5463 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5464 unsigned long cr2, int emulation_type)
5466 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5467 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5469 last_retry_eip = vcpu->arch.last_retry_eip;
5470 last_retry_addr = vcpu->arch.last_retry_addr;
5473 * If the emulation is caused by #PF and it is non-page_table
5474 * writing instruction, it means the VM-EXIT is caused by shadow
5475 * page protected, we can zap the shadow page and retry this
5476 * instruction directly.
5478 * Note: if the guest uses a non-page-table modifying instruction
5479 * on the PDE that points to the instruction, then we will unmap
5480 * the instruction and go to an infinite loop. So, we cache the
5481 * last retried eip and the last fault address, if we meet the eip
5482 * and the address again, we can break out of the potential infinite
5485 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5487 if (!(emulation_type & EMULTYPE_RETRY))
5490 if (x86_page_table_writing_insn(ctxt))
5493 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5496 vcpu->arch.last_retry_eip = ctxt->eip;
5497 vcpu->arch.last_retry_addr = cr2;
5499 if (!vcpu->arch.mmu.direct_map)
5500 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5502 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5507 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5508 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5510 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5512 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5513 /* This is a good place to trace that we are exiting SMM. */
5514 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5516 /* Process a latched INIT or SMI, if any. */
5517 kvm_make_request(KVM_REQ_EVENT, vcpu);
5520 kvm_mmu_reset_context(vcpu);
5523 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5525 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5527 vcpu->arch.hflags = emul_flags;
5529 if (changed & HF_SMM_MASK)
5530 kvm_smm_changed(vcpu);
5533 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5542 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5543 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5548 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
5550 struct kvm_run *kvm_run = vcpu->run;
5552 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5553 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5554 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5555 kvm_run->debug.arch.exception = DB_VECTOR;
5556 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5557 *r = EMULATE_USER_EXIT;
5560 * "Certain debug exceptions may clear bit 0-3. The
5561 * remaining contents of the DR6 register are never
5562 * cleared by the processor".
5564 vcpu->arch.dr6 &= ~15;
5565 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5566 kvm_queue_exception(vcpu, DB_VECTOR);
5570 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5572 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5573 int r = EMULATE_DONE;
5575 kvm_x86_ops->skip_emulated_instruction(vcpu);
5578 * rflags is the old, "raw" value of the flags. The new value has
5579 * not been saved yet.
5581 * This is correct even for TF set by the guest, because "the
5582 * processor will not generate this exception after the instruction
5583 * that sets the TF flag".
5585 if (unlikely(rflags & X86_EFLAGS_TF))
5586 kvm_vcpu_do_singlestep(vcpu, &r);
5587 return r == EMULATE_DONE;
5589 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5591 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5593 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5594 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5595 struct kvm_run *kvm_run = vcpu->run;
5596 unsigned long eip = kvm_get_linear_rip(vcpu);
5597 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5598 vcpu->arch.guest_debug_dr7,
5602 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5603 kvm_run->debug.arch.pc = eip;
5604 kvm_run->debug.arch.exception = DB_VECTOR;
5605 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5606 *r = EMULATE_USER_EXIT;
5611 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5612 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5613 unsigned long eip = kvm_get_linear_rip(vcpu);
5614 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5619 vcpu->arch.dr6 &= ~15;
5620 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5621 kvm_queue_exception(vcpu, DB_VECTOR);
5630 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5637 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5638 bool writeback = true;
5639 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5642 * Clear write_fault_to_shadow_pgtable here to ensure it is
5645 vcpu->arch.write_fault_to_shadow_pgtable = false;
5646 kvm_clear_exception_queue(vcpu);
5648 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5649 init_emulate_ctxt(vcpu);
5652 * We will reenter on the same instruction since
5653 * we do not set complete_userspace_io. This does not
5654 * handle watchpoints yet, those would be handled in
5657 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5660 ctxt->interruptibility = 0;
5661 ctxt->have_exception = false;
5662 ctxt->exception.vector = -1;
5663 ctxt->perm_ok = false;
5665 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5667 r = x86_decode_insn(ctxt, insn, insn_len);
5669 trace_kvm_emulate_insn_start(vcpu);
5670 ++vcpu->stat.insn_emulation;
5671 if (r != EMULATION_OK) {
5672 if (emulation_type & EMULTYPE_TRAP_UD)
5673 return EMULATE_FAIL;
5674 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5676 return EMULATE_DONE;
5677 if (emulation_type & EMULTYPE_SKIP)
5678 return EMULATE_FAIL;
5679 return handle_emulation_failure(vcpu);
5683 if (emulation_type & EMULTYPE_SKIP) {
5684 kvm_rip_write(vcpu, ctxt->_eip);
5685 if (ctxt->eflags & X86_EFLAGS_RF)
5686 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5687 return EMULATE_DONE;
5690 if (retry_instruction(ctxt, cr2, emulation_type))
5691 return EMULATE_DONE;
5693 /* this is needed for vmware backdoor interface to work since it
5694 changes registers values during IO operation */
5695 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5696 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5697 emulator_invalidate_register_cache(ctxt);
5701 /* Save the faulting GPA (cr2) in the address field */
5702 ctxt->exception.address = cr2;
5704 r = x86_emulate_insn(ctxt);
5706 if (r == EMULATION_INTERCEPTED)
5707 return EMULATE_DONE;
5709 if (r == EMULATION_FAILED) {
5710 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5712 return EMULATE_DONE;
5714 return handle_emulation_failure(vcpu);
5717 if (ctxt->have_exception) {
5719 if (inject_emulated_exception(vcpu))
5721 } else if (vcpu->arch.pio.count) {
5722 if (!vcpu->arch.pio.in) {
5723 /* FIXME: return into emulator if single-stepping. */
5724 vcpu->arch.pio.count = 0;
5727 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5729 r = EMULATE_USER_EXIT;
5730 } else if (vcpu->mmio_needed) {
5731 if (!vcpu->mmio_is_write)
5733 r = EMULATE_USER_EXIT;
5734 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5735 } else if (r == EMULATION_RESTART)
5741 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5742 toggle_interruptibility(vcpu, ctxt->interruptibility);
5743 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5744 kvm_rip_write(vcpu, ctxt->eip);
5745 if (r == EMULATE_DONE &&
5746 (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
5747 kvm_vcpu_do_singlestep(vcpu, &r);
5748 if (!ctxt->have_exception ||
5749 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5750 __kvm_set_rflags(vcpu, ctxt->eflags);
5753 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5754 * do nothing, and it will be requested again as soon as
5755 * the shadow expires. But we still need to check here,
5756 * because POPF has no interrupt shadow.
5758 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5759 kvm_make_request(KVM_REQ_EVENT, vcpu);
5761 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5765 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5767 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5769 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5770 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5771 size, port, &val, 1);
5772 /* do not return to emulator after return from userspace */
5773 vcpu->arch.pio.count = 0;
5776 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5778 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5782 /* We should only ever be called with arch.pio.count equal to 1 */
5783 BUG_ON(vcpu->arch.pio.count != 1);
5785 /* For size less than 4 we merge, else we zero extend */
5786 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5790 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5791 * the copy and tracing
5793 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5794 vcpu->arch.pio.port, &val, 1);
5795 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5800 int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5805 /* For size less than 4 we merge, else we zero extend */
5806 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5808 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5811 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5815 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5819 EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5821 static int kvmclock_cpu_down_prep(unsigned int cpu)
5823 __this_cpu_write(cpu_tsc_khz, 0);
5827 static void tsc_khz_changed(void *data)
5829 struct cpufreq_freqs *freq = data;
5830 unsigned long khz = 0;
5834 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5835 khz = cpufreq_quick_get(raw_smp_processor_id());
5838 __this_cpu_write(cpu_tsc_khz, khz);
5841 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5844 struct cpufreq_freqs *freq = data;
5846 struct kvm_vcpu *vcpu;
5847 int i, send_ipi = 0;
5850 * We allow guests to temporarily run on slowing clocks,
5851 * provided we notify them after, or to run on accelerating
5852 * clocks, provided we notify them before. Thus time never
5855 * However, we have a problem. We can't atomically update
5856 * the frequency of a given CPU from this function; it is
5857 * merely a notifier, which can be called from any CPU.
5858 * Changing the TSC frequency at arbitrary points in time
5859 * requires a recomputation of local variables related to
5860 * the TSC for each VCPU. We must flag these local variables
5861 * to be updated and be sure the update takes place with the
5862 * new frequency before any guests proceed.
