2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
30 #include "assigned-dev.h"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <trace/events/kvm.h>
59 #include <asm/debugreg.h>
63 #include <linux/kernel_stat.h>
64 #include <asm/fpu/internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
67 #include <asm/irq_remapping.h>
69 #define CREATE_TRACE_POINTS
72 #define MAX_IO_MSRS 256
73 #define KVM_MAX_MCE_BANKS 32
74 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
75 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
77 #define emul_to_vcpu(ctxt) \
78 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
81 * - enable syscall per default because its emulated by KVM
82 * - enable LME and LMA per default on 64 bit KVM
86 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
88 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
91 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
92 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
94 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
95 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
97 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
98 static void process_nmi(struct kvm_vcpu *vcpu);
99 static void enter_smm(struct kvm_vcpu *vcpu);
100 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
102 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
103 EXPORT_SYMBOL_GPL(kvm_x86_ops);
105 static bool __read_mostly ignore_msrs = 0;
106 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
108 unsigned int min_timer_period_us = 500;
109 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
111 static bool __read_mostly kvmclock_periodic_sync = true;
112 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
114 bool __read_mostly kvm_has_tsc_control;
115 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
116 u32 __read_mostly kvm_max_guest_tsc_khz;
117 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
118 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
119 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
120 u64 __read_mostly kvm_max_tsc_scaling_ratio;
121 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
122 u64 __read_mostly kvm_default_tsc_scaling_ratio;
123 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
125 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
126 static u32 __read_mostly tsc_tolerance_ppm = 250;
127 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
129 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
130 unsigned int __read_mostly lapic_timer_advance_ns = 0;
131 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
133 static bool __read_mostly vector_hashing = true;
134 module_param(vector_hashing, bool, S_IRUGO);
136 static bool __read_mostly backwards_tsc_observed = false;
138 #define KVM_NR_SHARED_MSRS 16
140 struct kvm_shared_msrs_global {
142 u32 msrs[KVM_NR_SHARED_MSRS];
145 struct kvm_shared_msrs {
146 struct user_return_notifier urn;
148 struct kvm_shared_msr_values {
151 } values[KVM_NR_SHARED_MSRS];
154 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
155 static struct kvm_shared_msrs __percpu *shared_msrs;
157 struct kvm_stats_debugfs_item debugfs_entries[] = {
158 { "pf_fixed", VCPU_STAT(pf_fixed) },
159 { "pf_guest", VCPU_STAT(pf_guest) },
160 { "tlb_flush", VCPU_STAT(tlb_flush) },
161 { "invlpg", VCPU_STAT(invlpg) },
162 { "exits", VCPU_STAT(exits) },
163 { "io_exits", VCPU_STAT(io_exits) },
164 { "mmio_exits", VCPU_STAT(mmio_exits) },
165 { "signal_exits", VCPU_STAT(signal_exits) },
166 { "irq_window", VCPU_STAT(irq_window_exits) },
167 { "nmi_window", VCPU_STAT(nmi_window_exits) },
168 { "halt_exits", VCPU_STAT(halt_exits) },
169 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
170 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
171 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
172 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
173 { "hypercalls", VCPU_STAT(hypercalls) },
174 { "request_irq", VCPU_STAT(request_irq_exits) },
175 { "irq_exits", VCPU_STAT(irq_exits) },
176 { "host_state_reload", VCPU_STAT(host_state_reload) },
177 { "efer_reload", VCPU_STAT(efer_reload) },
178 { "fpu_reload", VCPU_STAT(fpu_reload) },
179 { "insn_emulation", VCPU_STAT(insn_emulation) },
180 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
181 { "irq_injections", VCPU_STAT(irq_injections) },
182 { "nmi_injections", VCPU_STAT(nmi_injections) },
183 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
184 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
185 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
186 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
187 { "mmu_flooded", VM_STAT(mmu_flooded) },
188 { "mmu_recycled", VM_STAT(mmu_recycled) },
189 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
190 { "mmu_unsync", VM_STAT(mmu_unsync) },
191 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
192 { "largepages", VM_STAT(lpages) },
196 u64 __read_mostly host_xcr0;
198 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
200 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
203 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
204 vcpu->arch.apf.gfns[i] = ~0;
207 static void kvm_on_user_return(struct user_return_notifier *urn)
210 struct kvm_shared_msrs *locals
211 = container_of(urn, struct kvm_shared_msrs, urn);
212 struct kvm_shared_msr_values *values;
214 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
215 values = &locals->values[slot];
216 if (values->host != values->curr) {
217 wrmsrl(shared_msrs_global.msrs[slot], values->host);
218 values->curr = values->host;
221 locals->registered = false;
222 user_return_notifier_unregister(urn);
225 static void shared_msr_update(unsigned slot, u32 msr)
228 unsigned int cpu = smp_processor_id();
229 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
231 /* only read, and nobody should modify it at this time,
232 * so don't need lock */
233 if (slot >= shared_msrs_global.nr) {
234 printk(KERN_ERR "kvm: invalid MSR slot!");
237 rdmsrl_safe(msr, &value);
238 smsr->values[slot].host = value;
239 smsr->values[slot].curr = value;
242 void kvm_define_shared_msr(unsigned slot, u32 msr)
244 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
245 shared_msrs_global.msrs[slot] = msr;
246 if (slot >= shared_msrs_global.nr)
247 shared_msrs_global.nr = slot + 1;
249 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
251 static void kvm_shared_msr_cpu_online(void)
255 for (i = 0; i < shared_msrs_global.nr; ++i)
256 shared_msr_update(i, shared_msrs_global.msrs[i]);
259 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
261 unsigned int cpu = smp_processor_id();
262 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
265 if (((value ^ smsr->values[slot].curr) & mask) == 0)
267 smsr->values[slot].curr = value;
268 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
272 if (!smsr->registered) {
273 smsr->urn.on_user_return = kvm_on_user_return;
274 user_return_notifier_register(&smsr->urn);
275 smsr->registered = true;
279 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
281 static void drop_user_return_notifiers(void)
283 unsigned int cpu = smp_processor_id();
284 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
286 if (smsr->registered)
287 kvm_on_user_return(&smsr->urn);
290 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
292 return vcpu->arch.apic_base;
294 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
296 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
298 u64 old_state = vcpu->arch.apic_base &
299 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
300 u64 new_state = msr_info->data &
301 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
302 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
303 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
305 if (!msr_info->host_initiated &&
306 ((msr_info->data & reserved_bits) != 0 ||
307 new_state == X2APIC_ENABLE ||
308 (new_state == MSR_IA32_APICBASE_ENABLE &&
309 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
310 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
314 kvm_lapic_set_base(vcpu, msr_info->data);
317 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
319 asmlinkage __visible void kvm_spurious_fault(void)
321 /* Fault while not rebooting. We want the trace. */
324 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
326 #define EXCPT_BENIGN 0
327 #define EXCPT_CONTRIBUTORY 1
330 static int exception_class(int vector)
340 return EXCPT_CONTRIBUTORY;
347 #define EXCPT_FAULT 0
349 #define EXCPT_ABORT 2
350 #define EXCPT_INTERRUPT 3
352 static int exception_type(int vector)
356 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
357 return EXCPT_INTERRUPT;
361 /* #DB is trap, as instruction watchpoints are handled elsewhere */
362 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
365 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
368 /* Reserved exceptions will result in fault */
372 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
373 unsigned nr, bool has_error, u32 error_code,
379 kvm_make_request(KVM_REQ_EVENT, vcpu);
381 if (!vcpu->arch.exception.pending) {
383 if (has_error && !is_protmode(vcpu))
385 vcpu->arch.exception.pending = true;
386 vcpu->arch.exception.has_error_code = has_error;
387 vcpu->arch.exception.nr = nr;
388 vcpu->arch.exception.error_code = error_code;
389 vcpu->arch.exception.reinject = reinject;
393 /* to check exception */
394 prev_nr = vcpu->arch.exception.nr;
395 if (prev_nr == DF_VECTOR) {
396 /* triple fault -> shutdown */
397 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
400 class1 = exception_class(prev_nr);
401 class2 = exception_class(nr);
402 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
403 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
404 /* generate double fault per SDM Table 5-5 */
405 vcpu->arch.exception.pending = true;
406 vcpu->arch.exception.has_error_code = true;
407 vcpu->arch.exception.nr = DF_VECTOR;
408 vcpu->arch.exception.error_code = 0;
410 /* replace previous exception with a new one in a hope
411 that instruction re-execution will regenerate lost
416 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
418 kvm_multiple_exception(vcpu, nr, false, 0, false);
420 EXPORT_SYMBOL_GPL(kvm_queue_exception);
422 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
424 kvm_multiple_exception(vcpu, nr, false, 0, true);
426 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
428 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
431 kvm_inject_gp(vcpu, 0);
433 kvm_x86_ops->skip_emulated_instruction(vcpu);
435 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
437 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
439 ++vcpu->stat.pf_guest;
440 vcpu->arch.cr2 = fault->address;
441 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
443 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
445 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
447 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
448 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
450 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
452 return fault->nested_page_fault;
455 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
457 atomic_inc(&vcpu->arch.nmi_queued);
458 kvm_make_request(KVM_REQ_NMI, vcpu);
460 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
462 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
464 kvm_multiple_exception(vcpu, nr, true, error_code, false);
466 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
468 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
470 kvm_multiple_exception(vcpu, nr, true, error_code, true);
472 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
475 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
476 * a #GP and return false.
478 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
480 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
482 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
485 EXPORT_SYMBOL_GPL(kvm_require_cpl);
487 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
489 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
492 kvm_queue_exception(vcpu, UD_VECTOR);
495 EXPORT_SYMBOL_GPL(kvm_require_dr);
498 * This function will be used to read from the physical memory of the currently
499 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
500 * can read from guest physical or from the guest's guest physical memory.
502 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
503 gfn_t ngfn, void *data, int offset, int len,
506 struct x86_exception exception;
510 ngpa = gfn_to_gpa(ngfn);
511 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
512 if (real_gfn == UNMAPPED_GVA)
515 real_gfn = gpa_to_gfn(real_gfn);
517 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
519 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
521 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
522 void *data, int offset, int len, u32 access)
524 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
525 data, offset, len, access);
529 * Load the pae pdptrs. Return true is they are all valid.
531 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
533 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
534 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
537 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
539 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
540 offset * sizeof(u64), sizeof(pdpte),
541 PFERR_USER_MASK|PFERR_WRITE_MASK);
546 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
547 if ((pdpte[i] & PT_PRESENT_MASK) &&
549 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
556 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
557 __set_bit(VCPU_EXREG_PDPTR,
558 (unsigned long *)&vcpu->arch.regs_avail);
559 __set_bit(VCPU_EXREG_PDPTR,
560 (unsigned long *)&vcpu->arch.regs_dirty);
565 EXPORT_SYMBOL_GPL(load_pdptrs);
567 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
569 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
575 if (is_long_mode(vcpu) || !is_pae(vcpu))
578 if (!test_bit(VCPU_EXREG_PDPTR,
579 (unsigned long *)&vcpu->arch.regs_avail))
582 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
583 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
584 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
585 PFERR_USER_MASK | PFERR_WRITE_MASK);
588 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
594 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
596 unsigned long old_cr0 = kvm_read_cr0(vcpu);
597 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
602 if (cr0 & 0xffffffff00000000UL)
606 cr0 &= ~CR0_RESERVED_BITS;
608 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
611 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
614 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
616 if ((vcpu->arch.efer & EFER_LME)) {
621 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
626 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
631 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
634 kvm_x86_ops->set_cr0(vcpu, cr0);
636 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
637 kvm_clear_async_pf_completion_queue(vcpu);
638 kvm_async_pf_hash_reset(vcpu);
641 if ((cr0 ^ old_cr0) & update_bits)
642 kvm_mmu_reset_context(vcpu);
644 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
645 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
646 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
647 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
651 EXPORT_SYMBOL_GPL(kvm_set_cr0);
653 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
655 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
657 EXPORT_SYMBOL_GPL(kvm_lmsw);
659 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
661 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
662 !vcpu->guest_xcr0_loaded) {
663 /* kvm_set_xcr() also depends on this */
664 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
665 vcpu->guest_xcr0_loaded = 1;
669 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
671 if (vcpu->guest_xcr0_loaded) {
672 if (vcpu->arch.xcr0 != host_xcr0)
673 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
674 vcpu->guest_xcr0_loaded = 0;
678 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
681 u64 old_xcr0 = vcpu->arch.xcr0;
684 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
685 if (index != XCR_XFEATURE_ENABLED_MASK)
687 if (!(xcr0 & XFEATURE_MASK_FP))
689 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
693 * Do not allow the guest to set bits that we do not support
694 * saving. However, xcr0 bit 0 is always set, even if the
695 * emulated CPU does not support XSAVE (see fx_init).
697 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
698 if (xcr0 & ~valid_bits)
701 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
702 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
705 if (xcr0 & XFEATURE_MASK_AVX512) {
706 if (!(xcr0 & XFEATURE_MASK_YMM))
708 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
711 vcpu->arch.xcr0 = xcr0;
713 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
714 kvm_update_cpuid(vcpu);
718 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
720 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
721 __kvm_set_xcr(vcpu, index, xcr)) {
722 kvm_inject_gp(vcpu, 0);
727 EXPORT_SYMBOL_GPL(kvm_set_xcr);
729 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
731 unsigned long old_cr4 = kvm_read_cr4(vcpu);
732 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
733 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
735 if (cr4 & CR4_RESERVED_BITS)
738 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
741 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
744 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
747 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
750 if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
753 if (is_long_mode(vcpu)) {
754 if (!(cr4 & X86_CR4_PAE))
756 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
757 && ((cr4 ^ old_cr4) & pdptr_bits)
758 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
762 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
763 if (!guest_cpuid_has_pcid(vcpu))
766 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
767 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
771 if (kvm_x86_ops->set_cr4(vcpu, cr4))
774 if (((cr4 ^ old_cr4) & pdptr_bits) ||
775 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
776 kvm_mmu_reset_context(vcpu);
778 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
779 kvm_update_cpuid(vcpu);
783 EXPORT_SYMBOL_GPL(kvm_set_cr4);
785 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
788 cr3 &= ~CR3_PCID_INVD;
791 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
792 kvm_mmu_sync_roots(vcpu);
793 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
797 if (is_long_mode(vcpu)) {
798 if (cr3 & CR3_L_MODE_RESERVED_BITS)
800 } else if (is_pae(vcpu) && is_paging(vcpu) &&
801 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
804 vcpu->arch.cr3 = cr3;
805 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
806 kvm_mmu_new_cr3(vcpu);
809 EXPORT_SYMBOL_GPL(kvm_set_cr3);
811 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
813 if (cr8 & CR8_RESERVED_BITS)
815 if (lapic_in_kernel(vcpu))
816 kvm_lapic_set_tpr(vcpu, cr8);
818 vcpu->arch.cr8 = cr8;
821 EXPORT_SYMBOL_GPL(kvm_set_cr8);
823 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
825 if (lapic_in_kernel(vcpu))
826 return kvm_lapic_get_cr8(vcpu);
828 return vcpu->arch.cr8;
830 EXPORT_SYMBOL_GPL(kvm_get_cr8);
832 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
836 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
837 for (i = 0; i < KVM_NR_DB_REGS; i++)
838 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
839 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
843 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
845 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
846 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
849 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
853 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
854 dr7 = vcpu->arch.guest_debug_dr7;
856 dr7 = vcpu->arch.dr7;
857 kvm_x86_ops->set_dr7(vcpu, dr7);
858 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
859 if (dr7 & DR7_BP_EN_MASK)
860 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
863 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
865 u64 fixed = DR6_FIXED_1;
867 if (!guest_cpuid_has_rtm(vcpu))
872 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
876 vcpu->arch.db[dr] = val;
877 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
878 vcpu->arch.eff_db[dr] = val;
883 if (val & 0xffffffff00000000ULL)
885 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
886 kvm_update_dr6(vcpu);
891 if (val & 0xffffffff00000000ULL)
893 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
894 kvm_update_dr7(vcpu);
901 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
903 if (__kvm_set_dr(vcpu, dr, val)) {
904 kvm_inject_gp(vcpu, 0);
909 EXPORT_SYMBOL_GPL(kvm_set_dr);
911 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
915 *val = vcpu->arch.db[dr];
920 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
921 *val = vcpu->arch.dr6;
923 *val = kvm_x86_ops->get_dr6(vcpu);
928 *val = vcpu->arch.dr7;
933 EXPORT_SYMBOL_GPL(kvm_get_dr);
935 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
937 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
941 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
944 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
945 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
948 EXPORT_SYMBOL_GPL(kvm_rdpmc);
951 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
952 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
954 * This list is modified at module load time to reflect the
955 * capabilities of the host cpu. This capabilities test skips MSRs that are
956 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
957 * may depend on host virtualization features rather than host cpu features.
960 static u32 msrs_to_save[] = {
961 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
964 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
966 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
967 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
970 static unsigned num_msrs_to_save;
972 static u32 emulated_msrs[] = {
973 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
974 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
975 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
976 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
977 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
978 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
981 HV_X64_MSR_VP_RUNTIME,
983 HV_X64_MSR_STIMER0_CONFIG,
984 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
988 MSR_IA32_TSCDEADLINE,
989 MSR_IA32_MISC_ENABLE,
992 MSR_IA32_MCG_EXT_CTL,
996 static unsigned num_emulated_msrs;
998 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1000 if (efer & efer_reserved_bits)
1003 if (efer & EFER_FFXSR) {
1004 struct kvm_cpuid_entry2 *feat;
1006 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1007 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
1011 if (efer & EFER_SVME) {
1012 struct kvm_cpuid_entry2 *feat;
1014 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1015 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1021 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1023 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1025 u64 old_efer = vcpu->arch.efer;
1027 if (!kvm_valid_efer(vcpu, efer))
1031 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1035 efer |= vcpu->arch.efer & EFER_LMA;
1037 kvm_x86_ops->set_efer(vcpu, efer);
1039 /* Update reserved bits */
1040 if ((efer ^ old_efer) & EFER_NX)
1041 kvm_mmu_reset_context(vcpu);
1046 void kvm_enable_efer_bits(u64 mask)
1048 efer_reserved_bits &= ~mask;
1050 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1053 * Writes msr value into into the appropriate "register".
1054 * Returns 0 on success, non-0 otherwise.
1055 * Assumes vcpu_load() was already called.
1057 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1059 switch (msr->index) {
1062 case MSR_KERNEL_GS_BASE:
1065 if (is_noncanonical_address(msr->data))
1068 case MSR_IA32_SYSENTER_EIP:
1069 case MSR_IA32_SYSENTER_ESP:
1071 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1072 * non-canonical address is written on Intel but not on
1073 * AMD (which ignores the top 32-bits, because it does
1074 * not implement 64-bit SYSENTER).
1076 * 64-bit code should hence be able to write a non-canonical
1077 * value on AMD. Making the address canonical ensures that
1078 * vmentry does not fail on Intel after writing a non-canonical
1079 * value, and that something deterministic happens if the guest
1080 * invokes 64-bit SYSENTER.
1082 msr->data = get_canonical(msr->data);
1084 return kvm_x86_ops->set_msr(vcpu, msr);
1086 EXPORT_SYMBOL_GPL(kvm_set_msr);
1089 * Adapt set_msr() to msr_io()'s calling convention
1091 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1093 struct msr_data msr;
1097 msr.host_initiated = true;
1098 r = kvm_get_msr(vcpu, &msr);
1106 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1108 struct msr_data msr;
1112 msr.host_initiated = true;
1113 return kvm_set_msr(vcpu, &msr);
1116 #ifdef CONFIG_X86_64
1117 struct pvclock_gtod_data {
1120 struct { /* extract of a clocksource struct */
1132 static struct pvclock_gtod_data pvclock_gtod_data;
1134 static void update_pvclock_gtod(struct timekeeper *tk)
1136 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1139 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1141 write_seqcount_begin(&vdata->seq);
1143 /* copy pvclock gtod data */
1144 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1145 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1146 vdata->clock.mask = tk->tkr_mono.mask;
1147 vdata->clock.mult = tk->tkr_mono.mult;
1148 vdata->clock.shift = tk->tkr_mono.shift;
1150 vdata->boot_ns = boot_ns;
1151 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1153 write_seqcount_end(&vdata->seq);
1157 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1160 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1161 * vcpu_enter_guest. This function is only called from
1162 * the physical CPU that is running vcpu.
1164 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1167 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1171 struct pvclock_wall_clock wc;
1172 struct timespec64 boot;
1177 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1182 ++version; /* first time write, random junk */
1186 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1190 * The guest calculates current wall clock time by adding
1191 * system time (updated by kvm_guest_time_update below) to the
1192 * wall clock specified here. guest system time equals host
1193 * system time for us, thus we must fill in host boot time here.
