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"
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/sched/isolation.h>
58 #include <linux/mem_encrypt.h>
59 #include <linux/entry-kvm.h>
61 #include <trace/events/kvm.h>
63 #include <asm/debugreg.h>
67 #include <linux/kernel_stat.h>
68 #include <asm/fpu/internal.h> /* Ugh! */
69 #include <asm/pvclock.h>
70 #include <asm/div64.h>
71 #include <asm/irq_remapping.h>
72 #include <asm/mshyperv.h>
73 #include <asm/hypervisor.h>
74 #include <asm/tlbflush.h>
75 #include <asm/intel_pt.h>
76 #include <asm/emulate_prefix.h>
77 #include <clocksource/hyperv_timer.h>
79 #define CREATE_TRACE_POINTS
82 #define MAX_IO_MSRS 256
83 #define KVM_MAX_MCE_BANKS 32
84 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
85 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
87 #define emul_to_vcpu(ctxt) \
88 ((struct kvm_vcpu *)(ctxt)->vcpu)
91 * - enable syscall per default because its emulated by KVM
92 * - enable LME and LMA per default on 64 bit KVM
96 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
98 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
101 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
103 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
104 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
106 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
107 static void process_nmi(struct kvm_vcpu *vcpu);
108 static void enter_smm(struct kvm_vcpu *vcpu);
109 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
110 static void store_regs(struct kvm_vcpu *vcpu);
111 static int sync_regs(struct kvm_vcpu *vcpu);
113 struct kvm_x86_ops kvm_x86_ops __read_mostly;
114 EXPORT_SYMBOL_GPL(kvm_x86_ops);
116 static bool __read_mostly ignore_msrs = 0;
117 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
119 static bool __read_mostly report_ignored_msrs = true;
120 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
122 unsigned int min_timer_period_us = 200;
123 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
125 static bool __read_mostly kvmclock_periodic_sync = true;
126 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
128 bool __read_mostly kvm_has_tsc_control;
129 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
130 u32 __read_mostly kvm_max_guest_tsc_khz;
131 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
132 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
133 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
134 u64 __read_mostly kvm_max_tsc_scaling_ratio;
135 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
136 u64 __read_mostly kvm_default_tsc_scaling_ratio;
137 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
139 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
140 static u32 __read_mostly tsc_tolerance_ppm = 250;
141 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
144 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
145 * adaptive tuning starting from default advancment of 1000ns. '0' disables
146 * advancement entirely. Any other value is used as-is and disables adaptive
147 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
149 static int __read_mostly lapic_timer_advance_ns = -1;
150 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
152 static bool __read_mostly vector_hashing = true;
153 module_param(vector_hashing, bool, S_IRUGO);
155 bool __read_mostly enable_vmware_backdoor = false;
156 module_param(enable_vmware_backdoor, bool, S_IRUGO);
157 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
159 static bool __read_mostly force_emulation_prefix = false;
160 module_param(force_emulation_prefix, bool, S_IRUGO);
162 int __read_mostly pi_inject_timer = -1;
163 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
166 * Restoring the host value for MSRs that are only consumed when running in
167 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
168 * returns to userspace, i.e. the kernel can run with the guest's value.
170 #define KVM_MAX_NR_USER_RETURN_MSRS 16
172 struct kvm_user_return_msrs_global {
174 u32 msrs[KVM_MAX_NR_USER_RETURN_MSRS];
177 struct kvm_user_return_msrs {
178 struct user_return_notifier urn;
180 struct kvm_user_return_msr_values {
183 } values[KVM_MAX_NR_USER_RETURN_MSRS];
186 static struct kvm_user_return_msrs_global __read_mostly user_return_msrs_global;
187 static struct kvm_user_return_msrs __percpu *user_return_msrs;
189 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
190 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
191 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
192 | XFEATURE_MASK_PKRU)
194 u64 __read_mostly host_efer;
195 EXPORT_SYMBOL_GPL(host_efer);
197 bool __read_mostly allow_smaller_maxphyaddr = 0;
198 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
200 static u64 __read_mostly host_xss;
201 u64 __read_mostly supported_xss;
202 EXPORT_SYMBOL_GPL(supported_xss);
204 struct kvm_stats_debugfs_item debugfs_entries[] = {
205 VCPU_STAT("pf_fixed", pf_fixed),
206 VCPU_STAT("pf_guest", pf_guest),
207 VCPU_STAT("tlb_flush", tlb_flush),
208 VCPU_STAT("invlpg", invlpg),
209 VCPU_STAT("exits", exits),
210 VCPU_STAT("io_exits", io_exits),
211 VCPU_STAT("mmio_exits", mmio_exits),
212 VCPU_STAT("signal_exits", signal_exits),
213 VCPU_STAT("irq_window", irq_window_exits),
214 VCPU_STAT("nmi_window", nmi_window_exits),
215 VCPU_STAT("halt_exits", halt_exits),
216 VCPU_STAT("halt_successful_poll", halt_successful_poll),
217 VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
218 VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
219 VCPU_STAT("halt_wakeup", halt_wakeup),
220 VCPU_STAT("hypercalls", hypercalls),
221 VCPU_STAT("request_irq", request_irq_exits),
222 VCPU_STAT("irq_exits", irq_exits),
223 VCPU_STAT("host_state_reload", host_state_reload),
224 VCPU_STAT("fpu_reload", fpu_reload),
225 VCPU_STAT("insn_emulation", insn_emulation),
226 VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
227 VCPU_STAT("irq_injections", irq_injections),
228 VCPU_STAT("nmi_injections", nmi_injections),
229 VCPU_STAT("req_event", req_event),
230 VCPU_STAT("l1d_flush", l1d_flush),
231 VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
232 VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
233 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
234 VM_STAT("mmu_pte_write", mmu_pte_write),
235 VM_STAT("mmu_pte_updated", mmu_pte_updated),
236 VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
237 VM_STAT("mmu_flooded", mmu_flooded),
238 VM_STAT("mmu_recycled", mmu_recycled),
239 VM_STAT("mmu_cache_miss", mmu_cache_miss),
240 VM_STAT("mmu_unsync", mmu_unsync),
241 VM_STAT("remote_tlb_flush", remote_tlb_flush),
242 VM_STAT("largepages", lpages, .mode = 0444),
243 VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
244 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
248 u64 __read_mostly host_xcr0;
249 u64 __read_mostly supported_xcr0;
250 EXPORT_SYMBOL_GPL(supported_xcr0);
252 static struct kmem_cache *x86_fpu_cache;
254 static struct kmem_cache *x86_emulator_cache;
257 * When called, it means the previous get/set msr reached an invalid msr.
258 * Return true if we want to ignore/silent this failed msr access.
260 static bool kvm_msr_ignored_check(struct kvm_vcpu *vcpu, u32 msr,
261 u64 data, bool write)
263 const char *op = write ? "wrmsr" : "rdmsr";
266 if (report_ignored_msrs)
267 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
272 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
278 static struct kmem_cache *kvm_alloc_emulator_cache(void)
280 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
281 unsigned int size = sizeof(struct x86_emulate_ctxt);
283 return kmem_cache_create_usercopy("x86_emulator", size,
284 __alignof__(struct x86_emulate_ctxt),
285 SLAB_ACCOUNT, useroffset,
286 size - useroffset, NULL);
289 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
291 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
294 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
295 vcpu->arch.apf.gfns[i] = ~0;
298 static void kvm_on_user_return(struct user_return_notifier *urn)
301 struct kvm_user_return_msrs *msrs
302 = container_of(urn, struct kvm_user_return_msrs, urn);
303 struct kvm_user_return_msr_values *values;
307 * Disabling irqs at this point since the following code could be
308 * interrupted and executed through kvm_arch_hardware_disable()
310 local_irq_save(flags);
311 if (msrs->registered) {
312 msrs->registered = false;
313 user_return_notifier_unregister(urn);
315 local_irq_restore(flags);
316 for (slot = 0; slot < user_return_msrs_global.nr; ++slot) {
317 values = &msrs->values[slot];
318 if (values->host != values->curr) {
319 wrmsrl(user_return_msrs_global.msrs[slot], values->host);
320 values->curr = values->host;
325 void kvm_define_user_return_msr(unsigned slot, u32 msr)
327 BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS);
328 user_return_msrs_global.msrs[slot] = msr;
329 if (slot >= user_return_msrs_global.nr)
330 user_return_msrs_global.nr = slot + 1;
332 EXPORT_SYMBOL_GPL(kvm_define_user_return_msr);
334 static void kvm_user_return_msr_cpu_online(void)
336 unsigned int cpu = smp_processor_id();
337 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
341 for (i = 0; i < user_return_msrs_global.nr; ++i) {
342 rdmsrl_safe(user_return_msrs_global.msrs[i], &value);
343 msrs->values[i].host = value;
344 msrs->values[i].curr = value;
348 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
350 unsigned int cpu = smp_processor_id();
351 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
354 value = (value & mask) | (msrs->values[slot].host & ~mask);
355 if (value == msrs->values[slot].curr)
357 err = wrmsrl_safe(user_return_msrs_global.msrs[slot], value);
361 msrs->values[slot].curr = value;
362 if (!msrs->registered) {
363 msrs->urn.on_user_return = kvm_on_user_return;
364 user_return_notifier_register(&msrs->urn);
365 msrs->registered = true;
369 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
371 static void drop_user_return_notifiers(void)
373 unsigned int cpu = smp_processor_id();
374 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
376 if (msrs->registered)
377 kvm_on_user_return(&msrs->urn);
380 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
382 return vcpu->arch.apic_base;
384 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
386 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
388 return kvm_apic_mode(kvm_get_apic_base(vcpu));
390 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
392 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
394 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
395 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
396 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
397 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
399 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
401 if (!msr_info->host_initiated) {
402 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
404 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
408 kvm_lapic_set_base(vcpu, msr_info->data);
409 kvm_recalculate_apic_map(vcpu->kvm);
412 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
414 asmlinkage __visible noinstr void kvm_spurious_fault(void)
416 /* Fault while not rebooting. We want the trace. */
417 BUG_ON(!kvm_rebooting);
419 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
421 #define EXCPT_BENIGN 0
422 #define EXCPT_CONTRIBUTORY 1
425 static int exception_class(int vector)
435 return EXCPT_CONTRIBUTORY;
442 #define EXCPT_FAULT 0
444 #define EXCPT_ABORT 2
445 #define EXCPT_INTERRUPT 3
447 static int exception_type(int vector)
451 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
452 return EXCPT_INTERRUPT;
456 /* #DB is trap, as instruction watchpoints are handled elsewhere */
457 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
460 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
463 /* Reserved exceptions will result in fault */
467 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
469 unsigned nr = vcpu->arch.exception.nr;
470 bool has_payload = vcpu->arch.exception.has_payload;
471 unsigned long payload = vcpu->arch.exception.payload;
479 * "Certain debug exceptions may clear bit 0-3. The
480 * remaining contents of the DR6 register are never
481 * cleared by the processor".
483 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
485 * DR6.RTM is set by all #DB exceptions that don't clear it.
487 vcpu->arch.dr6 |= DR6_RTM;
488 vcpu->arch.dr6 |= payload;
490 * Bit 16 should be set in the payload whenever the #DB
491 * exception should clear DR6.RTM. This makes the payload
492 * compatible with the pending debug exceptions under VMX.
493 * Though not currently documented in the SDM, this also
494 * makes the payload compatible with the exit qualification
495 * for #DB exceptions under VMX.
497 vcpu->arch.dr6 ^= payload & DR6_RTM;
500 * The #DB payload is defined as compatible with the 'pending
501 * debug exceptions' field under VMX, not DR6. While bit 12 is
502 * defined in the 'pending debug exceptions' field (enabled
503 * breakpoint), it is reserved and must be zero in DR6.
505 vcpu->arch.dr6 &= ~BIT(12);
508 vcpu->arch.cr2 = payload;
512 vcpu->arch.exception.has_payload = false;
513 vcpu->arch.exception.payload = 0;
515 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
517 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
518 unsigned nr, bool has_error, u32 error_code,
519 bool has_payload, unsigned long payload, bool reinject)
524 kvm_make_request(KVM_REQ_EVENT, vcpu);
526 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
528 if (has_error && !is_protmode(vcpu))
532 * On vmentry, vcpu->arch.exception.pending is only
533 * true if an event injection was blocked by
534 * nested_run_pending. In that case, however,
535 * vcpu_enter_guest requests an immediate exit,
536 * and the guest shouldn't proceed far enough to
539 WARN_ON_ONCE(vcpu->arch.exception.pending);
540 vcpu->arch.exception.injected = true;
541 if (WARN_ON_ONCE(has_payload)) {
543 * A reinjected event has already
544 * delivered its payload.
550 vcpu->arch.exception.pending = true;
551 vcpu->arch.exception.injected = false;
553 vcpu->arch.exception.has_error_code = has_error;
554 vcpu->arch.exception.nr = nr;
555 vcpu->arch.exception.error_code = error_code;
556 vcpu->arch.exception.has_payload = has_payload;
557 vcpu->arch.exception.payload = payload;
558 if (!is_guest_mode(vcpu))
559 kvm_deliver_exception_payload(vcpu);
563 /* to check exception */
564 prev_nr = vcpu->arch.exception.nr;
565 if (prev_nr == DF_VECTOR) {
566 /* triple fault -> shutdown */
567 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
570 class1 = exception_class(prev_nr);
571 class2 = exception_class(nr);
572 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
573 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
575 * Generate double fault per SDM Table 5-5. Set
576 * exception.pending = true so that the double fault
577 * can trigger a nested vmexit.
579 vcpu->arch.exception.pending = true;
580 vcpu->arch.exception.injected = false;
581 vcpu->arch.exception.has_error_code = true;
582 vcpu->arch.exception.nr = DF_VECTOR;
583 vcpu->arch.exception.error_code = 0;
584 vcpu->arch.exception.has_payload = false;
585 vcpu->arch.exception.payload = 0;
587 /* replace previous exception with a new one in a hope
588 that instruction re-execution will regenerate lost
593 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
595 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
597 EXPORT_SYMBOL_GPL(kvm_queue_exception);
599 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
601 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
603 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
605 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
606 unsigned long payload)
608 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
610 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
612 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
613 u32 error_code, unsigned long payload)
615 kvm_multiple_exception(vcpu, nr, true, error_code,
616 true, payload, false);
619 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
622 kvm_inject_gp(vcpu, 0);
624 return kvm_skip_emulated_instruction(vcpu);
628 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
630 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
632 ++vcpu->stat.pf_guest;
633 vcpu->arch.exception.nested_apf =
634 is_guest_mode(vcpu) && fault->async_page_fault;
635 if (vcpu->arch.exception.nested_apf) {
636 vcpu->arch.apf.nested_apf_token = fault->address;
637 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
639 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
643 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
645 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
646 struct x86_exception *fault)
648 struct kvm_mmu *fault_mmu;
649 WARN_ON_ONCE(fault->vector != PF_VECTOR);
651 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
655 * Invalidate the TLB entry for the faulting address, if it exists,
656 * else the access will fault indefinitely (and to emulate hardware).
658 if ((fault->error_code & PFERR_PRESENT_MASK) &&
659 !(fault->error_code & PFERR_RSVD_MASK))
660 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
661 fault_mmu->root_hpa);
663 fault_mmu->inject_page_fault(vcpu, fault);
664 return fault->nested_page_fault;
666 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
668 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
670 atomic_inc(&vcpu->arch.nmi_queued);
671 kvm_make_request(KVM_REQ_NMI, vcpu);
673 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
675 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
677 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
679 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
681 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
683 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
685 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
688 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
689 * a #GP and return false.
691 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
693 if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
695 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
698 EXPORT_SYMBOL_GPL(kvm_require_cpl);
700 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
702 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
705 kvm_queue_exception(vcpu, UD_VECTOR);
708 EXPORT_SYMBOL_GPL(kvm_require_dr);
711 * This function will be used to read from the physical memory of the currently
712 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
713 * can read from guest physical or from the guest's guest physical memory.
715 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
716 gfn_t ngfn, void *data, int offset, int len,
719 struct x86_exception exception;
723 ngpa = gfn_to_gpa(ngfn);
724 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
725 if (real_gfn == UNMAPPED_GVA)
728 real_gfn = gpa_to_gfn(real_gfn);
730 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
732 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
734 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
735 void *data, int offset, int len, u32 access)
737 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
738 data, offset, len, access);
741 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
743 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
748 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
750 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
752 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
753 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
756 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
758 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
759 offset * sizeof(u64), sizeof(pdpte),
760 PFERR_USER_MASK|PFERR_WRITE_MASK);
765 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
766 if ((pdpte[i] & PT_PRESENT_MASK) &&
767 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
774 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
775 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
781 EXPORT_SYMBOL_GPL(load_pdptrs);
783 bool pdptrs_changed(struct kvm_vcpu *vcpu)
785 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
790 if (!is_pae_paging(vcpu))
793 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
796 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
797 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
798 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
799 PFERR_USER_MASK | PFERR_WRITE_MASK);
803 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
805 EXPORT_SYMBOL_GPL(pdptrs_changed);
807 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
809 unsigned long old_cr0 = kvm_read_cr0(vcpu);
810 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
811 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
816 if (cr0 & 0xffffffff00000000UL)
820 cr0 &= ~CR0_RESERVED_BITS;
822 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
825 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
829 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
830 (cr0 & X86_CR0_PG)) {
835 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
840 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
841 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
842 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
845 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
848 kvm_x86_ops.set_cr0(vcpu, cr0);
850 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
851 kvm_clear_async_pf_completion_queue(vcpu);
852 kvm_async_pf_hash_reset(vcpu);
855 if ((cr0 ^ old_cr0) & update_bits)
856 kvm_mmu_reset_context(vcpu);
858 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
859 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
860 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
861 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
865 EXPORT_SYMBOL_GPL(kvm_set_cr0);
867 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
869 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
871 EXPORT_SYMBOL_GPL(kvm_lmsw);
873 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
875 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
877 if (vcpu->arch.xcr0 != host_xcr0)
878 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
880 if (vcpu->arch.xsaves_enabled &&
881 vcpu->arch.ia32_xss != host_xss)
882 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
885 if (static_cpu_has(X86_FEATURE_PKU) &&
886 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
887 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
888 vcpu->arch.pkru != vcpu->arch.host_pkru)
889 __write_pkru(vcpu->arch.pkru);
891 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
893 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
895 if (static_cpu_has(X86_FEATURE_PKU) &&
896 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
897 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
898 vcpu->arch.pkru = rdpkru();
899 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
900 __write_pkru(vcpu->arch.host_pkru);
903 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
905 if (vcpu->arch.xcr0 != host_xcr0)
906 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
908 if (vcpu->arch.xsaves_enabled &&
909 vcpu->arch.ia32_xss != host_xss)
910 wrmsrl(MSR_IA32_XSS, host_xss);
914 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
916 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
919 u64 old_xcr0 = vcpu->arch.xcr0;
922 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
923 if (index != XCR_XFEATURE_ENABLED_MASK)
925 if (!(xcr0 & XFEATURE_MASK_FP))
927 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
931 * Do not allow the guest to set bits that we do not support
932 * saving. However, xcr0 bit 0 is always set, even if the
933 * emulated CPU does not support XSAVE (see fx_init).
935 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
936 if (xcr0 & ~valid_bits)
939 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
940 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
943 if (xcr0 & XFEATURE_MASK_AVX512) {
944 if (!(xcr0 & XFEATURE_MASK_YMM))
946 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
949 vcpu->arch.xcr0 = xcr0;
951 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
952 kvm_update_cpuid_runtime(vcpu);
956 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
958 if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
959 __kvm_set_xcr(vcpu, index, xcr)) {
960 kvm_inject_gp(vcpu, 0);
965 EXPORT_SYMBOL_GPL(kvm_set_xcr);
967 int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
969 if (cr4 & cr4_reserved_bits)
972 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
975 if (!kvm_x86_ops.is_valid_cr4(vcpu, cr4))
980 EXPORT_SYMBOL_GPL(kvm_valid_cr4);
982 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
984 unsigned long old_cr4 = kvm_read_cr4(vcpu);
985 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
987 unsigned long mmu_role_bits = pdptr_bits | X86_CR4_SMAP | X86_CR4_PKE;
989 if (kvm_valid_cr4(vcpu, cr4))
992 if (is_long_mode(vcpu)) {
993 if (!(cr4 & X86_CR4_PAE))
995 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
997 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
998 && ((cr4 ^ old_cr4) & pdptr_bits)
999 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1000 kvm_read_cr3(vcpu)))
1003 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1004 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1007 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1008 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1012 kvm_x86_ops.set_cr4(vcpu, cr4);
1014 if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1015 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1016 kvm_mmu_reset_context(vcpu);
1018 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
1019 kvm_update_cpuid_runtime(vcpu);
1023 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1025 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1027 bool skip_tlb_flush = false;
1028 #ifdef CONFIG_X86_64
1029 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1032 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1033 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1037 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1038 if (!skip_tlb_flush) {
1039 kvm_mmu_sync_roots(vcpu);
1040 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1045 if (is_long_mode(vcpu) &&
1046 (cr3 & vcpu->arch.cr3_lm_rsvd_bits))
1048 else if (is_pae_paging(vcpu) &&
1049 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1052 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1053 vcpu->arch.cr3 = cr3;
1054 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1058 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1060 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1062 if (cr8 & CR8_RESERVED_BITS)
1064 if (lapic_in_kernel(vcpu))
1065 kvm_lapic_set_tpr(vcpu, cr8);
1067 vcpu->arch.cr8 = cr8;
1070 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1072 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1074 if (lapic_in_kernel(vcpu))
1075 return kvm_lapic_get_cr8(vcpu);
1077 return vcpu->arch.cr8;
1079 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1081 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1085 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1086 for (i = 0; i < KVM_NR_DB_REGS; i++)
1087 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1088 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1092 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1096 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1097 dr7 = vcpu->arch.guest_debug_dr7;
1099 dr7 = vcpu->arch.dr7;
1100 kvm_x86_ops.set_dr7(vcpu, dr7);
1101 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1102 if (dr7 & DR7_BP_EN_MASK)
1103 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1105 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1107 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1109 u64 fixed = DR6_FIXED_1;
1111 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1116 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1118 size_t size = ARRAY_SIZE(vcpu->arch.db);
1122 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1123 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1124 vcpu->arch.eff_db[dr] = val;
1128 if (!kvm_dr6_valid(val))
1129 return -1; /* #GP */
1130 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1134 if (!kvm_dr7_valid(val))
1135 return -1; /* #GP */
1136 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1137 kvm_update_dr7(vcpu);
1144 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1146 if (__kvm_set_dr(vcpu, dr, val)) {
1147 kvm_inject_gp(vcpu, 0);
1152 EXPORT_SYMBOL_GPL(kvm_set_dr);
1154 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1156 size_t size = ARRAY_SIZE(vcpu->arch.db);
1160 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1164 *val = vcpu->arch.dr6;
1168 *val = vcpu->arch.dr7;
1173 EXPORT_SYMBOL_GPL(kvm_get_dr);
1175 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1177 u32 ecx = kvm_rcx_read(vcpu);
1181 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1184 kvm_rax_write(vcpu, (u32)data);
1185 kvm_rdx_write(vcpu, data >> 32);
1188 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1191 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1192 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1194 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1195 * extract the supported MSRs from the related const lists.
1196 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1197 * capabilities of the host cpu. This capabilities test skips MSRs that are
1198 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1199 * may depend on host virtualization features rather than host cpu features.
1202 static const u32 msrs_to_save_all[] = {
1203 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1205 #ifdef CONFIG_X86_64
1206 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1208 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1209 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1211 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1212 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1213 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1214 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1215 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1216 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1217 MSR_IA32_UMWAIT_CONTROL,
1219 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1220 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1221 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1222 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1223 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1224 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1225 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1226 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1227 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1228 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1229 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1230 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1231 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1232 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1233 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1234 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1235 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1236 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1237 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1238 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1239 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1240 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1243 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1244 static unsigned num_msrs_to_save;
1246 static const u32 emulated_msrs_all[] = {
1247 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1248 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1249 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1250 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1251 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1252 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1253 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1255 HV_X64_MSR_VP_INDEX,
1256 HV_X64_MSR_VP_RUNTIME,
1257 HV_X64_MSR_SCONTROL,
1258 HV_X64_MSR_STIMER0_CONFIG,
1259 HV_X64_MSR_VP_ASSIST_PAGE,
1260 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1261 HV_X64_MSR_TSC_EMULATION_STATUS,
1262 HV_X64_MSR_SYNDBG_OPTIONS,
1263 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1264 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1265 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1267 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1268 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1270 MSR_IA32_TSC_ADJUST,
1271 MSR_IA32_TSCDEADLINE,
1272 MSR_IA32_ARCH_CAPABILITIES,
1273 MSR_IA32_PERF_CAPABILITIES,
1274 MSR_IA32_MISC_ENABLE,
1275 MSR_IA32_MCG_STATUS,
1277 MSR_IA32_MCG_EXT_CTL,
1281 MSR_MISC_FEATURES_ENABLES,
1282 MSR_AMD64_VIRT_SPEC_CTRL,
1287 * The following list leaves out MSRs whose values are determined
1288 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1289 * We always support the "true" VMX control MSRs, even if the host
1290 * processor does not, so I am putting these registers here rather
1291 * than in msrs_to_save_all.
1294 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1295 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1296 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1297 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1299 MSR_IA32_VMX_CR0_FIXED0,
1300 MSR_IA32_VMX_CR4_FIXED0,
1301 MSR_IA32_VMX_VMCS_ENUM,
1302 MSR_IA32_VMX_PROCBASED_CTLS2,
1303 MSR_IA32_VMX_EPT_VPID_CAP,
1304 MSR_IA32_VMX_VMFUNC,
1307 MSR_KVM_POLL_CONTROL,
1310 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1311 static unsigned num_emulated_msrs;
1314 * List of msr numbers which are used to expose MSR-based features that
1315 * can be used by a hypervisor to validate requested CPU features.
1317 static const u32 msr_based_features_all[] = {
1319 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1320 MSR_IA32_VMX_PINBASED_CTLS,
1321 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1322 MSR_IA32_VMX_PROCBASED_CTLS,
1323 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1324 MSR_IA32_VMX_EXIT_CTLS,
1325 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1326 MSR_IA32_VMX_ENTRY_CTLS,
1328 MSR_IA32_VMX_CR0_FIXED0,
1329 MSR_IA32_VMX_CR0_FIXED1,
1330 MSR_IA32_VMX_CR4_FIXED0,
1331 MSR_IA32_VMX_CR4_FIXED1,
1332 MSR_IA32_VMX_VMCS_ENUM,
1333 MSR_IA32_VMX_PROCBASED_CTLS2,
1334 MSR_IA32_VMX_EPT_VPID_CAP,
1335 MSR_IA32_VMX_VMFUNC,
1339 MSR_IA32_ARCH_CAPABILITIES,
1340 MSR_IA32_PERF_CAPABILITIES,
1343 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1344 static unsigned int num_msr_based_features;
1346 static u64 kvm_get_arch_capabilities(void)
1350 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1351 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1354 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1355 * the nested hypervisor runs with NX huge pages. If it is not,
1356 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1357 * L1 guests, so it need not worry about its own (L2) guests.
1359 data |= ARCH_CAP_PSCHANGE_MC_NO;
1362 * If we're doing cache flushes (either "always" or "cond")
1363 * we will do one whenever the guest does a vmlaunch/vmresume.
1364 * If an outer hypervisor is doing the cache flush for us
1365 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1366 * capability to the guest too, and if EPT is disabled we're not
1367 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1368 * require a nested hypervisor to do a flush of its own.
1370 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1371 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1373 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1374 data |= ARCH_CAP_RDCL_NO;
1375 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1376 data |= ARCH_CAP_SSB_NO;
1377 if (!boot_cpu_has_bug(X86_BUG_MDS))
1378 data |= ARCH_CAP_MDS_NO;
1381 * On TAA affected systems:
1382 * - nothing to do if TSX is disabled on the host.
1383 * - we emulate TSX_CTRL if present on the host.
1384 * This lets the guest use VERW to clear CPU buffers.
