1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * derived from drivers/kvm/kvm_main.c
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright (C) 2008 Qumranet, Inc.
9 * Copyright IBM Corporation, 2008
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13 * Avi Kivity <avi@qumranet.com>
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Amit Shah <amit.shah@qumranet.com>
16 * Ben-Ami Yassour <benami@il.ibm.com>
19 #include <linux/kvm_host.h>
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
62 #include <trace/events/kvm.h>
64 #include <asm/debugreg.h>
68 #include <linux/kernel_stat.h>
69 #include <asm/fpu/internal.h> /* Ugh! */
70 #include <asm/pvclock.h>
71 #include <asm/div64.h>
72 #include <asm/irq_remapping.h>
73 #include <asm/mshyperv.h>
74 #include <asm/hypervisor.h>
75 #include <asm/tlbflush.h>
76 #include <asm/intel_pt.h>
77 #include <asm/emulate_prefix.h>
79 #include <clocksource/hyperv_timer.h>
81 #define CREATE_TRACE_POINTS
84 #define MAX_IO_MSRS 256
85 #define KVM_MAX_MCE_BANKS 32
86 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
87 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
89 #define emul_to_vcpu(ctxt) \
90 ((struct kvm_vcpu *)(ctxt)->vcpu)
93 * - enable syscall per default because its emulated by KVM
94 * - enable LME and LMA per default on 64 bit KVM
98 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
100 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
103 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
105 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
106 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
108 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
109 static void process_nmi(struct kvm_vcpu *vcpu);
110 static void process_smi(struct kvm_vcpu *vcpu);
111 static void enter_smm(struct kvm_vcpu *vcpu);
112 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
113 static void store_regs(struct kvm_vcpu *vcpu);
114 static int sync_regs(struct kvm_vcpu *vcpu);
116 struct kvm_x86_ops kvm_x86_ops __read_mostly;
117 EXPORT_SYMBOL_GPL(kvm_x86_ops);
119 #define KVM_X86_OP(func) \
120 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
121 *(((struct kvm_x86_ops *)0)->func));
122 #define KVM_X86_OP_NULL KVM_X86_OP
123 #include <asm/kvm-x86-ops.h>
124 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
125 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
126 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
128 static bool __read_mostly ignore_msrs = 0;
129 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
131 bool __read_mostly report_ignored_msrs = true;
132 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
133 EXPORT_SYMBOL_GPL(report_ignored_msrs);
135 unsigned int min_timer_period_us = 200;
136 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
138 static bool __read_mostly kvmclock_periodic_sync = true;
139 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
141 bool __read_mostly kvm_has_tsc_control;
142 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
143 u32 __read_mostly kvm_max_guest_tsc_khz;
144 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
145 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
146 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
147 u64 __read_mostly kvm_max_tsc_scaling_ratio;
148 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
149 u64 __read_mostly kvm_default_tsc_scaling_ratio;
150 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
151 bool __read_mostly kvm_has_bus_lock_exit;
152 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
154 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
155 static u32 __read_mostly tsc_tolerance_ppm = 250;
156 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
159 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
160 * adaptive tuning starting from default advancement of 1000ns. '0' disables
161 * advancement entirely. Any other value is used as-is and disables adaptive
162 * tuning, i.e. allows privileged userspace to set an exact advancement time.
164 static int __read_mostly lapic_timer_advance_ns = -1;
165 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
167 static bool __read_mostly vector_hashing = true;
168 module_param(vector_hashing, bool, S_IRUGO);
170 bool __read_mostly enable_vmware_backdoor = false;
171 module_param(enable_vmware_backdoor, bool, S_IRUGO);
172 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
174 static bool __read_mostly force_emulation_prefix = false;
175 module_param(force_emulation_prefix, bool, S_IRUGO);
177 int __read_mostly pi_inject_timer = -1;
178 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
181 * Restoring the host value for MSRs that are only consumed when running in
182 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
183 * returns to userspace, i.e. the kernel can run with the guest's value.
185 #define KVM_MAX_NR_USER_RETURN_MSRS 16
187 struct kvm_user_return_msrs_global {
189 u32 msrs[KVM_MAX_NR_USER_RETURN_MSRS];
192 struct kvm_user_return_msrs {
193 struct user_return_notifier urn;
195 struct kvm_user_return_msr_values {
198 } values[KVM_MAX_NR_USER_RETURN_MSRS];
201 static struct kvm_user_return_msrs_global __read_mostly user_return_msrs_global;
202 static struct kvm_user_return_msrs __percpu *user_return_msrs;
204 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
205 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
206 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
207 | XFEATURE_MASK_PKRU)
209 u64 __read_mostly host_efer;
210 EXPORT_SYMBOL_GPL(host_efer);
212 bool __read_mostly allow_smaller_maxphyaddr = 0;
213 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
215 u64 __read_mostly host_xss;
216 EXPORT_SYMBOL_GPL(host_xss);
217 u64 __read_mostly supported_xss;
218 EXPORT_SYMBOL_GPL(supported_xss);
220 struct kvm_stats_debugfs_item debugfs_entries[] = {
221 VCPU_STAT("pf_fixed", pf_fixed),
222 VCPU_STAT("pf_guest", pf_guest),
223 VCPU_STAT("tlb_flush", tlb_flush),
224 VCPU_STAT("invlpg", invlpg),
225 VCPU_STAT("exits", exits),
226 VCPU_STAT("io_exits", io_exits),
227 VCPU_STAT("mmio_exits", mmio_exits),
228 VCPU_STAT("signal_exits", signal_exits),
229 VCPU_STAT("irq_window", irq_window_exits),
230 VCPU_STAT("nmi_window", nmi_window_exits),
231 VCPU_STAT("halt_exits", halt_exits),
232 VCPU_STAT("halt_successful_poll", halt_successful_poll),
233 VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
234 VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
235 VCPU_STAT("halt_wakeup", halt_wakeup),
236 VCPU_STAT("hypercalls", hypercalls),
237 VCPU_STAT("request_irq", request_irq_exits),
238 VCPU_STAT("irq_exits", irq_exits),
239 VCPU_STAT("host_state_reload", host_state_reload),
240 VCPU_STAT("fpu_reload", fpu_reload),
241 VCPU_STAT("insn_emulation", insn_emulation),
242 VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
243 VCPU_STAT("irq_injections", irq_injections),
244 VCPU_STAT("nmi_injections", nmi_injections),
245 VCPU_STAT("req_event", req_event),
246 VCPU_STAT("l1d_flush", l1d_flush),
247 VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
248 VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
249 VCPU_STAT("nested_run", nested_run),
250 VCPU_STAT("directed_yield_attempted", directed_yield_attempted),
251 VCPU_STAT("directed_yield_successful", directed_yield_successful),
252 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
253 VM_STAT("mmu_pte_write", mmu_pte_write),
254 VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
255 VM_STAT("mmu_flooded", mmu_flooded),
256 VM_STAT("mmu_recycled", mmu_recycled),
257 VM_STAT("mmu_cache_miss", mmu_cache_miss),
258 VM_STAT("mmu_unsync", mmu_unsync),
259 VM_STAT("remote_tlb_flush", remote_tlb_flush),
260 VM_STAT("largepages", lpages, .mode = 0444),
261 VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
262 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
266 u64 __read_mostly host_xcr0;
267 u64 __read_mostly supported_xcr0;
268 EXPORT_SYMBOL_GPL(supported_xcr0);
270 static struct kmem_cache *x86_fpu_cache;
272 static struct kmem_cache *x86_emulator_cache;
275 * When called, it means the previous get/set msr reached an invalid msr.
276 * Return true if we want to ignore/silent this failed msr access.
278 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
280 const char *op = write ? "wrmsr" : "rdmsr";
283 if (report_ignored_msrs)
284 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
289 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
295 static struct kmem_cache *kvm_alloc_emulator_cache(void)
297 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
298 unsigned int size = sizeof(struct x86_emulate_ctxt);
300 return kmem_cache_create_usercopy("x86_emulator", size,
301 __alignof__(struct x86_emulate_ctxt),
302 SLAB_ACCOUNT, useroffset,
303 size - useroffset, NULL);
306 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
308 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
311 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
312 vcpu->arch.apf.gfns[i] = ~0;
315 static void kvm_on_user_return(struct user_return_notifier *urn)
318 struct kvm_user_return_msrs *msrs
319 = container_of(urn, struct kvm_user_return_msrs, urn);
320 struct kvm_user_return_msr_values *values;
324 * Disabling irqs at this point since the following code could be
325 * interrupted and executed through kvm_arch_hardware_disable()
327 local_irq_save(flags);
328 if (msrs->registered) {
329 msrs->registered = false;
330 user_return_notifier_unregister(urn);
332 local_irq_restore(flags);
333 for (slot = 0; slot < user_return_msrs_global.nr; ++slot) {
334 values = &msrs->values[slot];
335 if (values->host != values->curr) {
336 wrmsrl(user_return_msrs_global.msrs[slot], values->host);
337 values->curr = values->host;
342 int kvm_probe_user_return_msr(u32 msr)
348 ret = rdmsrl_safe(msr, &val);
351 ret = wrmsrl_safe(msr, val);
356 EXPORT_SYMBOL_GPL(kvm_probe_user_return_msr);
358 void kvm_define_user_return_msr(unsigned slot, u32 msr)
360 BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS);
361 user_return_msrs_global.msrs[slot] = msr;
362 if (slot >= user_return_msrs_global.nr)
363 user_return_msrs_global.nr = slot + 1;
365 EXPORT_SYMBOL_GPL(kvm_define_user_return_msr);
367 int kvm_find_user_return_msr(u32 msr)
371 for (i = 0; i < user_return_msrs_global.nr; ++i) {
372 if (user_return_msrs_global.msrs[i] == msr)
377 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
379 static void kvm_user_return_msr_cpu_online(void)
381 unsigned int cpu = smp_processor_id();
382 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
386 for (i = 0; i < user_return_msrs_global.nr; ++i) {
387 rdmsrl_safe(user_return_msrs_global.msrs[i], &value);
388 msrs->values[i].host = value;
389 msrs->values[i].curr = value;
393 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
395 unsigned int cpu = smp_processor_id();
396 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
399 value = (value & mask) | (msrs->values[slot].host & ~mask);
400 if (value == msrs->values[slot].curr)
402 err = wrmsrl_safe(user_return_msrs_global.msrs[slot], value);
406 msrs->values[slot].curr = value;
407 if (!msrs->registered) {
408 msrs->urn.on_user_return = kvm_on_user_return;
409 user_return_notifier_register(&msrs->urn);
410 msrs->registered = true;
414 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
416 static void drop_user_return_notifiers(void)
418 unsigned int cpu = smp_processor_id();
419 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
421 if (msrs->registered)
422 kvm_on_user_return(&msrs->urn);
425 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
427 return vcpu->arch.apic_base;
429 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
431 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
433 return kvm_apic_mode(kvm_get_apic_base(vcpu));
435 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
437 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
439 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
440 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
441 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
442 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
444 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
446 if (!msr_info->host_initiated) {
447 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
449 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
453 kvm_lapic_set_base(vcpu, msr_info->data);
454 kvm_recalculate_apic_map(vcpu->kvm);
457 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
459 asmlinkage __visible noinstr void kvm_spurious_fault(void)
461 /* Fault while not rebooting. We want the trace. */
462 BUG_ON(!kvm_rebooting);
464 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
466 #define EXCPT_BENIGN 0
467 #define EXCPT_CONTRIBUTORY 1
470 static int exception_class(int vector)
480 return EXCPT_CONTRIBUTORY;
487 #define EXCPT_FAULT 0
489 #define EXCPT_ABORT 2
490 #define EXCPT_INTERRUPT 3
492 static int exception_type(int vector)
496 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
497 return EXCPT_INTERRUPT;
501 /* #DB is trap, as instruction watchpoints are handled elsewhere */
502 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
505 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
508 /* Reserved exceptions will result in fault */
512 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
514 unsigned nr = vcpu->arch.exception.nr;
515 bool has_payload = vcpu->arch.exception.has_payload;
516 unsigned long payload = vcpu->arch.exception.payload;
524 * "Certain debug exceptions may clear bit 0-3. The
525 * remaining contents of the DR6 register are never
526 * cleared by the processor".
528 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
530 * In order to reflect the #DB exception payload in guest
531 * dr6, three components need to be considered: active low
532 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
534 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
535 * In the target guest dr6:
536 * FIXED_1 bits should always be set.
537 * Active low bits should be cleared if 1-setting in payload.
538 * Active high bits should be set if 1-setting in payload.
540 * Note, the payload is compatible with the pending debug
541 * exceptions/exit qualification under VMX, that active_low bits
542 * are active high in payload.
543 * So they need to be flipped for DR6.
545 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
546 vcpu->arch.dr6 |= payload;
547 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
550 * The #DB payload is defined as compatible with the 'pending
551 * debug exceptions' field under VMX, not DR6. While bit 12 is
552 * defined in the 'pending debug exceptions' field (enabled
553 * breakpoint), it is reserved and must be zero in DR6.
555 vcpu->arch.dr6 &= ~BIT(12);
558 vcpu->arch.cr2 = payload;
562 vcpu->arch.exception.has_payload = false;
563 vcpu->arch.exception.payload = 0;
565 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
567 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
568 unsigned nr, bool has_error, u32 error_code,
569 bool has_payload, unsigned long payload, bool reinject)
574 kvm_make_request(KVM_REQ_EVENT, vcpu);
576 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
580 * On vmentry, vcpu->arch.exception.pending is only
581 * true if an event injection was blocked by
582 * nested_run_pending. In that case, however,
583 * vcpu_enter_guest requests an immediate exit,
584 * and the guest shouldn't proceed far enough to
587 WARN_ON_ONCE(vcpu->arch.exception.pending);
588 vcpu->arch.exception.injected = true;
589 if (WARN_ON_ONCE(has_payload)) {
591 * A reinjected event has already
592 * delivered its payload.
598 vcpu->arch.exception.pending = true;
599 vcpu->arch.exception.injected = false;
601 vcpu->arch.exception.has_error_code = has_error;
602 vcpu->arch.exception.nr = nr;
603 vcpu->arch.exception.error_code = error_code;
604 vcpu->arch.exception.has_payload = has_payload;
605 vcpu->arch.exception.payload = payload;
606 if (!is_guest_mode(vcpu))
607 kvm_deliver_exception_payload(vcpu);
611 /* to check exception */
612 prev_nr = vcpu->arch.exception.nr;
613 if (prev_nr == DF_VECTOR) {
614 /* triple fault -> shutdown */
615 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
618 class1 = exception_class(prev_nr);
619 class2 = exception_class(nr);
620 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
621 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
623 * Generate double fault per SDM Table 5-5. Set
624 * exception.pending = true so that the double fault
625 * can trigger a nested vmexit.
627 vcpu->arch.exception.pending = true;
628 vcpu->arch.exception.injected = false;
629 vcpu->arch.exception.has_error_code = true;
630 vcpu->arch.exception.nr = DF_VECTOR;
631 vcpu->arch.exception.error_code = 0;
632 vcpu->arch.exception.has_payload = false;
633 vcpu->arch.exception.payload = 0;
635 /* replace previous exception with a new one in a hope
636 that instruction re-execution will regenerate lost
641 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
643 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
645 EXPORT_SYMBOL_GPL(kvm_queue_exception);
647 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
649 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
651 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
653 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
654 unsigned long payload)
656 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
658 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
660 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
661 u32 error_code, unsigned long payload)
663 kvm_multiple_exception(vcpu, nr, true, error_code,
664 true, payload, false);
667 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
670 kvm_inject_gp(vcpu, 0);
672 return kvm_skip_emulated_instruction(vcpu);
676 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
678 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
680 ++vcpu->stat.pf_guest;
681 vcpu->arch.exception.nested_apf =
682 is_guest_mode(vcpu) && fault->async_page_fault;
683 if (vcpu->arch.exception.nested_apf) {
684 vcpu->arch.apf.nested_apf_token = fault->address;
685 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
687 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
691 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
693 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
694 struct x86_exception *fault)
696 struct kvm_mmu *fault_mmu;
697 WARN_ON_ONCE(fault->vector != PF_VECTOR);
699 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
703 * Invalidate the TLB entry for the faulting address, if it exists,
704 * else the access will fault indefinitely (and to emulate hardware).
706 if ((fault->error_code & PFERR_PRESENT_MASK) &&
707 !(fault->error_code & PFERR_RSVD_MASK))
708 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
709 fault_mmu->root_hpa);
711 fault_mmu->inject_page_fault(vcpu, fault);
712 return fault->nested_page_fault;
714 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
716 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
718 atomic_inc(&vcpu->arch.nmi_queued);
719 kvm_make_request(KVM_REQ_NMI, vcpu);
721 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
723 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
725 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
727 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
729 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
731 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
733 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
736 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
737 * a #GP and return false.
739 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
741 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
743 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
746 EXPORT_SYMBOL_GPL(kvm_require_cpl);
748 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
750 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
753 kvm_queue_exception(vcpu, UD_VECTOR);
756 EXPORT_SYMBOL_GPL(kvm_require_dr);
759 * This function will be used to read from the physical memory of the currently
760 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
761 * can read from guest physical or from the guest's guest physical memory.
763 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
764 gfn_t ngfn, void *data, int offset, int len,
767 struct x86_exception exception;
771 ngpa = gfn_to_gpa(ngfn);
772 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
773 if (real_gfn == UNMAPPED_GVA)
776 real_gfn = gpa_to_gfn(real_gfn);
778 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
780 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
782 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
783 void *data, int offset, int len, u32 access)
785 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
786 data, offset, len, access);
789 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
791 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
795 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
797 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
799 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
800 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
803 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
805 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
806 offset * sizeof(u64), sizeof(pdpte),
807 PFERR_USER_MASK|PFERR_WRITE_MASK);
812 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
813 if ((pdpte[i] & PT_PRESENT_MASK) &&
814 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
821 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
822 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
828 EXPORT_SYMBOL_GPL(load_pdptrs);
830 bool pdptrs_changed(struct kvm_vcpu *vcpu)
832 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
837 if (!is_pae_paging(vcpu))
840 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
843 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
844 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
845 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
846 PFERR_USER_MASK | PFERR_WRITE_MASK);
850 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
852 EXPORT_SYMBOL_GPL(pdptrs_changed);
854 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
856 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
858 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
859 kvm_clear_async_pf_completion_queue(vcpu);
860 kvm_async_pf_hash_reset(vcpu);
863 if ((cr0 ^ old_cr0) & update_bits)
864 kvm_mmu_reset_context(vcpu);
866 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
867 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
868 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
869 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
871 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
873 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
875 unsigned long old_cr0 = kvm_read_cr0(vcpu);
876 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
881 if (cr0 & 0xffffffff00000000UL)
885 cr0 &= ~CR0_RESERVED_BITS;
887 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
890 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
894 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
895 (cr0 & X86_CR0_PG)) {
900 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
905 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
906 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
907 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
910 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
913 static_call(kvm_x86_set_cr0)(vcpu, cr0);
915 kvm_post_set_cr0(vcpu, old_cr0, cr0);
919 EXPORT_SYMBOL_GPL(kvm_set_cr0);
921 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
923 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
925 EXPORT_SYMBOL_GPL(kvm_lmsw);
927 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
929 if (vcpu->arch.guest_state_protected)
932 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
934 if (vcpu->arch.xcr0 != host_xcr0)
935 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
937 if (vcpu->arch.xsaves_enabled &&
938 vcpu->arch.ia32_xss != host_xss)
939 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
942 if (static_cpu_has(X86_FEATURE_PKU) &&
943 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
944 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
945 vcpu->arch.pkru != vcpu->arch.host_pkru)
946 __write_pkru(vcpu->arch.pkru);
948 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
950 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
952 if (vcpu->arch.guest_state_protected)
955 if (static_cpu_has(X86_FEATURE_PKU) &&
956 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
957 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
958 vcpu->arch.pkru = rdpkru();
959 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
960 __write_pkru(vcpu->arch.host_pkru);
963 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
965 if (vcpu->arch.xcr0 != host_xcr0)
966 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
968 if (vcpu->arch.xsaves_enabled &&
969 vcpu->arch.ia32_xss != host_xss)
970 wrmsrl(MSR_IA32_XSS, host_xss);
974 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
976 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
979 u64 old_xcr0 = vcpu->arch.xcr0;
982 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
983 if (index != XCR_XFEATURE_ENABLED_MASK)
985 if (!(xcr0 & XFEATURE_MASK_FP))
987 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
991 * Do not allow the guest to set bits that we do not support
992 * saving. However, xcr0 bit 0 is always set, even if the
993 * emulated CPU does not support XSAVE (see fx_init).
995 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
996 if (xcr0 & ~valid_bits)
999 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
1000 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1003 if (xcr0 & XFEATURE_MASK_AVX512) {
1004 if (!(xcr0 & XFEATURE_MASK_YMM))
1006 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1009 vcpu->arch.xcr0 = xcr0;
1011 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1012 kvm_update_cpuid_runtime(vcpu);
1016 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1018 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1019 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1020 kvm_inject_gp(vcpu, 0);
1024 return kvm_skip_emulated_instruction(vcpu);
1026 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1028 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1030 if (cr4 & cr4_reserved_bits)
1033 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1036 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1038 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1040 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1042 unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1043 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
1045 if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1046 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1047 kvm_mmu_reset_context(vcpu);
1049 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1051 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1053 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1054 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1057 if (!kvm_is_valid_cr4(vcpu, cr4))
1060 if (is_long_mode(vcpu)) {
1061 if (!(cr4 & X86_CR4_PAE))
1063 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1065 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1066 && ((cr4 ^ old_cr4) & pdptr_bits)
1067 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1068 kvm_read_cr3(vcpu)))
1071 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1072 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1075 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1076 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1080 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1082 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1086 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1088 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1090 bool skip_tlb_flush = false;
1091 #ifdef CONFIG_X86_64
1092 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1095 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1096 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1100 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1101 if (!skip_tlb_flush) {
1102 kvm_mmu_sync_roots(vcpu);
1103 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1109 * Do not condition the GPA check on long mode, this helper is used to
1110 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1111 * the current vCPU mode is accurate.
1113 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1116 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1119 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1120 vcpu->arch.cr3 = cr3;
1121 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1125 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1127 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1129 if (cr8 & CR8_RESERVED_BITS)
1131 if (lapic_in_kernel(vcpu))
1132 kvm_lapic_set_tpr(vcpu, cr8);
1134 vcpu->arch.cr8 = cr8;
1137 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1139 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1141 if (lapic_in_kernel(vcpu))
1142 return kvm_lapic_get_cr8(vcpu);
1144 return vcpu->arch.cr8;
1146 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1148 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1152 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1153 for (i = 0; i < KVM_NR_DB_REGS; i++)
1154 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1155 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1159 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1163 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1164 dr7 = vcpu->arch.guest_debug_dr7;
1166 dr7 = vcpu->arch.dr7;
1167 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1168 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1169 if (dr7 & DR7_BP_EN_MASK)
1170 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1172 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1174 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1176 u64 fixed = DR6_FIXED_1;
1178 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1183 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1185 size_t size = ARRAY_SIZE(vcpu->arch.db);
1189 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1190 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1191 vcpu->arch.eff_db[dr] = val;
1195 if (!kvm_dr6_valid(val))
1197 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1201 if (!kvm_dr7_valid(val))
1203 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1204 kvm_update_dr7(vcpu);
1210 EXPORT_SYMBOL_GPL(kvm_set_dr);
1212 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1214 size_t size = ARRAY_SIZE(vcpu->arch.db);
1218 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1222 *val = vcpu->arch.dr6;
1226 *val = vcpu->arch.dr7;
1230 EXPORT_SYMBOL_GPL(kvm_get_dr);
1232 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1234 u32 ecx = kvm_rcx_read(vcpu);
1237 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1238 kvm_inject_gp(vcpu, 0);
1242 kvm_rax_write(vcpu, (u32)data);
1243 kvm_rdx_write(vcpu, data >> 32);
1244 return kvm_skip_emulated_instruction(vcpu);
1246 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1249 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1250 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1252 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1253 * extract the supported MSRs from the related const lists.
1254 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1255 * capabilities of the host cpu. This capabilities test skips MSRs that are
1256 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1257 * may depend on host virtualization features rather than host cpu features.
1260 static const u32 msrs_to_save_all[] = {
1261 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1263 #ifdef CONFIG_X86_64
1264 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1266 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1267 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1269 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1270 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1271 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1272 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1273 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1274 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1275 MSR_IA32_UMWAIT_CONTROL,
1277 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1278 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1279 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1280 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1281 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1282 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1283 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1284 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1285 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1286 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1287 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1288 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1289 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1290 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1291 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1292 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1293 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1294 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1295 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1296 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1297 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1298 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1301 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1302 static unsigned num_msrs_to_save;
1304 static const u32 emulated_msrs_all[] = {
1305 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1306 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1307 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1308 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1309 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1310 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1311 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1313 HV_X64_MSR_VP_INDEX,
1314 HV_X64_MSR_VP_RUNTIME,
1315 HV_X64_MSR_SCONTROL,
1316 HV_X64_MSR_STIMER0_CONFIG,
1317 HV_X64_MSR_VP_ASSIST_PAGE,
1318 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1319 HV_X64_MSR_TSC_EMULATION_STATUS,
1320 HV_X64_MSR_SYNDBG_OPTIONS,
1321 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1322 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1323 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1325 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1326 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1328 MSR_IA32_TSC_ADJUST,
1329 MSR_IA32_TSC_DEADLINE,
1330 MSR_IA32_ARCH_CAPABILITIES,
1331 MSR_IA32_PERF_CAPABILITIES,
1332 MSR_IA32_MISC_ENABLE,
1333 MSR_IA32_MCG_STATUS,
1335 MSR_IA32_MCG_EXT_CTL,
1339 MSR_MISC_FEATURES_ENABLES,
1340 MSR_AMD64_VIRT_SPEC_CTRL,
1345 * The following list leaves out MSRs whose values are determined
1346 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1347 * We always support the "true" VMX control MSRs, even if the host
1348 * processor does not, so I am putting these registers here rather
1349 * than in msrs_to_save_all.
1352 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1353 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1354 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1355 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1357 MSR_IA32_VMX_CR0_FIXED0,
1358 MSR_IA32_VMX_CR4_FIXED0,
1359 MSR_IA32_VMX_VMCS_ENUM,
1360 MSR_IA32_VMX_PROCBASED_CTLS2,
1361 MSR_IA32_VMX_EPT_VPID_CAP,
1362 MSR_IA32_VMX_VMFUNC,
1365 MSR_KVM_POLL_CONTROL,
1368 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1369 static unsigned num_emulated_msrs;
1372 * List of msr numbers which are used to expose MSR-based features that
1373 * can be used by a hypervisor to validate requested CPU features.
1375 static const u32 msr_based_features_all[] = {
1377 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1378 MSR_IA32_VMX_PINBASED_CTLS,
1379 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1380 MSR_IA32_VMX_PROCBASED_CTLS,
1381 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1382 MSR_IA32_VMX_EXIT_CTLS,
1383 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1384 MSR_IA32_VMX_ENTRY_CTLS,
1386 MSR_IA32_VMX_CR0_FIXED0,
1387 MSR_IA32_VMX_CR0_FIXED1,
1388 MSR_IA32_VMX_CR4_FIXED0,
1389 MSR_IA32_VMX_CR4_FIXED1,
1390 MSR_IA32_VMX_VMCS_ENUM,
1391 MSR_IA32_VMX_PROCBASED_CTLS2,
1392 MSR_IA32_VMX_EPT_VPID_CAP,
1393 MSR_IA32_VMX_VMFUNC,
1397 MSR_IA32_ARCH_CAPABILITIES,
1398 MSR_IA32_PERF_CAPABILITIES,
1401 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1402 static unsigned int num_msr_based_features;
1404 static u64 kvm_get_arch_capabilities(void)
1408 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1409 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1412 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1413 * the nested hypervisor runs with NX huge pages. If it is not,
1414 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1415 * L1 guests, so it need not worry about its own (L2) guests.
1417 data |= ARCH_CAP_PSCHANGE_MC_NO;
1420 * If we're doing cache flushes (either "always" or "cond")
1421 * we will do one whenever the guest does a vmlaunch/vmresume.
1422 * If an outer hypervisor is doing the cache flush for us
1423 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1424 * capability to the guest too, and if EPT is disabled we're not
1425 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1426 * require a nested hypervisor to do a flush of its own.
1428 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1429 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1431 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1432 data |= ARCH_CAP_RDCL_NO;
1433 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1434 data |= ARCH_CAP_SSB_NO;
1435 if (!boot_cpu_has_bug(X86_BUG_MDS))
1436 data |= ARCH_CAP_MDS_NO;
1438 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1440 * If RTM=0 because the kernel has disabled TSX, the host might
1441 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1442 * and therefore knows that there cannot be TAA) but keep
1443 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1444 * and we want to allow migrating those guests to tsx=off hosts.
1446 data &= ~ARCH_CAP_TAA_NO;
1447 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1448 data |= ARCH_CAP_TAA_NO;
1451 * Nothing to do here; we emulate TSX_CTRL if present on the
1452 * host so the guest can choose between disabling TSX or
1453 * using VERW to clear CPU buffers.