1195 getboottime64(&boot);
1197 if (kvm->arch.kvmclock_offset) {
1198 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1199 boot = timespec64_sub(boot, ts);
1201 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1202 wc.nsec = boot.tv_nsec;
1203 wc.version = version;
1205 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1208 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1211 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1213 do_shl32_div32(dividend, divisor);
1217 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1218 s8 *pshift, u32 *pmultiplier)
1226 scaled64 = scaled_hz;
1227 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1232 tps32 = (uint32_t)tps64;
1233 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1234 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1242 *pmultiplier = div_frac(scaled64, tps32);
1244 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1245 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1248 #ifdef CONFIG_X86_64
1249 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1252 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1253 static unsigned long max_tsc_khz;
1255 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1257 u64 v = (u64)khz * (1000000 + ppm);
1262 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1266 /* Guest TSC same frequency as host TSC? */
1268 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1272 /* TSC scaling supported? */
1273 if (!kvm_has_tsc_control) {
1274 if (user_tsc_khz > tsc_khz) {
1275 vcpu->arch.tsc_catchup = 1;
1276 vcpu->arch.tsc_always_catchup = 1;
1279 WARN(1, "user requested TSC rate below hardware speed\n");
1284 /* TSC scaling required - calculate ratio */
1285 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1286 user_tsc_khz, tsc_khz);
1288 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1289 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1294 vcpu->arch.tsc_scaling_ratio = ratio;
1298 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1300 u32 thresh_lo, thresh_hi;
1301 int use_scaling = 0;
1303 /* tsc_khz can be zero if TSC calibration fails */
1304 if (user_tsc_khz == 0) {
1305 /* set tsc_scaling_ratio to a safe value */
1306 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1310 /* Compute a scale to convert nanoseconds in TSC cycles */
1311 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1312 &vcpu->arch.virtual_tsc_shift,
1313 &vcpu->arch.virtual_tsc_mult);
1314 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1317 * Compute the variation in TSC rate which is acceptable
1318 * within the range of tolerance and decide if the
1319 * rate being applied is within that bounds of the hardware
1320 * rate. If so, no scaling or compensation need be done.
1322 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1323 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1324 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1325 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1328 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1331 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1333 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1334 vcpu->arch.virtual_tsc_mult,
1335 vcpu->arch.virtual_tsc_shift);
1336 tsc += vcpu->arch.this_tsc_write;
1340 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1342 #ifdef CONFIG_X86_64
1344 struct kvm_arch *ka = &vcpu->kvm->arch;
1345 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1347 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1348 atomic_read(&vcpu->kvm->online_vcpus));
1351 * Once the masterclock is enabled, always perform request in
1352 * order to update it.
1354 * In order to enable masterclock, the host clocksource must be TSC
1355 * and the vcpus need to have matched TSCs. When that happens,
1356 * perform request to enable masterclock.
1358 if (ka->use_master_clock ||
1359 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1360 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1362 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1363 atomic_read(&vcpu->kvm->online_vcpus),
1364 ka->use_master_clock, gtod->clock.vclock_mode);
1368 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1370 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1371 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1375 * Multiply tsc by a fixed point number represented by ratio.
1377 * The most significant 64-N bits (mult) of ratio represent the
1378 * integral part of the fixed point number; the remaining N bits
1379 * (frac) represent the fractional part, ie. ratio represents a fixed
1380 * point number (mult + frac * 2^(-N)).
1382 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1384 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1386 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1389 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1392 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1394 if (ratio != kvm_default_tsc_scaling_ratio)
1395 _tsc = __scale_tsc(ratio, tsc);
1399 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1401 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1405 tsc = kvm_scale_tsc(vcpu, rdtsc());
1407 return target_tsc - tsc;
1410 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1412 return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
1414 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1416 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1418 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1419 vcpu->arch.tsc_offset = offset;
1422 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1424 struct kvm *kvm = vcpu->kvm;
1425 u64 offset, ns, elapsed;
1426 unsigned long flags;
1429 bool already_matched;
1430 u64 data = msr->data;
1432 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1433 offset = kvm_compute_tsc_offset(vcpu, data);
1434 ns = get_kernel_ns();
1435 elapsed = ns - kvm->arch.last_tsc_nsec;
1437 if (vcpu->arch.virtual_tsc_khz) {
1440 /* n.b - signed multiplication and division required */
1441 usdiff = data - kvm->arch.last_tsc_write;
1442 #ifdef CONFIG_X86_64
1443 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1445 /* do_div() only does unsigned */
1446 asm("1: idivl %[divisor]\n"
1447 "2: xor %%edx, %%edx\n"
1448 " movl $0, %[faulted]\n"
1450 ".section .fixup,\"ax\"\n"
1451 "4: movl $1, %[faulted]\n"
1455 _ASM_EXTABLE(1b, 4b)
1457 : "=A"(usdiff), [faulted] "=r" (faulted)
1458 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1461 do_div(elapsed, 1000);
1466 /* idivl overflow => difference is larger than USEC_PER_SEC */
1468 usdiff = USEC_PER_SEC;
1470 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1473 * Special case: TSC write with a small delta (1 second) of virtual
1474 * cycle time against real time is interpreted as an attempt to
1475 * synchronize the CPU.
1477 * For a reliable TSC, we can match TSC offsets, and for an unstable
1478 * TSC, we add elapsed time in this computation. We could let the
1479 * compensation code attempt to catch up if we fall behind, but
1480 * it's better to try to match offsets from the beginning.
1482 if (usdiff < USEC_PER_SEC &&
1483 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1484 if (!check_tsc_unstable()) {
1485 offset = kvm->arch.cur_tsc_offset;
1486 pr_debug("kvm: matched tsc offset for %llu\n", data);
1488 u64 delta = nsec_to_cycles(vcpu, elapsed);
1490 offset = kvm_compute_tsc_offset(vcpu, data);
1491 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1494 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1497 * We split periods of matched TSC writes into generations.
1498 * For each generation, we track the original measured
1499 * nanosecond time, offset, and write, so if TSCs are in
1500 * sync, we can match exact offset, and if not, we can match
1501 * exact software computation in compute_guest_tsc()
1503 * These values are tracked in kvm->arch.cur_xxx variables.
1505 kvm->arch.cur_tsc_generation++;
1506 kvm->arch.cur_tsc_nsec = ns;
1507 kvm->arch.cur_tsc_write = data;
1508 kvm->arch.cur_tsc_offset = offset;
1510 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1511 kvm->arch.cur_tsc_generation, data);
1515 * We also track th most recent recorded KHZ, write and time to
1516 * allow the matching interval to be extended at each write.
1518 kvm->arch.last_tsc_nsec = ns;
1519 kvm->arch.last_tsc_write = data;
1520 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1522 vcpu->arch.last_guest_tsc = data;
1524 /* Keep track of which generation this VCPU has synchronized to */
1525 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1526 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1527 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1529 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1530 update_ia32_tsc_adjust_msr(vcpu, offset);
1531 kvm_vcpu_write_tsc_offset(vcpu, offset);
1532 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1534 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1536 kvm->arch.nr_vcpus_matched_tsc = 0;
1537 } else if (!already_matched) {
1538 kvm->arch.nr_vcpus_matched_tsc++;
1541 kvm_track_tsc_matching(vcpu);
1542 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1545 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1547 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1550 kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1553 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1555 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1556 WARN_ON(adjustment < 0);
1557 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1558 kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1561 #ifdef CONFIG_X86_64
1563 static cycle_t read_tsc(void)
1565 cycle_t ret = (cycle_t)rdtsc_ordered();
1566 u64 last = pvclock_gtod_data.clock.cycle_last;
1568 if (likely(ret >= last))
1572 * GCC likes to generate cmov here, but this branch is extremely
1573 * predictable (it's just a function of time and the likely is
1574 * very likely) and there's a data dependence, so force GCC
1575 * to generate a branch instead. I don't barrier() because
1576 * we don't actually need a barrier, and if this function
1577 * ever gets inlined it will generate worse code.
1583 static inline u64 vgettsc(cycle_t *cycle_now)
1586 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1588 *cycle_now = read_tsc();
1590 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1591 return v * gtod->clock.mult;
1594 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1596 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1602 seq = read_seqcount_begin(>od->seq);
1603 mode = gtod->clock.vclock_mode;
1604 ns = gtod->nsec_base;
1605 ns += vgettsc(cycle_now);
1606 ns >>= gtod->clock.shift;
1607 ns += gtod->boot_ns;
1608 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1614 /* returns true if host is using tsc clocksource */
1615 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1617 /* checked again under seqlock below */
1618 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1621 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1627 * Assuming a stable TSC across physical CPUS, and a stable TSC
1628 * across virtual CPUs, the following condition is possible.
1629 * Each numbered line represents an event visible to both
1630 * CPUs at the next numbered event.
1632 * "timespecX" represents host monotonic time. "tscX" represents
1635 * VCPU0 on CPU0 | VCPU1 on CPU1
1637 * 1. read timespec0,tsc0
1638 * 2. | timespec1 = timespec0 + N
1640 * 3. transition to guest | transition to guest
1641 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1642 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1643 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1645 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1648 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1650 * - 0 < N - M => M < N
1652 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1653 * always the case (the difference between two distinct xtime instances
1654 * might be smaller then the difference between corresponding TSC reads,
1655 * when updating guest vcpus pvclock areas).
1657 * To avoid that problem, do not allow visibility of distinct
1658 * system_timestamp/tsc_timestamp values simultaneously: use a master
1659 * copy of host monotonic time values. Update that master copy
1662 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1666 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1668 #ifdef CONFIG_X86_64
1669 struct kvm_arch *ka = &kvm->arch;
1671 bool host_tsc_clocksource, vcpus_matched;
1673 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1674 atomic_read(&kvm->online_vcpus));
1677 * If the host uses TSC clock, then passthrough TSC as stable
1680 host_tsc_clocksource = kvm_get_time_and_clockread(
1681 &ka->master_kernel_ns,
1682 &ka->master_cycle_now);
1684 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1685 && !backwards_tsc_observed
1686 && !ka->boot_vcpu_runs_old_kvmclock;
1688 if (ka->use_master_clock)
1689 atomic_set(&kvm_guest_has_master_clock, 1);
1691 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1692 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1697 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1699 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1702 static void kvm_gen_update_masterclock(struct kvm *kvm)
1704 #ifdef CONFIG_X86_64
1706 struct kvm_vcpu *vcpu;
1707 struct kvm_arch *ka = &kvm->arch;
1709 spin_lock(&ka->pvclock_gtod_sync_lock);
1710 kvm_make_mclock_inprogress_request(kvm);
1711 /* no guest entries from this point */
1712 pvclock_update_vm_gtod_copy(kvm);
1714 kvm_for_each_vcpu(i, vcpu, kvm)
1715 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1717 /* guest entries allowed */
1718 kvm_for_each_vcpu(i, vcpu, kvm)
1719 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1721 spin_unlock(&ka->pvclock_gtod_sync_lock);
1725 static int kvm_guest_time_update(struct kvm_vcpu *v)
1727 unsigned long flags, tgt_tsc_khz;
1728 struct kvm_vcpu_arch *vcpu = &v->arch;
1729 struct kvm_arch *ka = &v->kvm->arch;
1731 u64 tsc_timestamp, host_tsc;
1732 struct pvclock_vcpu_time_info guest_hv_clock;
1734 bool use_master_clock;
1740 * If the host uses TSC clock, then passthrough TSC as stable
1743 spin_lock(&ka->pvclock_gtod_sync_lock);
1744 use_master_clock = ka->use_master_clock;
1745 if (use_master_clock) {
1746 host_tsc = ka->master_cycle_now;
1747 kernel_ns = ka->master_kernel_ns;
1749 spin_unlock(&ka->pvclock_gtod_sync_lock);
1751 /* Keep irq disabled to prevent changes to the clock */
1752 local_irq_save(flags);
1753 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1754 if (unlikely(tgt_tsc_khz == 0)) {
1755 local_irq_restore(flags);
1756 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1759 if (!use_master_clock) {
1761 kernel_ns = get_kernel_ns();
1764 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1767 * We may have to catch up the TSC to match elapsed wall clock
1768 * time for two reasons, even if kvmclock is used.
1769 * 1) CPU could have been running below the maximum TSC rate
1770 * 2) Broken TSC compensation resets the base at each VCPU
1771 * entry to avoid unknown leaps of TSC even when running
1772 * again on the same CPU. This may cause apparent elapsed
1773 * time to disappear, and the guest to stand still or run
1776 if (vcpu->tsc_catchup) {
1777 u64 tsc = compute_guest_tsc(v, kernel_ns);
1778 if (tsc > tsc_timestamp) {
1779 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1780 tsc_timestamp = tsc;
1784 local_irq_restore(flags);
1786 if (!vcpu->pv_time_enabled)
1789 if (kvm_has_tsc_control)
1790 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1792 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1793 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1794 &vcpu->hv_clock.tsc_shift,
1795 &vcpu->hv_clock.tsc_to_system_mul);
1796 vcpu->hw_tsc_khz = tgt_tsc_khz;
1799 /* With all the info we got, fill in the values */
1800 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1801 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1802 vcpu->last_guest_tsc = tsc_timestamp;
1804 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1805 &guest_hv_clock, sizeof(guest_hv_clock))))
1808 /* This VCPU is paused, but it's legal for a guest to read another
1809 * VCPU's kvmclock, so we really have to follow the specification where
1810 * it says that version is odd if data is being modified, and even after
1813 * Version field updates must be kept separate. This is because
1814 * kvm_write_guest_cached might use a "rep movs" instruction, and
1815 * writes within a string instruction are weakly ordered. So there
1816 * are three writes overall.
1818 * As a small optimization, only write the version field in the first
1819 * and third write. The vcpu->pv_time cache is still valid, because the
1820 * version field is the first in the struct.
1822 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1824 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1825 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1827 sizeof(vcpu->hv_clock.version));
1831 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1832 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1834 if (vcpu->pvclock_set_guest_stopped_request) {
1835 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1836 vcpu->pvclock_set_guest_stopped_request = false;
1839 /* If the host uses TSC clocksource, then it is stable */
1840 if (use_master_clock)
1841 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1843 vcpu->hv_clock.flags = pvclock_flags;
1845 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1847 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1849 sizeof(vcpu->hv_clock));
1853 vcpu->hv_clock.version++;
1854 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1856 sizeof(vcpu->hv_clock.version));
1861 * kvmclock updates which are isolated to a given vcpu, such as
1862 * vcpu->cpu migration, should not allow system_timestamp from
1863 * the rest of the vcpus to remain static. Otherwise ntp frequency
1864 * correction applies to one vcpu's system_timestamp but not
1867 * So in those cases, request a kvmclock update for all vcpus.
1868 * We need to rate-limit these requests though, as they can
1869 * considerably slow guests that have a large number of vcpus.
1870 * The time for a remote vcpu to update its kvmclock is bound
1871 * by the delay we use to rate-limit the updates.
1874 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1876 static void kvmclock_update_fn(struct work_struct *work)
1879 struct delayed_work *dwork = to_delayed_work(work);
1880 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1881 kvmclock_update_work);
1882 struct kvm *kvm = container_of(ka, struct kvm, arch);
1883 struct kvm_vcpu *vcpu;
1885 kvm_for_each_vcpu(i, vcpu, kvm) {
1886 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1887 kvm_vcpu_kick(vcpu);
1891 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1893 struct kvm *kvm = v->kvm;
1895 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1896 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1897 KVMCLOCK_UPDATE_DELAY);
1900 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1902 static void kvmclock_sync_fn(struct work_struct *work)
1904 struct delayed_work *dwork = to_delayed_work(work);
1905 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1906 kvmclock_sync_work);
1907 struct kvm *kvm = container_of(ka, struct kvm, arch);
1909 if (!kvmclock_periodic_sync)
1912 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1913 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1914 KVMCLOCK_SYNC_PERIOD);
1917 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1919 u64 mcg_cap = vcpu->arch.mcg_cap;
1920 unsigned bank_num = mcg_cap & 0xff;
1923 case MSR_IA32_MCG_STATUS:
1924 vcpu->arch.mcg_status = data;
1926 case MSR_IA32_MCG_CTL:
1927 if (!(mcg_cap & MCG_CTL_P))
1929 if (data != 0 && data != ~(u64)0)
1931 vcpu->arch.mcg_ctl = data;
1934 if (msr >= MSR_IA32_MC0_CTL &&
1935 msr < MSR_IA32_MCx_CTL(bank_num)) {
1936 u32 offset = msr - MSR_IA32_MC0_CTL;
1937 /* only 0 or all 1s can be written to IA32_MCi_CTL
1938 * some Linux kernels though clear bit 10 in bank 4 to
1939 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1940 * this to avoid an uncatched #GP in the guest
1942 if ((offset & 0x3) == 0 &&
1943 data != 0 && (data | (1 << 10)) != ~(u64)0)
1945 vcpu->arch.mce_banks[offset] = data;
1953 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1955 struct kvm *kvm = vcpu->kvm;
1956 int lm = is_long_mode(vcpu);
1957 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1958 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1959 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1960 : kvm->arch.xen_hvm_config.blob_size_32;
1961 u32 page_num = data & ~PAGE_MASK;
1962 u64 page_addr = data & PAGE_MASK;
1967 if (page_num >= blob_size)
1970 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1975 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1984 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1986 gpa_t gpa = data & ~0x3f;
1988 /* Bits 2:5 are reserved, Should be zero */
1992 vcpu->arch.apf.msr_val = data;
1994 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1995 kvm_clear_async_pf_completion_queue(vcpu);
1996 kvm_async_pf_hash_reset(vcpu);
2000 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2004 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2005 kvm_async_pf_wakeup_all(vcpu);
2009 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2011 vcpu->arch.pv_time_enabled = false;
2014 static void record_steal_time(struct kvm_vcpu *vcpu)
2016 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2019 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2020 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2023 if (vcpu->arch.st.steal.version & 1)
2024 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2026 vcpu->arch.st.steal.version += 1;
2028 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2029 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2033 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2034 vcpu->arch.st.last_steal;
2035 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2037 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2038 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2042 vcpu->arch.st.steal.version += 1;
2044 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2045 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2048 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2051 u32 msr = msr_info->index;
2052 u64 data = msr_info->data;
2055 case MSR_AMD64_NB_CFG:
2056 case MSR_IA32_UCODE_REV:
2057 case MSR_IA32_UCODE_WRITE:
2058 case MSR_VM_HSAVE_PA:
2059 case MSR_AMD64_PATCH_LOADER:
2060 case MSR_AMD64_BU_CFG2:
2064 return set_efer(vcpu, data);
2066 data &= ~(u64)0x40; /* ignore flush filter disable */
2067 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2068 data &= ~(u64)0x8; /* ignore TLB cache disable */
2069 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2071 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2076 case MSR_FAM10H_MMIO_CONF_BASE:
2078 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2083 case MSR_IA32_DEBUGCTLMSR:
2085 /* We support the non-activated case already */
2087 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2088 /* Values other than LBR and BTF are vendor-specific,
2089 thus reserved and should throw a #GP */
2092 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2095 case 0x200 ... 0x2ff:
2096 return kvm_mtrr_set_msr(vcpu, msr, data);
2097 case MSR_IA32_APICBASE:
2098 return kvm_set_apic_base(vcpu, msr_info);
2099 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2100 return kvm_x2apic_msr_write(vcpu, msr, data);
2101 case MSR_IA32_TSCDEADLINE:
2102 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2104 case MSR_IA32_TSC_ADJUST:
2105 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2106 if (!msr_info->host_initiated) {
2107 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2108 adjust_tsc_offset_guest(vcpu, adj);
2110 vcpu->arch.ia32_tsc_adjust_msr = data;
2113 case MSR_IA32_MISC_ENABLE:
2114 vcpu->arch.ia32_misc_enable_msr = data;
2116 case MSR_IA32_SMBASE:
2117 if (!msr_info->host_initiated)
2119 vcpu->arch.smbase = data;
2121 case MSR_KVM_WALL_CLOCK_NEW:
2122 case MSR_KVM_WALL_CLOCK:
2123 vcpu->kvm->arch.wall_clock = data;
2124 kvm_write_wall_clock(vcpu->kvm, data);
2126 case MSR_KVM_SYSTEM_TIME_NEW:
2127 case MSR_KVM_SYSTEM_TIME: {
2129 struct kvm_arch *ka = &vcpu->kvm->arch;
2131 kvmclock_reset(vcpu);
2133 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2134 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2136 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2137 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2140 ka->boot_vcpu_runs_old_kvmclock = tmp;
2143 vcpu->arch.time = data;
2144 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2146 /* we verify if the enable bit is set... */
2150 gpa_offset = data & ~(PAGE_MASK | 1);
2152 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2153 &vcpu->arch.pv_time, data & ~1ULL,
2154 sizeof(struct pvclock_vcpu_time_info)))
2155 vcpu->arch.pv_time_enabled = false;
2157 vcpu->arch.pv_time_enabled = true;
2161 case MSR_KVM_ASYNC_PF_EN:
2162 if (kvm_pv_enable_async_pf(vcpu, data))
2165 case MSR_KVM_STEAL_TIME:
2167 if (unlikely(!sched_info_on()))
2170 if (data & KVM_STEAL_RESERVED_MASK)
2173 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2174 data & KVM_STEAL_VALID_BITS,
2175 sizeof(struct kvm_steal_time)))
2178 vcpu->arch.st.msr_val = data;
2180 if (!(data & KVM_MSR_ENABLED))
2183 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2186 case MSR_KVM_PV_EOI_EN:
2187 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2191 case MSR_IA32_MCG_CTL:
2192 case MSR_IA32_MCG_STATUS:
2193 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2194 return set_msr_mce(vcpu, msr, data);
2196 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2197 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2198 pr = true; /* fall through */
2199 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2200 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2201 if (kvm_pmu_is_valid_msr(vcpu, msr))
2202 return kvm_pmu_set_msr(vcpu, msr_info);
2204 if (pr || data != 0)
2205 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2206 "0x%x data 0x%llx\n", msr, data);
2208 case MSR_K7_CLK_CTL:
2210 * Ignore all writes to this no longer documented MSR.