5864 * Unfortunately, the combination of hotplug CPU and frequency
5865 * change creates an intractable locking scenario; the order
5866 * of when these callouts happen is undefined with respect to
5867 * CPU hotplug, and they can race with each other. As such,
5868 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5869 * undefined; you can actually have a CPU frequency change take
5870 * place in between the computation of X and the setting of the
5871 * variable. To protect against this problem, all updates of
5872 * the per_cpu tsc_khz variable are done in an interrupt
5873 * protected IPI, and all callers wishing to update the value
5874 * must wait for a synchronous IPI to complete (which is trivial
5875 * if the caller is on the CPU already). This establishes the
5876 * necessary total order on variable updates.
5878 * Note that because a guest time update may take place
5879 * anytime after the setting of the VCPU's request bit, the
5880 * correct TSC value must be set before the request. However,
5881 * to ensure the update actually makes it to any guest which
5882 * starts running in hardware virtualization between the set
5883 * and the acquisition of the spinlock, we must also ping the
5884 * CPU after setting the request bit.
5888 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5890 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5893 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5895 spin_lock(&kvm_lock);
5896 list_for_each_entry(kvm, &vm_list, vm_list) {
5897 kvm_for_each_vcpu(i, vcpu, kvm) {
5898 if (vcpu->cpu != freq->cpu)
5900 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5901 if (vcpu->cpu != smp_processor_id())
5905 spin_unlock(&kvm_lock);
5907 if (freq->old < freq->new && send_ipi) {
5909 * We upscale the frequency. Must make the guest
5910 * doesn't see old kvmclock values while running with
5911 * the new frequency, otherwise we risk the guest sees
5912 * time go backwards.
5914 * In case we update the frequency for another cpu
5915 * (which might be in guest context) send an interrupt
5916 * to kick the cpu out of guest context. Next time
5917 * guest context is entered kvmclock will be updated,
5918 * so the guest will not see stale values.
5920 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5925 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5926 .notifier_call = kvmclock_cpufreq_notifier
5929 static int kvmclock_cpu_online(unsigned int cpu)
5931 tsc_khz_changed(NULL);
5935 static void kvm_timer_init(void)
5937 max_tsc_khz = tsc_khz;
5939 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5940 #ifdef CONFIG_CPU_FREQ
5941 struct cpufreq_policy policy;
5944 memset(&policy, 0, sizeof(policy));
5946 cpufreq_get_policy(&policy, cpu);
5947 if (policy.cpuinfo.max_freq)
5948 max_tsc_khz = policy.cpuinfo.max_freq;
5951 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5952 CPUFREQ_TRANSITION_NOTIFIER);
5954 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5956 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
5957 kvmclock_cpu_online, kvmclock_cpu_down_prep);
5960 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5962 int kvm_is_in_guest(void)
5964 return __this_cpu_read(current_vcpu) != NULL;
5967 static int kvm_is_user_mode(void)
5971 if (__this_cpu_read(current_vcpu))
5972 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5974 return user_mode != 0;
5977 static unsigned long kvm_get_guest_ip(void)
5979 unsigned long ip = 0;
5981 if (__this_cpu_read(current_vcpu))
5982 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5987 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5988 .is_in_guest = kvm_is_in_guest,
5989 .is_user_mode = kvm_is_user_mode,
5990 .get_guest_ip = kvm_get_guest_ip,
5993 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5995 __this_cpu_write(current_vcpu, vcpu);
5997 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5999 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
6001 __this_cpu_write(current_vcpu, NULL);
6003 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
6005 static void kvm_set_mmio_spte_mask(void)
6008 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6011 * Set the reserved bits and the present bit of an paging-structure
6012 * entry to generate page fault with PFER.RSV = 1.
6014 /* Mask the reserved physical address bits. */
6015 mask = rsvd_bits(maxphyaddr, 51);
6017 /* Set the present bit. */
6020 #ifdef CONFIG_X86_64
6022 * If reserved bit is not supported, clear the present bit to disable
6025 if (maxphyaddr == 52)
6029 kvm_mmu_set_mmio_spte_mask(mask, mask);
6032 #ifdef CONFIG_X86_64
6033 static void pvclock_gtod_update_fn(struct work_struct *work)
6037 struct kvm_vcpu *vcpu;
6040 spin_lock(&kvm_lock);
6041 list_for_each_entry(kvm, &vm_list, vm_list)
6042 kvm_for_each_vcpu(i, vcpu, kvm)
6043 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6044 atomic_set(&kvm_guest_has_master_clock, 0);
6045 spin_unlock(&kvm_lock);
6048 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6051 * Notification about pvclock gtod data update.
6053 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6056 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6057 struct timekeeper *tk = priv;
6059 update_pvclock_gtod(tk);
6061 /* disable master clock if host does not trust, or does not
6062 * use, TSC clocksource
6064 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6065 atomic_read(&kvm_guest_has_master_clock) != 0)
6066 queue_work(system_long_wq, &pvclock_gtod_work);
6071 static struct notifier_block pvclock_gtod_notifier = {
6072 .notifier_call = pvclock_gtod_notify,
6076 int kvm_arch_init(void *opaque)
6079 struct kvm_x86_ops *ops = opaque;
6082 printk(KERN_ERR "kvm: already loaded the other module\n");
6087 if (!ops->cpu_has_kvm_support()) {
6088 printk(KERN_ERR "kvm: no hardware support\n");
6092 if (ops->disabled_by_bios()) {
6093 printk(KERN_ERR "kvm: disabled by bios\n");
6099 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6101 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6105 r = kvm_mmu_module_init();
6107 goto out_free_percpu;
6109 kvm_set_mmio_spte_mask();
6113 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6114 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6115 PT_PRESENT_MASK, 0);
6118 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6120 if (boot_cpu_has(X86_FEATURE_XSAVE))
6121 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6124 #ifdef CONFIG_X86_64
6125 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6131 free_percpu(shared_msrs);
6136 void kvm_arch_exit(void)
6139 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6141 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6142 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6143 CPUFREQ_TRANSITION_NOTIFIER);
6144 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6145 #ifdef CONFIG_X86_64
6146 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6149 kvm_mmu_module_exit();
6150 free_percpu(shared_msrs);
6153 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6155 ++vcpu->stat.halt_exits;
6156 if (lapic_in_kernel(vcpu)) {
6157 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6160 vcpu->run->exit_reason = KVM_EXIT_HLT;
6164 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6166 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6168 int ret = kvm_skip_emulated_instruction(vcpu);
6170 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6171 * KVM_EXIT_DEBUG here.
6173 return kvm_vcpu_halt(vcpu) && ret;
6175 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6177 #ifdef CONFIG_X86_64
6178 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6179 unsigned long clock_type)
6181 struct kvm_clock_pairing clock_pairing;
6186 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6187 return -KVM_EOPNOTSUPP;
6189 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6190 return -KVM_EOPNOTSUPP;
6192 clock_pairing.sec = ts.tv_sec;
6193 clock_pairing.nsec = ts.tv_nsec;
6194 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6195 clock_pairing.flags = 0;
6198 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6199 sizeof(struct kvm_clock_pairing)))
6207 * kvm_pv_kick_cpu_op: Kick a vcpu.
6209 * @apicid - apicid of vcpu to be kicked.
6211 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6213 struct kvm_lapic_irq lapic_irq;
6215 lapic_irq.shorthand = 0;
6216 lapic_irq.dest_mode = 0;
6217 lapic_irq.level = 0;
6218 lapic_irq.dest_id = apicid;
6219 lapic_irq.msi_redir_hint = false;
6221 lapic_irq.delivery_mode = APIC_DM_REMRD;
6222 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6225 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6227 vcpu->arch.apicv_active = false;
6228 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6231 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6233 unsigned long nr, a0, a1, a2, a3, ret;
6236 r = kvm_skip_emulated_instruction(vcpu);
6238 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6239 return kvm_hv_hypercall(vcpu);
6241 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6242 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6243 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6244 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6245 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6247 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6249 op_64_bit = is_64_bit_mode(vcpu);
6258 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6264 case KVM_HC_VAPIC_POLL_IRQ:
6267 case KVM_HC_KICK_CPU:
6268 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6271 #ifdef CONFIG_X86_64
6272 case KVM_HC_CLOCK_PAIRING:
6273 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6283 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6284 ++vcpu->stat.hypercalls;
6287 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6289 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6291 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6292 char instruction[3];
6293 unsigned long rip = kvm_rip_read(vcpu);
6295 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6297 return emulator_write_emulated(ctxt, rip, instruction, 3,
6301 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6303 return vcpu->run->request_interrupt_window &&
6304 likely(!pic_in_kernel(vcpu->kvm));
6307 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6309 struct kvm_run *kvm_run = vcpu->run;
6311 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6312 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6313 kvm_run->cr8 = kvm_get_cr8(vcpu);
6314 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6315 kvm_run->ready_for_interrupt_injection =
6316 pic_in_kernel(vcpu->kvm) ||
6317 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6320 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6324 if (!kvm_x86_ops->update_cr8_intercept)
6327 if (!lapic_in_kernel(vcpu))
6330 if (vcpu->arch.apicv_active)
6333 if (!vcpu->arch.apic->vapic_addr)
6334 max_irr = kvm_lapic_find_highest_irr(vcpu);
6341 tpr = kvm_lapic_get_cr8(vcpu);
6343 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6346 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6350 /* try to reinject previous events if any */
6351 if (vcpu->arch.exception.pending) {
6352 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6353 vcpu->arch.exception.has_error_code,
6354 vcpu->arch.exception.error_code);
6356 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6357 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6360 if (vcpu->arch.exception.nr == DB_VECTOR &&
6361 (vcpu->arch.dr7 & DR7_GD)) {
6362 vcpu->arch.dr7 &= ~DR7_GD;
6363 kvm_update_dr7(vcpu);
6366 kvm_x86_ops->queue_exception(vcpu);
6370 if (vcpu->arch.nmi_injected) {
6371 kvm_x86_ops->set_nmi(vcpu);
6375 if (vcpu->arch.interrupt.pending) {
6376 kvm_x86_ops->set_irq(vcpu);
6380 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6381 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6386 /* try to inject new event if pending */
6387 if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6388 vcpu->arch.smi_pending = false;
6390 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6391 --vcpu->arch.nmi_pending;
6392 vcpu->arch.nmi_injected = true;
6393 kvm_x86_ops->set_nmi(vcpu);
6394 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6396 * Because interrupts can be injected asynchronously, we are
6397 * calling check_nested_events again here to avoid a race condition.