1386 if (!boot_cpu_has(X86_FEATURE_RTM))
1387 data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1388 else if (!boot_cpu_has_bug(X86_BUG_TAA))
1389 data |= ARCH_CAP_TAA_NO;
1394 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1396 switch (msr->index) {
1397 case MSR_IA32_ARCH_CAPABILITIES:
1398 msr->data = kvm_get_arch_capabilities();
1400 case MSR_IA32_UCODE_REV:
1401 rdmsrl_safe(msr->index, &msr->data);
1404 return kvm_x86_ops.get_msr_feature(msr);
1409 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1411 struct kvm_msr_entry msr;
1415 r = kvm_get_msr_feature(&msr);
1417 if (r == KVM_MSR_RET_INVALID) {
1418 /* Unconditionally clear the output for simplicity */
1420 if (kvm_msr_ignored_check(vcpu, index, 0, false))
1432 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1434 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1437 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1440 if (efer & (EFER_LME | EFER_LMA) &&
1441 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1444 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1450 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1452 if (efer & efer_reserved_bits)
1455 return __kvm_valid_efer(vcpu, efer);
1457 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1459 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1461 u64 old_efer = vcpu->arch.efer;
1462 u64 efer = msr_info->data;
1465 if (efer & efer_reserved_bits)
1468 if (!msr_info->host_initiated) {
1469 if (!__kvm_valid_efer(vcpu, efer))
1472 if (is_paging(vcpu) &&
1473 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1478 efer |= vcpu->arch.efer & EFER_LMA;
1480 r = kvm_x86_ops.set_efer(vcpu, efer);
1486 /* Update reserved bits */
1487 if ((efer ^ old_efer) & EFER_NX)
1488 kvm_mmu_reset_context(vcpu);
1493 void kvm_enable_efer_bits(u64 mask)
1495 efer_reserved_bits &= ~mask;
1497 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1499 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1501 struct kvm *kvm = vcpu->kvm;
1502 struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
1503 u32 count = kvm->arch.msr_filter.count;
1505 bool r = kvm->arch.msr_filter.default_allow;
1508 /* MSR filtering not set up or x2APIC enabled, allow everything */
1509 if (!count || (index >= 0x800 && index <= 0x8ff))
1512 /* Prevent collision with set_msr_filter */
1513 idx = srcu_read_lock(&kvm->srcu);
1515 for (i = 0; i < count; i++) {
1516 u32 start = ranges[i].base;
1517 u32 end = start + ranges[i].nmsrs;
1518 u32 flags = ranges[i].flags;
1519 unsigned long *bitmap = ranges[i].bitmap;
1521 if ((index >= start) && (index < end) && (flags & type)) {
1522 r = !!test_bit(index - start, bitmap);
1527 srcu_read_unlock(&kvm->srcu, idx);
1531 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1534 * Write @data into the MSR specified by @index. Select MSR specific fault
1535 * checks are bypassed if @host_initiated is %true.
1536 * Returns 0 on success, non-0 otherwise.
1537 * Assumes vcpu_load() was already called.
1539 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1540 bool host_initiated)
1542 struct msr_data msr;
1544 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1545 return KVM_MSR_RET_FILTERED;
1550 case MSR_KERNEL_GS_BASE:
1553 if (is_noncanonical_address(data, vcpu))
1556 case MSR_IA32_SYSENTER_EIP:
1557 case MSR_IA32_SYSENTER_ESP:
1559 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1560 * non-canonical address is written on Intel but not on
1561 * AMD (which ignores the top 32-bits, because it does
1562 * not implement 64-bit SYSENTER).
1564 * 64-bit code should hence be able to write a non-canonical
1565 * value on AMD. Making the address canonical ensures that
1566 * vmentry does not fail on Intel after writing a non-canonical
1567 * value, and that something deterministic happens if the guest
1568 * invokes 64-bit SYSENTER.
1570 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1575 msr.host_initiated = host_initiated;
1577 return kvm_x86_ops.set_msr(vcpu, &msr);
1580 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1581 u32 index, u64 data, bool host_initiated)
1583 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1585 if (ret == KVM_MSR_RET_INVALID)
1586 if (kvm_msr_ignored_check(vcpu, index, data, true))
1593 * Read the MSR specified by @index into @data. Select MSR specific fault
1594 * checks are bypassed if @host_initiated is %true.
1595 * Returns 0 on success, non-0 otherwise.
1596 * Assumes vcpu_load() was already called.
1598 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1599 bool host_initiated)
1601 struct msr_data msr;
1604 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1605 return KVM_MSR_RET_FILTERED;
1608 msr.host_initiated = host_initiated;
1610 ret = kvm_x86_ops.get_msr(vcpu, &msr);
1616 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1617 u32 index, u64 *data, bool host_initiated)
1619 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1621 if (ret == KVM_MSR_RET_INVALID) {
1622 /* Unconditionally clear *data for simplicity */
1624 if (kvm_msr_ignored_check(vcpu, index, 0, false))
1631 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1633 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1635 EXPORT_SYMBOL_GPL(kvm_get_msr);
1637 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1639 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1641 EXPORT_SYMBOL_GPL(kvm_set_msr);
1643 static int complete_emulated_msr(struct kvm_vcpu *vcpu, bool is_read)
1645 if (vcpu->run->msr.error) {
1646 kvm_inject_gp(vcpu, 0);
1648 } else if (is_read) {
1649 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1650 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1653 return kvm_skip_emulated_instruction(vcpu);
1656 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1658 return complete_emulated_msr(vcpu, true);
1661 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1663 return complete_emulated_msr(vcpu, false);
1666 static u64 kvm_msr_reason(int r)
1669 case KVM_MSR_RET_INVALID:
1670 return KVM_MSR_EXIT_REASON_UNKNOWN;
1671 case KVM_MSR_RET_FILTERED:
1672 return KVM_MSR_EXIT_REASON_FILTER;
1674 return KVM_MSR_EXIT_REASON_INVAL;
1678 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1679 u32 exit_reason, u64 data,
1680 int (*completion)(struct kvm_vcpu *vcpu),
1683 u64 msr_reason = kvm_msr_reason(r);
1685 /* Check if the user wanted to know about this MSR fault */
1686 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1689 vcpu->run->exit_reason = exit_reason;
1690 vcpu->run->msr.error = 0;
1691 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1692 vcpu->run->msr.reason = msr_reason;
1693 vcpu->run->msr.index = index;
1694 vcpu->run->msr.data = data;
1695 vcpu->arch.complete_userspace_io = completion;
1700 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1702 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1703 complete_emulated_rdmsr, r);
1706 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1708 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1709 complete_emulated_wrmsr, r);
1712 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1714 u32 ecx = kvm_rcx_read(vcpu);
1718 r = kvm_get_msr(vcpu, ecx, &data);
1720 /* MSR read failed? See if we should ask user space */
1721 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1722 /* Bounce to user space */
1726 /* MSR read failed? Inject a #GP */
1728 trace_kvm_msr_read_ex(ecx);
1729 kvm_inject_gp(vcpu, 0);
1733 trace_kvm_msr_read(ecx, data);
1735 kvm_rax_write(vcpu, data & -1u);
1736 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1737 return kvm_skip_emulated_instruction(vcpu);
1739 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1741 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1743 u32 ecx = kvm_rcx_read(vcpu);
1744 u64 data = kvm_read_edx_eax(vcpu);
1747 r = kvm_set_msr(vcpu, ecx, data);
1749 /* MSR write failed? See if we should ask user space */
1750 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1751 /* Bounce to user space */
1754 /* Signal all other negative errors to userspace */
1758 /* MSR write failed? Inject a #GP */
1760 trace_kvm_msr_write_ex(ecx, data);
1761 kvm_inject_gp(vcpu, 0);
1765 trace_kvm_msr_write(ecx, data);
1766 return kvm_skip_emulated_instruction(vcpu);
1768 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1770 bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1772 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1773 xfer_to_guest_mode_work_pending();
1775 EXPORT_SYMBOL_GPL(kvm_vcpu_exit_request);
1778 * The fast path for frequent and performance sensitive wrmsr emulation,
1779 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1780 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1781 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1782 * other cases which must be called after interrupts are enabled on the host.
1784 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1786 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1789 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1790 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1791 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1792 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1795 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1796 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1797 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1798 trace_kvm_apic_write(APIC_ICR, (u32)data);
1805 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1807 if (!kvm_can_use_hv_timer(vcpu))
1810 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1814 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1816 u32 msr = kvm_rcx_read(vcpu);
1818 fastpath_t ret = EXIT_FASTPATH_NONE;
1821 case APIC_BASE_MSR + (APIC_ICR >> 4):
1822 data = kvm_read_edx_eax(vcpu);
1823 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1824 kvm_skip_emulated_instruction(vcpu);
1825 ret = EXIT_FASTPATH_EXIT_HANDLED;
1828 case MSR_IA32_TSCDEADLINE:
1829 data = kvm_read_edx_eax(vcpu);
1830 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1831 kvm_skip_emulated_instruction(vcpu);
1832 ret = EXIT_FASTPATH_REENTER_GUEST;
1839 if (ret != EXIT_FASTPATH_NONE)
1840 trace_kvm_msr_write(msr, data);
1844 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1847 * Adapt set_msr() to msr_io()'s calling convention
1849 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1851 return kvm_get_msr_ignored_check(vcpu, index, data, true);
1854 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1856 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1859 #ifdef CONFIG_X86_64
1860 struct pvclock_clock {
1870 struct pvclock_gtod_data {
1873 struct pvclock_clock clock; /* extract of a clocksource struct */
1874 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1880 static struct pvclock_gtod_data pvclock_gtod_data;
1882 static void update_pvclock_gtod(struct timekeeper *tk)
1884 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1886 write_seqcount_begin(&vdata->seq);
1888 /* copy pvclock gtod data */
1889 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1890 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1891 vdata->clock.mask = tk->tkr_mono.mask;
1892 vdata->clock.mult = tk->tkr_mono.mult;
1893 vdata->clock.shift = tk->tkr_mono.shift;
1894 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1895 vdata->clock.offset = tk->tkr_mono.base;
1897 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1898 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1899 vdata->raw_clock.mask = tk->tkr_raw.mask;
1900 vdata->raw_clock.mult = tk->tkr_raw.mult;
1901 vdata->raw_clock.shift = tk->tkr_raw.shift;
1902 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1903 vdata->raw_clock.offset = tk->tkr_raw.base;
1905 vdata->wall_time_sec = tk->xtime_sec;
1907 vdata->offs_boot = tk->offs_boot;
1909 write_seqcount_end(&vdata->seq);
1912 static s64 get_kvmclock_base_ns(void)
1914 /* Count up from boot time, but with the frequency of the raw clock. */
1915 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1918 static s64 get_kvmclock_base_ns(void)
1920 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1921 return ktime_get_boottime_ns();
1925 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1929 struct pvclock_wall_clock wc;
1932 kvm->arch.wall_clock = wall_clock;
1937 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1942 ++version; /* first time write, random junk */
1946 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1950 * The guest calculates current wall clock time by adding
1951 * system time (updated by kvm_guest_time_update below) to the
1952 * wall clock specified here. We do the reverse here.
1954 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1956 wc.nsec = do_div(wall_nsec, 1000000000);
1957 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1958 wc.version = version;
1960 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1963 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1966 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
1967 bool old_msr, bool host_initiated)
1969 struct kvm_arch *ka = &vcpu->kvm->arch;
1971 if (vcpu->vcpu_id == 0 && !host_initiated) {
1972 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
1973 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1975 ka->boot_vcpu_runs_old_kvmclock = old_msr;
1978 vcpu->arch.time = system_time;
1979 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
1981 /* we verify if the enable bit is set... */
1982 vcpu->arch.pv_time_enabled = false;
1983 if (!(system_time & 1))
1986 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
1987 &vcpu->arch.pv_time, system_time & ~1ULL,
1988 sizeof(struct pvclock_vcpu_time_info)))
1989 vcpu->arch.pv_time_enabled = true;
1994 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1996 do_shl32_div32(dividend, divisor);
2000 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2001 s8 *pshift, u32 *pmultiplier)
2009 scaled64 = scaled_hz;
2010 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2015 tps32 = (uint32_t)tps64;
2016 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2017 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2025 *pmultiplier = div_frac(scaled64, tps32);
2028 #ifdef CONFIG_X86_64
2029 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2032 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2033 static unsigned long max_tsc_khz;
2035 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2037 u64 v = (u64)khz * (1000000 + ppm);
2042 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2046 /* Guest TSC same frequency as host TSC? */
2048 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2052 /* TSC scaling supported? */
2053 if (!kvm_has_tsc_control) {
2054 if (user_tsc_khz > tsc_khz) {
2055 vcpu->arch.tsc_catchup = 1;
2056 vcpu->arch.tsc_always_catchup = 1;
2059 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2064 /* TSC scaling required - calculate ratio */
2065 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2066 user_tsc_khz, tsc_khz);
2068 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2069 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2074 vcpu->arch.tsc_scaling_ratio = ratio;
2078 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2080 u32 thresh_lo, thresh_hi;
2081 int use_scaling = 0;
2083 /* tsc_khz can be zero if TSC calibration fails */
2084 if (user_tsc_khz == 0) {
2085 /* set tsc_scaling_ratio to a safe value */
2086 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2090 /* Compute a scale to convert nanoseconds in TSC cycles */
2091 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2092 &vcpu->arch.virtual_tsc_shift,
2093 &vcpu->arch.virtual_tsc_mult);
2094 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2097 * Compute the variation in TSC rate which is acceptable
2098 * within the range of tolerance and decide if the
2099 * rate being applied is within that bounds of the hardware
2100 * rate. If so, no scaling or compensation need be done.
2102 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2103 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2104 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2105 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2108 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2111 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2113 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2114 vcpu->arch.virtual_tsc_mult,
2115 vcpu->arch.virtual_tsc_shift);
2116 tsc += vcpu->arch.this_tsc_write;
2120 static inline int gtod_is_based_on_tsc(int mode)
2122 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2125 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2127 #ifdef CONFIG_X86_64
2129 struct kvm_arch *ka = &vcpu->kvm->arch;
2130 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2132 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2133 atomic_read(&vcpu->kvm->online_vcpus));
2136 * Once the masterclock is enabled, always perform request in
2137 * order to update it.
2139 * In order to enable masterclock, the host clocksource must be TSC
2140 * and the vcpus need to have matched TSCs. When that happens,
2141 * perform request to enable masterclock.
2143 if (ka->use_master_clock ||
2144 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2145 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2147 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2148 atomic_read(&vcpu->kvm->online_vcpus),
2149 ka->use_master_clock, gtod->clock.vclock_mode);
2154 * Multiply tsc by a fixed point number represented by ratio.
2156 * The most significant 64-N bits (mult) of ratio represent the
2157 * integral part of the fixed point number; the remaining N bits
2158 * (frac) represent the fractional part, ie. ratio represents a fixed
2159 * point number (mult + frac * 2^(-N)).
2161 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2163 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2165 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2168 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2171 u64 ratio = vcpu->arch.tsc_scaling_ratio;
2173 if (ratio != kvm_default_tsc_scaling_ratio)
2174 _tsc = __scale_tsc(ratio, tsc);
2178 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2180 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2184 tsc = kvm_scale_tsc(vcpu, rdtsc());
2186 return target_tsc - tsc;
2189 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2191 return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2193 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2195 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2197 vcpu->arch.l1_tsc_offset = offset;
2198 vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
2201 static inline bool kvm_check_tsc_unstable(void)
2203 #ifdef CONFIG_X86_64
2205 * TSC is marked unstable when we're running on Hyper-V,
2206 * 'TSC page' clocksource is good.
2208 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2211 return check_tsc_unstable();
2214 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2216 struct kvm *kvm = vcpu->kvm;
2217 u64 offset, ns, elapsed;
2218 unsigned long flags;
2220 bool already_matched;
2221 bool synchronizing = false;
2223 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2224 offset = kvm_compute_tsc_offset(vcpu, data);
2225 ns = get_kvmclock_base_ns();
2226 elapsed = ns - kvm->arch.last_tsc_nsec;
2228 if (vcpu->arch.virtual_tsc_khz) {
2231 * detection of vcpu initialization -- need to sync
2232 * with other vCPUs. This particularly helps to keep
2233 * kvm_clock stable after CPU hotplug
2235 synchronizing = true;
2237 u64 tsc_exp = kvm->arch.last_tsc_write +
2238 nsec_to_cycles(vcpu, elapsed);
2239 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2241 * Special case: TSC write with a small delta (1 second)
2242 * of virtual cycle time against real time is
2243 * interpreted as an attempt to synchronize the CPU.
2245 synchronizing = data < tsc_exp + tsc_hz &&
2246 data + tsc_hz > tsc_exp;
2251 * For a reliable TSC, we can match TSC offsets, and for an unstable
2252 * TSC, we add elapsed time in this computation. We could let the
2253 * compensation code attempt to catch up if we fall behind, but
2254 * it's better to try to match offsets from the beginning.
2256 if (synchronizing &&
2257 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2258 if (!kvm_check_tsc_unstable()) {
2259 offset = kvm->arch.cur_tsc_offset;
2261 u64 delta = nsec_to_cycles(vcpu, elapsed);
2263 offset = kvm_compute_tsc_offset(vcpu, data);
2266 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2269 * We split periods of matched TSC writes into generations.
2270 * For each generation, we track the original measured
2271 * nanosecond time, offset, and write, so if TSCs are in
2272 * sync, we can match exact offset, and if not, we can match
2273 * exact software computation in compute_guest_tsc()
2275 * These values are tracked in kvm->arch.cur_xxx variables.
2277 kvm->arch.cur_tsc_generation++;
2278 kvm->arch.cur_tsc_nsec = ns;
2279 kvm->arch.cur_tsc_write = data;
2280 kvm->arch.cur_tsc_offset = offset;
2285 * We also track th most recent recorded KHZ, write and time to
2286 * allow the matching interval to be extended at each write.
2288 kvm->arch.last_tsc_nsec = ns;
2289 kvm->arch.last_tsc_write = data;
2290 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2292 vcpu->arch.last_guest_tsc = data;
2294 /* Keep track of which generation this VCPU has synchronized to */
2295 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2296 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2297 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2299 kvm_vcpu_write_tsc_offset(vcpu, offset);
2300 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2302 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2304 kvm->arch.nr_vcpus_matched_tsc = 0;
2305 } else if (!already_matched) {
2306 kvm->arch.nr_vcpus_matched_tsc++;
2309 kvm_track_tsc_matching(vcpu);
2310 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2313 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2316 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2317 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2320 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2322 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2323 WARN_ON(adjustment < 0);
2324 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2325 adjust_tsc_offset_guest(vcpu, adjustment);
2328 #ifdef CONFIG_X86_64
2330 static u64 read_tsc(void)
2332 u64 ret = (u64)rdtsc_ordered();
2333 u64 last = pvclock_gtod_data.clock.cycle_last;
2335 if (likely(ret >= last))
2339 * GCC likes to generate cmov here, but this branch is extremely
2340 * predictable (it's just a function of time and the likely is
2341 * very likely) and there's a data dependence, so force GCC
2342 * to generate a branch instead. I don't barrier() because
2343 * we don't actually need a barrier, and if this function
2344 * ever gets inlined it will generate worse code.
2350 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2356 switch (clock->vclock_mode) {
2357 case VDSO_CLOCKMODE_HVCLOCK:
2358 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2360 if (tsc_pg_val != U64_MAX) {
2361 /* TSC page valid */
2362 *mode = VDSO_CLOCKMODE_HVCLOCK;
2363 v = (tsc_pg_val - clock->cycle_last) &
2366 /* TSC page invalid */
2367 *mode = VDSO_CLOCKMODE_NONE;
2370 case VDSO_CLOCKMODE_TSC:
2371 *mode = VDSO_CLOCKMODE_TSC;
2372 *tsc_timestamp = read_tsc();
2373 v = (*tsc_timestamp - clock->cycle_last) &
2377 *mode = VDSO_CLOCKMODE_NONE;
2380 if (*mode == VDSO_CLOCKMODE_NONE)
2381 *tsc_timestamp = v = 0;
2383 return v * clock->mult;
2386 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2388 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2394 seq = read_seqcount_begin(>od->seq);
2395 ns = gtod->raw_clock.base_cycles;
2396 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2397 ns >>= gtod->raw_clock.shift;
2398 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2399 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2405 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2407 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2413 seq = read_seqcount_begin(>od->seq);
2414 ts->tv_sec = gtod->wall_time_sec;
2415 ns = gtod->clock.base_cycles;
2416 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2417 ns >>= gtod->clock.shift;
2418 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2420 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2426 /* returns true if host is using TSC based clocksource */
2427 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2429 /* checked again under seqlock below */
2430 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2433 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2437 /* returns true if host is using TSC based clocksource */
2438 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2441 /* checked again under seqlock below */
2442 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2445 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2451 * Assuming a stable TSC across physical CPUS, and a stable TSC
2452 * across virtual CPUs, the following condition is possible.
2453 * Each numbered line represents an event visible to both
2454 * CPUs at the next numbered event.
2456 * "timespecX" represents host monotonic time. "tscX" represents
2459 * VCPU0 on CPU0 | VCPU1 on CPU1
2461 * 1. read timespec0,tsc0
2462 * 2. | timespec1 = timespec0 + N
2464 * 3. transition to guest | transition to guest
2465 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2466 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2467 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2469 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2472 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2474 * - 0 < N - M => M < N
2476 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2477 * always the case (the difference between two distinct xtime instances
2478 * might be smaller then the difference between corresponding TSC reads,
2479 * when updating guest vcpus pvclock areas).
2481 * To avoid that problem, do not allow visibility of distinct
2482 * system_timestamp/tsc_timestamp values simultaneously: use a master
2483 * copy of host monotonic time values. Update that master copy
2486 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2490 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2492 #ifdef CONFIG_X86_64
2493 struct kvm_arch *ka = &kvm->arch;
2495 bool host_tsc_clocksource, vcpus_matched;
2497 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2498 atomic_read(&kvm->online_vcpus));
2501 * If the host uses TSC clock, then passthrough TSC as stable
2504 host_tsc_clocksource = kvm_get_time_and_clockread(
2505 &ka->master_kernel_ns,
2506 &ka->master_cycle_now);
2508 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2509 && !ka->backwards_tsc_observed
2510 && !ka->boot_vcpu_runs_old_kvmclock;
2512 if (ka->use_master_clock)
2513 atomic_set(&kvm_guest_has_master_clock, 1);
2515 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2516 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2521 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2523 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2526 static void kvm_gen_update_masterclock(struct kvm *kvm)
2528 #ifdef CONFIG_X86_64
2530 struct kvm_vcpu *vcpu;
2531 struct kvm_arch *ka = &kvm->arch;
2533 spin_lock(&ka->pvclock_gtod_sync_lock);
2534 kvm_make_mclock_inprogress_request(kvm);
2535 /* no guest entries from this point */
2536 pvclock_update_vm_gtod_copy(kvm);
2538 kvm_for_each_vcpu(i, vcpu, kvm)
2539 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2541 /* guest entries allowed */
2542 kvm_for_each_vcpu(i, vcpu, kvm)
2543 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2545 spin_unlock(&ka->pvclock_gtod_sync_lock);
2549 u64 get_kvmclock_ns(struct kvm *kvm)
2551 struct kvm_arch *ka = &kvm->arch;
2552 struct pvclock_vcpu_time_info hv_clock;
2555 spin_lock(&ka->pvclock_gtod_sync_lock);
2556 if (!ka->use_master_clock) {
2557 spin_unlock(&ka->pvclock_gtod_sync_lock);
2558 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2561 hv_clock.tsc_timestamp = ka->master_cycle_now;
2562 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2563 spin_unlock(&ka->pvclock_gtod_sync_lock);
2565 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2568 if (__this_cpu_read(cpu_tsc_khz)) {
2569 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2570 &hv_clock.tsc_shift,
2571 &hv_clock.tsc_to_system_mul);
2572 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2574 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2581 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2583 struct kvm_vcpu_arch *vcpu = &v->arch;
2584 struct pvclock_vcpu_time_info guest_hv_clock;
2586 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2587 &guest_hv_clock, sizeof(guest_hv_clock))))
2590 /* This VCPU is paused, but it's legal for a guest to read another
2591 * VCPU's kvmclock, so we really have to follow the specification where
2592 * it says that version is odd if data is being modified, and even after
2595 * Version field updates must be kept separate. This is because
2596 * kvm_write_guest_cached might use a "rep movs" instruction, and
2597 * writes within a string instruction are weakly ordered. So there
2598 * are three writes overall.
2600 * As a small optimization, only write the version field in the first
2601 * and third write. The vcpu->pv_time cache is still valid, because the
2602 * version field is the first in the struct.
2604 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2606 if (guest_hv_clock.version & 1)
2607 ++guest_hv_clock.version; /* first time write, random junk */
2609 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2610 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2612 sizeof(vcpu->hv_clock.version));
2616 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2617 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2619 if (vcpu->pvclock_set_guest_stopped_request) {
2620 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2621 vcpu->pvclock_set_guest_stopped_request = false;
2624 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2626 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2628 sizeof(vcpu->hv_clock));
2632 vcpu->hv_clock.version++;
2633 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2635 sizeof(vcpu->hv_clock.version));
2638 static int kvm_guest_time_update(struct kvm_vcpu *v)
2640 unsigned long flags, tgt_tsc_khz;
2641 struct kvm_vcpu_arch *vcpu = &v->arch;
2642 struct kvm_arch *ka = &v->kvm->arch;
2644 u64 tsc_timestamp, host_tsc;
2646 bool use_master_clock;
2652 * If the host uses TSC clock, then passthrough TSC as stable
2655 spin_lock(&ka->pvclock_gtod_sync_lock);
2656 use_master_clock = ka->use_master_clock;
2657 if (use_master_clock) {
2658 host_tsc = ka->master_cycle_now;
2659 kernel_ns = ka->master_kernel_ns;
2661 spin_unlock(&ka->pvclock_gtod_sync_lock);
2663 /* Keep irq disabled to prevent changes to the clock */
2664 local_irq_save(flags);
2665 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2666 if (unlikely(tgt_tsc_khz == 0)) {
2667 local_irq_restore(flags);
2668 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2671 if (!use_master_clock) {
2673 kernel_ns = get_kvmclock_base_ns();
2676 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2679 * We may have to catch up the TSC to match elapsed wall clock
2680 * time for two reasons, even if kvmclock is used.
2681 * 1) CPU could have been running below the maximum TSC rate
2682 * 2) Broken TSC compensation resets the base at each VCPU
2683 * entry to avoid unknown leaps of TSC even when running
2684 * again on the same CPU. This may cause apparent elapsed
2685 * time to disappear, and the guest to stand still or run
2688 if (vcpu->tsc_catchup) {
2689 u64 tsc = compute_guest_tsc(v, kernel_ns);
2690 if (tsc > tsc_timestamp) {
2691 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2692 tsc_timestamp = tsc;
2696 local_irq_restore(flags);
2698 /* With all the info we got, fill in the values */
2700 if (kvm_has_tsc_control)
2701 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2703 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2704 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2705 &vcpu->hv_clock.tsc_shift,
2706 &vcpu->hv_clock.tsc_to_system_mul);
2707 vcpu->hw_tsc_khz = tgt_tsc_khz;
2710 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2711 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2712 vcpu->last_guest_tsc = tsc_timestamp;
2714 /* If the host uses TSC clocksource, then it is stable */
2716 if (use_master_clock)
2717 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2719 vcpu->hv_clock.flags = pvclock_flags;
2721 if (vcpu->pv_time_enabled)
2722 kvm_setup_pvclock_page(v);
2723 if (v == kvm_get_vcpu(v->kvm, 0))
2724 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2729 * kvmclock updates which are isolated to a given vcpu, such as
2730 * vcpu->cpu migration, should not allow system_timestamp from
2731 * the rest of the vcpus to remain static. Otherwise ntp frequency
2732 * correction applies to one vcpu's system_timestamp but not
2735 * So in those cases, request a kvmclock update for all vcpus.
2736 * We need to rate-limit these requests though, as they can
2737 * considerably slow guests that have a large number of vcpus.
2738 * The time for a remote vcpu to update its kvmclock is bound
2739 * by the delay we use to rate-limit the updates.