2211 * Writes are only relevant for old K7 processors,
2212 * all pre-dating SVM, but a recommended workaround from
2213 * AMD for these chips. It is possible to specify the
2214 * affected processor models on the command line, hence
2215 * the need to ignore the workaround.
2218 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2219 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2220 case HV_X64_MSR_CRASH_CTL:
2221 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2222 return kvm_hv_set_msr_common(vcpu, msr, data,
2223 msr_info->host_initiated);
2224 case MSR_IA32_BBL_CR_CTL3:
2225 /* Drop writes to this legacy MSR -- see rdmsr
2226 * counterpart for further detail.
2228 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2230 case MSR_AMD64_OSVW_ID_LENGTH:
2231 if (!guest_cpuid_has_osvw(vcpu))
2233 vcpu->arch.osvw.length = data;
2235 case MSR_AMD64_OSVW_STATUS:
2236 if (!guest_cpuid_has_osvw(vcpu))
2238 vcpu->arch.osvw.status = data;
2241 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2242 return xen_hvm_config(vcpu, data);
2243 if (kvm_pmu_is_valid_msr(vcpu, msr))
2244 return kvm_pmu_set_msr(vcpu, msr_info);
2246 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2250 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2257 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2261 * Reads an msr value (of 'msr_index') into 'pdata'.
2262 * Returns 0 on success, non-0 otherwise.
2263 * Assumes vcpu_load() was already called.
2265 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2267 return kvm_x86_ops->get_msr(vcpu, msr);
2269 EXPORT_SYMBOL_GPL(kvm_get_msr);
2271 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2274 u64 mcg_cap = vcpu->arch.mcg_cap;
2275 unsigned bank_num = mcg_cap & 0xff;
2278 case MSR_IA32_P5_MC_ADDR:
2279 case MSR_IA32_P5_MC_TYPE:
2282 case MSR_IA32_MCG_CAP:
2283 data = vcpu->arch.mcg_cap;
2285 case MSR_IA32_MCG_CTL:
2286 if (!(mcg_cap & MCG_CTL_P))
2288 data = vcpu->arch.mcg_ctl;
2290 case MSR_IA32_MCG_STATUS:
2291 data = vcpu->arch.mcg_status;
2294 if (msr >= MSR_IA32_MC0_CTL &&
2295 msr < MSR_IA32_MCx_CTL(bank_num)) {
2296 u32 offset = msr - MSR_IA32_MC0_CTL;
2297 data = vcpu->arch.mce_banks[offset];
2306 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2308 switch (msr_info->index) {
2309 case MSR_IA32_PLATFORM_ID:
2310 case MSR_IA32_EBL_CR_POWERON:
2311 case MSR_IA32_DEBUGCTLMSR:
2312 case MSR_IA32_LASTBRANCHFROMIP:
2313 case MSR_IA32_LASTBRANCHTOIP:
2314 case MSR_IA32_LASTINTFROMIP:
2315 case MSR_IA32_LASTINTTOIP:
2317 case MSR_K8_TSEG_ADDR:
2318 case MSR_K8_TSEG_MASK:
2320 case MSR_VM_HSAVE_PA:
2321 case MSR_K8_INT_PENDING_MSG:
2322 case MSR_AMD64_NB_CFG:
2323 case MSR_FAM10H_MMIO_CONF_BASE:
2324 case MSR_AMD64_BU_CFG2:
2325 case MSR_IA32_PERF_CTL:
2328 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2329 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2330 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2331 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2332 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2333 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2336 case MSR_IA32_UCODE_REV:
2337 msr_info->data = 0x100000000ULL;
2340 case 0x200 ... 0x2ff:
2341 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2342 case 0xcd: /* fsb frequency */
2346 * MSR_EBC_FREQUENCY_ID
2347 * Conservative value valid for even the basic CPU models.
2348 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2349 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2350 * and 266MHz for model 3, or 4. Set Core Clock
2351 * Frequency to System Bus Frequency Ratio to 1 (bits
2352 * 31:24) even though these are only valid for CPU
2353 * models > 2, however guests may end up dividing or
2354 * multiplying by zero otherwise.
2356 case MSR_EBC_FREQUENCY_ID:
2357 msr_info->data = 1 << 24;
2359 case MSR_IA32_APICBASE:
2360 msr_info->data = kvm_get_apic_base(vcpu);
2362 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2363 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2365 case MSR_IA32_TSCDEADLINE:
2366 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2368 case MSR_IA32_TSC_ADJUST:
2369 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2371 case MSR_IA32_MISC_ENABLE:
2372 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2374 case MSR_IA32_SMBASE:
2375 if (!msr_info->host_initiated)
2377 msr_info->data = vcpu->arch.smbase;
2379 case MSR_IA32_PERF_STATUS:
2380 /* TSC increment by tick */
2381 msr_info->data = 1000ULL;
2382 /* CPU multiplier */
2383 msr_info->data |= (((uint64_t)4ULL) << 40);
2386 msr_info->data = vcpu->arch.efer;
2388 case MSR_KVM_WALL_CLOCK:
2389 case MSR_KVM_WALL_CLOCK_NEW:
2390 msr_info->data = vcpu->kvm->arch.wall_clock;
2392 case MSR_KVM_SYSTEM_TIME:
2393 case MSR_KVM_SYSTEM_TIME_NEW:
2394 msr_info->data = vcpu->arch.time;
2396 case MSR_KVM_ASYNC_PF_EN:
2397 msr_info->data = vcpu->arch.apf.msr_val;
2399 case MSR_KVM_STEAL_TIME:
2400 msr_info->data = vcpu->arch.st.msr_val;
2402 case MSR_KVM_PV_EOI_EN:
2403 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2405 case MSR_IA32_P5_MC_ADDR:
2406 case MSR_IA32_P5_MC_TYPE:
2407 case MSR_IA32_MCG_CAP:
2408 case MSR_IA32_MCG_CTL:
2409 case MSR_IA32_MCG_STATUS:
2410 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2411 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2412 case MSR_K7_CLK_CTL:
2414 * Provide expected ramp-up count for K7. All other
2415 * are set to zero, indicating minimum divisors for
2418 * This prevents guest kernels on AMD host with CPU
2419 * type 6, model 8 and higher from exploding due to
2420 * the rdmsr failing.
2422 msr_info->data = 0x20000000;
2424 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2425 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2426 case HV_X64_MSR_CRASH_CTL:
2427 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2428 return kvm_hv_get_msr_common(vcpu,
2429 msr_info->index, &msr_info->data);
2431 case MSR_IA32_BBL_CR_CTL3:
2432 /* This legacy MSR exists but isn't fully documented in current
2433 * silicon. It is however accessed by winxp in very narrow
2434 * scenarios where it sets bit #19, itself documented as
2435 * a "reserved" bit. Best effort attempt to source coherent
2436 * read data here should the balance of the register be
2437 * interpreted by the guest:
2439 * L2 cache control register 3: 64GB range, 256KB size,
2440 * enabled, latency 0x1, configured
2442 msr_info->data = 0xbe702111;
2444 case MSR_AMD64_OSVW_ID_LENGTH:
2445 if (!guest_cpuid_has_osvw(vcpu))
2447 msr_info->data = vcpu->arch.osvw.length;
2449 case MSR_AMD64_OSVW_STATUS:
2450 if (!guest_cpuid_has_osvw(vcpu))
2452 msr_info->data = vcpu->arch.osvw.status;
2455 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2456 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2458 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2461 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2468 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2471 * Read or write a bunch of msrs. All parameters are kernel addresses.
2473 * @return number of msrs set successfully.
2475 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2476 struct kvm_msr_entry *entries,
2477 int (*do_msr)(struct kvm_vcpu *vcpu,
2478 unsigned index, u64 *data))
2482 idx = srcu_read_lock(&vcpu->kvm->srcu);
2483 for (i = 0; i < msrs->nmsrs; ++i)
2484 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2486 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2492 * Read or write a bunch of msrs. Parameters are user addresses.
2494 * @return number of msrs set successfully.
2496 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2497 int (*do_msr)(struct kvm_vcpu *vcpu,
2498 unsigned index, u64 *data),
2501 struct kvm_msrs msrs;
2502 struct kvm_msr_entry *entries;
2507 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2511 if (msrs.nmsrs >= MAX_IO_MSRS)
2514 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2515 entries = memdup_user(user_msrs->entries, size);
2516 if (IS_ERR(entries)) {
2517 r = PTR_ERR(entries);
2521 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2526 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2537 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2542 case KVM_CAP_IRQCHIP:
2544 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2545 case KVM_CAP_SET_TSS_ADDR:
2546 case KVM_CAP_EXT_CPUID:
2547 case KVM_CAP_EXT_EMUL_CPUID:
2548 case KVM_CAP_CLOCKSOURCE:
2550 case KVM_CAP_NOP_IO_DELAY:
2551 case KVM_CAP_MP_STATE:
2552 case KVM_CAP_SYNC_MMU:
2553 case KVM_CAP_USER_NMI:
2554 case KVM_CAP_REINJECT_CONTROL:
2555 case KVM_CAP_IRQ_INJECT_STATUS:
2556 case KVM_CAP_IOEVENTFD:
2557 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2559 case KVM_CAP_PIT_STATE2:
2560 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2561 case KVM_CAP_XEN_HVM:
2562 case KVM_CAP_ADJUST_CLOCK:
2563 case KVM_CAP_VCPU_EVENTS:
2564 case KVM_CAP_HYPERV:
2565 case KVM_CAP_HYPERV_VAPIC:
2566 case KVM_CAP_HYPERV_SPIN:
2567 case KVM_CAP_HYPERV_SYNIC:
2568 case KVM_CAP_PCI_SEGMENT:
2569 case KVM_CAP_DEBUGREGS:
2570 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2572 case KVM_CAP_ASYNC_PF:
2573 case KVM_CAP_GET_TSC_KHZ:
2574 case KVM_CAP_KVMCLOCK_CTRL:
2575 case KVM_CAP_READONLY_MEM:
2576 case KVM_CAP_HYPERV_TIME:
2577 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2578 case KVM_CAP_TSC_DEADLINE_TIMER:
2579 case KVM_CAP_ENABLE_CAP_VM:
2580 case KVM_CAP_DISABLE_QUIRKS:
2581 case KVM_CAP_SET_BOOT_CPU_ID:
2582 case KVM_CAP_SPLIT_IRQCHIP:
2583 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2584 case KVM_CAP_ASSIGN_DEV_IRQ:
2585 case KVM_CAP_PCI_2_3:
2589 case KVM_CAP_X86_SMM:
2590 /* SMBASE is usually relocated above 1M on modern chipsets,
2591 * and SMM handlers might indeed rely on 4G segment limits,
2592 * so do not report SMM to be available if real mode is
2593 * emulated via vm86 mode. Still, do not go to great lengths
2594 * to avoid userspace's usage of the feature, because it is a
2595 * fringe case that is not enabled except via specific settings
2596 * of the module parameters.
2598 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2600 case KVM_CAP_COALESCED_MMIO:
2601 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2604 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2606 case KVM_CAP_NR_VCPUS:
2607 r = KVM_SOFT_MAX_VCPUS;
2609 case KVM_CAP_MAX_VCPUS:
2612 case KVM_CAP_NR_MEMSLOTS:
2613 r = KVM_USER_MEM_SLOTS;
2615 case KVM_CAP_PV_MMU: /* obsolete */
2618 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2620 r = iommu_present(&pci_bus_type);
2624 r = KVM_MAX_MCE_BANKS;
2627 r = boot_cpu_has(X86_FEATURE_XSAVE);
2629 case KVM_CAP_TSC_CONTROL:
2630 r = kvm_has_tsc_control;
2632 case KVM_CAP_X2APIC_API:
2633 r = KVM_X2APIC_API_VALID_FLAGS;
2643 long kvm_arch_dev_ioctl(struct file *filp,
2644 unsigned int ioctl, unsigned long arg)
2646 void __user *argp = (void __user *)arg;
2650 case KVM_GET_MSR_INDEX_LIST: {
2651 struct kvm_msr_list __user *user_msr_list = argp;
2652 struct kvm_msr_list msr_list;
2656 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2659 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2660 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2663 if (n < msr_list.nmsrs)
2666 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2667 num_msrs_to_save * sizeof(u32)))
2669 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2671 num_emulated_msrs * sizeof(u32)))
2676 case KVM_GET_SUPPORTED_CPUID:
2677 case KVM_GET_EMULATED_CPUID: {
2678 struct kvm_cpuid2 __user *cpuid_arg = argp;
2679 struct kvm_cpuid2 cpuid;
2682 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2685 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2691 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2696 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2698 if (copy_to_user(argp, &kvm_mce_cap_supported,
2699 sizeof(kvm_mce_cap_supported)))
2711 static void wbinvd_ipi(void *garbage)
2716 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2718 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2721 static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
2723 set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
2726 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2728 /* Address WBINVD may be executed by guest */
2729 if (need_emulate_wbinvd(vcpu)) {
2730 if (kvm_x86_ops->has_wbinvd_exit())
2731 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2732 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2733 smp_call_function_single(vcpu->cpu,
2734 wbinvd_ipi, NULL, 1);
2737 kvm_x86_ops->vcpu_load(vcpu, cpu);
2739 /* Apply any externally detected TSC adjustments (due to suspend) */
2740 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2741 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2742 vcpu->arch.tsc_offset_adjustment = 0;
2743 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2746 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2747 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2748 rdtsc() - vcpu->arch.last_host_tsc;
2750 mark_tsc_unstable("KVM discovered backwards TSC");
2752 if (kvm_lapic_hv_timer_in_use(vcpu) &&
2753 kvm_x86_ops->set_hv_timer(vcpu,
2754 kvm_get_lapic_tscdeadline_msr(vcpu)))
2755 kvm_lapic_switch_to_sw_timer(vcpu);
2756 if (check_tsc_unstable()) {
2757 u64 offset = kvm_compute_tsc_offset(vcpu,
2758 vcpu->arch.last_guest_tsc);
2759 kvm_vcpu_write_tsc_offset(vcpu, offset);
2760 vcpu->arch.tsc_catchup = 1;
2763 * On a host with synchronized TSC, there is no need to update
2764 * kvmclock on vcpu->cpu migration
2766 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2767 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2768 if (vcpu->cpu != cpu)
2769 kvm_migrate_timers(vcpu);
2773 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2776 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2778 kvm_x86_ops->vcpu_put(vcpu);
2779 kvm_put_guest_fpu(vcpu);
2780 vcpu->arch.last_host_tsc = rdtsc();
2783 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2784 struct kvm_lapic_state *s)
2786 if (vcpu->arch.apicv_active)
2787 kvm_x86_ops->sync_pir_to_irr(vcpu);
2789 return kvm_apic_get_state(vcpu, s);
2792 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2793 struct kvm_lapic_state *s)
2797 r = kvm_apic_set_state(vcpu, s);
2800 update_cr8_intercept(vcpu);
2805 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2807 return (!lapic_in_kernel(vcpu) ||
2808 kvm_apic_accept_pic_intr(vcpu));
2812 * if userspace requested an interrupt window, check that the
2813 * interrupt window is open.
2815 * No need to exit to userspace if we already have an interrupt queued.
2817 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2819 return kvm_arch_interrupt_allowed(vcpu) &&
2820 !kvm_cpu_has_interrupt(vcpu) &&
2821 !kvm_event_needs_reinjection(vcpu) &&
2822 kvm_cpu_accept_dm_intr(vcpu);
2825 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2826 struct kvm_interrupt *irq)
2828 if (irq->irq >= KVM_NR_INTERRUPTS)
2831 if (!irqchip_in_kernel(vcpu->kvm)) {
2832 kvm_queue_interrupt(vcpu, irq->irq, false);
2833 kvm_make_request(KVM_REQ_EVENT, vcpu);
2838 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2839 * fail for in-kernel 8259.
2841 if (pic_in_kernel(vcpu->kvm))
2844 if (vcpu->arch.pending_external_vector != -1)
2847 vcpu->arch.pending_external_vector = irq->irq;
2848 kvm_make_request(KVM_REQ_EVENT, vcpu);
2852 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2854 kvm_inject_nmi(vcpu);
2859 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2861 kvm_make_request(KVM_REQ_SMI, vcpu);
2866 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2867 struct kvm_tpr_access_ctl *tac)
2871 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2875 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2879 unsigned bank_num = mcg_cap & 0xff, bank;
2882 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2884 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
2887 vcpu->arch.mcg_cap = mcg_cap;
2888 /* Init IA32_MCG_CTL to all 1s */
2889 if (mcg_cap & MCG_CTL_P)
2890 vcpu->arch.mcg_ctl = ~(u64)0;
2891 /* Init IA32_MCi_CTL to all 1s */
2892 for (bank = 0; bank < bank_num; bank++)
2893 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2895 if (kvm_x86_ops->setup_mce)
2896 kvm_x86_ops->setup_mce(vcpu);
2901 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2902 struct kvm_x86_mce *mce)
2904 u64 mcg_cap = vcpu->arch.mcg_cap;
2905 unsigned bank_num = mcg_cap & 0xff;
2906 u64 *banks = vcpu->arch.mce_banks;
2908 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2911 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2912 * reporting is disabled
2914 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2915 vcpu->arch.mcg_ctl != ~(u64)0)
2917 banks += 4 * mce->bank;
2919 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2920 * reporting is disabled for the bank
2922 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2924 if (mce->status & MCI_STATUS_UC) {
2925 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2926 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2927 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2930 if (banks[1] & MCI_STATUS_VAL)
2931 mce->status |= MCI_STATUS_OVER;
2932 banks[2] = mce->addr;
2933 banks[3] = mce->misc;
2934 vcpu->arch.mcg_status = mce->mcg_status;
2935 banks[1] = mce->status;
2936 kvm_queue_exception(vcpu, MC_VECTOR);
2937 } else if (!(banks[1] & MCI_STATUS_VAL)
2938 || !(banks[1] & MCI_STATUS_UC)) {
2939 if (banks[1] & MCI_STATUS_VAL)
2940 mce->status |= MCI_STATUS_OVER;
2941 banks[2] = mce->addr;
2942 banks[3] = mce->misc;
2943 banks[1] = mce->status;
2945 banks[1] |= MCI_STATUS_OVER;
2949 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2950 struct kvm_vcpu_events *events)
2953 events->exception.injected =
2954 vcpu->arch.exception.pending &&
2955 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2956 events->exception.nr = vcpu->arch.exception.nr;
2957 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2958 events->exception.pad = 0;
2959 events->exception.error_code = vcpu->arch.exception.error_code;
2961 events->interrupt.injected =
2962 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2963 events->interrupt.nr = vcpu->arch.interrupt.nr;
2964 events->interrupt.soft = 0;
2965 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2967 events->nmi.injected = vcpu->arch.nmi_injected;
2968 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2969 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2970 events->nmi.pad = 0;
2972 events->sipi_vector = 0; /* never valid when reporting to user space */
2974 events->smi.smm = is_smm(vcpu);
2975 events->smi.pending = vcpu->arch.smi_pending;
2976 events->smi.smm_inside_nmi =
2977 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2978 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2980 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2981 | KVM_VCPUEVENT_VALID_SHADOW
2982 | KVM_VCPUEVENT_VALID_SMM);
2983 memset(&events->reserved, 0, sizeof(events->reserved));
2986 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2987 struct kvm_vcpu_events *events)
2989 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2990 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2991 | KVM_VCPUEVENT_VALID_SHADOW
2992 | KVM_VCPUEVENT_VALID_SMM))
2995 if (events->exception.injected &&
2996 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3000 vcpu->arch.exception.pending = events->exception.injected;
3001 vcpu->arch.exception.nr = events->exception.nr;
3002 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3003 vcpu->arch.exception.error_code = events->exception.error_code;
3005 vcpu->arch.interrupt.pending = events->interrupt.injected;
3006 vcpu->arch.interrupt.nr = events->interrupt.nr;
3007 vcpu->arch.interrupt.soft = events->interrupt.soft;
3008 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3009 kvm_x86_ops->set_interrupt_shadow(vcpu,
3010 events->interrupt.shadow);
3012 vcpu->arch.nmi_injected = events->nmi.injected;
3013 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3014 vcpu->arch.nmi_pending = events->nmi.pending;
3015 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3017 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3018 lapic_in_kernel(vcpu))
3019 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3021 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3022 if (events->smi.smm)
3023 vcpu->arch.hflags |= HF_SMM_MASK;
3025 vcpu->arch.hflags &= ~HF_SMM_MASK;
3026 vcpu->arch.smi_pending = events->smi.pending;
3027 if (events->smi.smm_inside_nmi)
3028 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3030 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3031 if (lapic_in_kernel(vcpu)) {
3032 if (events->smi.latched_init)
3033 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3035 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3039 kvm_make_request(KVM_REQ_EVENT, vcpu);
3044 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3045 struct kvm_debugregs *dbgregs)
3049 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3050 kvm_get_dr(vcpu, 6, &val);
3052 dbgregs->dr7 = vcpu->arch.dr7;
3054 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3057 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3058 struct kvm_debugregs *dbgregs)
3063 if (dbgregs->dr6 & ~0xffffffffull)
3065 if (dbgregs->dr7 & ~0xffffffffull)
3068 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3069 kvm_update_dr0123(vcpu);
3070 vcpu->arch.dr6 = dbgregs->dr6;
3071 kvm_update_dr6(vcpu);
3072 vcpu->arch.dr7 = dbgregs->dr7;
3073 kvm_update_dr7(vcpu);
3078 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3080 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3082 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3083 u64 xstate_bv = xsave->header.xfeatures;
3087 * Copy legacy XSAVE area, to avoid complications with CPUID
3088 * leaves 0 and 1 in the loop below.