1460 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1462 switch (msr->index) {
1463 case MSR_IA32_ARCH_CAPABILITIES:
1464 msr->data = kvm_get_arch_capabilities();
1466 case MSR_IA32_UCODE_REV:
1467 rdmsrl_safe(msr->index, &msr->data);
1470 return static_call(kvm_x86_get_msr_feature)(msr);
1475 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1477 struct kvm_msr_entry msr;
1481 r = kvm_get_msr_feature(&msr);
1483 if (r == KVM_MSR_RET_INVALID) {
1484 /* Unconditionally clear the output for simplicity */
1486 if (kvm_msr_ignored_check(index, 0, false))
1498 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1500 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1503 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1506 if (efer & (EFER_LME | EFER_LMA) &&
1507 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1510 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1516 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1518 if (efer & efer_reserved_bits)
1521 return __kvm_valid_efer(vcpu, efer);
1523 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1525 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1527 u64 old_efer = vcpu->arch.efer;
1528 u64 efer = msr_info->data;
1531 if (efer & efer_reserved_bits)
1534 if (!msr_info->host_initiated) {
1535 if (!__kvm_valid_efer(vcpu, efer))
1538 if (is_paging(vcpu) &&
1539 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1544 efer |= vcpu->arch.efer & EFER_LMA;
1546 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1552 /* Update reserved bits */
1553 if ((efer ^ old_efer) & EFER_NX)
1554 kvm_mmu_reset_context(vcpu);
1559 void kvm_enable_efer_bits(u64 mask)
1561 efer_reserved_bits &= ~mask;
1563 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1565 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1567 struct kvm_x86_msr_filter *msr_filter;
1568 struct msr_bitmap_range *ranges;
1569 struct kvm *kvm = vcpu->kvm;
1574 /* x2APIC MSRs do not support filtering. */
1575 if (index >= 0x800 && index <= 0x8ff)
1578 idx = srcu_read_lock(&kvm->srcu);
1580 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1586 allowed = msr_filter->default_allow;
1587 ranges = msr_filter->ranges;
1589 for (i = 0; i < msr_filter->count; i++) {
1590 u32 start = ranges[i].base;
1591 u32 end = start + ranges[i].nmsrs;
1592 u32 flags = ranges[i].flags;
1593 unsigned long *bitmap = ranges[i].bitmap;
1595 if ((index >= start) && (index < end) && (flags & type)) {
1596 allowed = !!test_bit(index - start, bitmap);
1602 srcu_read_unlock(&kvm->srcu, idx);
1606 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1609 * Write @data into the MSR specified by @index. Select MSR specific fault
1610 * checks are bypassed if @host_initiated is %true.
1611 * Returns 0 on success, non-0 otherwise.
1612 * Assumes vcpu_load() was already called.
1614 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1615 bool host_initiated)
1617 struct msr_data msr;
1619 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1620 return KVM_MSR_RET_FILTERED;
1625 case MSR_KERNEL_GS_BASE:
1628 if (is_noncanonical_address(data, vcpu))
1631 case MSR_IA32_SYSENTER_EIP:
1632 case MSR_IA32_SYSENTER_ESP:
1634 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1635 * non-canonical address is written on Intel but not on
1636 * AMD (which ignores the top 32-bits, because it does
1637 * not implement 64-bit SYSENTER).
1639 * 64-bit code should hence be able to write a non-canonical
1640 * value on AMD. Making the address canonical ensures that
1641 * vmentry does not fail on Intel after writing a non-canonical
1642 * value, and that something deterministic happens if the guest
1643 * invokes 64-bit SYSENTER.
1645 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1650 msr.host_initiated = host_initiated;
1652 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1655 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1656 u32 index, u64 data, bool host_initiated)
1658 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1660 if (ret == KVM_MSR_RET_INVALID)
1661 if (kvm_msr_ignored_check(index, data, true))
1668 * Read the MSR specified by @index into @data. Select MSR specific fault
1669 * checks are bypassed if @host_initiated is %true.
1670 * Returns 0 on success, non-0 otherwise.
1671 * Assumes vcpu_load() was already called.
1673 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1674 bool host_initiated)
1676 struct msr_data msr;
1679 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1680 return KVM_MSR_RET_FILTERED;
1683 msr.host_initiated = host_initiated;
1685 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1691 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1692 u32 index, u64 *data, bool host_initiated)
1694 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1696 if (ret == KVM_MSR_RET_INVALID) {
1697 /* Unconditionally clear *data for simplicity */
1699 if (kvm_msr_ignored_check(index, 0, false))
1706 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1708 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1710 EXPORT_SYMBOL_GPL(kvm_get_msr);
1712 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1714 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1716 EXPORT_SYMBOL_GPL(kvm_set_msr);
1718 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1720 int err = vcpu->run->msr.error;
1722 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1723 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1726 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1729 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1731 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1734 static u64 kvm_msr_reason(int r)
1737 case KVM_MSR_RET_INVALID:
1738 return KVM_MSR_EXIT_REASON_UNKNOWN;
1739 case KVM_MSR_RET_FILTERED:
1740 return KVM_MSR_EXIT_REASON_FILTER;
1742 return KVM_MSR_EXIT_REASON_INVAL;
1746 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1747 u32 exit_reason, u64 data,
1748 int (*completion)(struct kvm_vcpu *vcpu),
1751 u64 msr_reason = kvm_msr_reason(r);
1753 /* Check if the user wanted to know about this MSR fault */
1754 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1757 vcpu->run->exit_reason = exit_reason;
1758 vcpu->run->msr.error = 0;
1759 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1760 vcpu->run->msr.reason = msr_reason;
1761 vcpu->run->msr.index = index;
1762 vcpu->run->msr.data = data;
1763 vcpu->arch.complete_userspace_io = completion;
1768 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1770 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1771 complete_emulated_rdmsr, r);
1774 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1776 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1777 complete_emulated_wrmsr, r);
1780 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1782 u32 ecx = kvm_rcx_read(vcpu);
1786 r = kvm_get_msr(vcpu, ecx, &data);
1788 /* MSR read failed? See if we should ask user space */
1789 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1790 /* Bounce to user space */
1795 trace_kvm_msr_read(ecx, data);
1797 kvm_rax_write(vcpu, data & -1u);
1798 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1800 trace_kvm_msr_read_ex(ecx);
1803 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1805 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1807 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1809 u32 ecx = kvm_rcx_read(vcpu);
1810 u64 data = kvm_read_edx_eax(vcpu);
1813 r = kvm_set_msr(vcpu, ecx, data);
1815 /* MSR write failed? See if we should ask user space */
1816 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1817 /* Bounce to user space */
1820 /* Signal all other negative errors to userspace */
1825 trace_kvm_msr_write(ecx, data);
1827 trace_kvm_msr_write_ex(ecx, data);
1829 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1831 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1833 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1835 return kvm_skip_emulated_instruction(vcpu);
1837 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1839 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1841 /* Treat an INVD instruction as a NOP and just skip it. */
1842 return kvm_emulate_as_nop(vcpu);
1844 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1846 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1848 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1849 return kvm_emulate_as_nop(vcpu);
1851 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1853 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1855 kvm_queue_exception(vcpu, UD_VECTOR);
1858 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1860 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1862 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1863 return kvm_emulate_as_nop(vcpu);
1865 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1867 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1869 xfer_to_guest_mode_prepare();
1870 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1871 xfer_to_guest_mode_work_pending();
1875 * The fast path for frequent and performance sensitive wrmsr emulation,
1876 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1877 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1878 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1879 * other cases which must be called after interrupts are enabled on the host.
1881 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1883 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1886 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1887 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1888 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1889 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1892 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1893 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1894 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1895 trace_kvm_apic_write(APIC_ICR, (u32)data);
1902 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1904 if (!kvm_can_use_hv_timer(vcpu))
1907 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1911 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1913 u32 msr = kvm_rcx_read(vcpu);
1915 fastpath_t ret = EXIT_FASTPATH_NONE;
1918 case APIC_BASE_MSR + (APIC_ICR >> 4):
1919 data = kvm_read_edx_eax(vcpu);
1920 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1921 kvm_skip_emulated_instruction(vcpu);
1922 ret = EXIT_FASTPATH_EXIT_HANDLED;
1925 case MSR_IA32_TSC_DEADLINE:
1926 data = kvm_read_edx_eax(vcpu);
1927 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1928 kvm_skip_emulated_instruction(vcpu);
1929 ret = EXIT_FASTPATH_REENTER_GUEST;
1936 if (ret != EXIT_FASTPATH_NONE)
1937 trace_kvm_msr_write(msr, data);
1941 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1944 * Adapt set_msr() to msr_io()'s calling convention
1946 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1948 return kvm_get_msr_ignored_check(vcpu, index, data, true);
1951 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1953 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1956 #ifdef CONFIG_X86_64
1957 struct pvclock_clock {
1967 struct pvclock_gtod_data {
1970 struct pvclock_clock clock; /* extract of a clocksource struct */
1971 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1977 static struct pvclock_gtod_data pvclock_gtod_data;
1979 static void update_pvclock_gtod(struct timekeeper *tk)
1981 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1983 write_seqcount_begin(&vdata->seq);
1985 /* copy pvclock gtod data */
1986 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1987 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1988 vdata->clock.mask = tk->tkr_mono.mask;
1989 vdata->clock.mult = tk->tkr_mono.mult;
1990 vdata->clock.shift = tk->tkr_mono.shift;
1991 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1992 vdata->clock.offset = tk->tkr_mono.base;
1994 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1995 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1996 vdata->raw_clock.mask = tk->tkr_raw.mask;
1997 vdata->raw_clock.mult = tk->tkr_raw.mult;
1998 vdata->raw_clock.shift = tk->tkr_raw.shift;
1999 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2000 vdata->raw_clock.offset = tk->tkr_raw.base;
2002 vdata->wall_time_sec = tk->xtime_sec;
2004 vdata->offs_boot = tk->offs_boot;
2006 write_seqcount_end(&vdata->seq);
2009 static s64 get_kvmclock_base_ns(void)
2011 /* Count up from boot time, but with the frequency of the raw clock. */
2012 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2015 static s64 get_kvmclock_base_ns(void)
2017 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2018 return ktime_get_boottime_ns();
2022 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2026 struct pvclock_wall_clock wc;
2033 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2038 ++version; /* first time write, random junk */
2042 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2046 * The guest calculates current wall clock time by adding
2047 * system time (updated by kvm_guest_time_update below) to the
2048 * wall clock specified here. We do the reverse here.
2050 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2052 wc.nsec = do_div(wall_nsec, 1000000000);
2053 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2054 wc.version = version;
2056 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2059 wc_sec_hi = wall_nsec >> 32;
2060 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2061 &wc_sec_hi, sizeof(wc_sec_hi));
2065 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2068 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2069 bool old_msr, bool host_initiated)
2071 struct kvm_arch *ka = &vcpu->kvm->arch;
2073 if (vcpu->vcpu_id == 0 && !host_initiated) {
2074 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2075 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2077 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2080 vcpu->arch.time = system_time;
2081 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2083 /* we verify if the enable bit is set... */
2084 vcpu->arch.pv_time_enabled = false;
2085 if (!(system_time & 1))
2088 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2089 &vcpu->arch.pv_time, system_time & ~1ULL,
2090 sizeof(struct pvclock_vcpu_time_info)))
2091 vcpu->arch.pv_time_enabled = true;
2096 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2098 do_shl32_div32(dividend, divisor);
2102 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2103 s8 *pshift, u32 *pmultiplier)
2111 scaled64 = scaled_hz;
2112 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2117 tps32 = (uint32_t)tps64;
2118 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2119 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2127 *pmultiplier = div_frac(scaled64, tps32);
2130 #ifdef CONFIG_X86_64
2131 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2134 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2135 static unsigned long max_tsc_khz;
2137 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2139 u64 v = (u64)khz * (1000000 + ppm);
2144 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2148 /* Guest TSC same frequency as host TSC? */
2150 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2154 /* TSC scaling supported? */
2155 if (!kvm_has_tsc_control) {
2156 if (user_tsc_khz > tsc_khz) {
2157 vcpu->arch.tsc_catchup = 1;
2158 vcpu->arch.tsc_always_catchup = 1;
2161 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2166 /* TSC scaling required - calculate ratio */
2167 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2168 user_tsc_khz, tsc_khz);
2170 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2171 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2176 vcpu->arch.tsc_scaling_ratio = ratio;
2180 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2182 u32 thresh_lo, thresh_hi;
2183 int use_scaling = 0;
2185 /* tsc_khz can be zero if TSC calibration fails */
2186 if (user_tsc_khz == 0) {
2187 /* set tsc_scaling_ratio to a safe value */
2188 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2192 /* Compute a scale to convert nanoseconds in TSC cycles */
2193 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2194 &vcpu->arch.virtual_tsc_shift,
2195 &vcpu->arch.virtual_tsc_mult);
2196 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2199 * Compute the variation in TSC rate which is acceptable
2200 * within the range of tolerance and decide if the
2201 * rate being applied is within that bounds of the hardware
2202 * rate. If so, no scaling or compensation need be done.
2204 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2205 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2206 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2207 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2210 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2213 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2215 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2216 vcpu->arch.virtual_tsc_mult,
2217 vcpu->arch.virtual_tsc_shift);
2218 tsc += vcpu->arch.this_tsc_write;
2222 static inline int gtod_is_based_on_tsc(int mode)
2224 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2227 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2229 #ifdef CONFIG_X86_64
2231 struct kvm_arch *ka = &vcpu->kvm->arch;
2232 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2234 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2235 atomic_read(&vcpu->kvm->online_vcpus));
2238 * Once the masterclock is enabled, always perform request in
2239 * order to update it.
2241 * In order to enable masterclock, the host clocksource must be TSC
2242 * and the vcpus need to have matched TSCs. When that happens,
2243 * perform request to enable masterclock.
2245 if (ka->use_master_clock ||
2246 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2247 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2249 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2250 atomic_read(&vcpu->kvm->online_vcpus),
2251 ka->use_master_clock, gtod->clock.vclock_mode);
2256 * Multiply tsc by a fixed point number represented by ratio.
2258 * The most significant 64-N bits (mult) of ratio represent the
2259 * integral part of the fixed point number; the remaining N bits
2260 * (frac) represent the fractional part, ie. ratio represents a fixed
2261 * point number (mult + frac * 2^(-N)).
2263 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2265 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2267 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2270 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2273 u64 ratio = vcpu->arch.tsc_scaling_ratio;
2275 if (ratio != kvm_default_tsc_scaling_ratio)
2276 _tsc = __scale_tsc(ratio, tsc);
2280 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2282 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2286 tsc = kvm_scale_tsc(vcpu, rdtsc());
2288 return target_tsc - tsc;
2291 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2293 return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2295 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2297 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2299 vcpu->arch.l1_tsc_offset = offset;
2300 vcpu->arch.tsc_offset = static_call(kvm_x86_write_l1_tsc_offset)(vcpu, offset);
2303 static inline bool kvm_check_tsc_unstable(void)
2305 #ifdef CONFIG_X86_64
2307 * TSC is marked unstable when we're running on Hyper-V,
2308 * 'TSC page' clocksource is good.
2310 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2313 return check_tsc_unstable();
2316 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2318 struct kvm *kvm = vcpu->kvm;
2319 u64 offset, ns, elapsed;
2320 unsigned long flags;
2322 bool already_matched;
2323 bool synchronizing = false;
2325 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2326 offset = kvm_compute_tsc_offset(vcpu, data);
2327 ns = get_kvmclock_base_ns();
2328 elapsed = ns - kvm->arch.last_tsc_nsec;
2330 if (vcpu->arch.virtual_tsc_khz) {
2333 * detection of vcpu initialization -- need to sync
2334 * with other vCPUs. This particularly helps to keep
2335 * kvm_clock stable after CPU hotplug
2337 synchronizing = true;
2339 u64 tsc_exp = kvm->arch.last_tsc_write +
2340 nsec_to_cycles(vcpu, elapsed);
2341 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2343 * Special case: TSC write with a small delta (1 second)
2344 * of virtual cycle time against real time is
2345 * interpreted as an attempt to synchronize the CPU.
2347 synchronizing = data < tsc_exp + tsc_hz &&
2348 data + tsc_hz > tsc_exp;
2353 * For a reliable TSC, we can match TSC offsets, and for an unstable
2354 * TSC, we add elapsed time in this computation. We could let the
2355 * compensation code attempt to catch up if we fall behind, but
2356 * it's better to try to match offsets from the beginning.
2358 if (synchronizing &&
2359 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2360 if (!kvm_check_tsc_unstable()) {
2361 offset = kvm->arch.cur_tsc_offset;
2363 u64 delta = nsec_to_cycles(vcpu, elapsed);
2365 offset = kvm_compute_tsc_offset(vcpu, data);
2368 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2371 * We split periods of matched TSC writes into generations.
2372 * For each generation, we track the original measured
2373 * nanosecond time, offset, and write, so if TSCs are in
2374 * sync, we can match exact offset, and if not, we can match
2375 * exact software computation in compute_guest_tsc()
2377 * These values are tracked in kvm->arch.cur_xxx variables.
2379 kvm->arch.cur_tsc_generation++;
2380 kvm->arch.cur_tsc_nsec = ns;
2381 kvm->arch.cur_tsc_write = data;
2382 kvm->arch.cur_tsc_offset = offset;
2387 * We also track th most recent recorded KHZ, write and time to
2388 * allow the matching interval to be extended at each write.
2390 kvm->arch.last_tsc_nsec = ns;
2391 kvm->arch.last_tsc_write = data;
2392 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2394 vcpu->arch.last_guest_tsc = data;
2396 /* Keep track of which generation this VCPU has synchronized to */
2397 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2398 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2399 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2401 kvm_vcpu_write_tsc_offset(vcpu, offset);
2402 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2404 spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2406 kvm->arch.nr_vcpus_matched_tsc = 0;
2407 } else if (!already_matched) {
2408 kvm->arch.nr_vcpus_matched_tsc++;
2411 kvm_track_tsc_matching(vcpu);
2412 spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2415 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2418 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2419 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2422 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2424 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2425 WARN_ON(adjustment < 0);
2426 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2427 adjust_tsc_offset_guest(vcpu, adjustment);
2430 #ifdef CONFIG_X86_64
2432 static u64 read_tsc(void)
2434 u64 ret = (u64)rdtsc_ordered();
2435 u64 last = pvclock_gtod_data.clock.cycle_last;
2437 if (likely(ret >= last))
2441 * GCC likes to generate cmov here, but this branch is extremely
2442 * predictable (it's just a function of time and the likely is
2443 * very likely) and there's a data dependence, so force GCC
2444 * to generate a branch instead. I don't barrier() because
2445 * we don't actually need a barrier, and if this function
2446 * ever gets inlined it will generate worse code.
2452 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2458 switch (clock->vclock_mode) {
2459 case VDSO_CLOCKMODE_HVCLOCK:
2460 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2462 if (tsc_pg_val != U64_MAX) {
2463 /* TSC page valid */
2464 *mode = VDSO_CLOCKMODE_HVCLOCK;
2465 v = (tsc_pg_val - clock->cycle_last) &
2468 /* TSC page invalid */
2469 *mode = VDSO_CLOCKMODE_NONE;
2472 case VDSO_CLOCKMODE_TSC:
2473 *mode = VDSO_CLOCKMODE_TSC;
2474 *tsc_timestamp = read_tsc();
2475 v = (*tsc_timestamp - clock->cycle_last) &
2479 *mode = VDSO_CLOCKMODE_NONE;
2482 if (*mode == VDSO_CLOCKMODE_NONE)
2483 *tsc_timestamp = v = 0;
2485 return v * clock->mult;
2488 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2490 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2496 seq = read_seqcount_begin(>od->seq);
2497 ns = gtod->raw_clock.base_cycles;
2498 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2499 ns >>= gtod->raw_clock.shift;
2500 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2501 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2507 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2509 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2515 seq = read_seqcount_begin(>od->seq);
2516 ts->tv_sec = gtod->wall_time_sec;
2517 ns = gtod->clock.base_cycles;
2518 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2519 ns >>= gtod->clock.shift;
2520 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2522 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2528 /* returns true if host is using TSC based clocksource */
2529 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2531 /* checked again under seqlock below */
2532 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2535 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2539 /* returns true if host is using TSC based clocksource */
2540 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2543 /* checked again under seqlock below */
2544 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2547 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2553 * Assuming a stable TSC across physical CPUS, and a stable TSC
2554 * across virtual CPUs, the following condition is possible.
2555 * Each numbered line represents an event visible to both
2556 * CPUs at the next numbered event.
2558 * "timespecX" represents host monotonic time. "tscX" represents
2561 * VCPU0 on CPU0 | VCPU1 on CPU1
2563 * 1. read timespec0,tsc0
2564 * 2. | timespec1 = timespec0 + N
2566 * 3. transition to guest | transition to guest
2567 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2568 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2569 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2571 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2574 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2576 * - 0 < N - M => M < N
2578 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2579 * always the case (the difference between two distinct xtime instances
2580 * might be smaller then the difference between corresponding TSC reads,
2581 * when updating guest vcpus pvclock areas).
2583 * To avoid that problem, do not allow visibility of distinct
2584 * system_timestamp/tsc_timestamp values simultaneously: use a master
2585 * copy of host monotonic time values. Update that master copy
2588 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2592 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2594 #ifdef CONFIG_X86_64
2595 struct kvm_arch *ka = &kvm->arch;
2597 bool host_tsc_clocksource, vcpus_matched;
2599 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2600 atomic_read(&kvm->online_vcpus));
2603 * If the host uses TSC clock, then passthrough TSC as stable
2606 host_tsc_clocksource = kvm_get_time_and_clockread(
2607 &ka->master_kernel_ns,
2608 &ka->master_cycle_now);
2610 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2611 && !ka->backwards_tsc_observed
2612 && !ka->boot_vcpu_runs_old_kvmclock;
2614 if (ka->use_master_clock)
2615 atomic_set(&kvm_guest_has_master_clock, 1);
2617 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2618 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2623 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2625 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2628 static void kvm_gen_update_masterclock(struct kvm *kvm)
2630 #ifdef CONFIG_X86_64
2632 struct kvm_vcpu *vcpu;
2633 struct kvm_arch *ka = &kvm->arch;
2634 unsigned long flags;
2636 kvm_hv_invalidate_tsc_page(kvm);
2638 kvm_make_mclock_inprogress_request(kvm);
2640 /* no guest entries from this point */
2641 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2642 pvclock_update_vm_gtod_copy(kvm);
2643 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2645 kvm_for_each_vcpu(i, vcpu, kvm)
2646 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2648 /* guest entries allowed */
2649 kvm_for_each_vcpu(i, vcpu, kvm)
2650 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2654 u64 get_kvmclock_ns(struct kvm *kvm)
2656 struct kvm_arch *ka = &kvm->arch;
2657 struct pvclock_vcpu_time_info hv_clock;
2658 unsigned long flags;
2661 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2662 if (!ka->use_master_clock) {
2663 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2664 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2667 hv_clock.tsc_timestamp = ka->master_cycle_now;
2668 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2669 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2671 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2674 if (__this_cpu_read(cpu_tsc_khz)) {
2675 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2676 &hv_clock.tsc_shift,
2677 &hv_clock.tsc_to_system_mul);
2678 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2680 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2687 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2688 struct gfn_to_hva_cache *cache,
2689 unsigned int offset)
2691 struct kvm_vcpu_arch *vcpu = &v->arch;
2692 struct pvclock_vcpu_time_info guest_hv_clock;
2694 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2695 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2698 /* This VCPU is paused, but it's legal for a guest to read another
2699 * VCPU's kvmclock, so we really have to follow the specification where
2700 * it says that version is odd if data is being modified, and even after
2703 * Version field updates must be kept separate. This is because
2704 * kvm_write_guest_cached might use a "rep movs" instruction, and
2705 * writes within a string instruction are weakly ordered. So there
2706 * are three writes overall.
2708 * As a small optimization, only write the version field in the first
2709 * and third write. The vcpu->pv_time cache is still valid, because the
2710 * version field is the first in the struct.
2712 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2714 if (guest_hv_clock.version & 1)
2715 ++guest_hv_clock.version; /* first time write, random junk */
2717 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2718 kvm_write_guest_offset_cached(v->kvm, cache,
2719 &vcpu->hv_clock, offset,
2720 sizeof(vcpu->hv_clock.version));
2724 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2725 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2727 if (vcpu->pvclock_set_guest_stopped_request) {
2728 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2729 vcpu->pvclock_set_guest_stopped_request = false;
2732 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2734 kvm_write_guest_offset_cached(v->kvm, cache,
2735 &vcpu->hv_clock, offset,
2736 sizeof(vcpu->hv_clock));
2740 vcpu->hv_clock.version++;
2741 kvm_write_guest_offset_cached(v->kvm, cache,
2742 &vcpu->hv_clock, offset,
2743 sizeof(vcpu->hv_clock.version));
2746 static int kvm_guest_time_update(struct kvm_vcpu *v)
2748 unsigned long flags, tgt_tsc_khz;
2749 struct kvm_vcpu_arch *vcpu = &v->arch;
2750 struct kvm_arch *ka = &v->kvm->arch;
2752 u64 tsc_timestamp, host_tsc;
2754 bool use_master_clock;
2760 * If the host uses TSC clock, then passthrough TSC as stable
2763 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2764 use_master_clock = ka->use_master_clock;
2765 if (use_master_clock) {
2766 host_tsc = ka->master_cycle_now;
2767 kernel_ns = ka->master_kernel_ns;
2769 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2771 /* Keep irq disabled to prevent changes to the clock */
2772 local_irq_save(flags);
2773 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2774 if (unlikely(tgt_tsc_khz == 0)) {
2775 local_irq_restore(flags);
2776 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2779 if (!use_master_clock) {
2781 kernel_ns = get_kvmclock_base_ns();
2784 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2787 * We may have to catch up the TSC to match elapsed wall clock
2788 * time for two reasons, even if kvmclock is used.
2789 * 1) CPU could have been running below the maximum TSC rate
2790 * 2) Broken TSC compensation resets the base at each VCPU
2791 * entry to avoid unknown leaps of TSC even when running
2792 * again on the same CPU. This may cause apparent elapsed
2793 * time to disappear, and the guest to stand still or run
2796 if (vcpu->tsc_catchup) {
2797 u64 tsc = compute_guest_tsc(v, kernel_ns);
2798 if (tsc > tsc_timestamp) {
2799 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2800 tsc_timestamp = tsc;
2804 local_irq_restore(flags);
2806 /* With all the info we got, fill in the values */
2808 if (kvm_has_tsc_control)
2809 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2811 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2812 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2813 &vcpu->hv_clock.tsc_shift,
2814 &vcpu->hv_clock.tsc_to_system_mul);
2815 vcpu->hw_tsc_khz = tgt_tsc_khz;
2818 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2819 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2820 vcpu->last_guest_tsc = tsc_timestamp;
2822 /* If the host uses TSC clocksource, then it is stable */
2824 if (use_master_clock)
2825 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2827 vcpu->hv_clock.flags = pvclock_flags;
2829 if (vcpu->pv_time_enabled)
2830 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2831 if (vcpu->xen.vcpu_info_set)
2832 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2833 offsetof(struct compat_vcpu_info, time));
2834 if (vcpu->xen.vcpu_time_info_set)
2835 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2836 if (v == kvm_get_vcpu(v->kvm, 0))
2837 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2842 * kvmclock updates which are isolated to a given vcpu, such as
2843 * vcpu->cpu migration, should not allow system_timestamp from
2844 * the rest of the vcpus to remain static. Otherwise ntp frequency
2845 * correction applies to one vcpu's system_timestamp but not
2848 * So in those cases, request a kvmclock update for all vcpus.
2849 * We need to rate-limit these requests though, as they can
2850 * considerably slow guests that have a large number of vcpus.
2851 * The time for a remote vcpu to update its kvmclock is bound
2852 * by the delay we use to rate-limit the updates.