6398 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6399 * proposal and current concerns. Perhaps we should be setting
6400 * KVM_REQ_EVENT only on certain events and not unconditionally?
6402 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6403 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6407 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6408 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6410 kvm_x86_ops->set_irq(vcpu);
6417 static void process_nmi(struct kvm_vcpu *vcpu)
6422 * x86 is limited to one NMI running, and one NMI pending after it.
6423 * If an NMI is already in progress, limit further NMIs to just one.
6424 * Otherwise, allow two (and we'll inject the first one immediately).
6426 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6429 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6430 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6431 kvm_make_request(KVM_REQ_EVENT, vcpu);
6434 #define put_smstate(type, buf, offset, val) \
6435 *(type *)((buf) + (offset) - 0x7e00) = val
6437 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6440 flags |= seg->g << 23;
6441 flags |= seg->db << 22;
6442 flags |= seg->l << 21;
6443 flags |= seg->avl << 20;
6444 flags |= seg->present << 15;
6445 flags |= seg->dpl << 13;
6446 flags |= seg->s << 12;
6447 flags |= seg->type << 8;
6451 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6453 struct kvm_segment seg;
6456 kvm_get_segment(vcpu, &seg, n);
6457 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6460 offset = 0x7f84 + n * 12;
6462 offset = 0x7f2c + (n - 3) * 12;
6464 put_smstate(u32, buf, offset + 8, seg.base);
6465 put_smstate(u32, buf, offset + 4, seg.limit);
6466 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6469 #ifdef CONFIG_X86_64
6470 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6472 struct kvm_segment seg;
6476 kvm_get_segment(vcpu, &seg, n);
6477 offset = 0x7e00 + n * 16;
6479 flags = enter_smm_get_segment_flags(&seg) >> 8;
6480 put_smstate(u16, buf, offset, seg.selector);
6481 put_smstate(u16, buf, offset + 2, flags);
6482 put_smstate(u32, buf, offset + 4, seg.limit);
6483 put_smstate(u64, buf, offset + 8, seg.base);
6487 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6490 struct kvm_segment seg;
6494 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6495 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6496 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6497 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6499 for (i = 0; i < 8; i++)
6500 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6502 kvm_get_dr(vcpu, 6, &val);
6503 put_smstate(u32, buf, 0x7fcc, (u32)val);
6504 kvm_get_dr(vcpu, 7, &val);
6505 put_smstate(u32, buf, 0x7fc8, (u32)val);
6507 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6508 put_smstate(u32, buf, 0x7fc4, seg.selector);
6509 put_smstate(u32, buf, 0x7f64, seg.base);
6510 put_smstate(u32, buf, 0x7f60, seg.limit);
6511 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6513 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6514 put_smstate(u32, buf, 0x7fc0, seg.selector);
6515 put_smstate(u32, buf, 0x7f80, seg.base);
6516 put_smstate(u32, buf, 0x7f7c, seg.limit);
6517 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6519 kvm_x86_ops->get_gdt(vcpu, &dt);
6520 put_smstate(u32, buf, 0x7f74, dt.address);
6521 put_smstate(u32, buf, 0x7f70, dt.size);
6523 kvm_x86_ops->get_idt(vcpu, &dt);
6524 put_smstate(u32, buf, 0x7f58, dt.address);
6525 put_smstate(u32, buf, 0x7f54, dt.size);
6527 for (i = 0; i < 6; i++)
6528 enter_smm_save_seg_32(vcpu, buf, i);
6530 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6533 put_smstate(u32, buf, 0x7efc, 0x00020000);
6534 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6537 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6539 #ifdef CONFIG_X86_64
6541 struct kvm_segment seg;
6545 for (i = 0; i < 16; i++)
6546 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6548 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6549 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6551 kvm_get_dr(vcpu, 6, &val);
6552 put_smstate(u64, buf, 0x7f68, val);
6553 kvm_get_dr(vcpu, 7, &val);
6554 put_smstate(u64, buf, 0x7f60, val);
6556 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6557 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6558 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6560 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6563 put_smstate(u32, buf, 0x7efc, 0x00020064);
6565 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6567 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6568 put_smstate(u16, buf, 0x7e90, seg.selector);
6569 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6570 put_smstate(u32, buf, 0x7e94, seg.limit);
6571 put_smstate(u64, buf, 0x7e98, seg.base);
6573 kvm_x86_ops->get_idt(vcpu, &dt);
6574 put_smstate(u32, buf, 0x7e84, dt.size);
6575 put_smstate(u64, buf, 0x7e88, dt.address);
6577 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6578 put_smstate(u16, buf, 0x7e70, seg.selector);
6579 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6580 put_smstate(u32, buf, 0x7e74, seg.limit);
6581 put_smstate(u64, buf, 0x7e78, seg.base);
6583 kvm_x86_ops->get_gdt(vcpu, &dt);
6584 put_smstate(u32, buf, 0x7e64, dt.size);
6585 put_smstate(u64, buf, 0x7e68, dt.address);
6587 for (i = 0; i < 6; i++)
6588 enter_smm_save_seg_64(vcpu, buf, i);
6594 static void enter_smm(struct kvm_vcpu *vcpu)
6596 struct kvm_segment cs, ds;
6601 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6602 vcpu->arch.hflags |= HF_SMM_MASK;
6603 memset(buf, 0, 512);
6604 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
6605 enter_smm_save_state_64(vcpu, buf);
6607 enter_smm_save_state_32(vcpu, buf);
6609 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6611 if (kvm_x86_ops->get_nmi_mask(vcpu))
6612 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6614 kvm_x86_ops->set_nmi_mask(vcpu, true);
6616 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6617 kvm_rip_write(vcpu, 0x8000);
6619 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6620 kvm_x86_ops->set_cr0(vcpu, cr0);
6621 vcpu->arch.cr0 = cr0;
6623 kvm_x86_ops->set_cr4(vcpu, 0);
6625 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6626 dt.address = dt.size = 0;
6627 kvm_x86_ops->set_idt(vcpu, &dt);
6629 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6631 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6632 cs.base = vcpu->arch.smbase;
6637 cs.limit = ds.limit = 0xffffffff;
6638 cs.type = ds.type = 0x3;
6639 cs.dpl = ds.dpl = 0;
6644 cs.avl = ds.avl = 0;
6645 cs.present = ds.present = 1;
6646 cs.unusable = ds.unusable = 0;
6647 cs.padding = ds.padding = 0;
6649 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6650 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6651 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6652 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6653 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6654 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6656 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
6657 kvm_x86_ops->set_efer(vcpu, 0);
6659 kvm_update_cpuid(vcpu);
6660 kvm_mmu_reset_context(vcpu);
6663 static void process_smi(struct kvm_vcpu *vcpu)
6665 vcpu->arch.smi_pending = true;
6666 kvm_make_request(KVM_REQ_EVENT, vcpu);
6669 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6671 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6674 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6676 u64 eoi_exit_bitmap[4];
6678 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6681 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6683 if (irqchip_split(vcpu->kvm))
6684 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6686 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6687 kvm_x86_ops->sync_pir_to_irr(vcpu);
6688 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6690 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6691 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6692 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6695 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6697 ++vcpu->stat.tlb_flush;
6698 kvm_x86_ops->tlb_flush(vcpu);
6701 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6703 struct page *page = NULL;
6705 if (!lapic_in_kernel(vcpu))
6708 if (!kvm_x86_ops->set_apic_access_page_addr)
6711 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6712 if (is_error_page(page))
6714 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6717 * Do not pin apic access page in memory, the MMU notifier
6718 * will call us again if it is migrated or swapped out.