3090 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3093 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3096 * Copy each region from the possibly compacted offset to the
3097 * non-compacted offset.
3099 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3101 u64 feature = valid & -valid;
3102 int index = fls64(feature) - 1;
3103 void *src = get_xsave_addr(xsave, feature);
3106 u32 size, offset, ecx, edx;
3107 cpuid_count(XSTATE_CPUID, index,
3108 &size, &offset, &ecx, &edx);
3109 memcpy(dest + offset, src, size);
3116 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3118 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3119 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3123 * Copy legacy XSAVE area, to avoid complications with CPUID
3124 * leaves 0 and 1 in the loop below.
3126 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3128 /* Set XSTATE_BV and possibly XCOMP_BV. */
3129 xsave->header.xfeatures = xstate_bv;
3130 if (boot_cpu_has(X86_FEATURE_XSAVES))
3131 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3134 * Copy each region from the non-compacted offset to the
3135 * possibly compacted offset.
3137 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3139 u64 feature = valid & -valid;
3140 int index = fls64(feature) - 1;
3141 void *dest = get_xsave_addr(xsave, feature);
3144 u32 size, offset, ecx, edx;
3145 cpuid_count(XSTATE_CPUID, index,
3146 &size, &offset, &ecx, &edx);
3147 memcpy(dest, src + offset, size);
3154 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3155 struct kvm_xsave *guest_xsave)
3157 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3158 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3159 fill_xsave((u8 *) guest_xsave->region, vcpu);
3161 memcpy(guest_xsave->region,
3162 &vcpu->arch.guest_fpu.state.fxsave,
3163 sizeof(struct fxregs_state));
3164 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3165 XFEATURE_MASK_FPSSE;
3169 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3170 struct kvm_xsave *guest_xsave)
3173 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3175 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3177 * Here we allow setting states that are not present in
3178 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3179 * with old userspace.
3181 if (xstate_bv & ~kvm_supported_xcr0())
3183 load_xsave(vcpu, (u8 *)guest_xsave->region);
3185 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3187 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3188 guest_xsave->region, sizeof(struct fxregs_state));
3193 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3194 struct kvm_xcrs *guest_xcrs)
3196 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3197 guest_xcrs->nr_xcrs = 0;
3201 guest_xcrs->nr_xcrs = 1;
3202 guest_xcrs->flags = 0;
3203 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3204 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3207 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3208 struct kvm_xcrs *guest_xcrs)
3212 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3215 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3218 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3219 /* Only support XCR0 currently */
3220 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3221 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3222 guest_xcrs->xcrs[i].value);
3231 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3232 * stopped by the hypervisor. This function will be called from the host only.
3233 * EINVAL is returned when the host attempts to set the flag for a guest that
3234 * does not support pv clocks.
3236 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3238 if (!vcpu->arch.pv_time_enabled)
3240 vcpu->arch.pvclock_set_guest_stopped_request = true;
3241 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3245 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3246 struct kvm_enable_cap *cap)
3252 case KVM_CAP_HYPERV_SYNIC:
3253 return kvm_hv_activate_synic(vcpu);
3259 long kvm_arch_vcpu_ioctl(struct file *filp,
3260 unsigned int ioctl, unsigned long arg)
3262 struct kvm_vcpu *vcpu = filp->private_data;
3263 void __user *argp = (void __user *)arg;
3266 struct kvm_lapic_state *lapic;
3267 struct kvm_xsave *xsave;
3268 struct kvm_xcrs *xcrs;
3274 case KVM_GET_LAPIC: {
3276 if (!lapic_in_kernel(vcpu))
3278 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3283 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3287 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3292 case KVM_SET_LAPIC: {
3294 if (!lapic_in_kernel(vcpu))
3296 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3297 if (IS_ERR(u.lapic))
3298 return PTR_ERR(u.lapic);
3300 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3303 case KVM_INTERRUPT: {
3304 struct kvm_interrupt irq;
3307 if (copy_from_user(&irq, argp, sizeof irq))
3309 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3313 r = kvm_vcpu_ioctl_nmi(vcpu);
3317 r = kvm_vcpu_ioctl_smi(vcpu);
3320 case KVM_SET_CPUID: {
3321 struct kvm_cpuid __user *cpuid_arg = argp;
3322 struct kvm_cpuid cpuid;
3325 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3327 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3330 case KVM_SET_CPUID2: {
3331 struct kvm_cpuid2 __user *cpuid_arg = argp;
3332 struct kvm_cpuid2 cpuid;
3335 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3337 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3338 cpuid_arg->entries);
3341 case KVM_GET_CPUID2: {
3342 struct kvm_cpuid2 __user *cpuid_arg = argp;
3343 struct kvm_cpuid2 cpuid;
3346 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3348 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3349 cpuid_arg->entries);
3353 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3359 r = msr_io(vcpu, argp, do_get_msr, 1);
3362 r = msr_io(vcpu, argp, do_set_msr, 0);
3364 case KVM_TPR_ACCESS_REPORTING: {
3365 struct kvm_tpr_access_ctl tac;
3368 if (copy_from_user(&tac, argp, sizeof tac))
3370 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3374 if (copy_to_user(argp, &tac, sizeof tac))
3379 case KVM_SET_VAPIC_ADDR: {
3380 struct kvm_vapic_addr va;
3383 if (!lapic_in_kernel(vcpu))
3386 if (copy_from_user(&va, argp, sizeof va))
3388 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3391 case KVM_X86_SETUP_MCE: {
3395 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3397 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3400 case KVM_X86_SET_MCE: {
3401 struct kvm_x86_mce mce;
3404 if (copy_from_user(&mce, argp, sizeof mce))
3406 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3409 case KVM_GET_VCPU_EVENTS: {
3410 struct kvm_vcpu_events events;
3412 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3415 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3420 case KVM_SET_VCPU_EVENTS: {
3421 struct kvm_vcpu_events events;
3424 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3427 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3430 case KVM_GET_DEBUGREGS: {
3431 struct kvm_debugregs dbgregs;
3433 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3436 if (copy_to_user(argp, &dbgregs,
3437 sizeof(struct kvm_debugregs)))
3442 case KVM_SET_DEBUGREGS: {
3443 struct kvm_debugregs dbgregs;
3446 if (copy_from_user(&dbgregs, argp,
3447 sizeof(struct kvm_debugregs)))
3450 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3453 case KVM_GET_XSAVE: {
3454 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3459 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3462 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3467 case KVM_SET_XSAVE: {
3468 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3469 if (IS_ERR(u.xsave))
3470 return PTR_ERR(u.xsave);
3472 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3475 case KVM_GET_XCRS: {
3476 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3481 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3484 if (copy_to_user(argp, u.xcrs,
3485 sizeof(struct kvm_xcrs)))
3490 case KVM_SET_XCRS: {
3491 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3493 return PTR_ERR(u.xcrs);
3495 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3498 case KVM_SET_TSC_KHZ: {
3502 user_tsc_khz = (u32)arg;
3504 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3507 if (user_tsc_khz == 0)
3508 user_tsc_khz = tsc_khz;
3510 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3515 case KVM_GET_TSC_KHZ: {
3516 r = vcpu->arch.virtual_tsc_khz;
3519 case KVM_KVMCLOCK_CTRL: {
3520 r = kvm_set_guest_paused(vcpu);
3523 case KVM_ENABLE_CAP: {
3524 struct kvm_enable_cap cap;
3527 if (copy_from_user(&cap, argp, sizeof(cap)))
3529 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3540 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3542 return VM_FAULT_SIGBUS;
3545 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3549 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3551 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3555 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3558 kvm->arch.ept_identity_map_addr = ident_addr;
3562 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3563 u32 kvm_nr_mmu_pages)
3565 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3568 mutex_lock(&kvm->slots_lock);
3570 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3571 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3573 mutex_unlock(&kvm->slots_lock);
3577 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3579 return kvm->arch.n_max_mmu_pages;
3582 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3587 switch (chip->chip_id) {
3588 case KVM_IRQCHIP_PIC_MASTER:
3589 memcpy(&chip->chip.pic,
3590 &pic_irqchip(kvm)->pics[0],
3591 sizeof(struct kvm_pic_state));
3593 case KVM_IRQCHIP_PIC_SLAVE:
3594 memcpy(&chip->chip.pic,
3595 &pic_irqchip(kvm)->pics[1],
3596 sizeof(struct kvm_pic_state));
3598 case KVM_IRQCHIP_IOAPIC:
3599 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3608 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3613 switch (chip->chip_id) {
3614 case KVM_IRQCHIP_PIC_MASTER:
3615 spin_lock(&pic_irqchip(kvm)->lock);
3616 memcpy(&pic_irqchip(kvm)->pics[0],
3618 sizeof(struct kvm_pic_state));
3619 spin_unlock(&pic_irqchip(kvm)->lock);
3621 case KVM_IRQCHIP_PIC_SLAVE:
3622 spin_lock(&pic_irqchip(kvm)->lock);
3623 memcpy(&pic_irqchip(kvm)->pics[1],
3625 sizeof(struct kvm_pic_state));
3626 spin_unlock(&pic_irqchip(kvm)->lock);
3628 case KVM_IRQCHIP_IOAPIC:
3629 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3635 kvm_pic_update_irq(pic_irqchip(kvm));
3639 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3641 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3643 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3645 mutex_lock(&kps->lock);
3646 memcpy(ps, &kps->channels, sizeof(*ps));
3647 mutex_unlock(&kps->lock);
3651 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3654 struct kvm_pit *pit = kvm->arch.vpit;
3656 mutex_lock(&pit->pit_state.lock);
3657 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3658 for (i = 0; i < 3; i++)
3659 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3660 mutex_unlock(&pit->pit_state.lock);
3664 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3666 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3667 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3668 sizeof(ps->channels));
3669 ps->flags = kvm->arch.vpit->pit_state.flags;
3670 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3671 memset(&ps->reserved, 0, sizeof(ps->reserved));
3675 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3679 u32 prev_legacy, cur_legacy;
3680 struct kvm_pit *pit = kvm->arch.vpit;
3682 mutex_lock(&pit->pit_state.lock);
3683 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3684 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3685 if (!prev_legacy && cur_legacy)
3687 memcpy(&pit->pit_state.channels, &ps->channels,
3688 sizeof(pit->pit_state.channels));
3689 pit->pit_state.flags = ps->flags;
3690 for (i = 0; i < 3; i++)
3691 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3693 mutex_unlock(&pit->pit_state.lock);
3697 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3698 struct kvm_reinject_control *control)
3700 struct kvm_pit *pit = kvm->arch.vpit;
3705 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3706 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3707 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3709 mutex_lock(&pit->pit_state.lock);
3710 kvm_pit_set_reinject(pit, control->pit_reinject);
3711 mutex_unlock(&pit->pit_state.lock);
3717 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3718 * @kvm: kvm instance
3719 * @log: slot id and address to which we copy the log
3721 * Steps 1-4 below provide general overview of dirty page logging. See
3722 * kvm_get_dirty_log_protect() function description for additional details.
3724 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3725 * always flush the TLB (step 4) even if previous step failed and the dirty
3726 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3727 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3728 * writes will be marked dirty for next log read.
3730 * 1. Take a snapshot of the bit and clear it if needed.
3731 * 2. Write protect the corresponding page.
3732 * 3. Copy the snapshot to the userspace.
3733 * 4. Flush TLB's if needed.
3735 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3737 bool is_dirty = false;
3740 mutex_lock(&kvm->slots_lock);
3743 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3745 if (kvm_x86_ops->flush_log_dirty)
3746 kvm_x86_ops->flush_log_dirty(kvm);
3748 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3751 * All the TLBs can be flushed out of mmu lock, see the comments in
3752 * kvm_mmu_slot_remove_write_access().
3754 lockdep_assert_held(&kvm->slots_lock);
3756 kvm_flush_remote_tlbs(kvm);
3758 mutex_unlock(&kvm->slots_lock);
3762 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3765 if (!irqchip_in_kernel(kvm))
3768 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3769 irq_event->irq, irq_event->level,
3774 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3775 struct kvm_enable_cap *cap)
3783 case KVM_CAP_DISABLE_QUIRKS:
3784 kvm->arch.disabled_quirks = cap->args[0];
3787 case KVM_CAP_SPLIT_IRQCHIP: {
3788 mutex_lock(&kvm->lock);
3790 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3791 goto split_irqchip_unlock;
3793 if (irqchip_in_kernel(kvm))
3794 goto split_irqchip_unlock;
3795 if (kvm->created_vcpus)
3796 goto split_irqchip_unlock;
3797 r = kvm_setup_empty_irq_routing(kvm);
3799 goto split_irqchip_unlock;
3800 /* Pairs with irqchip_in_kernel. */
3802 kvm->arch.irqchip_split = true;
3803 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3805 split_irqchip_unlock:
3806 mutex_unlock(&kvm->lock);
3809 case KVM_CAP_X2APIC_API:
3811 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3814 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3815 kvm->arch.x2apic_format = true;
3816 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3817 kvm->arch.x2apic_broadcast_quirk_disabled = true;
3828 long kvm_arch_vm_ioctl(struct file *filp,
3829 unsigned int ioctl, unsigned long arg)
3831 struct kvm *kvm = filp->private_data;
3832 void __user *argp = (void __user *)arg;
3835 * This union makes it completely explicit to gcc-3.x
3836 * that these two variables' stack usage should be
3837 * combined, not added together.
3840 struct kvm_pit_state ps;
3841 struct kvm_pit_state2 ps2;
3842 struct kvm_pit_config pit_config;
3846 case KVM_SET_TSS_ADDR:
3847 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3849 case KVM_SET_IDENTITY_MAP_ADDR: {
3853 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3855 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3858 case KVM_SET_NR_MMU_PAGES:
3859 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3861 case KVM_GET_NR_MMU_PAGES:
3862 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3864 case KVM_CREATE_IRQCHIP: {
3865 struct kvm_pic *vpic;
3867 mutex_lock(&kvm->lock);
3870 goto create_irqchip_unlock;
3872 if (kvm->created_vcpus)
3873 goto create_irqchip_unlock;
3875 vpic = kvm_create_pic(kvm);
3877 r = kvm_ioapic_init(kvm);
3879 mutex_lock(&kvm->slots_lock);
3880 kvm_destroy_pic(vpic);
3881 mutex_unlock(&kvm->slots_lock);
3882 goto create_irqchip_unlock;
3885 goto create_irqchip_unlock;
3886 r = kvm_setup_default_irq_routing(kvm);
3888 mutex_lock(&kvm->slots_lock);
3889 mutex_lock(&kvm->irq_lock);
3890 kvm_ioapic_destroy(kvm);
3891 kvm_destroy_pic(vpic);
3892 mutex_unlock(&kvm->irq_lock);
3893 mutex_unlock(&kvm->slots_lock);
3894 goto create_irqchip_unlock;
3896 /* Write kvm->irq_routing before kvm->arch.vpic. */
3898 kvm->arch.vpic = vpic;
3899 create_irqchip_unlock:
3900 mutex_unlock(&kvm->lock);
3903 case KVM_CREATE_PIT:
3904 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3906 case KVM_CREATE_PIT2:
3908 if (copy_from_user(&u.pit_config, argp,
3909 sizeof(struct kvm_pit_config)))
3912 mutex_lock(&kvm->lock);
3915 goto create_pit_unlock;
3917 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3921 mutex_unlock(&kvm->lock);
3923 case KVM_GET_IRQCHIP: {
3924 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3925 struct kvm_irqchip *chip;
3927 chip = memdup_user(argp, sizeof(*chip));
3934 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3935 goto get_irqchip_out;
3936 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3938 goto get_irqchip_out;
3940 if (copy_to_user(argp, chip, sizeof *chip))
3941 goto get_irqchip_out;
3947 case KVM_SET_IRQCHIP: {
3948 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3949 struct kvm_irqchip *chip;
3951 chip = memdup_user(argp, sizeof(*chip));
3958 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3959 goto set_irqchip_out;
3960 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3962 goto set_irqchip_out;
3970 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3973 if (!kvm->arch.vpit)
3975 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3979 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3986 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3989 if (!kvm->arch.vpit)
3991 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3994 case KVM_GET_PIT2: {
3996 if (!kvm->arch.vpit)
3998 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4002 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4007 case KVM_SET_PIT2: {
4009 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4012 if (!kvm->arch.vpit)
4014 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4017 case KVM_REINJECT_CONTROL: {
4018 struct kvm_reinject_control control;
4020 if (copy_from_user(&control, argp, sizeof(control)))
4022 r = kvm_vm_ioctl_reinject(kvm, &control);
4025 case KVM_SET_BOOT_CPU_ID:
4027 mutex_lock(&kvm->lock);
4028 if (kvm->created_vcpus)
4031 kvm->arch.bsp_vcpu_id = arg;
4032 mutex_unlock(&kvm->lock);
4034 case KVM_XEN_HVM_CONFIG: {
4036 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4037 sizeof(struct kvm_xen_hvm_config)))
4040 if (kvm->arch.xen_hvm_config.flags)
4045 case KVM_SET_CLOCK: {
4046 struct kvm_clock_data user_ns;
4051 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4059 local_irq_disable();
4060 now_ns = get_kernel_ns();
4061 delta = user_ns.clock - now_ns;
4063 kvm->arch.kvmclock_offset = delta;
4064 kvm_gen_update_masterclock(kvm);
4067 case KVM_GET_CLOCK: {
4068 struct kvm_clock_data user_ns;
4071 local_irq_disable();
4072 now_ns = get_kernel_ns();
4073 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4076 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4079 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4084 case KVM_ENABLE_CAP: {
4085 struct kvm_enable_cap cap;
4088 if (copy_from_user(&cap, argp, sizeof(cap)))
4090 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4094 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4100 static void kvm_init_msr_list(void)
4105 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4106 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4110 * Even MSRs that are valid in the host may not be exposed
4111 * to the guests in some cases.
4113 switch (msrs_to_save[i]) {
4114 case MSR_IA32_BNDCFGS:
4115 if (!kvm_x86_ops->mpx_supported())
4119 if (!kvm_x86_ops->rdtscp_supported())
4127 msrs_to_save[j] = msrs_to_save[i];
4130 num_msrs_to_save = j;
4132 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4133 switch (emulated_msrs[i]) {
4134 case MSR_IA32_SMBASE:
4135 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4143 emulated_msrs[j] = emulated_msrs[i];
4146 num_emulated_msrs = j;
4149 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4157 if (!(lapic_in_kernel(vcpu) &&
4158 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4159 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4170 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4177 if (!(lapic_in_kernel(vcpu) &&
4178 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4180 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4182 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4192 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4193 struct kvm_segment *var, int seg)
4195 kvm_x86_ops->set_segment(vcpu, var, seg);
4198 void kvm_get_segment(struct kvm_vcpu *vcpu,
4199 struct kvm_segment *var, int seg)
4201 kvm_x86_ops->get_segment(vcpu, var, seg);
4204 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4205 struct x86_exception *exception)
4209 BUG_ON(!mmu_is_nested(vcpu));
4211 /* NPT walks are always user-walks */
4212 access |= PFERR_USER_MASK;
4213 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4218 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4219 struct x86_exception *exception)
4221 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4222 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4225 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4226 struct x86_exception *exception)
4228 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4229 access |= PFERR_FETCH_MASK;
4230 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4233 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4234 struct x86_exception *exception)
4236 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4237 access |= PFERR_WRITE_MASK;
4238 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4241 /* uses this to access any guest's mapped memory without checking CPL */
4242 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4243 struct x86_exception *exception)
4245 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4248 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4249 struct kvm_vcpu *vcpu, u32 access,
4250 struct x86_exception *exception)
4253 int r = X86EMUL_CONTINUE;
4256 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4258 unsigned offset = addr & (PAGE_SIZE-1);
4259 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4262 if (gpa == UNMAPPED_GVA)
4263 return X86EMUL_PROPAGATE_FAULT;
4264 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4267 r = X86EMUL_IO_NEEDED;
4279 /* used for instruction fetching */
4280 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4281 gva_t addr, void *val, unsigned int bytes,
4282 struct x86_exception *exception)
4284 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4285 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4289 /* Inline kvm_read_guest_virt_helper for speed. */
4290 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4292 if (unlikely(gpa == UNMAPPED_GVA))
4293 return X86EMUL_PROPAGATE_FAULT;
4295 offset = addr & (PAGE_SIZE-1);
4296 if (WARN_ON(offset + bytes > PAGE_SIZE))
4297 bytes = (unsigned)PAGE_SIZE - offset;
4298 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4300 if (unlikely(ret < 0))
4301 return X86EMUL_IO_NEEDED;
4303 return X86EMUL_CONTINUE;
4306 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4307 gva_t addr, void *val, unsigned int bytes,
4308 struct x86_exception *exception)
4310 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4311 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4313 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4316 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4318 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4319 gva_t addr, void *val, unsigned int bytes,
4320 struct x86_exception *exception)
4322 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4323 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4326 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4327 unsigned long addr, void *val, unsigned int bytes)
4329 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4330 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4332 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4335 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4336 gva_t addr, void *val,
4338 struct x86_exception *exception)
4340 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4342 int r = X86EMUL_CONTINUE;
4345 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4348 unsigned offset = addr & (PAGE_SIZE-1);
4349 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4352 if (gpa == UNMAPPED_GVA)
4353 return X86EMUL_PROPAGATE_FAULT;
4354 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4356 r = X86EMUL_IO_NEEDED;
4367 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4369 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4370 gpa_t *gpa, struct x86_exception *exception,
4373 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4374 | (write ? PFERR_WRITE_MASK : 0);
4377 * currently PKRU is only applied to ept enabled guest so
4378 * there is no pkey in EPT page table for L1 guest or EPT
4379 * shadow page table for L2 guest.