2742 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2744 static void kvmclock_update_fn(struct work_struct *work)
2747 struct delayed_work *dwork = to_delayed_work(work);
2748 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2749 kvmclock_update_work);
2750 struct kvm *kvm = container_of(ka, struct kvm, arch);
2751 struct kvm_vcpu *vcpu;
2753 kvm_for_each_vcpu(i, vcpu, kvm) {
2754 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2755 kvm_vcpu_kick(vcpu);
2759 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2761 struct kvm *kvm = v->kvm;
2763 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2764 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2765 KVMCLOCK_UPDATE_DELAY);
2768 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2770 static void kvmclock_sync_fn(struct work_struct *work)
2772 struct delayed_work *dwork = to_delayed_work(work);
2773 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2774 kvmclock_sync_work);
2775 struct kvm *kvm = container_of(ka, struct kvm, arch);
2777 if (!kvmclock_periodic_sync)
2780 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2781 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2782 KVMCLOCK_SYNC_PERIOD);
2786 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2788 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2790 /* McStatusWrEn enabled? */
2791 if (guest_cpuid_is_amd_or_hygon(vcpu))
2792 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2797 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2799 u64 mcg_cap = vcpu->arch.mcg_cap;
2800 unsigned bank_num = mcg_cap & 0xff;
2801 u32 msr = msr_info->index;
2802 u64 data = msr_info->data;
2805 case MSR_IA32_MCG_STATUS:
2806 vcpu->arch.mcg_status = data;
2808 case MSR_IA32_MCG_CTL:
2809 if (!(mcg_cap & MCG_CTL_P) &&
2810 (data || !msr_info->host_initiated))
2812 if (data != 0 && data != ~(u64)0)
2814 vcpu->arch.mcg_ctl = data;
2817 if (msr >= MSR_IA32_MC0_CTL &&
2818 msr < MSR_IA32_MCx_CTL(bank_num)) {
2819 u32 offset = array_index_nospec(
2820 msr - MSR_IA32_MC0_CTL,
2821 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2823 /* only 0 or all 1s can be written to IA32_MCi_CTL
2824 * some Linux kernels though clear bit 10 in bank 4 to
2825 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2826 * this to avoid an uncatched #GP in the guest
2828 if ((offset & 0x3) == 0 &&
2829 data != 0 && (data | (1 << 10)) != ~(u64)0)
2833 if (!msr_info->host_initiated &&
2834 (offset & 0x3) == 1 && data != 0) {
2835 if (!can_set_mci_status(vcpu))
2839 vcpu->arch.mce_banks[offset] = data;
2847 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2849 struct kvm *kvm = vcpu->kvm;
2850 int lm = is_long_mode(vcpu);
2851 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2852 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2853 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2854 : kvm->arch.xen_hvm_config.blob_size_32;
2855 u32 page_num = data & ~PAGE_MASK;
2856 u64 page_addr = data & PAGE_MASK;
2859 if (page_num >= blob_size)
2862 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2864 return PTR_ERR(page);
2866 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) {
2873 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2875 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2877 return (vcpu->arch.apf.msr_en_val & mask) == mask;
2880 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2882 gpa_t gpa = data & ~0x3f;
2884 /* Bits 4:5 are reserved, Should be zero */
2888 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
2889 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
2892 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
2893 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
2896 if (!lapic_in_kernel(vcpu))
2897 return data ? 1 : 0;
2899 vcpu->arch.apf.msr_en_val = data;
2901 if (!kvm_pv_async_pf_enabled(vcpu)) {
2902 kvm_clear_async_pf_completion_queue(vcpu);
2903 kvm_async_pf_hash_reset(vcpu);
2907 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2911 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2912 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2914 kvm_async_pf_wakeup_all(vcpu);
2919 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2921 /* Bits 8-63 are reserved */
2925 if (!lapic_in_kernel(vcpu))
2928 vcpu->arch.apf.msr_int_val = data;
2930 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
2935 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2937 vcpu->arch.pv_time_enabled = false;
2938 vcpu->arch.time = 0;
2941 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
2943 ++vcpu->stat.tlb_flush;
2944 kvm_x86_ops.tlb_flush_all(vcpu);
2947 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
2949 ++vcpu->stat.tlb_flush;
2950 kvm_x86_ops.tlb_flush_guest(vcpu);
2953 static void record_steal_time(struct kvm_vcpu *vcpu)
2955 struct kvm_host_map map;
2956 struct kvm_steal_time *st;
2958 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2961 /* -EAGAIN is returned in atomic context so we can just return. */
2962 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2963 &map, &vcpu->arch.st.cache, false))
2967 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2970 * Doing a TLB flush here, on the guest's behalf, can avoid
2973 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
2974 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2975 st->preempted & KVM_VCPU_FLUSH_TLB);
2976 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2977 kvm_vcpu_flush_tlb_guest(vcpu);
2980 vcpu->arch.st.preempted = 0;
2982 if (st->version & 1)
2983 st->version += 1; /* first time write, random junk */
2989 st->steal += current->sched_info.run_delay -
2990 vcpu->arch.st.last_steal;
2991 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2997 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3000 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3003 u32 msr = msr_info->index;
3004 u64 data = msr_info->data;
3007 case MSR_AMD64_NB_CFG:
3008 case MSR_IA32_UCODE_WRITE:
3009 case MSR_VM_HSAVE_PA:
3010 case MSR_AMD64_PATCH_LOADER:
3011 case MSR_AMD64_BU_CFG2:
3012 case MSR_AMD64_DC_CFG:
3013 case MSR_F15H_EX_CFG:
3016 case MSR_IA32_UCODE_REV:
3017 if (msr_info->host_initiated)
3018 vcpu->arch.microcode_version = data;
3020 case MSR_IA32_ARCH_CAPABILITIES:
3021 if (!msr_info->host_initiated)
3023 vcpu->arch.arch_capabilities = data;
3025 case MSR_IA32_PERF_CAPABILITIES: {
3026 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3028 if (!msr_info->host_initiated)
3030 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3032 if (data & ~msr_ent.data)
3035 vcpu->arch.perf_capabilities = data;
3040 return set_efer(vcpu, msr_info);
3042 data &= ~(u64)0x40; /* ignore flush filter disable */
3043 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3044 data &= ~(u64)0x8; /* ignore TLB cache disable */
3046 /* Handle McStatusWrEn */
3047 if (data == BIT_ULL(18)) {
3048 vcpu->arch.msr_hwcr = data;
3049 } else if (data != 0) {
3050 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3055 case MSR_FAM10H_MMIO_CONF_BASE:
3057 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3062 case MSR_IA32_DEBUGCTLMSR:
3064 /* We support the non-activated case already */
3066 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
3067 /* Values other than LBR and BTF are vendor-specific,
3068 thus reserved and should throw a #GP */
3070 } else if (report_ignored_msrs)
3071 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
3074 case 0x200 ... 0x2ff:
3075 return kvm_mtrr_set_msr(vcpu, msr, data);
3076 case MSR_IA32_APICBASE:
3077 return kvm_set_apic_base(vcpu, msr_info);
3078 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3079 return kvm_x2apic_msr_write(vcpu, msr, data);
3080 case MSR_IA32_TSCDEADLINE:
3081 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3083 case MSR_IA32_TSC_ADJUST:
3084 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3085 if (!msr_info->host_initiated) {
3086 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3087 adjust_tsc_offset_guest(vcpu, adj);
3089 vcpu->arch.ia32_tsc_adjust_msr = data;
3092 case MSR_IA32_MISC_ENABLE:
3093 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3094 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3095 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3097 vcpu->arch.ia32_misc_enable_msr = data;
3098 kvm_update_cpuid_runtime(vcpu);
3100 vcpu->arch.ia32_misc_enable_msr = data;
3103 case MSR_IA32_SMBASE:
3104 if (!msr_info->host_initiated)
3106 vcpu->arch.smbase = data;
3108 case MSR_IA32_POWER_CTL:
3109 vcpu->arch.msr_ia32_power_ctl = data;
3112 if (msr_info->host_initiated) {
3113 kvm_synchronize_tsc(vcpu, data);
3115 u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3116 adjust_tsc_offset_guest(vcpu, adj);
3117 vcpu->arch.ia32_tsc_adjust_msr += adj;
3121 if (!msr_info->host_initiated &&
3122 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3125 * KVM supports exposing PT to the guest, but does not support
3126 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3127 * XSAVES/XRSTORS to save/restore PT MSRs.
3129 if (data & ~supported_xss)
3131 vcpu->arch.ia32_xss = data;
3134 if (!msr_info->host_initiated)
3136 vcpu->arch.smi_count = data;
3138 case MSR_KVM_WALL_CLOCK_NEW:
3139 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3142 kvm_write_wall_clock(vcpu->kvm, data);
3144 case MSR_KVM_WALL_CLOCK:
3145 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3148 kvm_write_wall_clock(vcpu->kvm, data);
3150 case MSR_KVM_SYSTEM_TIME_NEW:
3151 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3154 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3156 case MSR_KVM_SYSTEM_TIME:
3157 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3160 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3162 case MSR_KVM_ASYNC_PF_EN:
3163 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3166 if (kvm_pv_enable_async_pf(vcpu, data))
3169 case MSR_KVM_ASYNC_PF_INT:
3170 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3173 if (kvm_pv_enable_async_pf_int(vcpu, data))
3176 case MSR_KVM_ASYNC_PF_ACK:
3177 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3180 vcpu->arch.apf.pageready_pending = false;
3181 kvm_check_async_pf_completion(vcpu);
3184 case MSR_KVM_STEAL_TIME:
3185 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3188 if (unlikely(!sched_info_on()))
3191 if (data & KVM_STEAL_RESERVED_MASK)
3194 vcpu->arch.st.msr_val = data;
3196 if (!(data & KVM_MSR_ENABLED))
3199 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3202 case MSR_KVM_PV_EOI_EN:
3203 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3206 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3210 case MSR_KVM_POLL_CONTROL:
3211 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3214 /* only enable bit supported */
3215 if (data & (-1ULL << 1))
3218 vcpu->arch.msr_kvm_poll_control = data;
3221 case MSR_IA32_MCG_CTL:
3222 case MSR_IA32_MCG_STATUS:
3223 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3224 return set_msr_mce(vcpu, msr_info);
3226 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3227 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3230 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3231 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3232 if (kvm_pmu_is_valid_msr(vcpu, msr))
3233 return kvm_pmu_set_msr(vcpu, msr_info);
3235 if (pr || data != 0)
3236 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3237 "0x%x data 0x%llx\n", msr, data);
3239 case MSR_K7_CLK_CTL:
3241 * Ignore all writes to this no longer documented MSR.
3242 * Writes are only relevant for old K7 processors,
3243 * all pre-dating SVM, but a recommended workaround from
3244 * AMD for these chips. It is possible to specify the
3245 * affected processor models on the command line, hence
3246 * the need to ignore the workaround.
3249 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3250 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3251 case HV_X64_MSR_SYNDBG_OPTIONS:
3252 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3253 case HV_X64_MSR_CRASH_CTL:
3254 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3255 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3256 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3257 case HV_X64_MSR_TSC_EMULATION_STATUS:
3258 return kvm_hv_set_msr_common(vcpu, msr, data,
3259 msr_info->host_initiated);
3260 case MSR_IA32_BBL_CR_CTL3:
3261 /* Drop writes to this legacy MSR -- see rdmsr
3262 * counterpart for further detail.
3264 if (report_ignored_msrs)
3265 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3268 case MSR_AMD64_OSVW_ID_LENGTH:
3269 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3271 vcpu->arch.osvw.length = data;
3273 case MSR_AMD64_OSVW_STATUS:
3274 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3276 vcpu->arch.osvw.status = data;
3278 case MSR_PLATFORM_INFO:
3279 if (!msr_info->host_initiated ||
3280 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3281 cpuid_fault_enabled(vcpu)))
3283 vcpu->arch.msr_platform_info = data;
3285 case MSR_MISC_FEATURES_ENABLES:
3286 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3287 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3288 !supports_cpuid_fault(vcpu)))
3290 vcpu->arch.msr_misc_features_enables = data;
3293 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
3294 return xen_hvm_config(vcpu, data);
3295 if (kvm_pmu_is_valid_msr(vcpu, msr))
3296 return kvm_pmu_set_msr(vcpu, msr_info);
3297 return KVM_MSR_RET_INVALID;
3301 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3303 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3306 u64 mcg_cap = vcpu->arch.mcg_cap;
3307 unsigned bank_num = mcg_cap & 0xff;
3310 case MSR_IA32_P5_MC_ADDR:
3311 case MSR_IA32_P5_MC_TYPE:
3314 case MSR_IA32_MCG_CAP:
3315 data = vcpu->arch.mcg_cap;
3317 case MSR_IA32_MCG_CTL:
3318 if (!(mcg_cap & MCG_CTL_P) && !host)
3320 data = vcpu->arch.mcg_ctl;
3322 case MSR_IA32_MCG_STATUS:
3323 data = vcpu->arch.mcg_status;
3326 if (msr >= MSR_IA32_MC0_CTL &&
3327 msr < MSR_IA32_MCx_CTL(bank_num)) {
3328 u32 offset = array_index_nospec(
3329 msr - MSR_IA32_MC0_CTL,
3330 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3332 data = vcpu->arch.mce_banks[offset];
3341 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3343 switch (msr_info->index) {
3344 case MSR_IA32_PLATFORM_ID:
3345 case MSR_IA32_EBL_CR_POWERON:
3346 case MSR_IA32_DEBUGCTLMSR:
3347 case MSR_IA32_LASTBRANCHFROMIP:
3348 case MSR_IA32_LASTBRANCHTOIP:
3349 case MSR_IA32_LASTINTFROMIP:
3350 case MSR_IA32_LASTINTTOIP:
3352 case MSR_K8_TSEG_ADDR:
3353 case MSR_K8_TSEG_MASK:
3354 case MSR_VM_HSAVE_PA:
3355 case MSR_K8_INT_PENDING_MSG:
3356 case MSR_AMD64_NB_CFG:
3357 case MSR_FAM10H_MMIO_CONF_BASE:
3358 case MSR_AMD64_BU_CFG2:
3359 case MSR_IA32_PERF_CTL:
3360 case MSR_AMD64_DC_CFG:
3361 case MSR_F15H_EX_CFG:
3363 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3364 * limit) MSRs. Just return 0, as we do not want to expose the host
3365 * data here. Do not conditionalize this on CPUID, as KVM does not do
3366 * so for existing CPU-specific MSRs.
3368 case MSR_RAPL_POWER_UNIT:
3369 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3370 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3371 case MSR_PKG_ENERGY_STATUS: /* Total package */
3372 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3375 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3376 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3377 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3378 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3379 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3380 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3381 return kvm_pmu_get_msr(vcpu, msr_info);
3384 case MSR_IA32_UCODE_REV:
3385 msr_info->data = vcpu->arch.microcode_version;
3387 case MSR_IA32_ARCH_CAPABILITIES:
3388 if (!msr_info->host_initiated &&
3389 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3391 msr_info->data = vcpu->arch.arch_capabilities;
3393 case MSR_IA32_PERF_CAPABILITIES:
3394 if (!msr_info->host_initiated &&
3395 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3397 msr_info->data = vcpu->arch.perf_capabilities;
3399 case MSR_IA32_POWER_CTL:
3400 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3402 case MSR_IA32_TSC: {
3404 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3405 * even when not intercepted. AMD manual doesn't explicitly
3406 * state this but appears to behave the same.
3408 * On userspace reads and writes, however, we unconditionally
3409 * return L1's TSC value to ensure backwards-compatible
3410 * behavior for migration.
3412 u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3413 vcpu->arch.tsc_offset;
3415 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3419 case 0x200 ... 0x2ff:
3420 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3421 case 0xcd: /* fsb frequency */
3425 * MSR_EBC_FREQUENCY_ID
3426 * Conservative value valid for even the basic CPU models.
3427 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3428 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3429 * and 266MHz for model 3, or 4. Set Core Clock
3430 * Frequency to System Bus Frequency Ratio to 1 (bits
3431 * 31:24) even though these are only valid for CPU
3432 * models > 2, however guests may end up dividing or
3433 * multiplying by zero otherwise.
3435 case MSR_EBC_FREQUENCY_ID:
3436 msr_info->data = 1 << 24;
3438 case MSR_IA32_APICBASE:
3439 msr_info->data = kvm_get_apic_base(vcpu);
3441 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3442 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3443 case MSR_IA32_TSCDEADLINE:
3444 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3446 case MSR_IA32_TSC_ADJUST:
3447 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3449 case MSR_IA32_MISC_ENABLE:
3450 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3452 case MSR_IA32_SMBASE:
3453 if (!msr_info->host_initiated)
3455 msr_info->data = vcpu->arch.smbase;
3458 msr_info->data = vcpu->arch.smi_count;
3460 case MSR_IA32_PERF_STATUS:
3461 /* TSC increment by tick */
3462 msr_info->data = 1000ULL;
3463 /* CPU multiplier */
3464 msr_info->data |= (((uint64_t)4ULL) << 40);
3467 msr_info->data = vcpu->arch.efer;
3469 case MSR_KVM_WALL_CLOCK:
3470 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3473 msr_info->data = vcpu->kvm->arch.wall_clock;
3475 case MSR_KVM_WALL_CLOCK_NEW:
3476 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3479 msr_info->data = vcpu->kvm->arch.wall_clock;
3481 case MSR_KVM_SYSTEM_TIME:
3482 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3485 msr_info->data = vcpu->arch.time;
3487 case MSR_KVM_SYSTEM_TIME_NEW:
3488 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3491 msr_info->data = vcpu->arch.time;
3493 case MSR_KVM_ASYNC_PF_EN:
3494 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3497 msr_info->data = vcpu->arch.apf.msr_en_val;
3499 case MSR_KVM_ASYNC_PF_INT:
3500 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3503 msr_info->data = vcpu->arch.apf.msr_int_val;
3505 case MSR_KVM_ASYNC_PF_ACK:
3506 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3511 case MSR_KVM_STEAL_TIME:
3512 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3515 msr_info->data = vcpu->arch.st.msr_val;
3517 case MSR_KVM_PV_EOI_EN:
3518 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3521 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3523 case MSR_KVM_POLL_CONTROL:
3524 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3527 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3529 case MSR_IA32_P5_MC_ADDR:
3530 case MSR_IA32_P5_MC_TYPE:
3531 case MSR_IA32_MCG_CAP:
3532 case MSR_IA32_MCG_CTL:
3533 case MSR_IA32_MCG_STATUS:
3534 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3535 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3536 msr_info->host_initiated);
3538 if (!msr_info->host_initiated &&
3539 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3541 msr_info->data = vcpu->arch.ia32_xss;
3543 case MSR_K7_CLK_CTL:
3545 * Provide expected ramp-up count for K7. All other
3546 * are set to zero, indicating minimum divisors for
3549 * This prevents guest kernels on AMD host with CPU
3550 * type 6, model 8 and higher from exploding due to
3551 * the rdmsr failing.
3553 msr_info->data = 0x20000000;
3555 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3556 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3557 case HV_X64_MSR_SYNDBG_OPTIONS:
3558 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3559 case HV_X64_MSR_CRASH_CTL:
3560 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3561 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3562 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3563 case HV_X64_MSR_TSC_EMULATION_STATUS:
3564 return kvm_hv_get_msr_common(vcpu,
3565 msr_info->index, &msr_info->data,
3566 msr_info->host_initiated);
3567 case MSR_IA32_BBL_CR_CTL3:
3568 /* This legacy MSR exists but isn't fully documented in current
3569 * silicon. It is however accessed by winxp in very narrow
3570 * scenarios where it sets bit #19, itself documented as
3571 * a "reserved" bit. Best effort attempt to source coherent
3572 * read data here should the balance of the register be
3573 * interpreted by the guest:
3575 * L2 cache control register 3: 64GB range, 256KB size,
3576 * enabled, latency 0x1, configured
3578 msr_info->data = 0xbe702111;
3580 case MSR_AMD64_OSVW_ID_LENGTH:
3581 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3583 msr_info->data = vcpu->arch.osvw.length;
3585 case MSR_AMD64_OSVW_STATUS:
3586 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3588 msr_info->data = vcpu->arch.osvw.status;
3590 case MSR_PLATFORM_INFO:
3591 if (!msr_info->host_initiated &&
3592 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3594 msr_info->data = vcpu->arch.msr_platform_info;
3596 case MSR_MISC_FEATURES_ENABLES:
3597 msr_info->data = vcpu->arch.msr_misc_features_enables;
3600 msr_info->data = vcpu->arch.msr_hwcr;
3603 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3604 return kvm_pmu_get_msr(vcpu, msr_info);
3605 return KVM_MSR_RET_INVALID;
3609 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3612 * Read or write a bunch of msrs. All parameters are kernel addresses.
3614 * @return number of msrs set successfully.
3616 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3617 struct kvm_msr_entry *entries,
3618 int (*do_msr)(struct kvm_vcpu *vcpu,
3619 unsigned index, u64 *data))
3623 for (i = 0; i < msrs->nmsrs; ++i)
3624 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3631 * Read or write a bunch of msrs. Parameters are user addresses.
3633 * @return number of msrs set successfully.
3635 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3636 int (*do_msr)(struct kvm_vcpu *vcpu,
3637 unsigned index, u64 *data),
3640 struct kvm_msrs msrs;
3641 struct kvm_msr_entry *entries;
3646 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3650 if (msrs.nmsrs >= MAX_IO_MSRS)
3653 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3654 entries = memdup_user(user_msrs->entries, size);
3655 if (IS_ERR(entries)) {
3656 r = PTR_ERR(entries);
3660 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3665 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3676 static inline bool kvm_can_mwait_in_guest(void)
3678 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3679 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3680 boot_cpu_has(X86_FEATURE_ARAT);
3683 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3688 case KVM_CAP_IRQCHIP:
3690 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3691 case KVM_CAP_SET_TSS_ADDR:
3692 case KVM_CAP_EXT_CPUID:
3693 case KVM_CAP_EXT_EMUL_CPUID:
3694 case KVM_CAP_CLOCKSOURCE:
3696 case KVM_CAP_NOP_IO_DELAY:
3697 case KVM_CAP_MP_STATE:
3698 case KVM_CAP_SYNC_MMU:
3699 case KVM_CAP_USER_NMI:
3700 case KVM_CAP_REINJECT_CONTROL:
3701 case KVM_CAP_IRQ_INJECT_STATUS:
3702 case KVM_CAP_IOEVENTFD:
3703 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3705 case KVM_CAP_PIT_STATE2:
3706 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3707 case KVM_CAP_XEN_HVM:
3708 case KVM_CAP_VCPU_EVENTS:
3709 case KVM_CAP_HYPERV:
3710 case KVM_CAP_HYPERV_VAPIC:
3711 case KVM_CAP_HYPERV_SPIN:
3712 case KVM_CAP_HYPERV_SYNIC:
3713 case KVM_CAP_HYPERV_SYNIC2:
3714 case KVM_CAP_HYPERV_VP_INDEX:
3715 case KVM_CAP_HYPERV_EVENTFD:
3716 case KVM_CAP_HYPERV_TLBFLUSH:
3717 case KVM_CAP_HYPERV_SEND_IPI:
3718 case KVM_CAP_HYPERV_CPUID:
3719 case KVM_CAP_PCI_SEGMENT:
3720 case KVM_CAP_DEBUGREGS:
3721 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3723 case KVM_CAP_ASYNC_PF:
3724 case KVM_CAP_ASYNC_PF_INT:
3725 case KVM_CAP_GET_TSC_KHZ:
3726 case KVM_CAP_KVMCLOCK_CTRL:
3727 case KVM_CAP_READONLY_MEM:
3728 case KVM_CAP_HYPERV_TIME:
3729 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3730 case KVM_CAP_TSC_DEADLINE_TIMER:
3731 case KVM_CAP_DISABLE_QUIRKS:
3732 case KVM_CAP_SET_BOOT_CPU_ID:
3733 case KVM_CAP_SPLIT_IRQCHIP:
3734 case KVM_CAP_IMMEDIATE_EXIT:
3735 case KVM_CAP_PMU_EVENT_FILTER:
3736 case KVM_CAP_GET_MSR_FEATURES:
3737 case KVM_CAP_MSR_PLATFORM_INFO:
3738 case KVM_CAP_EXCEPTION_PAYLOAD:
3739 case KVM_CAP_SET_GUEST_DEBUG:
3740 case KVM_CAP_LAST_CPU:
3741 case KVM_CAP_X86_USER_SPACE_MSR:
3742 case KVM_CAP_X86_MSR_FILTER:
3743 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3746 case KVM_CAP_SYNC_REGS:
3747 r = KVM_SYNC_X86_VALID_FIELDS;
3749 case KVM_CAP_ADJUST_CLOCK:
3750 r = KVM_CLOCK_TSC_STABLE;
3752 case KVM_CAP_X86_DISABLE_EXITS:
3753 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3754 KVM_X86_DISABLE_EXITS_CSTATE;
3755 if(kvm_can_mwait_in_guest())
3756 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3758 case KVM_CAP_X86_SMM:
3759 /* SMBASE is usually relocated above 1M on modern chipsets,
3760 * and SMM handlers might indeed rely on 4G segment limits,
3761 * so do not report SMM to be available if real mode is
3762 * emulated via vm86 mode. Still, do not go to great lengths
3763 * to avoid userspace's usage of the feature, because it is a
3764 * fringe case that is not enabled except via specific settings
3765 * of the module parameters.
3767 r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE);
3770 r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3772 case KVM_CAP_NR_VCPUS:
3773 r = KVM_SOFT_MAX_VCPUS;
3775 case KVM_CAP_MAX_VCPUS:
3778 case KVM_CAP_MAX_VCPU_ID:
3779 r = KVM_MAX_VCPU_ID;
3781 case KVM_CAP_PV_MMU: /* obsolete */
3785 r = KVM_MAX_MCE_BANKS;
3788 r = boot_cpu_has(X86_FEATURE_XSAVE);
3790 case KVM_CAP_TSC_CONTROL:
3791 r = kvm_has_tsc_control;
3793 case KVM_CAP_X2APIC_API:
3794 r = KVM_X2APIC_API_VALID_FLAGS;
3796 case KVM_CAP_NESTED_STATE:
3797 r = kvm_x86_ops.nested_ops->get_state ?
3798 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3800 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3801 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3803 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3804 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3806 case KVM_CAP_SMALLER_MAXPHYADDR:
3807 r = (int) allow_smaller_maxphyaddr;
3809 case KVM_CAP_STEAL_TIME:
3810 r = sched_info_on();
3819 long kvm_arch_dev_ioctl(struct file *filp,
3820 unsigned int ioctl, unsigned long arg)
3822 void __user *argp = (void __user *)arg;
3826 case KVM_GET_MSR_INDEX_LIST: {
3827 struct kvm_msr_list __user *user_msr_list = argp;
3828 struct kvm_msr_list msr_list;
3832 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3835 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3836 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3839 if (n < msr_list.nmsrs)
3842 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3843 num_msrs_to_save * sizeof(u32)))
3845 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3847 num_emulated_msrs * sizeof(u32)))
3852 case KVM_GET_SUPPORTED_CPUID:
3853 case KVM_GET_EMULATED_CPUID: {
3854 struct kvm_cpuid2 __user *cpuid_arg = argp;
3855 struct kvm_cpuid2 cpuid;
3858 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3861 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3867 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3872 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3874 if (copy_to_user(argp, &kvm_mce_cap_supported,
3875 sizeof(kvm_mce_cap_supported)))
3879 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3880 struct kvm_msr_list __user *user_msr_list = argp;
3881 struct kvm_msr_list msr_list;
3885 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3888 msr_list.nmsrs = num_msr_based_features;
3889 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3892 if (n < msr_list.nmsrs)
3895 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3896 num_msr_based_features * sizeof(u32)))
3902 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3912 static void wbinvd_ipi(void *garbage)
3917 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3919 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3922 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3924 /* Address WBINVD may be executed by guest */
3925 if (need_emulate_wbinvd(vcpu)) {
3926 if (kvm_x86_ops.has_wbinvd_exit())
3927 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3928 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3929 smp_call_function_single(vcpu->cpu,
3930 wbinvd_ipi, NULL, 1);
3933 kvm_x86_ops.vcpu_load(vcpu, cpu);
3935 /* Save host pkru register if supported */
3936 vcpu->arch.host_pkru = read_pkru();
3938 /* Apply any externally detected TSC adjustments (due to suspend) */
3939 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3940 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3941 vcpu->arch.tsc_offset_adjustment = 0;
3942 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3945 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3946 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3947 rdtsc() - vcpu->arch.last_host_tsc;
3949 mark_tsc_unstable("KVM discovered backwards TSC");
3951 if (kvm_check_tsc_unstable()) {
3952 u64 offset = kvm_compute_tsc_offset(vcpu,
3953 vcpu->arch.last_guest_tsc);
3954 kvm_vcpu_write_tsc_offset(vcpu, offset);
3955 vcpu->arch.tsc_catchup = 1;
3958 if (kvm_lapic_hv_timer_in_use(vcpu))
3959 kvm_lapic_restart_hv_timer(vcpu);
3962 * On a host with synchronized TSC, there is no need to update
3963 * kvmclock on vcpu->cpu migration
3965 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3966 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3967 if (vcpu->cpu != cpu)
3968 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3972 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3975 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3977 struct kvm_host_map map;
3978 struct kvm_steal_time *st;
3980 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3983 if (vcpu->arch.st.preempted)
3986 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3987 &vcpu->arch.st.cache, true))
3991 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3993 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
3995 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3998 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4002 if (vcpu->preempted)
4003 vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
4006 * Disable page faults because we're in atomic context here.
4007 * kvm_write_guest_offset_cached() would call might_fault()
4008 * that relies on pagefault_disable() to tell if there's a
4009 * bug. NOTE: the write to guest memory may not go through if
4010 * during postcopy live migration or if there's heavy guest
4013 pagefault_disable();
4015 * kvm_memslots() will be called by
4016 * kvm_write_guest_offset_cached() so take the srcu lock.
4018 idx = srcu_read_lock(&vcpu->kvm->srcu);
4019 kvm_steal_time_set_preempted(vcpu);
4020 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4022 kvm_x86_ops.vcpu_put(vcpu);
4023 vcpu->arch.last_host_tsc = rdtsc();
4025 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4026 * on every vmexit, but if not, we might have a stale dr6 from the
4027 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4032 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4033 struct kvm_lapic_state *s)
4035 if (vcpu->arch.apicv_active)
4036 kvm_x86_ops.sync_pir_to_irr(vcpu);
4038 return kvm_apic_get_state(vcpu, s);
4041 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4042 struct kvm_lapic_state *s)
4046 r = kvm_apic_set_state(vcpu, s);
4049 update_cr8_intercept(vcpu);
4054 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4056 return (!lapic_in_kernel(vcpu) ||
4057 kvm_apic_accept_pic_intr(vcpu));
4061 * if userspace requested an interrupt window, check that the
4062 * interrupt window is open.