6722 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6724 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6725 unsigned long address)
6728 * The physical address of apic access page is stored in the VMCS.
6729 * Update it when it becomes invalid.
6731 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6732 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6736 * Returns 1 to let vcpu_run() continue the guest execution loop without
6737 * exiting to the userspace. Otherwise, the value will be returned to the
6740 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6744 dm_request_for_irq_injection(vcpu) &&
6745 kvm_cpu_accept_dm_intr(vcpu);
6747 bool req_immediate_exit = false;
6749 if (kvm_request_pending(vcpu)) {
6750 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6751 kvm_mmu_unload(vcpu);
6752 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6753 __kvm_migrate_timers(vcpu);
6754 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6755 kvm_gen_update_masterclock(vcpu->kvm);
6756 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6757 kvm_gen_kvmclock_update(vcpu);
6758 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6759 r = kvm_guest_time_update(vcpu);
6763 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6764 kvm_mmu_sync_roots(vcpu);
6765 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6766 kvm_vcpu_flush_tlb(vcpu);
6767 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6768 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6772 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6773 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6777 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6778 /* Page is swapped out. Do synthetic halt */
6779 vcpu->arch.apf.halted = true;
6783 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6784 record_steal_time(vcpu);
6785 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6787 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6789 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6790 kvm_pmu_handle_event(vcpu);
6791 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6792 kvm_pmu_deliver_pmi(vcpu);
6793 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6794 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6795 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6796 vcpu->arch.ioapic_handled_vectors)) {
6797 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6798 vcpu->run->eoi.vector =
6799 vcpu->arch.pending_ioapic_eoi;
6804 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6805 vcpu_scan_ioapic(vcpu);
6806 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6807 kvm_vcpu_reload_apic_access_page(vcpu);
6808 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6809 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6810 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6814 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6815 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6816 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6820 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6821 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6822 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6828 * KVM_REQ_HV_STIMER has to be processed after
6829 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6830 * depend on the guest clock being up-to-date
6832 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6833 kvm_hv_process_stimers(vcpu);
6836 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6837 ++vcpu->stat.req_event;
6838 kvm_apic_accept_events(vcpu);
6839 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6844 if (inject_pending_event(vcpu, req_int_win) != 0)
6845 req_immediate_exit = true;
6847 /* Enable NMI/IRQ window open exits if needed.
6849 * SMIs have two cases: 1) they can be nested, and
6850 * then there is nothing to do here because RSM will
6851 * cause a vmexit anyway; 2) or the SMI can be pending
6852 * because inject_pending_event has completed the
6853 * injection of an IRQ or NMI from the previous vmexit,
6854 * and then we request an immediate exit to inject the SMI.
6856 if (vcpu->arch.smi_pending && !is_smm(vcpu))
6857 req_immediate_exit = true;
6858 if (vcpu->arch.nmi_pending)
6859 kvm_x86_ops->enable_nmi_window(vcpu);
6860 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6861 kvm_x86_ops->enable_irq_window(vcpu);
6864 if (kvm_lapic_enabled(vcpu)) {
6865 update_cr8_intercept(vcpu);
6866 kvm_lapic_sync_to_vapic(vcpu);
6870 r = kvm_mmu_reload(vcpu);
6872 goto cancel_injection;
6877 kvm_x86_ops->prepare_guest_switch(vcpu);
6878 kvm_load_guest_fpu(vcpu);
6881 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
6882 * IPI are then delayed after guest entry, which ensures that they
6883 * result in virtual interrupt delivery.
6885 local_irq_disable();
6886 vcpu->mode = IN_GUEST_MODE;
6888 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6891 * 1) We should set ->mode before checking ->requests. Please see
6892 * the comment in kvm_vcpu_exiting_guest_mode().
6894 * 2) For APICv, we should set ->mode before checking PIR.ON. This
6895 * pairs with the memory barrier implicit in pi_test_and_set_on
6896 * (see vmx_deliver_posted_interrupt).
6898 * 3) This also orders the write to mode from any reads to the page
6899 * tables done while the VCPU is running. Please see the comment
6900 * in kvm_flush_remote_tlbs.
6902 smp_mb__after_srcu_read_unlock();
6905 * This handles the case where a posted interrupt was
6906 * notified with kvm_vcpu_kick.
6908 if (kvm_lapic_enabled(vcpu)) {
6909 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6910 kvm_x86_ops->sync_pir_to_irr(vcpu);
6913 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
6914 || need_resched() || signal_pending(current)) {
6915 vcpu->mode = OUTSIDE_GUEST_MODE;
6919 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6921 goto cancel_injection;
6924 kvm_load_guest_xcr0(vcpu);
6926 if (req_immediate_exit) {
6927 kvm_make_request(KVM_REQ_EVENT, vcpu);
6928 smp_send_reschedule(vcpu->cpu);
6931 trace_kvm_entry(vcpu->vcpu_id);
6932 wait_lapic_expire(vcpu);
6933 guest_enter_irqoff();
6935 if (unlikely(vcpu->arch.switch_db_regs)) {
6937 set_debugreg(vcpu->arch.eff_db[0], 0);
6938 set_debugreg(vcpu->arch.eff_db[1], 1);
6939 set_debugreg(vcpu->arch.eff_db[2], 2);
6940 set_debugreg(vcpu->arch.eff_db[3], 3);
6941 set_debugreg(vcpu->arch.dr6, 6);
6942 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6945 kvm_x86_ops->run(vcpu);
6948 * Do this here before restoring debug registers on the host. And
6949 * since we do this before handling the vmexit, a DR access vmexit
6950 * can (a) read the correct value of the debug registers, (b) set
6951 * KVM_DEBUGREG_WONT_EXIT again.
6953 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6954 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6955 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6956 kvm_update_dr0123(vcpu);
6957 kvm_update_dr6(vcpu);
6958 kvm_update_dr7(vcpu);
6959 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6963 * If the guest has used debug registers, at least dr7
6964 * will be disabled while returning to the host.
6965 * If we don't have active breakpoints in the host, we don't
6966 * care about the messed up debug address registers. But if
6967 * we have some of them active, restore the old state.