2855 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2857 static void kvmclock_update_fn(struct work_struct *work)
2860 struct delayed_work *dwork = to_delayed_work(work);
2861 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2862 kvmclock_update_work);
2863 struct kvm *kvm = container_of(ka, struct kvm, arch);
2864 struct kvm_vcpu *vcpu;
2866 kvm_for_each_vcpu(i, vcpu, kvm) {
2867 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2868 kvm_vcpu_kick(vcpu);
2872 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2874 struct kvm *kvm = v->kvm;
2876 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2877 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2878 KVMCLOCK_UPDATE_DELAY);
2881 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2883 static void kvmclock_sync_fn(struct work_struct *work)
2885 struct delayed_work *dwork = to_delayed_work(work);
2886 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2887 kvmclock_sync_work);
2888 struct kvm *kvm = container_of(ka, struct kvm, arch);
2890 if (!kvmclock_periodic_sync)
2893 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2894 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2895 KVMCLOCK_SYNC_PERIOD);
2899 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2901 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2903 /* McStatusWrEn enabled? */
2904 if (guest_cpuid_is_amd_or_hygon(vcpu))
2905 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2910 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2912 u64 mcg_cap = vcpu->arch.mcg_cap;
2913 unsigned bank_num = mcg_cap & 0xff;
2914 u32 msr = msr_info->index;
2915 u64 data = msr_info->data;
2918 case MSR_IA32_MCG_STATUS:
2919 vcpu->arch.mcg_status = data;
2921 case MSR_IA32_MCG_CTL:
2922 if (!(mcg_cap & MCG_CTL_P) &&
2923 (data || !msr_info->host_initiated))
2925 if (data != 0 && data != ~(u64)0)
2927 vcpu->arch.mcg_ctl = data;
2930 if (msr >= MSR_IA32_MC0_CTL &&
2931 msr < MSR_IA32_MCx_CTL(bank_num)) {
2932 u32 offset = array_index_nospec(
2933 msr - MSR_IA32_MC0_CTL,
2934 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2936 /* only 0 or all 1s can be written to IA32_MCi_CTL
2937 * some Linux kernels though clear bit 10 in bank 4 to
2938 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2939 * this to avoid an uncatched #GP in the guest
2941 if ((offset & 0x3) == 0 &&
2942 data != 0 && (data | (1 << 10)) != ~(u64)0)
2946 if (!msr_info->host_initiated &&
2947 (offset & 0x3) == 1 && data != 0) {
2948 if (!can_set_mci_status(vcpu))
2952 vcpu->arch.mce_banks[offset] = data;
2960 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2962 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2964 return (vcpu->arch.apf.msr_en_val & mask) == mask;
2967 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2969 gpa_t gpa = data & ~0x3f;
2971 /* Bits 4:5 are reserved, Should be zero */
2975 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
2976 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
2979 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
2980 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
2983 if (!lapic_in_kernel(vcpu))
2984 return data ? 1 : 0;
2986 vcpu->arch.apf.msr_en_val = data;
2988 if (!kvm_pv_async_pf_enabled(vcpu)) {
2989 kvm_clear_async_pf_completion_queue(vcpu);
2990 kvm_async_pf_hash_reset(vcpu);
2994 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2998 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2999 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3001 kvm_async_pf_wakeup_all(vcpu);
3006 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3008 /* Bits 8-63 are reserved */
3012 if (!lapic_in_kernel(vcpu))
3015 vcpu->arch.apf.msr_int_val = data;
3017 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3022 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3024 vcpu->arch.pv_time_enabled = false;
3025 vcpu->arch.time = 0;
3028 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3030 ++vcpu->stat.tlb_flush;
3031 static_call(kvm_x86_tlb_flush_all)(vcpu);
3034 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3036 ++vcpu->stat.tlb_flush;
3037 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3040 static void record_steal_time(struct kvm_vcpu *vcpu)
3042 struct kvm_host_map map;
3043 struct kvm_steal_time *st;
3045 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3046 kvm_xen_runstate_set_running(vcpu);
3050 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3053 /* -EAGAIN is returned in atomic context so we can just return. */
3054 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3055 &map, &vcpu->arch.st.cache, false))
3059 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3062 * Doing a TLB flush here, on the guest's behalf, can avoid
3065 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3066 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3067 st->preempted & KVM_VCPU_FLUSH_TLB);
3068 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
3069 kvm_vcpu_flush_tlb_guest(vcpu);
3072 vcpu->arch.st.preempted = 0;
3074 if (st->version & 1)
3075 st->version += 1; /* first time write, random junk */
3081 st->steal += current->sched_info.run_delay -
3082 vcpu->arch.st.last_steal;
3083 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3089 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3092 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3095 u32 msr = msr_info->index;
3096 u64 data = msr_info->data;
3098 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3099 return kvm_xen_write_hypercall_page(vcpu, data);
3102 case MSR_AMD64_NB_CFG:
3103 case MSR_IA32_UCODE_WRITE:
3104 case MSR_VM_HSAVE_PA:
3105 case MSR_AMD64_PATCH_LOADER:
3106 case MSR_AMD64_BU_CFG2:
3107 case MSR_AMD64_DC_CFG:
3108 case MSR_F15H_EX_CFG:
3111 case MSR_IA32_UCODE_REV:
3112 if (msr_info->host_initiated)
3113 vcpu->arch.microcode_version = data;
3115 case MSR_IA32_ARCH_CAPABILITIES:
3116 if (!msr_info->host_initiated)
3118 vcpu->arch.arch_capabilities = data;
3120 case MSR_IA32_PERF_CAPABILITIES: {
3121 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3123 if (!msr_info->host_initiated)
3125 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3127 if (data & ~msr_ent.data)
3130 vcpu->arch.perf_capabilities = data;
3135 return set_efer(vcpu, msr_info);
3137 data &= ~(u64)0x40; /* ignore flush filter disable */
3138 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3139 data &= ~(u64)0x8; /* ignore TLB cache disable */
3141 /* Handle McStatusWrEn */
3142 if (data == BIT_ULL(18)) {
3143 vcpu->arch.msr_hwcr = data;
3144 } else if (data != 0) {
3145 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3150 case MSR_FAM10H_MMIO_CONF_BASE:
3152 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3157 case 0x200 ... 0x2ff:
3158 return kvm_mtrr_set_msr(vcpu, msr, data);
3159 case MSR_IA32_APICBASE:
3160 return kvm_set_apic_base(vcpu, msr_info);
3161 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3162 return kvm_x2apic_msr_write(vcpu, msr, data);
3163 case MSR_IA32_TSC_DEADLINE:
3164 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3166 case MSR_IA32_TSC_ADJUST:
3167 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3168 if (!msr_info->host_initiated) {
3169 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3170 adjust_tsc_offset_guest(vcpu, adj);
3172 vcpu->arch.ia32_tsc_adjust_msr = data;
3175 case MSR_IA32_MISC_ENABLE:
3176 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3177 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3178 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3180 vcpu->arch.ia32_misc_enable_msr = data;
3181 kvm_update_cpuid_runtime(vcpu);
3183 vcpu->arch.ia32_misc_enable_msr = data;
3186 case MSR_IA32_SMBASE:
3187 if (!msr_info->host_initiated)
3189 vcpu->arch.smbase = data;
3191 case MSR_IA32_POWER_CTL:
3192 vcpu->arch.msr_ia32_power_ctl = data;
3195 if (msr_info->host_initiated) {
3196 kvm_synchronize_tsc(vcpu, data);
3198 u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3199 adjust_tsc_offset_guest(vcpu, adj);
3200 vcpu->arch.ia32_tsc_adjust_msr += adj;
3204 if (!msr_info->host_initiated &&
3205 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3208 * KVM supports exposing PT to the guest, but does not support
3209 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3210 * XSAVES/XRSTORS to save/restore PT MSRs.
3212 if (data & ~supported_xss)
3214 vcpu->arch.ia32_xss = data;
3217 if (!msr_info->host_initiated)
3219 vcpu->arch.smi_count = data;
3221 case MSR_KVM_WALL_CLOCK_NEW:
3222 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3225 vcpu->kvm->arch.wall_clock = data;
3226 kvm_write_wall_clock(vcpu->kvm, data, 0);
3228 case MSR_KVM_WALL_CLOCK:
3229 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3232 vcpu->kvm->arch.wall_clock = data;
3233 kvm_write_wall_clock(vcpu->kvm, data, 0);
3235 case MSR_KVM_SYSTEM_TIME_NEW:
3236 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3239 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3241 case MSR_KVM_SYSTEM_TIME:
3242 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3245 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3247 case MSR_KVM_ASYNC_PF_EN:
3248 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3251 if (kvm_pv_enable_async_pf(vcpu, data))
3254 case MSR_KVM_ASYNC_PF_INT:
3255 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3258 if (kvm_pv_enable_async_pf_int(vcpu, data))
3261 case MSR_KVM_ASYNC_PF_ACK:
3262 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3265 vcpu->arch.apf.pageready_pending = false;
3266 kvm_check_async_pf_completion(vcpu);
3269 case MSR_KVM_STEAL_TIME:
3270 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3273 if (unlikely(!sched_info_on()))
3276 if (data & KVM_STEAL_RESERVED_MASK)
3279 vcpu->arch.st.msr_val = data;
3281 if (!(data & KVM_MSR_ENABLED))
3284 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3287 case MSR_KVM_PV_EOI_EN:
3288 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3291 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3295 case MSR_KVM_POLL_CONTROL:
3296 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3299 /* only enable bit supported */
3300 if (data & (-1ULL << 1))
3303 vcpu->arch.msr_kvm_poll_control = data;
3306 case MSR_IA32_MCG_CTL:
3307 case MSR_IA32_MCG_STATUS:
3308 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3309 return set_msr_mce(vcpu, msr_info);
3311 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3312 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3315 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3316 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3317 if (kvm_pmu_is_valid_msr(vcpu, msr))
3318 return kvm_pmu_set_msr(vcpu, msr_info);
3320 if (pr || data != 0)
3321 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3322 "0x%x data 0x%llx\n", msr, data);
3324 case MSR_K7_CLK_CTL:
3326 * Ignore all writes to this no longer documented MSR.
3327 * Writes are only relevant for old K7 processors,
3328 * all pre-dating SVM, but a recommended workaround from
3329 * AMD for these chips. It is possible to specify the
3330 * affected processor models on the command line, hence
3331 * the need to ignore the workaround.
3334 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3335 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3336 case HV_X64_MSR_SYNDBG_OPTIONS:
3337 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3338 case HV_X64_MSR_CRASH_CTL:
3339 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3340 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3341 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3342 case HV_X64_MSR_TSC_EMULATION_STATUS:
3343 return kvm_hv_set_msr_common(vcpu, msr, data,
3344 msr_info->host_initiated);
3345 case MSR_IA32_BBL_CR_CTL3:
3346 /* Drop writes to this legacy MSR -- see rdmsr
3347 * counterpart for further detail.
3349 if (report_ignored_msrs)
3350 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3353 case MSR_AMD64_OSVW_ID_LENGTH:
3354 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3356 vcpu->arch.osvw.length = data;
3358 case MSR_AMD64_OSVW_STATUS:
3359 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3361 vcpu->arch.osvw.status = data;
3363 case MSR_PLATFORM_INFO:
3364 if (!msr_info->host_initiated ||
3365 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3366 cpuid_fault_enabled(vcpu)))
3368 vcpu->arch.msr_platform_info = data;
3370 case MSR_MISC_FEATURES_ENABLES:
3371 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3372 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3373 !supports_cpuid_fault(vcpu)))
3375 vcpu->arch.msr_misc_features_enables = data;
3378 if (kvm_pmu_is_valid_msr(vcpu, msr))
3379 return kvm_pmu_set_msr(vcpu, msr_info);
3380 return KVM_MSR_RET_INVALID;
3384 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3386 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3389 u64 mcg_cap = vcpu->arch.mcg_cap;
3390 unsigned bank_num = mcg_cap & 0xff;
3393 case MSR_IA32_P5_MC_ADDR:
3394 case MSR_IA32_P5_MC_TYPE:
3397 case MSR_IA32_MCG_CAP:
3398 data = vcpu->arch.mcg_cap;
3400 case MSR_IA32_MCG_CTL:
3401 if (!(mcg_cap & MCG_CTL_P) && !host)
3403 data = vcpu->arch.mcg_ctl;
3405 case MSR_IA32_MCG_STATUS:
3406 data = vcpu->arch.mcg_status;
3409 if (msr >= MSR_IA32_MC0_CTL &&
3410 msr < MSR_IA32_MCx_CTL(bank_num)) {
3411 u32 offset = array_index_nospec(
3412 msr - MSR_IA32_MC0_CTL,
3413 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3415 data = vcpu->arch.mce_banks[offset];
3424 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3426 switch (msr_info->index) {
3427 case MSR_IA32_PLATFORM_ID:
3428 case MSR_IA32_EBL_CR_POWERON:
3429 case MSR_IA32_LASTBRANCHFROMIP:
3430 case MSR_IA32_LASTBRANCHTOIP:
3431 case MSR_IA32_LASTINTFROMIP:
3432 case MSR_IA32_LASTINTTOIP:
3434 case MSR_K8_TSEG_ADDR:
3435 case MSR_K8_TSEG_MASK:
3436 case MSR_VM_HSAVE_PA:
3437 case MSR_K8_INT_PENDING_MSG:
3438 case MSR_AMD64_NB_CFG:
3439 case MSR_FAM10H_MMIO_CONF_BASE:
3440 case MSR_AMD64_BU_CFG2:
3441 case MSR_IA32_PERF_CTL:
3442 case MSR_AMD64_DC_CFG:
3443 case MSR_F15H_EX_CFG:
3445 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3446 * limit) MSRs. Just return 0, as we do not want to expose the host
3447 * data here. Do not conditionalize this on CPUID, as KVM does not do
3448 * so for existing CPU-specific MSRs.
3450 case MSR_RAPL_POWER_UNIT:
3451 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3452 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3453 case MSR_PKG_ENERGY_STATUS: /* Total package */
3454 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3457 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3458 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3459 return kvm_pmu_get_msr(vcpu, msr_info);
3460 if (!msr_info->host_initiated)
3464 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3465 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3466 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3467 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3468 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3469 return kvm_pmu_get_msr(vcpu, msr_info);
3472 case MSR_IA32_UCODE_REV:
3473 msr_info->data = vcpu->arch.microcode_version;
3475 case MSR_IA32_ARCH_CAPABILITIES:
3476 if (!msr_info->host_initiated &&
3477 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3479 msr_info->data = vcpu->arch.arch_capabilities;
3481 case MSR_IA32_PERF_CAPABILITIES:
3482 if (!msr_info->host_initiated &&
3483 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3485 msr_info->data = vcpu->arch.perf_capabilities;
3487 case MSR_IA32_POWER_CTL:
3488 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3490 case MSR_IA32_TSC: {
3492 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3493 * even when not intercepted. AMD manual doesn't explicitly
3494 * state this but appears to behave the same.
3496 * On userspace reads and writes, however, we unconditionally
3497 * return L1's TSC value to ensure backwards-compatible
3498 * behavior for migration.
3500 u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3501 vcpu->arch.tsc_offset;
3503 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3507 case 0x200 ... 0x2ff:
3508 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3509 case 0xcd: /* fsb frequency */
3513 * MSR_EBC_FREQUENCY_ID
3514 * Conservative value valid for even the basic CPU models.
3515 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3516 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3517 * and 266MHz for model 3, or 4. Set Core Clock
3518 * Frequency to System Bus Frequency Ratio to 1 (bits
3519 * 31:24) even though these are only valid for CPU
3520 * models > 2, however guests may end up dividing or
3521 * multiplying by zero otherwise.
3523 case MSR_EBC_FREQUENCY_ID:
3524 msr_info->data = 1 << 24;
3526 case MSR_IA32_APICBASE:
3527 msr_info->data = kvm_get_apic_base(vcpu);
3529 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3530 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3531 case MSR_IA32_TSC_DEADLINE:
3532 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3534 case MSR_IA32_TSC_ADJUST:
3535 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3537 case MSR_IA32_MISC_ENABLE:
3538 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3540 case MSR_IA32_SMBASE:
3541 if (!msr_info->host_initiated)
3543 msr_info->data = vcpu->arch.smbase;
3546 msr_info->data = vcpu->arch.smi_count;
3548 case MSR_IA32_PERF_STATUS:
3549 /* TSC increment by tick */
3550 msr_info->data = 1000ULL;
3551 /* CPU multiplier */
3552 msr_info->data |= (((uint64_t)4ULL) << 40);
3555 msr_info->data = vcpu->arch.efer;
3557 case MSR_KVM_WALL_CLOCK:
3558 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3561 msr_info->data = vcpu->kvm->arch.wall_clock;
3563 case MSR_KVM_WALL_CLOCK_NEW:
3564 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3567 msr_info->data = vcpu->kvm->arch.wall_clock;
3569 case MSR_KVM_SYSTEM_TIME:
3570 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3573 msr_info->data = vcpu->arch.time;
3575 case MSR_KVM_SYSTEM_TIME_NEW:
3576 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3579 msr_info->data = vcpu->arch.time;
3581 case MSR_KVM_ASYNC_PF_EN:
3582 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3585 msr_info->data = vcpu->arch.apf.msr_en_val;
3587 case MSR_KVM_ASYNC_PF_INT:
3588 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3591 msr_info->data = vcpu->arch.apf.msr_int_val;
3593 case MSR_KVM_ASYNC_PF_ACK:
3594 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3599 case MSR_KVM_STEAL_TIME:
3600 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3603 msr_info->data = vcpu->arch.st.msr_val;
3605 case MSR_KVM_PV_EOI_EN:
3606 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3609 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3611 case MSR_KVM_POLL_CONTROL:
3612 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3615 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3617 case MSR_IA32_P5_MC_ADDR:
3618 case MSR_IA32_P5_MC_TYPE:
3619 case MSR_IA32_MCG_CAP:
3620 case MSR_IA32_MCG_CTL:
3621 case MSR_IA32_MCG_STATUS:
3622 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3623 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3624 msr_info->host_initiated);
3626 if (!msr_info->host_initiated &&
3627 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3629 msr_info->data = vcpu->arch.ia32_xss;
3631 case MSR_K7_CLK_CTL:
3633 * Provide expected ramp-up count for K7. All other
3634 * are set to zero, indicating minimum divisors for
3637 * This prevents guest kernels on AMD host with CPU
3638 * type 6, model 8 and higher from exploding due to
3639 * the rdmsr failing.
3641 msr_info->data = 0x20000000;
3643 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3644 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3645 case HV_X64_MSR_SYNDBG_OPTIONS:
3646 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3647 case HV_X64_MSR_CRASH_CTL:
3648 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3649 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3650 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3651 case HV_X64_MSR_TSC_EMULATION_STATUS:
3652 return kvm_hv_get_msr_common(vcpu,
3653 msr_info->index, &msr_info->data,
3654 msr_info->host_initiated);
3655 case MSR_IA32_BBL_CR_CTL3:
3656 /* This legacy MSR exists but isn't fully documented in current
3657 * silicon. It is however accessed by winxp in very narrow
3658 * scenarios where it sets bit #19, itself documented as
3659 * a "reserved" bit. Best effort attempt to source coherent
3660 * read data here should the balance of the register be
3661 * interpreted by the guest:
3663 * L2 cache control register 3: 64GB range, 256KB size,
3664 * enabled, latency 0x1, configured
3666 msr_info->data = 0xbe702111;
3668 case MSR_AMD64_OSVW_ID_LENGTH:
3669 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3671 msr_info->data = vcpu->arch.osvw.length;
3673 case MSR_AMD64_OSVW_STATUS:
3674 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3676 msr_info->data = vcpu->arch.osvw.status;
3678 case MSR_PLATFORM_INFO:
3679 if (!msr_info->host_initiated &&
3680 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3682 msr_info->data = vcpu->arch.msr_platform_info;
3684 case MSR_MISC_FEATURES_ENABLES:
3685 msr_info->data = vcpu->arch.msr_misc_features_enables;
3688 msr_info->data = vcpu->arch.msr_hwcr;
3691 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3692 return kvm_pmu_get_msr(vcpu, msr_info);
3693 return KVM_MSR_RET_INVALID;
3697 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3700 * Read or write a bunch of msrs. All parameters are kernel addresses.
3702 * @return number of msrs set successfully.
3704 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3705 struct kvm_msr_entry *entries,
3706 int (*do_msr)(struct kvm_vcpu *vcpu,
3707 unsigned index, u64 *data))
3711 for (i = 0; i < msrs->nmsrs; ++i)
3712 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3719 * Read or write a bunch of msrs. Parameters are user addresses.
3721 * @return number of msrs set successfully.
3723 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3724 int (*do_msr)(struct kvm_vcpu *vcpu,
3725 unsigned index, u64 *data),
3728 struct kvm_msrs msrs;
3729 struct kvm_msr_entry *entries;
3734 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3738 if (msrs.nmsrs >= MAX_IO_MSRS)
3741 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3742 entries = memdup_user(user_msrs->entries, size);
3743 if (IS_ERR(entries)) {
3744 r = PTR_ERR(entries);
3748 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3753 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3764 static inline bool kvm_can_mwait_in_guest(void)
3766 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3767 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3768 boot_cpu_has(X86_FEATURE_ARAT);
3771 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3772 struct kvm_cpuid2 __user *cpuid_arg)
3774 struct kvm_cpuid2 cpuid;
3778 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3781 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3786 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3792 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3797 case KVM_CAP_IRQCHIP:
3799 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3800 case KVM_CAP_SET_TSS_ADDR:
3801 case KVM_CAP_EXT_CPUID:
3802 case KVM_CAP_EXT_EMUL_CPUID:
3803 case KVM_CAP_CLOCKSOURCE:
3805 case KVM_CAP_NOP_IO_DELAY:
3806 case KVM_CAP_MP_STATE:
3807 case KVM_CAP_SYNC_MMU:
3808 case KVM_CAP_USER_NMI:
3809 case KVM_CAP_REINJECT_CONTROL:
3810 case KVM_CAP_IRQ_INJECT_STATUS:
3811 case KVM_CAP_IOEVENTFD:
3812 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3814 case KVM_CAP_PIT_STATE2:
3815 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3816 case KVM_CAP_VCPU_EVENTS:
3817 case KVM_CAP_HYPERV:
3818 case KVM_CAP_HYPERV_VAPIC:
3819 case KVM_CAP_HYPERV_SPIN:
3820 case KVM_CAP_HYPERV_SYNIC:
3821 case KVM_CAP_HYPERV_SYNIC2:
3822 case KVM_CAP_HYPERV_VP_INDEX:
3823 case KVM_CAP_HYPERV_EVENTFD:
3824 case KVM_CAP_HYPERV_TLBFLUSH:
3825 case KVM_CAP_HYPERV_SEND_IPI:
3826 case KVM_CAP_HYPERV_CPUID:
3827 case KVM_CAP_SYS_HYPERV_CPUID:
3828 case KVM_CAP_PCI_SEGMENT:
3829 case KVM_CAP_DEBUGREGS:
3830 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3832 case KVM_CAP_ASYNC_PF:
3833 case KVM_CAP_ASYNC_PF_INT:
3834 case KVM_CAP_GET_TSC_KHZ:
3835 case KVM_CAP_KVMCLOCK_CTRL:
3836 case KVM_CAP_READONLY_MEM:
3837 case KVM_CAP_HYPERV_TIME:
3838 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3839 case KVM_CAP_TSC_DEADLINE_TIMER:
3840 case KVM_CAP_DISABLE_QUIRKS:
3841 case KVM_CAP_SET_BOOT_CPU_ID:
3842 case KVM_CAP_SPLIT_IRQCHIP:
3843 case KVM_CAP_IMMEDIATE_EXIT:
3844 case KVM_CAP_PMU_EVENT_FILTER:
3845 case KVM_CAP_GET_MSR_FEATURES:
3846 case KVM_CAP_MSR_PLATFORM_INFO:
3847 case KVM_CAP_EXCEPTION_PAYLOAD:
3848 case KVM_CAP_SET_GUEST_DEBUG:
3849 case KVM_CAP_LAST_CPU:
3850 case KVM_CAP_X86_USER_SPACE_MSR:
3851 case KVM_CAP_X86_MSR_FILTER:
3852 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3853 #ifdef CONFIG_X86_SGX_KVM
3854 case KVM_CAP_SGX_ATTRIBUTE:
3856 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
3859 case KVM_CAP_SET_GUEST_DEBUG2:
3860 return KVM_GUESTDBG_VALID_MASK;
3861 #ifdef CONFIG_KVM_XEN
3862 case KVM_CAP_XEN_HVM:
3863 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
3864 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
3865 KVM_XEN_HVM_CONFIG_SHARED_INFO;
3866 if (sched_info_on())
3867 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
3870 case KVM_CAP_SYNC_REGS:
3871 r = KVM_SYNC_X86_VALID_FIELDS;
3873 case KVM_CAP_ADJUST_CLOCK:
3874 r = KVM_CLOCK_TSC_STABLE;
3876 case KVM_CAP_X86_DISABLE_EXITS:
3877 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3878 KVM_X86_DISABLE_EXITS_CSTATE;
3879 if(kvm_can_mwait_in_guest())
3880 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3882 case KVM_CAP_X86_SMM:
3883 /* SMBASE is usually relocated above 1M on modern chipsets,
3884 * and SMM handlers might indeed rely on 4G segment limits,
3885 * so do not report SMM to be available if real mode is
3886 * emulated via vm86 mode. Still, do not go to great lengths
3887 * to avoid userspace's usage of the feature, because it is a
3888 * fringe case that is not enabled except via specific settings
3889 * of the module parameters.
3891 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
3894 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
3896 case KVM_CAP_NR_VCPUS:
3897 r = KVM_SOFT_MAX_VCPUS;
3899 case KVM_CAP_MAX_VCPUS:
3902 case KVM_CAP_MAX_VCPU_ID:
3903 r = KVM_MAX_VCPU_ID;
3905 case KVM_CAP_PV_MMU: /* obsolete */
3909 r = KVM_MAX_MCE_BANKS;
3912 r = boot_cpu_has(X86_FEATURE_XSAVE);
3914 case KVM_CAP_TSC_CONTROL:
3915 r = kvm_has_tsc_control;
3917 case KVM_CAP_X2APIC_API:
3918 r = KVM_X2APIC_API_VALID_FLAGS;
3920 case KVM_CAP_NESTED_STATE:
3921 r = kvm_x86_ops.nested_ops->get_state ?
3922 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3924 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3925 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3927 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3928 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3930 case KVM_CAP_SMALLER_MAXPHYADDR:
3931 r = (int) allow_smaller_maxphyaddr;
3933 case KVM_CAP_STEAL_TIME:
3934 r = sched_info_on();
3936 case KVM_CAP_X86_BUS_LOCK_EXIT:
3937 if (kvm_has_bus_lock_exit)
3938 r = KVM_BUS_LOCK_DETECTION_OFF |
3939 KVM_BUS_LOCK_DETECTION_EXIT;
3950 long kvm_arch_dev_ioctl(struct file *filp,
3951 unsigned int ioctl, unsigned long arg)
3953 void __user *argp = (void __user *)arg;
3957 case KVM_GET_MSR_INDEX_LIST: {
3958 struct kvm_msr_list __user *user_msr_list = argp;
3959 struct kvm_msr_list msr_list;
3963 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3966 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3967 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3970 if (n < msr_list.nmsrs)
3973 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3974 num_msrs_to_save * sizeof(u32)))
3976 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3978 num_emulated_msrs * sizeof(u32)))
3983 case KVM_GET_SUPPORTED_CPUID:
3984 case KVM_GET_EMULATED_CPUID: {
3985 struct kvm_cpuid2 __user *cpuid_arg = argp;
3986 struct kvm_cpuid2 cpuid;
3989 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3992 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3998 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4003 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4005 if (copy_to_user(argp, &kvm_mce_cap_supported,
4006 sizeof(kvm_mce_cap_supported)))
4010 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4011 struct kvm_msr_list __user *user_msr_list = argp;
4012 struct kvm_msr_list msr_list;
4016 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4019 msr_list.nmsrs = num_msr_based_features;
4020 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4023 if (n < msr_list.nmsrs)
4026 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4027 num_msr_based_features * sizeof(u32)))
4033 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4035 case KVM_GET_SUPPORTED_HV_CPUID:
4036 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4046 static void wbinvd_ipi(void *garbage)
4051 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4053 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4056 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4058 /* Address WBINVD may be executed by guest */
4059 if (need_emulate_wbinvd(vcpu)) {
4060 if (static_call(kvm_x86_has_wbinvd_exit)())
4061 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4062 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4063 smp_call_function_single(vcpu->cpu,
4064 wbinvd_ipi, NULL, 1);
4067 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4069 /* Save host pkru register if supported */
4070 vcpu->arch.host_pkru = read_pkru();
4072 /* Apply any externally detected TSC adjustments (due to suspend) */
4073 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4074 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4075 vcpu->arch.tsc_offset_adjustment = 0;
4076 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4079 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4080 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4081 rdtsc() - vcpu->arch.last_host_tsc;
4083 mark_tsc_unstable("KVM discovered backwards TSC");
4085 if (kvm_check_tsc_unstable()) {
4086 u64 offset = kvm_compute_tsc_offset(vcpu,
4087 vcpu->arch.last_guest_tsc);
4088 kvm_vcpu_write_tsc_offset(vcpu, offset);
4089 vcpu->arch.tsc_catchup = 1;
4092 if (kvm_lapic_hv_timer_in_use(vcpu))
4093 kvm_lapic_restart_hv_timer(vcpu);
4096 * On a host with synchronized TSC, there is no need to update
4097 * kvmclock on vcpu->cpu migration
4099 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4100 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4101 if (vcpu->cpu != cpu)
4102 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4106 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4109 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4111 struct kvm_host_map map;
4112 struct kvm_steal_time *st;
4114 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4117 if (vcpu->arch.st.preempted)
4120 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4121 &vcpu->arch.st.cache, true))
4125 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4127 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4129 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4132 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4136 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4137 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4140 * Take the srcu lock as memslots will be accessed to check the gfn
4141 * cache generation against the memslots generation.
4143 idx = srcu_read_lock(&vcpu->kvm->srcu);
4144 if (kvm_xen_msr_enabled(vcpu->kvm))
4145 kvm_xen_runstate_set_preempted(vcpu);
4147 kvm_steal_time_set_preempted(vcpu);
4148 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4150 static_call(kvm_x86_vcpu_put)(vcpu);
4151 vcpu->arch.last_host_tsc = rdtsc();
4153 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4154 * on every vmexit, but if not, we might have a stale dr6 from the
4155 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4160 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4161 struct kvm_lapic_state *s)
4163 if (vcpu->arch.apicv_active)
4164 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4166 return kvm_apic_get_state(vcpu, s);
4169 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4170 struct kvm_lapic_state *s)
4174 r = kvm_apic_set_state(vcpu, s);
4177 update_cr8_intercept(vcpu);
4182 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4185 * We can accept userspace's request for interrupt injection
4186 * as long as we have a place to store the interrupt number.
4187 * The actual injection will happen when the CPU is able to
4188 * deliver the interrupt.
4190 if (kvm_cpu_has_extint(vcpu))
4193 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4194 return (!lapic_in_kernel(vcpu) ||
4195 kvm_apic_accept_pic_intr(vcpu));
4198 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4200 return kvm_arch_interrupt_allowed(vcpu) &&
4201 kvm_cpu_accept_dm_intr(vcpu);
4204 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4205 struct kvm_interrupt *irq)
4207 if (irq->irq >= KVM_NR_INTERRUPTS)
4210 if (!irqchip_in_kernel(vcpu->kvm)) {
4211 kvm_queue_interrupt(vcpu, irq->irq, false);
4212 kvm_make_request(KVM_REQ_EVENT, vcpu);
4217 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4218 * fail for in-kernel 8259.