4064 * No need to exit to userspace if we already have an interrupt queued.
4066 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4068 return kvm_arch_interrupt_allowed(vcpu) &&
4069 !kvm_cpu_has_interrupt(vcpu) &&
4070 !kvm_event_needs_reinjection(vcpu) &&
4071 kvm_cpu_accept_dm_intr(vcpu);
4074 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4075 struct kvm_interrupt *irq)
4077 if (irq->irq >= KVM_NR_INTERRUPTS)
4080 if (!irqchip_in_kernel(vcpu->kvm)) {
4081 kvm_queue_interrupt(vcpu, irq->irq, false);
4082 kvm_make_request(KVM_REQ_EVENT, vcpu);
4087 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4088 * fail for in-kernel 8259.
4090 if (pic_in_kernel(vcpu->kvm))
4093 if (vcpu->arch.pending_external_vector != -1)
4096 vcpu->arch.pending_external_vector = irq->irq;
4097 kvm_make_request(KVM_REQ_EVENT, vcpu);
4101 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4103 kvm_inject_nmi(vcpu);
4108 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4110 kvm_make_request(KVM_REQ_SMI, vcpu);
4115 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4116 struct kvm_tpr_access_ctl *tac)
4120 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4124 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4128 unsigned bank_num = mcg_cap & 0xff, bank;
4131 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4133 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4136 vcpu->arch.mcg_cap = mcg_cap;
4137 /* Init IA32_MCG_CTL to all 1s */
4138 if (mcg_cap & MCG_CTL_P)
4139 vcpu->arch.mcg_ctl = ~(u64)0;
4140 /* Init IA32_MCi_CTL to all 1s */
4141 for (bank = 0; bank < bank_num; bank++)
4142 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4144 kvm_x86_ops.setup_mce(vcpu);
4149 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4150 struct kvm_x86_mce *mce)
4152 u64 mcg_cap = vcpu->arch.mcg_cap;
4153 unsigned bank_num = mcg_cap & 0xff;
4154 u64 *banks = vcpu->arch.mce_banks;
4156 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4159 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4160 * reporting is disabled
4162 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4163 vcpu->arch.mcg_ctl != ~(u64)0)
4165 banks += 4 * mce->bank;
4167 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4168 * reporting is disabled for the bank
4170 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4172 if (mce->status & MCI_STATUS_UC) {
4173 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4174 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4175 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4178 if (banks[1] & MCI_STATUS_VAL)
4179 mce->status |= MCI_STATUS_OVER;
4180 banks[2] = mce->addr;
4181 banks[3] = mce->misc;
4182 vcpu->arch.mcg_status = mce->mcg_status;
4183 banks[1] = mce->status;
4184 kvm_queue_exception(vcpu, MC_VECTOR);
4185 } else if (!(banks[1] & MCI_STATUS_VAL)
4186 || !(banks[1] & MCI_STATUS_UC)) {
4187 if (banks[1] & MCI_STATUS_VAL)
4188 mce->status |= MCI_STATUS_OVER;
4189 banks[2] = mce->addr;
4190 banks[3] = mce->misc;
4191 banks[1] = mce->status;
4193 banks[1] |= MCI_STATUS_OVER;
4197 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4198 struct kvm_vcpu_events *events)
4203 * In guest mode, payload delivery should be deferred,
4204 * so that the L1 hypervisor can intercept #PF before
4205 * CR2 is modified (or intercept #DB before DR6 is
4206 * modified under nVMX). Unless the per-VM capability,
4207 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4208 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4209 * opportunistically defer the exception payload, deliver it if the
4210 * capability hasn't been requested before processing a
4211 * KVM_GET_VCPU_EVENTS.
4213 if (!vcpu->kvm->arch.exception_payload_enabled &&
4214 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4215 kvm_deliver_exception_payload(vcpu);
4218 * The API doesn't provide the instruction length for software
4219 * exceptions, so don't report them. As long as the guest RIP
4220 * isn't advanced, we should expect to encounter the exception
4223 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4224 events->exception.injected = 0;
4225 events->exception.pending = 0;
4227 events->exception.injected = vcpu->arch.exception.injected;
4228 events->exception.pending = vcpu->arch.exception.pending;
4230 * For ABI compatibility, deliberately conflate
4231 * pending and injected exceptions when
4232 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4234 if (!vcpu->kvm->arch.exception_payload_enabled)
4235 events->exception.injected |=
4236 vcpu->arch.exception.pending;
4238 events->exception.nr = vcpu->arch.exception.nr;
4239 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4240 events->exception.error_code = vcpu->arch.exception.error_code;
4241 events->exception_has_payload = vcpu->arch.exception.has_payload;
4242 events->exception_payload = vcpu->arch.exception.payload;
4244 events->interrupt.injected =
4245 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4246 events->interrupt.nr = vcpu->arch.interrupt.nr;
4247 events->interrupt.soft = 0;
4248 events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
4250 events->nmi.injected = vcpu->arch.nmi_injected;
4251 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4252 events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
4253 events->nmi.pad = 0;
4255 events->sipi_vector = 0; /* never valid when reporting to user space */
4257 events->smi.smm = is_smm(vcpu);
4258 events->smi.pending = vcpu->arch.smi_pending;
4259 events->smi.smm_inside_nmi =
4260 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4261 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4263 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4264 | KVM_VCPUEVENT_VALID_SHADOW
4265 | KVM_VCPUEVENT_VALID_SMM);
4266 if (vcpu->kvm->arch.exception_payload_enabled)
4267 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4269 memset(&events->reserved, 0, sizeof(events->reserved));
4272 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4274 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4275 struct kvm_vcpu_events *events)
4277 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4278 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4279 | KVM_VCPUEVENT_VALID_SHADOW
4280 | KVM_VCPUEVENT_VALID_SMM
4281 | KVM_VCPUEVENT_VALID_PAYLOAD))
4284 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4285 if (!vcpu->kvm->arch.exception_payload_enabled)
4287 if (events->exception.pending)
4288 events->exception.injected = 0;
4290 events->exception_has_payload = 0;
4292 events->exception.pending = 0;
4293 events->exception_has_payload = 0;
4296 if ((events->exception.injected || events->exception.pending) &&
4297 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4300 /* INITs are latched while in SMM */
4301 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4302 (events->smi.smm || events->smi.pending) &&
4303 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4307 vcpu->arch.exception.injected = events->exception.injected;
4308 vcpu->arch.exception.pending = events->exception.pending;
4309 vcpu->arch.exception.nr = events->exception.nr;
4310 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4311 vcpu->arch.exception.error_code = events->exception.error_code;
4312 vcpu->arch.exception.has_payload = events->exception_has_payload;
4313 vcpu->arch.exception.payload = events->exception_payload;
4315 vcpu->arch.interrupt.injected = events->interrupt.injected;
4316 vcpu->arch.interrupt.nr = events->interrupt.nr;
4317 vcpu->arch.interrupt.soft = events->interrupt.soft;
4318 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4319 kvm_x86_ops.set_interrupt_shadow(vcpu,
4320 events->interrupt.shadow);
4322 vcpu->arch.nmi_injected = events->nmi.injected;
4323 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4324 vcpu->arch.nmi_pending = events->nmi.pending;
4325 kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
4327 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4328 lapic_in_kernel(vcpu))
4329 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4331 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4332 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4333 if (events->smi.smm)
4334 vcpu->arch.hflags |= HF_SMM_MASK;
4336 vcpu->arch.hflags &= ~HF_SMM_MASK;
4337 kvm_smm_changed(vcpu);
4340 vcpu->arch.smi_pending = events->smi.pending;
4342 if (events->smi.smm) {
4343 if (events->smi.smm_inside_nmi)
4344 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4346 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4349 if (lapic_in_kernel(vcpu)) {
4350 if (events->smi.latched_init)
4351 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4353 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4357 kvm_make_request(KVM_REQ_EVENT, vcpu);
4362 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4363 struct kvm_debugregs *dbgregs)
4367 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4368 kvm_get_dr(vcpu, 6, &val);
4370 dbgregs->dr7 = vcpu->arch.dr7;
4372 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4375 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4376 struct kvm_debugregs *dbgregs)
4381 if (dbgregs->dr6 & ~0xffffffffull)
4383 if (dbgregs->dr7 & ~0xffffffffull)
4386 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4387 kvm_update_dr0123(vcpu);
4388 vcpu->arch.dr6 = dbgregs->dr6;
4389 vcpu->arch.dr7 = dbgregs->dr7;
4390 kvm_update_dr7(vcpu);
4395 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4397 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4399 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4400 u64 xstate_bv = xsave->header.xfeatures;
4404 * Copy legacy XSAVE area, to avoid complications with CPUID
4405 * leaves 0 and 1 in the loop below.
4407 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4410 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4411 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4414 * Copy each region from the possibly compacted offset to the
4415 * non-compacted offset.
4417 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4419 u64 xfeature_mask = valid & -valid;
4420 int xfeature_nr = fls64(xfeature_mask) - 1;
4421 void *src = get_xsave_addr(xsave, xfeature_nr);
4424 u32 size, offset, ecx, edx;
4425 cpuid_count(XSTATE_CPUID, xfeature_nr,
4426 &size, &offset, &ecx, &edx);
4427 if (xfeature_nr == XFEATURE_PKRU)
4428 memcpy(dest + offset, &vcpu->arch.pkru,
4429 sizeof(vcpu->arch.pkru));
4431 memcpy(dest + offset, src, size);
4435 valid -= xfeature_mask;
4439 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4441 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4442 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4446 * Copy legacy XSAVE area, to avoid complications with CPUID
4447 * leaves 0 and 1 in the loop below.
4449 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4451 /* Set XSTATE_BV and possibly XCOMP_BV. */
4452 xsave->header.xfeatures = xstate_bv;
4453 if (boot_cpu_has(X86_FEATURE_XSAVES))
4454 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4457 * Copy each region from the non-compacted offset to the
4458 * possibly compacted offset.
4460 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4462 u64 xfeature_mask = valid & -valid;
4463 int xfeature_nr = fls64(xfeature_mask) - 1;
4464 void *dest = get_xsave_addr(xsave, xfeature_nr);
4467 u32 size, offset, ecx, edx;
4468 cpuid_count(XSTATE_CPUID, xfeature_nr,
4469 &size, &offset, &ecx, &edx);
4470 if (xfeature_nr == XFEATURE_PKRU)
4471 memcpy(&vcpu->arch.pkru, src + offset,
4472 sizeof(vcpu->arch.pkru));
4474 memcpy(dest, src + offset, size);
4477 valid -= xfeature_mask;
4481 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4482 struct kvm_xsave *guest_xsave)
4484 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4485 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4486 fill_xsave((u8 *) guest_xsave->region, vcpu);
4488 memcpy(guest_xsave->region,
4489 &vcpu->arch.guest_fpu->state.fxsave,
4490 sizeof(struct fxregs_state));
4491 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4492 XFEATURE_MASK_FPSSE;
4496 #define XSAVE_MXCSR_OFFSET 24
4498 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4499 struct kvm_xsave *guest_xsave)
4502 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4503 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4505 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4507 * Here we allow setting states that are not present in
4508 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4509 * with old userspace.
4511 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4513 load_xsave(vcpu, (u8 *)guest_xsave->region);
4515 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4516 mxcsr & ~mxcsr_feature_mask)
4518 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4519 guest_xsave->region, sizeof(struct fxregs_state));
4524 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4525 struct kvm_xcrs *guest_xcrs)
4527 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4528 guest_xcrs->nr_xcrs = 0;
4532 guest_xcrs->nr_xcrs = 1;
4533 guest_xcrs->flags = 0;
4534 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4535 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4538 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4539 struct kvm_xcrs *guest_xcrs)
4543 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4546 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4549 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4550 /* Only support XCR0 currently */
4551 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4552 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4553 guest_xcrs->xcrs[i].value);
4562 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4563 * stopped by the hypervisor. This function will be called from the host only.
4564 * EINVAL is returned when the host attempts to set the flag for a guest that
4565 * does not support pv clocks.
4567 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4569 if (!vcpu->arch.pv_time_enabled)
4571 vcpu->arch.pvclock_set_guest_stopped_request = true;
4572 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4576 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4577 struct kvm_enable_cap *cap)
4580 uint16_t vmcs_version;
4581 void __user *user_ptr;
4587 case KVM_CAP_HYPERV_SYNIC2:
4592 case KVM_CAP_HYPERV_SYNIC:
4593 if (!irqchip_in_kernel(vcpu->kvm))
4595 return kvm_hv_activate_synic(vcpu, cap->cap ==
4596 KVM_CAP_HYPERV_SYNIC2);
4597 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4598 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4600 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4602 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4603 if (copy_to_user(user_ptr, &vmcs_version,
4604 sizeof(vmcs_version)))
4608 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4609 if (!kvm_x86_ops.enable_direct_tlbflush)
4612 return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4614 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4615 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4616 if (vcpu->arch.pv_cpuid.enforce)
4617 kvm_update_pv_runtime(vcpu);
4626 long kvm_arch_vcpu_ioctl(struct file *filp,
4627 unsigned int ioctl, unsigned long arg)
4629 struct kvm_vcpu *vcpu = filp->private_data;
4630 void __user *argp = (void __user *)arg;
4633 struct kvm_lapic_state *lapic;
4634 struct kvm_xsave *xsave;
4635 struct kvm_xcrs *xcrs;
4643 case KVM_GET_LAPIC: {
4645 if (!lapic_in_kernel(vcpu))
4647 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4648 GFP_KERNEL_ACCOUNT);
4653 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4657 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4662 case KVM_SET_LAPIC: {
4664 if (!lapic_in_kernel(vcpu))
4666 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4667 if (IS_ERR(u.lapic)) {
4668 r = PTR_ERR(u.lapic);
4672 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4675 case KVM_INTERRUPT: {
4676 struct kvm_interrupt irq;
4679 if (copy_from_user(&irq, argp, sizeof(irq)))
4681 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4685 r = kvm_vcpu_ioctl_nmi(vcpu);
4689 r = kvm_vcpu_ioctl_smi(vcpu);
4692 case KVM_SET_CPUID: {
4693 struct kvm_cpuid __user *cpuid_arg = argp;
4694 struct kvm_cpuid cpuid;
4697 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4699 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4702 case KVM_SET_CPUID2: {
4703 struct kvm_cpuid2 __user *cpuid_arg = argp;
4704 struct kvm_cpuid2 cpuid;
4707 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4709 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4710 cpuid_arg->entries);
4713 case KVM_GET_CPUID2: {
4714 struct kvm_cpuid2 __user *cpuid_arg = argp;
4715 struct kvm_cpuid2 cpuid;
4718 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4720 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4721 cpuid_arg->entries);
4725 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4730 case KVM_GET_MSRS: {
4731 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4732 r = msr_io(vcpu, argp, do_get_msr, 1);
4733 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4736 case KVM_SET_MSRS: {
4737 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4738 r = msr_io(vcpu, argp, do_set_msr, 0);
4739 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4742 case KVM_TPR_ACCESS_REPORTING: {
4743 struct kvm_tpr_access_ctl tac;
4746 if (copy_from_user(&tac, argp, sizeof(tac)))
4748 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4752 if (copy_to_user(argp, &tac, sizeof(tac)))
4757 case KVM_SET_VAPIC_ADDR: {
4758 struct kvm_vapic_addr va;
4762 if (!lapic_in_kernel(vcpu))
4765 if (copy_from_user(&va, argp, sizeof(va)))
4767 idx = srcu_read_lock(&vcpu->kvm->srcu);
4768 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4769 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4772 case KVM_X86_SETUP_MCE: {
4776 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4778 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4781 case KVM_X86_SET_MCE: {
4782 struct kvm_x86_mce mce;
4785 if (copy_from_user(&mce, argp, sizeof(mce)))
4787 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4790 case KVM_GET_VCPU_EVENTS: {
4791 struct kvm_vcpu_events events;
4793 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4796 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4801 case KVM_SET_VCPU_EVENTS: {
4802 struct kvm_vcpu_events events;
4805 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4808 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4811 case KVM_GET_DEBUGREGS: {
4812 struct kvm_debugregs dbgregs;
4814 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4817 if (copy_to_user(argp, &dbgregs,
4818 sizeof(struct kvm_debugregs)))
4823 case KVM_SET_DEBUGREGS: {
4824 struct kvm_debugregs dbgregs;
4827 if (copy_from_user(&dbgregs, argp,
4828 sizeof(struct kvm_debugregs)))
4831 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4834 case KVM_GET_XSAVE: {
4835 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4840 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4843 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4848 case KVM_SET_XSAVE: {
4849 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4850 if (IS_ERR(u.xsave)) {
4851 r = PTR_ERR(u.xsave);
4855 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4858 case KVM_GET_XCRS: {
4859 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4864 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4867 if (copy_to_user(argp, u.xcrs,
4868 sizeof(struct kvm_xcrs)))
4873 case KVM_SET_XCRS: {
4874 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4875 if (IS_ERR(u.xcrs)) {
4876 r = PTR_ERR(u.xcrs);
4880 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4883 case KVM_SET_TSC_KHZ: {
4887 user_tsc_khz = (u32)arg;
4889 if (kvm_has_tsc_control &&
4890 user_tsc_khz >= kvm_max_guest_tsc_khz)
4893 if (user_tsc_khz == 0)
4894 user_tsc_khz = tsc_khz;
4896 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4901 case KVM_GET_TSC_KHZ: {
4902 r = vcpu->arch.virtual_tsc_khz;
4905 case KVM_KVMCLOCK_CTRL: {
4906 r = kvm_set_guest_paused(vcpu);
4909 case KVM_ENABLE_CAP: {
4910 struct kvm_enable_cap cap;
4913 if (copy_from_user(&cap, argp, sizeof(cap)))
4915 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4918 case KVM_GET_NESTED_STATE: {
4919 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4923 if (!kvm_x86_ops.nested_ops->get_state)
4926 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4928 if (get_user(user_data_size, &user_kvm_nested_state->size))
4931 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
4936 if (r > user_data_size) {
4937 if (put_user(r, &user_kvm_nested_state->size))
4947 case KVM_SET_NESTED_STATE: {
4948 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4949 struct kvm_nested_state kvm_state;
4953 if (!kvm_x86_ops.nested_ops->set_state)
4957 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4961 if (kvm_state.size < sizeof(kvm_state))
4964 if (kvm_state.flags &
4965 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4966 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
4967 | KVM_STATE_NESTED_GIF_SET))
4970 /* nested_run_pending implies guest_mode. */
4971 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4972 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4975 idx = srcu_read_lock(&vcpu->kvm->srcu);
4976 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
4977 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4980 case KVM_GET_SUPPORTED_HV_CPUID: {
4981 struct kvm_cpuid2 __user *cpuid_arg = argp;
4982 struct kvm_cpuid2 cpuid;
4985 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4988 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4989 cpuid_arg->entries);
4994 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5009 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5011 return VM_FAULT_SIGBUS;
5014 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5018 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5020 ret = kvm_x86_ops.set_tss_addr(kvm, addr);
5024 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5027 return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
5030 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5031 unsigned long kvm_nr_mmu_pages)
5033 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5036 mutex_lock(&kvm->slots_lock);
5038 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5039 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5041 mutex_unlock(&kvm->slots_lock);
5045 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5047 return kvm->arch.n_max_mmu_pages;
5050 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5052 struct kvm_pic *pic = kvm->arch.vpic;
5056 switch (chip->chip_id) {
5057 case KVM_IRQCHIP_PIC_MASTER:
5058 memcpy(&chip->chip.pic, &pic->pics[0],
5059 sizeof(struct kvm_pic_state));
5061 case KVM_IRQCHIP_PIC_SLAVE:
5062 memcpy(&chip->chip.pic, &pic->pics[1],
5063 sizeof(struct kvm_pic_state));
5065 case KVM_IRQCHIP_IOAPIC:
5066 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5075 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5077 struct kvm_pic *pic = kvm->arch.vpic;
5081 switch (chip->chip_id) {
5082 case KVM_IRQCHIP_PIC_MASTER:
5083 spin_lock(&pic->lock);
5084 memcpy(&pic->pics[0], &chip->chip.pic,
5085 sizeof(struct kvm_pic_state));
5086 spin_unlock(&pic->lock);
5088 case KVM_IRQCHIP_PIC_SLAVE:
5089 spin_lock(&pic->lock);
5090 memcpy(&pic->pics[1], &chip->chip.pic,
5091 sizeof(struct kvm_pic_state));
5092 spin_unlock(&pic->lock);
5094 case KVM_IRQCHIP_IOAPIC:
5095 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5101 kvm_pic_update_irq(pic);
5105 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5107 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5109 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5111 mutex_lock(&kps->lock);
5112 memcpy(ps, &kps->channels, sizeof(*ps));
5113 mutex_unlock(&kps->lock);
5117 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5120 struct kvm_pit *pit = kvm->arch.vpit;
5122 mutex_lock(&pit->pit_state.lock);
5123 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5124 for (i = 0; i < 3; i++)
5125 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5126 mutex_unlock(&pit->pit_state.lock);
5130 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5132 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5133 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5134 sizeof(ps->channels));
5135 ps->flags = kvm->arch.vpit->pit_state.flags;
5136 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5137 memset(&ps->reserved, 0, sizeof(ps->reserved));
5141 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5145 u32 prev_legacy, cur_legacy;
5146 struct kvm_pit *pit = kvm->arch.vpit;
5148 mutex_lock(&pit->pit_state.lock);
5149 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5150 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5151 if (!prev_legacy && cur_legacy)
5153 memcpy(&pit->pit_state.channels, &ps->channels,
5154 sizeof(pit->pit_state.channels));
5155 pit->pit_state.flags = ps->flags;
5156 for (i = 0; i < 3; i++)
5157 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5159 mutex_unlock(&pit->pit_state.lock);
5163 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5164 struct kvm_reinject_control *control)
5166 struct kvm_pit *pit = kvm->arch.vpit;
5168 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5169 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5170 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5172 mutex_lock(&pit->pit_state.lock);
5173 kvm_pit_set_reinject(pit, control->pit_reinject);
5174 mutex_unlock(&pit->pit_state.lock);
5179 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5182 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
5184 if (kvm_x86_ops.flush_log_dirty)
5185 kvm_x86_ops.flush_log_dirty(kvm);
5188 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5191 if (!irqchip_in_kernel(kvm))
5194 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5195 irq_event->irq, irq_event->level,
5200 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5201 struct kvm_enable_cap *cap)
5209 case KVM_CAP_DISABLE_QUIRKS:
5210 kvm->arch.disabled_quirks = cap->args[0];
5213 case KVM_CAP_SPLIT_IRQCHIP: {
5214 mutex_lock(&kvm->lock);
5216 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5217 goto split_irqchip_unlock;
5219 if (irqchip_in_kernel(kvm))
5220 goto split_irqchip_unlock;
5221 if (kvm->created_vcpus)
5222 goto split_irqchip_unlock;
5223 r = kvm_setup_empty_irq_routing(kvm);
5225 goto split_irqchip_unlock;
5226 /* Pairs with irqchip_in_kernel. */
5228 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5229 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5231 split_irqchip_unlock:
5232 mutex_unlock(&kvm->lock);
5235 case KVM_CAP_X2APIC_API:
5237 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5240 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5241 kvm->arch.x2apic_format = true;
5242 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5243 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5247 case KVM_CAP_X86_DISABLE_EXITS:
5249 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5252 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5253 kvm_can_mwait_in_guest())
5254 kvm->arch.mwait_in_guest = true;
5255 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5256 kvm->arch.hlt_in_guest = true;
5257 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5258 kvm->arch.pause_in_guest = true;
5259 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5260 kvm->arch.cstate_in_guest = true;
5263 case KVM_CAP_MSR_PLATFORM_INFO:
5264 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5267 case KVM_CAP_EXCEPTION_PAYLOAD:
5268 kvm->arch.exception_payload_enabled = cap->args[0];
5271 case KVM_CAP_X86_USER_SPACE_MSR:
5272 kvm->arch.user_space_msr_mask = cap->args[0];
5282 static void kvm_clear_msr_filter(struct kvm *kvm)
5285 u32 count = kvm->arch.msr_filter.count;
5286 struct msr_bitmap_range ranges[16];
5288 mutex_lock(&kvm->lock);
5289 kvm->arch.msr_filter.count = 0;
5290 memcpy(ranges, kvm->arch.msr_filter.ranges, count * sizeof(ranges[0]));
5291 mutex_unlock(&kvm->lock);
5292 synchronize_srcu(&kvm->srcu);
5294 for (i = 0; i < count; i++)
5295 kfree(ranges[i].bitmap);
5298 static int kvm_add_msr_filter(struct kvm *kvm, struct kvm_msr_filter_range *user_range)
5300 struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
5301 struct msr_bitmap_range range;
5302 unsigned long *bitmap = NULL;
5306 if (!user_range->nmsrs)
5309 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5310 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5313 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5315 return PTR_ERR(bitmap);
5317 range = (struct msr_bitmap_range) {
5318 .flags = user_range->flags,
5319 .base = user_range->base,
5320 .nmsrs = user_range->nmsrs,
5324 if (range.flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE)) {
5334 /* Everything ok, add this range identifier to our global pool */
5335 ranges[kvm->arch.msr_filter.count] = range;
5336 /* Make sure we filled the array before we tell anyone to walk it */
5338 kvm->arch.msr_filter.count++;
5346 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5348 struct kvm_msr_filter __user *user_msr_filter = argp;
5349 struct kvm_msr_filter filter;
5355 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5358 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5359 empty &= !filter.ranges[i].nmsrs;
5361 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5362 if (empty && !default_allow)
5365 kvm_clear_msr_filter(kvm);
5367 kvm->arch.msr_filter.default_allow = default_allow;
5370 * Protect from concurrent calls to this function that could trigger
5371 * a TOCTOU violation on kvm->arch.msr_filter.count.
5373 mutex_lock(&kvm->lock);
5374 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5375 r = kvm_add_msr_filter(kvm, &filter.ranges[i]);
5380 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5381 mutex_unlock(&kvm->lock);
5386 long kvm_arch_vm_ioctl(struct file *filp,
5387 unsigned int ioctl, unsigned long arg)
5389 struct kvm *kvm = filp->private_data;
5390 void __user *argp = (void __user *)arg;
5393 * This union makes it completely explicit to gcc-3.x
5394 * that these two variables' stack usage should be
5395 * combined, not added together.
5398 struct kvm_pit_state ps;
5399 struct kvm_pit_state2 ps2;
5400 struct kvm_pit_config pit_config;
5404 case KVM_SET_TSS_ADDR:
5405 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5407 case KVM_SET_IDENTITY_MAP_ADDR: {
5410 mutex_lock(&kvm->lock);
5412 if (kvm->created_vcpus)
5413 goto set_identity_unlock;
5415 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5416 goto set_identity_unlock;
5417 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5418 set_identity_unlock:
5419 mutex_unlock(&kvm->lock);
5422 case KVM_SET_NR_MMU_PAGES:
5423 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5425 case KVM_GET_NR_MMU_PAGES:
5426 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5428 case KVM_CREATE_IRQCHIP: {
5429 mutex_lock(&kvm->lock);
5432 if (irqchip_in_kernel(kvm))
5433 goto create_irqchip_unlock;
5436 if (kvm->created_vcpus)
5437 goto create_irqchip_unlock;
5439 r = kvm_pic_init(kvm);
5441 goto create_irqchip_unlock;
5443 r = kvm_ioapic_init(kvm);
5445 kvm_pic_destroy(kvm);
5446 goto create_irqchip_unlock;
5449 r = kvm_setup_default_irq_routing(kvm);
5451 kvm_ioapic_destroy(kvm);
5452 kvm_pic_destroy(kvm);
5453 goto create_irqchip_unlock;
5455 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5457 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5458 create_irqchip_unlock:
5459 mutex_unlock(&kvm->lock);
5462 case KVM_CREATE_PIT:
5463 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5465 case KVM_CREATE_PIT2:
5467 if (copy_from_user(&u.pit_config, argp,
5468 sizeof(struct kvm_pit_config)))
5471 mutex_lock(&kvm->lock);
5474 goto create_pit_unlock;
5476 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5480 mutex_unlock(&kvm->lock);
5482 case KVM_GET_IRQCHIP: {
5483 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5484 struct kvm_irqchip *chip;
5486 chip = memdup_user(argp, sizeof(*chip));
5493 if (!irqchip_kernel(kvm))
5494 goto get_irqchip_out;
5495 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5497 goto get_irqchip_out;
5499 if (copy_to_user(argp, chip, sizeof(*chip)))
5500 goto get_irqchip_out;
5506 case KVM_SET_IRQCHIP: {
5507 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5508 struct kvm_irqchip *chip;
5510 chip = memdup_user(argp, sizeof(*chip));
5517 if (!irqchip_kernel(kvm))
5518 goto set_irqchip_out;
5519 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5526 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5529 if (!kvm->arch.vpit)
5531 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5535 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5542 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5544 mutex_lock(&kvm->lock);
5546 if (!kvm->arch.vpit)
5548 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5550 mutex_unlock(&kvm->lock);
5553 case KVM_GET_PIT2: {
5555 if (!kvm->arch.vpit)
5557 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5561 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5566 case KVM_SET_PIT2: {
5568 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5570 mutex_lock(&kvm->lock);
5572 if (!kvm->arch.vpit)
5574 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5576 mutex_unlock(&kvm->lock);
5579 case KVM_REINJECT_CONTROL: {
5580 struct kvm_reinject_control control;
5582 if (copy_from_user(&control, argp, sizeof(control)))
5585 if (!kvm->arch.vpit)
5587 r = kvm_vm_ioctl_reinject(kvm, &control);
5590 case KVM_SET_BOOT_CPU_ID:
5592 mutex_lock(&kvm->lock);
5593 if (kvm->created_vcpus)
5596 kvm->arch.bsp_vcpu_id = arg;
5597 mutex_unlock(&kvm->lock);
5599 case KVM_XEN_HVM_CONFIG: {
5600 struct kvm_xen_hvm_config xhc;
5602 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5607 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5611 case KVM_SET_CLOCK: {
5612 struct kvm_clock_data user_ns;
5616 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5625 * TODO: userspace has to take care of races with VCPU_RUN, so
5626 * kvm_gen_update_masterclock() can be cut down to locked
5627 * pvclock_update_vm_gtod_copy().