6969 if (hw_breakpoint_active())
6970 hw_breakpoint_restore();
6972 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6974 vcpu->mode = OUTSIDE_GUEST_MODE;
6977 kvm_put_guest_xcr0(vcpu);
6979 kvm_x86_ops->handle_external_intr(vcpu);
6983 guest_exit_irqoff();
6988 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6991 * Profile KVM exit RIPs:
6993 if (unlikely(prof_on == KVM_PROFILING)) {
6994 unsigned long rip = kvm_rip_read(vcpu);
6995 profile_hit(KVM_PROFILING, (void *)rip);
6998 if (unlikely(vcpu->arch.tsc_always_catchup))
6999 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7001 if (vcpu->arch.apic_attention)
7002 kvm_lapic_sync_from_vapic(vcpu);
7004 r = kvm_x86_ops->handle_exit(vcpu);
7008 kvm_x86_ops->cancel_injection(vcpu);
7009 if (unlikely(vcpu->arch.apic_attention))
7010 kvm_lapic_sync_from_vapic(vcpu);
7015 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7017 if (!kvm_arch_vcpu_runnable(vcpu) &&
7018 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7019 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7020 kvm_vcpu_block(vcpu);
7021 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7023 if (kvm_x86_ops->post_block)
7024 kvm_x86_ops->post_block(vcpu);
7026 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7030 kvm_apic_accept_events(vcpu);
7031 switch(vcpu->arch.mp_state) {
7032 case KVM_MP_STATE_HALTED:
7033 vcpu->arch.pv.pv_unhalted = false;
7034 vcpu->arch.mp_state =
7035 KVM_MP_STATE_RUNNABLE;
7036 case KVM_MP_STATE_RUNNABLE:
7037 vcpu->arch.apf.halted = false;
7039 case KVM_MP_STATE_INIT_RECEIVED:
7048 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7050 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7051 kvm_x86_ops->check_nested_events(vcpu, false);
7053 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7054 !vcpu->arch.apf.halted);
7057 static int vcpu_run(struct kvm_vcpu *vcpu)
7060 struct kvm *kvm = vcpu->kvm;
7062 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7065 if (kvm_vcpu_running(vcpu)) {
7066 r = vcpu_enter_guest(vcpu);
7068 r = vcpu_block(kvm, vcpu);
7074 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7075 if (kvm_cpu_has_pending_timer(vcpu))
7076 kvm_inject_pending_timer_irqs(vcpu);
7078 if (dm_request_for_irq_injection(vcpu) &&
7079 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7081 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7082 ++vcpu->stat.request_irq_exits;
7086 kvm_check_async_pf_completion(vcpu);
7088 if (signal_pending(current)) {
7090 vcpu->run->exit_reason = KVM_EXIT_INTR;
7091 ++vcpu->stat.signal_exits;
7094 if (need_resched()) {
7095 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7097 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7101 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7106 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7109 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7110 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7111 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7112 if (r != EMULATE_DONE)
7117 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7119 BUG_ON(!vcpu->arch.pio.count);
7121 return complete_emulated_io(vcpu);
7125 * Implements the following, as a state machine:
7129 * for each mmio piece in the fragment
7137 * for each mmio piece in the fragment
7142 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7144 struct kvm_run *run = vcpu->run;
7145 struct kvm_mmio_fragment *frag;
7148 BUG_ON(!vcpu->mmio_needed);
7150 /* Complete previous fragment */
7151 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7152 len = min(8u, frag->len);
7153 if (!vcpu->mmio_is_write)
7154 memcpy(frag->data, run->mmio.data, len);
7156 if (frag->len <= 8) {
7157 /* Switch to the next fragment. */
7159 vcpu->mmio_cur_fragment++;
7161 /* Go forward to the next mmio piece. */
7167 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7168 vcpu->mmio_needed = 0;
7170 /* FIXME: return into emulator if single-stepping. */
7171 if (vcpu->mmio_is_write)
7173 vcpu->mmio_read_completed = 1;
7174 return complete_emulated_io(vcpu);
7177 run->exit_reason = KVM_EXIT_MMIO;
7178 run->mmio.phys_addr = frag->gpa;
7179 if (vcpu->mmio_is_write)
7180 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7181 run->mmio.len = min(8u, frag->len);
7182 run->mmio.is_write = vcpu->mmio_is_write;
7183 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7188 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7190 struct fpu *fpu = ¤t->thread.fpu;
7194 fpu__activate_curr(fpu);
7196 if (vcpu->sigset_active)
7197 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
7199 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7200 kvm_vcpu_block(vcpu);
7201 kvm_apic_accept_events(vcpu);
7202 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7207 /* re-sync apic's tpr */
7208 if (!lapic_in_kernel(vcpu)) {
7209 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7215 if (unlikely(vcpu->arch.complete_userspace_io)) {
7216 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7217 vcpu->arch.complete_userspace_io = NULL;
7222 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7224 if (kvm_run->immediate_exit)
7230 post_kvm_run_save(vcpu);
7231 if (vcpu->sigset_active)
7232 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
7237 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7239 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7241 * We are here if userspace calls get_regs() in the middle of
7242 * instruction emulation. Registers state needs to be copied
7243 * back from emulation context to vcpu. Userspace shouldn't do
7244 * that usually, but some bad designed PV devices (vmware
7245 * backdoor interface) need this to work
7247 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7248 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7250 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7251 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7252 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7253 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7254 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7255 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7256 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7257 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7258 #ifdef CONFIG_X86_64
7259 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7260 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7261 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7262 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7263 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7264 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7265 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7266 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7269 regs->rip = kvm_rip_read(vcpu);
7270 regs->rflags = kvm_get_rflags(vcpu);
7275 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7277 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7278 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7280 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7281 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7282 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7283 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7284 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7285 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7286 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7287 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7288 #ifdef CONFIG_X86_64
7289 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7290 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7291 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7292 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7293 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7294 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7295 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7296 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7299 kvm_rip_write(vcpu, regs->rip);
7300 kvm_set_rflags(vcpu, regs->rflags);
7302 vcpu->arch.exception.pending = false;
7304 kvm_make_request(KVM_REQ_EVENT, vcpu);
7309 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7311 struct kvm_segment cs;
7313 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7317 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7319 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7320 struct kvm_sregs *sregs)
7324 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7325 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7326 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7327 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7328 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7329 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7331 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7332 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7334 kvm_x86_ops->get_idt(vcpu, &dt);
7335 sregs->idt.limit = dt.size;
7336 sregs->idt.base = dt.address;
7337 kvm_x86_ops->get_gdt(vcpu, &dt);
7338 sregs->gdt.limit = dt.size;
7339 sregs->gdt.base = dt.address;
7341 sregs->cr0 = kvm_read_cr0(vcpu);
7342 sregs->cr2 = vcpu->arch.cr2;
7343 sregs->cr3 = kvm_read_cr3(vcpu);
7344 sregs->cr4 = kvm_read_cr4(vcpu);
7345 sregs->cr8 = kvm_get_cr8(vcpu);
7346 sregs->efer = vcpu->arch.efer;
7347 sregs->apic_base = kvm_get_apic_base(vcpu);
7349 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7351 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7352 set_bit(vcpu->arch.interrupt.nr,
7353 (unsigned long *)sregs->interrupt_bitmap);
7358 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7359 struct kvm_mp_state *mp_state)
7361 kvm_apic_accept_events(vcpu);
7362 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7363 vcpu->arch.pv.pv_unhalted)
7364 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7366 mp_state->mp_state = vcpu->arch.mp_state;
7371 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7372 struct kvm_mp_state *mp_state)
7374 if (!lapic_in_kernel(vcpu) &&
7375 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7378 /* INITs are latched while in SMM */
7379 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7380 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7381 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7384 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7385 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7386 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7388 vcpu->arch.mp_state = mp_state->mp_state;
7389 kvm_make_request(KVM_REQ_EVENT, vcpu);
7393 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7394 int reason, bool has_error_code, u32 error_code)
7396 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7399 init_emulate_ctxt(vcpu);
7401 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7402 has_error_code, error_code);
7405 return EMULATE_FAIL;
7407 kvm_rip_write(vcpu, ctxt->eip);
7408 kvm_set_rflags(vcpu, ctxt->eflags);
7409 kvm_make_request(KVM_REQ_EVENT, vcpu);
7410 return EMULATE_DONE;
7412 EXPORT_SYMBOL_GPL(kvm_task_switch);
7414 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7415 struct kvm_sregs *sregs)
7417 struct msr_data apic_base_msr;
7418 int mmu_reset_needed = 0;
7419 int pending_vec, max_bits, idx;
7422 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7423 (sregs->cr4 & X86_CR4_OSXSAVE))
7426 dt.size = sregs->idt.limit;
7427 dt.address = sregs->idt.base;
7428 kvm_x86_ops->set_idt(vcpu, &dt);
7429 dt.size = sregs->gdt.limit;
7430 dt.address = sregs->gdt.base;
7431 kvm_x86_ops->set_gdt(vcpu, &dt);
7433 vcpu->arch.cr2 = sregs->cr2;
7434 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7435 vcpu->arch.cr3 = sregs->cr3;
7436 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7438 kvm_set_cr8(vcpu, sregs->cr8);
7440 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7441 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7442 apic_base_msr.data = sregs->apic_base;
7443 apic_base_msr.host_initiated = true;
7444 kvm_set_apic_base(vcpu, &apic_base_msr);
7446 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7447 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7448 vcpu->arch.cr0 = sregs->cr0;
7450 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7451 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7452 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7453 kvm_update_cpuid(vcpu);
7455 idx = srcu_read_lock(&vcpu->kvm->srcu);
7456 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7457 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7458 mmu_reset_needed = 1;
7460 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7462 if (mmu_reset_needed)
7463 kvm_mmu_reset_context(vcpu);
7465 max_bits = KVM_NR_INTERRUPTS;
7466 pending_vec = find_first_bit(
7467 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7468 if (pending_vec < max_bits) {
7469 kvm_queue_interrupt(vcpu, pending_vec, false);
7470 pr_debug("Set back pending irq %d\n", pending_vec);
7473 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7474 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7475 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7476 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7477 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7478 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7480 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7481 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7483 update_cr8_intercept(vcpu);
7485 /* Older userspace won't unhalt the vcpu on reset. */
7486 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7487 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7489 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7491 kvm_make_request(KVM_REQ_EVENT, vcpu);
7496 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7497 struct kvm_guest_debug *dbg)
7499 unsigned long rflags;
7502 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7504 if (vcpu->arch.exception.pending)
7506 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7507 kvm_queue_exception(vcpu, DB_VECTOR);
7509 kvm_queue_exception(vcpu, BP_VECTOR);
7513 * Read rflags as long as potentially injected trace flags are still
7516 rflags = kvm_get_rflags(vcpu);
7518 vcpu->guest_debug = dbg->control;
7519 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7520 vcpu->guest_debug = 0;
7522 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7523 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7524 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7525 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7527 for (i = 0; i < KVM_NR_DB_REGS; i++)
7528 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7530 kvm_update_dr7(vcpu);
7532 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7533 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7534 get_segment_base(vcpu, VCPU_SREG_CS);
7537 * Trigger an rflags update that will inject or remove the trace
7540 kvm_set_rflags(vcpu, rflags);
7542 kvm_x86_ops->update_bp_intercept(vcpu);
7552 * Translate a guest virtual address to a guest physical address.