4220 if (pic_in_kernel(vcpu->kvm))
4223 if (vcpu->arch.pending_external_vector != -1)
4226 vcpu->arch.pending_external_vector = irq->irq;
4227 kvm_make_request(KVM_REQ_EVENT, vcpu);
4231 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4233 kvm_inject_nmi(vcpu);
4238 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4240 kvm_make_request(KVM_REQ_SMI, vcpu);
4245 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4246 struct kvm_tpr_access_ctl *tac)
4250 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4254 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4258 unsigned bank_num = mcg_cap & 0xff, bank;
4261 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4263 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4266 vcpu->arch.mcg_cap = mcg_cap;
4267 /* Init IA32_MCG_CTL to all 1s */
4268 if (mcg_cap & MCG_CTL_P)
4269 vcpu->arch.mcg_ctl = ~(u64)0;
4270 /* Init IA32_MCi_CTL to all 1s */
4271 for (bank = 0; bank < bank_num; bank++)
4272 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4274 static_call(kvm_x86_setup_mce)(vcpu);
4279 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4280 struct kvm_x86_mce *mce)
4282 u64 mcg_cap = vcpu->arch.mcg_cap;
4283 unsigned bank_num = mcg_cap & 0xff;
4284 u64 *banks = vcpu->arch.mce_banks;
4286 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4289 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4290 * reporting is disabled
4292 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4293 vcpu->arch.mcg_ctl != ~(u64)0)
4295 banks += 4 * mce->bank;
4297 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4298 * reporting is disabled for the bank
4300 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4302 if (mce->status & MCI_STATUS_UC) {
4303 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4304 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4305 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4308 if (banks[1] & MCI_STATUS_VAL)
4309 mce->status |= MCI_STATUS_OVER;
4310 banks[2] = mce->addr;
4311 banks[3] = mce->misc;
4312 vcpu->arch.mcg_status = mce->mcg_status;
4313 banks[1] = mce->status;
4314 kvm_queue_exception(vcpu, MC_VECTOR);
4315 } else if (!(banks[1] & MCI_STATUS_VAL)
4316 || !(banks[1] & MCI_STATUS_UC)) {
4317 if (banks[1] & MCI_STATUS_VAL)
4318 mce->status |= MCI_STATUS_OVER;
4319 banks[2] = mce->addr;
4320 banks[3] = mce->misc;
4321 banks[1] = mce->status;
4323 banks[1] |= MCI_STATUS_OVER;
4327 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4328 struct kvm_vcpu_events *events)
4332 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4336 * In guest mode, payload delivery should be deferred,
4337 * so that the L1 hypervisor can intercept #PF before
4338 * CR2 is modified (or intercept #DB before DR6 is
4339 * modified under nVMX). Unless the per-VM capability,
4340 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4341 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4342 * opportunistically defer the exception payload, deliver it if the
4343 * capability hasn't been requested before processing a
4344 * KVM_GET_VCPU_EVENTS.
4346 if (!vcpu->kvm->arch.exception_payload_enabled &&
4347 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4348 kvm_deliver_exception_payload(vcpu);
4351 * The API doesn't provide the instruction length for software
4352 * exceptions, so don't report them. As long as the guest RIP
4353 * isn't advanced, we should expect to encounter the exception
4356 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4357 events->exception.injected = 0;
4358 events->exception.pending = 0;
4360 events->exception.injected = vcpu->arch.exception.injected;
4361 events->exception.pending = vcpu->arch.exception.pending;
4363 * For ABI compatibility, deliberately conflate
4364 * pending and injected exceptions when
4365 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4367 if (!vcpu->kvm->arch.exception_payload_enabled)
4368 events->exception.injected |=
4369 vcpu->arch.exception.pending;
4371 events->exception.nr = vcpu->arch.exception.nr;
4372 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4373 events->exception.error_code = vcpu->arch.exception.error_code;
4374 events->exception_has_payload = vcpu->arch.exception.has_payload;
4375 events->exception_payload = vcpu->arch.exception.payload;
4377 events->interrupt.injected =
4378 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4379 events->interrupt.nr = vcpu->arch.interrupt.nr;
4380 events->interrupt.soft = 0;
4381 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4383 events->nmi.injected = vcpu->arch.nmi_injected;
4384 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4385 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4386 events->nmi.pad = 0;
4388 events->sipi_vector = 0; /* never valid when reporting to user space */
4390 events->smi.smm = is_smm(vcpu);
4391 events->smi.pending = vcpu->arch.smi_pending;
4392 events->smi.smm_inside_nmi =
4393 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4394 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4396 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4397 | KVM_VCPUEVENT_VALID_SHADOW
4398 | KVM_VCPUEVENT_VALID_SMM);
4399 if (vcpu->kvm->arch.exception_payload_enabled)
4400 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4402 memset(&events->reserved, 0, sizeof(events->reserved));
4405 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4407 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4408 struct kvm_vcpu_events *events)
4410 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4411 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4412 | KVM_VCPUEVENT_VALID_SHADOW
4413 | KVM_VCPUEVENT_VALID_SMM
4414 | KVM_VCPUEVENT_VALID_PAYLOAD))
4417 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4418 if (!vcpu->kvm->arch.exception_payload_enabled)
4420 if (events->exception.pending)
4421 events->exception.injected = 0;
4423 events->exception_has_payload = 0;
4425 events->exception.pending = 0;
4426 events->exception_has_payload = 0;
4429 if ((events->exception.injected || events->exception.pending) &&
4430 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4433 /* INITs are latched while in SMM */
4434 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4435 (events->smi.smm || events->smi.pending) &&
4436 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4440 vcpu->arch.exception.injected = events->exception.injected;
4441 vcpu->arch.exception.pending = events->exception.pending;
4442 vcpu->arch.exception.nr = events->exception.nr;
4443 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4444 vcpu->arch.exception.error_code = events->exception.error_code;
4445 vcpu->arch.exception.has_payload = events->exception_has_payload;
4446 vcpu->arch.exception.payload = events->exception_payload;
4448 vcpu->arch.interrupt.injected = events->interrupt.injected;
4449 vcpu->arch.interrupt.nr = events->interrupt.nr;
4450 vcpu->arch.interrupt.soft = events->interrupt.soft;
4451 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4452 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4453 events->interrupt.shadow);
4455 vcpu->arch.nmi_injected = events->nmi.injected;
4456 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4457 vcpu->arch.nmi_pending = events->nmi.pending;
4458 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4460 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4461 lapic_in_kernel(vcpu))
4462 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4464 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4465 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4466 if (events->smi.smm)
4467 vcpu->arch.hflags |= HF_SMM_MASK;
4469 vcpu->arch.hflags &= ~HF_SMM_MASK;
4470 kvm_smm_changed(vcpu);
4473 vcpu->arch.smi_pending = events->smi.pending;
4475 if (events->smi.smm) {
4476 if (events->smi.smm_inside_nmi)
4477 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4479 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4482 if (lapic_in_kernel(vcpu)) {
4483 if (events->smi.latched_init)
4484 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4486 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4490 kvm_make_request(KVM_REQ_EVENT, vcpu);
4495 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4496 struct kvm_debugregs *dbgregs)
4500 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4501 kvm_get_dr(vcpu, 6, &val);
4503 dbgregs->dr7 = vcpu->arch.dr7;
4505 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4508 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4509 struct kvm_debugregs *dbgregs)
4514 if (!kvm_dr6_valid(dbgregs->dr6))
4516 if (!kvm_dr7_valid(dbgregs->dr7))
4519 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4520 kvm_update_dr0123(vcpu);
4521 vcpu->arch.dr6 = dbgregs->dr6;
4522 vcpu->arch.dr7 = dbgregs->dr7;
4523 kvm_update_dr7(vcpu);
4528 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4530 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4532 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4533 u64 xstate_bv = xsave->header.xfeatures;
4537 * Copy legacy XSAVE area, to avoid complications with CPUID
4538 * leaves 0 and 1 in the loop below.
4540 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4543 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4544 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4547 * Copy each region from the possibly compacted offset to the
4548 * non-compacted offset.
4550 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4552 u64 xfeature_mask = valid & -valid;
4553 int xfeature_nr = fls64(xfeature_mask) - 1;
4554 void *src = get_xsave_addr(xsave, xfeature_nr);
4557 u32 size, offset, ecx, edx;
4558 cpuid_count(XSTATE_CPUID, xfeature_nr,
4559 &size, &offset, &ecx, &edx);
4560 if (xfeature_nr == XFEATURE_PKRU)
4561 memcpy(dest + offset, &vcpu->arch.pkru,
4562 sizeof(vcpu->arch.pkru));
4564 memcpy(dest + offset, src, size);
4568 valid -= xfeature_mask;
4572 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4574 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4575 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4579 * Copy legacy XSAVE area, to avoid complications with CPUID
4580 * leaves 0 and 1 in the loop below.
4582 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4584 /* Set XSTATE_BV and possibly XCOMP_BV. */
4585 xsave->header.xfeatures = xstate_bv;
4586 if (boot_cpu_has(X86_FEATURE_XSAVES))
4587 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4590 * Copy each region from the non-compacted offset to the
4591 * possibly compacted offset.
4593 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4595 u64 xfeature_mask = valid & -valid;
4596 int xfeature_nr = fls64(xfeature_mask) - 1;
4597 void *dest = get_xsave_addr(xsave, xfeature_nr);
4600 u32 size, offset, ecx, edx;
4601 cpuid_count(XSTATE_CPUID, xfeature_nr,
4602 &size, &offset, &ecx, &edx);
4603 if (xfeature_nr == XFEATURE_PKRU)
4604 memcpy(&vcpu->arch.pkru, src + offset,
4605 sizeof(vcpu->arch.pkru));
4607 memcpy(dest, src + offset, size);
4610 valid -= xfeature_mask;
4614 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4615 struct kvm_xsave *guest_xsave)
4617 if (!vcpu->arch.guest_fpu)
4620 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4621 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4622 fill_xsave((u8 *) guest_xsave->region, vcpu);
4624 memcpy(guest_xsave->region,
4625 &vcpu->arch.guest_fpu->state.fxsave,
4626 sizeof(struct fxregs_state));
4627 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4628 XFEATURE_MASK_FPSSE;
4632 #define XSAVE_MXCSR_OFFSET 24
4634 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4635 struct kvm_xsave *guest_xsave)
4640 if (!vcpu->arch.guest_fpu)
4643 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4644 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4646 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4648 * Here we allow setting states that are not present in
4649 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4650 * with old userspace.
4652 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4654 load_xsave(vcpu, (u8 *)guest_xsave->region);
4656 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4657 mxcsr & ~mxcsr_feature_mask)
4659 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4660 guest_xsave->region, sizeof(struct fxregs_state));
4665 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4666 struct kvm_xcrs *guest_xcrs)
4668 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4669 guest_xcrs->nr_xcrs = 0;
4673 guest_xcrs->nr_xcrs = 1;
4674 guest_xcrs->flags = 0;
4675 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4676 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4679 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4680 struct kvm_xcrs *guest_xcrs)
4684 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4687 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4690 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4691 /* Only support XCR0 currently */
4692 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4693 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4694 guest_xcrs->xcrs[i].value);
4703 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4704 * stopped by the hypervisor. This function will be called from the host only.
4705 * EINVAL is returned when the host attempts to set the flag for a guest that
4706 * does not support pv clocks.
4708 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4710 if (!vcpu->arch.pv_time_enabled)
4712 vcpu->arch.pvclock_set_guest_stopped_request = true;
4713 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4717 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4718 struct kvm_enable_cap *cap)
4721 uint16_t vmcs_version;
4722 void __user *user_ptr;
4728 case KVM_CAP_HYPERV_SYNIC2:
4733 case KVM_CAP_HYPERV_SYNIC:
4734 if (!irqchip_in_kernel(vcpu->kvm))
4736 return kvm_hv_activate_synic(vcpu, cap->cap ==
4737 KVM_CAP_HYPERV_SYNIC2);
4738 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4739 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4741 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4743 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4744 if (copy_to_user(user_ptr, &vmcs_version,
4745 sizeof(vmcs_version)))
4749 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4750 if (!kvm_x86_ops.enable_direct_tlbflush)
4753 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4755 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4756 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4757 if (vcpu->arch.pv_cpuid.enforce)
4758 kvm_update_pv_runtime(vcpu);
4766 long kvm_arch_vcpu_ioctl(struct file *filp,
4767 unsigned int ioctl, unsigned long arg)
4769 struct kvm_vcpu *vcpu = filp->private_data;
4770 void __user *argp = (void __user *)arg;
4773 struct kvm_lapic_state *lapic;
4774 struct kvm_xsave *xsave;
4775 struct kvm_xcrs *xcrs;
4783 case KVM_GET_LAPIC: {
4785 if (!lapic_in_kernel(vcpu))
4787 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4788 GFP_KERNEL_ACCOUNT);
4793 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4797 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4802 case KVM_SET_LAPIC: {
4804 if (!lapic_in_kernel(vcpu))
4806 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4807 if (IS_ERR(u.lapic)) {
4808 r = PTR_ERR(u.lapic);
4812 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4815 case KVM_INTERRUPT: {
4816 struct kvm_interrupt irq;
4819 if (copy_from_user(&irq, argp, sizeof(irq)))
4821 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4825 r = kvm_vcpu_ioctl_nmi(vcpu);
4829 r = kvm_vcpu_ioctl_smi(vcpu);
4832 case KVM_SET_CPUID: {
4833 struct kvm_cpuid __user *cpuid_arg = argp;
4834 struct kvm_cpuid cpuid;
4837 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4839 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4842 case KVM_SET_CPUID2: {
4843 struct kvm_cpuid2 __user *cpuid_arg = argp;
4844 struct kvm_cpuid2 cpuid;
4847 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4849 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4850 cpuid_arg->entries);
4853 case KVM_GET_CPUID2: {
4854 struct kvm_cpuid2 __user *cpuid_arg = argp;
4855 struct kvm_cpuid2 cpuid;
4858 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4860 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4861 cpuid_arg->entries);
4865 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4870 case KVM_GET_MSRS: {
4871 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4872 r = msr_io(vcpu, argp, do_get_msr, 1);
4873 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4876 case KVM_SET_MSRS: {
4877 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4878 r = msr_io(vcpu, argp, do_set_msr, 0);
4879 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4882 case KVM_TPR_ACCESS_REPORTING: {
4883 struct kvm_tpr_access_ctl tac;
4886 if (copy_from_user(&tac, argp, sizeof(tac)))
4888 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4892 if (copy_to_user(argp, &tac, sizeof(tac)))
4897 case KVM_SET_VAPIC_ADDR: {
4898 struct kvm_vapic_addr va;
4902 if (!lapic_in_kernel(vcpu))
4905 if (copy_from_user(&va, argp, sizeof(va)))
4907 idx = srcu_read_lock(&vcpu->kvm->srcu);
4908 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4909 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4912 case KVM_X86_SETUP_MCE: {
4916 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4918 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4921 case KVM_X86_SET_MCE: {
4922 struct kvm_x86_mce mce;
4925 if (copy_from_user(&mce, argp, sizeof(mce)))
4927 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4930 case KVM_GET_VCPU_EVENTS: {
4931 struct kvm_vcpu_events events;
4933 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4936 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4941 case KVM_SET_VCPU_EVENTS: {
4942 struct kvm_vcpu_events events;
4945 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4948 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4951 case KVM_GET_DEBUGREGS: {
4952 struct kvm_debugregs dbgregs;
4954 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4957 if (copy_to_user(argp, &dbgregs,
4958 sizeof(struct kvm_debugregs)))
4963 case KVM_SET_DEBUGREGS: {
4964 struct kvm_debugregs dbgregs;
4967 if (copy_from_user(&dbgregs, argp,
4968 sizeof(struct kvm_debugregs)))
4971 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4974 case KVM_GET_XSAVE: {
4975 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4980 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4983 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4988 case KVM_SET_XSAVE: {
4989 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4990 if (IS_ERR(u.xsave)) {
4991 r = PTR_ERR(u.xsave);
4995 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4998 case KVM_GET_XCRS: {
4999 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5004 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5007 if (copy_to_user(argp, u.xcrs,
5008 sizeof(struct kvm_xcrs)))
5013 case KVM_SET_XCRS: {
5014 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5015 if (IS_ERR(u.xcrs)) {
5016 r = PTR_ERR(u.xcrs);
5020 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5023 case KVM_SET_TSC_KHZ: {
5027 user_tsc_khz = (u32)arg;
5029 if (kvm_has_tsc_control &&
5030 user_tsc_khz >= kvm_max_guest_tsc_khz)
5033 if (user_tsc_khz == 0)
5034 user_tsc_khz = tsc_khz;
5036 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5041 case KVM_GET_TSC_KHZ: {
5042 r = vcpu->arch.virtual_tsc_khz;
5045 case KVM_KVMCLOCK_CTRL: {
5046 r = kvm_set_guest_paused(vcpu);
5049 case KVM_ENABLE_CAP: {
5050 struct kvm_enable_cap cap;
5053 if (copy_from_user(&cap, argp, sizeof(cap)))
5055 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5058 case KVM_GET_NESTED_STATE: {
5059 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5063 if (!kvm_x86_ops.nested_ops->get_state)
5066 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5068 if (get_user(user_data_size, &user_kvm_nested_state->size))
5071 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5076 if (r > user_data_size) {
5077 if (put_user(r, &user_kvm_nested_state->size))
5087 case KVM_SET_NESTED_STATE: {
5088 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5089 struct kvm_nested_state kvm_state;
5093 if (!kvm_x86_ops.nested_ops->set_state)
5097 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5101 if (kvm_state.size < sizeof(kvm_state))
5104 if (kvm_state.flags &
5105 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5106 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5107 | KVM_STATE_NESTED_GIF_SET))
5110 /* nested_run_pending implies guest_mode. */
5111 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5112 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5115 idx = srcu_read_lock(&vcpu->kvm->srcu);
5116 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5117 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5120 case KVM_GET_SUPPORTED_HV_CPUID:
5121 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5123 #ifdef CONFIG_KVM_XEN
5124 case KVM_XEN_VCPU_GET_ATTR: {
5125 struct kvm_xen_vcpu_attr xva;
5128 if (copy_from_user(&xva, argp, sizeof(xva)))
5130 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5131 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5135 case KVM_XEN_VCPU_SET_ATTR: {
5136 struct kvm_xen_vcpu_attr xva;
5139 if (copy_from_user(&xva, argp, sizeof(xva)))
5141 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5155 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5157 return VM_FAULT_SIGBUS;
5160 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5164 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5166 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5170 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5173 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5176 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5177 unsigned long kvm_nr_mmu_pages)
5179 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5182 mutex_lock(&kvm->slots_lock);
5184 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5185 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5187 mutex_unlock(&kvm->slots_lock);
5191 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5193 return kvm->arch.n_max_mmu_pages;
5196 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5198 struct kvm_pic *pic = kvm->arch.vpic;
5202 switch (chip->chip_id) {
5203 case KVM_IRQCHIP_PIC_MASTER:
5204 memcpy(&chip->chip.pic, &pic->pics[0],
5205 sizeof(struct kvm_pic_state));
5207 case KVM_IRQCHIP_PIC_SLAVE:
5208 memcpy(&chip->chip.pic, &pic->pics[1],
5209 sizeof(struct kvm_pic_state));
5211 case KVM_IRQCHIP_IOAPIC:
5212 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5221 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5223 struct kvm_pic *pic = kvm->arch.vpic;
5227 switch (chip->chip_id) {
5228 case KVM_IRQCHIP_PIC_MASTER:
5229 spin_lock(&pic->lock);
5230 memcpy(&pic->pics[0], &chip->chip.pic,
5231 sizeof(struct kvm_pic_state));
5232 spin_unlock(&pic->lock);
5234 case KVM_IRQCHIP_PIC_SLAVE:
5235 spin_lock(&pic->lock);
5236 memcpy(&pic->pics[1], &chip->chip.pic,
5237 sizeof(struct kvm_pic_state));
5238 spin_unlock(&pic->lock);
5240 case KVM_IRQCHIP_IOAPIC:
5241 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5247 kvm_pic_update_irq(pic);
5251 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5253 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5255 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5257 mutex_lock(&kps->lock);
5258 memcpy(ps, &kps->channels, sizeof(*ps));
5259 mutex_unlock(&kps->lock);
5263 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5266 struct kvm_pit *pit = kvm->arch.vpit;
5268 mutex_lock(&pit->pit_state.lock);
5269 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5270 for (i = 0; i < 3; i++)
5271 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5272 mutex_unlock(&pit->pit_state.lock);
5276 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5278 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5279 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5280 sizeof(ps->channels));
5281 ps->flags = kvm->arch.vpit->pit_state.flags;
5282 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5283 memset(&ps->reserved, 0, sizeof(ps->reserved));
5287 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5291 u32 prev_legacy, cur_legacy;
5292 struct kvm_pit *pit = kvm->arch.vpit;
5294 mutex_lock(&pit->pit_state.lock);
5295 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5296 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5297 if (!prev_legacy && cur_legacy)
5299 memcpy(&pit->pit_state.channels, &ps->channels,
5300 sizeof(pit->pit_state.channels));
5301 pit->pit_state.flags = ps->flags;
5302 for (i = 0; i < 3; i++)
5303 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5305 mutex_unlock(&pit->pit_state.lock);
5309 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5310 struct kvm_reinject_control *control)
5312 struct kvm_pit *pit = kvm->arch.vpit;
5314 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5315 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5316 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5318 mutex_lock(&pit->pit_state.lock);
5319 kvm_pit_set_reinject(pit, control->pit_reinject);
5320 mutex_unlock(&pit->pit_state.lock);
5325 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5329 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5330 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5331 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5334 struct kvm_vcpu *vcpu;
5337 kvm_for_each_vcpu(i, vcpu, kvm)
5338 kvm_vcpu_kick(vcpu);
5341 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5344 if (!irqchip_in_kernel(kvm))
5347 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5348 irq_event->irq, irq_event->level,
5353 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5354 struct kvm_enable_cap *cap)
5362 case KVM_CAP_DISABLE_QUIRKS:
5363 kvm->arch.disabled_quirks = cap->args[0];
5366 case KVM_CAP_SPLIT_IRQCHIP: {
5367 mutex_lock(&kvm->lock);
5369 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5370 goto split_irqchip_unlock;
5372 if (irqchip_in_kernel(kvm))
5373 goto split_irqchip_unlock;
5374 if (kvm->created_vcpus)
5375 goto split_irqchip_unlock;
5376 r = kvm_setup_empty_irq_routing(kvm);
5378 goto split_irqchip_unlock;
5379 /* Pairs with irqchip_in_kernel. */
5381 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5382 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5384 split_irqchip_unlock:
5385 mutex_unlock(&kvm->lock);
5388 case KVM_CAP_X2APIC_API:
5390 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5393 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5394 kvm->arch.x2apic_format = true;
5395 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5396 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5400 case KVM_CAP_X86_DISABLE_EXITS:
5402 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5405 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5406 kvm_can_mwait_in_guest())
5407 kvm->arch.mwait_in_guest = true;
5408 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5409 kvm->arch.hlt_in_guest = true;
5410 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5411 kvm->arch.pause_in_guest = true;
5412 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5413 kvm->arch.cstate_in_guest = true;
5416 case KVM_CAP_MSR_PLATFORM_INFO:
5417 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5420 case KVM_CAP_EXCEPTION_PAYLOAD:
5421 kvm->arch.exception_payload_enabled = cap->args[0];
5424 case KVM_CAP_X86_USER_SPACE_MSR:
5425 kvm->arch.user_space_msr_mask = cap->args[0];
5428 case KVM_CAP_X86_BUS_LOCK_EXIT:
5430 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5433 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5434 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5437 if (kvm_has_bus_lock_exit &&
5438 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5439 kvm->arch.bus_lock_detection_enabled = true;
5442 #ifdef CONFIG_X86_SGX_KVM
5443 case KVM_CAP_SGX_ATTRIBUTE: {
5444 unsigned long allowed_attributes = 0;
5446 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5450 /* KVM only supports the PROVISIONKEY privileged attribute. */
5451 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5452 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5453 kvm->arch.sgx_provisioning_allowed = true;
5459 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5461 if (kvm_x86_ops.vm_copy_enc_context_from)
5462 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5471 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5473 struct kvm_x86_msr_filter *msr_filter;
5475 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5479 msr_filter->default_allow = default_allow;
5483 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5490 for (i = 0; i < msr_filter->count; i++)
5491 kfree(msr_filter->ranges[i].bitmap);
5496 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5497 struct kvm_msr_filter_range *user_range)
5499 unsigned long *bitmap = NULL;
5502 if (!user_range->nmsrs)
5505 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5508 if (!user_range->flags)
5511 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5512 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5515 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5517 return PTR_ERR(bitmap);
5519 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5520 .flags = user_range->flags,
5521 .base = user_range->base,
5522 .nmsrs = user_range->nmsrs,
5526 msr_filter->count++;
5530 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5532 struct kvm_msr_filter __user *user_msr_filter = argp;
5533 struct kvm_x86_msr_filter *new_filter, *old_filter;
5534 struct kvm_msr_filter filter;
5540 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5543 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5544 empty &= !filter.ranges[i].nmsrs;
5546 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5547 if (empty && !default_allow)
5550 new_filter = kvm_alloc_msr_filter(default_allow);
5554 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5555 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5557 kvm_free_msr_filter(new_filter);
5562 mutex_lock(&kvm->lock);
5564 /* The per-VM filter is protected by kvm->lock... */
5565 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5567 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5568 synchronize_srcu(&kvm->srcu);
5570 kvm_free_msr_filter(old_filter);
5572 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5573 mutex_unlock(&kvm->lock);
5578 long kvm_arch_vm_ioctl(struct file *filp,
5579 unsigned int ioctl, unsigned long arg)
5581 struct kvm *kvm = filp->private_data;
5582 void __user *argp = (void __user *)arg;
5585 * This union makes it completely explicit to gcc-3.x
5586 * that these two variables' stack usage should be
5587 * combined, not added together.
5590 struct kvm_pit_state ps;
5591 struct kvm_pit_state2 ps2;
5592 struct kvm_pit_config pit_config;
5596 case KVM_SET_TSS_ADDR:
5597 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5599 case KVM_SET_IDENTITY_MAP_ADDR: {
5602 mutex_lock(&kvm->lock);
5604 if (kvm->created_vcpus)
5605 goto set_identity_unlock;
5607 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5608 goto set_identity_unlock;
5609 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5610 set_identity_unlock:
5611 mutex_unlock(&kvm->lock);
5614 case KVM_SET_NR_MMU_PAGES:
5615 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5617 case KVM_GET_NR_MMU_PAGES:
5618 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5620 case KVM_CREATE_IRQCHIP: {
5621 mutex_lock(&kvm->lock);
5624 if (irqchip_in_kernel(kvm))
5625 goto create_irqchip_unlock;
5628 if (kvm->created_vcpus)
5629 goto create_irqchip_unlock;
5631 r = kvm_pic_init(kvm);
5633 goto create_irqchip_unlock;
5635 r = kvm_ioapic_init(kvm);
5637 kvm_pic_destroy(kvm);
5638 goto create_irqchip_unlock;
5641 r = kvm_setup_default_irq_routing(kvm);
5643 kvm_ioapic_destroy(kvm);
5644 kvm_pic_destroy(kvm);
5645 goto create_irqchip_unlock;
5647 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5649 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5650 create_irqchip_unlock:
5651 mutex_unlock(&kvm->lock);
5654 case KVM_CREATE_PIT:
5655 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5657 case KVM_CREATE_PIT2:
5659 if (copy_from_user(&u.pit_config, argp,
5660 sizeof(struct kvm_pit_config)))
5663 mutex_lock(&kvm->lock);
5666 goto create_pit_unlock;
5668 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5672 mutex_unlock(&kvm->lock);
5674 case KVM_GET_IRQCHIP: {
5675 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5676 struct kvm_irqchip *chip;
5678 chip = memdup_user(argp, sizeof(*chip));
5685 if (!irqchip_kernel(kvm))
5686 goto get_irqchip_out;
5687 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5689 goto get_irqchip_out;
5691 if (copy_to_user(argp, chip, sizeof(*chip)))
5692 goto get_irqchip_out;
5698 case KVM_SET_IRQCHIP: {
5699 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5700 struct kvm_irqchip *chip;
5702 chip = memdup_user(argp, sizeof(*chip));
5709 if (!irqchip_kernel(kvm))
5710 goto set_irqchip_out;
5711 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5718 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5721 if (!kvm->arch.vpit)
5723 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5727 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5734 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5736 mutex_lock(&kvm->lock);
5738 if (!kvm->arch.vpit)
5740 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5742 mutex_unlock(&kvm->lock);
5745 case KVM_GET_PIT2: {
5747 if (!kvm->arch.vpit)
5749 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5753 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5758 case KVM_SET_PIT2: {
5760 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5762 mutex_lock(&kvm->lock);
5764 if (!kvm->arch.vpit)
5766 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5768 mutex_unlock(&kvm->lock);
5771 case KVM_REINJECT_CONTROL: {
5772 struct kvm_reinject_control control;
5774 if (copy_from_user(&control, argp, sizeof(control)))
5777 if (!kvm->arch.vpit)
5779 r = kvm_vm_ioctl_reinject(kvm, &control);
5782 case KVM_SET_BOOT_CPU_ID:
5784 mutex_lock(&kvm->lock);
5785 if (kvm->created_vcpus)
5788 kvm->arch.bsp_vcpu_id = arg;
5789 mutex_unlock(&kvm->lock);
5791 #ifdef CONFIG_KVM_XEN
5792 case KVM_XEN_HVM_CONFIG: {
5793 struct kvm_xen_hvm_config xhc;
5795 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5797 r = kvm_xen_hvm_config(kvm, &xhc);
5800 case KVM_XEN_HVM_GET_ATTR: {
5801 struct kvm_xen_hvm_attr xha;
5804 if (copy_from_user(&xha, argp, sizeof(xha)))
5806 r = kvm_xen_hvm_get_attr(kvm, &xha);
5807 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
5811 case KVM_XEN_HVM_SET_ATTR: {
5812 struct kvm_xen_hvm_attr xha;
5815 if (copy_from_user(&xha, argp, sizeof(xha)))
5817 r = kvm_xen_hvm_set_attr(kvm, &xha);
5821 case KVM_SET_CLOCK: {
5822 struct kvm_arch *ka = &kvm->arch;
5823 struct kvm_clock_data user_ns;
5827 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5836 * TODO: userspace has to take care of races with VCPU_RUN, so
5837 * kvm_gen_update_masterclock() can be cut down to locked
5838 * pvclock_update_vm_gtod_copy().