5629 kvm_gen_update_masterclock(kvm);
5630 now_ns = get_kvmclock_ns(kvm);
5631 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5632 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5635 case KVM_GET_CLOCK: {
5636 struct kvm_clock_data user_ns;
5639 now_ns = get_kvmclock_ns(kvm);
5640 user_ns.clock = now_ns;
5641 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5642 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5645 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5650 case KVM_MEMORY_ENCRYPT_OP: {
5652 if (kvm_x86_ops.mem_enc_op)
5653 r = kvm_x86_ops.mem_enc_op(kvm, argp);
5656 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5657 struct kvm_enc_region region;
5660 if (copy_from_user(®ion, argp, sizeof(region)))
5664 if (kvm_x86_ops.mem_enc_reg_region)
5665 r = kvm_x86_ops.mem_enc_reg_region(kvm, ®ion);
5668 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5669 struct kvm_enc_region region;
5672 if (copy_from_user(®ion, argp, sizeof(region)))
5676 if (kvm_x86_ops.mem_enc_unreg_region)
5677 r = kvm_x86_ops.mem_enc_unreg_region(kvm, ®ion);
5680 case KVM_HYPERV_EVENTFD: {
5681 struct kvm_hyperv_eventfd hvevfd;
5684 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5686 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5689 case KVM_SET_PMU_EVENT_FILTER:
5690 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5692 case KVM_X86_SET_MSR_FILTER:
5693 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5702 static void kvm_init_msr_list(void)
5704 struct x86_pmu_capability x86_pmu;
5708 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5709 "Please update the fixed PMCs in msrs_to_saved_all[]");
5711 perf_get_x86_pmu_capability(&x86_pmu);
5713 num_msrs_to_save = 0;
5714 num_emulated_msrs = 0;
5715 num_msr_based_features = 0;
5717 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5718 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5722 * Even MSRs that are valid in the host may not be exposed
5723 * to the guests in some cases.
5725 switch (msrs_to_save_all[i]) {
5726 case MSR_IA32_BNDCFGS:
5727 if (!kvm_mpx_supported())
5731 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5734 case MSR_IA32_UMWAIT_CONTROL:
5735 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5738 case MSR_IA32_RTIT_CTL:
5739 case MSR_IA32_RTIT_STATUS:
5740 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5743 case MSR_IA32_RTIT_CR3_MATCH:
5744 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5745 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5748 case MSR_IA32_RTIT_OUTPUT_BASE:
5749 case MSR_IA32_RTIT_OUTPUT_MASK:
5750 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5751 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5752 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5755 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5756 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5757 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5758 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5761 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5762 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5763 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5766 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5767 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5768 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5775 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5778 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5779 if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i]))
5782 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5785 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5786 struct kvm_msr_entry msr;
5788 msr.index = msr_based_features_all[i];
5789 if (kvm_get_msr_feature(&msr))
5792 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5796 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5804 if (!(lapic_in_kernel(vcpu) &&
5805 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5806 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5817 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5824 if (!(lapic_in_kernel(vcpu) &&
5825 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5827 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5829 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5839 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5840 struct kvm_segment *var, int seg)
5842 kvm_x86_ops.set_segment(vcpu, var, seg);
5845 void kvm_get_segment(struct kvm_vcpu *vcpu,
5846 struct kvm_segment *var, int seg)
5848 kvm_x86_ops.get_segment(vcpu, var, seg);
5851 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5852 struct x86_exception *exception)
5856 BUG_ON(!mmu_is_nested(vcpu));
5858 /* NPT walks are always user-walks */
5859 access |= PFERR_USER_MASK;
5860 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5865 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5866 struct x86_exception *exception)
5868 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5869 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5872 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5873 struct x86_exception *exception)
5875 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5876 access |= PFERR_FETCH_MASK;
5877 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5880 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5881 struct x86_exception *exception)
5883 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5884 access |= PFERR_WRITE_MASK;
5885 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5888 /* uses this to access any guest's mapped memory without checking CPL */
5889 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5890 struct x86_exception *exception)
5892 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5895 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5896 struct kvm_vcpu *vcpu, u32 access,
5897 struct x86_exception *exception)
5900 int r = X86EMUL_CONTINUE;
5903 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5905 unsigned offset = addr & (PAGE_SIZE-1);
5906 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5909 if (gpa == UNMAPPED_GVA)
5910 return X86EMUL_PROPAGATE_FAULT;
5911 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5914 r = X86EMUL_IO_NEEDED;
5926 /* used for instruction fetching */
5927 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5928 gva_t addr, void *val, unsigned int bytes,
5929 struct x86_exception *exception)
5931 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5932 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5936 /* Inline kvm_read_guest_virt_helper for speed. */
5937 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5939 if (unlikely(gpa == UNMAPPED_GVA))
5940 return X86EMUL_PROPAGATE_FAULT;
5942 offset = addr & (PAGE_SIZE-1);
5943 if (WARN_ON(offset + bytes > PAGE_SIZE))
5944 bytes = (unsigned)PAGE_SIZE - offset;
5945 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5947 if (unlikely(ret < 0))
5948 return X86EMUL_IO_NEEDED;
5950 return X86EMUL_CONTINUE;
5953 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5954 gva_t addr, void *val, unsigned int bytes,
5955 struct x86_exception *exception)
5957 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5960 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5961 * is returned, but our callers are not ready for that and they blindly
5962 * call kvm_inject_page_fault. Ensure that they at least do not leak
5963 * uninitialized kernel stack memory into cr2 and error code.
5965 memset(exception, 0, sizeof(*exception));
5966 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5969 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5971 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5972 gva_t addr, void *val, unsigned int bytes,
5973 struct x86_exception *exception, bool system)
5975 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5978 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5979 access |= PFERR_USER_MASK;
5981 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5984 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5985 unsigned long addr, void *val, unsigned int bytes)
5987 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5988 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5990 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5993 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5994 struct kvm_vcpu *vcpu, u32 access,
5995 struct x86_exception *exception)
5998 int r = X86EMUL_CONTINUE;
6001 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6004 unsigned offset = addr & (PAGE_SIZE-1);
6005 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6008 if (gpa == UNMAPPED_GVA)
6009 return X86EMUL_PROPAGATE_FAULT;
6010 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6012 r = X86EMUL_IO_NEEDED;
6024 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6025 unsigned int bytes, struct x86_exception *exception,
6028 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6029 u32 access = PFERR_WRITE_MASK;
6031 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
6032 access |= PFERR_USER_MASK;
6034 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6038 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6039 unsigned int bytes, struct x86_exception *exception)
6041 /* kvm_write_guest_virt_system can pull in tons of pages. */
6042 vcpu->arch.l1tf_flush_l1d = true;
6044 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6045 PFERR_WRITE_MASK, exception);
6047 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6049 int handle_ud(struct kvm_vcpu *vcpu)
6051 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6052 int emul_type = EMULTYPE_TRAP_UD;
6053 char sig[5]; /* ud2; .ascii "kvm" */
6054 struct x86_exception e;
6056 if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, NULL, 0)))
6059 if (force_emulation_prefix &&
6060 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6061 sig, sizeof(sig), &e) == 0 &&
6062 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6063 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6064 emul_type = EMULTYPE_TRAP_UD_FORCED;
6067 return kvm_emulate_instruction(vcpu, emul_type);
6069 EXPORT_SYMBOL_GPL(handle_ud);
6071 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6072 gpa_t gpa, bool write)
6074 /* For APIC access vmexit */
6075 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6078 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6079 trace_vcpu_match_mmio(gva, gpa, write, true);
6086 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6087 gpa_t *gpa, struct x86_exception *exception,
6090 u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
6091 | (write ? PFERR_WRITE_MASK : 0);
6094 * currently PKRU is only applied to ept enabled guest so
6095 * there is no pkey in EPT page table for L1 guest or EPT
6096 * shadow page table for L2 guest.
6098 if (vcpu_match_mmio_gva(vcpu, gva)
6099 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6100 vcpu->arch.mmio_access, 0, access)) {
6101 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6102 (gva & (PAGE_SIZE - 1));
6103 trace_vcpu_match_mmio(gva, *gpa, write, false);
6107 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6109 if (*gpa == UNMAPPED_GVA)
6112 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6115 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6116 const void *val, int bytes)
6120 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6123 kvm_page_track_write(vcpu, gpa, val, bytes);
6127 struct read_write_emulator_ops {
6128 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6130 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6131 void *val, int bytes);
6132 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6133 int bytes, void *val);
6134 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6135 void *val, int bytes);
6139 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6141 if (vcpu->mmio_read_completed) {
6142 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6143 vcpu->mmio_fragments[0].gpa, val);
6144 vcpu->mmio_read_completed = 0;
6151 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6152 void *val, int bytes)
6154 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6157 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6158 void *val, int bytes)
6160 return emulator_write_phys(vcpu, gpa, val, bytes);
6163 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6165 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6166 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6169 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6170 void *val, int bytes)
6172 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6173 return X86EMUL_IO_NEEDED;
6176 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6177 void *val, int bytes)
6179 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6181 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6182 return X86EMUL_CONTINUE;
6185 static const struct read_write_emulator_ops read_emultor = {
6186 .read_write_prepare = read_prepare,
6187 .read_write_emulate = read_emulate,
6188 .read_write_mmio = vcpu_mmio_read,
6189 .read_write_exit_mmio = read_exit_mmio,
6192 static const struct read_write_emulator_ops write_emultor = {
6193 .read_write_emulate = write_emulate,
6194 .read_write_mmio = write_mmio,
6195 .read_write_exit_mmio = write_exit_mmio,
6199 static int emulator_read_write_onepage(unsigned long addr, void *val,
6201 struct x86_exception *exception,
6202 struct kvm_vcpu *vcpu,
6203 const struct read_write_emulator_ops *ops)
6207 bool write = ops->write;
6208 struct kvm_mmio_fragment *frag;
6209 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6212 * If the exit was due to a NPF we may already have a GPA.
6213 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6214 * Note, this cannot be used on string operations since string
6215 * operation using rep will only have the initial GPA from the NPF
6218 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6219 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6220 gpa = ctxt->gpa_val;
6221 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6223 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6225 return X86EMUL_PROPAGATE_FAULT;
6228 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6229 return X86EMUL_CONTINUE;
6232 * Is this MMIO handled locally?
6234 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6235 if (handled == bytes)
6236 return X86EMUL_CONTINUE;
6242 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6243 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6247 return X86EMUL_CONTINUE;
6250 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6252 void *val, unsigned int bytes,
6253 struct x86_exception *exception,
6254 const struct read_write_emulator_ops *ops)
6256 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6260 if (ops->read_write_prepare &&
6261 ops->read_write_prepare(vcpu, val, bytes))
6262 return X86EMUL_CONTINUE;
6264 vcpu->mmio_nr_fragments = 0;
6266 /* Crossing a page boundary? */
6267 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6270 now = -addr & ~PAGE_MASK;
6271 rc = emulator_read_write_onepage(addr, val, now, exception,
6274 if (rc != X86EMUL_CONTINUE)
6277 if (ctxt->mode != X86EMUL_MODE_PROT64)
6283 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6285 if (rc != X86EMUL_CONTINUE)
6288 if (!vcpu->mmio_nr_fragments)
6291 gpa = vcpu->mmio_fragments[0].gpa;
6293 vcpu->mmio_needed = 1;
6294 vcpu->mmio_cur_fragment = 0;
6296 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6297 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6298 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6299 vcpu->run->mmio.phys_addr = gpa;
6301 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6304 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6308 struct x86_exception *exception)
6310 return emulator_read_write(ctxt, addr, val, bytes,
6311 exception, &read_emultor);
6314 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6318 struct x86_exception *exception)
6320 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6321 exception, &write_emultor);
6324 #define CMPXCHG_TYPE(t, ptr, old, new) \
6325 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6327 #ifdef CONFIG_X86_64
6328 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6330 # define CMPXCHG64(ptr, old, new) \
6331 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6334 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6339 struct x86_exception *exception)
6341 struct kvm_host_map map;
6342 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6348 /* guests cmpxchg8b have to be emulated atomically */
6349 if (bytes > 8 || (bytes & (bytes - 1)))
6352 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6354 if (gpa == UNMAPPED_GVA ||
6355 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6359 * Emulate the atomic as a straight write to avoid #AC if SLD is
6360 * enabled in the host and the access splits a cache line.
6362 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6363 page_line_mask = ~(cache_line_size() - 1);
6365 page_line_mask = PAGE_MASK;
6367 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6370 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6373 kaddr = map.hva + offset_in_page(gpa);
6377 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6380 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6383 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6386 exchanged = CMPXCHG64(kaddr, old, new);
6392 kvm_vcpu_unmap(vcpu, &map, true);
6395 return X86EMUL_CMPXCHG_FAILED;
6397 kvm_page_track_write(vcpu, gpa, new, bytes);
6399 return X86EMUL_CONTINUE;
6402 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6404 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6407 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6411 for (i = 0; i < vcpu->arch.pio.count; i++) {
6412 if (vcpu->arch.pio.in)
6413 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6414 vcpu->arch.pio.size, pd);
6416 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6417 vcpu->arch.pio.port, vcpu->arch.pio.size,
6421 pd += vcpu->arch.pio.size;
6426 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6427 unsigned short port, void *val,
6428 unsigned int count, bool in)
6430 vcpu->arch.pio.port = port;
6431 vcpu->arch.pio.in = in;
6432 vcpu->arch.pio.count = count;
6433 vcpu->arch.pio.size = size;
6435 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6436 vcpu->arch.pio.count = 0;
6440 vcpu->run->exit_reason = KVM_EXIT_IO;
6441 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6442 vcpu->run->io.size = size;
6443 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6444 vcpu->run->io.count = count;
6445 vcpu->run->io.port = port;
6450 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6451 unsigned short port, void *val, unsigned int count)
6455 if (vcpu->arch.pio.count)
6458 memset(vcpu->arch.pio_data, 0, size * count);
6460 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6463 memcpy(val, vcpu->arch.pio_data, size * count);
6464 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6465 vcpu->arch.pio.count = 0;
6472 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6473 int size, unsigned short port, void *val,
6476 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6480 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6481 unsigned short port, const void *val,
6484 memcpy(vcpu->arch.pio_data, val, size * count);
6485 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6486 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6489 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6490 int size, unsigned short port,
6491 const void *val, unsigned int count)
6493 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6496 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6498 return kvm_x86_ops.get_segment_base(vcpu, seg);
6501 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6503 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6506 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6508 if (!need_emulate_wbinvd(vcpu))
6509 return X86EMUL_CONTINUE;
6511 if (kvm_x86_ops.has_wbinvd_exit()) {
6512 int cpu = get_cpu();
6514 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6515 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6516 wbinvd_ipi, NULL, 1);
6518 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6521 return X86EMUL_CONTINUE;
6524 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6526 kvm_emulate_wbinvd_noskip(vcpu);
6527 return kvm_skip_emulated_instruction(vcpu);
6529 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6533 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6535 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6538 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6539 unsigned long *dest)
6541 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6544 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6545 unsigned long value)
6548 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6551 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6553 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6556 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6558 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6559 unsigned long value;
6563 value = kvm_read_cr0(vcpu);
6566 value = vcpu->arch.cr2;
6569 value = kvm_read_cr3(vcpu);
6572 value = kvm_read_cr4(vcpu);
6575 value = kvm_get_cr8(vcpu);
6578 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6585 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6587 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6592 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6595 vcpu->arch.cr2 = val;
6598 res = kvm_set_cr3(vcpu, val);
6601 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6604 res = kvm_set_cr8(vcpu, val);
6607 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6614 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6616 return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6619 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6621 kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6624 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6626 kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6629 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6631 kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6634 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6636 kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6639 static unsigned long emulator_get_cached_segment_base(
6640 struct x86_emulate_ctxt *ctxt, int seg)
6642 return get_segment_base(emul_to_vcpu(ctxt), seg);
6645 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6646 struct desc_struct *desc, u32 *base3,
6649 struct kvm_segment var;
6651 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6652 *selector = var.selector;
6655 memset(desc, 0, sizeof(*desc));
6663 set_desc_limit(desc, var.limit);
6664 set_desc_base(desc, (unsigned long)var.base);
6665 #ifdef CONFIG_X86_64
6667 *base3 = var.base >> 32;
6669 desc->type = var.type;
6671 desc->dpl = var.dpl;
6672 desc->p = var.present;
6673 desc->avl = var.avl;
6681 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6682 struct desc_struct *desc, u32 base3,
6685 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6686 struct kvm_segment var;
6688 var.selector = selector;
6689 var.base = get_desc_base(desc);
6690 #ifdef CONFIG_X86_64
6691 var.base |= ((u64)base3) << 32;
6693 var.limit = get_desc_limit(desc);
6695 var.limit = (var.limit << 12) | 0xfff;
6696 var.type = desc->type;
6697 var.dpl = desc->dpl;
6702 var.avl = desc->avl;
6703 var.present = desc->p;
6704 var.unusable = !var.present;
6707 kvm_set_segment(vcpu, &var, seg);
6711 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6712 u32 msr_index, u64 *pdata)
6714 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6717 r = kvm_get_msr(vcpu, msr_index, pdata);
6719 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6720 /* Bounce to user space */
6721 return X86EMUL_IO_NEEDED;
6727 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6728 u32 msr_index, u64 data)
6730 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6733 r = kvm_set_msr(vcpu, msr_index, data);
6735 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
6736 /* Bounce to user space */
6737 return X86EMUL_IO_NEEDED;
6743 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6745 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6747 return vcpu->arch.smbase;
6750 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6752 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6754 vcpu->arch.smbase = smbase;
6757 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6760 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6763 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6764 u32 pmc, u64 *pdata)
6766 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6769 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6771 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6774 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6775 struct x86_instruction_info *info,
6776 enum x86_intercept_stage stage)
6778 return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6782 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6783 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6786 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6789 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6791 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6794 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6796 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6799 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6801 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6804 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6806 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6809 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6811 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6814 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6816 kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6819 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6821 return emul_to_vcpu(ctxt)->arch.hflags;
6824 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6826 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6829 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6830 const char *smstate)
6832 return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6835 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6837 kvm_smm_changed(emul_to_vcpu(ctxt));
6840 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6842 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6845 static const struct x86_emulate_ops emulate_ops = {
6846 .read_gpr = emulator_read_gpr,
6847 .write_gpr = emulator_write_gpr,
6848 .read_std = emulator_read_std,
6849 .write_std = emulator_write_std,
6850 .read_phys = kvm_read_guest_phys_system,
6851 .fetch = kvm_fetch_guest_virt,
6852 .read_emulated = emulator_read_emulated,
6853 .write_emulated = emulator_write_emulated,
6854 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6855 .invlpg = emulator_invlpg,
6856 .pio_in_emulated = emulator_pio_in_emulated,
6857 .pio_out_emulated = emulator_pio_out_emulated,
6858 .get_segment = emulator_get_segment,
6859 .set_segment = emulator_set_segment,
6860 .get_cached_segment_base = emulator_get_cached_segment_base,
6861 .get_gdt = emulator_get_gdt,
6862 .get_idt = emulator_get_idt,
6863 .set_gdt = emulator_set_gdt,
6864 .set_idt = emulator_set_idt,
6865 .get_cr = emulator_get_cr,
6866 .set_cr = emulator_set_cr,
6867 .cpl = emulator_get_cpl,
6868 .get_dr = emulator_get_dr,
6869 .set_dr = emulator_set_dr,
6870 .get_smbase = emulator_get_smbase,
6871 .set_smbase = emulator_set_smbase,
6872 .set_msr = emulator_set_msr,
6873 .get_msr = emulator_get_msr,
6874 .check_pmc = emulator_check_pmc,
6875 .read_pmc = emulator_read_pmc,
6876 .halt = emulator_halt,
6877 .wbinvd = emulator_wbinvd,
6878 .fix_hypercall = emulator_fix_hypercall,
6879 .intercept = emulator_intercept,
6880 .get_cpuid = emulator_get_cpuid,
6881 .guest_has_long_mode = emulator_guest_has_long_mode,
6882 .guest_has_movbe = emulator_guest_has_movbe,
6883 .guest_has_fxsr = emulator_guest_has_fxsr,
6884 .set_nmi_mask = emulator_set_nmi_mask,
6885 .get_hflags = emulator_get_hflags,
6886 .set_hflags = emulator_set_hflags,
6887 .pre_leave_smm = emulator_pre_leave_smm,
6888 .post_leave_smm = emulator_post_leave_smm,
6889 .set_xcr = emulator_set_xcr,
6892 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6894 u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6896 * an sti; sti; sequence only disable interrupts for the first
6897 * instruction. So, if the last instruction, be it emulated or
6898 * not, left the system with the INT_STI flag enabled, it
6899 * means that the last instruction is an sti. We should not
6900 * leave the flag on in this case. The same goes for mov ss
6902 if (int_shadow & mask)
6904 if (unlikely(int_shadow || mask)) {
6905 kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6907 kvm_make_request(KVM_REQ_EVENT, vcpu);
6911 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6913 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6914 if (ctxt->exception.vector == PF_VECTOR)
6915 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
6917 if (ctxt->exception.error_code_valid)
6918 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6919 ctxt->exception.error_code);
6921 kvm_queue_exception(vcpu, ctxt->exception.vector);
6925 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6927 struct x86_emulate_ctxt *ctxt;
6929 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6931 pr_err("kvm: failed to allocate vcpu's emulator\n");
6936 ctxt->ops = &emulate_ops;
6937 vcpu->arch.emulate_ctxt = ctxt;
6942 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6944 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6947 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6949 ctxt->gpa_available = false;
6950 ctxt->eflags = kvm_get_rflags(vcpu);
6951 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6953 ctxt->eip = kvm_rip_read(vcpu);
6954 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6955 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6956 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6957 cs_db ? X86EMUL_MODE_PROT32 :
6958 X86EMUL_MODE_PROT16;
6959 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6960 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6961 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6963 init_decode_cache(ctxt);
6964 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6967 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6969 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6972 init_emulate_ctxt(vcpu);
6976 ctxt->_eip = ctxt->eip + inc_eip;
6977 ret = emulate_int_real(ctxt, irq);
6979 if (ret != X86EMUL_CONTINUE) {
6980 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6982 ctxt->eip = ctxt->_eip;
6983 kvm_rip_write(vcpu, ctxt->eip);
6984 kvm_set_rflags(vcpu, ctxt->eflags);
6987 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6989 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6991 ++vcpu->stat.insn_emulation_fail;
6992 trace_kvm_emulate_insn_failed(vcpu);
6994 if (emulation_type & EMULTYPE_VMWARE_GP) {
6995 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6999 if (emulation_type & EMULTYPE_SKIP) {
7000 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7001 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7002 vcpu->run->internal.ndata = 0;
7006 kvm_queue_exception(vcpu, UD_VECTOR);
7008 if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
7009 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7010 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7011 vcpu->run->internal.ndata = 0;
7018 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7019 bool write_fault_to_shadow_pgtable,
7022 gpa_t gpa = cr2_or_gpa;
7025 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7028 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7029 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7032 if (!vcpu->arch.mmu->direct_map) {
7034 * Write permission should be allowed since only
7035 * write access need to be emulated.
7037 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7040 * If the mapping is invalid in guest, let cpu retry
7041 * it to generate fault.
7043 if (gpa == UNMAPPED_GVA)
7048 * Do not retry the unhandleable instruction if it faults on the
7049 * readonly host memory, otherwise it will goto a infinite loop:
7050 * retry instruction -> write #PF -> emulation fail -> retry
7051 * instruction -> ...
7053 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7056 * If the instruction failed on the error pfn, it can not be fixed,
7057 * report the error to userspace.
7059 if (is_error_noslot_pfn(pfn))
7062 kvm_release_pfn_clean(pfn);
7064 /* The instructions are well-emulated on direct mmu. */
7065 if (vcpu->arch.mmu->direct_map) {
7066 unsigned int indirect_shadow_pages;
7068 spin_lock(&vcpu->kvm->mmu_lock);
7069 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7070 spin_unlock(&vcpu->kvm->mmu_lock);
7072 if (indirect_shadow_pages)
7073 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7079 * if emulation was due to access to shadowed page table
7080 * and it failed try to unshadow page and re-enter the
7081 * guest to let CPU execute the instruction.
7083 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7086 * If the access faults on its page table, it can not
7087 * be fixed by unprotecting shadow page and it should
7088 * be reported to userspace.
7090 return !write_fault_to_shadow_pgtable;
7093 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7094 gpa_t cr2_or_gpa, int emulation_type)
7096 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7097 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7099 last_retry_eip = vcpu->arch.last_retry_eip;
7100 last_retry_addr = vcpu->arch.last_retry_addr;
7103 * If the emulation is caused by #PF and it is non-page_table
7104 * writing instruction, it means the VM-EXIT is caused by shadow
7105 * page protected, we can zap the shadow page and retry this
7106 * instruction directly.
7108 * Note: if the guest uses a non-page-table modifying instruction
7109 * on the PDE that points to the instruction, then we will unmap
7110 * the instruction and go to an infinite loop. So, we cache the
7111 * last retried eip and the last fault address, if we meet the eip
7112 * and the address again, we can break out of the potential infinite
7115 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7117 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7120 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7121 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7124 if (x86_page_table_writing_insn(ctxt))
7127 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7130 vcpu->arch.last_retry_eip = ctxt->eip;
7131 vcpu->arch.last_retry_addr = cr2_or_gpa;
7133 if (!vcpu->arch.mmu->direct_map)
7134 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7136 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7141 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7142 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7144 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7146 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7147 /* This is a good place to trace that we are exiting SMM. */
7148 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7150 /* Process a latched INIT or SMI, if any. */
7151 kvm_make_request(KVM_REQ_EVENT, vcpu);
7154 kvm_mmu_reset_context(vcpu);
7157 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7166 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7167 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7172 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7174 struct kvm_run *kvm_run = vcpu->run;
7176 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7177 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
7178 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7179 kvm_run->debug.arch.exception = DB_VECTOR;
7180 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7183 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7187 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7189 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7192 r = kvm_x86_ops.skip_emulated_instruction(vcpu);
7197 * rflags is the old, "raw" value of the flags. The new value has
7198 * not been saved yet.
7200 * This is correct even for TF set by the guest, because "the
7201 * processor will not generate this exception after the instruction
7202 * that sets the TF flag".
7204 if (unlikely(rflags & X86_EFLAGS_TF))
7205 r = kvm_vcpu_do_singlestep(vcpu);
7208 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7210 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7212 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7213 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7214 struct kvm_run *kvm_run = vcpu->run;
7215 unsigned long eip = kvm_get_linear_rip(vcpu);
7216 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7217 vcpu->arch.guest_debug_dr7,
7221 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
7222 kvm_run->debug.arch.pc = eip;
7223 kvm_run->debug.arch.exception = DB_VECTOR;
7224 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7230 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7231 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7232 unsigned long eip = kvm_get_linear_rip(vcpu);
7233 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7238 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7247 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7249 switch (ctxt->opcode_len) {
7256 case 0xe6: /* OUT */
7260 case 0x6c: /* INS */
7262 case 0x6e: /* OUTS */
7269 case 0x33: /* RDPMC */
7278 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7279 int emulation_type, void *insn, int insn_len)
7282 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7283 bool writeback = true;
7284 bool write_fault_to_spt;
7286 if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, insn, insn_len)))
7289 vcpu->arch.l1tf_flush_l1d = true;
7292 * Clear write_fault_to_shadow_pgtable here to ensure it is
7295 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7296 vcpu->arch.write_fault_to_shadow_pgtable = false;
7297 kvm_clear_exception_queue(vcpu);
7299 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7300 init_emulate_ctxt(vcpu);
7303 * We will reenter on the same instruction since
7304 * we do not set complete_userspace_io. This does not
7305 * handle watchpoints yet, those would be handled in
7308 if (!(emulation_type & EMULTYPE_SKIP) &&
7309 kvm_vcpu_check_breakpoint(vcpu, &r))
7312 ctxt->interruptibility = 0;
7313 ctxt->have_exception = false;
7314 ctxt->exception.vector = -1;
7315 ctxt->perm_ok = false;
7317 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7319 r = x86_decode_insn(ctxt, insn, insn_len);
7321 trace_kvm_emulate_insn_start(vcpu);
7322 ++vcpu->stat.insn_emulation;
7323 if (r != EMULATION_OK) {
7324 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7325 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7326 kvm_queue_exception(vcpu, UD_VECTOR);
7329 if (reexecute_instruction(vcpu, cr2_or_gpa,
7333 if (ctxt->have_exception) {
7335 * #UD should result in just EMULATION_FAILED, and trap-like
7336 * exception should not be encountered during decode.