7554 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7555 struct kvm_translation *tr)
7557 unsigned long vaddr = tr->linear_address;
7561 idx = srcu_read_lock(&vcpu->kvm->srcu);
7562 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7563 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7564 tr->physical_address = gpa;
7565 tr->valid = gpa != UNMAPPED_GVA;
7572 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7574 struct fxregs_state *fxsave =
7575 &vcpu->arch.guest_fpu.state.fxsave;
7577 memcpy(fpu->fpr, fxsave->st_space, 128);
7578 fpu->fcw = fxsave->cwd;
7579 fpu->fsw = fxsave->swd;
7580 fpu->ftwx = fxsave->twd;
7581 fpu->last_opcode = fxsave->fop;
7582 fpu->last_ip = fxsave->rip;
7583 fpu->last_dp = fxsave->rdp;
7584 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7589 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7591 struct fxregs_state *fxsave =
7592 &vcpu->arch.guest_fpu.state.fxsave;
7594 memcpy(fxsave->st_space, fpu->fpr, 128);
7595 fxsave->cwd = fpu->fcw;
7596 fxsave->swd = fpu->fsw;
7597 fxsave->twd = fpu->ftwx;
7598 fxsave->fop = fpu->last_opcode;
7599 fxsave->rip = fpu->last_ip;
7600 fxsave->rdp = fpu->last_dp;
7601 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7606 static void fx_init(struct kvm_vcpu *vcpu)
7608 fpstate_init(&vcpu->arch.guest_fpu.state);
7609 if (boot_cpu_has(X86_FEATURE_XSAVES))
7610 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7611 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7614 * Ensure guest xcr0 is valid for loading
7616 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7618 vcpu->arch.cr0 |= X86_CR0_ET;
7621 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7623 if (vcpu->guest_fpu_loaded)
7627 * Restore all possible states in the guest,
7628 * and assume host would use all available bits.
7629 * Guest xcr0 would be loaded later.
7631 vcpu->guest_fpu_loaded = 1;
7632 __kernel_fpu_begin();
7633 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7637 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7639 if (!vcpu->guest_fpu_loaded)
7642 vcpu->guest_fpu_loaded = 0;
7643 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7645 ++vcpu->stat.fpu_reload;
7649 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7651 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7653 kvmclock_reset(vcpu);
7655 kvm_x86_ops->vcpu_free(vcpu);
7656 free_cpumask_var(wbinvd_dirty_mask);
7659 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7662 struct kvm_vcpu *vcpu;
7664 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7665 printk_once(KERN_WARNING
7666 "kvm: SMP vm created on host with unstable TSC; "
7667 "guest TSC will not be reliable\n");
7669 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7674 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7678 kvm_vcpu_mtrr_init(vcpu);
7679 r = vcpu_load(vcpu);
7682 kvm_vcpu_reset(vcpu, false);
7683 kvm_mmu_setup(vcpu);
7688 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7690 struct msr_data msr;
7691 struct kvm *kvm = vcpu->kvm;
7693 kvm_hv_vcpu_postcreate(vcpu);
7695 if (vcpu_load(vcpu))
7698 msr.index = MSR_IA32_TSC;
7699 msr.host_initiated = true;
7700 kvm_write_tsc(vcpu, &msr);
7703 if (!kvmclock_periodic_sync)
7706 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7707 KVMCLOCK_SYNC_PERIOD);
7710 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7713 vcpu->arch.apf.msr_val = 0;
7715 r = vcpu_load(vcpu);
7717 kvm_mmu_unload(vcpu);
7720 kvm_x86_ops->vcpu_free(vcpu);
7723 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7725 vcpu->arch.hflags = 0;
7727 vcpu->arch.smi_pending = 0;
7728 atomic_set(&vcpu->arch.nmi_queued, 0);
7729 vcpu->arch.nmi_pending = 0;
7730 vcpu->arch.nmi_injected = false;
7731 kvm_clear_interrupt_queue(vcpu);
7732 kvm_clear_exception_queue(vcpu);
7734 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7735 kvm_update_dr0123(vcpu);
7736 vcpu->arch.dr6 = DR6_INIT;
7737 kvm_update_dr6(vcpu);
7738 vcpu->arch.dr7 = DR7_FIXED_1;
7739 kvm_update_dr7(vcpu);
7743 kvm_make_request(KVM_REQ_EVENT, vcpu);
7744 vcpu->arch.apf.msr_val = 0;
7745 vcpu->arch.st.msr_val = 0;
7747 kvmclock_reset(vcpu);
7749 kvm_clear_async_pf_completion_queue(vcpu);
7750 kvm_async_pf_hash_reset(vcpu);
7751 vcpu->arch.apf.halted = false;
7754 kvm_pmu_reset(vcpu);
7755 vcpu->arch.smbase = 0x30000;
7757 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
7758 vcpu->arch.msr_misc_features_enables = 0;
7761 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7762 vcpu->arch.regs_avail = ~0;
7763 vcpu->arch.regs_dirty = ~0;
7765 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7768 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7770 struct kvm_segment cs;
7772 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7773 cs.selector = vector << 8;
7774 cs.base = vector << 12;
7775 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7776 kvm_rip_write(vcpu, 0);
7779 int kvm_arch_hardware_enable(void)
7782 struct kvm_vcpu *vcpu;
7787 bool stable, backwards_tsc = false;
7789 kvm_shared_msr_cpu_online();
7790 ret = kvm_x86_ops->hardware_enable();
7794 local_tsc = rdtsc();
7795 stable = !check_tsc_unstable();
7796 list_for_each_entry(kvm, &vm_list, vm_list) {
7797 kvm_for_each_vcpu(i, vcpu, kvm) {
7798 if (!stable && vcpu->cpu == smp_processor_id())
7799 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7800 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7801 backwards_tsc = true;
7802 if (vcpu->arch.last_host_tsc > max_tsc)
7803 max_tsc = vcpu->arch.last_host_tsc;
7809 * Sometimes, even reliable TSCs go backwards. This happens on
7810 * platforms that reset TSC during suspend or hibernate actions, but
7811 * maintain synchronization. We must compensate. Fortunately, we can
7812 * detect that condition here, which happens early in CPU bringup,
7813 * before any KVM threads can be running. Unfortunately, we can't
7814 * bring the TSCs fully up to date with real time, as we aren't yet far
7815 * enough into CPU bringup that we know how much real time has actually
7816 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
7817 * variables that haven't been updated yet.
7819 * So we simply find the maximum observed TSC above, then record the
7820 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7821 * the adjustment will be applied. Note that we accumulate
7822 * adjustments, in case multiple suspend cycles happen before some VCPU
7823 * gets a chance to run again. In the event that no KVM threads get a
7824 * chance to run, we will miss the entire elapsed period, as we'll have
7825 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7826 * loose cycle time. This isn't too big a deal, since the loss will be
7827 * uniform across all VCPUs (not to mention the scenario is extremely
7828 * unlikely). It is possible that a second hibernate recovery happens
7829 * much faster than a first, causing the observed TSC here to be
7830 * smaller; this would require additional padding adjustment, which is
7831 * why we set last_host_tsc to the local tsc observed here.
7833 * N.B. - this code below runs only on platforms with reliable TSC,
7834 * as that is the only way backwards_tsc is set above. Also note
7835 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7836 * have the same delta_cyc adjustment applied if backwards_tsc
7837 * is detected. Note further, this adjustment is only done once,
7838 * as we reset last_host_tsc on all VCPUs to stop this from being
7839 * called multiple times (one for each physical CPU bringup).
7841 * Platforms with unreliable TSCs don't have to deal with this, they
7842 * will be compensated by the logic in vcpu_load, which sets the TSC to
7843 * catchup mode. This will catchup all VCPUs to real time, but cannot
7844 * guarantee that they stay in perfect synchronization.
7846 if (backwards_tsc) {
7847 u64 delta_cyc = max_tsc - local_tsc;
7848 list_for_each_entry(kvm, &vm_list, vm_list) {
7849 kvm->arch.backwards_tsc_observed = true;
7850 kvm_for_each_vcpu(i, vcpu, kvm) {
7851 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7852 vcpu->arch.last_host_tsc = local_tsc;
7853 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7857 * We have to disable TSC offset matching.. if you were
7858 * booting a VM while issuing an S4 host suspend....