5840 kvm_gen_update_masterclock(kvm);
5843 * This pairs with kvm_guest_time_update(): when masterclock is
5844 * in use, we use master_kernel_ns + kvmclock_offset to set
5845 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
5846 * is slightly ahead) here we risk going negative on unsigned
5847 * 'system_time' when 'user_ns.clock' is very small.
5849 spin_lock_irq(&ka->pvclock_gtod_sync_lock);
5850 if (kvm->arch.use_master_clock)
5851 now_ns = ka->master_kernel_ns;
5853 now_ns = get_kvmclock_base_ns();
5854 ka->kvmclock_offset = user_ns.clock - now_ns;
5855 spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
5857 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5860 case KVM_GET_CLOCK: {
5861 struct kvm_clock_data user_ns;
5864 now_ns = get_kvmclock_ns(kvm);
5865 user_ns.clock = now_ns;
5866 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5867 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5870 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5875 case KVM_MEMORY_ENCRYPT_OP: {
5877 if (kvm_x86_ops.mem_enc_op)
5878 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
5881 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5882 struct kvm_enc_region region;
5885 if (copy_from_user(®ion, argp, sizeof(region)))
5889 if (kvm_x86_ops.mem_enc_reg_region)
5890 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
5893 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5894 struct kvm_enc_region region;
5897 if (copy_from_user(®ion, argp, sizeof(region)))
5901 if (kvm_x86_ops.mem_enc_unreg_region)
5902 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
5905 case KVM_HYPERV_EVENTFD: {
5906 struct kvm_hyperv_eventfd hvevfd;
5909 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5911 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5914 case KVM_SET_PMU_EVENT_FILTER:
5915 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5917 case KVM_X86_SET_MSR_FILTER:
5918 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5927 static void kvm_init_msr_list(void)
5929 struct x86_pmu_capability x86_pmu;
5933 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5934 "Please update the fixed PMCs in msrs_to_saved_all[]");
5936 perf_get_x86_pmu_capability(&x86_pmu);
5938 num_msrs_to_save = 0;
5939 num_emulated_msrs = 0;
5940 num_msr_based_features = 0;
5942 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5943 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5947 * Even MSRs that are valid in the host may not be exposed
5948 * to the guests in some cases.
5950 switch (msrs_to_save_all[i]) {
5951 case MSR_IA32_BNDCFGS:
5952 if (!kvm_mpx_supported())
5956 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
5957 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
5960 case MSR_IA32_UMWAIT_CONTROL:
5961 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5964 case MSR_IA32_RTIT_CTL:
5965 case MSR_IA32_RTIT_STATUS:
5966 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5969 case MSR_IA32_RTIT_CR3_MATCH:
5970 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5971 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5974 case MSR_IA32_RTIT_OUTPUT_BASE:
5975 case MSR_IA32_RTIT_OUTPUT_MASK:
5976 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5977 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5978 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5981 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5982 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5983 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5984 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5987 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5988 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5989 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5992 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5993 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5994 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6001 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6004 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6005 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6008 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6011 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6012 struct kvm_msr_entry msr;
6014 msr.index = msr_based_features_all[i];
6015 if (kvm_get_msr_feature(&msr))
6018 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6022 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6030 if (!(lapic_in_kernel(vcpu) &&
6031 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6032 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6043 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6050 if (!(lapic_in_kernel(vcpu) &&
6051 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6053 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6055 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6065 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6066 struct kvm_segment *var, int seg)
6068 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6071 void kvm_get_segment(struct kvm_vcpu *vcpu,
6072 struct kvm_segment *var, int seg)
6074 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6077 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6078 struct x86_exception *exception)
6082 BUG_ON(!mmu_is_nested(vcpu));
6084 /* NPT walks are always user-walks */
6085 access |= PFERR_USER_MASK;
6086 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6091 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6092 struct x86_exception *exception)
6094 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6095 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6097 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6099 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6100 struct x86_exception *exception)
6102 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6103 access |= PFERR_FETCH_MASK;
6104 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6107 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6108 struct x86_exception *exception)
6110 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6111 access |= PFERR_WRITE_MASK;
6112 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6114 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6116 /* uses this to access any guest's mapped memory without checking CPL */
6117 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6118 struct x86_exception *exception)
6120 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6123 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6124 struct kvm_vcpu *vcpu, u32 access,
6125 struct x86_exception *exception)
6128 int r = X86EMUL_CONTINUE;
6131 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6133 unsigned offset = addr & (PAGE_SIZE-1);
6134 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6137 if (gpa == UNMAPPED_GVA)
6138 return X86EMUL_PROPAGATE_FAULT;
6139 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6142 r = X86EMUL_IO_NEEDED;
6154 /* used for instruction fetching */
6155 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6156 gva_t addr, void *val, unsigned int bytes,
6157 struct x86_exception *exception)
6159 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6160 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6164 /* Inline kvm_read_guest_virt_helper for speed. */
6165 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6167 if (unlikely(gpa == UNMAPPED_GVA))
6168 return X86EMUL_PROPAGATE_FAULT;
6170 offset = addr & (PAGE_SIZE-1);
6171 if (WARN_ON(offset + bytes > PAGE_SIZE))
6172 bytes = (unsigned)PAGE_SIZE - offset;
6173 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6175 if (unlikely(ret < 0))
6176 return X86EMUL_IO_NEEDED;
6178 return X86EMUL_CONTINUE;
6181 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6182 gva_t addr, void *val, unsigned int bytes,
6183 struct x86_exception *exception)
6185 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6188 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6189 * is returned, but our callers are not ready for that and they blindly
6190 * call kvm_inject_page_fault. Ensure that they at least do not leak
6191 * uninitialized kernel stack memory into cr2 and error code.
6193 memset(exception, 0, sizeof(*exception));
6194 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6197 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6199 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6200 gva_t addr, void *val, unsigned int bytes,
6201 struct x86_exception *exception, bool system)
6203 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6206 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6207 access |= PFERR_USER_MASK;
6209 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6212 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6213 unsigned long addr, void *val, unsigned int bytes)
6215 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6216 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6218 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6221 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6222 struct kvm_vcpu *vcpu, u32 access,
6223 struct x86_exception *exception)
6226 int r = X86EMUL_CONTINUE;
6229 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6232 unsigned offset = addr & (PAGE_SIZE-1);
6233 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6236 if (gpa == UNMAPPED_GVA)
6237 return X86EMUL_PROPAGATE_FAULT;
6238 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6240 r = X86EMUL_IO_NEEDED;
6252 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6253 unsigned int bytes, struct x86_exception *exception,
6256 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6257 u32 access = PFERR_WRITE_MASK;
6259 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6260 access |= PFERR_USER_MASK;
6262 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6266 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6267 unsigned int bytes, struct x86_exception *exception)
6269 /* kvm_write_guest_virt_system can pull in tons of pages. */
6270 vcpu->arch.l1tf_flush_l1d = true;
6272 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6273 PFERR_WRITE_MASK, exception);
6275 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6277 int handle_ud(struct kvm_vcpu *vcpu)
6279 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6280 int emul_type = EMULTYPE_TRAP_UD;
6281 char sig[5]; /* ud2; .ascii "kvm" */
6282 struct x86_exception e;
6284 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6287 if (force_emulation_prefix &&
6288 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6289 sig, sizeof(sig), &e) == 0 &&
6290 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6291 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6292 emul_type = EMULTYPE_TRAP_UD_FORCED;
6295 return kvm_emulate_instruction(vcpu, emul_type);
6297 EXPORT_SYMBOL_GPL(handle_ud);
6299 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6300 gpa_t gpa, bool write)
6302 /* For APIC access vmexit */
6303 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6306 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6307 trace_vcpu_match_mmio(gva, gpa, write, true);
6314 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6315 gpa_t *gpa, struct x86_exception *exception,
6318 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6319 | (write ? PFERR_WRITE_MASK : 0);
6322 * currently PKRU is only applied to ept enabled guest so
6323 * there is no pkey in EPT page table for L1 guest or EPT
6324 * shadow page table for L2 guest.
6326 if (vcpu_match_mmio_gva(vcpu, gva)
6327 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6328 vcpu->arch.mmio_access, 0, access)) {
6329 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6330 (gva & (PAGE_SIZE - 1));
6331 trace_vcpu_match_mmio(gva, *gpa, write, false);
6335 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6337 if (*gpa == UNMAPPED_GVA)
6340 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6343 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6344 const void *val, int bytes)
6348 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6351 kvm_page_track_write(vcpu, gpa, val, bytes);
6355 struct read_write_emulator_ops {
6356 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6358 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6359 void *val, int bytes);
6360 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6361 int bytes, void *val);
6362 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6363 void *val, int bytes);
6367 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6369 if (vcpu->mmio_read_completed) {
6370 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6371 vcpu->mmio_fragments[0].gpa, val);
6372 vcpu->mmio_read_completed = 0;
6379 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6380 void *val, int bytes)
6382 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6385 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6386 void *val, int bytes)
6388 return emulator_write_phys(vcpu, gpa, val, bytes);
6391 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6393 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6394 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6397 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6398 void *val, int bytes)
6400 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6401 return X86EMUL_IO_NEEDED;
6404 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6405 void *val, int bytes)
6407 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6409 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6410 return X86EMUL_CONTINUE;
6413 static const struct read_write_emulator_ops read_emultor = {
6414 .read_write_prepare = read_prepare,
6415 .read_write_emulate = read_emulate,
6416 .read_write_mmio = vcpu_mmio_read,
6417 .read_write_exit_mmio = read_exit_mmio,
6420 static const struct read_write_emulator_ops write_emultor = {
6421 .read_write_emulate = write_emulate,
6422 .read_write_mmio = write_mmio,
6423 .read_write_exit_mmio = write_exit_mmio,
6427 static int emulator_read_write_onepage(unsigned long addr, void *val,
6429 struct x86_exception *exception,
6430 struct kvm_vcpu *vcpu,
6431 const struct read_write_emulator_ops *ops)
6435 bool write = ops->write;
6436 struct kvm_mmio_fragment *frag;
6437 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6440 * If the exit was due to a NPF we may already have a GPA.
6441 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6442 * Note, this cannot be used on string operations since string
6443 * operation using rep will only have the initial GPA from the NPF
6446 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6447 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6448 gpa = ctxt->gpa_val;
6449 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6451 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6453 return X86EMUL_PROPAGATE_FAULT;
6456 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6457 return X86EMUL_CONTINUE;
6460 * Is this MMIO handled locally?
6462 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6463 if (handled == bytes)
6464 return X86EMUL_CONTINUE;
6470 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6471 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6475 return X86EMUL_CONTINUE;
6478 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6480 void *val, unsigned int bytes,
6481 struct x86_exception *exception,
6482 const struct read_write_emulator_ops *ops)
6484 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6488 if (ops->read_write_prepare &&
6489 ops->read_write_prepare(vcpu, val, bytes))
6490 return X86EMUL_CONTINUE;
6492 vcpu->mmio_nr_fragments = 0;
6494 /* Crossing a page boundary? */
6495 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6498 now = -addr & ~PAGE_MASK;
6499 rc = emulator_read_write_onepage(addr, val, now, exception,
6502 if (rc != X86EMUL_CONTINUE)
6505 if (ctxt->mode != X86EMUL_MODE_PROT64)
6511 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6513 if (rc != X86EMUL_CONTINUE)
6516 if (!vcpu->mmio_nr_fragments)
6519 gpa = vcpu->mmio_fragments[0].gpa;
6521 vcpu->mmio_needed = 1;
6522 vcpu->mmio_cur_fragment = 0;
6524 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6525 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6526 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6527 vcpu->run->mmio.phys_addr = gpa;
6529 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6532 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6536 struct x86_exception *exception)
6538 return emulator_read_write(ctxt, addr, val, bytes,
6539 exception, &read_emultor);
6542 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6546 struct x86_exception *exception)
6548 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6549 exception, &write_emultor);
6552 #define CMPXCHG_TYPE(t, ptr, old, new) \
6553 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6555 #ifdef CONFIG_X86_64
6556 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6558 # define CMPXCHG64(ptr, old, new) \
6559 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6562 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6567 struct x86_exception *exception)
6569 struct kvm_host_map map;
6570 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6576 /* guests cmpxchg8b have to be emulated atomically */
6577 if (bytes > 8 || (bytes & (bytes - 1)))
6580 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6582 if (gpa == UNMAPPED_GVA ||
6583 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6587 * Emulate the atomic as a straight write to avoid #AC if SLD is
6588 * enabled in the host and the access splits a cache line.
6590 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6591 page_line_mask = ~(cache_line_size() - 1);
6593 page_line_mask = PAGE_MASK;
6595 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6598 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6601 kaddr = map.hva + offset_in_page(gpa);
6605 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6608 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6611 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6614 exchanged = CMPXCHG64(kaddr, old, new);
6620 kvm_vcpu_unmap(vcpu, &map, true);
6623 return X86EMUL_CMPXCHG_FAILED;
6625 kvm_page_track_write(vcpu, gpa, new, bytes);
6627 return X86EMUL_CONTINUE;
6630 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6632 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6635 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6639 for (i = 0; i < vcpu->arch.pio.count; i++) {
6640 if (vcpu->arch.pio.in)
6641 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6642 vcpu->arch.pio.size, pd);
6644 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6645 vcpu->arch.pio.port, vcpu->arch.pio.size,
6649 pd += vcpu->arch.pio.size;
6654 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6655 unsigned short port, void *val,
6656 unsigned int count, bool in)
6658 vcpu->arch.pio.port = port;
6659 vcpu->arch.pio.in = in;
6660 vcpu->arch.pio.count = count;
6661 vcpu->arch.pio.size = size;
6663 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6664 vcpu->arch.pio.count = 0;
6668 vcpu->run->exit_reason = KVM_EXIT_IO;
6669 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6670 vcpu->run->io.size = size;
6671 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6672 vcpu->run->io.count = count;
6673 vcpu->run->io.port = port;
6678 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6679 unsigned short port, void *val, unsigned int count)
6683 if (vcpu->arch.pio.count)
6686 memset(vcpu->arch.pio_data, 0, size * count);
6688 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6691 memcpy(val, vcpu->arch.pio_data, size * count);
6692 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6693 vcpu->arch.pio.count = 0;
6700 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6701 int size, unsigned short port, void *val,
6704 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6708 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6709 unsigned short port, const void *val,
6712 memcpy(vcpu->arch.pio_data, val, size * count);
6713 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6714 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6717 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6718 int size, unsigned short port,
6719 const void *val, unsigned int count)
6721 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6724 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6726 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6729 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6731 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6734 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6736 if (!need_emulate_wbinvd(vcpu))
6737 return X86EMUL_CONTINUE;
6739 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6740 int cpu = get_cpu();
6742 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6743 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6744 wbinvd_ipi, NULL, 1);
6746 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6749 return X86EMUL_CONTINUE;
6752 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6754 kvm_emulate_wbinvd_noskip(vcpu);
6755 return kvm_skip_emulated_instruction(vcpu);
6757 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6761 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6763 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6766 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6767 unsigned long *dest)
6769 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6772 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6773 unsigned long value)
6776 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6779 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6781 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6784 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6786 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6787 unsigned long value;
6791 value = kvm_read_cr0(vcpu);
6794 value = vcpu->arch.cr2;
6797 value = kvm_read_cr3(vcpu);
6800 value = kvm_read_cr4(vcpu);
6803 value = kvm_get_cr8(vcpu);
6806 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6813 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6815 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6820 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6823 vcpu->arch.cr2 = val;
6826 res = kvm_set_cr3(vcpu, val);
6829 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6832 res = kvm_set_cr8(vcpu, val);
6835 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6842 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6844 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
6847 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6849 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
6852 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6854 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
6857 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6859 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
6862 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6864 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
6867 static unsigned long emulator_get_cached_segment_base(
6868 struct x86_emulate_ctxt *ctxt, int seg)
6870 return get_segment_base(emul_to_vcpu(ctxt), seg);
6873 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6874 struct desc_struct *desc, u32 *base3,
6877 struct kvm_segment var;
6879 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6880 *selector = var.selector;
6883 memset(desc, 0, sizeof(*desc));
6891 set_desc_limit(desc, var.limit);
6892 set_desc_base(desc, (unsigned long)var.base);
6893 #ifdef CONFIG_X86_64
6895 *base3 = var.base >> 32;
6897 desc->type = var.type;
6899 desc->dpl = var.dpl;
6900 desc->p = var.present;
6901 desc->avl = var.avl;
6909 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6910 struct desc_struct *desc, u32 base3,
6913 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6914 struct kvm_segment var;
6916 var.selector = selector;
6917 var.base = get_desc_base(desc);
6918 #ifdef CONFIG_X86_64
6919 var.base |= ((u64)base3) << 32;
6921 var.limit = get_desc_limit(desc);
6923 var.limit = (var.limit << 12) | 0xfff;
6924 var.type = desc->type;
6925 var.dpl = desc->dpl;
6930 var.avl = desc->avl;
6931 var.present = desc->p;
6932 var.unusable = !var.present;
6935 kvm_set_segment(vcpu, &var, seg);
6939 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6940 u32 msr_index, u64 *pdata)
6942 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6945 r = kvm_get_msr(vcpu, msr_index, pdata);
6947 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6948 /* Bounce to user space */
6949 return X86EMUL_IO_NEEDED;
6955 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6956 u32 msr_index, u64 data)
6958 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6961 r = kvm_set_msr(vcpu, msr_index, data);
6963 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
6964 /* Bounce to user space */
6965 return X86EMUL_IO_NEEDED;
6971 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6973 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6975 return vcpu->arch.smbase;
6978 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6980 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6982 vcpu->arch.smbase = smbase;
6985 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6988 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6991 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6992 u32 pmc, u64 *pdata)
6994 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6997 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6999 emul_to_vcpu(ctxt)->arch.halt_request = 1;
7002 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7003 struct x86_instruction_info *info,
7004 enum x86_intercept_stage stage)
7006 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7010 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7011 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7014 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7017 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7019 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7022 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7024 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7027 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7029 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7032 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7034 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7037 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7039 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7042 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7044 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7047 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7049 return emul_to_vcpu(ctxt)->arch.hflags;
7052 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
7054 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
7057 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
7058 const char *smstate)
7060 return static_call(kvm_x86_pre_leave_smm)(emul_to_vcpu(ctxt), smstate);
7063 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
7065 kvm_smm_changed(emul_to_vcpu(ctxt));
7068 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7070 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7073 static const struct x86_emulate_ops emulate_ops = {
7074 .read_gpr = emulator_read_gpr,
7075 .write_gpr = emulator_write_gpr,
7076 .read_std = emulator_read_std,
7077 .write_std = emulator_write_std,
7078 .read_phys = kvm_read_guest_phys_system,
7079 .fetch = kvm_fetch_guest_virt,
7080 .read_emulated = emulator_read_emulated,
7081 .write_emulated = emulator_write_emulated,
7082 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7083 .invlpg = emulator_invlpg,
7084 .pio_in_emulated = emulator_pio_in_emulated,
7085 .pio_out_emulated = emulator_pio_out_emulated,
7086 .get_segment = emulator_get_segment,
7087 .set_segment = emulator_set_segment,
7088 .get_cached_segment_base = emulator_get_cached_segment_base,
7089 .get_gdt = emulator_get_gdt,
7090 .get_idt = emulator_get_idt,
7091 .set_gdt = emulator_set_gdt,
7092 .set_idt = emulator_set_idt,
7093 .get_cr = emulator_get_cr,
7094 .set_cr = emulator_set_cr,
7095 .cpl = emulator_get_cpl,
7096 .get_dr = emulator_get_dr,
7097 .set_dr = emulator_set_dr,
7098 .get_smbase = emulator_get_smbase,
7099 .set_smbase = emulator_set_smbase,
7100 .set_msr = emulator_set_msr,
7101 .get_msr = emulator_get_msr,
7102 .check_pmc = emulator_check_pmc,
7103 .read_pmc = emulator_read_pmc,
7104 .halt = emulator_halt,
7105 .wbinvd = emulator_wbinvd,
7106 .fix_hypercall = emulator_fix_hypercall,
7107 .intercept = emulator_intercept,
7108 .get_cpuid = emulator_get_cpuid,
7109 .guest_has_long_mode = emulator_guest_has_long_mode,
7110 .guest_has_movbe = emulator_guest_has_movbe,
7111 .guest_has_fxsr = emulator_guest_has_fxsr,
7112 .set_nmi_mask = emulator_set_nmi_mask,
7113 .get_hflags = emulator_get_hflags,
7114 .set_hflags = emulator_set_hflags,
7115 .pre_leave_smm = emulator_pre_leave_smm,
7116 .post_leave_smm = emulator_post_leave_smm,
7117 .set_xcr = emulator_set_xcr,
7120 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7122 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7124 * an sti; sti; sequence only disable interrupts for the first
7125 * instruction. So, if the last instruction, be it emulated or
7126 * not, left the system with the INT_STI flag enabled, it
7127 * means that the last instruction is an sti. We should not
7128 * leave the flag on in this case. The same goes for mov ss
7130 if (int_shadow & mask)
7132 if (unlikely(int_shadow || mask)) {
7133 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7135 kvm_make_request(KVM_REQ_EVENT, vcpu);
7139 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7141 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7142 if (ctxt->exception.vector == PF_VECTOR)
7143 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7145 if (ctxt->exception.error_code_valid)
7146 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7147 ctxt->exception.error_code);
7149 kvm_queue_exception(vcpu, ctxt->exception.vector);
7153 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7155 struct x86_emulate_ctxt *ctxt;
7157 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7159 pr_err("kvm: failed to allocate vcpu's emulator\n");
7164 ctxt->ops = &emulate_ops;
7165 vcpu->arch.emulate_ctxt = ctxt;
7170 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7172 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7175 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7177 ctxt->gpa_available = false;
7178 ctxt->eflags = kvm_get_rflags(vcpu);
7179 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7181 ctxt->eip = kvm_rip_read(vcpu);
7182 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7183 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7184 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7185 cs_db ? X86EMUL_MODE_PROT32 :
7186 X86EMUL_MODE_PROT16;
7187 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7188 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7189 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7191 init_decode_cache(ctxt);
7192 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7195 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7197 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7200 init_emulate_ctxt(vcpu);
7204 ctxt->_eip = ctxt->eip + inc_eip;
7205 ret = emulate_int_real(ctxt, irq);
7207 if (ret != X86EMUL_CONTINUE) {
7208 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7210 ctxt->eip = ctxt->_eip;
7211 kvm_rip_write(vcpu, ctxt->eip);
7212 kvm_set_rflags(vcpu, ctxt->eflags);
7215 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7217 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7219 ++vcpu->stat.insn_emulation_fail;
7220 trace_kvm_emulate_insn_failed(vcpu);
7222 if (emulation_type & EMULTYPE_VMWARE_GP) {
7223 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7227 if (emulation_type & EMULTYPE_SKIP) {
7228 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7229 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7230 vcpu->run->internal.ndata = 0;
7234 kvm_queue_exception(vcpu, UD_VECTOR);
7236 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7237 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7238 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7239 vcpu->run->internal.ndata = 0;
7246 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7247 bool write_fault_to_shadow_pgtable,
7250 gpa_t gpa = cr2_or_gpa;
7253 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7256 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7257 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7260 if (!vcpu->arch.mmu->direct_map) {
7262 * Write permission should be allowed since only
7263 * write access need to be emulated.
7265 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7268 * If the mapping is invalid in guest, let cpu retry
7269 * it to generate fault.
7271 if (gpa == UNMAPPED_GVA)
7276 * Do not retry the unhandleable instruction if it faults on the
7277 * readonly host memory, otherwise it will goto a infinite loop:
7278 * retry instruction -> write #PF -> emulation fail -> retry
7279 * instruction -> ...
7281 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7284 * If the instruction failed on the error pfn, it can not be fixed,
7285 * report the error to userspace.
7287 if (is_error_noslot_pfn(pfn))
7290 kvm_release_pfn_clean(pfn);
7292 /* The instructions are well-emulated on direct mmu. */
7293 if (vcpu->arch.mmu->direct_map) {
7294 unsigned int indirect_shadow_pages;
7296 write_lock(&vcpu->kvm->mmu_lock);
7297 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7298 write_unlock(&vcpu->kvm->mmu_lock);
7300 if (indirect_shadow_pages)
7301 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7307 * if emulation was due to access to shadowed page table
7308 * and it failed try to unshadow page and re-enter the
7309 * guest to let CPU execute the instruction.
7311 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7314 * If the access faults on its page table, it can not
7315 * be fixed by unprotecting shadow page and it should
7316 * be reported to userspace.
7318 return !write_fault_to_shadow_pgtable;
7321 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7322 gpa_t cr2_or_gpa, int emulation_type)
7324 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7325 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7327 last_retry_eip = vcpu->arch.last_retry_eip;
7328 last_retry_addr = vcpu->arch.last_retry_addr;
7331 * If the emulation is caused by #PF and it is non-page_table
7332 * writing instruction, it means the VM-EXIT is caused by shadow
7333 * page protected, we can zap the shadow page and retry this
7334 * instruction directly.
7336 * Note: if the guest uses a non-page-table modifying instruction
7337 * on the PDE that points to the instruction, then we will unmap
7338 * the instruction and go to an infinite loop. So, we cache the
7339 * last retried eip and the last fault address, if we meet the eip
7340 * and the address again, we can break out of the potential infinite
7343 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7345 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7348 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7349 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7352 if (x86_page_table_writing_insn(ctxt))
7355 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7358 vcpu->arch.last_retry_eip = ctxt->eip;
7359 vcpu->arch.last_retry_addr = cr2_or_gpa;
7361 if (!vcpu->arch.mmu->direct_map)
7362 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7364 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7369 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7370 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7372 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7374 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7375 /* This is a good place to trace that we are exiting SMM. */
7376 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7378 /* Process a latched INIT or SMI, if any. */
7379 kvm_make_request(KVM_REQ_EVENT, vcpu);
7382 kvm_mmu_reset_context(vcpu);
7385 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7394 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7395 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7400 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7402 struct kvm_run *kvm_run = vcpu->run;
7404 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7405 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7406 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7407 kvm_run->debug.arch.exception = DB_VECTOR;
7408 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7411 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7415 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7417 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7420 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7425 * rflags is the old, "raw" value of the flags. The new value has
7426 * not been saved yet.
7428 * This is correct even for TF set by the guest, because "the
7429 * processor will not generate this exception after the instruction
7430 * that sets the TF flag".
7432 if (unlikely(rflags & X86_EFLAGS_TF))
7433 r = kvm_vcpu_do_singlestep(vcpu);
7436 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7438 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7440 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7441 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7442 struct kvm_run *kvm_run = vcpu->run;
7443 unsigned long eip = kvm_get_linear_rip(vcpu);
7444 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7445 vcpu->arch.guest_debug_dr7,
7449 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7450 kvm_run->debug.arch.pc = eip;
7451 kvm_run->debug.arch.exception = DB_VECTOR;
7452 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7458 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7459 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7460 unsigned long eip = kvm_get_linear_rip(vcpu);
7461 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7466 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7475 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7477 switch (ctxt->opcode_len) {
7484 case 0xe6: /* OUT */
7488 case 0x6c: /* INS */
7490 case 0x6e: /* OUTS */
7497 case 0x33: /* RDPMC */
7507 * Decode to be emulated instruction. Return EMULATION_OK if success.
7509 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7510 void *insn, int insn_len)
7512 int r = EMULATION_OK;
7513 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7515 init_emulate_ctxt(vcpu);
7518 * We will reenter on the same instruction since we do not set
7519 * complete_userspace_io. This does not handle watchpoints yet,
7520 * those would be handled in the emulate_ops.
7522 if (!(emulation_type & EMULTYPE_SKIP) &&
7523 kvm_vcpu_check_breakpoint(vcpu, &r))
7526 ctxt->interruptibility = 0;
7527 ctxt->have_exception = false;
7528 ctxt->exception.vector = -1;
7529 ctxt->perm_ok = false;
7531 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7533 r = x86_decode_insn(ctxt, insn, insn_len);
7535 trace_kvm_emulate_insn_start(vcpu);
7536 ++vcpu->stat.insn_emulation;
7540 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7542 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7543 int emulation_type, void *insn, int insn_len)
7546 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7547 bool writeback = true;
7548 bool write_fault_to_spt;
7550 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7553 vcpu->arch.l1tf_flush_l1d = true;
7556 * Clear write_fault_to_shadow_pgtable here to ensure it is
7559 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7560 vcpu->arch.write_fault_to_shadow_pgtable = false;
7562 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7563 kvm_clear_exception_queue(vcpu);
7565 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7567 if (r != EMULATION_OK) {
7568 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7569 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7570 kvm_queue_exception(vcpu, UD_VECTOR);
7573 if (reexecute_instruction(vcpu, cr2_or_gpa,
7577 if (ctxt->have_exception) {
7579 * #UD should result in just EMULATION_FAILED, and trap-like
7580 * exception should not be encountered during decode.
7582 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7583 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7584 inject_emulated_exception(vcpu);
7587 return handle_emulation_failure(vcpu, emulation_type);
7591 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7592 !is_vmware_backdoor_opcode(ctxt)) {
7593 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7598 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7599 * for kvm_skip_emulated_instruction(). The caller is responsible for
7600 * updating interruptibility state and injecting single-step #DBs.