4381 if (vcpu_match_mmio_gva(vcpu, gva)
4382 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4383 vcpu->arch.access, 0, access)) {
4384 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4385 (gva & (PAGE_SIZE - 1));
4386 trace_vcpu_match_mmio(gva, *gpa, write, false);
4390 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4392 if (*gpa == UNMAPPED_GVA)
4395 /* For APIC access vmexit */
4396 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4399 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4400 trace_vcpu_match_mmio(gva, *gpa, write, true);
4407 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4408 const void *val, int bytes)
4412 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4415 kvm_page_track_write(vcpu, gpa, val, bytes);
4419 struct read_write_emulator_ops {
4420 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4422 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4423 void *val, int bytes);
4424 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4425 int bytes, void *val);
4426 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4427 void *val, int bytes);
4431 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4433 if (vcpu->mmio_read_completed) {
4434 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4435 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4436 vcpu->mmio_read_completed = 0;
4443 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4444 void *val, int bytes)
4446 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4449 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4450 void *val, int bytes)
4452 return emulator_write_phys(vcpu, gpa, val, bytes);
4455 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4457 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4458 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4461 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4462 void *val, int bytes)
4464 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4465 return X86EMUL_IO_NEEDED;
4468 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4469 void *val, int bytes)
4471 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4473 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4474 return X86EMUL_CONTINUE;
4477 static const struct read_write_emulator_ops read_emultor = {
4478 .read_write_prepare = read_prepare,
4479 .read_write_emulate = read_emulate,
4480 .read_write_mmio = vcpu_mmio_read,
4481 .read_write_exit_mmio = read_exit_mmio,
4484 static const struct read_write_emulator_ops write_emultor = {
4485 .read_write_emulate = write_emulate,
4486 .read_write_mmio = write_mmio,
4487 .read_write_exit_mmio = write_exit_mmio,
4491 static int emulator_read_write_onepage(unsigned long addr, void *val,
4493 struct x86_exception *exception,
4494 struct kvm_vcpu *vcpu,
4495 const struct read_write_emulator_ops *ops)
4499 bool write = ops->write;
4500 struct kvm_mmio_fragment *frag;
4502 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4505 return X86EMUL_PROPAGATE_FAULT;
4507 /* For APIC access vmexit */
4511 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4512 return X86EMUL_CONTINUE;
4516 * Is this MMIO handled locally?
4518 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4519 if (handled == bytes)
4520 return X86EMUL_CONTINUE;
4526 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4527 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4531 return X86EMUL_CONTINUE;
4534 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4536 void *val, unsigned int bytes,
4537 struct x86_exception *exception,
4538 const struct read_write_emulator_ops *ops)
4540 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4544 if (ops->read_write_prepare &&
4545 ops->read_write_prepare(vcpu, val, bytes))
4546 return X86EMUL_CONTINUE;
4548 vcpu->mmio_nr_fragments = 0;
4550 /* Crossing a page boundary? */
4551 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4554 now = -addr & ~PAGE_MASK;
4555 rc = emulator_read_write_onepage(addr, val, now, exception,
4558 if (rc != X86EMUL_CONTINUE)
4561 if (ctxt->mode != X86EMUL_MODE_PROT64)
4567 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4569 if (rc != X86EMUL_CONTINUE)
4572 if (!vcpu->mmio_nr_fragments)
4575 gpa = vcpu->mmio_fragments[0].gpa;
4577 vcpu->mmio_needed = 1;
4578 vcpu->mmio_cur_fragment = 0;
4580 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4581 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4582 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4583 vcpu->run->mmio.phys_addr = gpa;
4585 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4588 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4592 struct x86_exception *exception)
4594 return emulator_read_write(ctxt, addr, val, bytes,
4595 exception, &read_emultor);
4598 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4602 struct x86_exception *exception)
4604 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4605 exception, &write_emultor);
4608 #define CMPXCHG_TYPE(t, ptr, old, new) \
4609 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4611 #ifdef CONFIG_X86_64
4612 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4614 # define CMPXCHG64(ptr, old, new) \
4615 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4618 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4623 struct x86_exception *exception)
4625 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4631 /* guests cmpxchg8b have to be emulated atomically */
4632 if (bytes > 8 || (bytes & (bytes - 1)))
4635 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4637 if (gpa == UNMAPPED_GVA ||
4638 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4641 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4644 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4645 if (is_error_page(page))
4648 kaddr = kmap_atomic(page);
4649 kaddr += offset_in_page(gpa);
4652 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4655 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4658 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4661 exchanged = CMPXCHG64(kaddr, old, new);
4666 kunmap_atomic(kaddr);
4667 kvm_release_page_dirty(page);
4670 return X86EMUL_CMPXCHG_FAILED;
4672 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4673 kvm_page_track_write(vcpu, gpa, new, bytes);
4675 return X86EMUL_CONTINUE;
4678 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4680 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4683 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4685 /* TODO: String I/O for in kernel device */
4688 if (vcpu->arch.pio.in)
4689 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4690 vcpu->arch.pio.size, pd);
4692 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4693 vcpu->arch.pio.port, vcpu->arch.pio.size,
4698 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4699 unsigned short port, void *val,
4700 unsigned int count, bool in)
4702 vcpu->arch.pio.port = port;
4703 vcpu->arch.pio.in = in;
4704 vcpu->arch.pio.count = count;
4705 vcpu->arch.pio.size = size;
4707 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4708 vcpu->arch.pio.count = 0;
4712 vcpu->run->exit_reason = KVM_EXIT_IO;
4713 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4714 vcpu->run->io.size = size;
4715 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4716 vcpu->run->io.count = count;
4717 vcpu->run->io.port = port;
4722 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4723 int size, unsigned short port, void *val,
4726 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4729 if (vcpu->arch.pio.count)
4732 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4735 memcpy(val, vcpu->arch.pio_data, size * count);
4736 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4737 vcpu->arch.pio.count = 0;
4744 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4745 int size, unsigned short port,
4746 const void *val, unsigned int count)
4748 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4750 memcpy(vcpu->arch.pio_data, val, size * count);
4751 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4752 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4755 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4757 return kvm_x86_ops->get_segment_base(vcpu, seg);
4760 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4762 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4765 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4767 if (!need_emulate_wbinvd(vcpu))
4768 return X86EMUL_CONTINUE;
4770 if (kvm_x86_ops->has_wbinvd_exit()) {
4771 int cpu = get_cpu();
4773 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4774 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4775 wbinvd_ipi, NULL, 1);
4777 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4780 return X86EMUL_CONTINUE;
4783 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4785 kvm_x86_ops->skip_emulated_instruction(vcpu);
4786 return kvm_emulate_wbinvd_noskip(vcpu);
4788 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4792 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4794 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4797 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4798 unsigned long *dest)
4800 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4803 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4804 unsigned long value)
4807 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4810 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4812 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4815 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4817 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4818 unsigned long value;
4822 value = kvm_read_cr0(vcpu);
4825 value = vcpu->arch.cr2;
4828 value = kvm_read_cr3(vcpu);
4831 value = kvm_read_cr4(vcpu);
4834 value = kvm_get_cr8(vcpu);
4837 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4844 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4846 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4851 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4854 vcpu->arch.cr2 = val;
4857 res = kvm_set_cr3(vcpu, val);
4860 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4863 res = kvm_set_cr8(vcpu, val);
4866 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4873 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4875 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4878 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4880 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4883 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4885 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4888 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4890 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4893 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4895 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4898 static unsigned long emulator_get_cached_segment_base(
4899 struct x86_emulate_ctxt *ctxt, int seg)
4901 return get_segment_base(emul_to_vcpu(ctxt), seg);
4904 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4905 struct desc_struct *desc, u32 *base3,
4908 struct kvm_segment var;
4910 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4911 *selector = var.selector;
4914 memset(desc, 0, sizeof(*desc));
4920 set_desc_limit(desc, var.limit);
4921 set_desc_base(desc, (unsigned long)var.base);
4922 #ifdef CONFIG_X86_64
4924 *base3 = var.base >> 32;
4926 desc->type = var.type;
4928 desc->dpl = var.dpl;
4929 desc->p = var.present;
4930 desc->avl = var.avl;
4938 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4939 struct desc_struct *desc, u32 base3,
4942 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4943 struct kvm_segment var;
4945 var.selector = selector;
4946 var.base = get_desc_base(desc);
4947 #ifdef CONFIG_X86_64
4948 var.base |= ((u64)base3) << 32;
4950 var.limit = get_desc_limit(desc);
4952 var.limit = (var.limit << 12) | 0xfff;
4953 var.type = desc->type;
4954 var.dpl = desc->dpl;
4959 var.avl = desc->avl;
4960 var.present = desc->p;
4961 var.unusable = !var.present;
4964 kvm_set_segment(vcpu, &var, seg);
4968 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4969 u32 msr_index, u64 *pdata)
4971 struct msr_data msr;
4974 msr.index = msr_index;
4975 msr.host_initiated = false;
4976 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4984 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4985 u32 msr_index, u64 data)
4987 struct msr_data msr;
4990 msr.index = msr_index;
4991 msr.host_initiated = false;
4992 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4995 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4997 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4999 return vcpu->arch.smbase;
5002 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5004 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5006 vcpu->arch.smbase = smbase;
5009 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5012 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5015 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5016 u32 pmc, u64 *pdata)
5018 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5021 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5023 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5026 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5029 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5031 * CR0.TS may reference the host fpu state, not the guest fpu state,
5032 * so it may be clear at this point.
5037 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5042 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5043 struct x86_instruction_info *info,
5044 enum x86_intercept_stage stage)
5046 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5049 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5050 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5052 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5055 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5057 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5060 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5062 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5065 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5067 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5070 static const struct x86_emulate_ops emulate_ops = {
5071 .read_gpr = emulator_read_gpr,
5072 .write_gpr = emulator_write_gpr,
5073 .read_std = kvm_read_guest_virt_system,
5074 .write_std = kvm_write_guest_virt_system,
5075 .read_phys = kvm_read_guest_phys_system,
5076 .fetch = kvm_fetch_guest_virt,
5077 .read_emulated = emulator_read_emulated,
5078 .write_emulated = emulator_write_emulated,
5079 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5080 .invlpg = emulator_invlpg,
5081 .pio_in_emulated = emulator_pio_in_emulated,
5082 .pio_out_emulated = emulator_pio_out_emulated,
5083 .get_segment = emulator_get_segment,
5084 .set_segment = emulator_set_segment,
5085 .get_cached_segment_base = emulator_get_cached_segment_base,
5086 .get_gdt = emulator_get_gdt,
5087 .get_idt = emulator_get_idt,
5088 .set_gdt = emulator_set_gdt,
5089 .set_idt = emulator_set_idt,
5090 .get_cr = emulator_get_cr,
5091 .set_cr = emulator_set_cr,
5092 .cpl = emulator_get_cpl,
5093 .get_dr = emulator_get_dr,
5094 .set_dr = emulator_set_dr,
5095 .get_smbase = emulator_get_smbase,
5096 .set_smbase = emulator_set_smbase,
5097 .set_msr = emulator_set_msr,
5098 .get_msr = emulator_get_msr,
5099 .check_pmc = emulator_check_pmc,
5100 .read_pmc = emulator_read_pmc,
5101 .halt = emulator_halt,
5102 .wbinvd = emulator_wbinvd,
5103 .fix_hypercall = emulator_fix_hypercall,
5104 .get_fpu = emulator_get_fpu,
5105 .put_fpu = emulator_put_fpu,
5106 .intercept = emulator_intercept,
5107 .get_cpuid = emulator_get_cpuid,
5108 .set_nmi_mask = emulator_set_nmi_mask,
5111 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5113 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5115 * an sti; sti; sequence only disable interrupts for the first
5116 * instruction. So, if the last instruction, be it emulated or
5117 * not, left the system with the INT_STI flag enabled, it
5118 * means that the last instruction is an sti. We should not
5119 * leave the flag on in this case. The same goes for mov ss
5121 if (int_shadow & mask)
5123 if (unlikely(int_shadow || mask)) {
5124 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5126 kvm_make_request(KVM_REQ_EVENT, vcpu);
5130 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5132 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5133 if (ctxt->exception.vector == PF_VECTOR)
5134 return kvm_propagate_fault(vcpu, &ctxt->exception);
5136 if (ctxt->exception.error_code_valid)
5137 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5138 ctxt->exception.error_code);
5140 kvm_queue_exception(vcpu, ctxt->exception.vector);
5144 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5146 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5149 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5151 ctxt->eflags = kvm_get_rflags(vcpu);
5152 ctxt->eip = kvm_rip_read(vcpu);
5153 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5154 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5155 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5156 cs_db ? X86EMUL_MODE_PROT32 :
5157 X86EMUL_MODE_PROT16;
5158 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5159 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5160 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5161 ctxt->emul_flags = vcpu->arch.hflags;
5163 init_decode_cache(ctxt);
5164 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5167 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5169 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5172 init_emulate_ctxt(vcpu);
5176 ctxt->_eip = ctxt->eip + inc_eip;
5177 ret = emulate_int_real(ctxt, irq);
5179 if (ret != X86EMUL_CONTINUE)
5180 return EMULATE_FAIL;
5182 ctxt->eip = ctxt->_eip;
5183 kvm_rip_write(vcpu, ctxt->eip);
5184 kvm_set_rflags(vcpu, ctxt->eflags);
5186 if (irq == NMI_VECTOR)
5187 vcpu->arch.nmi_pending = 0;
5189 vcpu->arch.interrupt.pending = false;
5191 return EMULATE_DONE;
5193 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5195 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5197 int r = EMULATE_DONE;
5199 ++vcpu->stat.insn_emulation_fail;
5200 trace_kvm_emulate_insn_failed(vcpu);
5201 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5202 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5203 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5204 vcpu->run->internal.ndata = 0;
5207 kvm_queue_exception(vcpu, UD_VECTOR);
5212 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5213 bool write_fault_to_shadow_pgtable,
5219 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5222 if (!vcpu->arch.mmu.direct_map) {
5224 * Write permission should be allowed since only
5225 * write access need to be emulated.
5227 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5230 * If the mapping is invalid in guest, let cpu retry
5231 * it to generate fault.
5233 if (gpa == UNMAPPED_GVA)
5238 * Do not retry the unhandleable instruction if it faults on the
5239 * readonly host memory, otherwise it will goto a infinite loop:
5240 * retry instruction -> write #PF -> emulation fail -> retry
5241 * instruction -> ...
5243 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5246 * If the instruction failed on the error pfn, it can not be fixed,
5247 * report the error to userspace.
5249 if (is_error_noslot_pfn(pfn))
5252 kvm_release_pfn_clean(pfn);
5254 /* The instructions are well-emulated on direct mmu. */
5255 if (vcpu->arch.mmu.direct_map) {
5256 unsigned int indirect_shadow_pages;
5258 spin_lock(&vcpu->kvm->mmu_lock);
5259 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5260 spin_unlock(&vcpu->kvm->mmu_lock);
5262 if (indirect_shadow_pages)
5263 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5269 * if emulation was due to access to shadowed page table
5270 * and it failed try to unshadow page and re-enter the
5271 * guest to let CPU execute the instruction.
5273 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5276 * If the access faults on its page table, it can not
5277 * be fixed by unprotecting shadow page and it should
5278 * be reported to userspace.
5280 return !write_fault_to_shadow_pgtable;
5283 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5284 unsigned long cr2, int emulation_type)
5286 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5287 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5289 last_retry_eip = vcpu->arch.last_retry_eip;
5290 last_retry_addr = vcpu->arch.last_retry_addr;
5293 * If the emulation is caused by #PF and it is non-page_table
5294 * writing instruction, it means the VM-EXIT is caused by shadow
5295 * page protected, we can zap the shadow page and retry this
5296 * instruction directly.
5298 * Note: if the guest uses a non-page-table modifying instruction
5299 * on the PDE that points to the instruction, then we will unmap
5300 * the instruction and go to an infinite loop. So, we cache the
5301 * last retried eip and the last fault address, if we meet the eip
5302 * and the address again, we can break out of the potential infinite
5305 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5307 if (!(emulation_type & EMULTYPE_RETRY))
5310 if (x86_page_table_writing_insn(ctxt))
5313 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5316 vcpu->arch.last_retry_eip = ctxt->eip;
5317 vcpu->arch.last_retry_addr = cr2;
5319 if (!vcpu->arch.mmu.direct_map)
5320 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5322 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5327 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5328 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5330 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5332 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5333 /* This is a good place to trace that we are exiting SMM. */
5334 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5336 /* Process a latched INIT or SMI, if any. */
5337 kvm_make_request(KVM_REQ_EVENT, vcpu);
5340 kvm_mmu_reset_context(vcpu);
5343 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5345 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5347 vcpu->arch.hflags = emul_flags;
5349 if (changed & HF_SMM_MASK)
5350 kvm_smm_changed(vcpu);
5353 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5362 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5363 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5368 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5370 struct kvm_run *kvm_run = vcpu->run;
5373 * rflags is the old, "raw" value of the flags. The new value has
5374 * not been saved yet.
5376 * This is correct even for TF set by the guest, because "the
5377 * processor will not generate this exception after the instruction
5378 * that sets the TF flag".
5380 if (unlikely(rflags & X86_EFLAGS_TF)) {
5381 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5382 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5384 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5385 kvm_run->debug.arch.exception = DB_VECTOR;
5386 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5387 *r = EMULATE_USER_EXIT;
5389 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5391 * "Certain debug exceptions may clear bit 0-3. The
5392 * remaining contents of the DR6 register are never
5393 * cleared by the processor".