7338 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7339 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7340 inject_emulated_exception(vcpu);
7343 return handle_emulation_failure(vcpu, emulation_type);
7347 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7348 !is_vmware_backdoor_opcode(ctxt)) {
7349 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7354 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7355 * for kvm_skip_emulated_instruction(). The caller is responsible for
7356 * updating interruptibility state and injecting single-step #DBs.
7358 if (emulation_type & EMULTYPE_SKIP) {
7359 kvm_rip_write(vcpu, ctxt->_eip);
7360 if (ctxt->eflags & X86_EFLAGS_RF)
7361 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7365 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7368 /* this is needed for vmware backdoor interface to work since it
7369 changes registers values during IO operation */
7370 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7371 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7372 emulator_invalidate_register_cache(ctxt);
7376 if (emulation_type & EMULTYPE_PF) {
7377 /* Save the faulting GPA (cr2) in the address field */
7378 ctxt->exception.address = cr2_or_gpa;
7380 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7381 if (vcpu->arch.mmu->direct_map) {
7382 ctxt->gpa_available = true;
7383 ctxt->gpa_val = cr2_or_gpa;
7386 /* Sanitize the address out of an abundance of paranoia. */
7387 ctxt->exception.address = 0;
7390 r = x86_emulate_insn(ctxt);
7392 if (r == EMULATION_INTERCEPTED)
7395 if (r == EMULATION_FAILED) {
7396 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7400 return handle_emulation_failure(vcpu, emulation_type);
7403 if (ctxt->have_exception) {
7405 if (inject_emulated_exception(vcpu))
7407 } else if (vcpu->arch.pio.count) {
7408 if (!vcpu->arch.pio.in) {
7409 /* FIXME: return into emulator if single-stepping. */
7410 vcpu->arch.pio.count = 0;
7413 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7416 } else if (vcpu->mmio_needed) {
7417 ++vcpu->stat.mmio_exits;
7419 if (!vcpu->mmio_is_write)
7422 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7423 } else if (r == EMULATION_RESTART)
7429 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7430 toggle_interruptibility(vcpu, ctxt->interruptibility);
7431 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7432 if (!ctxt->have_exception ||
7433 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7434 kvm_rip_write(vcpu, ctxt->eip);
7435 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7436 r = kvm_vcpu_do_singlestep(vcpu);
7437 if (kvm_x86_ops.update_emulated_instruction)
7438 kvm_x86_ops.update_emulated_instruction(vcpu);
7439 __kvm_set_rflags(vcpu, ctxt->eflags);
7443 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7444 * do nothing, and it will be requested again as soon as
7445 * the shadow expires. But we still need to check here,
7446 * because POPF has no interrupt shadow.
7448 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7449 kvm_make_request(KVM_REQ_EVENT, vcpu);
7451 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7456 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7458 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7460 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7462 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7463 void *insn, int insn_len)
7465 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7467 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7469 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7471 vcpu->arch.pio.count = 0;
7475 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7477 vcpu->arch.pio.count = 0;
7479 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7482 return kvm_skip_emulated_instruction(vcpu);
7485 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7486 unsigned short port)
7488 unsigned long val = kvm_rax_read(vcpu);
7489 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7495 * Workaround userspace that relies on old KVM behavior of %rip being
7496 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7499 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7500 vcpu->arch.complete_userspace_io =
7501 complete_fast_pio_out_port_0x7e;
7502 kvm_skip_emulated_instruction(vcpu);
7504 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7505 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7510 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7514 /* We should only ever be called with arch.pio.count equal to 1 */
7515 BUG_ON(vcpu->arch.pio.count != 1);
7517 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7518 vcpu->arch.pio.count = 0;
7522 /* For size less than 4 we merge, else we zero extend */
7523 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7526 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7527 * the copy and tracing
7529 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7530 kvm_rax_write(vcpu, val);
7532 return kvm_skip_emulated_instruction(vcpu);
7535 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7536 unsigned short port)
7541 /* For size less than 4 we merge, else we zero extend */
7542 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7544 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7546 kvm_rax_write(vcpu, val);
7550 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7551 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7556 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7561 ret = kvm_fast_pio_in(vcpu, size, port);
7563 ret = kvm_fast_pio_out(vcpu, size, port);
7564 return ret && kvm_skip_emulated_instruction(vcpu);
7566 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7568 static int kvmclock_cpu_down_prep(unsigned int cpu)
7570 __this_cpu_write(cpu_tsc_khz, 0);
7574 static void tsc_khz_changed(void *data)
7576 struct cpufreq_freqs *freq = data;
7577 unsigned long khz = 0;
7581 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7582 khz = cpufreq_quick_get(raw_smp_processor_id());
7585 __this_cpu_write(cpu_tsc_khz, khz);
7588 #ifdef CONFIG_X86_64
7589 static void kvm_hyperv_tsc_notifier(void)
7592 struct kvm_vcpu *vcpu;
7595 mutex_lock(&kvm_lock);
7596 list_for_each_entry(kvm, &vm_list, vm_list)
7597 kvm_make_mclock_inprogress_request(kvm);
7599 hyperv_stop_tsc_emulation();
7601 /* TSC frequency always matches when on Hyper-V */
7602 for_each_present_cpu(cpu)
7603 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7604 kvm_max_guest_tsc_khz = tsc_khz;
7606 list_for_each_entry(kvm, &vm_list, vm_list) {
7607 struct kvm_arch *ka = &kvm->arch;
7609 spin_lock(&ka->pvclock_gtod_sync_lock);
7611 pvclock_update_vm_gtod_copy(kvm);
7613 kvm_for_each_vcpu(cpu, vcpu, kvm)
7614 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7616 kvm_for_each_vcpu(cpu, vcpu, kvm)
7617 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7619 spin_unlock(&ka->pvclock_gtod_sync_lock);
7621 mutex_unlock(&kvm_lock);
7625 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7628 struct kvm_vcpu *vcpu;
7629 int i, send_ipi = 0;
7632 * We allow guests to temporarily run on slowing clocks,
7633 * provided we notify them after, or to run on accelerating
7634 * clocks, provided we notify them before. Thus time never
7637 * However, we have a problem. We can't atomically update
7638 * the frequency of a given CPU from this function; it is
7639 * merely a notifier, which can be called from any CPU.
7640 * Changing the TSC frequency at arbitrary points in time
7641 * requires a recomputation of local variables related to
7642 * the TSC for each VCPU. We must flag these local variables
7643 * to be updated and be sure the update takes place with the
7644 * new frequency before any guests proceed.
7646 * Unfortunately, the combination of hotplug CPU and frequency
7647 * change creates an intractable locking scenario; the order
7648 * of when these callouts happen is undefined with respect to
7649 * CPU hotplug, and they can race with each other. As such,
7650 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7651 * undefined; you can actually have a CPU frequency change take
7652 * place in between the computation of X and the setting of the
7653 * variable. To protect against this problem, all updates of
7654 * the per_cpu tsc_khz variable are done in an interrupt
7655 * protected IPI, and all callers wishing to update the value
7656 * must wait for a synchronous IPI to complete (which is trivial
7657 * if the caller is on the CPU already). This establishes the
7658 * necessary total order on variable updates.
7660 * Note that because a guest time update may take place
7661 * anytime after the setting of the VCPU's request bit, the
7662 * correct TSC value must be set before the request. However,
7663 * to ensure the update actually makes it to any guest which
7664 * starts running in hardware virtualization between the set
7665 * and the acquisition of the spinlock, we must also ping the
7666 * CPU after setting the request bit.
7670 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7672 mutex_lock(&kvm_lock);
7673 list_for_each_entry(kvm, &vm_list, vm_list) {
7674 kvm_for_each_vcpu(i, vcpu, kvm) {
7675 if (vcpu->cpu != cpu)
7677 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7678 if (vcpu->cpu != raw_smp_processor_id())
7682 mutex_unlock(&kvm_lock);
7684 if (freq->old < freq->new && send_ipi) {
7686 * We upscale the frequency. Must make the guest
7687 * doesn't see old kvmclock values while running with
7688 * the new frequency, otherwise we risk the guest sees
7689 * time go backwards.
7691 * In case we update the frequency for another cpu
7692 * (which might be in guest context) send an interrupt
7693 * to kick the cpu out of guest context. Next time
7694 * guest context is entered kvmclock will be updated,
7695 * so the guest will not see stale values.
7697 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7701 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7704 struct cpufreq_freqs *freq = data;
7707 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7709 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7712 for_each_cpu(cpu, freq->policy->cpus)
7713 __kvmclock_cpufreq_notifier(freq, cpu);
7718 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7719 .notifier_call = kvmclock_cpufreq_notifier
7722 static int kvmclock_cpu_online(unsigned int cpu)
7724 tsc_khz_changed(NULL);
7728 static void kvm_timer_init(void)
7730 max_tsc_khz = tsc_khz;
7732 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7733 #ifdef CONFIG_CPU_FREQ
7734 struct cpufreq_policy *policy;
7738 policy = cpufreq_cpu_get(cpu);
7740 if (policy->cpuinfo.max_freq)
7741 max_tsc_khz = policy->cpuinfo.max_freq;
7742 cpufreq_cpu_put(policy);
7746 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7747 CPUFREQ_TRANSITION_NOTIFIER);
7750 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7751 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7754 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7755 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7757 int kvm_is_in_guest(void)
7759 return __this_cpu_read(current_vcpu) != NULL;
7762 static int kvm_is_user_mode(void)
7766 if (__this_cpu_read(current_vcpu))
7767 user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7769 return user_mode != 0;
7772 static unsigned long kvm_get_guest_ip(void)
7774 unsigned long ip = 0;
7776 if (__this_cpu_read(current_vcpu))
7777 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7782 static void kvm_handle_intel_pt_intr(void)
7784 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7786 kvm_make_request(KVM_REQ_PMI, vcpu);
7787 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7788 (unsigned long *)&vcpu->arch.pmu.global_status);
7791 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7792 .is_in_guest = kvm_is_in_guest,
7793 .is_user_mode = kvm_is_user_mode,
7794 .get_guest_ip = kvm_get_guest_ip,
7795 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7798 #ifdef CONFIG_X86_64
7799 static void pvclock_gtod_update_fn(struct work_struct *work)
7803 struct kvm_vcpu *vcpu;
7806 mutex_lock(&kvm_lock);
7807 list_for_each_entry(kvm, &vm_list, vm_list)
7808 kvm_for_each_vcpu(i, vcpu, kvm)
7809 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7810 atomic_set(&kvm_guest_has_master_clock, 0);
7811 mutex_unlock(&kvm_lock);
7814 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7817 * Notification about pvclock gtod data update.
7819 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7822 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7823 struct timekeeper *tk = priv;
7825 update_pvclock_gtod(tk);
7827 /* disable master clock if host does not trust, or does not
7828 * use, TSC based clocksource.
7830 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7831 atomic_read(&kvm_guest_has_master_clock) != 0)
7832 queue_work(system_long_wq, &pvclock_gtod_work);
7837 static struct notifier_block pvclock_gtod_notifier = {
7838 .notifier_call = pvclock_gtod_notify,
7842 int kvm_arch_init(void *opaque)
7844 struct kvm_x86_init_ops *ops = opaque;
7847 if (kvm_x86_ops.hardware_enable) {
7848 printk(KERN_ERR "kvm: already loaded the other module\n");
7853 if (!ops->cpu_has_kvm_support()) {
7854 pr_err_ratelimited("kvm: no hardware support\n");
7858 if (ops->disabled_by_bios()) {
7859 pr_err_ratelimited("kvm: disabled by bios\n");
7865 * KVM explicitly assumes that the guest has an FPU and
7866 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7867 * vCPU's FPU state as a fxregs_state struct.
7869 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7870 printk(KERN_ERR "kvm: inadequate fpu\n");
7876 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7877 __alignof__(struct fpu), SLAB_ACCOUNT,
7879 if (!x86_fpu_cache) {
7880 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7884 x86_emulator_cache = kvm_alloc_emulator_cache();
7885 if (!x86_emulator_cache) {
7886 pr_err("kvm: failed to allocate cache for x86 emulator\n");
7887 goto out_free_x86_fpu_cache;
7890 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
7891 if (!user_return_msrs) {
7892 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
7893 goto out_free_x86_emulator_cache;
7896 r = kvm_mmu_module_init();
7898 goto out_free_percpu;
7900 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7901 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7902 PT_PRESENT_MASK, 0, sme_me_mask);
7905 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7907 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7908 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7909 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7913 if (pi_inject_timer == -1)
7914 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7915 #ifdef CONFIG_X86_64
7916 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7918 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7919 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7925 free_percpu(user_return_msrs);
7926 out_free_x86_emulator_cache:
7927 kmem_cache_destroy(x86_emulator_cache);
7928 out_free_x86_fpu_cache:
7929 kmem_cache_destroy(x86_fpu_cache);
7934 void kvm_arch_exit(void)
7936 #ifdef CONFIG_X86_64
7937 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7938 clear_hv_tscchange_cb();
7941 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7943 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7944 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7945 CPUFREQ_TRANSITION_NOTIFIER);
7946 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7947 #ifdef CONFIG_X86_64
7948 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7950 kvm_x86_ops.hardware_enable = NULL;
7951 kvm_mmu_module_exit();
7952 free_percpu(user_return_msrs);
7953 kmem_cache_destroy(x86_fpu_cache);
7956 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7958 ++vcpu->stat.halt_exits;
7959 if (lapic_in_kernel(vcpu)) {
7960 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7963 vcpu->run->exit_reason = KVM_EXIT_HLT;
7967 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7969 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7971 int ret = kvm_skip_emulated_instruction(vcpu);
7973 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7974 * KVM_EXIT_DEBUG here.
7976 return kvm_vcpu_halt(vcpu) && ret;
7978 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7980 #ifdef CONFIG_X86_64
7981 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7982 unsigned long clock_type)
7984 struct kvm_clock_pairing clock_pairing;
7985 struct timespec64 ts;
7989 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7990 return -KVM_EOPNOTSUPP;
7992 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7993 return -KVM_EOPNOTSUPP;
7995 clock_pairing.sec = ts.tv_sec;
7996 clock_pairing.nsec = ts.tv_nsec;
7997 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7998 clock_pairing.flags = 0;
7999 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8002 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8003 sizeof(struct kvm_clock_pairing)))
8011 * kvm_pv_kick_cpu_op: Kick a vcpu.
8013 * @apicid - apicid of vcpu to be kicked.
8015 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8017 struct kvm_lapic_irq lapic_irq;
8019 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8020 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8021 lapic_irq.level = 0;
8022 lapic_irq.dest_id = apicid;
8023 lapic_irq.msi_redir_hint = false;
8025 lapic_irq.delivery_mode = APIC_DM_REMRD;
8026 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8029 bool kvm_apicv_activated(struct kvm *kvm)
8031 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8033 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8035 void kvm_apicv_init(struct kvm *kvm, bool enable)
8038 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8039 &kvm->arch.apicv_inhibit_reasons);
8041 set_bit(APICV_INHIBIT_REASON_DISABLE,
8042 &kvm->arch.apicv_inhibit_reasons);
8044 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8046 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
8048 struct kvm_vcpu *target = NULL;
8049 struct kvm_apic_map *map;
8052 map = rcu_dereference(kvm->arch.apic_map);
8054 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8055 target = map->phys_map[dest_id]->vcpu;
8059 if (target && READ_ONCE(target->ready))
8060 kvm_vcpu_yield_to(target);
8063 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8065 unsigned long nr, a0, a1, a2, a3, ret;
8068 if (kvm_hv_hypercall_enabled(vcpu->kvm))
8069 return kvm_hv_hypercall(vcpu);
8071 nr = kvm_rax_read(vcpu);
8072 a0 = kvm_rbx_read(vcpu);
8073 a1 = kvm_rcx_read(vcpu);
8074 a2 = kvm_rdx_read(vcpu);
8075 a3 = kvm_rsi_read(vcpu);
8077 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8079 op_64_bit = is_64_bit_mode(vcpu);
8088 if (kvm_x86_ops.get_cpl(vcpu) != 0) {
8096 case KVM_HC_VAPIC_POLL_IRQ:
8099 case KVM_HC_KICK_CPU:
8100 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8103 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8104 kvm_sched_yield(vcpu->kvm, a1);
8107 #ifdef CONFIG_X86_64
8108 case KVM_HC_CLOCK_PAIRING:
8109 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8112 case KVM_HC_SEND_IPI:
8113 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8116 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8118 case KVM_HC_SCHED_YIELD:
8119 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8122 kvm_sched_yield(vcpu->kvm, a0);
8132 kvm_rax_write(vcpu, ret);
8134 ++vcpu->stat.hypercalls;
8135 return kvm_skip_emulated_instruction(vcpu);
8137 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8139 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8141 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8142 char instruction[3];
8143 unsigned long rip = kvm_rip_read(vcpu);
8145 kvm_x86_ops.patch_hypercall(vcpu, instruction);
8147 return emulator_write_emulated(ctxt, rip, instruction, 3,
8151 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8153 return vcpu->run->request_interrupt_window &&
8154 likely(!pic_in_kernel(vcpu->kvm));
8157 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8159 struct kvm_run *kvm_run = vcpu->run;
8161 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8162 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
8163 kvm_run->cr8 = kvm_get_cr8(vcpu);
8164 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8165 kvm_run->ready_for_interrupt_injection =
8166 pic_in_kernel(vcpu->kvm) ||
8167 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8170 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8174 if (!kvm_x86_ops.update_cr8_intercept)
8177 if (!lapic_in_kernel(vcpu))
8180 if (vcpu->arch.apicv_active)
8183 if (!vcpu->arch.apic->vapic_addr)
8184 max_irr = kvm_lapic_find_highest_irr(vcpu);
8191 tpr = kvm_lapic_get_cr8(vcpu);
8193 kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
8196 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8199 bool can_inject = true;
8201 /* try to reinject previous events if any */
8203 if (vcpu->arch.exception.injected) {
8204 kvm_x86_ops.queue_exception(vcpu);
8208 * Do not inject an NMI or interrupt if there is a pending
8209 * exception. Exceptions and interrupts are recognized at
8210 * instruction boundaries, i.e. the start of an instruction.
8211 * Trap-like exceptions, e.g. #DB, have higher priority than
8212 * NMIs and interrupts, i.e. traps are recognized before an
8213 * NMI/interrupt that's pending on the same instruction.
8214 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8215 * priority, but are only generated (pended) during instruction
8216 * execution, i.e. a pending fault-like exception means the
8217 * fault occurred on the *previous* instruction and must be
8218 * serviced prior to recognizing any new events in order to
8219 * fully complete the previous instruction.
8221 else if (!vcpu->arch.exception.pending) {
8222 if (vcpu->arch.nmi_injected) {
8223 kvm_x86_ops.set_nmi(vcpu);
8225 } else if (vcpu->arch.interrupt.injected) {
8226 kvm_x86_ops.set_irq(vcpu);
8231 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8232 vcpu->arch.exception.pending);
8235 * Call check_nested_events() even if we reinjected a previous event
8236 * in order for caller to determine if it should require immediate-exit
8237 * from L2 to L1 due to pending L1 events which require exit
8240 if (is_guest_mode(vcpu)) {
8241 r = kvm_x86_ops.nested_ops->check_events(vcpu);
8246 /* try to inject new event if pending */
8247 if (vcpu->arch.exception.pending) {
8248 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8249 vcpu->arch.exception.has_error_code,
8250 vcpu->arch.exception.error_code);
8252 vcpu->arch.exception.pending = false;
8253 vcpu->arch.exception.injected = true;
8255 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8256 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8259 if (vcpu->arch.exception.nr == DB_VECTOR) {
8260 kvm_deliver_exception_payload(vcpu);
8261 if (vcpu->arch.dr7 & DR7_GD) {
8262 vcpu->arch.dr7 &= ~DR7_GD;
8263 kvm_update_dr7(vcpu);
8267 kvm_x86_ops.queue_exception(vcpu);
8272 * Finally, inject interrupt events. If an event cannot be injected
8273 * due to architectural conditions (e.g. IF=0) a window-open exit
8274 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8275 * and can architecturally be injected, but we cannot do it right now:
8276 * an interrupt could have arrived just now and we have to inject it
8277 * as a vmexit, or there could already an event in the queue, which is
8278 * indicated by can_inject. In that case we request an immediate exit
8279 * in order to make progress and get back here for another iteration.
8280 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8282 if (vcpu->arch.smi_pending) {
8283 r = can_inject ? kvm_x86_ops.smi_allowed(vcpu, true) : -EBUSY;
8287 vcpu->arch.smi_pending = false;
8288 ++vcpu->arch.smi_count;
8292 kvm_x86_ops.enable_smi_window(vcpu);
8295 if (vcpu->arch.nmi_pending) {
8296 r = can_inject ? kvm_x86_ops.nmi_allowed(vcpu, true) : -EBUSY;
8300 --vcpu->arch.nmi_pending;
8301 vcpu->arch.nmi_injected = true;
8302 kvm_x86_ops.set_nmi(vcpu);
8304 WARN_ON(kvm_x86_ops.nmi_allowed(vcpu, true) < 0);
8306 if (vcpu->arch.nmi_pending)
8307 kvm_x86_ops.enable_nmi_window(vcpu);
8310 if (kvm_cpu_has_injectable_intr(vcpu)) {
8311 r = can_inject ? kvm_x86_ops.interrupt_allowed(vcpu, true) : -EBUSY;
8315 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8316 kvm_x86_ops.set_irq(vcpu);
8317 WARN_ON(kvm_x86_ops.interrupt_allowed(vcpu, true) < 0);
8319 if (kvm_cpu_has_injectable_intr(vcpu))
8320 kvm_x86_ops.enable_irq_window(vcpu);
8323 if (is_guest_mode(vcpu) &&
8324 kvm_x86_ops.nested_ops->hv_timer_pending &&
8325 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8326 *req_immediate_exit = true;
8328 WARN_ON(vcpu->arch.exception.pending);
8332 *req_immediate_exit = true;
8336 static void process_nmi(struct kvm_vcpu *vcpu)
8341 * x86 is limited to one NMI running, and one NMI pending after it.
8342 * If an NMI is already in progress, limit further NMIs to just one.
8343 * Otherwise, allow two (and we'll inject the first one immediately).
8345 if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
8348 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8349 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8350 kvm_make_request(KVM_REQ_EVENT, vcpu);
8353 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8356 flags |= seg->g << 23;
8357 flags |= seg->db << 22;
8358 flags |= seg->l << 21;
8359 flags |= seg->avl << 20;
8360 flags |= seg->present << 15;
8361 flags |= seg->dpl << 13;
8362 flags |= seg->s << 12;
8363 flags |= seg->type << 8;
8367 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8369 struct kvm_segment seg;
8372 kvm_get_segment(vcpu, &seg, n);
8373 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8376 offset = 0x7f84 + n * 12;
8378 offset = 0x7f2c + (n - 3) * 12;
8380 put_smstate(u32, buf, offset + 8, seg.base);
8381 put_smstate(u32, buf, offset + 4, seg.limit);
8382 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8385 #ifdef CONFIG_X86_64
8386 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8388 struct kvm_segment seg;
8392 kvm_get_segment(vcpu, &seg, n);
8393 offset = 0x7e00 + n * 16;
8395 flags = enter_smm_get_segment_flags(&seg) >> 8;
8396 put_smstate(u16, buf, offset, seg.selector);
8397 put_smstate(u16, buf, offset + 2, flags);
8398 put_smstate(u32, buf, offset + 4, seg.limit);
8399 put_smstate(u64, buf, offset + 8, seg.base);
8403 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8406 struct kvm_segment seg;
8410 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8411 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8412 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8413 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8415 for (i = 0; i < 8; i++)
8416 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
8418 kvm_get_dr(vcpu, 6, &val);
8419 put_smstate(u32, buf, 0x7fcc, (u32)val);
8420 kvm_get_dr(vcpu, 7, &val);
8421 put_smstate(u32, buf, 0x7fc8, (u32)val);
8423 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8424 put_smstate(u32, buf, 0x7fc4, seg.selector);
8425 put_smstate(u32, buf, 0x7f64, seg.base);
8426 put_smstate(u32, buf, 0x7f60, seg.limit);
8427 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8429 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8430 put_smstate(u32, buf, 0x7fc0, seg.selector);
8431 put_smstate(u32, buf, 0x7f80, seg.base);
8432 put_smstate(u32, buf, 0x7f7c, seg.limit);
8433 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8435 kvm_x86_ops.get_gdt(vcpu, &dt);
8436 put_smstate(u32, buf, 0x7f74, dt.address);
8437 put_smstate(u32, buf, 0x7f70, dt.size);
8439 kvm_x86_ops.get_idt(vcpu, &dt);
8440 put_smstate(u32, buf, 0x7f58, dt.address);
8441 put_smstate(u32, buf, 0x7f54, dt.size);
8443 for (i = 0; i < 6; i++)
8444 enter_smm_save_seg_32(vcpu, buf, i);
8446 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8449 put_smstate(u32, buf, 0x7efc, 0x00020000);
8450 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8453 #ifdef CONFIG_X86_64
8454 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8457 struct kvm_segment seg;
8461 for (i = 0; i < 16; i++)
8462 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
8464 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8465 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8467 kvm_get_dr(vcpu, 6, &val);
8468 put_smstate(u64, buf, 0x7f68, val);
8469 kvm_get_dr(vcpu, 7, &val);
8470 put_smstate(u64, buf, 0x7f60, val);
8472 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8473 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8474 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8476 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8479 put_smstate(u32, buf, 0x7efc, 0x00020064);
8481 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8483 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8484 put_smstate(u16, buf, 0x7e90, seg.selector);
8485 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8486 put_smstate(u32, buf, 0x7e94, seg.limit);
8487 put_smstate(u64, buf, 0x7e98, seg.base);
8489 kvm_x86_ops.get_idt(vcpu, &dt);
8490 put_smstate(u32, buf, 0x7e84, dt.size);
8491 put_smstate(u64, buf, 0x7e88, dt.address);
8493 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8494 put_smstate(u16, buf, 0x7e70, seg.selector);
8495 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8496 put_smstate(u32, buf, 0x7e74, seg.limit);
8497 put_smstate(u64, buf, 0x7e78, seg.base);
8499 kvm_x86_ops.get_gdt(vcpu, &dt);
8500 put_smstate(u32, buf, 0x7e64, dt.size);
8501 put_smstate(u64, buf, 0x7e68, dt.address);
8503 for (i = 0; i < 6; i++)
8504 enter_smm_save_seg_64(vcpu, buf, i);
8508 static void enter_smm(struct kvm_vcpu *vcpu)
8510 struct kvm_segment cs, ds;
8515 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8516 memset(buf, 0, 512);
8517 #ifdef CONFIG_X86_64
8518 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8519 enter_smm_save_state_64(vcpu, buf);
8522 enter_smm_save_state_32(vcpu, buf);
8525 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8526 * vCPU state (e.g. leave guest mode) after we've saved the state into
8527 * the SMM state-save area.
8529 kvm_x86_ops.pre_enter_smm(vcpu, buf);
8531 vcpu->arch.hflags |= HF_SMM_MASK;
8532 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8534 if (kvm_x86_ops.get_nmi_mask(vcpu))
8535 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8537 kvm_x86_ops.set_nmi_mask(vcpu, true);
8539 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8540 kvm_rip_write(vcpu, 0x8000);
8542 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8543 kvm_x86_ops.set_cr0(vcpu, cr0);
8544 vcpu->arch.cr0 = cr0;
8546 kvm_x86_ops.set_cr4(vcpu, 0);
8548 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8549 dt.address = dt.size = 0;
8550 kvm_x86_ops.set_idt(vcpu, &dt);
8552 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8554 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8555 cs.base = vcpu->arch.smbase;
8560 cs.limit = ds.limit = 0xffffffff;
8561 cs.type = ds.type = 0x3;
8562 cs.dpl = ds.dpl = 0;
8567 cs.avl = ds.avl = 0;
8568 cs.present = ds.present = 1;
8569 cs.unusable = ds.unusable = 0;
8570 cs.padding = ds.padding = 0;
8572 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8573 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8574 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8575 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8576 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8577 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8579 #ifdef CONFIG_X86_64
8580 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8581 kvm_x86_ops.set_efer(vcpu, 0);
8584 kvm_update_cpuid_runtime(vcpu);
8585 kvm_mmu_reset_context(vcpu);
8588 static void process_smi(struct kvm_vcpu *vcpu)
8590 vcpu->arch.smi_pending = true;
8591 kvm_make_request(KVM_REQ_EVENT, vcpu);
8594 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8595 unsigned long *vcpu_bitmap)
8599 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8601 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8602 NULL, vcpu_bitmap, cpus);
8604 free_cpumask_var(cpus);
8607 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8609 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8612 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8614 if (!lapic_in_kernel(vcpu))
8617 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8618 kvm_apic_update_apicv(vcpu);
8619 kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8621 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8624 * NOTE: Do not hold any lock prior to calling this.