7602 if (emulation_type & EMULTYPE_SKIP) {
7603 kvm_rip_write(vcpu, ctxt->_eip);
7604 if (ctxt->eflags & X86_EFLAGS_RF)
7605 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7609 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7612 /* this is needed for vmware backdoor interface to work since it
7613 changes registers values during IO operation */
7614 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7615 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7616 emulator_invalidate_register_cache(ctxt);
7620 if (emulation_type & EMULTYPE_PF) {
7621 /* Save the faulting GPA (cr2) in the address field */
7622 ctxt->exception.address = cr2_or_gpa;
7624 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7625 if (vcpu->arch.mmu->direct_map) {
7626 ctxt->gpa_available = true;
7627 ctxt->gpa_val = cr2_or_gpa;
7630 /* Sanitize the address out of an abundance of paranoia. */
7631 ctxt->exception.address = 0;
7634 r = x86_emulate_insn(ctxt);
7636 if (r == EMULATION_INTERCEPTED)
7639 if (r == EMULATION_FAILED) {
7640 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7644 return handle_emulation_failure(vcpu, emulation_type);
7647 if (ctxt->have_exception) {
7649 if (inject_emulated_exception(vcpu))
7651 } else if (vcpu->arch.pio.count) {
7652 if (!vcpu->arch.pio.in) {
7653 /* FIXME: return into emulator if single-stepping. */
7654 vcpu->arch.pio.count = 0;
7657 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7660 } else if (vcpu->mmio_needed) {
7661 ++vcpu->stat.mmio_exits;
7663 if (!vcpu->mmio_is_write)
7666 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7667 } else if (r == EMULATION_RESTART)
7673 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7674 toggle_interruptibility(vcpu, ctxt->interruptibility);
7675 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7676 if (!ctxt->have_exception ||
7677 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7678 kvm_rip_write(vcpu, ctxt->eip);
7679 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7680 r = kvm_vcpu_do_singlestep(vcpu);
7681 if (kvm_x86_ops.update_emulated_instruction)
7682 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7683 __kvm_set_rflags(vcpu, ctxt->eflags);
7687 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7688 * do nothing, and it will be requested again as soon as
7689 * the shadow expires. But we still need to check here,
7690 * because POPF has no interrupt shadow.
7692 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7693 kvm_make_request(KVM_REQ_EVENT, vcpu);
7695 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7700 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7702 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7704 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7706 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7707 void *insn, int insn_len)
7709 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7711 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7713 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7715 vcpu->arch.pio.count = 0;
7719 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7721 vcpu->arch.pio.count = 0;
7723 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7726 return kvm_skip_emulated_instruction(vcpu);
7729 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7730 unsigned short port)
7732 unsigned long val = kvm_rax_read(vcpu);
7733 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7739 * Workaround userspace that relies on old KVM behavior of %rip being
7740 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7743 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7744 vcpu->arch.complete_userspace_io =
7745 complete_fast_pio_out_port_0x7e;
7746 kvm_skip_emulated_instruction(vcpu);
7748 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7749 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7754 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7758 /* We should only ever be called with arch.pio.count equal to 1 */
7759 BUG_ON(vcpu->arch.pio.count != 1);
7761 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7762 vcpu->arch.pio.count = 0;
7766 /* For size less than 4 we merge, else we zero extend */
7767 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7770 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7771 * the copy and tracing
7773 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7774 kvm_rax_write(vcpu, val);
7776 return kvm_skip_emulated_instruction(vcpu);
7779 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7780 unsigned short port)
7785 /* For size less than 4 we merge, else we zero extend */
7786 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7788 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7790 kvm_rax_write(vcpu, val);
7794 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7795 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7800 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7805 ret = kvm_fast_pio_in(vcpu, size, port);
7807 ret = kvm_fast_pio_out(vcpu, size, port);
7808 return ret && kvm_skip_emulated_instruction(vcpu);
7810 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7812 static int kvmclock_cpu_down_prep(unsigned int cpu)
7814 __this_cpu_write(cpu_tsc_khz, 0);
7818 static void tsc_khz_changed(void *data)
7820 struct cpufreq_freqs *freq = data;
7821 unsigned long khz = 0;
7825 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7826 khz = cpufreq_quick_get(raw_smp_processor_id());
7829 __this_cpu_write(cpu_tsc_khz, khz);
7832 #ifdef CONFIG_X86_64
7833 static void kvm_hyperv_tsc_notifier(void)
7836 struct kvm_vcpu *vcpu;
7838 unsigned long flags;
7840 mutex_lock(&kvm_lock);
7841 list_for_each_entry(kvm, &vm_list, vm_list)
7842 kvm_make_mclock_inprogress_request(kvm);
7844 hyperv_stop_tsc_emulation();
7846 /* TSC frequency always matches when on Hyper-V */
7847 for_each_present_cpu(cpu)
7848 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7849 kvm_max_guest_tsc_khz = tsc_khz;
7851 list_for_each_entry(kvm, &vm_list, vm_list) {
7852 struct kvm_arch *ka = &kvm->arch;
7854 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
7855 pvclock_update_vm_gtod_copy(kvm);
7856 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
7858 kvm_for_each_vcpu(cpu, vcpu, kvm)
7859 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7861 kvm_for_each_vcpu(cpu, vcpu, kvm)
7862 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7864 mutex_unlock(&kvm_lock);
7868 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7871 struct kvm_vcpu *vcpu;
7872 int i, send_ipi = 0;
7875 * We allow guests to temporarily run on slowing clocks,
7876 * provided we notify them after, or to run on accelerating
7877 * clocks, provided we notify them before. Thus time never
7880 * However, we have a problem. We can't atomically update
7881 * the frequency of a given CPU from this function; it is
7882 * merely a notifier, which can be called from any CPU.
7883 * Changing the TSC frequency at arbitrary points in time
7884 * requires a recomputation of local variables related to
7885 * the TSC for each VCPU. We must flag these local variables
7886 * to be updated and be sure the update takes place with the
7887 * new frequency before any guests proceed.
7889 * Unfortunately, the combination of hotplug CPU and frequency
7890 * change creates an intractable locking scenario; the order
7891 * of when these callouts happen is undefined with respect to
7892 * CPU hotplug, and they can race with each other. As such,
7893 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7894 * undefined; you can actually have a CPU frequency change take
7895 * place in between the computation of X and the setting of the
7896 * variable. To protect against this problem, all updates of
7897 * the per_cpu tsc_khz variable are done in an interrupt
7898 * protected IPI, and all callers wishing to update the value
7899 * must wait for a synchronous IPI to complete (which is trivial
7900 * if the caller is on the CPU already). This establishes the
7901 * necessary total order on variable updates.
7903 * Note that because a guest time update may take place
7904 * anytime after the setting of the VCPU's request bit, the
7905 * correct TSC value must be set before the request. However,
7906 * to ensure the update actually makes it to any guest which
7907 * starts running in hardware virtualization between the set
7908 * and the acquisition of the spinlock, we must also ping the
7909 * CPU after setting the request bit.
7913 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7915 mutex_lock(&kvm_lock);
7916 list_for_each_entry(kvm, &vm_list, vm_list) {
7917 kvm_for_each_vcpu(i, vcpu, kvm) {
7918 if (vcpu->cpu != cpu)
7920 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7921 if (vcpu->cpu != raw_smp_processor_id())
7925 mutex_unlock(&kvm_lock);
7927 if (freq->old < freq->new && send_ipi) {
7929 * We upscale the frequency. Must make the guest
7930 * doesn't see old kvmclock values while running with
7931 * the new frequency, otherwise we risk the guest sees
7932 * time go backwards.
7934 * In case we update the frequency for another cpu
7935 * (which might be in guest context) send an interrupt
7936 * to kick the cpu out of guest context. Next time
7937 * guest context is entered kvmclock will be updated,
7938 * so the guest will not see stale values.
7940 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7944 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7947 struct cpufreq_freqs *freq = data;
7950 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7952 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7955 for_each_cpu(cpu, freq->policy->cpus)
7956 __kvmclock_cpufreq_notifier(freq, cpu);
7961 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7962 .notifier_call = kvmclock_cpufreq_notifier
7965 static int kvmclock_cpu_online(unsigned int cpu)
7967 tsc_khz_changed(NULL);
7971 static void kvm_timer_init(void)
7973 max_tsc_khz = tsc_khz;
7975 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7976 #ifdef CONFIG_CPU_FREQ
7977 struct cpufreq_policy *policy;
7981 policy = cpufreq_cpu_get(cpu);
7983 if (policy->cpuinfo.max_freq)
7984 max_tsc_khz = policy->cpuinfo.max_freq;
7985 cpufreq_cpu_put(policy);
7989 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7990 CPUFREQ_TRANSITION_NOTIFIER);
7993 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7994 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7997 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7998 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8000 int kvm_is_in_guest(void)
8002 return __this_cpu_read(current_vcpu) != NULL;
8005 static int kvm_is_user_mode(void)
8009 if (__this_cpu_read(current_vcpu))
8010 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8012 return user_mode != 0;
8015 static unsigned long kvm_get_guest_ip(void)
8017 unsigned long ip = 0;
8019 if (__this_cpu_read(current_vcpu))
8020 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8025 static void kvm_handle_intel_pt_intr(void)
8027 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8029 kvm_make_request(KVM_REQ_PMI, vcpu);
8030 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8031 (unsigned long *)&vcpu->arch.pmu.global_status);
8034 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8035 .is_in_guest = kvm_is_in_guest,
8036 .is_user_mode = kvm_is_user_mode,
8037 .get_guest_ip = kvm_get_guest_ip,
8038 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8041 #ifdef CONFIG_X86_64
8042 static void pvclock_gtod_update_fn(struct work_struct *work)
8046 struct kvm_vcpu *vcpu;
8049 mutex_lock(&kvm_lock);
8050 list_for_each_entry(kvm, &vm_list, vm_list)
8051 kvm_for_each_vcpu(i, vcpu, kvm)
8052 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8053 atomic_set(&kvm_guest_has_master_clock, 0);
8054 mutex_unlock(&kvm_lock);
8057 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8060 * Notification about pvclock gtod data update.
8062 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8065 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8066 struct timekeeper *tk = priv;
8068 update_pvclock_gtod(tk);
8070 /* disable master clock if host does not trust, or does not
8071 * use, TSC based clocksource.
8073 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8074 atomic_read(&kvm_guest_has_master_clock) != 0)
8075 queue_work(system_long_wq, &pvclock_gtod_work);
8080 static struct notifier_block pvclock_gtod_notifier = {
8081 .notifier_call = pvclock_gtod_notify,
8085 int kvm_arch_init(void *opaque)
8087 struct kvm_x86_init_ops *ops = opaque;
8090 if (kvm_x86_ops.hardware_enable) {
8091 printk(KERN_ERR "kvm: already loaded the other module\n");
8096 if (!ops->cpu_has_kvm_support()) {
8097 pr_err_ratelimited("kvm: no hardware support\n");
8101 if (ops->disabled_by_bios()) {
8102 pr_err_ratelimited("kvm: disabled by bios\n");
8108 * KVM explicitly assumes that the guest has an FPU and
8109 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8110 * vCPU's FPU state as a fxregs_state struct.
8112 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8113 printk(KERN_ERR "kvm: inadequate fpu\n");
8119 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8120 __alignof__(struct fpu), SLAB_ACCOUNT,
8122 if (!x86_fpu_cache) {
8123 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8127 x86_emulator_cache = kvm_alloc_emulator_cache();
8128 if (!x86_emulator_cache) {
8129 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8130 goto out_free_x86_fpu_cache;
8133 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8134 if (!user_return_msrs) {
8135 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8136 goto out_free_x86_emulator_cache;
8139 r = kvm_mmu_module_init();
8141 goto out_free_percpu;
8145 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8147 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8148 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8149 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8152 if (pi_inject_timer == -1)
8153 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8154 #ifdef CONFIG_X86_64
8155 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8157 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8158 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8164 free_percpu(user_return_msrs);
8165 out_free_x86_emulator_cache:
8166 kmem_cache_destroy(x86_emulator_cache);
8167 out_free_x86_fpu_cache:
8168 kmem_cache_destroy(x86_fpu_cache);
8173 void kvm_arch_exit(void)
8175 #ifdef CONFIG_X86_64
8176 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8177 clear_hv_tscchange_cb();
8180 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8182 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8183 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8184 CPUFREQ_TRANSITION_NOTIFIER);
8185 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8186 #ifdef CONFIG_X86_64
8187 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8189 kvm_x86_ops.hardware_enable = NULL;
8190 kvm_mmu_module_exit();
8191 free_percpu(user_return_msrs);
8192 kmem_cache_destroy(x86_fpu_cache);
8193 #ifdef CONFIG_KVM_XEN
8194 static_key_deferred_flush(&kvm_xen_enabled);
8195 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8199 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8201 ++vcpu->stat.halt_exits;
8202 if (lapic_in_kernel(vcpu)) {
8203 vcpu->arch.mp_state = state;
8206 vcpu->run->exit_reason = reason;
8211 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8213 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8215 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8217 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8219 int ret = kvm_skip_emulated_instruction(vcpu);
8221 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8222 * KVM_EXIT_DEBUG here.
8224 return kvm_vcpu_halt(vcpu) && ret;
8226 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8228 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8230 int ret = kvm_skip_emulated_instruction(vcpu);
8232 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8234 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8236 #ifdef CONFIG_X86_64
8237 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8238 unsigned long clock_type)
8240 struct kvm_clock_pairing clock_pairing;
8241 struct timespec64 ts;
8245 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8246 return -KVM_EOPNOTSUPP;
8248 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8249 return -KVM_EOPNOTSUPP;
8251 clock_pairing.sec = ts.tv_sec;
8252 clock_pairing.nsec = ts.tv_nsec;
8253 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8254 clock_pairing.flags = 0;
8255 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8258 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8259 sizeof(struct kvm_clock_pairing)))
8267 * kvm_pv_kick_cpu_op: Kick a vcpu.
8269 * @apicid - apicid of vcpu to be kicked.
8271 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8273 struct kvm_lapic_irq lapic_irq;
8275 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8276 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8277 lapic_irq.level = 0;
8278 lapic_irq.dest_id = apicid;
8279 lapic_irq.msi_redir_hint = false;
8281 lapic_irq.delivery_mode = APIC_DM_REMRD;
8282 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8285 bool kvm_apicv_activated(struct kvm *kvm)
8287 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8289 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8291 void kvm_apicv_init(struct kvm *kvm, bool enable)
8294 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8295 &kvm->arch.apicv_inhibit_reasons);
8297 set_bit(APICV_INHIBIT_REASON_DISABLE,
8298 &kvm->arch.apicv_inhibit_reasons);
8300 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8302 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8304 struct kvm_vcpu *target = NULL;
8305 struct kvm_apic_map *map;
8307 vcpu->stat.directed_yield_attempted++;
8310 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8312 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8313 target = map->phys_map[dest_id]->vcpu;
8317 if (!target || !READ_ONCE(target->ready))
8320 /* Ignore requests to yield to self */
8324 if (kvm_vcpu_yield_to(target) <= 0)
8327 vcpu->stat.directed_yield_successful++;
8333 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8335 unsigned long nr, a0, a1, a2, a3, ret;
8338 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8339 return kvm_xen_hypercall(vcpu);
8341 if (kvm_hv_hypercall_enabled(vcpu))
8342 return kvm_hv_hypercall(vcpu);
8344 nr = kvm_rax_read(vcpu);
8345 a0 = kvm_rbx_read(vcpu);
8346 a1 = kvm_rcx_read(vcpu);
8347 a2 = kvm_rdx_read(vcpu);
8348 a3 = kvm_rsi_read(vcpu);
8350 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8352 op_64_bit = is_64_bit_mode(vcpu);
8361 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8369 case KVM_HC_VAPIC_POLL_IRQ:
8372 case KVM_HC_KICK_CPU:
8373 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8376 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8377 kvm_sched_yield(vcpu, a1);
8380 #ifdef CONFIG_X86_64
8381 case KVM_HC_CLOCK_PAIRING:
8382 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8385 case KVM_HC_SEND_IPI:
8386 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8389 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8391 case KVM_HC_SCHED_YIELD:
8392 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8395 kvm_sched_yield(vcpu, a0);
8405 kvm_rax_write(vcpu, ret);
8407 ++vcpu->stat.hypercalls;
8408 return kvm_skip_emulated_instruction(vcpu);
8410 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8412 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8414 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8415 char instruction[3];
8416 unsigned long rip = kvm_rip_read(vcpu);
8418 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8420 return emulator_write_emulated(ctxt, rip, instruction, 3,
8424 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8426 return vcpu->run->request_interrupt_window &&
8427 likely(!pic_in_kernel(vcpu->kvm));
8430 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8432 struct kvm_run *kvm_run = vcpu->run;
8435 * if_flag is obsolete and useless, so do not bother
8436 * setting it for SEV-ES guests. Userspace can just
8437 * use kvm_run->ready_for_interrupt_injection.
8439 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8440 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8442 kvm_run->cr8 = kvm_get_cr8(vcpu);
8443 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8444 kvm_run->ready_for_interrupt_injection =
8445 pic_in_kernel(vcpu->kvm) ||
8446 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8449 kvm_run->flags |= KVM_RUN_X86_SMM;
8452 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8456 if (!kvm_x86_ops.update_cr8_intercept)
8459 if (!lapic_in_kernel(vcpu))
8462 if (vcpu->arch.apicv_active)
8465 if (!vcpu->arch.apic->vapic_addr)
8466 max_irr = kvm_lapic_find_highest_irr(vcpu);
8473 tpr = kvm_lapic_get_cr8(vcpu);
8475 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8479 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8481 if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
8484 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8485 kvm_x86_ops.nested_ops->triple_fault(vcpu);
8489 return kvm_x86_ops.nested_ops->check_events(vcpu);
8492 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8494 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8495 vcpu->arch.exception.error_code = false;
8496 static_call(kvm_x86_queue_exception)(vcpu);
8499 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8502 bool can_inject = true;
8504 /* try to reinject previous events if any */
8506 if (vcpu->arch.exception.injected) {
8507 kvm_inject_exception(vcpu);
8511 * Do not inject an NMI or interrupt if there is a pending
8512 * exception. Exceptions and interrupts are recognized at
8513 * instruction boundaries, i.e. the start of an instruction.
8514 * Trap-like exceptions, e.g. #DB, have higher priority than
8515 * NMIs and interrupts, i.e. traps are recognized before an
8516 * NMI/interrupt that's pending on the same instruction.
8517 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8518 * priority, but are only generated (pended) during instruction
8519 * execution, i.e. a pending fault-like exception means the
8520 * fault occurred on the *previous* instruction and must be
8521 * serviced prior to recognizing any new events in order to
8522 * fully complete the previous instruction.
8524 else if (!vcpu->arch.exception.pending) {
8525 if (vcpu->arch.nmi_injected) {
8526 static_call(kvm_x86_set_nmi)(vcpu);
8528 } else if (vcpu->arch.interrupt.injected) {
8529 static_call(kvm_x86_set_irq)(vcpu);
8534 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8535 vcpu->arch.exception.pending);
8538 * Call check_nested_events() even if we reinjected a previous event
8539 * in order for caller to determine if it should require immediate-exit
8540 * from L2 to L1 due to pending L1 events which require exit
8543 if (is_guest_mode(vcpu)) {
8544 r = kvm_check_nested_events(vcpu);
8549 /* try to inject new event if pending */
8550 if (vcpu->arch.exception.pending) {
8551 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8552 vcpu->arch.exception.has_error_code,
8553 vcpu->arch.exception.error_code);
8555 vcpu->arch.exception.pending = false;
8556 vcpu->arch.exception.injected = true;
8558 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8559 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8562 if (vcpu->arch.exception.nr == DB_VECTOR) {
8563 kvm_deliver_exception_payload(vcpu);
8564 if (vcpu->arch.dr7 & DR7_GD) {
8565 vcpu->arch.dr7 &= ~DR7_GD;
8566 kvm_update_dr7(vcpu);
8570 kvm_inject_exception(vcpu);
8575 * Finally, inject interrupt events. If an event cannot be injected
8576 * due to architectural conditions (e.g. IF=0) a window-open exit
8577 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8578 * and can architecturally be injected, but we cannot do it right now:
8579 * an interrupt could have arrived just now and we have to inject it
8580 * as a vmexit, or there could already an event in the queue, which is
8581 * indicated by can_inject. In that case we request an immediate exit
8582 * in order to make progress and get back here for another iteration.
8583 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8585 if (vcpu->arch.smi_pending) {
8586 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8590 vcpu->arch.smi_pending = false;
8591 ++vcpu->arch.smi_count;
8595 static_call(kvm_x86_enable_smi_window)(vcpu);
8598 if (vcpu->arch.nmi_pending) {
8599 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8603 --vcpu->arch.nmi_pending;
8604 vcpu->arch.nmi_injected = true;
8605 static_call(kvm_x86_set_nmi)(vcpu);
8607 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8609 if (vcpu->arch.nmi_pending)
8610 static_call(kvm_x86_enable_nmi_window)(vcpu);
8613 if (kvm_cpu_has_injectable_intr(vcpu)) {
8614 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8618 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8619 static_call(kvm_x86_set_irq)(vcpu);
8620 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8622 if (kvm_cpu_has_injectable_intr(vcpu))
8623 static_call(kvm_x86_enable_irq_window)(vcpu);
8626 if (is_guest_mode(vcpu) &&
8627 kvm_x86_ops.nested_ops->hv_timer_pending &&
8628 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8629 *req_immediate_exit = true;
8631 WARN_ON(vcpu->arch.exception.pending);
8635 *req_immediate_exit = true;
8639 static void process_nmi(struct kvm_vcpu *vcpu)
8644 * x86 is limited to one NMI running, and one NMI pending after it.
8645 * If an NMI is already in progress, limit further NMIs to just one.
8646 * Otherwise, allow two (and we'll inject the first one immediately).
8648 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8651 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8652 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8653 kvm_make_request(KVM_REQ_EVENT, vcpu);
8656 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8659 flags |= seg->g << 23;
8660 flags |= seg->db << 22;
8661 flags |= seg->l << 21;
8662 flags |= seg->avl << 20;
8663 flags |= seg->present << 15;
8664 flags |= seg->dpl << 13;
8665 flags |= seg->s << 12;
8666 flags |= seg->type << 8;
8670 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8672 struct kvm_segment seg;
8675 kvm_get_segment(vcpu, &seg, n);
8676 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8679 offset = 0x7f84 + n * 12;
8681 offset = 0x7f2c + (n - 3) * 12;
8683 put_smstate(u32, buf, offset + 8, seg.base);
8684 put_smstate(u32, buf, offset + 4, seg.limit);
8685 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8688 #ifdef CONFIG_X86_64
8689 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8691 struct kvm_segment seg;
8695 kvm_get_segment(vcpu, &seg, n);
8696 offset = 0x7e00 + n * 16;
8698 flags = enter_smm_get_segment_flags(&seg) >> 8;
8699 put_smstate(u16, buf, offset, seg.selector);
8700 put_smstate(u16, buf, offset + 2, flags);
8701 put_smstate(u32, buf, offset + 4, seg.limit);
8702 put_smstate(u64, buf, offset + 8, seg.base);
8706 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8709 struct kvm_segment seg;
8713 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8714 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8715 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8716 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8718 for (i = 0; i < 8; i++)
8719 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
8721 kvm_get_dr(vcpu, 6, &val);
8722 put_smstate(u32, buf, 0x7fcc, (u32)val);
8723 kvm_get_dr(vcpu, 7, &val);
8724 put_smstate(u32, buf, 0x7fc8, (u32)val);
8726 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8727 put_smstate(u32, buf, 0x7fc4, seg.selector);
8728 put_smstate(u32, buf, 0x7f64, seg.base);
8729 put_smstate(u32, buf, 0x7f60, seg.limit);
8730 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8732 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8733 put_smstate(u32, buf, 0x7fc0, seg.selector);
8734 put_smstate(u32, buf, 0x7f80, seg.base);
8735 put_smstate(u32, buf, 0x7f7c, seg.limit);
8736 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8738 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8739 put_smstate(u32, buf, 0x7f74, dt.address);
8740 put_smstate(u32, buf, 0x7f70, dt.size);
8742 static_call(kvm_x86_get_idt)(vcpu, &dt);
8743 put_smstate(u32, buf, 0x7f58, dt.address);
8744 put_smstate(u32, buf, 0x7f54, dt.size);
8746 for (i = 0; i < 6; i++)
8747 enter_smm_save_seg_32(vcpu, buf, i);
8749 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8752 put_smstate(u32, buf, 0x7efc, 0x00020000);
8753 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8756 #ifdef CONFIG_X86_64
8757 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8760 struct kvm_segment seg;
8764 for (i = 0; i < 16; i++)
8765 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
8767 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8768 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8770 kvm_get_dr(vcpu, 6, &val);
8771 put_smstate(u64, buf, 0x7f68, val);
8772 kvm_get_dr(vcpu, 7, &val);
8773 put_smstate(u64, buf, 0x7f60, val);
8775 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8776 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8777 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8779 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8782 put_smstate(u32, buf, 0x7efc, 0x00020064);
8784 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8786 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8787 put_smstate(u16, buf, 0x7e90, seg.selector);
8788 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8789 put_smstate(u32, buf, 0x7e94, seg.limit);
8790 put_smstate(u64, buf, 0x7e98, seg.base);
8792 static_call(kvm_x86_get_idt)(vcpu, &dt);
8793 put_smstate(u32, buf, 0x7e84, dt.size);
8794 put_smstate(u64, buf, 0x7e88, dt.address);
8796 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8797 put_smstate(u16, buf, 0x7e70, seg.selector);
8798 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8799 put_smstate(u32, buf, 0x7e74, seg.limit);
8800 put_smstate(u64, buf, 0x7e78, seg.base);
8802 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8803 put_smstate(u32, buf, 0x7e64, dt.size);
8804 put_smstate(u64, buf, 0x7e68, dt.address);
8806 for (i = 0; i < 6; i++)
8807 enter_smm_save_seg_64(vcpu, buf, i);
8811 static void enter_smm(struct kvm_vcpu *vcpu)
8813 struct kvm_segment cs, ds;
8818 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8819 memset(buf, 0, 512);
8820 #ifdef CONFIG_X86_64
8821 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8822 enter_smm_save_state_64(vcpu, buf);
8825 enter_smm_save_state_32(vcpu, buf);
8828 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8829 * vCPU state (e.g. leave guest mode) after we've saved the state into
8830 * the SMM state-save area.
8832 static_call(kvm_x86_pre_enter_smm)(vcpu, buf);
8834 vcpu->arch.hflags |= HF_SMM_MASK;
8835 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8837 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
8838 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8840 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
8842 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8843 kvm_rip_write(vcpu, 0x8000);
8845 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8846 static_call(kvm_x86_set_cr0)(vcpu, cr0);
8847 vcpu->arch.cr0 = cr0;
8849 static_call(kvm_x86_set_cr4)(vcpu, 0);
8851 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8852 dt.address = dt.size = 0;
8853 static_call(kvm_x86_set_idt)(vcpu, &dt);
8855 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8857 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8858 cs.base = vcpu->arch.smbase;
8863 cs.limit = ds.limit = 0xffffffff;
8864 cs.type = ds.type = 0x3;
8865 cs.dpl = ds.dpl = 0;
8870 cs.avl = ds.avl = 0;
8871 cs.present = ds.present = 1;
8872 cs.unusable = ds.unusable = 0;
8873 cs.padding = ds.padding = 0;
8875 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8876 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8877 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8878 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8879 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8880 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8882 #ifdef CONFIG_X86_64
8883 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8884 static_call(kvm_x86_set_efer)(vcpu, 0);
8887 kvm_update_cpuid_runtime(vcpu);
8888 kvm_mmu_reset_context(vcpu);
8891 static void process_smi(struct kvm_vcpu *vcpu)
8893 vcpu->arch.smi_pending = true;
8894 kvm_make_request(KVM_REQ_EVENT, vcpu);
8897 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8898 unsigned long *vcpu_bitmap)
8902 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8904 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8905 NULL, vcpu_bitmap, cpus);
8907 free_cpumask_var(cpus);
8910 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8912 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8915 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8917 if (!lapic_in_kernel(vcpu))
8920 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8921 kvm_apic_update_apicv(vcpu);
8922 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
8924 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8927 * NOTE: Do not hold any lock prior to calling this.
8929 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8930 * locked, because it calls __x86_set_memory_region() which does
8931 * synchronize_srcu(&kvm->srcu).
8933 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8935 struct kvm_vcpu *except;
8936 unsigned long old, new, expected;
8938 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8939 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
8942 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8944 expected = new = old;
8946 __clear_bit(bit, &new);
8948 __set_bit(bit, &new);
8951 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8952 } while (old != expected);
8957 trace_kvm_apicv_update_request(activate, bit);
8958 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8959 static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
8962 * Sending request to update APICV for all other vcpus,
8963 * while update the calling vcpu immediately instead of
8964 * waiting for another #VMEXIT to handle the request.
8966 except = kvm_get_running_vcpu();
8967 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8970 kvm_vcpu_update_apicv(except);
8972 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8974 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8976 if (!kvm_apic_present(vcpu))
8979 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8981 if (irqchip_split(vcpu->kvm))
8982 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8984 if (vcpu->arch.apicv_active)
8985 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
8986 if (ioapic_in_kernel(vcpu->kvm))
8987 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8990 if (is_guest_mode(vcpu))
8991 vcpu->arch.load_eoi_exitmap_pending = true;
8993 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8996 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8998 u64 eoi_exit_bitmap[4];
9000 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9003 if (to_hv_vcpu(vcpu))
9004 bitmap_or((ulong *)eoi_exit_bitmap,
9005 vcpu->arch.ioapic_handled_vectors,
9006 to_hv_synic(vcpu)->vec_bitmap, 256);
9008 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9011 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9012 unsigned long start, unsigned long end)
9014 unsigned long apic_address;
9017 * The physical address of apic access page is stored in the VMCS.
9018 * Update it when it becomes invalid.