5395 vcpu->arch.dr6 &= ~15;
5396 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5397 kvm_queue_exception(vcpu, DB_VECTOR);
5402 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5404 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5405 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5406 struct kvm_run *kvm_run = vcpu->run;
5407 unsigned long eip = kvm_get_linear_rip(vcpu);
5408 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5409 vcpu->arch.guest_debug_dr7,
5413 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5414 kvm_run->debug.arch.pc = eip;
5415 kvm_run->debug.arch.exception = DB_VECTOR;
5416 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5417 *r = EMULATE_USER_EXIT;
5422 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5423 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5424 unsigned long eip = kvm_get_linear_rip(vcpu);
5425 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5430 vcpu->arch.dr6 &= ~15;
5431 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5432 kvm_queue_exception(vcpu, DB_VECTOR);
5441 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5448 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5449 bool writeback = true;
5450 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5453 * Clear write_fault_to_shadow_pgtable here to ensure it is
5456 vcpu->arch.write_fault_to_shadow_pgtable = false;
5457 kvm_clear_exception_queue(vcpu);
5459 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5460 init_emulate_ctxt(vcpu);
5463 * We will reenter on the same instruction since
5464 * we do not set complete_userspace_io. This does not
5465 * handle watchpoints yet, those would be handled in
5468 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5471 ctxt->interruptibility = 0;
5472 ctxt->have_exception = false;
5473 ctxt->exception.vector = -1;
5474 ctxt->perm_ok = false;
5476 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5478 r = x86_decode_insn(ctxt, insn, insn_len);
5480 trace_kvm_emulate_insn_start(vcpu);
5481 ++vcpu->stat.insn_emulation;
5482 if (r != EMULATION_OK) {
5483 if (emulation_type & EMULTYPE_TRAP_UD)
5484 return EMULATE_FAIL;
5485 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5487 return EMULATE_DONE;
5488 if (emulation_type & EMULTYPE_SKIP)
5489 return EMULATE_FAIL;
5490 return handle_emulation_failure(vcpu);
5494 if (emulation_type & EMULTYPE_SKIP) {
5495 kvm_rip_write(vcpu, ctxt->_eip);
5496 if (ctxt->eflags & X86_EFLAGS_RF)
5497 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5498 return EMULATE_DONE;
5501 if (retry_instruction(ctxt, cr2, emulation_type))
5502 return EMULATE_DONE;
5504 /* this is needed for vmware backdoor interface to work since it
5505 changes registers values during IO operation */
5506 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5507 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5508 emulator_invalidate_register_cache(ctxt);
5512 r = x86_emulate_insn(ctxt);
5514 if (r == EMULATION_INTERCEPTED)
5515 return EMULATE_DONE;
5517 if (r == EMULATION_FAILED) {
5518 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5520 return EMULATE_DONE;
5522 return handle_emulation_failure(vcpu);
5525 if (ctxt->have_exception) {
5527 if (inject_emulated_exception(vcpu))
5529 } else if (vcpu->arch.pio.count) {
5530 if (!vcpu->arch.pio.in) {
5531 /* FIXME: return into emulator if single-stepping. */
5532 vcpu->arch.pio.count = 0;
5535 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5537 r = EMULATE_USER_EXIT;
5538 } else if (vcpu->mmio_needed) {
5539 if (!vcpu->mmio_is_write)
5541 r = EMULATE_USER_EXIT;
5542 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5543 } else if (r == EMULATION_RESTART)
5549 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5550 toggle_interruptibility(vcpu, ctxt->interruptibility);
5551 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5552 if (vcpu->arch.hflags != ctxt->emul_flags)
5553 kvm_set_hflags(vcpu, ctxt->emul_flags);
5554 kvm_rip_write(vcpu, ctxt->eip);
5555 if (r == EMULATE_DONE)
5556 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5557 if (!ctxt->have_exception ||
5558 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5559 __kvm_set_rflags(vcpu, ctxt->eflags);
5562 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5563 * do nothing, and it will be requested again as soon as
5564 * the shadow expires. But we still need to check here,
5565 * because POPF has no interrupt shadow.
5567 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5568 kvm_make_request(KVM_REQ_EVENT, vcpu);
5570 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5574 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5576 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5578 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5579 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5580 size, port, &val, 1);
5581 /* do not return to emulator after return from userspace */
5582 vcpu->arch.pio.count = 0;
5585 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5587 static int kvmclock_cpu_down_prep(unsigned int cpu)
5589 __this_cpu_write(cpu_tsc_khz, 0);
5593 static void tsc_khz_changed(void *data)
5595 struct cpufreq_freqs *freq = data;
5596 unsigned long khz = 0;
5600 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5601 khz = cpufreq_quick_get(raw_smp_processor_id());
5604 __this_cpu_write(cpu_tsc_khz, khz);
5607 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5610 struct cpufreq_freqs *freq = data;
5612 struct kvm_vcpu *vcpu;
5613 int i, send_ipi = 0;
5616 * We allow guests to temporarily run on slowing clocks,
5617 * provided we notify them after, or to run on accelerating
5618 * clocks, provided we notify them before. Thus time never
5621 * However, we have a problem. We can't atomically update
5622 * the frequency of a given CPU from this function; it is
5623 * merely a notifier, which can be called from any CPU.
5624 * Changing the TSC frequency at arbitrary points in time
5625 * requires a recomputation of local variables related to
5626 * the TSC for each VCPU. We must flag these local variables
5627 * to be updated and be sure the update takes place with the
5628 * new frequency before any guests proceed.
5630 * Unfortunately, the combination of hotplug CPU and frequency
5631 * change creates an intractable locking scenario; the order
5632 * of when these callouts happen is undefined with respect to
5633 * CPU hotplug, and they can race with each other. As such,
5634 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5635 * undefined; you can actually have a CPU frequency change take
5636 * place in between the computation of X and the setting of the
5637 * variable. To protect against this problem, all updates of
5638 * the per_cpu tsc_khz variable are done in an interrupt
5639 * protected IPI, and all callers wishing to update the value
5640 * must wait for a synchronous IPI to complete (which is trivial
5641 * if the caller is on the CPU already). This establishes the
5642 * necessary total order on variable updates.
5644 * Note that because a guest time update may take place
5645 * anytime after the setting of the VCPU's request bit, the
5646 * correct TSC value must be set before the request. However,
5647 * to ensure the update actually makes it to any guest which
5648 * starts running in hardware virtualization between the set
5649 * and the acquisition of the spinlock, we must also ping the
5650 * CPU after setting the request bit.
5654 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5656 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5659 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5661 spin_lock(&kvm_lock);
5662 list_for_each_entry(kvm, &vm_list, vm_list) {
5663 kvm_for_each_vcpu(i, vcpu, kvm) {
5664 if (vcpu->cpu != freq->cpu)
5666 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5667 if (vcpu->cpu != smp_processor_id())
5671 spin_unlock(&kvm_lock);
5673 if (freq->old < freq->new && send_ipi) {
5675 * We upscale the frequency. Must make the guest
5676 * doesn't see old kvmclock values while running with
5677 * the new frequency, otherwise we risk the guest sees
5678 * time go backwards.
5680 * In case we update the frequency for another cpu
5681 * (which might be in guest context) send an interrupt
5682 * to kick the cpu out of guest context. Next time
5683 * guest context is entered kvmclock will be updated,
5684 * so the guest will not see stale values.
5686 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5691 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5692 .notifier_call = kvmclock_cpufreq_notifier
5695 static int kvmclock_cpu_online(unsigned int cpu)
5697 tsc_khz_changed(NULL);
5701 static void kvm_timer_init(void)
5705 max_tsc_khz = tsc_khz;
5707 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5708 #ifdef CONFIG_CPU_FREQ
5709 struct cpufreq_policy policy;
5710 memset(&policy, 0, sizeof(policy));
5712 cpufreq_get_policy(&policy, cpu);
5713 if (policy.cpuinfo.max_freq)
5714 max_tsc_khz = policy.cpuinfo.max_freq;
5717 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5718 CPUFREQ_TRANSITION_NOTIFIER);
5720 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5722 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "AP_X86_KVM_CLK_ONLINE",
5723 kvmclock_cpu_online, kvmclock_cpu_down_prep);
5726 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5728 int kvm_is_in_guest(void)
5730 return __this_cpu_read(current_vcpu) != NULL;
5733 static int kvm_is_user_mode(void)
5737 if (__this_cpu_read(current_vcpu))
5738 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5740 return user_mode != 0;
5743 static unsigned long kvm_get_guest_ip(void)
5745 unsigned long ip = 0;
5747 if (__this_cpu_read(current_vcpu))
5748 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5753 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5754 .is_in_guest = kvm_is_in_guest,
5755 .is_user_mode = kvm_is_user_mode,
5756 .get_guest_ip = kvm_get_guest_ip,
5759 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5761 __this_cpu_write(current_vcpu, vcpu);
5763 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5765 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5767 __this_cpu_write(current_vcpu, NULL);
5769 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5771 static void kvm_set_mmio_spte_mask(void)
5774 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5777 * Set the reserved bits and the present bit of an paging-structure
5778 * entry to generate page fault with PFER.RSV = 1.
5780 /* Mask the reserved physical address bits. */
5781 mask = rsvd_bits(maxphyaddr, 51);
5783 /* Bit 62 is always reserved for 32bit host. */
5784 mask |= 0x3ull << 62;
5786 /* Set the present bit. */
5789 #ifdef CONFIG_X86_64
5791 * If reserved bit is not supported, clear the present bit to disable
5794 if (maxphyaddr == 52)
5798 kvm_mmu_set_mmio_spte_mask(mask);
5801 #ifdef CONFIG_X86_64
5802 static void pvclock_gtod_update_fn(struct work_struct *work)
5806 struct kvm_vcpu *vcpu;
5809 spin_lock(&kvm_lock);
5810 list_for_each_entry(kvm, &vm_list, vm_list)
5811 kvm_for_each_vcpu(i, vcpu, kvm)
5812 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5813 atomic_set(&kvm_guest_has_master_clock, 0);
5814 spin_unlock(&kvm_lock);
5817 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5820 * Notification about pvclock gtod data update.
5822 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5825 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5826 struct timekeeper *tk = priv;
5828 update_pvclock_gtod(tk);
5830 /* disable master clock if host does not trust, or does not
5831 * use, TSC clocksource
5833 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5834 atomic_read(&kvm_guest_has_master_clock) != 0)
5835 queue_work(system_long_wq, &pvclock_gtod_work);
5840 static struct notifier_block pvclock_gtod_notifier = {
5841 .notifier_call = pvclock_gtod_notify,
5845 int kvm_arch_init(void *opaque)
5848 struct kvm_x86_ops *ops = opaque;
5851 printk(KERN_ERR "kvm: already loaded the other module\n");
5856 if (!ops->cpu_has_kvm_support()) {
5857 printk(KERN_ERR "kvm: no hardware support\n");
5861 if (ops->disabled_by_bios()) {
5862 printk(KERN_ERR "kvm: disabled by bios\n");
5868 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5870 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5874 r = kvm_mmu_module_init();
5876 goto out_free_percpu;
5878 kvm_set_mmio_spte_mask();
5882 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5883 PT_DIRTY_MASK, PT64_NX_MASK, 0,
5887 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5889 if (boot_cpu_has(X86_FEATURE_XSAVE))
5890 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5893 #ifdef CONFIG_X86_64
5894 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5900 free_percpu(shared_msrs);
5905 void kvm_arch_exit(void)
5907 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5909 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5910 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5911 CPUFREQ_TRANSITION_NOTIFIER);
5912 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
5913 #ifdef CONFIG_X86_64
5914 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5917 kvm_mmu_module_exit();
5918 free_percpu(shared_msrs);
5921 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5923 ++vcpu->stat.halt_exits;
5924 if (lapic_in_kernel(vcpu)) {
5925 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5928 vcpu->run->exit_reason = KVM_EXIT_HLT;
5932 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5934 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5936 kvm_x86_ops->skip_emulated_instruction(vcpu);
5937 return kvm_vcpu_halt(vcpu);
5939 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5942 * kvm_pv_kick_cpu_op: Kick a vcpu.
5944 * @apicid - apicid of vcpu to be kicked.
5946 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5948 struct kvm_lapic_irq lapic_irq;
5950 lapic_irq.shorthand = 0;
5951 lapic_irq.dest_mode = 0;
5952 lapic_irq.dest_id = apicid;
5953 lapic_irq.msi_redir_hint = false;
5955 lapic_irq.delivery_mode = APIC_DM_REMRD;
5956 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5959 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
5961 vcpu->arch.apicv_active = false;
5962 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
5965 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5967 unsigned long nr, a0, a1, a2, a3, ret;
5968 int op_64_bit, r = 1;
5970 kvm_x86_ops->skip_emulated_instruction(vcpu);
5972 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5973 return kvm_hv_hypercall(vcpu);
5975 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5976 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5977 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5978 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5979 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5981 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5983 op_64_bit = is_64_bit_mode(vcpu);
5992 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5998 case KVM_HC_VAPIC_POLL_IRQ:
6001 case KVM_HC_KICK_CPU:
6002 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6012 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6013 ++vcpu->stat.hypercalls;
6016 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6018 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6020 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6021 char instruction[3];
6022 unsigned long rip = kvm_rip_read(vcpu);
6024 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6026 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
6029 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6031 return vcpu->run->request_interrupt_window &&
6032 likely(!pic_in_kernel(vcpu->kvm));
6035 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6037 struct kvm_run *kvm_run = vcpu->run;
6039 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6040 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6041 kvm_run->cr8 = kvm_get_cr8(vcpu);
6042 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6043 kvm_run->ready_for_interrupt_injection =
6044 pic_in_kernel(vcpu->kvm) ||
6045 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6048 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6052 if (!kvm_x86_ops->update_cr8_intercept)
6055 if (!lapic_in_kernel(vcpu))
6058 if (vcpu->arch.apicv_active)
6061 if (!vcpu->arch.apic->vapic_addr)
6062 max_irr = kvm_lapic_find_highest_irr(vcpu);
6069 tpr = kvm_lapic_get_cr8(vcpu);
6071 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6074 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6078 /* try to reinject previous events if any */
6079 if (vcpu->arch.exception.pending) {
6080 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6081 vcpu->arch.exception.has_error_code,
6082 vcpu->arch.exception.error_code);
6084 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6085 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6088 if (vcpu->arch.exception.nr == DB_VECTOR &&
6089 (vcpu->arch.dr7 & DR7_GD)) {
6090 vcpu->arch.dr7 &= ~DR7_GD;
6091 kvm_update_dr7(vcpu);
6094 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6095 vcpu->arch.exception.has_error_code,
6096 vcpu->arch.exception.error_code,
6097 vcpu->arch.exception.reinject);
6101 if (vcpu->arch.nmi_injected) {
6102 kvm_x86_ops->set_nmi(vcpu);
6106 if (vcpu->arch.interrupt.pending) {
6107 kvm_x86_ops->set_irq(vcpu);
6111 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6112 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6117 /* try to inject new event if pending */
6118 if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6119 vcpu->arch.smi_pending = false;
6121 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6122 --vcpu->arch.nmi_pending;
6123 vcpu->arch.nmi_injected = true;
6124 kvm_x86_ops->set_nmi(vcpu);
6125 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6127 * Because interrupts can be injected asynchronously, we are
6128 * calling check_nested_events again here to avoid a race condition.
6129 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6130 * proposal and current concerns. Perhaps we should be setting
6131 * KVM_REQ_EVENT only on certain events and not unconditionally?
6133 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6134 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6138 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6139 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6141 kvm_x86_ops->set_irq(vcpu);
6148 static void process_nmi(struct kvm_vcpu *vcpu)
6153 * x86 is limited to one NMI running, and one NMI pending after it.
6154 * If an NMI is already in progress, limit further NMIs to just one.
6155 * Otherwise, allow two (and we'll inject the first one immediately).
6157 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6160 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6161 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6162 kvm_make_request(KVM_REQ_EVENT, vcpu);
6165 #define put_smstate(type, buf, offset, val) \
6166 *(type *)((buf) + (offset) - 0x7e00) = val
6168 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6171 flags |= seg->g << 23;
6172 flags |= seg->db << 22;
6173 flags |= seg->l << 21;
6174 flags |= seg->avl << 20;
6175 flags |= seg->present << 15;
6176 flags |= seg->dpl << 13;
6177 flags |= seg->s << 12;
6178 flags |= seg->type << 8;
6182 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6184 struct kvm_segment seg;
6187 kvm_get_segment(vcpu, &seg, n);
6188 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6191 offset = 0x7f84 + n * 12;
6193 offset = 0x7f2c + (n - 3) * 12;
6195 put_smstate(u32, buf, offset + 8, seg.base);
6196 put_smstate(u32, buf, offset + 4, seg.limit);
6197 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6200 #ifdef CONFIG_X86_64
6201 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6203 struct kvm_segment seg;
6207 kvm_get_segment(vcpu, &seg, n);
6208 offset = 0x7e00 + n * 16;
6210 flags = enter_smm_get_segment_flags(&seg) >> 8;
6211 put_smstate(u16, buf, offset, seg.selector);
6212 put_smstate(u16, buf, offset + 2, flags);
6213 put_smstate(u32, buf, offset + 4, seg.limit);
6214 put_smstate(u64, buf, offset + 8, seg.base);
6218 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6221 struct kvm_segment seg;
6225 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6226 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6227 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6228 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6230 for (i = 0; i < 8; i++)
6231 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6233 kvm_get_dr(vcpu, 6, &val);
6234 put_smstate(u32, buf, 0x7fcc, (u32)val);
6235 kvm_get_dr(vcpu, 7, &val);
6236 put_smstate(u32, buf, 0x7fc8, (u32)val);
6238 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6239 put_smstate(u32, buf, 0x7fc4, seg.selector);
6240 put_smstate(u32, buf, 0x7f64, seg.base);
6241 put_smstate(u32, buf, 0x7f60, seg.limit);
6242 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6244 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6245 put_smstate(u32, buf, 0x7fc0, seg.selector);
6246 put_smstate(u32, buf, 0x7f80, seg.base);
6247 put_smstate(u32, buf, 0x7f7c, seg.limit);
6248 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6250 kvm_x86_ops->get_gdt(vcpu, &dt);
6251 put_smstate(u32, buf, 0x7f74, dt.address);
6252 put_smstate(u32, buf, 0x7f70, dt.size);
6254 kvm_x86_ops->get_idt(vcpu, &dt);
6255 put_smstate(u32, buf, 0x7f58, dt.address);
6256 put_smstate(u32, buf, 0x7f54, dt.size);
6258 for (i = 0; i < 6; i++)
6259 enter_smm_save_seg_32(vcpu, buf, i);
6261 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6264 put_smstate(u32, buf, 0x7efc, 0x00020000);
6265 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6268 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6270 #ifdef CONFIG_X86_64
6272 struct kvm_segment seg;
6276 for (i = 0; i < 16; i++)
6277 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6279 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6280 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6282 kvm_get_dr(vcpu, 6, &val);
6283 put_smstate(u64, buf, 0x7f68, val);
6284 kvm_get_dr(vcpu, 7, &val);
6285 put_smstate(u64, buf, 0x7f60, val);
6287 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6288 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6289 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6291 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6294 put_smstate(u32, buf, 0x7efc, 0x00020064);
6296 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6298 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6299 put_smstate(u16, buf, 0x7e90, seg.selector);
6300 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6301 put_smstate(u32, buf, 0x7e94, seg.limit);
6302 put_smstate(u64, buf, 0x7e98, seg.base);
6304 kvm_x86_ops->get_idt(vcpu, &dt);
6305 put_smstate(u32, buf, 0x7e84, dt.size);
6306 put_smstate(u64, buf, 0x7e88, dt.address);
6308 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6309 put_smstate(u16, buf, 0x7e70, seg.selector);
6310 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6311 put_smstate(u32, buf, 0x7e74, seg.limit);
6312 put_smstate(u64, buf, 0x7e78, seg.base);
6314 kvm_x86_ops->get_gdt(vcpu, &dt);
6315 put_smstate(u32, buf, 0x7e64, dt.size);
6316 put_smstate(u64, buf, 0x7e68, dt.address);
6318 for (i = 0; i < 6; i++)
6319 enter_smm_save_seg_64(vcpu, buf, i);
6325 static void enter_smm(struct kvm_vcpu *vcpu)
6327 struct kvm_segment cs, ds;
6332 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6333 vcpu->arch.hflags |= HF_SMM_MASK;
6334 memset(buf, 0, 512);
6335 if (guest_cpuid_has_longmode(vcpu))
6336 enter_smm_save_state_64(vcpu, buf);
6338 enter_smm_save_state_32(vcpu, buf);
6340 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6342 if (kvm_x86_ops->get_nmi_mask(vcpu))
6343 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6345 kvm_x86_ops->set_nmi_mask(vcpu, true);
6347 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6348 kvm_rip_write(vcpu, 0x8000);
6350 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6351 kvm_x86_ops->set_cr0(vcpu, cr0);
6352 vcpu->arch.cr0 = cr0;
6354 kvm_x86_ops->set_cr4(vcpu, 0);
6356 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6357 dt.address = dt.size = 0;
6358 kvm_x86_ops->set_idt(vcpu, &dt);
6360 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6362 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6363 cs.base = vcpu->arch.smbase;
6368 cs.limit = ds.limit = 0xffffffff;
6369 cs.type = ds.type = 0x3;
6370 cs.dpl = ds.dpl = 0;
6375 cs.avl = ds.avl = 0;
6376 cs.present = ds.present = 1;
6377 cs.unusable = ds.unusable = 0;
6378 cs.padding = ds.padding = 0;
6380 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6381 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6382 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6383 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6384 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6385 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6387 if (guest_cpuid_has_longmode(vcpu))
6388 kvm_x86_ops->set_efer(vcpu, 0);
6390 kvm_update_cpuid(vcpu);
6391 kvm_mmu_reset_context(vcpu);
6394 static void process_smi(struct kvm_vcpu *vcpu)
6396 vcpu->arch.smi_pending = true;
6397 kvm_make_request(KVM_REQ_EVENT, vcpu);
6400 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6402 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6405 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6407 u64 eoi_exit_bitmap[4];
6409 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6412 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6414 if (irqchip_split(vcpu->kvm))
6415 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6417 if (vcpu->arch.apicv_active)
6418 kvm_x86_ops->sync_pir_to_irr(vcpu);
6419 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6421 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6422 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6423 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6426 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6428 ++vcpu->stat.tlb_flush;
6429 kvm_x86_ops->tlb_flush(vcpu);
6432 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6434 struct page *page = NULL;
6436 if (!lapic_in_kernel(vcpu))
6439 if (!kvm_x86_ops->set_apic_access_page_addr)
6442 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6443 if (is_error_page(page))
6445 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6448 * Do not pin apic access page in memory, the MMU notifier
6449 * will call us again if it is migrated or swapped out.