8626 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8627 * locked, because it calls __x86_set_memory_region() which does
8628 * synchronize_srcu(&kvm->srcu).
8630 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8632 struct kvm_vcpu *except;
8633 unsigned long old, new, expected;
8635 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8636 !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8639 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8641 expected = new = old;
8643 __clear_bit(bit, &new);
8645 __set_bit(bit, &new);
8648 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8649 } while (old != expected);
8654 trace_kvm_apicv_update_request(activate, bit);
8655 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8656 kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8659 * Sending request to update APICV for all other vcpus,
8660 * while update the calling vcpu immediately instead of
8661 * waiting for another #VMEXIT to handle the request.
8663 except = kvm_get_running_vcpu();
8664 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8667 kvm_vcpu_update_apicv(except);
8669 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8671 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8673 if (!kvm_apic_present(vcpu))
8676 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8678 if (irqchip_split(vcpu->kvm))
8679 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8681 if (vcpu->arch.apicv_active)
8682 kvm_x86_ops.sync_pir_to_irr(vcpu);
8683 if (ioapic_in_kernel(vcpu->kvm))
8684 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8687 if (is_guest_mode(vcpu))
8688 vcpu->arch.load_eoi_exitmap_pending = true;
8690 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8693 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8695 u64 eoi_exit_bitmap[4];
8697 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8700 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8701 vcpu_to_synic(vcpu)->vec_bitmap, 256);
8702 kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8705 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8706 unsigned long start, unsigned long end)
8708 unsigned long apic_address;
8711 * The physical address of apic access page is stored in the VMCS.
8712 * Update it when it becomes invalid.
8714 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8715 if (start <= apic_address && apic_address < end)
8716 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8719 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8721 if (!lapic_in_kernel(vcpu))
8724 if (!kvm_x86_ops.set_apic_access_page_addr)
8727 kvm_x86_ops.set_apic_access_page_addr(vcpu);
8730 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8732 smp_send_reschedule(vcpu->cpu);
8734 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8737 * Returns 1 to let vcpu_run() continue the guest execution loop without
8738 * exiting to the userspace. Otherwise, the value will be returned to the
8741 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8745 dm_request_for_irq_injection(vcpu) &&
8746 kvm_cpu_accept_dm_intr(vcpu);
8747 fastpath_t exit_fastpath;
8749 bool req_immediate_exit = false;
8751 if (kvm_request_pending(vcpu)) {
8752 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
8753 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
8758 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8759 kvm_mmu_unload(vcpu);
8760 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8761 __kvm_migrate_timers(vcpu);
8762 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8763 kvm_gen_update_masterclock(vcpu->kvm);
8764 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8765 kvm_gen_kvmclock_update(vcpu);
8766 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8767 r = kvm_guest_time_update(vcpu);
8771 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8772 kvm_mmu_sync_roots(vcpu);
8773 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8774 kvm_mmu_load_pgd(vcpu);
8775 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
8776 kvm_vcpu_flush_tlb_all(vcpu);
8778 /* Flushing all ASIDs flushes the current ASID... */
8779 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
8781 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
8782 kvm_vcpu_flush_tlb_current(vcpu);
8783 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
8784 kvm_vcpu_flush_tlb_guest(vcpu);
8786 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8787 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8791 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8792 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8793 vcpu->mmio_needed = 0;
8797 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8798 /* Page is swapped out. Do synthetic halt */
8799 vcpu->arch.apf.halted = true;
8803 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8804 record_steal_time(vcpu);
8805 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8807 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8809 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8810 kvm_pmu_handle_event(vcpu);
8811 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8812 kvm_pmu_deliver_pmi(vcpu);
8813 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8814 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8815 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8816 vcpu->arch.ioapic_handled_vectors)) {
8817 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8818 vcpu->run->eoi.vector =
8819 vcpu->arch.pending_ioapic_eoi;
8824 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8825 vcpu_scan_ioapic(vcpu);
8826 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8827 vcpu_load_eoi_exitmap(vcpu);
8828 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8829 kvm_vcpu_reload_apic_access_page(vcpu);
8830 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8831 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8832 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8836 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8837 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8838 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8842 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8843 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8844 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8850 * KVM_REQ_HV_STIMER has to be processed after
8851 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8852 * depend on the guest clock being up-to-date
8854 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8855 kvm_hv_process_stimers(vcpu);
8856 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8857 kvm_vcpu_update_apicv(vcpu);
8858 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
8859 kvm_check_async_pf_completion(vcpu);
8860 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
8861 kvm_x86_ops.msr_filter_changed(vcpu);
8864 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8865 ++vcpu->stat.req_event;
8866 kvm_apic_accept_events(vcpu);
8867 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8872 inject_pending_event(vcpu, &req_immediate_exit);
8874 kvm_x86_ops.enable_irq_window(vcpu);
8876 if (kvm_lapic_enabled(vcpu)) {
8877 update_cr8_intercept(vcpu);
8878 kvm_lapic_sync_to_vapic(vcpu);
8882 r = kvm_mmu_reload(vcpu);
8884 goto cancel_injection;
8889 kvm_x86_ops.prepare_guest_switch(vcpu);
8892 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8893 * IPI are then delayed after guest entry, which ensures that they
8894 * result in virtual interrupt delivery.
8896 local_irq_disable();
8897 vcpu->mode = IN_GUEST_MODE;
8899 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8902 * 1) We should set ->mode before checking ->requests. Please see
8903 * the comment in kvm_vcpu_exiting_guest_mode().
8905 * 2) For APICv, we should set ->mode before checking PID.ON. This
8906 * pairs with the memory barrier implicit in pi_test_and_set_on
8907 * (see vmx_deliver_posted_interrupt).
8909 * 3) This also orders the write to mode from any reads to the page
8910 * tables done while the VCPU is running. Please see the comment
8911 * in kvm_flush_remote_tlbs.
8913 smp_mb__after_srcu_read_unlock();
8916 * This handles the case where a posted interrupt was
8917 * notified with kvm_vcpu_kick.
8919 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8920 kvm_x86_ops.sync_pir_to_irr(vcpu);
8922 if (kvm_vcpu_exit_request(vcpu)) {
8923 vcpu->mode = OUTSIDE_GUEST_MODE;
8927 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8929 goto cancel_injection;
8932 if (req_immediate_exit) {
8933 kvm_make_request(KVM_REQ_EVENT, vcpu);
8934 kvm_x86_ops.request_immediate_exit(vcpu);
8937 trace_kvm_entry(vcpu);
8939 fpregs_assert_state_consistent();
8940 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8941 switch_fpu_return();
8943 if (unlikely(vcpu->arch.switch_db_regs)) {
8945 set_debugreg(vcpu->arch.eff_db[0], 0);
8946 set_debugreg(vcpu->arch.eff_db[1], 1);
8947 set_debugreg(vcpu->arch.eff_db[2], 2);
8948 set_debugreg(vcpu->arch.eff_db[3], 3);
8949 set_debugreg(vcpu->arch.dr6, 6);
8950 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8953 exit_fastpath = kvm_x86_ops.run(vcpu);
8956 * Do this here before restoring debug registers on the host. And
8957 * since we do this before handling the vmexit, a DR access vmexit
8958 * can (a) read the correct value of the debug registers, (b) set
8959 * KVM_DEBUGREG_WONT_EXIT again.
8961 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8962 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8963 kvm_x86_ops.sync_dirty_debug_regs(vcpu);
8964 kvm_update_dr0123(vcpu);
8965 kvm_update_dr7(vcpu);
8966 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8970 * If the guest has used debug registers, at least dr7
8971 * will be disabled while returning to the host.
8972 * If we don't have active breakpoints in the host, we don't
8973 * care about the messed up debug address registers. But if
8974 * we have some of them active, restore the old state.
8976 if (hw_breakpoint_active())
8977 hw_breakpoint_restore();
8979 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
8980 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8982 vcpu->mode = OUTSIDE_GUEST_MODE;
8985 kvm_x86_ops.handle_exit_irqoff(vcpu);
8988 * Consume any pending interrupts, including the possible source of
8989 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8990 * An instruction is required after local_irq_enable() to fully unblock
8991 * interrupts on processors that implement an interrupt shadow, the
8992 * stat.exits increment will do nicely.
8994 kvm_before_interrupt(vcpu);
8997 local_irq_disable();
8998 kvm_after_interrupt(vcpu);
9000 if (lapic_in_kernel(vcpu)) {
9001 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9002 if (delta != S64_MIN) {
9003 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9004 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9011 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9014 * Profile KVM exit RIPs:
9016 if (unlikely(prof_on == KVM_PROFILING)) {
9017 unsigned long rip = kvm_rip_read(vcpu);
9018 profile_hit(KVM_PROFILING, (void *)rip);
9021 if (unlikely(vcpu->arch.tsc_always_catchup))
9022 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9024 if (vcpu->arch.apic_attention)
9025 kvm_lapic_sync_from_vapic(vcpu);
9027 r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
9031 if (req_immediate_exit)
9032 kvm_make_request(KVM_REQ_EVENT, vcpu);
9033 kvm_x86_ops.cancel_injection(vcpu);
9034 if (unlikely(vcpu->arch.apic_attention))
9035 kvm_lapic_sync_from_vapic(vcpu);
9040 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9042 if (!kvm_arch_vcpu_runnable(vcpu) &&
9043 (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
9044 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9045 kvm_vcpu_block(vcpu);
9046 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9048 if (kvm_x86_ops.post_block)
9049 kvm_x86_ops.post_block(vcpu);
9051 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9055 kvm_apic_accept_events(vcpu);
9056 switch(vcpu->arch.mp_state) {
9057 case KVM_MP_STATE_HALTED:
9058 vcpu->arch.pv.pv_unhalted = false;
9059 vcpu->arch.mp_state =
9060 KVM_MP_STATE_RUNNABLE;
9062 case KVM_MP_STATE_RUNNABLE:
9063 vcpu->arch.apf.halted = false;
9065 case KVM_MP_STATE_INIT_RECEIVED:
9073 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9075 if (is_guest_mode(vcpu))
9076 kvm_x86_ops.nested_ops->check_events(vcpu);
9078 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9079 !vcpu->arch.apf.halted);
9082 static int vcpu_run(struct kvm_vcpu *vcpu)
9085 struct kvm *kvm = vcpu->kvm;
9087 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9088 vcpu->arch.l1tf_flush_l1d = true;
9091 if (kvm_vcpu_running(vcpu)) {
9092 r = vcpu_enter_guest(vcpu);
9094 r = vcpu_block(kvm, vcpu);
9100 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
9101 if (kvm_cpu_has_pending_timer(vcpu))
9102 kvm_inject_pending_timer_irqs(vcpu);
9104 if (dm_request_for_irq_injection(vcpu) &&
9105 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9107 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9108 ++vcpu->stat.request_irq_exits;
9112 if (__xfer_to_guest_mode_work_pending()) {
9113 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9114 r = xfer_to_guest_mode_handle_work(vcpu);
9117 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9121 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9126 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9130 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9131 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9132 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9136 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9138 BUG_ON(!vcpu->arch.pio.count);
9140 return complete_emulated_io(vcpu);
9144 * Implements the following, as a state machine:
9148 * for each mmio piece in the fragment
9156 * for each mmio piece in the fragment
9161 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9163 struct kvm_run *run = vcpu->run;
9164 struct kvm_mmio_fragment *frag;
9167 BUG_ON(!vcpu->mmio_needed);
9169 /* Complete previous fragment */
9170 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9171 len = min(8u, frag->len);
9172 if (!vcpu->mmio_is_write)
9173 memcpy(frag->data, run->mmio.data, len);
9175 if (frag->len <= 8) {
9176 /* Switch to the next fragment. */
9178 vcpu->mmio_cur_fragment++;
9180 /* Go forward to the next mmio piece. */
9186 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9187 vcpu->mmio_needed = 0;
9189 /* FIXME: return into emulator if single-stepping. */
9190 if (vcpu->mmio_is_write)
9192 vcpu->mmio_read_completed = 1;
9193 return complete_emulated_io(vcpu);
9196 run->exit_reason = KVM_EXIT_MMIO;
9197 run->mmio.phys_addr = frag->gpa;
9198 if (vcpu->mmio_is_write)
9199 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9200 run->mmio.len = min(8u, frag->len);
9201 run->mmio.is_write = vcpu->mmio_is_write;
9202 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9206 static void kvm_save_current_fpu(struct fpu *fpu)
9209 * If the target FPU state is not resident in the CPU registers, just
9210 * memcpy() from current, else save CPU state directly to the target.
9212 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9213 memcpy(&fpu->state, ¤t->thread.fpu.state,
9214 fpu_kernel_xstate_size);
9216 copy_fpregs_to_fpstate(fpu);
9219 /* Swap (qemu) user FPU context for the guest FPU context. */
9220 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9224 kvm_save_current_fpu(vcpu->arch.user_fpu);
9226 /* PKRU is separately restored in kvm_x86_ops.run. */
9227 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9228 ~XFEATURE_MASK_PKRU);
9230 fpregs_mark_activate();
9236 /* When vcpu_run ends, restore user space FPU context. */
9237 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9241 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9243 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9245 fpregs_mark_activate();
9248 ++vcpu->stat.fpu_reload;
9252 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9254 struct kvm_run *kvm_run = vcpu->run;
9258 kvm_sigset_activate(vcpu);
9259 kvm_load_guest_fpu(vcpu);
9261 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9262 if (kvm_run->immediate_exit) {
9266 kvm_vcpu_block(vcpu);
9267 kvm_apic_accept_events(vcpu);
9268 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9270 if (signal_pending(current)) {
9272 kvm_run->exit_reason = KVM_EXIT_INTR;
9273 ++vcpu->stat.signal_exits;
9278 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9283 if (kvm_run->kvm_dirty_regs) {
9284 r = sync_regs(vcpu);
9289 /* re-sync apic's tpr */
9290 if (!lapic_in_kernel(vcpu)) {
9291 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9297 if (unlikely(vcpu->arch.complete_userspace_io)) {
9298 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9299 vcpu->arch.complete_userspace_io = NULL;
9304 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9306 if (kvm_run->immediate_exit)
9312 kvm_put_guest_fpu(vcpu);
9313 if (kvm_run->kvm_valid_regs)
9315 post_kvm_run_save(vcpu);
9316 kvm_sigset_deactivate(vcpu);
9322 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9324 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9326 * We are here if userspace calls get_regs() in the middle of
9327 * instruction emulation. Registers state needs to be copied
9328 * back from emulation context to vcpu. Userspace shouldn't do
9329 * that usually, but some bad designed PV devices (vmware
9330 * backdoor interface) need this to work
9332 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9333 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9335 regs->rax = kvm_rax_read(vcpu);
9336 regs->rbx = kvm_rbx_read(vcpu);
9337 regs->rcx = kvm_rcx_read(vcpu);
9338 regs->rdx = kvm_rdx_read(vcpu);
9339 regs->rsi = kvm_rsi_read(vcpu);
9340 regs->rdi = kvm_rdi_read(vcpu);
9341 regs->rsp = kvm_rsp_read(vcpu);
9342 regs->rbp = kvm_rbp_read(vcpu);
9343 #ifdef CONFIG_X86_64
9344 regs->r8 = kvm_r8_read(vcpu);
9345 regs->r9 = kvm_r9_read(vcpu);
9346 regs->r10 = kvm_r10_read(vcpu);
9347 regs->r11 = kvm_r11_read(vcpu);
9348 regs->r12 = kvm_r12_read(vcpu);
9349 regs->r13 = kvm_r13_read(vcpu);
9350 regs->r14 = kvm_r14_read(vcpu);
9351 regs->r15 = kvm_r15_read(vcpu);
9354 regs->rip = kvm_rip_read(vcpu);
9355 regs->rflags = kvm_get_rflags(vcpu);
9358 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9361 __get_regs(vcpu, regs);
9366 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9368 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9369 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9371 kvm_rax_write(vcpu, regs->rax);
9372 kvm_rbx_write(vcpu, regs->rbx);
9373 kvm_rcx_write(vcpu, regs->rcx);
9374 kvm_rdx_write(vcpu, regs->rdx);
9375 kvm_rsi_write(vcpu, regs->rsi);
9376 kvm_rdi_write(vcpu, regs->rdi);
9377 kvm_rsp_write(vcpu, regs->rsp);
9378 kvm_rbp_write(vcpu, regs->rbp);
9379 #ifdef CONFIG_X86_64
9380 kvm_r8_write(vcpu, regs->r8);
9381 kvm_r9_write(vcpu, regs->r9);
9382 kvm_r10_write(vcpu, regs->r10);
9383 kvm_r11_write(vcpu, regs->r11);
9384 kvm_r12_write(vcpu, regs->r12);
9385 kvm_r13_write(vcpu, regs->r13);
9386 kvm_r14_write(vcpu, regs->r14);
9387 kvm_r15_write(vcpu, regs->r15);
9390 kvm_rip_write(vcpu, regs->rip);
9391 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9393 vcpu->arch.exception.pending = false;
9395 kvm_make_request(KVM_REQ_EVENT, vcpu);
9398 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9401 __set_regs(vcpu, regs);
9406 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9408 struct kvm_segment cs;
9410 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9414 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9416 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9420 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9421 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9422 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9423 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9424 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9425 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9427 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9428 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9430 kvm_x86_ops.get_idt(vcpu, &dt);
9431 sregs->idt.limit = dt.size;
9432 sregs->idt.base = dt.address;
9433 kvm_x86_ops.get_gdt(vcpu, &dt);
9434 sregs->gdt.limit = dt.size;
9435 sregs->gdt.base = dt.address;
9437 sregs->cr0 = kvm_read_cr0(vcpu);
9438 sregs->cr2 = vcpu->arch.cr2;
9439 sregs->cr3 = kvm_read_cr3(vcpu);
9440 sregs->cr4 = kvm_read_cr4(vcpu);
9441 sregs->cr8 = kvm_get_cr8(vcpu);
9442 sregs->efer = vcpu->arch.efer;
9443 sregs->apic_base = kvm_get_apic_base(vcpu);
9445 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9447 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9448 set_bit(vcpu->arch.interrupt.nr,
9449 (unsigned long *)sregs->interrupt_bitmap);
9452 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9453 struct kvm_sregs *sregs)
9456 __get_sregs(vcpu, sregs);
9461 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9462 struct kvm_mp_state *mp_state)
9465 if (kvm_mpx_supported())
9466 kvm_load_guest_fpu(vcpu);
9468 kvm_apic_accept_events(vcpu);
9469 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
9470 vcpu->arch.pv.pv_unhalted)
9471 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9473 mp_state->mp_state = vcpu->arch.mp_state;
9475 if (kvm_mpx_supported())
9476 kvm_put_guest_fpu(vcpu);
9481 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9482 struct kvm_mp_state *mp_state)
9488 if (!lapic_in_kernel(vcpu) &&
9489 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9493 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9494 * INIT state; latched init should be reported using
9495 * KVM_SET_VCPU_EVENTS, so reject it here.
9497 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9498 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9499 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9502 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9503 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9504 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9506 vcpu->arch.mp_state = mp_state->mp_state;
9507 kvm_make_request(KVM_REQ_EVENT, vcpu);
9515 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9516 int reason, bool has_error_code, u32 error_code)
9518 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9521 init_emulate_ctxt(vcpu);
9523 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9524 has_error_code, error_code);
9526 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9527 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9528 vcpu->run->internal.ndata = 0;
9532 kvm_rip_write(vcpu, ctxt->eip);
9533 kvm_set_rflags(vcpu, ctxt->eflags);
9536 EXPORT_SYMBOL_GPL(kvm_task_switch);
9538 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9540 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9542 * When EFER.LME and CR0.PG are set, the processor is in
9543 * 64-bit mode (though maybe in a 32-bit code segment).
9544 * CR4.PAE and EFER.LMA must be set.
9546 if (!(sregs->cr4 & X86_CR4_PAE)
9547 || !(sregs->efer & EFER_LMA))
9551 * Not in 64-bit mode: EFER.LMA is clear and the code
9552 * segment cannot be 64-bit.
9554 if (sregs->efer & EFER_LMA || sregs->cs.l)
9558 return kvm_valid_cr4(vcpu, sregs->cr4);
9561 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9563 struct msr_data apic_base_msr;
9564 int mmu_reset_needed = 0;
9565 int cpuid_update_needed = 0;
9566 int pending_vec, max_bits, idx;
9570 if (kvm_valid_sregs(vcpu, sregs))
9573 apic_base_msr.data = sregs->apic_base;
9574 apic_base_msr.host_initiated = true;
9575 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9578 dt.size = sregs->idt.limit;
9579 dt.address = sregs->idt.base;
9580 kvm_x86_ops.set_idt(vcpu, &dt);
9581 dt.size = sregs->gdt.limit;
9582 dt.address = sregs->gdt.base;
9583 kvm_x86_ops.set_gdt(vcpu, &dt);
9585 vcpu->arch.cr2 = sregs->cr2;
9586 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9587 vcpu->arch.cr3 = sregs->cr3;
9588 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9590 kvm_set_cr8(vcpu, sregs->cr8);
9592 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9593 kvm_x86_ops.set_efer(vcpu, sregs->efer);
9595 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9596 kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9597 vcpu->arch.cr0 = sregs->cr0;
9599 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9600 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
9601 (X86_CR4_OSXSAVE | X86_CR4_PKE));
9602 kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9603 if (cpuid_update_needed)
9604 kvm_update_cpuid_runtime(vcpu);
9606 idx = srcu_read_lock(&vcpu->kvm->srcu);
9607 if (is_pae_paging(vcpu)) {
9608 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9609 mmu_reset_needed = 1;
9611 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9613 if (mmu_reset_needed)
9614 kvm_mmu_reset_context(vcpu);
9616 max_bits = KVM_NR_INTERRUPTS;
9617 pending_vec = find_first_bit(
9618 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9619 if (pending_vec < max_bits) {
9620 kvm_queue_interrupt(vcpu, pending_vec, false);
9621 pr_debug("Set back pending irq %d\n", pending_vec);
9624 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9625 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9626 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9627 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9628 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9629 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9631 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9632 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9634 update_cr8_intercept(vcpu);
9636 /* Older userspace won't unhalt the vcpu on reset. */
9637 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9638 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9640 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9642 kvm_make_request(KVM_REQ_EVENT, vcpu);
9649 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9650 struct kvm_sregs *sregs)
9655 ret = __set_sregs(vcpu, sregs);
9660 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9661 struct kvm_guest_debug *dbg)
9663 unsigned long rflags;
9668 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9670 if (vcpu->arch.exception.pending)
9672 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9673 kvm_queue_exception(vcpu, DB_VECTOR);
9675 kvm_queue_exception(vcpu, BP_VECTOR);
9679 * Read rflags as long as potentially injected trace flags are still
9682 rflags = kvm_get_rflags(vcpu);
9684 vcpu->guest_debug = dbg->control;
9685 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9686 vcpu->guest_debug = 0;
9688 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9689 for (i = 0; i < KVM_NR_DB_REGS; ++i)
9690 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9691 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9693 for (i = 0; i < KVM_NR_DB_REGS; i++)
9694 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9696 kvm_update_dr7(vcpu);
9698 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9699 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9700 get_segment_base(vcpu, VCPU_SREG_CS);
9703 * Trigger an rflags update that will inject or remove the trace
9706 kvm_set_rflags(vcpu, rflags);
9708 kvm_x86_ops.update_exception_bitmap(vcpu);
9718 * Translate a guest virtual address to a guest physical address.
9720 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9721 struct kvm_translation *tr)
9723 unsigned long vaddr = tr->linear_address;
9729 idx = srcu_read_lock(&vcpu->kvm->srcu);
9730 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9731 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9732 tr->physical_address = gpa;
9733 tr->valid = gpa != UNMAPPED_GVA;
9741 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9743 struct fxregs_state *fxsave;
9747 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9748 memcpy(fpu->fpr, fxsave->st_space, 128);
9749 fpu->fcw = fxsave->cwd;
9750 fpu->fsw = fxsave->swd;
9751 fpu->ftwx = fxsave->twd;
9752 fpu->last_opcode = fxsave->fop;
9753 fpu->last_ip = fxsave->rip;
9754 fpu->last_dp = fxsave->rdp;
9755 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9761 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9763 struct fxregs_state *fxsave;
9767 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9769 memcpy(fxsave->st_space, fpu->fpr, 128);
9770 fxsave->cwd = fpu->fcw;
9771 fxsave->swd = fpu->fsw;
9772 fxsave->twd = fpu->ftwx;
9773 fxsave->fop = fpu->last_opcode;
9774 fxsave->rip = fpu->last_ip;
9775 fxsave->rdp = fpu->last_dp;
9776 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9782 static void store_regs(struct kvm_vcpu *vcpu)
9784 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9786 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9787 __get_regs(vcpu, &vcpu->run->s.regs.regs);
9789 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9790 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9792 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9793 kvm_vcpu_ioctl_x86_get_vcpu_events(
9794 vcpu, &vcpu->run->s.regs.events);
9797 static int sync_regs(struct kvm_vcpu *vcpu)
9799 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9802 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9803 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9804 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9806 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9807 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9809 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9811 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9812 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9813 vcpu, &vcpu->run->s.regs.events))
9815 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9821 static void fx_init(struct kvm_vcpu *vcpu)
9823 fpstate_init(&vcpu->arch.guest_fpu->state);
9824 if (boot_cpu_has(X86_FEATURE_XSAVES))
9825 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9826 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9829 * Ensure guest xcr0 is valid for loading
9831 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9833 vcpu->arch.cr0 |= X86_CR0_ET;
9836 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9838 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9839 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9840 "guest TSC will not be reliable\n");
9845 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9850 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9851 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9853 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9855 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9857 r = kvm_mmu_create(vcpu);
9861 if (irqchip_in_kernel(vcpu->kvm)) {
9862 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9864 goto fail_mmu_destroy;
9865 if (kvm_apicv_activated(vcpu->kvm))
9866 vcpu->arch.apicv_active = true;
9868 static_key_slow_inc(&kvm_no_apic_vcpu);
9872 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9874 goto fail_free_lapic;
9875 vcpu->arch.pio_data = page_address(page);
9877 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9878 GFP_KERNEL_ACCOUNT);
9879 if (!vcpu->arch.mce_banks)
9880 goto fail_free_pio_data;
9881 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9883 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9884 GFP_KERNEL_ACCOUNT))
9885 goto fail_free_mce_banks;
9887 if (!alloc_emulate_ctxt(vcpu))
9888 goto free_wbinvd_dirty_mask;
9890 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9891 GFP_KERNEL_ACCOUNT);
9892 if (!vcpu->arch.user_fpu) {
9893 pr_err("kvm: failed to allocate userspace's fpu\n");
9894 goto free_emulate_ctxt;
9897 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9898 GFP_KERNEL_ACCOUNT);
9899 if (!vcpu->arch.guest_fpu) {
9900 pr_err("kvm: failed to allocate vcpu's fpu\n");
9905 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9907 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9909 kvm_async_pf_hash_reset(vcpu);
9912 vcpu->arch.pending_external_vector = -1;
9913 vcpu->arch.preempted_in_kernel = false;
9915 kvm_hv_vcpu_init(vcpu);
9917 r = kvm_x86_ops.vcpu_create(vcpu);
9919 goto free_guest_fpu;
9921 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9922 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9923 kvm_vcpu_mtrr_init(vcpu);
9925 kvm_vcpu_reset(vcpu, false);
9926 kvm_init_mmu(vcpu, false);
9931 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9933 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9935 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9936 free_wbinvd_dirty_mask:
9937 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9938 fail_free_mce_banks:
9939 kfree(vcpu->arch.mce_banks);
9941 free_page((unsigned long)vcpu->arch.pio_data);
9943 kvm_free_lapic(vcpu);
9945 kvm_mmu_destroy(vcpu);
9949 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9951 struct kvm *kvm = vcpu->kvm;
9953 kvm_hv_vcpu_postcreate(vcpu);
9955 if (mutex_lock_killable(&vcpu->mutex))
9958 kvm_synchronize_tsc(vcpu, 0);
9961 /* poll control enabled by default */
9962 vcpu->arch.msr_kvm_poll_control = 1;
9964 mutex_unlock(&vcpu->mutex);
9966 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
9967 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9968 KVMCLOCK_SYNC_PERIOD);
9971 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9973 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9976 kvm_release_pfn(cache->pfn, cache->dirty, cache);
9978 kvmclock_reset(vcpu);
9980 kvm_x86_ops.vcpu_free(vcpu);
9982 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9983 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9984 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9985 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9987 kvm_hv_vcpu_uninit(vcpu);
9988 kvm_pmu_destroy(vcpu);
9989 kfree(vcpu->arch.mce_banks);
9990 kvm_free_lapic(vcpu);
9991 idx = srcu_read_lock(&vcpu->kvm->srcu);
9992 kvm_mmu_destroy(vcpu);
9993 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9994 free_page((unsigned long)vcpu->arch.pio_data);
9995 kvfree(vcpu->arch.cpuid_entries);
9996 if (!lapic_in_kernel(vcpu))
9997 static_key_slow_dec(&kvm_no_apic_vcpu);
10000 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10002 kvm_lapic_reset(vcpu, init_event);
10004 vcpu->arch.hflags = 0;
10006 vcpu->arch.smi_pending = 0;
10007 vcpu->arch.smi_count = 0;
10008 atomic_set(&vcpu->arch.nmi_queued, 0);
10009 vcpu->arch.nmi_pending = 0;
10010 vcpu->arch.nmi_injected = false;
10011 kvm_clear_interrupt_queue(vcpu);
10012 kvm_clear_exception_queue(vcpu);
10014 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10015 kvm_update_dr0123(vcpu);
10016 vcpu->arch.dr6 = DR6_INIT;
10017 vcpu->arch.dr7 = DR7_FIXED_1;
10018 kvm_update_dr7(vcpu);
10020 vcpu->arch.cr2 = 0;
10022 kvm_make_request(KVM_REQ_EVENT, vcpu);
10023 vcpu->arch.apf.msr_en_val = 0;
10024 vcpu->arch.apf.msr_int_val = 0;
10025 vcpu->arch.st.msr_val = 0;
10027 kvmclock_reset(vcpu);
10029 kvm_clear_async_pf_completion_queue(vcpu);
10030 kvm_async_pf_hash_reset(vcpu);
10031 vcpu->arch.apf.halted = false;
10033 if (kvm_mpx_supported()) {
10034 void *mpx_state_buffer;
10037 * To avoid have the INIT path from kvm_apic_has_events() that be
10038 * called with loaded FPU and does not let userspace fix the state.