7859 * you may have some problem. Solving this issue is
7860 * left as an exercise to the reader.
7862 kvm->arch.last_tsc_nsec = 0;
7863 kvm->arch.last_tsc_write = 0;
7870 void kvm_arch_hardware_disable(void)
7872 kvm_x86_ops->hardware_disable();
7873 drop_user_return_notifiers();
7876 int kvm_arch_hardware_setup(void)
7880 r = kvm_x86_ops->hardware_setup();
7884 if (kvm_has_tsc_control) {
7886 * Make sure the user can only configure tsc_khz values that
7887 * fit into a signed integer.
7888 * A min value is not calculated needed because it will always
7889 * be 1 on all machines.
7891 u64 max = min(0x7fffffffULL,
7892 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7893 kvm_max_guest_tsc_khz = max;
7895 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7898 kvm_init_msr_list();
7902 void kvm_arch_hardware_unsetup(void)
7904 kvm_x86_ops->hardware_unsetup();
7907 void kvm_arch_check_processor_compat(void *rtn)
7909 kvm_x86_ops->check_processor_compatibility(rtn);
7912 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7914 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7916 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7918 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7920 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7923 struct static_key kvm_no_apic_vcpu __read_mostly;
7924 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7926 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7932 BUG_ON(vcpu->kvm == NULL);
7935 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7936 vcpu->arch.pv.pv_unhalted = false;
7937 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7938 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7939 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7941 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7943 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7948 vcpu->arch.pio_data = page_address(page);
7950 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7952 r = kvm_mmu_create(vcpu);
7954 goto fail_free_pio_data;
7956 if (irqchip_in_kernel(kvm)) {
7957 r = kvm_create_lapic(vcpu);
7959 goto fail_mmu_destroy;
7961 static_key_slow_inc(&kvm_no_apic_vcpu);
7963 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7965 if (!vcpu->arch.mce_banks) {
7967 goto fail_free_lapic;
7969 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7971 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7973 goto fail_free_mce_banks;
7978 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7979 vcpu->arch.pv_time_enabled = false;
7981 vcpu->arch.guest_supported_xcr0 = 0;
7982 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7984 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7986 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7988 kvm_async_pf_hash_reset(vcpu);
7991 vcpu->arch.pending_external_vector = -1;
7992 vcpu->arch.preempted_in_kernel = false;
7994 kvm_hv_vcpu_init(vcpu);
7998 fail_free_mce_banks:
7999 kfree(vcpu->arch.mce_banks);
8001 kvm_free_lapic(vcpu);
8003 kvm_mmu_destroy(vcpu);
8005 free_page((unsigned long)vcpu->arch.pio_data);
8010 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8014 kvm_hv_vcpu_uninit(vcpu);
8015 kvm_pmu_destroy(vcpu);
8016 kfree(vcpu->arch.mce_banks);
8017 kvm_free_lapic(vcpu);
8018 idx = srcu_read_lock(&vcpu->kvm->srcu);
8019 kvm_mmu_destroy(vcpu);
8020 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8021 free_page((unsigned long)vcpu->arch.pio_data);
8022 if (!lapic_in_kernel(vcpu))
8023 static_key_slow_dec(&kvm_no_apic_vcpu);
8026 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8028 kvm_x86_ops->sched_in(vcpu, cpu);
8031 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8036 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8037 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8038 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8039 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8040 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8042 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8043 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8044 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8045 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8046 &kvm->arch.irq_sources_bitmap);
8048 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8049 mutex_init(&kvm->arch.apic_map_lock);
8050 mutex_init(&kvm->arch.hyperv.hv_lock);
8051 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8053 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8054 pvclock_update_vm_gtod_copy(kvm);
8056 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8057 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8059 kvm_page_track_init(kvm);
8060 kvm_mmu_init_vm(kvm);
8062 if (kvm_x86_ops->vm_init)
8063 return kvm_x86_ops->vm_init(kvm);
8068 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8071 r = vcpu_load(vcpu);
8073 kvm_mmu_unload(vcpu);
8077 static void kvm_free_vcpus(struct kvm *kvm)
8080 struct kvm_vcpu *vcpu;
8083 * Unpin any mmu pages first.
8085 kvm_for_each_vcpu(i, vcpu, kvm) {
8086 kvm_clear_async_pf_completion_queue(vcpu);
8087 kvm_unload_vcpu_mmu(vcpu);
8089 kvm_for_each_vcpu(i, vcpu, kvm)
8090 kvm_arch_vcpu_free(vcpu);
8092 mutex_lock(&kvm->lock);
8093 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8094 kvm->vcpus[i] = NULL;
8096 atomic_set(&kvm->online_vcpus, 0);
8097 mutex_unlock(&kvm->lock);
8100 void kvm_arch_sync_events(struct kvm *kvm)
8102 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8103 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8107 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8111 struct kvm_memslots *slots = kvm_memslots(kvm);
8112 struct kvm_memory_slot *slot, old;
8114 /* Called with kvm->slots_lock held. */
8115 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8118 slot = id_to_memslot(slots, id);
8124 * MAP_SHARED to prevent internal slot pages from being moved
8127 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8128 MAP_SHARED | MAP_ANONYMOUS, 0);
8129 if (IS_ERR((void *)hva))
8130 return PTR_ERR((void *)hva);
8139 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8140 struct kvm_userspace_memory_region m;
8142 m.slot = id | (i << 16);
8144 m.guest_phys_addr = gpa;
8145 m.userspace_addr = hva;
8146 m.memory_size = size;
8147 r = __kvm_set_memory_region(kvm, &m);
8153 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8159 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8161 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8165 mutex_lock(&kvm->slots_lock);
8166 r = __x86_set_memory_region(kvm, id, gpa, size);
8167 mutex_unlock(&kvm->slots_lock);
8171 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8173 void kvm_arch_destroy_vm(struct kvm *kvm)
8175 if (current->mm == kvm->mm) {
8177 * Free memory regions allocated on behalf of userspace,
8178 * unless the the memory map has changed due to process exit
8181 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8182 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8183 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8185 if (kvm_x86_ops->vm_destroy)
8186 kvm_x86_ops->vm_destroy(kvm);
8187 kvm_pic_destroy(kvm);
8188 kvm_ioapic_destroy(kvm);
8189 kvm_free_vcpus(kvm);
8190 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8191 kvm_mmu_uninit_vm(kvm);
8192 kvm_page_track_cleanup(kvm);
8195 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8196 struct kvm_memory_slot *dont)
8200 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8201 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8202 kvfree(free->arch.rmap[i]);
8203 free->arch.rmap[i] = NULL;
8208 if (!dont || free->arch.lpage_info[i - 1] !=
8209 dont->arch.lpage_info[i - 1]) {
8210 kvfree(free->arch.lpage_info[i - 1]);
8211 free->arch.lpage_info[i - 1] = NULL;
8215 kvm_page_track_free_memslot(free, dont);
8218 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8219 unsigned long npages)
8223 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8224 struct kvm_lpage_info *linfo;
8229 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8230 slot->base_gfn, level) + 1;
8232 slot->arch.rmap[i] =
8233 kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8234 if (!slot->arch.rmap[i])
8239 linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8243 slot->arch.lpage_info[i - 1] = linfo;
8245 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8246 linfo[0].disallow_lpage = 1;
8247 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8248 linfo[lpages - 1].disallow_lpage = 1;
8249 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8251 * If the gfn and userspace address are not aligned wrt each
8252 * other, or if explicitly asked to, disable large page
8253 * support for this slot
8255 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8256 !kvm_largepages_enabled()) {
8259 for (j = 0; j < lpages; ++j)
8260 linfo[j].disallow_lpage = 1;
8264 if (kvm_page_track_create_memslot(slot, npages))
8270 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8271 kvfree(slot->arch.rmap[i]);
8272 slot->arch.rmap[i] = NULL;
8276 kvfree(slot->arch.lpage_info[i - 1]);
8277 slot->arch.lpage_info[i - 1] = NULL;
8282 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8285 * memslots->generation has been incremented.
8286 * mmio generation may have reached its maximum value.
8288 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8291 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8292 struct kvm_memory_slot *memslot,
8293 const struct kvm_userspace_memory_region *mem,
8294 enum kvm_mr_change change)
8299 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8300 struct kvm_memory_slot *new)
8302 /* Still write protect RO slot */
8303 if (new->flags & KVM_MEM_READONLY) {
8304 kvm_mmu_slot_remove_write_access(kvm, new);
8309 * Call kvm_x86_ops dirty logging hooks when they are valid.