9020 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9021 if (start <= apic_address && apic_address < end)
9022 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9025 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9027 if (!lapic_in_kernel(vcpu))
9030 if (!kvm_x86_ops.set_apic_access_page_addr)
9033 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9036 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9038 smp_send_reschedule(vcpu->cpu);
9040 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9043 * Returns 1 to let vcpu_run() continue the guest execution loop without
9044 * exiting to the userspace. Otherwise, the value will be returned to the
9047 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9051 dm_request_for_irq_injection(vcpu) &&
9052 kvm_cpu_accept_dm_intr(vcpu);
9053 fastpath_t exit_fastpath;
9055 bool req_immediate_exit = false;
9057 /* Forbid vmenter if vcpu dirty ring is soft-full */
9058 if (unlikely(vcpu->kvm->dirty_ring_size &&
9059 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9060 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9061 trace_kvm_dirty_ring_exit(vcpu);
9066 if (kvm_request_pending(vcpu)) {
9067 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9068 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9073 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9074 kvm_mmu_unload(vcpu);
9075 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9076 __kvm_migrate_timers(vcpu);
9077 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9078 kvm_gen_update_masterclock(vcpu->kvm);
9079 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9080 kvm_gen_kvmclock_update(vcpu);
9081 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9082 r = kvm_guest_time_update(vcpu);
9086 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9087 kvm_mmu_sync_roots(vcpu);
9088 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9089 kvm_mmu_load_pgd(vcpu);
9090 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9091 kvm_vcpu_flush_tlb_all(vcpu);
9093 /* Flushing all ASIDs flushes the current ASID... */
9094 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9096 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9097 kvm_vcpu_flush_tlb_current(vcpu);
9098 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
9099 kvm_vcpu_flush_tlb_guest(vcpu);
9101 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9102 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9106 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9107 if (is_guest_mode(vcpu)) {
9108 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9110 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9111 vcpu->mmio_needed = 0;
9116 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9117 /* Page is swapped out. Do synthetic halt */
9118 vcpu->arch.apf.halted = true;
9122 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9123 record_steal_time(vcpu);
9124 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9126 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9128 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9129 kvm_pmu_handle_event(vcpu);
9130 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9131 kvm_pmu_deliver_pmi(vcpu);
9132 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9133 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9134 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9135 vcpu->arch.ioapic_handled_vectors)) {
9136 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9137 vcpu->run->eoi.vector =
9138 vcpu->arch.pending_ioapic_eoi;
9143 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9144 vcpu_scan_ioapic(vcpu);
9145 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9146 vcpu_load_eoi_exitmap(vcpu);
9147 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9148 kvm_vcpu_reload_apic_access_page(vcpu);
9149 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9150 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9151 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9155 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9156 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9157 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9161 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9162 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9164 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9165 vcpu->run->hyperv = hv_vcpu->exit;
9171 * KVM_REQ_HV_STIMER has to be processed after
9172 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9173 * depend on the guest clock being up-to-date
9175 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9176 kvm_hv_process_stimers(vcpu);
9177 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9178 kvm_vcpu_update_apicv(vcpu);
9179 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9180 kvm_check_async_pf_completion(vcpu);
9181 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9182 static_call(kvm_x86_msr_filter_changed)(vcpu);
9184 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9185 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9188 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9189 kvm_xen_has_interrupt(vcpu)) {
9190 ++vcpu->stat.req_event;
9191 kvm_apic_accept_events(vcpu);
9192 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9197 inject_pending_event(vcpu, &req_immediate_exit);
9199 static_call(kvm_x86_enable_irq_window)(vcpu);
9201 if (kvm_lapic_enabled(vcpu)) {
9202 update_cr8_intercept(vcpu);
9203 kvm_lapic_sync_to_vapic(vcpu);
9207 r = kvm_mmu_reload(vcpu);
9209 goto cancel_injection;
9214 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9217 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9218 * IPI are then delayed after guest entry, which ensures that they
9219 * result in virtual interrupt delivery.
9221 local_irq_disable();
9222 vcpu->mode = IN_GUEST_MODE;
9224 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9227 * 1) We should set ->mode before checking ->requests. Please see
9228 * the comment in kvm_vcpu_exiting_guest_mode().
9230 * 2) For APICv, we should set ->mode before checking PID.ON. This
9231 * pairs with the memory barrier implicit in pi_test_and_set_on
9232 * (see vmx_deliver_posted_interrupt).
9234 * 3) This also orders the write to mode from any reads to the page
9235 * tables done while the VCPU is running. Please see the comment
9236 * in kvm_flush_remote_tlbs.
9238 smp_mb__after_srcu_read_unlock();
9241 * This handles the case where a posted interrupt was
9242 * notified with kvm_vcpu_kick.
9244 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9245 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9247 if (kvm_vcpu_exit_request(vcpu)) {
9248 vcpu->mode = OUTSIDE_GUEST_MODE;
9252 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9254 goto cancel_injection;
9257 if (req_immediate_exit) {
9258 kvm_make_request(KVM_REQ_EVENT, vcpu);
9259 static_call(kvm_x86_request_immediate_exit)(vcpu);
9262 fpregs_assert_state_consistent();
9263 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9264 switch_fpu_return();
9266 if (unlikely(vcpu->arch.switch_db_regs)) {
9268 set_debugreg(vcpu->arch.eff_db[0], 0);
9269 set_debugreg(vcpu->arch.eff_db[1], 1);
9270 set_debugreg(vcpu->arch.eff_db[2], 2);
9271 set_debugreg(vcpu->arch.eff_db[3], 3);
9272 set_debugreg(vcpu->arch.dr6, 6);
9273 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9277 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9278 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9281 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9282 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9286 if (vcpu->arch.apicv_active)
9287 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9291 * Do this here before restoring debug registers on the host. And
9292 * since we do this before handling the vmexit, a DR access vmexit
9293 * can (a) read the correct value of the debug registers, (b) set
9294 * KVM_DEBUGREG_WONT_EXIT again.
9296 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9297 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9298 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9299 kvm_update_dr0123(vcpu);
9300 kvm_update_dr7(vcpu);
9301 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9305 * If the guest has used debug registers, at least dr7
9306 * will be disabled while returning to the host.
9307 * If we don't have active breakpoints in the host, we don't
9308 * care about the messed up debug address registers. But if
9309 * we have some of them active, restore the old state.
9311 if (hw_breakpoint_active())
9312 hw_breakpoint_restore();
9314 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9315 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9317 vcpu->mode = OUTSIDE_GUEST_MODE;
9320 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9323 * Consume any pending interrupts, including the possible source of
9324 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9325 * An instruction is required after local_irq_enable() to fully unblock
9326 * interrupts on processors that implement an interrupt shadow, the
9327 * stat.exits increment will do nicely.
9329 kvm_before_interrupt(vcpu);
9332 local_irq_disable();
9333 kvm_after_interrupt(vcpu);
9335 if (lapic_in_kernel(vcpu)) {
9336 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9337 if (delta != S64_MIN) {
9338 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9339 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9346 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9349 * Profile KVM exit RIPs:
9351 if (unlikely(prof_on == KVM_PROFILING)) {
9352 unsigned long rip = kvm_rip_read(vcpu);
9353 profile_hit(KVM_PROFILING, (void *)rip);
9356 if (unlikely(vcpu->arch.tsc_always_catchup))
9357 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9359 if (vcpu->arch.apic_attention)
9360 kvm_lapic_sync_from_vapic(vcpu);
9362 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9366 if (req_immediate_exit)
9367 kvm_make_request(KVM_REQ_EVENT, vcpu);
9368 static_call(kvm_x86_cancel_injection)(vcpu);
9369 if (unlikely(vcpu->arch.apic_attention))
9370 kvm_lapic_sync_from_vapic(vcpu);
9375 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9377 if (!kvm_arch_vcpu_runnable(vcpu) &&
9378 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9379 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9380 kvm_vcpu_block(vcpu);
9381 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9383 if (kvm_x86_ops.post_block)
9384 static_call(kvm_x86_post_block)(vcpu);
9386 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9390 kvm_apic_accept_events(vcpu);
9391 switch(vcpu->arch.mp_state) {
9392 case KVM_MP_STATE_HALTED:
9393 case KVM_MP_STATE_AP_RESET_HOLD:
9394 vcpu->arch.pv.pv_unhalted = false;
9395 vcpu->arch.mp_state =
9396 KVM_MP_STATE_RUNNABLE;
9398 case KVM_MP_STATE_RUNNABLE:
9399 vcpu->arch.apf.halted = false;
9401 case KVM_MP_STATE_INIT_RECEIVED:
9409 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9411 if (is_guest_mode(vcpu))
9412 kvm_check_nested_events(vcpu);
9414 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9415 !vcpu->arch.apf.halted);
9418 static int vcpu_run(struct kvm_vcpu *vcpu)
9421 struct kvm *kvm = vcpu->kvm;
9423 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9424 vcpu->arch.l1tf_flush_l1d = true;
9427 if (kvm_vcpu_running(vcpu)) {
9428 r = vcpu_enter_guest(vcpu);
9430 r = vcpu_block(kvm, vcpu);
9436 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
9437 if (kvm_cpu_has_pending_timer(vcpu))
9438 kvm_inject_pending_timer_irqs(vcpu);
9440 if (dm_request_for_irq_injection(vcpu) &&
9441 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9443 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9444 ++vcpu->stat.request_irq_exits;
9448 if (__xfer_to_guest_mode_work_pending()) {
9449 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9450 r = xfer_to_guest_mode_handle_work(vcpu);
9453 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9457 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9462 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9466 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9467 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9468 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9472 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9474 BUG_ON(!vcpu->arch.pio.count);
9476 return complete_emulated_io(vcpu);
9480 * Implements the following, as a state machine:
9484 * for each mmio piece in the fragment
9492 * for each mmio piece in the fragment
9497 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9499 struct kvm_run *run = vcpu->run;
9500 struct kvm_mmio_fragment *frag;
9503 BUG_ON(!vcpu->mmio_needed);
9505 /* Complete previous fragment */
9506 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9507 len = min(8u, frag->len);
9508 if (!vcpu->mmio_is_write)
9509 memcpy(frag->data, run->mmio.data, len);
9511 if (frag->len <= 8) {
9512 /* Switch to the next fragment. */
9514 vcpu->mmio_cur_fragment++;
9516 /* Go forward to the next mmio piece. */
9522 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9523 vcpu->mmio_needed = 0;
9525 /* FIXME: return into emulator if single-stepping. */
9526 if (vcpu->mmio_is_write)
9528 vcpu->mmio_read_completed = 1;
9529 return complete_emulated_io(vcpu);
9532 run->exit_reason = KVM_EXIT_MMIO;
9533 run->mmio.phys_addr = frag->gpa;
9534 if (vcpu->mmio_is_write)
9535 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9536 run->mmio.len = min(8u, frag->len);
9537 run->mmio.is_write = vcpu->mmio_is_write;
9538 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9542 static void kvm_save_current_fpu(struct fpu *fpu)
9545 * If the target FPU state is not resident in the CPU registers, just
9546 * memcpy() from current, else save CPU state directly to the target.
9548 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9549 memcpy(&fpu->state, ¤t->thread.fpu.state,
9550 fpu_kernel_xstate_size);
9552 copy_fpregs_to_fpstate(fpu);
9555 /* Swap (qemu) user FPU context for the guest FPU context. */
9556 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9560 kvm_save_current_fpu(vcpu->arch.user_fpu);
9563 * Guests with protected state can't have it set by the hypervisor,
9564 * so skip trying to set it.
9566 if (vcpu->arch.guest_fpu)
9567 /* PKRU is separately restored in kvm_x86_ops.run. */
9568 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9569 ~XFEATURE_MASK_PKRU);
9571 fpregs_mark_activate();
9577 /* When vcpu_run ends, restore user space FPU context. */
9578 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9583 * Guests with protected state can't have it read by the hypervisor,
9584 * so skip trying to save it.
9586 if (vcpu->arch.guest_fpu)
9587 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9589 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9591 fpregs_mark_activate();
9594 ++vcpu->stat.fpu_reload;
9598 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9600 struct kvm_run *kvm_run = vcpu->run;
9604 kvm_sigset_activate(vcpu);
9606 kvm_load_guest_fpu(vcpu);
9608 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9609 if (kvm_run->immediate_exit) {
9613 kvm_vcpu_block(vcpu);
9614 kvm_apic_accept_events(vcpu);
9615 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9617 if (signal_pending(current)) {
9619 kvm_run->exit_reason = KVM_EXIT_INTR;
9620 ++vcpu->stat.signal_exits;
9625 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9630 if (kvm_run->kvm_dirty_regs) {
9631 r = sync_regs(vcpu);
9636 /* re-sync apic's tpr */
9637 if (!lapic_in_kernel(vcpu)) {
9638 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9644 if (unlikely(vcpu->arch.complete_userspace_io)) {
9645 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9646 vcpu->arch.complete_userspace_io = NULL;
9651 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9653 if (kvm_run->immediate_exit)
9659 kvm_put_guest_fpu(vcpu);
9660 if (kvm_run->kvm_valid_regs)
9662 post_kvm_run_save(vcpu);
9663 kvm_sigset_deactivate(vcpu);
9669 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9671 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9673 * We are here if userspace calls get_regs() in the middle of
9674 * instruction emulation. Registers state needs to be copied
9675 * back from emulation context to vcpu. Userspace shouldn't do
9676 * that usually, but some bad designed PV devices (vmware
9677 * backdoor interface) need this to work
9679 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9680 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9682 regs->rax = kvm_rax_read(vcpu);
9683 regs->rbx = kvm_rbx_read(vcpu);
9684 regs->rcx = kvm_rcx_read(vcpu);
9685 regs->rdx = kvm_rdx_read(vcpu);
9686 regs->rsi = kvm_rsi_read(vcpu);
9687 regs->rdi = kvm_rdi_read(vcpu);
9688 regs->rsp = kvm_rsp_read(vcpu);
9689 regs->rbp = kvm_rbp_read(vcpu);
9690 #ifdef CONFIG_X86_64
9691 regs->r8 = kvm_r8_read(vcpu);
9692 regs->r9 = kvm_r9_read(vcpu);
9693 regs->r10 = kvm_r10_read(vcpu);
9694 regs->r11 = kvm_r11_read(vcpu);
9695 regs->r12 = kvm_r12_read(vcpu);
9696 regs->r13 = kvm_r13_read(vcpu);
9697 regs->r14 = kvm_r14_read(vcpu);
9698 regs->r15 = kvm_r15_read(vcpu);
9701 regs->rip = kvm_rip_read(vcpu);
9702 regs->rflags = kvm_get_rflags(vcpu);
9705 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9708 __get_regs(vcpu, regs);
9713 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9715 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9716 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9718 kvm_rax_write(vcpu, regs->rax);
9719 kvm_rbx_write(vcpu, regs->rbx);
9720 kvm_rcx_write(vcpu, regs->rcx);
9721 kvm_rdx_write(vcpu, regs->rdx);
9722 kvm_rsi_write(vcpu, regs->rsi);
9723 kvm_rdi_write(vcpu, regs->rdi);
9724 kvm_rsp_write(vcpu, regs->rsp);
9725 kvm_rbp_write(vcpu, regs->rbp);
9726 #ifdef CONFIG_X86_64
9727 kvm_r8_write(vcpu, regs->r8);
9728 kvm_r9_write(vcpu, regs->r9);
9729 kvm_r10_write(vcpu, regs->r10);
9730 kvm_r11_write(vcpu, regs->r11);
9731 kvm_r12_write(vcpu, regs->r12);
9732 kvm_r13_write(vcpu, regs->r13);
9733 kvm_r14_write(vcpu, regs->r14);
9734 kvm_r15_write(vcpu, regs->r15);
9737 kvm_rip_write(vcpu, regs->rip);
9738 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9740 vcpu->arch.exception.pending = false;
9742 kvm_make_request(KVM_REQ_EVENT, vcpu);
9745 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9748 __set_regs(vcpu, regs);
9753 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9755 struct kvm_segment cs;
9757 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9761 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9763 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9767 if (vcpu->arch.guest_state_protected)
9768 goto skip_protected_regs;
9770 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9771 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9772 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9773 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9774 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9775 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9777 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9778 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9780 static_call(kvm_x86_get_idt)(vcpu, &dt);
9781 sregs->idt.limit = dt.size;
9782 sregs->idt.base = dt.address;
9783 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9784 sregs->gdt.limit = dt.size;
9785 sregs->gdt.base = dt.address;
9787 sregs->cr2 = vcpu->arch.cr2;
9788 sregs->cr3 = kvm_read_cr3(vcpu);
9790 skip_protected_regs:
9791 sregs->cr0 = kvm_read_cr0(vcpu);
9792 sregs->cr4 = kvm_read_cr4(vcpu);
9793 sregs->cr8 = kvm_get_cr8(vcpu);
9794 sregs->efer = vcpu->arch.efer;
9795 sregs->apic_base = kvm_get_apic_base(vcpu);
9797 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9799 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9800 set_bit(vcpu->arch.interrupt.nr,
9801 (unsigned long *)sregs->interrupt_bitmap);
9804 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9805 struct kvm_sregs *sregs)
9808 __get_sregs(vcpu, sregs);
9813 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9814 struct kvm_mp_state *mp_state)
9817 if (kvm_mpx_supported())
9818 kvm_load_guest_fpu(vcpu);
9820 kvm_apic_accept_events(vcpu);
9821 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
9822 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
9823 vcpu->arch.pv.pv_unhalted)
9824 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9826 mp_state->mp_state = vcpu->arch.mp_state;
9828 if (kvm_mpx_supported())
9829 kvm_put_guest_fpu(vcpu);
9834 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9835 struct kvm_mp_state *mp_state)
9841 if (!lapic_in_kernel(vcpu) &&
9842 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9846 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9847 * INIT state; latched init should be reported using
9848 * KVM_SET_VCPU_EVENTS, so reject it here.
9850 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9851 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9852 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9855 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9856 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9857 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9859 vcpu->arch.mp_state = mp_state->mp_state;
9860 kvm_make_request(KVM_REQ_EVENT, vcpu);
9868 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9869 int reason, bool has_error_code, u32 error_code)
9871 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9874 init_emulate_ctxt(vcpu);
9876 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9877 has_error_code, error_code);
9879 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9880 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9881 vcpu->run->internal.ndata = 0;
9885 kvm_rip_write(vcpu, ctxt->eip);
9886 kvm_set_rflags(vcpu, ctxt->eflags);
9889 EXPORT_SYMBOL_GPL(kvm_task_switch);
9891 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9893 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9895 * When EFER.LME and CR0.PG are set, the processor is in
9896 * 64-bit mode (though maybe in a 32-bit code segment).
9897 * CR4.PAE and EFER.LMA must be set.
9899 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
9901 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
9905 * Not in 64-bit mode: EFER.LMA is clear and the code
9906 * segment cannot be 64-bit.
9908 if (sregs->efer & EFER_LMA || sregs->cs.l)
9912 return kvm_is_valid_cr4(vcpu, sregs->cr4);
9915 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9917 struct msr_data apic_base_msr;
9918 int mmu_reset_needed = 0;
9919 int pending_vec, max_bits, idx;
9923 if (!kvm_is_valid_sregs(vcpu, sregs))
9926 apic_base_msr.data = sregs->apic_base;
9927 apic_base_msr.host_initiated = true;
9928 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9931 if (vcpu->arch.guest_state_protected)
9932 goto skip_protected_regs;
9934 dt.size = sregs->idt.limit;
9935 dt.address = sregs->idt.base;
9936 static_call(kvm_x86_set_idt)(vcpu, &dt);
9937 dt.size = sregs->gdt.limit;
9938 dt.address = sregs->gdt.base;
9939 static_call(kvm_x86_set_gdt)(vcpu, &dt);
9941 vcpu->arch.cr2 = sregs->cr2;
9942 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9943 vcpu->arch.cr3 = sregs->cr3;
9944 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9946 kvm_set_cr8(vcpu, sregs->cr8);
9948 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9949 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
9951 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9952 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
9953 vcpu->arch.cr0 = sregs->cr0;
9955 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9956 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
9958 idx = srcu_read_lock(&vcpu->kvm->srcu);
9959 if (is_pae_paging(vcpu)) {
9960 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9961 mmu_reset_needed = 1;
9963 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9965 if (mmu_reset_needed)
9966 kvm_mmu_reset_context(vcpu);
9968 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9969 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9970 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9971 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9972 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9973 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9975 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9976 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9978 update_cr8_intercept(vcpu);
9980 /* Older userspace won't unhalt the vcpu on reset. */
9981 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9982 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9984 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9986 skip_protected_regs:
9987 max_bits = KVM_NR_INTERRUPTS;
9988 pending_vec = find_first_bit(
9989 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9990 if (pending_vec < max_bits) {
9991 kvm_queue_interrupt(vcpu, pending_vec, false);
9992 pr_debug("Set back pending irq %d\n", pending_vec);
9995 kvm_make_request(KVM_REQ_EVENT, vcpu);
10002 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10003 struct kvm_sregs *sregs)
10008 ret = __set_sregs(vcpu, sregs);
10013 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10014 struct kvm_guest_debug *dbg)
10016 unsigned long rflags;
10019 if (vcpu->arch.guest_state_protected)
10024 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10026 if (vcpu->arch.exception.pending)
10028 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10029 kvm_queue_exception(vcpu, DB_VECTOR);
10031 kvm_queue_exception(vcpu, BP_VECTOR);
10035 * Read rflags as long as potentially injected trace flags are still
10038 rflags = kvm_get_rflags(vcpu);
10040 vcpu->guest_debug = dbg->control;
10041 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10042 vcpu->guest_debug = 0;
10044 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10045 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10046 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10047 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10049 for (i = 0; i < KVM_NR_DB_REGS; i++)
10050 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10052 kvm_update_dr7(vcpu);
10054 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10055 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
10056 get_segment_base(vcpu, VCPU_SREG_CS);
10059 * Trigger an rflags update that will inject or remove the trace
10062 kvm_set_rflags(vcpu, rflags);
10064 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10074 * Translate a guest virtual address to a guest physical address.
10076 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10077 struct kvm_translation *tr)
10079 unsigned long vaddr = tr->linear_address;
10085 idx = srcu_read_lock(&vcpu->kvm->srcu);
10086 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10087 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10088 tr->physical_address = gpa;
10089 tr->valid = gpa != UNMAPPED_GVA;
10097 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10099 struct fxregs_state *fxsave;
10101 if (!vcpu->arch.guest_fpu)
10106 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10107 memcpy(fpu->fpr, fxsave->st_space, 128);
10108 fpu->fcw = fxsave->cwd;
10109 fpu->fsw = fxsave->swd;
10110 fpu->ftwx = fxsave->twd;
10111 fpu->last_opcode = fxsave->fop;
10112 fpu->last_ip = fxsave->rip;
10113 fpu->last_dp = fxsave->rdp;
10114 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10120 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10122 struct fxregs_state *fxsave;
10124 if (!vcpu->arch.guest_fpu)
10129 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10131 memcpy(fxsave->st_space, fpu->fpr, 128);
10132 fxsave->cwd = fpu->fcw;
10133 fxsave->swd = fpu->fsw;
10134 fxsave->twd = fpu->ftwx;
10135 fxsave->fop = fpu->last_opcode;
10136 fxsave->rip = fpu->last_ip;
10137 fxsave->rdp = fpu->last_dp;
10138 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10144 static void store_regs(struct kvm_vcpu *vcpu)
10146 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10148 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10149 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10151 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10152 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10154 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10155 kvm_vcpu_ioctl_x86_get_vcpu_events(
10156 vcpu, &vcpu->run->s.regs.events);
10159 static int sync_regs(struct kvm_vcpu *vcpu)
10161 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
10164 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10165 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10166 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10168 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10169 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10171 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10173 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10174 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10175 vcpu, &vcpu->run->s.regs.events))
10177 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10183 static void fx_init(struct kvm_vcpu *vcpu)
10185 if (!vcpu->arch.guest_fpu)
10188 fpstate_init(&vcpu->arch.guest_fpu->state);
10189 if (boot_cpu_has(X86_FEATURE_XSAVES))
10190 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10191 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10194 * Ensure guest xcr0 is valid for loading
10196 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10198 vcpu->arch.cr0 |= X86_CR0_ET;
10201 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10203 if (vcpu->arch.guest_fpu) {
10204 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10205 vcpu->arch.guest_fpu = NULL;
10208 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10210 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10212 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10213 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10214 "guest TSC will not be reliable\n");
10219 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10224 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10225 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10227 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10229 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10231 r = kvm_mmu_create(vcpu);
10235 if (irqchip_in_kernel(vcpu->kvm)) {
10236 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10238 goto fail_mmu_destroy;
10239 if (kvm_apicv_activated(vcpu->kvm))
10240 vcpu->arch.apicv_active = true;
10242 static_branch_inc(&kvm_has_noapic_vcpu);
10246 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10248 goto fail_free_lapic;
10249 vcpu->arch.pio_data = page_address(page);
10251 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10252 GFP_KERNEL_ACCOUNT);
10253 if (!vcpu->arch.mce_banks)
10254 goto fail_free_pio_data;
10255 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10257 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10258 GFP_KERNEL_ACCOUNT))
10259 goto fail_free_mce_banks;
10261 if (!alloc_emulate_ctxt(vcpu))
10262 goto free_wbinvd_dirty_mask;
10264 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10265 GFP_KERNEL_ACCOUNT);
10266 if (!vcpu->arch.user_fpu) {
10267 pr_err("kvm: failed to allocate userspace's fpu\n");
10268 goto free_emulate_ctxt;
10271 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10272 GFP_KERNEL_ACCOUNT);
10273 if (!vcpu->arch.guest_fpu) {
10274 pr_err("kvm: failed to allocate vcpu's fpu\n");
10275 goto free_user_fpu;
10279 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10280 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10282 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10284 kvm_async_pf_hash_reset(vcpu);
10285 kvm_pmu_init(vcpu);
10287 vcpu->arch.pending_external_vector = -1;
10288 vcpu->arch.preempted_in_kernel = false;
10290 r = static_call(kvm_x86_vcpu_create)(vcpu);
10292 goto free_guest_fpu;
10294 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10295 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10296 kvm_vcpu_mtrr_init(vcpu);
10298 kvm_vcpu_reset(vcpu, false);
10299 kvm_init_mmu(vcpu, false);
10304 kvm_free_guest_fpu(vcpu);
10306 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10308 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10309 free_wbinvd_dirty_mask:
10310 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10311 fail_free_mce_banks:
10312 kfree(vcpu->arch.mce_banks);
10313 fail_free_pio_data:
10314 free_page((unsigned long)vcpu->arch.pio_data);
10316 kvm_free_lapic(vcpu);
10318 kvm_mmu_destroy(vcpu);
10322 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10324 struct kvm *kvm = vcpu->kvm;
10326 if (mutex_lock_killable(&vcpu->mutex))
10329 kvm_synchronize_tsc(vcpu, 0);
10332 /* poll control enabled by default */
10333 vcpu->arch.msr_kvm_poll_control = 1;
10335 mutex_unlock(&vcpu->mutex);
10337 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10338 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10339 KVMCLOCK_SYNC_PERIOD);
10342 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10344 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10347 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10349 kvmclock_reset(vcpu);
10351 static_call(kvm_x86_vcpu_free)(vcpu);
10353 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10354 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10355 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10356 kvm_free_guest_fpu(vcpu);
10358 kvm_hv_vcpu_uninit(vcpu);
10359 kvm_pmu_destroy(vcpu);
10360 kfree(vcpu->arch.mce_banks);
10361 kvm_free_lapic(vcpu);
10362 idx = srcu_read_lock(&vcpu->kvm->srcu);
10363 kvm_mmu_destroy(vcpu);
10364 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10365 free_page((unsigned long)vcpu->arch.pio_data);
10366 kvfree(vcpu->arch.cpuid_entries);
10367 if (!lapic_in_kernel(vcpu))
10368 static_branch_dec(&kvm_has_noapic_vcpu);
10371 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10373 kvm_lapic_reset(vcpu, init_event);
10375 vcpu->arch.hflags = 0;
10377 vcpu->arch.smi_pending = 0;
10378 vcpu->arch.smi_count = 0;
10379 atomic_set(&vcpu->arch.nmi_queued, 0);
10380 vcpu->arch.nmi_pending = 0;
10381 vcpu->arch.nmi_injected = false;
10382 kvm_clear_interrupt_queue(vcpu);
10383 kvm_clear_exception_queue(vcpu);
10385 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10386 kvm_update_dr0123(vcpu);
10387 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10388 vcpu->arch.dr7 = DR7_FIXED_1;
10389 kvm_update_dr7(vcpu);
10391 vcpu->arch.cr2 = 0;
10393 kvm_make_request(KVM_REQ_EVENT, vcpu);
10394 vcpu->arch.apf.msr_en_val = 0;
10395 vcpu->arch.apf.msr_int_val = 0;
10396 vcpu->arch.st.msr_val = 0;
10398 kvmclock_reset(vcpu);
10400 kvm_clear_async_pf_completion_queue(vcpu);
10401 kvm_async_pf_hash_reset(vcpu);
10402 vcpu->arch.apf.halted = false;
10404 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10405 void *mpx_state_buffer;
10408 * To avoid have the INIT path from kvm_apic_has_events() that be
10409 * called with loaded FPU and does not let userspace fix the state.