6453 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6455 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6456 unsigned long address)
6459 * The physical address of apic access page is stored in the VMCS.
6460 * Update it when it becomes invalid.
6462 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6463 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6467 * Returns 1 to let vcpu_run() continue the guest execution loop without
6468 * exiting to the userspace. Otherwise, the value will be returned to the
6471 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6475 dm_request_for_irq_injection(vcpu) &&
6476 kvm_cpu_accept_dm_intr(vcpu);
6478 bool req_immediate_exit = false;
6480 if (vcpu->requests) {
6481 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6482 kvm_mmu_unload(vcpu);
6483 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6484 __kvm_migrate_timers(vcpu);
6485 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6486 kvm_gen_update_masterclock(vcpu->kvm);
6487 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6488 kvm_gen_kvmclock_update(vcpu);
6489 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6490 r = kvm_guest_time_update(vcpu);
6494 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6495 kvm_mmu_sync_roots(vcpu);
6496 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6497 kvm_vcpu_flush_tlb(vcpu);
6498 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6499 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6503 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6504 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6508 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6509 vcpu->fpu_active = 0;
6510 kvm_x86_ops->fpu_deactivate(vcpu);
6512 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6513 /* Page is swapped out. Do synthetic halt */
6514 vcpu->arch.apf.halted = true;
6518 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6519 record_steal_time(vcpu);
6520 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6522 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6524 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6525 kvm_pmu_handle_event(vcpu);
6526 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6527 kvm_pmu_deliver_pmi(vcpu);
6528 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6529 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6530 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6531 vcpu->arch.ioapic_handled_vectors)) {
6532 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6533 vcpu->run->eoi.vector =
6534 vcpu->arch.pending_ioapic_eoi;
6539 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6540 vcpu_scan_ioapic(vcpu);
6541 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6542 kvm_vcpu_reload_apic_access_page(vcpu);
6543 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6544 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6545 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6549 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6550 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6551 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6555 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6556 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6557 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6563 * KVM_REQ_HV_STIMER has to be processed after
6564 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6565 * depend on the guest clock being up-to-date
6567 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6568 kvm_hv_process_stimers(vcpu);
6572 * KVM_REQ_EVENT is not set when posted interrupts are set by
6573 * VT-d hardware, so we have to update RVI unconditionally.
6575 if (kvm_lapic_enabled(vcpu)) {
6577 * Update architecture specific hints for APIC
6578 * virtual interrupt delivery.
6580 if (vcpu->arch.apicv_active)
6581 kvm_x86_ops->hwapic_irr_update(vcpu,
6582 kvm_lapic_find_highest_irr(vcpu));
6585 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6586 kvm_apic_accept_events(vcpu);
6587 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6592 if (inject_pending_event(vcpu, req_int_win) != 0)
6593 req_immediate_exit = true;
6595 /* Enable NMI/IRQ window open exits if needed.
6597 * SMIs have two cases: 1) they can be nested, and
6598 * then there is nothing to do here because RSM will
6599 * cause a vmexit anyway; 2) or the SMI can be pending
6600 * because inject_pending_event has completed the
6601 * injection of an IRQ or NMI from the previous vmexit,
6602 * and then we request an immediate exit to inject the SMI.
6604 if (vcpu->arch.smi_pending && !is_smm(vcpu))
6605 req_immediate_exit = true;
6606 if (vcpu->arch.nmi_pending)
6607 kvm_x86_ops->enable_nmi_window(vcpu);
6608 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6609 kvm_x86_ops->enable_irq_window(vcpu);
6612 if (kvm_lapic_enabled(vcpu)) {
6613 update_cr8_intercept(vcpu);
6614 kvm_lapic_sync_to_vapic(vcpu);
6618 r = kvm_mmu_reload(vcpu);
6620 goto cancel_injection;
6625 kvm_x86_ops->prepare_guest_switch(vcpu);
6626 if (vcpu->fpu_active)
6627 kvm_load_guest_fpu(vcpu);
6628 vcpu->mode = IN_GUEST_MODE;
6630 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6633 * We should set ->mode before check ->requests,
6634 * Please see the comment in kvm_make_all_cpus_request.
6635 * This also orders the write to mode from any reads
6636 * to the page tables done while the VCPU is running.
6637 * Please see the comment in kvm_flush_remote_tlbs.
6639 smp_mb__after_srcu_read_unlock();
6641 local_irq_disable();
6643 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6644 || need_resched() || signal_pending(current)) {
6645 vcpu->mode = OUTSIDE_GUEST_MODE;
6649 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6651 goto cancel_injection;
6654 kvm_load_guest_xcr0(vcpu);
6656 if (req_immediate_exit) {
6657 kvm_make_request(KVM_REQ_EVENT, vcpu);
6658 smp_send_reschedule(vcpu->cpu);
6661 trace_kvm_entry(vcpu->vcpu_id);
6662 wait_lapic_expire(vcpu);
6663 guest_enter_irqoff();
6665 if (unlikely(vcpu->arch.switch_db_regs)) {
6667 set_debugreg(vcpu->arch.eff_db[0], 0);
6668 set_debugreg(vcpu->arch.eff_db[1], 1);
6669 set_debugreg(vcpu->arch.eff_db[2], 2);
6670 set_debugreg(vcpu->arch.eff_db[3], 3);
6671 set_debugreg(vcpu->arch.dr6, 6);
6672 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6675 kvm_x86_ops->run(vcpu);
6678 * Do this here before restoring debug registers on the host. And
6679 * since we do this before handling the vmexit, a DR access vmexit
6680 * can (a) read the correct value of the debug registers, (b) set
6681 * KVM_DEBUGREG_WONT_EXIT again.
6683 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6684 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6685 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6686 kvm_update_dr0123(vcpu);
6687 kvm_update_dr6(vcpu);
6688 kvm_update_dr7(vcpu);
6689 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6693 * If the guest has used debug registers, at least dr7
6694 * will be disabled while returning to the host.
6695 * If we don't have active breakpoints in the host, we don't
6696 * care about the messed up debug address registers. But if
6697 * we have some of them active, restore the old state.
6699 if (hw_breakpoint_active())
6700 hw_breakpoint_restore();
6702 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6704 vcpu->mode = OUTSIDE_GUEST_MODE;
6707 kvm_put_guest_xcr0(vcpu);
6709 kvm_x86_ops->handle_external_intr(vcpu);
6713 guest_exit_irqoff();
6718 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6721 * Profile KVM exit RIPs:
6723 if (unlikely(prof_on == KVM_PROFILING)) {
6724 unsigned long rip = kvm_rip_read(vcpu);
6725 profile_hit(KVM_PROFILING, (void *)rip);
6728 if (unlikely(vcpu->arch.tsc_always_catchup))
6729 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6731 if (vcpu->arch.apic_attention)
6732 kvm_lapic_sync_from_vapic(vcpu);
6734 r = kvm_x86_ops->handle_exit(vcpu);
6738 kvm_x86_ops->cancel_injection(vcpu);
6739 if (unlikely(vcpu->arch.apic_attention))
6740 kvm_lapic_sync_from_vapic(vcpu);
6745 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6747 if (!kvm_arch_vcpu_runnable(vcpu) &&
6748 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6749 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6750 kvm_vcpu_block(vcpu);
6751 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6753 if (kvm_x86_ops->post_block)
6754 kvm_x86_ops->post_block(vcpu);
6756 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6760 kvm_apic_accept_events(vcpu);
6761 switch(vcpu->arch.mp_state) {
6762 case KVM_MP_STATE_HALTED:
6763 vcpu->arch.pv.pv_unhalted = false;
6764 vcpu->arch.mp_state =
6765 KVM_MP_STATE_RUNNABLE;
6766 case KVM_MP_STATE_RUNNABLE:
6767 vcpu->arch.apf.halted = false;
6769 case KVM_MP_STATE_INIT_RECEIVED:
6778 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6780 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6781 !vcpu->arch.apf.halted);
6784 static int vcpu_run(struct kvm_vcpu *vcpu)
6787 struct kvm *kvm = vcpu->kvm;
6789 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6792 if (kvm_vcpu_running(vcpu)) {
6793 r = vcpu_enter_guest(vcpu);
6795 r = vcpu_block(kvm, vcpu);
6801 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6802 if (kvm_cpu_has_pending_timer(vcpu))
6803 kvm_inject_pending_timer_irqs(vcpu);
6805 if (dm_request_for_irq_injection(vcpu) &&
6806 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
6808 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6809 ++vcpu->stat.request_irq_exits;
6813 kvm_check_async_pf_completion(vcpu);
6815 if (signal_pending(current)) {
6817 vcpu->run->exit_reason = KVM_EXIT_INTR;
6818 ++vcpu->stat.signal_exits;
6821 if (need_resched()) {
6822 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6824 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6828 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6833 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6836 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6837 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6838 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6839 if (r != EMULATE_DONE)
6844 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6846 BUG_ON(!vcpu->arch.pio.count);
6848 return complete_emulated_io(vcpu);
6852 * Implements the following, as a state machine:
6856 * for each mmio piece in the fragment
6864 * for each mmio piece in the fragment
6869 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6871 struct kvm_run *run = vcpu->run;
6872 struct kvm_mmio_fragment *frag;
6875 BUG_ON(!vcpu->mmio_needed);
6877 /* Complete previous fragment */
6878 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6879 len = min(8u, frag->len);
6880 if (!vcpu->mmio_is_write)
6881 memcpy(frag->data, run->mmio.data, len);
6883 if (frag->len <= 8) {
6884 /* Switch to the next fragment. */
6886 vcpu->mmio_cur_fragment++;
6888 /* Go forward to the next mmio piece. */
6894 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6895 vcpu->mmio_needed = 0;
6897 /* FIXME: return into emulator if single-stepping. */
6898 if (vcpu->mmio_is_write)
6900 vcpu->mmio_read_completed = 1;
6901 return complete_emulated_io(vcpu);
6904 run->exit_reason = KVM_EXIT_MMIO;
6905 run->mmio.phys_addr = frag->gpa;
6906 if (vcpu->mmio_is_write)
6907 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6908 run->mmio.len = min(8u, frag->len);
6909 run->mmio.is_write = vcpu->mmio_is_write;
6910 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6915 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6917 struct fpu *fpu = ¤t->thread.fpu;
6921 fpu__activate_curr(fpu);
6923 if (vcpu->sigset_active)
6924 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6926 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6927 kvm_vcpu_block(vcpu);
6928 kvm_apic_accept_events(vcpu);
6929 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6934 /* re-sync apic's tpr */
6935 if (!lapic_in_kernel(vcpu)) {
6936 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6942 if (unlikely(vcpu->arch.complete_userspace_io)) {
6943 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6944 vcpu->arch.complete_userspace_io = NULL;
6949 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6954 post_kvm_run_save(vcpu);
6955 if (vcpu->sigset_active)
6956 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6961 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6963 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6965 * We are here if userspace calls get_regs() in the middle of
6966 * instruction emulation. Registers state needs to be copied
6967 * back from emulation context to vcpu. Userspace shouldn't do
6968 * that usually, but some bad designed PV devices (vmware
6969 * backdoor interface) need this to work
6971 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6972 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6974 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6975 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6976 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6977 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6978 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6979 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6980 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6981 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6982 #ifdef CONFIG_X86_64
6983 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6984 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6985 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6986 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6987 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6988 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6989 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6990 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6993 regs->rip = kvm_rip_read(vcpu);
6994 regs->rflags = kvm_get_rflags(vcpu);
6999 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7001 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7002 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7004 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7005 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7006 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7007 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7008 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7009 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7010 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7011 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7012 #ifdef CONFIG_X86_64
7013 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7014 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7015 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7016 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7017 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7018 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7019 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7020 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7023 kvm_rip_write(vcpu, regs->rip);
7024 kvm_set_rflags(vcpu, regs->rflags);
7026 vcpu->arch.exception.pending = false;
7028 kvm_make_request(KVM_REQ_EVENT, vcpu);
7033 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7035 struct kvm_segment cs;
7037 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7041 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7043 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7044 struct kvm_sregs *sregs)
7048 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7049 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7050 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7051 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7052 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7053 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7055 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7056 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7058 kvm_x86_ops->get_idt(vcpu, &dt);
7059 sregs->idt.limit = dt.size;
7060 sregs->idt.base = dt.address;
7061 kvm_x86_ops->get_gdt(vcpu, &dt);
7062 sregs->gdt.limit = dt.size;
7063 sregs->gdt.base = dt.address;
7065 sregs->cr0 = kvm_read_cr0(vcpu);
7066 sregs->cr2 = vcpu->arch.cr2;
7067 sregs->cr3 = kvm_read_cr3(vcpu);
7068 sregs->cr4 = kvm_read_cr4(vcpu);
7069 sregs->cr8 = kvm_get_cr8(vcpu);
7070 sregs->efer = vcpu->arch.efer;
7071 sregs->apic_base = kvm_get_apic_base(vcpu);
7073 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7075 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7076 set_bit(vcpu->arch.interrupt.nr,
7077 (unsigned long *)sregs->interrupt_bitmap);
7082 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7083 struct kvm_mp_state *mp_state)
7085 kvm_apic_accept_events(vcpu);
7086 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7087 vcpu->arch.pv.pv_unhalted)
7088 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7090 mp_state->mp_state = vcpu->arch.mp_state;
7095 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7096 struct kvm_mp_state *mp_state)
7098 if (!lapic_in_kernel(vcpu) &&
7099 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7102 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7103 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7104 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7106 vcpu->arch.mp_state = mp_state->mp_state;
7107 kvm_make_request(KVM_REQ_EVENT, vcpu);
7111 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7112 int reason, bool has_error_code, u32 error_code)
7114 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7117 init_emulate_ctxt(vcpu);
7119 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7120 has_error_code, error_code);
7123 return EMULATE_FAIL;
7125 kvm_rip_write(vcpu, ctxt->eip);
7126 kvm_set_rflags(vcpu, ctxt->eflags);
7127 kvm_make_request(KVM_REQ_EVENT, vcpu);
7128 return EMULATE_DONE;
7130 EXPORT_SYMBOL_GPL(kvm_task_switch);
7132 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7133 struct kvm_sregs *sregs)
7135 struct msr_data apic_base_msr;
7136 int mmu_reset_needed = 0;
7137 int pending_vec, max_bits, idx;
7140 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7143 dt.size = sregs->idt.limit;
7144 dt.address = sregs->idt.base;
7145 kvm_x86_ops->set_idt(vcpu, &dt);
7146 dt.size = sregs->gdt.limit;
7147 dt.address = sregs->gdt.base;
7148 kvm_x86_ops->set_gdt(vcpu, &dt);
7150 vcpu->arch.cr2 = sregs->cr2;
7151 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7152 vcpu->arch.cr3 = sregs->cr3;
7153 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7155 kvm_set_cr8(vcpu, sregs->cr8);
7157 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7158 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7159 apic_base_msr.data = sregs->apic_base;
7160 apic_base_msr.host_initiated = true;
7161 kvm_set_apic_base(vcpu, &apic_base_msr);
7163 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7164 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7165 vcpu->arch.cr0 = sregs->cr0;
7167 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7168 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7169 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7170 kvm_update_cpuid(vcpu);
7172 idx = srcu_read_lock(&vcpu->kvm->srcu);
7173 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7174 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7175 mmu_reset_needed = 1;
7177 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7179 if (mmu_reset_needed)
7180 kvm_mmu_reset_context(vcpu);
7182 max_bits = KVM_NR_INTERRUPTS;
7183 pending_vec = find_first_bit(
7184 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7185 if (pending_vec < max_bits) {
7186 kvm_queue_interrupt(vcpu, pending_vec, false);
7187 pr_debug("Set back pending irq %d\n", pending_vec);
7190 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7191 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7192 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7193 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7194 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7195 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7197 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7198 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7200 update_cr8_intercept(vcpu);
7202 /* Older userspace won't unhalt the vcpu on reset. */
7203 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7204 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7206 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7208 kvm_make_request(KVM_REQ_EVENT, vcpu);
7213 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7214 struct kvm_guest_debug *dbg)
7216 unsigned long rflags;
7219 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7221 if (vcpu->arch.exception.pending)
7223 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7224 kvm_queue_exception(vcpu, DB_VECTOR);
7226 kvm_queue_exception(vcpu, BP_VECTOR);
7230 * Read rflags as long as potentially injected trace flags are still
7233 rflags = kvm_get_rflags(vcpu);
7235 vcpu->guest_debug = dbg->control;
7236 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7237 vcpu->guest_debug = 0;
7239 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7240 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7241 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7242 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7244 for (i = 0; i < KVM_NR_DB_REGS; i++)
7245 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7247 kvm_update_dr7(vcpu);
7249 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7250 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7251 get_segment_base(vcpu, VCPU_SREG_CS);
7254 * Trigger an rflags update that will inject or remove the trace
7257 kvm_set_rflags(vcpu, rflags);
7259 kvm_x86_ops->update_bp_intercept(vcpu);
7269 * Translate a guest virtual address to a guest physical address.
7271 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7272 struct kvm_translation *tr)
7274 unsigned long vaddr = tr->linear_address;
7278 idx = srcu_read_lock(&vcpu->kvm->srcu);
7279 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7280 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7281 tr->physical_address = gpa;
7282 tr->valid = gpa != UNMAPPED_GVA;
7289 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7291 struct fxregs_state *fxsave =
7292 &vcpu->arch.guest_fpu.state.fxsave;
7294 memcpy(fpu->fpr, fxsave->st_space, 128);
7295 fpu->fcw = fxsave->cwd;
7296 fpu->fsw = fxsave->swd;
7297 fpu->ftwx = fxsave->twd;
7298 fpu->last_opcode = fxsave->fop;
7299 fpu->last_ip = fxsave->rip;
7300 fpu->last_dp = fxsave->rdp;
7301 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7306 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7308 struct fxregs_state *fxsave =
7309 &vcpu->arch.guest_fpu.state.fxsave;
7311 memcpy(fxsave->st_space, fpu->fpr, 128);
7312 fxsave->cwd = fpu->fcw;
7313 fxsave->swd = fpu->fsw;
7314 fxsave->twd = fpu->ftwx;
7315 fxsave->fop = fpu->last_opcode;
7316 fxsave->rip = fpu->last_ip;
7317 fxsave->rdp = fpu->last_dp;
7318 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7323 static void fx_init(struct kvm_vcpu *vcpu)
7325 fpstate_init(&vcpu->arch.guest_fpu.state);
7326 if (boot_cpu_has(X86_FEATURE_XSAVES))
7327 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7328 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7331 * Ensure guest xcr0 is valid for loading
7333 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7335 vcpu->arch.cr0 |= X86_CR0_ET;
7338 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7340 if (vcpu->guest_fpu_loaded)
7344 * Restore all possible states in the guest,
7345 * and assume host would use all available bits.
7346 * Guest xcr0 would be loaded later.
7348 vcpu->guest_fpu_loaded = 1;
7349 __kernel_fpu_begin();
7350 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7354 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7356 if (!vcpu->guest_fpu_loaded) {
7357 vcpu->fpu_counter = 0;
7361 vcpu->guest_fpu_loaded = 0;
7362 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7364 ++vcpu->stat.fpu_reload;
7366 * If using eager FPU mode, or if the guest is a frequent user
7367 * of the FPU, just leave the FPU active for next time.
7368 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7369 * the FPU in bursts will revert to loading it on demand.