8311 * kvm_x86_ops->slot_disable_log_dirty is called when:
8313 * - KVM_MR_CREATE with dirty logging is disabled
8314 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8316 * The reason is, in case of PML, we need to set D-bit for any slots
8317 * with dirty logging disabled in order to eliminate unnecessary GPA
8318 * logging in PML buffer (and potential PML buffer full VMEXT). This
8319 * guarantees leaving PML enabled during guest's lifetime won't have
8320 * any additonal overhead from PML when guest is running with dirty
8321 * logging disabled for memory slots.
8323 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8324 * to dirty logging mode.
8326 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8328 * In case of write protect:
8330 * Write protect all pages for dirty logging.
8332 * All the sptes including the large sptes which point to this
8333 * slot are set to readonly. We can not create any new large
8334 * spte on this slot until the end of the logging.
8336 * See the comments in fast_page_fault().
8338 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8339 if (kvm_x86_ops->slot_enable_log_dirty)
8340 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8342 kvm_mmu_slot_remove_write_access(kvm, new);
8344 if (kvm_x86_ops->slot_disable_log_dirty)
8345 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8349 void kvm_arch_commit_memory_region(struct kvm *kvm,
8350 const struct kvm_userspace_memory_region *mem,
8351 const struct kvm_memory_slot *old,
8352 const struct kvm_memory_slot *new,
8353 enum kvm_mr_change change)
8355 int nr_mmu_pages = 0;
8357 if (!kvm->arch.n_requested_mmu_pages)
8358 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8361 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8364 * Dirty logging tracks sptes in 4k granularity, meaning that large
8365 * sptes have to be split. If live migration is successful, the guest
8366 * in the source machine will be destroyed and large sptes will be
8367 * created in the destination. However, if the guest continues to run
8368 * in the source machine (for example if live migration fails), small
8369 * sptes will remain around and cause bad performance.
8371 * Scan sptes if dirty logging has been stopped, dropping those
8372 * which can be collapsed into a single large-page spte. Later
8373 * page faults will create the large-page sptes.
8375 if ((change != KVM_MR_DELETE) &&
8376 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8377 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8378 kvm_mmu_zap_collapsible_sptes(kvm, new);
8381 * Set up write protection and/or dirty logging for the new slot.
8383 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8384 * been zapped so no dirty logging staff is needed for old slot. For
8385 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8386 * new and it's also covered when dealing with the new slot.
8388 * FIXME: const-ify all uses of struct kvm_memory_slot.
8390 if (change != KVM_MR_DELETE)
8391 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8394 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8396 kvm_mmu_invalidate_zap_all_pages(kvm);
8399 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8400 struct kvm_memory_slot *slot)
8402 kvm_page_track_flush_slot(kvm, slot);
8405 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8407 if (!list_empty_careful(&vcpu->async_pf.done))
8410 if (kvm_apic_has_events(vcpu))
8413 if (vcpu->arch.pv.pv_unhalted)
8416 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
8417 (vcpu->arch.nmi_pending &&
8418 kvm_x86_ops->nmi_allowed(vcpu)))
8421 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
8422 (vcpu->arch.smi_pending && !is_smm(vcpu)))
8425 if (kvm_arch_interrupt_allowed(vcpu) &&
8426 kvm_cpu_has_interrupt(vcpu))
8429 if (kvm_hv_has_stimer_pending(vcpu))
8435 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8437 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8440 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
8442 return vcpu->arch.preempted_in_kernel;
8445 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8447 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8450 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8452 return kvm_x86_ops->interrupt_allowed(vcpu);
8455 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8457 if (is_64_bit_mode(vcpu))
8458 return kvm_rip_read(vcpu);
8459 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8460 kvm_rip_read(vcpu));
8462 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8464 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8466 return kvm_get_linear_rip(vcpu) == linear_rip;
8468 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8470 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8472 unsigned long rflags;
8474 rflags = kvm_x86_ops->get_rflags(vcpu);
8475 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8476 rflags &= ~X86_EFLAGS_TF;
8479 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8481 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8483 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8484 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8485 rflags |= X86_EFLAGS_TF;
8486 kvm_x86_ops->set_rflags(vcpu, rflags);
8489 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8491 __kvm_set_rflags(vcpu, rflags);
8492 kvm_make_request(KVM_REQ_EVENT, vcpu);
8494 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8496 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8500 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8504 r = kvm_mmu_reload(vcpu);
8508 if (!vcpu->arch.mmu.direct_map &&
8509 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8512 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8515 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8517 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8520 static inline u32 kvm_async_pf_next_probe(u32 key)
8522 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8525 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8527 u32 key = kvm_async_pf_hash_fn(gfn);
8529 while (vcpu->arch.apf.gfns[key] != ~0)
8530 key = kvm_async_pf_next_probe(key);
8532 vcpu->arch.apf.gfns[key] = gfn;
8535 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8538 u32 key = kvm_async_pf_hash_fn(gfn);
8540 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8541 (vcpu->arch.apf.gfns[key] != gfn &&
8542 vcpu->arch.apf.gfns[key] != ~0); i++)
8543 key = kvm_async_pf_next_probe(key);
8548 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8550 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8553 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8557 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8559 vcpu->arch.apf.gfns[i] = ~0;
8561 j = kvm_async_pf_next_probe(j);
8562 if (vcpu->arch.apf.gfns[j] == ~0)
8564 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8566 * k lies cyclically in ]i,j]
8568 * |....j i.k.| or |.k..j i...|
8570 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8571 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8576 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8579 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8583 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8584 struct kvm_async_pf *work)
8586 struct x86_exception fault;
8588 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8589 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8591 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8592 (vcpu->arch.apf.send_user_only &&
8593 kvm_x86_ops->get_cpl(vcpu) == 0))
8594 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8595 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8596 fault.vector = PF_VECTOR;
8597 fault.error_code_valid = true;
8598 fault.error_code = 0;
8599 fault.nested_page_fault = false;
8600 fault.address = work->arch.token;
8601 fault.async_page_fault = true;
8602 kvm_inject_page_fault(vcpu, &fault);
8606 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8607 struct kvm_async_pf *work)
8609 struct x86_exception fault;
8611 if (work->wakeup_all)
8612 work->arch.token = ~0; /* broadcast wakeup */
8614 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8615 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8617 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8618 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8619 fault.vector = PF_VECTOR;
8620 fault.error_code_valid = true;
8621 fault.error_code = 0;
8622 fault.nested_page_fault = false;
8623 fault.address = work->arch.token;
8624 fault.async_page_fault = true;
8625 kvm_inject_page_fault(vcpu, &fault);
8627 vcpu->arch.apf.halted = false;
8628 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8631 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8633 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8636 return kvm_can_do_async_pf(vcpu);
8639 void kvm_arch_start_assignment(struct kvm *kvm)
8641 atomic_inc(&kvm->arch.assigned_device_count);
8643 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8645 void kvm_arch_end_assignment(struct kvm *kvm)
8647 atomic_dec(&kvm->arch.assigned_device_count);
8649 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8651 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8653 return atomic_read(&kvm->arch.assigned_device_count);
8655 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8657 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8659 atomic_inc(&kvm->arch.noncoherent_dma_count);
8661 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8663 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8665 atomic_dec(&kvm->arch.noncoherent_dma_count);
8667 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8669 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8671 return atomic_read(&kvm->arch.noncoherent_dma_count);
8673 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8675 bool kvm_arch_has_irq_bypass(void)
8677 return kvm_x86_ops->update_pi_irte != NULL;
8680 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8681 struct irq_bypass_producer *prod)
8683 struct kvm_kernel_irqfd *irqfd =
8684 container_of(cons, struct kvm_kernel_irqfd, consumer);
8686 irqfd->producer = prod;
8688 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8689 prod->irq, irqfd->gsi, 1);
8692 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8693 struct irq_bypass_producer *prod)
8696 struct kvm_kernel_irqfd *irqfd =
8697 container_of(cons, struct kvm_kernel_irqfd, consumer);
8699 WARN_ON(irqfd->producer != prod);
8700 irqfd->producer = NULL;
8703 * When producer of consumer is unregistered, we change back to
8704 * remapped mode, so we can re-use the current implementation
8705 * when the irq is masked/disabled or the consumer side (KVM
8706 * int this case doesn't want to receive the interrupts.
8708 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8710 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8711 " fails: %d\n", irqfd->consumer.token, ret);
8714 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8715 uint32_t guest_irq, bool set)
8717 if (!kvm_x86_ops->update_pi_irte)
8720 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8723 bool kvm_vector_hashing_enabled(void)
8725 return vector_hashing;
8727 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8729 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8730 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8731 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8732 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8733 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8734 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8735 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8736 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8737 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8738 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8739 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8740 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8741 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8742 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8743 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8744 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8745 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8746 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8747 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);