10041 kvm_put_guest_fpu(vcpu);
10042 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10044 if (mpx_state_buffer)
10045 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10046 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10048 if (mpx_state_buffer)
10049 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10051 kvm_load_guest_fpu(vcpu);
10055 kvm_pmu_reset(vcpu);
10056 vcpu->arch.smbase = 0x30000;
10058 vcpu->arch.msr_misc_features_enables = 0;
10060 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10063 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10064 vcpu->arch.regs_avail = ~0;
10065 vcpu->arch.regs_dirty = ~0;
10067 vcpu->arch.ia32_xss = 0;
10069 kvm_x86_ops.vcpu_reset(vcpu, init_event);
10072 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10074 struct kvm_segment cs;
10076 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10077 cs.selector = vector << 8;
10078 cs.base = vector << 12;
10079 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10080 kvm_rip_write(vcpu, 0);
10083 int kvm_arch_hardware_enable(void)
10086 struct kvm_vcpu *vcpu;
10091 bool stable, backwards_tsc = false;
10093 kvm_user_return_msr_cpu_online();
10094 ret = kvm_x86_ops.hardware_enable();
10098 local_tsc = rdtsc();
10099 stable = !kvm_check_tsc_unstable();
10100 list_for_each_entry(kvm, &vm_list, vm_list) {
10101 kvm_for_each_vcpu(i, vcpu, kvm) {
10102 if (!stable && vcpu->cpu == smp_processor_id())
10103 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10104 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10105 backwards_tsc = true;
10106 if (vcpu->arch.last_host_tsc > max_tsc)
10107 max_tsc = vcpu->arch.last_host_tsc;
10113 * Sometimes, even reliable TSCs go backwards. This happens on
10114 * platforms that reset TSC during suspend or hibernate actions, but
10115 * maintain synchronization. We must compensate. Fortunately, we can
10116 * detect that condition here, which happens early in CPU bringup,
10117 * before any KVM threads can be running. Unfortunately, we can't
10118 * bring the TSCs fully up to date with real time, as we aren't yet far
10119 * enough into CPU bringup that we know how much real time has actually
10120 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10121 * variables that haven't been updated yet.
10123 * So we simply find the maximum observed TSC above, then record the
10124 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10125 * the adjustment will be applied. Note that we accumulate
10126 * adjustments, in case multiple suspend cycles happen before some VCPU
10127 * gets a chance to run again. In the event that no KVM threads get a
10128 * chance to run, we will miss the entire elapsed period, as we'll have
10129 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10130 * loose cycle time. This isn't too big a deal, since the loss will be
10131 * uniform across all VCPUs (not to mention the scenario is extremely
10132 * unlikely). It is possible that a second hibernate recovery happens
10133 * much faster than a first, causing the observed TSC here to be
10134 * smaller; this would require additional padding adjustment, which is
10135 * why we set last_host_tsc to the local tsc observed here.
10137 * N.B. - this code below runs only on platforms with reliable TSC,
10138 * as that is the only way backwards_tsc is set above. Also note
10139 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10140 * have the same delta_cyc adjustment applied if backwards_tsc
10141 * is detected. Note further, this adjustment is only done once,
10142 * as we reset last_host_tsc on all VCPUs to stop this from being
10143 * called multiple times (one for each physical CPU bringup).
10145 * Platforms with unreliable TSCs don't have to deal with this, they
10146 * will be compensated by the logic in vcpu_load, which sets the TSC to
10147 * catchup mode. This will catchup all VCPUs to real time, but cannot
10148 * guarantee that they stay in perfect synchronization.
10150 if (backwards_tsc) {
10151 u64 delta_cyc = max_tsc - local_tsc;
10152 list_for_each_entry(kvm, &vm_list, vm_list) {
10153 kvm->arch.backwards_tsc_observed = true;
10154 kvm_for_each_vcpu(i, vcpu, kvm) {
10155 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10156 vcpu->arch.last_host_tsc = local_tsc;
10157 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10161 * We have to disable TSC offset matching.. if you were
10162 * booting a VM while issuing an S4 host suspend....
10163 * you may have some problem. Solving this issue is
10164 * left as an exercise to the reader.
10166 kvm->arch.last_tsc_nsec = 0;
10167 kvm->arch.last_tsc_write = 0;
10174 void kvm_arch_hardware_disable(void)
10176 kvm_x86_ops.hardware_disable();
10177 drop_user_return_notifiers();
10180 int kvm_arch_hardware_setup(void *opaque)
10182 struct kvm_x86_init_ops *ops = opaque;
10185 rdmsrl_safe(MSR_EFER, &host_efer);
10187 if (boot_cpu_has(X86_FEATURE_XSAVES))
10188 rdmsrl(MSR_IA32_XSS, host_xss);
10190 r = ops->hardware_setup();
10194 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10196 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10199 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10200 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10201 #undef __kvm_cpu_cap_has
10203 if (kvm_has_tsc_control) {
10205 * Make sure the user can only configure tsc_khz values that
10206 * fit into a signed integer.
10207 * A min value is not calculated because it will always
10208 * be 1 on all machines.
10210 u64 max = min(0x7fffffffULL,
10211 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10212 kvm_max_guest_tsc_khz = max;
10214 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10217 kvm_init_msr_list();
10221 void kvm_arch_hardware_unsetup(void)
10223 kvm_x86_ops.hardware_unsetup();
10226 int kvm_arch_check_processor_compat(void *opaque)
10228 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10229 struct kvm_x86_init_ops *ops = opaque;
10231 WARN_ON(!irqs_disabled());
10233 if (__cr4_reserved_bits(cpu_has, c) !=
10234 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10237 return ops->check_processor_compatibility();
10240 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10242 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10244 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10246 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10248 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10251 struct static_key kvm_no_apic_vcpu __read_mostly;
10252 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
10254 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10256 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10258 vcpu->arch.l1tf_flush_l1d = true;
10259 if (pmu->version && unlikely(pmu->event_count)) {
10260 pmu->need_cleanup = true;
10261 kvm_make_request(KVM_REQ_PMU, vcpu);
10263 kvm_x86_ops.sched_in(vcpu, cpu);
10266 void kvm_arch_free_vm(struct kvm *kvm)
10268 kfree(kvm->arch.hyperv.hv_pa_pg);
10273 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10278 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10279 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10280 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10281 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10282 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10283 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10285 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10286 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10287 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10288 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10289 &kvm->arch.irq_sources_bitmap);
10291 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10292 mutex_init(&kvm->arch.apic_map_lock);
10293 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10295 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10296 pvclock_update_vm_gtod_copy(kvm);
10298 kvm->arch.guest_can_read_msr_platform_info = true;
10300 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10301 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10303 kvm_hv_init_vm(kvm);
10304 kvm_page_track_init(kvm);
10305 kvm_mmu_init_vm(kvm);
10307 return kvm_x86_ops.vm_init(kvm);
10310 int kvm_arch_post_init_vm(struct kvm *kvm)
10312 return kvm_mmu_post_init_vm(kvm);
10315 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10318 kvm_mmu_unload(vcpu);
10322 static void kvm_free_vcpus(struct kvm *kvm)
10325 struct kvm_vcpu *vcpu;
10328 * Unpin any mmu pages first.
10330 kvm_for_each_vcpu(i, vcpu, kvm) {
10331 kvm_clear_async_pf_completion_queue(vcpu);
10332 kvm_unload_vcpu_mmu(vcpu);
10334 kvm_for_each_vcpu(i, vcpu, kvm)
10335 kvm_vcpu_destroy(vcpu);
10337 mutex_lock(&kvm->lock);
10338 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10339 kvm->vcpus[i] = NULL;
10341 atomic_set(&kvm->online_vcpus, 0);
10342 mutex_unlock(&kvm->lock);
10345 void kvm_arch_sync_events(struct kvm *kvm)
10347 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10348 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10352 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
10355 unsigned long hva, old_npages;
10356 struct kvm_memslots *slots = kvm_memslots(kvm);
10357 struct kvm_memory_slot *slot;
10359 /* Called with kvm->slots_lock held. */
10360 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10363 slot = id_to_memslot(slots, id);
10365 if (slot && slot->npages)
10369 * MAP_SHARED to prevent internal slot pages from being moved
10372 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10373 MAP_SHARED | MAP_ANONYMOUS, 0);
10374 if (IS_ERR((void *)hva))
10375 return PTR_ERR((void *)hva);
10377 if (!slot || !slot->npages)
10380 old_npages = slot->npages;
10384 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10385 struct kvm_userspace_memory_region m;
10387 m.slot = id | (i << 16);
10389 m.guest_phys_addr = gpa;
10390 m.userspace_addr = hva;
10391 m.memory_size = size;
10392 r = __kvm_set_memory_region(kvm, &m);
10398 vm_munmap(hva, old_npages * PAGE_SIZE);
10402 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10404 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10406 kvm_mmu_pre_destroy_vm(kvm);
10409 void kvm_arch_destroy_vm(struct kvm *kvm)
10413 if (current->mm == kvm->mm) {
10415 * Free memory regions allocated on behalf of userspace,
10416 * unless the the memory map has changed due to process exit
10419 mutex_lock(&kvm->slots_lock);
10420 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10422 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10424 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10425 mutex_unlock(&kvm->slots_lock);
10427 if (kvm_x86_ops.vm_destroy)
10428 kvm_x86_ops.vm_destroy(kvm);
10429 for (i = 0; i < kvm->arch.msr_filter.count; i++)
10430 kfree(kvm->arch.msr_filter.ranges[i].bitmap);
10431 kvm_pic_destroy(kvm);
10432 kvm_ioapic_destroy(kvm);
10433 kvm_free_vcpus(kvm);
10434 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10435 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10436 kvm_mmu_uninit_vm(kvm);
10437 kvm_page_track_cleanup(kvm);
10438 kvm_hv_destroy_vm(kvm);
10441 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10445 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10446 kvfree(slot->arch.rmap[i]);
10447 slot->arch.rmap[i] = NULL;
10452 kvfree(slot->arch.lpage_info[i - 1]);
10453 slot->arch.lpage_info[i - 1] = NULL;
10456 kvm_page_track_free_memslot(slot);
10459 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10460 unsigned long npages)
10465 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
10466 * old arrays will be freed by __kvm_set_memory_region() if installing
10467 * the new memslot is successful.
10469 memset(&slot->arch, 0, sizeof(slot->arch));
10471 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10472 struct kvm_lpage_info *linfo;
10473 unsigned long ugfn;
10477 lpages = gfn_to_index(slot->base_gfn + npages - 1,
10478 slot->base_gfn, level) + 1;
10480 slot->arch.rmap[i] =
10481 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10482 GFP_KERNEL_ACCOUNT);
10483 if (!slot->arch.rmap[i])
10488 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10492 slot->arch.lpage_info[i - 1] = linfo;
10494 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10495 linfo[0].disallow_lpage = 1;
10496 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10497 linfo[lpages - 1].disallow_lpage = 1;
10498 ugfn = slot->userspace_addr >> PAGE_SHIFT;
10500 * If the gfn and userspace address are not aligned wrt each
10501 * other, disable large page support for this slot.
10503 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10506 for (j = 0; j < lpages; ++j)
10507 linfo[j].disallow_lpage = 1;
10511 if (kvm_page_track_create_memslot(slot, npages))
10517 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10518 kvfree(slot->arch.rmap[i]);
10519 slot->arch.rmap[i] = NULL;
10523 kvfree(slot->arch.lpage_info[i - 1]);
10524 slot->arch.lpage_info[i - 1] = NULL;
10529 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10531 struct kvm_vcpu *vcpu;
10535 * memslots->generation has been incremented.
10536 * mmio generation may have reached its maximum value.
10538 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10540 /* Force re-initialization of steal_time cache */
10541 kvm_for_each_vcpu(i, vcpu, kvm)
10542 kvm_vcpu_kick(vcpu);
10545 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10546 struct kvm_memory_slot *memslot,
10547 const struct kvm_userspace_memory_region *mem,
10548 enum kvm_mr_change change)
10550 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10551 return kvm_alloc_memslot_metadata(memslot,
10552 mem->memory_size >> PAGE_SHIFT);
10556 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10557 struct kvm_memory_slot *old,
10558 struct kvm_memory_slot *new,
10559 enum kvm_mr_change change)
10562 * Nothing to do for RO slots or CREATE/MOVE/DELETE of a slot.
10563 * See comments below.
10565 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10569 * Dirty logging tracks sptes in 4k granularity, meaning that large
10570 * sptes have to be split. If live migration is successful, the guest
10571 * in the source machine will be destroyed and large sptes will be
10572 * created in the destination. However, if the guest continues to run
10573 * in the source machine (for example if live migration fails), small
10574 * sptes will remain around and cause bad performance.
10576 * Scan sptes if dirty logging has been stopped, dropping those
10577 * which can be collapsed into a single large-page spte. Later
10578 * page faults will create the large-page sptes.
10580 * There is no need to do this in any of the following cases:
10581 * CREATE: No dirty mappings will already exist.
10582 * MOVE/DELETE: The old mappings will already have been cleaned up by
10583 * kvm_arch_flush_shadow_memslot()
10585 if ((old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10586 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10587 kvm_mmu_zap_collapsible_sptes(kvm, new);
10590 * Enable or disable dirty logging for the slot.
10592 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of the old
10593 * slot have been zapped so no dirty logging updates are needed for
10595 * For KVM_MR_CREATE and KVM_MR_MOVE, once the new slot is visible
10596 * any mappings that might be created in it will consume the
10597 * properties of the new slot and do not need to be updated here.
10599 * When PML is enabled, the kvm_x86_ops dirty logging hooks are
10600 * called to enable/disable dirty logging.
10602 * When disabling dirty logging with PML enabled, the D-bit is set
10603 * for sptes in the slot in order to prevent unnecessary GPA
10604 * logging in the PML buffer (and potential PML buffer full VMEXIT).
10605 * This guarantees leaving PML enabled for the guest's lifetime
10606 * won't have any additional overhead from PML when the guest is
10607 * running with dirty logging disabled.
10609 * When enabling dirty logging, large sptes are write-protected
10610 * so they can be split on first write. New large sptes cannot
10611 * be created for this slot until the end of the logging.
10612 * See the comments in fast_page_fault().
10613 * For small sptes, nothing is done if the dirty log is in the
10614 * initial-all-set state. Otherwise, depending on whether pml
10615 * is enabled the D-bit or the W-bit will be cleared.
10617 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10618 if (kvm_x86_ops.slot_enable_log_dirty) {
10619 kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10622 kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10623 PG_LEVEL_2M : PG_LEVEL_4K;
10626 * If we're with initial-all-set, we don't need
10627 * to write protect any small page because
10628 * they're reported as dirty already. However
10629 * we still need to write-protect huge pages
10630 * so that the page split can happen lazily on
10631 * the first write to the huge page.
10633 kvm_mmu_slot_remove_write_access(kvm, new, level);
10636 if (kvm_x86_ops.slot_disable_log_dirty)
10637 kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10641 void kvm_arch_commit_memory_region(struct kvm *kvm,
10642 const struct kvm_userspace_memory_region *mem,
10643 struct kvm_memory_slot *old,
10644 const struct kvm_memory_slot *new,
10645 enum kvm_mr_change change)
10647 if (!kvm->arch.n_requested_mmu_pages)
10648 kvm_mmu_change_mmu_pages(kvm,
10649 kvm_mmu_calculate_default_mmu_pages(kvm));
10652 * FIXME: const-ify all uses of struct kvm_memory_slot.
10654 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
10656 /* Free the arrays associated with the old memslot. */
10657 if (change == KVM_MR_MOVE)
10658 kvm_arch_free_memslot(kvm, old);
10661 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10663 kvm_mmu_zap_all(kvm);
10666 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10667 struct kvm_memory_slot *slot)
10669 kvm_page_track_flush_slot(kvm, slot);
10672 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10674 return (is_guest_mode(vcpu) &&
10675 kvm_x86_ops.guest_apic_has_interrupt &&
10676 kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10679 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10681 if (!list_empty_careful(&vcpu->async_pf.done))
10684 if (kvm_apic_has_events(vcpu))
10687 if (vcpu->arch.pv.pv_unhalted)
10690 if (vcpu->arch.exception.pending)
10693 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10694 (vcpu->arch.nmi_pending &&
10695 kvm_x86_ops.nmi_allowed(vcpu, false)))
10698 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10699 (vcpu->arch.smi_pending &&
10700 kvm_x86_ops.smi_allowed(vcpu, false)))
10703 if (kvm_arch_interrupt_allowed(vcpu) &&
10704 (kvm_cpu_has_interrupt(vcpu) ||
10705 kvm_guest_apic_has_interrupt(vcpu)))
10708 if (kvm_hv_has_stimer_pending(vcpu))
10711 if (is_guest_mode(vcpu) &&
10712 kvm_x86_ops.nested_ops->hv_timer_pending &&
10713 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
10719 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10721 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10724 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10726 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10729 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10730 kvm_test_request(KVM_REQ_SMI, vcpu) ||
10731 kvm_test_request(KVM_REQ_EVENT, vcpu))
10734 if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10740 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10742 return vcpu->arch.preempted_in_kernel;
10745 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10747 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10750 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10752 return kvm_x86_ops.interrupt_allowed(vcpu, false);
10755 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10757 if (is_64_bit_mode(vcpu))
10758 return kvm_rip_read(vcpu);
10759 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10760 kvm_rip_read(vcpu));
10762 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10764 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10766 return kvm_get_linear_rip(vcpu) == linear_rip;
10768 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10770 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10772 unsigned long rflags;
10774 rflags = kvm_x86_ops.get_rflags(vcpu);
10775 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10776 rflags &= ~X86_EFLAGS_TF;
10779 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10781 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10783 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10784 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10785 rflags |= X86_EFLAGS_TF;
10786 kvm_x86_ops.set_rflags(vcpu, rflags);
10789 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10791 __kvm_set_rflags(vcpu, rflags);
10792 kvm_make_request(KVM_REQ_EVENT, vcpu);
10794 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10796 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10800 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10804 r = kvm_mmu_reload(vcpu);
10808 if (!vcpu->arch.mmu->direct_map &&
10809 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10812 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10815 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10817 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
10819 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10822 static inline u32 kvm_async_pf_next_probe(u32 key)
10824 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
10827 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10829 u32 key = kvm_async_pf_hash_fn(gfn);
10831 while (vcpu->arch.apf.gfns[key] != ~0)
10832 key = kvm_async_pf_next_probe(key);
10834 vcpu->arch.apf.gfns[key] = gfn;
10837 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10840 u32 key = kvm_async_pf_hash_fn(gfn);
10842 for (i = 0; i < ASYNC_PF_PER_VCPU &&
10843 (vcpu->arch.apf.gfns[key] != gfn &&
10844 vcpu->arch.apf.gfns[key] != ~0); i++)
10845 key = kvm_async_pf_next_probe(key);
10850 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10852 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10855 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10859 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10861 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
10865 vcpu->arch.apf.gfns[i] = ~0;
10867 j = kvm_async_pf_next_probe(j);
10868 if (vcpu->arch.apf.gfns[j] == ~0)
10870 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10872 * k lies cyclically in ]i,j]
10874 * |....j i.k.| or |.k..j i...|
10876 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10877 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10882 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
10884 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
10886 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
10890 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
10892 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
10894 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
10895 &token, offset, sizeof(token));
10898 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
10900 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
10903 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
10904 &val, offset, sizeof(val)))
10910 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10912 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10915 if (!kvm_pv_async_pf_enabled(vcpu) ||
10916 (vcpu->arch.apf.send_user_only && kvm_x86_ops.get_cpl(vcpu) == 0))
10922 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10924 if (unlikely(!lapic_in_kernel(vcpu) ||
10925 kvm_event_needs_reinjection(vcpu) ||
10926 vcpu->arch.exception.pending))
10929 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10933 * If interrupts are off we cannot even use an artificial
10936 return kvm_arch_interrupt_allowed(vcpu);
10939 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10940 struct kvm_async_pf *work)
10942 struct x86_exception fault;
10944 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10945 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10947 if (kvm_can_deliver_async_pf(vcpu) &&
10948 !apf_put_user_notpresent(vcpu)) {
10949 fault.vector = PF_VECTOR;
10950 fault.error_code_valid = true;
10951 fault.error_code = 0;
10952 fault.nested_page_fault = false;
10953 fault.address = work->arch.token;
10954 fault.async_page_fault = true;
10955 kvm_inject_page_fault(vcpu, &fault);
10959 * It is not possible to deliver a paravirtualized asynchronous
10960 * page fault, but putting the guest in an artificial halt state
10961 * can be beneficial nevertheless: if an interrupt arrives, we
10962 * can deliver it timely and perhaps the guest will schedule
10963 * another process. When the instruction that triggered a page
10964 * fault is retried, hopefully the page will be ready in the host.
10966 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10971 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10972 struct kvm_async_pf *work)
10974 struct kvm_lapic_irq irq = {
10975 .delivery_mode = APIC_DM_FIXED,
10976 .vector = vcpu->arch.apf.vec
10979 if (work->wakeup_all)
10980 work->arch.token = ~0; /* broadcast wakeup */
10982 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10983 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10985 if ((work->wakeup_all || work->notpresent_injected) &&
10986 kvm_pv_async_pf_enabled(vcpu) &&
10987 !apf_put_user_ready(vcpu, work->arch.token)) {
10988 vcpu->arch.apf.pageready_pending = true;
10989 kvm_apic_set_irq(vcpu, &irq, NULL);
10992 vcpu->arch.apf.halted = false;
10993 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10996 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
10998 kvm_make_request(KVM_REQ_APF_READY, vcpu);
10999 if (!vcpu->arch.apf.pageready_pending)
11000 kvm_vcpu_kick(vcpu);
11003 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11005 if (!kvm_pv_async_pf_enabled(vcpu))
11008 return apf_pageready_slot_free(vcpu);
11011 void kvm_arch_start_assignment(struct kvm *kvm)
11013 atomic_inc(&kvm->arch.assigned_device_count);
11015 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11017 void kvm_arch_end_assignment(struct kvm *kvm)
11019 atomic_dec(&kvm->arch.assigned_device_count);
11021 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11023 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11025 return atomic_read(&kvm->arch.assigned_device_count);
11027 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11029 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11031 atomic_inc(&kvm->arch.noncoherent_dma_count);
11033 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11035 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11037 atomic_dec(&kvm->arch.noncoherent_dma_count);
11039 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11041 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11043 return atomic_read(&kvm->arch.noncoherent_dma_count);
11045 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11047 bool kvm_arch_has_irq_bypass(void)
11052 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11053 struct irq_bypass_producer *prod)
11055 struct kvm_kernel_irqfd *irqfd =
11056 container_of(cons, struct kvm_kernel_irqfd, consumer);
11059 irqfd->producer = prod;
11060 kvm_arch_start_assignment(irqfd->kvm);
11061 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm,
11062 prod->irq, irqfd->gsi, 1);
11065 kvm_arch_end_assignment(irqfd->kvm);
11070 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11071 struct irq_bypass_producer *prod)
11074 struct kvm_kernel_irqfd *irqfd =
11075 container_of(cons, struct kvm_kernel_irqfd, consumer);
11077 WARN_ON(irqfd->producer != prod);
11078 irqfd->producer = NULL;
11081 * When producer of consumer is unregistered, we change back to
11082 * remapped mode, so we can re-use the current implementation
11083 * when the irq is masked/disabled or the consumer side (KVM
11084 * int this case doesn't want to receive the interrupts.
11086 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11088 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11089 " fails: %d\n", irqfd->consumer.token, ret);
11091 kvm_arch_end_assignment(irqfd->kvm);
11094 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11095 uint32_t guest_irq, bool set)
11097 return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
11100 bool kvm_vector_hashing_enabled(void)
11102 return vector_hashing;
11105 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11107 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11109 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11112 int kvm_spec_ctrl_test_value(u64 value)
11115 * test that setting IA32_SPEC_CTRL to given value
11116 * is allowed by the host processor
11120 unsigned long flags;
11123 local_irq_save(flags);
11125 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11127 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11130 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11132 local_irq_restore(flags);
11136 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11138 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11140 struct x86_exception fault;
11141 u32 access = error_code &
11142 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11144 if (!(error_code & PFERR_PRESENT_MASK) ||
11145 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11147 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11148 * tables probably do not match the TLB. Just proceed
11149 * with the error code that the processor gave.
11151 fault.vector = PF_VECTOR;
11152 fault.error_code_valid = true;
11153 fault.error_code = error_code;
11154 fault.nested_page_fault = false;
11155 fault.address = gva;
11157 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11159 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11162 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11163 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11164 * indicates whether exit to userspace is needed.
11166 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11167 struct x86_exception *e)
11169 if (r == X86EMUL_PROPAGATE_FAULT) {
11170 kvm_inject_emulated_page_fault(vcpu, e);
11175 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11176 * while handling a VMX instruction KVM could've handled the request
11177 * correctly by exiting to userspace and performing I/O but there
11178 * doesn't seem to be a real use-case behind such requests, just return
11179 * KVM_EXIT_INTERNAL_ERROR for now.
11181 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11182 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11183 vcpu->run->internal.ndata = 0;
11187 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11189 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11192 struct x86_exception e;
11194 unsigned long roots_to_free = 0;
11201 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11202 if (r != X86EMUL_CONTINUE)
11203 return kvm_handle_memory_failure(vcpu, r, &e);
11205 if (operand.pcid >> 12 != 0) {
11206 kvm_inject_gp(vcpu, 0);
11210 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11213 case INVPCID_TYPE_INDIV_ADDR:
11214 if ((!pcid_enabled && (operand.pcid != 0)) ||
11215 is_noncanonical_address(operand.gla, vcpu)) {
11216 kvm_inject_gp(vcpu, 0);
11219 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11220 return kvm_skip_emulated_instruction(vcpu);
11222 case INVPCID_TYPE_SINGLE_CTXT:
11223 if (!pcid_enabled && (operand.pcid != 0)) {
11224 kvm_inject_gp(vcpu, 0);
11228 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11229 kvm_mmu_sync_roots(vcpu);
11230 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11233 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11234 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11236 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11238 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11240 * If neither the current cr3 nor any of the prev_roots use the
11241 * given PCID, then nothing needs to be done here because a
11242 * resync will happen anyway before switching to any other CR3.
11245 return kvm_skip_emulated_instruction(vcpu);
11247 case INVPCID_TYPE_ALL_NON_GLOBAL:
11249 * Currently, KVM doesn't mark global entries in the shadow
11250 * page tables, so a non-global flush just degenerates to a
11251 * global flush. If needed, we could optimize this later by
11252 * keeping track of global entries in shadow page tables.
11256 case INVPCID_TYPE_ALL_INCL_GLOBAL:
11257 kvm_mmu_unload(vcpu);
11258 return kvm_skip_emulated_instruction(vcpu);
11261 BUG(); /* We have already checked above that type <= 3 */
11264 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11266 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11267 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11268 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11269 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11270 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11271 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11272 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11273 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11274 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11275 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11276 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11277 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11278 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11279 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11280 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11281 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11282 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11283 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11284 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11285 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11286 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11287 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);