7371 if (!use_eager_fpu()) {
7372 if (++vcpu->fpu_counter < 5)
7373 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7378 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7380 kvmclock_reset(vcpu);
7382 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7383 kvm_x86_ops->vcpu_free(vcpu);
7386 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7389 struct kvm_vcpu *vcpu;
7391 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7392 printk_once(KERN_WARNING
7393 "kvm: SMP vm created on host with unstable TSC; "
7394 "guest TSC will not be reliable\n");
7396 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7401 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7405 kvm_vcpu_mtrr_init(vcpu);
7406 r = vcpu_load(vcpu);
7409 kvm_vcpu_reset(vcpu, false);
7410 kvm_mmu_setup(vcpu);
7415 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7417 struct msr_data msr;
7418 struct kvm *kvm = vcpu->kvm;
7420 if (vcpu_load(vcpu))
7423 msr.index = MSR_IA32_TSC;
7424 msr.host_initiated = true;
7425 kvm_write_tsc(vcpu, &msr);
7428 if (!kvmclock_periodic_sync)
7431 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7432 KVMCLOCK_SYNC_PERIOD);
7435 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7438 vcpu->arch.apf.msr_val = 0;
7440 r = vcpu_load(vcpu);
7442 kvm_mmu_unload(vcpu);
7445 kvm_x86_ops->vcpu_free(vcpu);
7448 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7450 vcpu->arch.hflags = 0;
7452 vcpu->arch.smi_pending = 0;
7453 atomic_set(&vcpu->arch.nmi_queued, 0);
7454 vcpu->arch.nmi_pending = 0;
7455 vcpu->arch.nmi_injected = false;
7456 kvm_clear_interrupt_queue(vcpu);
7457 kvm_clear_exception_queue(vcpu);
7459 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7460 kvm_update_dr0123(vcpu);
7461 vcpu->arch.dr6 = DR6_INIT;
7462 kvm_update_dr6(vcpu);
7463 vcpu->arch.dr7 = DR7_FIXED_1;
7464 kvm_update_dr7(vcpu);
7468 kvm_make_request(KVM_REQ_EVENT, vcpu);
7469 vcpu->arch.apf.msr_val = 0;
7470 vcpu->arch.st.msr_val = 0;
7472 kvmclock_reset(vcpu);
7474 kvm_clear_async_pf_completion_queue(vcpu);
7475 kvm_async_pf_hash_reset(vcpu);
7476 vcpu->arch.apf.halted = false;
7479 kvm_pmu_reset(vcpu);
7480 vcpu->arch.smbase = 0x30000;
7483 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7484 vcpu->arch.regs_avail = ~0;
7485 vcpu->arch.regs_dirty = ~0;
7487 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7490 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7492 struct kvm_segment cs;
7494 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7495 cs.selector = vector << 8;
7496 cs.base = vector << 12;
7497 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7498 kvm_rip_write(vcpu, 0);
7501 int kvm_arch_hardware_enable(void)
7504 struct kvm_vcpu *vcpu;
7509 bool stable, backwards_tsc = false;
7511 kvm_shared_msr_cpu_online();
7512 ret = kvm_x86_ops->hardware_enable();
7516 local_tsc = rdtsc();
7517 stable = !check_tsc_unstable();
7518 list_for_each_entry(kvm, &vm_list, vm_list) {
7519 kvm_for_each_vcpu(i, vcpu, kvm) {
7520 if (!stable && vcpu->cpu == smp_processor_id())
7521 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7522 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7523 backwards_tsc = true;
7524 if (vcpu->arch.last_host_tsc > max_tsc)
7525 max_tsc = vcpu->arch.last_host_tsc;
7531 * Sometimes, even reliable TSCs go backwards. This happens on
7532 * platforms that reset TSC during suspend or hibernate actions, but
7533 * maintain synchronization. We must compensate. Fortunately, we can
7534 * detect that condition here, which happens early in CPU bringup,
7535 * before any KVM threads can be running. Unfortunately, we can't
7536 * bring the TSCs fully up to date with real time, as we aren't yet far
7537 * enough into CPU bringup that we know how much real time has actually
7538 * elapsed; our helper function, get_kernel_ns() will be using boot
7539 * variables that haven't been updated yet.
7541 * So we simply find the maximum observed TSC above, then record the
7542 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7543 * the adjustment will be applied. Note that we accumulate
7544 * adjustments, in case multiple suspend cycles happen before some VCPU
7545 * gets a chance to run again. In the event that no KVM threads get a
7546 * chance to run, we will miss the entire elapsed period, as we'll have
7547 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7548 * loose cycle time. This isn't too big a deal, since the loss will be
7549 * uniform across all VCPUs (not to mention the scenario is extremely
7550 * unlikely). It is possible that a second hibernate recovery happens
7551 * much faster than a first, causing the observed TSC here to be
7552 * smaller; this would require additional padding adjustment, which is
7553 * why we set last_host_tsc to the local tsc observed here.
7555 * N.B. - this code below runs only on platforms with reliable TSC,
7556 * as that is the only way backwards_tsc is set above. Also note
7557 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7558 * have the same delta_cyc adjustment applied if backwards_tsc
7559 * is detected. Note further, this adjustment is only done once,
7560 * as we reset last_host_tsc on all VCPUs to stop this from being
7561 * called multiple times (one for each physical CPU bringup).
7563 * Platforms with unreliable TSCs don't have to deal with this, they
7564 * will be compensated by the logic in vcpu_load, which sets the TSC to
7565 * catchup mode. This will catchup all VCPUs to real time, but cannot
7566 * guarantee that they stay in perfect synchronization.
7568 if (backwards_tsc) {
7569 u64 delta_cyc = max_tsc - local_tsc;
7570 backwards_tsc_observed = true;
7571 list_for_each_entry(kvm, &vm_list, vm_list) {
7572 kvm_for_each_vcpu(i, vcpu, kvm) {
7573 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7574 vcpu->arch.last_host_tsc = local_tsc;
7575 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7579 * We have to disable TSC offset matching.. if you were
7580 * booting a VM while issuing an S4 host suspend....
7581 * you may have some problem. Solving this issue is
7582 * left as an exercise to the reader.
7584 kvm->arch.last_tsc_nsec = 0;
7585 kvm->arch.last_tsc_write = 0;
7592 void kvm_arch_hardware_disable(void)
7594 kvm_x86_ops->hardware_disable();
7595 drop_user_return_notifiers();
7598 int kvm_arch_hardware_setup(void)
7602 r = kvm_x86_ops->hardware_setup();
7606 if (kvm_has_tsc_control) {
7608 * Make sure the user can only configure tsc_khz values that
7609 * fit into a signed integer.
7610 * A min value is not calculated needed because it will always
7611 * be 1 on all machines.
7613 u64 max = min(0x7fffffffULL,
7614 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7615 kvm_max_guest_tsc_khz = max;
7617 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7620 kvm_init_msr_list();
7624 void kvm_arch_hardware_unsetup(void)
7626 kvm_x86_ops->hardware_unsetup();
7629 void kvm_arch_check_processor_compat(void *rtn)
7631 kvm_x86_ops->check_processor_compatibility(rtn);
7634 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7636 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7638 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7640 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7642 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7645 struct static_key kvm_no_apic_vcpu __read_mostly;
7646 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7648 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7654 BUG_ON(vcpu->kvm == NULL);
7657 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7658 vcpu->arch.pv.pv_unhalted = false;
7659 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7660 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7661 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7663 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7665 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7670 vcpu->arch.pio_data = page_address(page);
7672 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7674 r = kvm_mmu_create(vcpu);
7676 goto fail_free_pio_data;
7678 if (irqchip_in_kernel(kvm)) {
7679 r = kvm_create_lapic(vcpu);
7681 goto fail_mmu_destroy;
7683 static_key_slow_inc(&kvm_no_apic_vcpu);
7685 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7687 if (!vcpu->arch.mce_banks) {
7689 goto fail_free_lapic;
7691 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7693 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7695 goto fail_free_mce_banks;
7700 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7701 vcpu->arch.pv_time_enabled = false;
7703 vcpu->arch.guest_supported_xcr0 = 0;
7704 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7706 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7708 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7710 kvm_async_pf_hash_reset(vcpu);
7713 vcpu->arch.pending_external_vector = -1;
7715 kvm_hv_vcpu_init(vcpu);
7719 fail_free_mce_banks:
7720 kfree(vcpu->arch.mce_banks);
7722 kvm_free_lapic(vcpu);
7724 kvm_mmu_destroy(vcpu);
7726 free_page((unsigned long)vcpu->arch.pio_data);
7731 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7735 kvm_hv_vcpu_uninit(vcpu);
7736 kvm_pmu_destroy(vcpu);
7737 kfree(vcpu->arch.mce_banks);
7738 kvm_free_lapic(vcpu);
7739 idx = srcu_read_lock(&vcpu->kvm->srcu);
7740 kvm_mmu_destroy(vcpu);
7741 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7742 free_page((unsigned long)vcpu->arch.pio_data);
7743 if (!lapic_in_kernel(vcpu))
7744 static_key_slow_dec(&kvm_no_apic_vcpu);
7747 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7749 kvm_x86_ops->sched_in(vcpu, cpu);
7752 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7757 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7758 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7759 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7760 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7761 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7763 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7764 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7765 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7766 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7767 &kvm->arch.irq_sources_bitmap);
7769 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7770 mutex_init(&kvm->arch.apic_map_lock);
7771 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7773 pvclock_update_vm_gtod_copy(kvm);
7775 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7776 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7778 kvm_page_track_init(kvm);
7779 kvm_mmu_init_vm(kvm);
7781 if (kvm_x86_ops->vm_init)
7782 return kvm_x86_ops->vm_init(kvm);
7787 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7790 r = vcpu_load(vcpu);
7792 kvm_mmu_unload(vcpu);
7796 static void kvm_free_vcpus(struct kvm *kvm)
7799 struct kvm_vcpu *vcpu;
7802 * Unpin any mmu pages first.
7804 kvm_for_each_vcpu(i, vcpu, kvm) {
7805 kvm_clear_async_pf_completion_queue(vcpu);
7806 kvm_unload_vcpu_mmu(vcpu);
7808 kvm_for_each_vcpu(i, vcpu, kvm)
7809 kvm_arch_vcpu_free(vcpu);
7811 mutex_lock(&kvm->lock);
7812 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7813 kvm->vcpus[i] = NULL;
7815 atomic_set(&kvm->online_vcpus, 0);
7816 mutex_unlock(&kvm->lock);
7819 void kvm_arch_sync_events(struct kvm *kvm)
7821 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7822 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7823 kvm_free_all_assigned_devices(kvm);
7827 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7831 struct kvm_memslots *slots = kvm_memslots(kvm);
7832 struct kvm_memory_slot *slot, old;
7834 /* Called with kvm->slots_lock held. */
7835 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7838 slot = id_to_memslot(slots, id);
7844 * MAP_SHARED to prevent internal slot pages from being moved
7847 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7848 MAP_SHARED | MAP_ANONYMOUS, 0);
7849 if (IS_ERR((void *)hva))
7850 return PTR_ERR((void *)hva);
7859 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7860 struct kvm_userspace_memory_region m;
7862 m.slot = id | (i << 16);
7864 m.guest_phys_addr = gpa;
7865 m.userspace_addr = hva;
7866 m.memory_size = size;
7867 r = __kvm_set_memory_region(kvm, &m);
7873 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7879 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7881 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7885 mutex_lock(&kvm->slots_lock);
7886 r = __x86_set_memory_region(kvm, id, gpa, size);
7887 mutex_unlock(&kvm->slots_lock);
7891 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7893 void kvm_arch_destroy_vm(struct kvm *kvm)
7895 if (current->mm == kvm->mm) {
7897 * Free memory regions allocated on behalf of userspace,
7898 * unless the the memory map has changed due to process exit
7901 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7902 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7903 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7905 if (kvm_x86_ops->vm_destroy)
7906 kvm_x86_ops->vm_destroy(kvm);
7907 kvm_iommu_unmap_guest(kvm);
7908 kfree(kvm->arch.vpic);
7909 kfree(kvm->arch.vioapic);
7910 kvm_free_vcpus(kvm);
7911 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7912 kvm_mmu_uninit_vm(kvm);
7915 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7916 struct kvm_memory_slot *dont)
7920 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7921 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7922 kvfree(free->arch.rmap[i]);
7923 free->arch.rmap[i] = NULL;
7928 if (!dont || free->arch.lpage_info[i - 1] !=
7929 dont->arch.lpage_info[i - 1]) {
7930 kvfree(free->arch.lpage_info[i - 1]);
7931 free->arch.lpage_info[i - 1] = NULL;
7935 kvm_page_track_free_memslot(free, dont);
7938 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7939 unsigned long npages)
7943 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7944 struct kvm_lpage_info *linfo;
7949 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7950 slot->base_gfn, level) + 1;
7952 slot->arch.rmap[i] =
7953 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7954 if (!slot->arch.rmap[i])
7959 linfo = kvm_kvzalloc(lpages * sizeof(*linfo));
7963 slot->arch.lpage_info[i - 1] = linfo;
7965 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7966 linfo[0].disallow_lpage = 1;
7967 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7968 linfo[lpages - 1].disallow_lpage = 1;
7969 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7971 * If the gfn and userspace address are not aligned wrt each
7972 * other, or if explicitly asked to, disable large page
7973 * support for this slot
7975 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7976 !kvm_largepages_enabled()) {
7979 for (j = 0; j < lpages; ++j)
7980 linfo[j].disallow_lpage = 1;
7984 if (kvm_page_track_create_memslot(slot, npages))
7990 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7991 kvfree(slot->arch.rmap[i]);
7992 slot->arch.rmap[i] = NULL;
7996 kvfree(slot->arch.lpage_info[i - 1]);
7997 slot->arch.lpage_info[i - 1] = NULL;
8002 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8005 * memslots->generation has been incremented.
8006 * mmio generation may have reached its maximum value.
8008 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8011 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8012 struct kvm_memory_slot *memslot,
8013 const struct kvm_userspace_memory_region *mem,
8014 enum kvm_mr_change change)
8019 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8020 struct kvm_memory_slot *new)
8022 /* Still write protect RO slot */
8023 if (new->flags & KVM_MEM_READONLY) {
8024 kvm_mmu_slot_remove_write_access(kvm, new);
8029 * Call kvm_x86_ops dirty logging hooks when they are valid.
8031 * kvm_x86_ops->slot_disable_log_dirty is called when:
8033 * - KVM_MR_CREATE with dirty logging is disabled
8034 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8036 * The reason is, in case of PML, we need to set D-bit for any slots
8037 * with dirty logging disabled in order to eliminate unnecessary GPA
8038 * logging in PML buffer (and potential PML buffer full VMEXT). This
8039 * guarantees leaving PML enabled during guest's lifetime won't have
8040 * any additonal overhead from PML when guest is running with dirty
8041 * logging disabled for memory slots.
8043 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8044 * to dirty logging mode.
8046 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8048 * In case of write protect:
8050 * Write protect all pages for dirty logging.
8052 * All the sptes including the large sptes which point to this
8053 * slot are set to readonly. We can not create any new large
8054 * spte on this slot until the end of the logging.
8056 * See the comments in fast_page_fault().
8058 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8059 if (kvm_x86_ops->slot_enable_log_dirty)
8060 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8062 kvm_mmu_slot_remove_write_access(kvm, new);
8064 if (kvm_x86_ops->slot_disable_log_dirty)
8065 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8069 void kvm_arch_commit_memory_region(struct kvm *kvm,
8070 const struct kvm_userspace_memory_region *mem,
8071 const struct kvm_memory_slot *old,
8072 const struct kvm_memory_slot *new,
8073 enum kvm_mr_change change)
8075 int nr_mmu_pages = 0;
8077 if (!kvm->arch.n_requested_mmu_pages)
8078 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8081 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8084 * Dirty logging tracks sptes in 4k granularity, meaning that large
8085 * sptes have to be split. If live migration is successful, the guest
8086 * in the source machine will be destroyed and large sptes will be
8087 * created in the destination. However, if the guest continues to run
8088 * in the source machine (for example if live migration fails), small
8089 * sptes will remain around and cause bad performance.
8091 * Scan sptes if dirty logging has been stopped, dropping those
8092 * which can be collapsed into a single large-page spte. Later
8093 * page faults will create the large-page sptes.
8095 if ((change != KVM_MR_DELETE) &&
8096 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8097 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8098 kvm_mmu_zap_collapsible_sptes(kvm, new);
8101 * Set up write protection and/or dirty logging for the new slot.
8103 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8104 * been zapped so no dirty logging staff is needed for old slot. For
8105 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8106 * new and it's also covered when dealing with the new slot.
8108 * FIXME: const-ify all uses of struct kvm_memory_slot.
8110 if (change != KVM_MR_DELETE)
8111 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8114 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8116 kvm_mmu_invalidate_zap_all_pages(kvm);
8119 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8120 struct kvm_memory_slot *slot)
8122 kvm_mmu_invalidate_zap_all_pages(kvm);
8125 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8127 if (!list_empty_careful(&vcpu->async_pf.done))
8130 if (kvm_apic_has_events(vcpu))
8133 if (vcpu->arch.pv.pv_unhalted)
8136 if (atomic_read(&vcpu->arch.nmi_queued))
8139 if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8142 if (kvm_arch_interrupt_allowed(vcpu) &&
8143 kvm_cpu_has_interrupt(vcpu))
8146 if (kvm_hv_has_stimer_pending(vcpu))
8152 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8154 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8155 kvm_x86_ops->check_nested_events(vcpu, false);
8157 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8160 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8162 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8165 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8167 return kvm_x86_ops->interrupt_allowed(vcpu);
8170 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8172 if (is_64_bit_mode(vcpu))
8173 return kvm_rip_read(vcpu);
8174 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8175 kvm_rip_read(vcpu));
8177 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8179 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8181 return kvm_get_linear_rip(vcpu) == linear_rip;
8183 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8185 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8187 unsigned long rflags;
8189 rflags = kvm_x86_ops->get_rflags(vcpu);
8190 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8191 rflags &= ~X86_EFLAGS_TF;
8194 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8196 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8198 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8199 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8200 rflags |= X86_EFLAGS_TF;
8201 kvm_x86_ops->set_rflags(vcpu, rflags);
8204 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8206 __kvm_set_rflags(vcpu, rflags);
8207 kvm_make_request(KVM_REQ_EVENT, vcpu);
8209 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8211 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8215 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8219 r = kvm_mmu_reload(vcpu);
8223 if (!vcpu->arch.mmu.direct_map &&
8224 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8227 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8230 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8232 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8235 static inline u32 kvm_async_pf_next_probe(u32 key)
8237 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8240 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8242 u32 key = kvm_async_pf_hash_fn(gfn);
8244 while (vcpu->arch.apf.gfns[key] != ~0)
8245 key = kvm_async_pf_next_probe(key);
8247 vcpu->arch.apf.gfns[key] = gfn;
8250 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8253 u32 key = kvm_async_pf_hash_fn(gfn);
8255 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8256 (vcpu->arch.apf.gfns[key] != gfn &&
8257 vcpu->arch.apf.gfns[key] != ~0); i++)
8258 key = kvm_async_pf_next_probe(key);
8263 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8265 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8268 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8272 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8274 vcpu->arch.apf.gfns[i] = ~0;
8276 j = kvm_async_pf_next_probe(j);
8277 if (vcpu->arch.apf.gfns[j] == ~0)
8279 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8281 * k lies cyclically in ]i,j]
8283 * |....j i.k.| or |.k..j i...|
8285 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8286 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8291 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8294 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8298 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8299 struct kvm_async_pf *work)
8301 struct x86_exception fault;
8303 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8304 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8306 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8307 (vcpu->arch.apf.send_user_only &&
8308 kvm_x86_ops->get_cpl(vcpu) == 0))
8309 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8310 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8311 fault.vector = PF_VECTOR;
8312 fault.error_code_valid = true;
8313 fault.error_code = 0;
8314 fault.nested_page_fault = false;
8315 fault.address = work->arch.token;
8316 kvm_inject_page_fault(vcpu, &fault);
8320 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8321 struct kvm_async_pf *work)
8323 struct x86_exception fault;
8325 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8326 if (work->wakeup_all)
8327 work->arch.token = ~0; /* broadcast wakeup */
8329 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8331 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8332 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8333 fault.vector = PF_VECTOR;
8334 fault.error_code_valid = true;
8335 fault.error_code = 0;
8336 fault.nested_page_fault = false;
8337 fault.address = work->arch.token;
8338 kvm_inject_page_fault(vcpu, &fault);
8340 vcpu->arch.apf.halted = false;
8341 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8344 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8346 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8349 return !kvm_event_needs_reinjection(vcpu) &&
8350 kvm_x86_ops->interrupt_allowed(vcpu);
8353 void kvm_arch_start_assignment(struct kvm *kvm)
8355 atomic_inc(&kvm->arch.assigned_device_count);
8357 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8359 void kvm_arch_end_assignment(struct kvm *kvm)
8361 atomic_dec(&kvm->arch.assigned_device_count);
8363 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8365 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8367 return atomic_read(&kvm->arch.assigned_device_count);
8369 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8371 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8373 atomic_inc(&kvm->arch.noncoherent_dma_count);
8375 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8377 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8379 atomic_dec(&kvm->arch.noncoherent_dma_count);
8381 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8383 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8385 return atomic_read(&kvm->arch.noncoherent_dma_count);
8387 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8389 bool kvm_arch_has_irq_bypass(void)
8391 return kvm_x86_ops->update_pi_irte != NULL;
8394 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8395 struct irq_bypass_producer *prod)
8397 struct kvm_kernel_irqfd *irqfd =
8398 container_of(cons, struct kvm_kernel_irqfd, consumer);
8400 irqfd->producer = prod;
8402 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8403 prod->irq, irqfd->gsi, 1);
8406 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8407 struct irq_bypass_producer *prod)
8410 struct kvm_kernel_irqfd *irqfd =
8411 container_of(cons, struct kvm_kernel_irqfd, consumer);
8413 WARN_ON(irqfd->producer != prod);
8414 irqfd->producer = NULL;
8417 * When producer of consumer is unregistered, we change back to
8418 * remapped mode, so we can re-use the current implementation
8419 * when the irq is masked/disabled or the consumer side (KVM
8420 * int this case doesn't want to receive the interrupts.
8422 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8424 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8425 " fails: %d\n", irqfd->consumer.token, ret);
8428 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8429 uint32_t guest_irq, bool set)
8431 if (!kvm_x86_ops->update_pi_irte)
8434 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8437 bool kvm_vector_hashing_enabled(void)
8439 return vector_hashing;
8441 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8443 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8444 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8445 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8446 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8447 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8448 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8449 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8450 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8451 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8452 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8453 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8454 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8455 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8456 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8457 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8458 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8459 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8460 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8461 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);