10412 kvm_put_guest_fpu(vcpu);
10413 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10415 if (mpx_state_buffer)
10416 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10417 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10419 if (mpx_state_buffer)
10420 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10422 kvm_load_guest_fpu(vcpu);
10426 kvm_pmu_reset(vcpu);
10427 vcpu->arch.smbase = 0x30000;
10429 vcpu->arch.msr_misc_features_enables = 0;
10431 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10434 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10435 vcpu->arch.regs_avail = ~0;
10436 vcpu->arch.regs_dirty = ~0;
10438 vcpu->arch.ia32_xss = 0;
10440 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10443 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10445 struct kvm_segment cs;
10447 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10448 cs.selector = vector << 8;
10449 cs.base = vector << 12;
10450 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10451 kvm_rip_write(vcpu, 0);
10453 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10455 int kvm_arch_hardware_enable(void)
10458 struct kvm_vcpu *vcpu;
10463 bool stable, backwards_tsc = false;
10465 kvm_user_return_msr_cpu_online();
10466 ret = static_call(kvm_x86_hardware_enable)();
10470 local_tsc = rdtsc();
10471 stable = !kvm_check_tsc_unstable();
10472 list_for_each_entry(kvm, &vm_list, vm_list) {
10473 kvm_for_each_vcpu(i, vcpu, kvm) {
10474 if (!stable && vcpu->cpu == smp_processor_id())
10475 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10476 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10477 backwards_tsc = true;
10478 if (vcpu->arch.last_host_tsc > max_tsc)
10479 max_tsc = vcpu->arch.last_host_tsc;
10485 * Sometimes, even reliable TSCs go backwards. This happens on
10486 * platforms that reset TSC during suspend or hibernate actions, but
10487 * maintain synchronization. We must compensate. Fortunately, we can
10488 * detect that condition here, which happens early in CPU bringup,
10489 * before any KVM threads can be running. Unfortunately, we can't
10490 * bring the TSCs fully up to date with real time, as we aren't yet far
10491 * enough into CPU bringup that we know how much real time has actually
10492 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10493 * variables that haven't been updated yet.
10495 * So we simply find the maximum observed TSC above, then record the
10496 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10497 * the adjustment will be applied. Note that we accumulate
10498 * adjustments, in case multiple suspend cycles happen before some VCPU
10499 * gets a chance to run again. In the event that no KVM threads get a
10500 * chance to run, we will miss the entire elapsed period, as we'll have
10501 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10502 * loose cycle time. This isn't too big a deal, since the loss will be
10503 * uniform across all VCPUs (not to mention the scenario is extremely
10504 * unlikely). It is possible that a second hibernate recovery happens
10505 * much faster than a first, causing the observed TSC here to be
10506 * smaller; this would require additional padding adjustment, which is
10507 * why we set last_host_tsc to the local tsc observed here.
10509 * N.B. - this code below runs only on platforms with reliable TSC,
10510 * as that is the only way backwards_tsc is set above. Also note
10511 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10512 * have the same delta_cyc adjustment applied if backwards_tsc
10513 * is detected. Note further, this adjustment is only done once,
10514 * as we reset last_host_tsc on all VCPUs to stop this from being
10515 * called multiple times (one for each physical CPU bringup).
10517 * Platforms with unreliable TSCs don't have to deal with this, they
10518 * will be compensated by the logic in vcpu_load, which sets the TSC to
10519 * catchup mode. This will catchup all VCPUs to real time, but cannot
10520 * guarantee that they stay in perfect synchronization.
10522 if (backwards_tsc) {
10523 u64 delta_cyc = max_tsc - local_tsc;
10524 list_for_each_entry(kvm, &vm_list, vm_list) {
10525 kvm->arch.backwards_tsc_observed = true;
10526 kvm_for_each_vcpu(i, vcpu, kvm) {
10527 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10528 vcpu->arch.last_host_tsc = local_tsc;
10529 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10533 * We have to disable TSC offset matching.. if you were
10534 * booting a VM while issuing an S4 host suspend....
10535 * you may have some problem. Solving this issue is
10536 * left as an exercise to the reader.
10538 kvm->arch.last_tsc_nsec = 0;
10539 kvm->arch.last_tsc_write = 0;
10546 void kvm_arch_hardware_disable(void)
10548 static_call(kvm_x86_hardware_disable)();
10549 drop_user_return_notifiers();
10552 int kvm_arch_hardware_setup(void *opaque)
10554 struct kvm_x86_init_ops *ops = opaque;
10557 rdmsrl_safe(MSR_EFER, &host_efer);
10559 if (boot_cpu_has(X86_FEATURE_XSAVES))
10560 rdmsrl(MSR_IA32_XSS, host_xss);
10562 r = ops->hardware_setup();
10566 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10567 kvm_ops_static_call_update();
10569 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10572 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10573 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10574 #undef __kvm_cpu_cap_has
10576 if (kvm_has_tsc_control) {
10578 * Make sure the user can only configure tsc_khz values that
10579 * fit into a signed integer.
10580 * A min value is not calculated because it will always
10581 * be 1 on all machines.
10583 u64 max = min(0x7fffffffULL,
10584 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10585 kvm_max_guest_tsc_khz = max;
10587 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10590 kvm_init_msr_list();
10594 void kvm_arch_hardware_unsetup(void)
10596 static_call(kvm_x86_hardware_unsetup)();
10599 int kvm_arch_check_processor_compat(void *opaque)
10601 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10602 struct kvm_x86_init_ops *ops = opaque;
10604 WARN_ON(!irqs_disabled());
10606 if (__cr4_reserved_bits(cpu_has, c) !=
10607 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10610 return ops->check_processor_compatibility();
10613 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10615 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10617 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10619 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10621 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10624 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10625 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10627 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10629 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10631 vcpu->arch.l1tf_flush_l1d = true;
10632 if (pmu->version && unlikely(pmu->event_count)) {
10633 pmu->need_cleanup = true;
10634 kvm_make_request(KVM_REQ_PMU, vcpu);
10636 static_call(kvm_x86_sched_in)(vcpu, cpu);
10639 void kvm_arch_free_vm(struct kvm *kvm)
10641 kfree(to_kvm_hv(kvm)->hv_pa_pg);
10646 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10651 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10652 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10653 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10654 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10655 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10656 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10658 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10659 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10660 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10661 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10662 &kvm->arch.irq_sources_bitmap);
10664 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10665 mutex_init(&kvm->arch.apic_map_lock);
10666 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10668 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10669 pvclock_update_vm_gtod_copy(kvm);
10671 kvm->arch.guest_can_read_msr_platform_info = true;
10673 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10674 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10676 kvm_hv_init_vm(kvm);
10677 kvm_page_track_init(kvm);
10678 kvm_mmu_init_vm(kvm);
10680 return static_call(kvm_x86_vm_init)(kvm);
10683 int kvm_arch_post_init_vm(struct kvm *kvm)
10685 return kvm_mmu_post_init_vm(kvm);
10688 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10691 kvm_mmu_unload(vcpu);
10695 static void kvm_free_vcpus(struct kvm *kvm)
10698 struct kvm_vcpu *vcpu;
10701 * Unpin any mmu pages first.
10703 kvm_for_each_vcpu(i, vcpu, kvm) {
10704 kvm_clear_async_pf_completion_queue(vcpu);
10705 kvm_unload_vcpu_mmu(vcpu);
10707 kvm_for_each_vcpu(i, vcpu, kvm)
10708 kvm_vcpu_destroy(vcpu);
10710 mutex_lock(&kvm->lock);
10711 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10712 kvm->vcpus[i] = NULL;
10714 atomic_set(&kvm->online_vcpus, 0);
10715 mutex_unlock(&kvm->lock);
10718 void kvm_arch_sync_events(struct kvm *kvm)
10720 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10721 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10725 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
10728 * __x86_set_memory_region: Setup KVM internal memory slot
10730 * @kvm: the kvm pointer to the VM.
10731 * @id: the slot ID to setup.
10732 * @gpa: the GPA to install the slot (unused when @size == 0).
10733 * @size: the size of the slot. Set to zero to uninstall a slot.
10735 * This function helps to setup a KVM internal memory slot. Specify
10736 * @size > 0 to install a new slot, while @size == 0 to uninstall a
10737 * slot. The return code can be one of the following:
10739 * HVA: on success (uninstall will return a bogus HVA)
10742 * The caller should always use IS_ERR() to check the return value
10743 * before use. Note, the KVM internal memory slots are guaranteed to
10744 * remain valid and unchanged until the VM is destroyed, i.e., the
10745 * GPA->HVA translation will not change. However, the HVA is a user
10746 * address, i.e. its accessibility is not guaranteed, and must be
10747 * accessed via __copy_{to,from}_user().
10749 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
10753 unsigned long hva, old_npages;
10754 struct kvm_memslots *slots = kvm_memslots(kvm);
10755 struct kvm_memory_slot *slot;
10757 /* Called with kvm->slots_lock held. */
10758 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10759 return ERR_PTR_USR(-EINVAL);
10761 slot = id_to_memslot(slots, id);
10763 if (slot && slot->npages)
10764 return ERR_PTR_USR(-EEXIST);
10767 * MAP_SHARED to prevent internal slot pages from being moved
10770 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10771 MAP_SHARED | MAP_ANONYMOUS, 0);
10772 if (IS_ERR((void *)hva))
10773 return (void __user *)hva;
10775 if (!slot || !slot->npages)
10778 old_npages = slot->npages;
10779 hva = slot->userspace_addr;
10782 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10783 struct kvm_userspace_memory_region m;
10785 m.slot = id | (i << 16);
10787 m.guest_phys_addr = gpa;
10788 m.userspace_addr = hva;
10789 m.memory_size = size;
10790 r = __kvm_set_memory_region(kvm, &m);
10792 return ERR_PTR_USR(r);
10796 vm_munmap(hva, old_npages * PAGE_SIZE);
10798 return (void __user *)hva;
10800 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10802 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10804 kvm_mmu_pre_destroy_vm(kvm);
10807 void kvm_arch_destroy_vm(struct kvm *kvm)
10809 if (current->mm == kvm->mm) {
10811 * Free memory regions allocated on behalf of userspace,
10812 * unless the the memory map has changed due to process exit
10815 mutex_lock(&kvm->slots_lock);
10816 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10818 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10820 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10821 mutex_unlock(&kvm->slots_lock);
10823 static_call_cond(kvm_x86_vm_destroy)(kvm);
10824 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
10825 kvm_pic_destroy(kvm);
10826 kvm_ioapic_destroy(kvm);
10827 kvm_free_vcpus(kvm);
10828 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10829 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10830 kvm_mmu_uninit_vm(kvm);
10831 kvm_page_track_cleanup(kvm);
10832 kvm_xen_destroy_vm(kvm);
10833 kvm_hv_destroy_vm(kvm);
10836 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10840 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10841 kvfree(slot->arch.rmap[i]);
10842 slot->arch.rmap[i] = NULL;
10847 kvfree(slot->arch.lpage_info[i - 1]);
10848 slot->arch.lpage_info[i - 1] = NULL;
10851 kvm_page_track_free_memslot(slot);
10854 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10855 unsigned long npages)
10860 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
10861 * old arrays will be freed by __kvm_set_memory_region() if installing
10862 * the new memslot is successful.
10864 memset(&slot->arch, 0, sizeof(slot->arch));
10866 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10867 struct kvm_lpage_info *linfo;
10868 unsigned long ugfn;
10872 lpages = gfn_to_index(slot->base_gfn + npages - 1,
10873 slot->base_gfn, level) + 1;
10875 slot->arch.rmap[i] =
10876 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10877 GFP_KERNEL_ACCOUNT);
10878 if (!slot->arch.rmap[i])
10883 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10887 slot->arch.lpage_info[i - 1] = linfo;
10889 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10890 linfo[0].disallow_lpage = 1;
10891 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10892 linfo[lpages - 1].disallow_lpage = 1;
10893 ugfn = slot->userspace_addr >> PAGE_SHIFT;
10895 * If the gfn and userspace address are not aligned wrt each
10896 * other, disable large page support for this slot.
10898 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10901 for (j = 0; j < lpages; ++j)
10902 linfo[j].disallow_lpage = 1;
10906 if (kvm_page_track_create_memslot(slot, npages))
10912 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10913 kvfree(slot->arch.rmap[i]);
10914 slot->arch.rmap[i] = NULL;
10918 kvfree(slot->arch.lpage_info[i - 1]);
10919 slot->arch.lpage_info[i - 1] = NULL;
10924 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10926 struct kvm_vcpu *vcpu;
10930 * memslots->generation has been incremented.
10931 * mmio generation may have reached its maximum value.
10933 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10935 /* Force re-initialization of steal_time cache */
10936 kvm_for_each_vcpu(i, vcpu, kvm)
10937 kvm_vcpu_kick(vcpu);
10940 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10941 struct kvm_memory_slot *memslot,
10942 const struct kvm_userspace_memory_region *mem,
10943 enum kvm_mr_change change)
10945 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10946 return kvm_alloc_memslot_metadata(memslot,
10947 mem->memory_size >> PAGE_SHIFT);
10952 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
10954 struct kvm_arch *ka = &kvm->arch;
10956 if (!kvm_x86_ops.cpu_dirty_log_size)
10959 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
10960 (!enable && --ka->cpu_dirty_logging_count == 0))
10961 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
10963 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
10966 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10967 struct kvm_memory_slot *old,
10968 struct kvm_memory_slot *new,
10969 enum kvm_mr_change change)
10971 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
10974 * Update CPU dirty logging if dirty logging is being toggled. This
10975 * applies to all operations.
10977 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
10978 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
10981 * Nothing more to do for RO slots (which can't be dirtied and can't be
10982 * made writable) or CREATE/MOVE/DELETE of a slot.
10984 * For a memslot with dirty logging disabled:
10985 * CREATE: No dirty mappings will already exist.
10986 * MOVE/DELETE: The old mappings will already have been cleaned up by
10987 * kvm_arch_flush_shadow_memslot()
10989 * For a memslot with dirty logging enabled:
10990 * CREATE: No shadow pages exist, thus nothing to write-protect
10991 * and no dirty bits to clear.
10992 * MOVE/DELETE: The old mappings will already have been cleaned up by
10993 * kvm_arch_flush_shadow_memslot().
10995 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10999 * READONLY and non-flags changes were filtered out above, and the only
11000 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11001 * logging isn't being toggled on or off.
11003 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11006 if (!log_dirty_pages) {
11008 * Dirty logging tracks sptes in 4k granularity, meaning that
11009 * large sptes have to be split. If live migration succeeds,
11010 * the guest in the source machine will be destroyed and large
11011 * sptes will be created in the destination. However, if the
11012 * guest continues to run in the source machine (for example if
11013 * live migration fails), small sptes will remain around and
11014 * cause bad performance.
11016 * Scan sptes if dirty logging has been stopped, dropping those
11017 * which can be collapsed into a single large-page spte. Later
11018 * page faults will create the large-page sptes.
11020 kvm_mmu_zap_collapsible_sptes(kvm, new);
11022 /* By default, write-protect everything to log writes. */
11023 int level = PG_LEVEL_4K;
11025 if (kvm_x86_ops.cpu_dirty_log_size) {
11027 * Clear all dirty bits, unless pages are treated as
11028 * dirty from the get-go.
11030 if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
11031 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11034 * Write-protect large pages on write so that dirty
11035 * logging happens at 4k granularity. No need to
11036 * write-protect small SPTEs since write accesses are
11037 * logged by the CPU via dirty bits.
11039 level = PG_LEVEL_2M;
11040 } else if (kvm_dirty_log_manual_protect_and_init_set(kvm)) {
11042 * If we're with initial-all-set, we don't need
11043 * to write protect any small page because
11044 * they're reported as dirty already. However
11045 * we still need to write-protect huge pages
11046 * so that the page split can happen lazily on
11047 * the first write to the huge page.
11049 level = PG_LEVEL_2M;
11051 kvm_mmu_slot_remove_write_access(kvm, new, level);
11055 void kvm_arch_commit_memory_region(struct kvm *kvm,
11056 const struct kvm_userspace_memory_region *mem,
11057 struct kvm_memory_slot *old,
11058 const struct kvm_memory_slot *new,
11059 enum kvm_mr_change change)
11061 if (!kvm->arch.n_requested_mmu_pages)
11062 kvm_mmu_change_mmu_pages(kvm,
11063 kvm_mmu_calculate_default_mmu_pages(kvm));
11066 * FIXME: const-ify all uses of struct kvm_memory_slot.
11068 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
11070 /* Free the arrays associated with the old memslot. */
11071 if (change == KVM_MR_MOVE)
11072 kvm_arch_free_memslot(kvm, old);
11075 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11077 kvm_mmu_zap_all(kvm);
11080 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11081 struct kvm_memory_slot *slot)
11083 kvm_page_track_flush_slot(kvm, slot);
11086 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11088 return (is_guest_mode(vcpu) &&
11089 kvm_x86_ops.guest_apic_has_interrupt &&
11090 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11093 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11095 if (!list_empty_careful(&vcpu->async_pf.done))
11098 if (kvm_apic_has_events(vcpu))
11101 if (vcpu->arch.pv.pv_unhalted)
11104 if (vcpu->arch.exception.pending)
11107 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11108 (vcpu->arch.nmi_pending &&
11109 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11112 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11113 (vcpu->arch.smi_pending &&
11114 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11117 if (kvm_arch_interrupt_allowed(vcpu) &&
11118 (kvm_cpu_has_interrupt(vcpu) ||
11119 kvm_guest_apic_has_interrupt(vcpu)))
11122 if (kvm_hv_has_stimer_pending(vcpu))
11125 if (is_guest_mode(vcpu) &&
11126 kvm_x86_ops.nested_ops->hv_timer_pending &&
11127 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11133 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11135 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11138 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11140 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11146 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11148 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11151 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11152 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11153 kvm_test_request(KVM_REQ_EVENT, vcpu))
11156 return kvm_arch_dy_has_pending_interrupt(vcpu);
11159 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11161 if (vcpu->arch.guest_state_protected)
11164 return vcpu->arch.preempted_in_kernel;
11167 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11169 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11172 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11174 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11177 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11179 /* Can't read the RIP when guest state is protected, just return 0 */
11180 if (vcpu->arch.guest_state_protected)
11183 if (is_64_bit_mode(vcpu))
11184 return kvm_rip_read(vcpu);
11185 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11186 kvm_rip_read(vcpu));
11188 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11190 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11192 return kvm_get_linear_rip(vcpu) == linear_rip;
11194 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11196 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11198 unsigned long rflags;
11200 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11201 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11202 rflags &= ~X86_EFLAGS_TF;
11205 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11207 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11209 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11210 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11211 rflags |= X86_EFLAGS_TF;
11212 static_call(kvm_x86_set_rflags)(vcpu, rflags);
11215 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11217 __kvm_set_rflags(vcpu, rflags);
11218 kvm_make_request(KVM_REQ_EVENT, vcpu);
11220 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11222 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11226 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11230 r = kvm_mmu_reload(vcpu);
11234 if (!vcpu->arch.mmu->direct_map &&
11235 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11238 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11241 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11243 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11245 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11248 static inline u32 kvm_async_pf_next_probe(u32 key)
11250 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11253 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11255 u32 key = kvm_async_pf_hash_fn(gfn);
11257 while (vcpu->arch.apf.gfns[key] != ~0)
11258 key = kvm_async_pf_next_probe(key);
11260 vcpu->arch.apf.gfns[key] = gfn;
11263 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11266 u32 key = kvm_async_pf_hash_fn(gfn);
11268 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11269 (vcpu->arch.apf.gfns[key] != gfn &&
11270 vcpu->arch.apf.gfns[key] != ~0); i++)
11271 key = kvm_async_pf_next_probe(key);
11276 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11278 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11281 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11285 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11287 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11291 vcpu->arch.apf.gfns[i] = ~0;
11293 j = kvm_async_pf_next_probe(j);
11294 if (vcpu->arch.apf.gfns[j] == ~0)
11296 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11298 * k lies cyclically in ]i,j]
11300 * |....j i.k.| or |.k..j i...|
11302 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11303 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11308 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11310 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11312 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11316 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11318 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11320 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11321 &token, offset, sizeof(token));
11324 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11326 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11329 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11330 &val, offset, sizeof(val)))
11336 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11338 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11341 if (!kvm_pv_async_pf_enabled(vcpu) ||
11342 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11348 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11350 if (unlikely(!lapic_in_kernel(vcpu) ||
11351 kvm_event_needs_reinjection(vcpu) ||
11352 vcpu->arch.exception.pending))
11355 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11359 * If interrupts are off we cannot even use an artificial
11362 return kvm_arch_interrupt_allowed(vcpu);
11365 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11366 struct kvm_async_pf *work)
11368 struct x86_exception fault;
11370 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11371 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11373 if (kvm_can_deliver_async_pf(vcpu) &&
11374 !apf_put_user_notpresent(vcpu)) {
11375 fault.vector = PF_VECTOR;
11376 fault.error_code_valid = true;
11377 fault.error_code = 0;
11378 fault.nested_page_fault = false;
11379 fault.address = work->arch.token;
11380 fault.async_page_fault = true;
11381 kvm_inject_page_fault(vcpu, &fault);
11385 * It is not possible to deliver a paravirtualized asynchronous
11386 * page fault, but putting the guest in an artificial halt state
11387 * can be beneficial nevertheless: if an interrupt arrives, we
11388 * can deliver it timely and perhaps the guest will schedule
11389 * another process. When the instruction that triggered a page
11390 * fault is retried, hopefully the page will be ready in the host.
11392 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11397 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11398 struct kvm_async_pf *work)
11400 struct kvm_lapic_irq irq = {
11401 .delivery_mode = APIC_DM_FIXED,
11402 .vector = vcpu->arch.apf.vec
11405 if (work->wakeup_all)
11406 work->arch.token = ~0; /* broadcast wakeup */
11408 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11409 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11411 if ((work->wakeup_all || work->notpresent_injected) &&
11412 kvm_pv_async_pf_enabled(vcpu) &&
11413 !apf_put_user_ready(vcpu, work->arch.token)) {
11414 vcpu->arch.apf.pageready_pending = true;
11415 kvm_apic_set_irq(vcpu, &irq, NULL);
11418 vcpu->arch.apf.halted = false;
11419 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11422 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11424 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11425 if (!vcpu->arch.apf.pageready_pending)
11426 kvm_vcpu_kick(vcpu);
11429 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11431 if (!kvm_pv_async_pf_enabled(vcpu))
11434 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11437 void kvm_arch_start_assignment(struct kvm *kvm)
11439 atomic_inc(&kvm->arch.assigned_device_count);
11441 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11443 void kvm_arch_end_assignment(struct kvm *kvm)
11445 atomic_dec(&kvm->arch.assigned_device_count);
11447 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11449 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11451 return atomic_read(&kvm->arch.assigned_device_count);
11453 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11455 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11457 atomic_inc(&kvm->arch.noncoherent_dma_count);
11459 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11461 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11463 atomic_dec(&kvm->arch.noncoherent_dma_count);
11465 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11467 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11469 return atomic_read(&kvm->arch.noncoherent_dma_count);
11471 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11473 bool kvm_arch_has_irq_bypass(void)
11478 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11479 struct irq_bypass_producer *prod)
11481 struct kvm_kernel_irqfd *irqfd =
11482 container_of(cons, struct kvm_kernel_irqfd, consumer);
11485 irqfd->producer = prod;
11486 kvm_arch_start_assignment(irqfd->kvm);
11487 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11488 prod->irq, irqfd->gsi, 1);
11491 kvm_arch_end_assignment(irqfd->kvm);
11496 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11497 struct irq_bypass_producer *prod)
11500 struct kvm_kernel_irqfd *irqfd =
11501 container_of(cons, struct kvm_kernel_irqfd, consumer);
11503 WARN_ON(irqfd->producer != prod);
11504 irqfd->producer = NULL;
11507 * When producer of consumer is unregistered, we change back to
11508 * remapped mode, so we can re-use the current implementation
11509 * when the irq is masked/disabled or the consumer side (KVM
11510 * int this case doesn't want to receive the interrupts.
11512 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11514 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11515 " fails: %d\n", irqfd->consumer.token, ret);
11517 kvm_arch_end_assignment(irqfd->kvm);
11520 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11521 uint32_t guest_irq, bool set)
11523 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11526 bool kvm_vector_hashing_enabled(void)
11528 return vector_hashing;
11531 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11533 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11535 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11538 int kvm_spec_ctrl_test_value(u64 value)
11541 * test that setting IA32_SPEC_CTRL to given value
11542 * is allowed by the host processor
11546 unsigned long flags;
11549 local_irq_save(flags);
11551 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11553 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11556 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11558 local_irq_restore(flags);
11562 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11564 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11566 struct x86_exception fault;
11567 u32 access = error_code &
11568 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11570 if (!(error_code & PFERR_PRESENT_MASK) ||
11571 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11573 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11574 * tables probably do not match the TLB. Just proceed
11575 * with the error code that the processor gave.
11577 fault.vector = PF_VECTOR;
11578 fault.error_code_valid = true;
11579 fault.error_code = error_code;
11580 fault.nested_page_fault = false;
11581 fault.address = gva;
11583 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11585 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11588 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11589 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11590 * indicates whether exit to userspace is needed.
11592 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11593 struct x86_exception *e)
11595 if (r == X86EMUL_PROPAGATE_FAULT) {
11596 kvm_inject_emulated_page_fault(vcpu, e);
11601 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11602 * while handling a VMX instruction KVM could've handled the request
11603 * correctly by exiting to userspace and performing I/O but there
11604 * doesn't seem to be a real use-case behind such requests, just return
11605 * KVM_EXIT_INTERNAL_ERROR for now.
11607 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11608 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11609 vcpu->run->internal.ndata = 0;
11613 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11615 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11618 struct x86_exception e;
11620 unsigned long roots_to_free = 0;
11627 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11628 if (r != X86EMUL_CONTINUE)
11629 return kvm_handle_memory_failure(vcpu, r, &e);
11631 if (operand.pcid >> 12 != 0) {
11632 kvm_inject_gp(vcpu, 0);
11636 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11639 case INVPCID_TYPE_INDIV_ADDR:
11640 if ((!pcid_enabled && (operand.pcid != 0)) ||
11641 is_noncanonical_address(operand.gla, vcpu)) {
11642 kvm_inject_gp(vcpu, 0);
11645 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11646 return kvm_skip_emulated_instruction(vcpu);
11648 case INVPCID_TYPE_SINGLE_CTXT:
11649 if (!pcid_enabled && (operand.pcid != 0)) {
11650 kvm_inject_gp(vcpu, 0);
11654 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11655 kvm_mmu_sync_roots(vcpu);
11656 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11659 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11660 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11662 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11664 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11666 * If neither the current cr3 nor any of the prev_roots use the
11667 * given PCID, then nothing needs to be done here because a
11668 * resync will happen anyway before switching to any other CR3.
11671 return kvm_skip_emulated_instruction(vcpu);
11673 case INVPCID_TYPE_ALL_NON_GLOBAL:
11675 * Currently, KVM doesn't mark global entries in the shadow
11676 * page tables, so a non-global flush just degenerates to a
11677 * global flush. If needed, we could optimize this later by
11678 * keeping track of global entries in shadow page tables.
11682 case INVPCID_TYPE_ALL_INCL_GLOBAL:
11683 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
11684 return kvm_skip_emulated_instruction(vcpu);
11687 BUG(); /* We have already checked above that type <= 3 */
11690 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11692 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
11694 struct kvm_run *run = vcpu->run;
11695 struct kvm_mmio_fragment *frag;
11698 BUG_ON(!vcpu->mmio_needed);
11700 /* Complete previous fragment */
11701 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11702 len = min(8u, frag->len);
11703 if (!vcpu->mmio_is_write)
11704 memcpy(frag->data, run->mmio.data, len);
11706 if (frag->len <= 8) {
11707 /* Switch to the next fragment. */
11709 vcpu->mmio_cur_fragment++;
11711 /* Go forward to the next mmio piece. */
11717 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11718 vcpu->mmio_needed = 0;
11720 // VMG change, at this point, we're always done
11721 // RIP has already been advanced
11725 // More MMIO is needed
11726 run->mmio.phys_addr = frag->gpa;
11727 run->mmio.len = min(8u, frag->len);
11728 run->mmio.is_write = vcpu->mmio_is_write;
11729 if (run->mmio.is_write)
11730 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11731 run->exit_reason = KVM_EXIT_MMIO;
11733 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11738 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11742 struct kvm_mmio_fragment *frag;
11747 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11748 if (handled == bytes)
11755 /*TODO: Check if need to increment number of frags */
11756 frag = vcpu->mmio_fragments;
11757 vcpu->mmio_nr_fragments = 1;
11762 vcpu->mmio_needed = 1;
11763 vcpu->mmio_cur_fragment = 0;
11765 vcpu->run->mmio.phys_addr = gpa;
11766 vcpu->run->mmio.len = min(8u, frag->len);
11767 vcpu->run->mmio.is_write = 1;
11768 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
11769 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11771 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11775 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
11777 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11781 struct kvm_mmio_fragment *frag;
11786 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11787 if (handled == bytes)
11794 /*TODO: Check if need to increment number of frags */
11795 frag = vcpu->mmio_fragments;
11796 vcpu->mmio_nr_fragments = 1;
11801 vcpu->mmio_needed = 1;
11802 vcpu->mmio_cur_fragment = 0;
11804 vcpu->run->mmio.phys_addr = gpa;
11805 vcpu->run->mmio.len = min(8u, frag->len);
11806 vcpu->run->mmio.is_write = 0;
11807 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11809 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11813 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
11815 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
11817 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
11818 vcpu->arch.pio.count * vcpu->arch.pio.size);
11819 vcpu->arch.pio.count = 0;
11824 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
11825 unsigned int port, void *data, unsigned int count)
11829 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
11834 vcpu->arch.pio.count = 0;
11839 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
11840 unsigned int port, void *data, unsigned int count)
11844 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
11847 vcpu->arch.pio.count = 0;
11849 vcpu->arch.guest_ins_data = data;
11850 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
11856 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
11857 unsigned int port, void *data, unsigned int count,
11860 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
11861 : kvm_sev_es_outs(vcpu, size, port, data, count);
11863 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
11865 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
11866 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11867 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11868 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11869 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11870 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11871 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11872 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11873 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11874 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11875 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11876 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11877 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11878 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11879 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11880 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11881 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11882 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11883 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11884 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11885 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11886 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11887 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
11888 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
11889 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
11890 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
11891 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);