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"
31 #include <linux/clocksource.h>
32 #include <linux/interrupt.h>
33 #include <linux/kvm.h>
35 #include <linux/vmalloc.h>
36 #include <linux/module.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <trace/events/kvm.h>
53 #define CREATE_TRACE_POINTS
56 #include <asm/debugreg.h>
62 #include <asm/fpu-internal.h> /* Ugh! */
64 #include <asm/pvclock.h>
65 #include <asm/div64.h>
67 #define MAX_IO_MSRS 256
68 #define KVM_MAX_MCE_BANKS 32
69 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
71 #define emul_to_vcpu(ctxt) \
72 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
75 * - enable syscall per default because its emulated by KVM
76 * - enable LME and LMA per default on 64 bit KVM
80 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
82 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
85 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
86 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
88 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
89 static void process_nmi(struct kvm_vcpu *vcpu);
90 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
92 struct kvm_x86_ops *kvm_x86_ops;
93 EXPORT_SYMBOL_GPL(kvm_x86_ops);
95 static bool ignore_msrs = 0;
96 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
98 unsigned int min_timer_period_us = 500;
99 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
101 bool kvm_has_tsc_control;
102 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
103 u32 kvm_max_guest_tsc_khz;
104 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
106 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
107 static u32 tsc_tolerance_ppm = 250;
108 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
110 static bool backwards_tsc_observed = false;
112 #define KVM_NR_SHARED_MSRS 16
114 struct kvm_shared_msrs_global {
116 u32 msrs[KVM_NR_SHARED_MSRS];
119 struct kvm_shared_msrs {
120 struct user_return_notifier urn;
122 struct kvm_shared_msr_values {
125 } values[KVM_NR_SHARED_MSRS];
128 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
129 static struct kvm_shared_msrs __percpu *shared_msrs;
131 struct kvm_stats_debugfs_item debugfs_entries[] = {
132 { "pf_fixed", VCPU_STAT(pf_fixed) },
133 { "pf_guest", VCPU_STAT(pf_guest) },
134 { "tlb_flush", VCPU_STAT(tlb_flush) },
135 { "invlpg", VCPU_STAT(invlpg) },
136 { "exits", VCPU_STAT(exits) },
137 { "io_exits", VCPU_STAT(io_exits) },
138 { "mmio_exits", VCPU_STAT(mmio_exits) },
139 { "signal_exits", VCPU_STAT(signal_exits) },
140 { "irq_window", VCPU_STAT(irq_window_exits) },
141 { "nmi_window", VCPU_STAT(nmi_window_exits) },
142 { "halt_exits", VCPU_STAT(halt_exits) },
143 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
144 { "hypercalls", VCPU_STAT(hypercalls) },
145 { "request_irq", VCPU_STAT(request_irq_exits) },
146 { "irq_exits", VCPU_STAT(irq_exits) },
147 { "host_state_reload", VCPU_STAT(host_state_reload) },
148 { "efer_reload", VCPU_STAT(efer_reload) },
149 { "fpu_reload", VCPU_STAT(fpu_reload) },
150 { "insn_emulation", VCPU_STAT(insn_emulation) },
151 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
152 { "irq_injections", VCPU_STAT(irq_injections) },
153 { "nmi_injections", VCPU_STAT(nmi_injections) },
154 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
155 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
156 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
157 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
158 { "mmu_flooded", VM_STAT(mmu_flooded) },
159 { "mmu_recycled", VM_STAT(mmu_recycled) },
160 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
161 { "mmu_unsync", VM_STAT(mmu_unsync) },
162 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
163 { "largepages", VM_STAT(lpages) },
167 u64 __read_mostly host_xcr0;
169 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
171 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
174 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
175 vcpu->arch.apf.gfns[i] = ~0;
178 static void kvm_on_user_return(struct user_return_notifier *urn)
181 struct kvm_shared_msrs *locals
182 = container_of(urn, struct kvm_shared_msrs, urn);
183 struct kvm_shared_msr_values *values;
185 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
186 values = &locals->values[slot];
187 if (values->host != values->curr) {
188 wrmsrl(shared_msrs_global.msrs[slot], values->host);
189 values->curr = values->host;
192 locals->registered = false;
193 user_return_notifier_unregister(urn);
196 static void shared_msr_update(unsigned slot, u32 msr)
199 unsigned int cpu = smp_processor_id();
200 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
202 /* only read, and nobody should modify it at this time,
203 * so don't need lock */
204 if (slot >= shared_msrs_global.nr) {
205 printk(KERN_ERR "kvm: invalid MSR slot!");
208 rdmsrl_safe(msr, &value);
209 smsr->values[slot].host = value;
210 smsr->values[slot].curr = value;
213 void kvm_define_shared_msr(unsigned slot, u32 msr)
215 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
216 if (slot >= shared_msrs_global.nr)
217 shared_msrs_global.nr = slot + 1;
218 shared_msrs_global.msrs[slot] = msr;
219 /* we need ensured the shared_msr_global have been updated */
222 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
224 static void kvm_shared_msr_cpu_online(void)
228 for (i = 0; i < shared_msrs_global.nr; ++i)
229 shared_msr_update(i, shared_msrs_global.msrs[i]);
232 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
234 unsigned int cpu = smp_processor_id();
235 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
237 if (((value ^ smsr->values[slot].curr) & mask) == 0)
239 smsr->values[slot].curr = value;
240 wrmsrl(shared_msrs_global.msrs[slot], value);
241 if (!smsr->registered) {
242 smsr->urn.on_user_return = kvm_on_user_return;
243 user_return_notifier_register(&smsr->urn);
244 smsr->registered = true;
247 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
249 static void drop_user_return_notifiers(void *ignore)
251 unsigned int cpu = smp_processor_id();
252 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
254 if (smsr->registered)
255 kvm_on_user_return(&smsr->urn);
258 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
260 return vcpu->arch.apic_base;
262 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
264 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
266 u64 old_state = vcpu->arch.apic_base &
267 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
268 u64 new_state = msr_info->data &
269 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
270 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
271 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
273 if (!msr_info->host_initiated &&
274 ((msr_info->data & reserved_bits) != 0 ||
275 new_state == X2APIC_ENABLE ||
276 (new_state == MSR_IA32_APICBASE_ENABLE &&
277 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
278 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
282 kvm_lapic_set_base(vcpu, msr_info->data);
285 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
287 asmlinkage __visible void kvm_spurious_fault(void)
289 /* Fault while not rebooting. We want the trace. */
292 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
294 #define EXCPT_BENIGN 0
295 #define EXCPT_CONTRIBUTORY 1
298 static int exception_class(int vector)
308 return EXCPT_CONTRIBUTORY;
315 #define EXCPT_FAULT 0
317 #define EXCPT_ABORT 2
318 #define EXCPT_INTERRUPT 3
320 static int exception_type(int vector)
324 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
325 return EXCPT_INTERRUPT;
329 /* #DB is trap, as instruction watchpoints are handled elsewhere */
330 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
333 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
336 /* Reserved exceptions will result in fault */
340 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
341 unsigned nr, bool has_error, u32 error_code,
347 kvm_make_request(KVM_REQ_EVENT, vcpu);
349 if (!vcpu->arch.exception.pending) {
351 vcpu->arch.exception.pending = true;
352 vcpu->arch.exception.has_error_code = has_error;
353 vcpu->arch.exception.nr = nr;
354 vcpu->arch.exception.error_code = error_code;
355 vcpu->arch.exception.reinject = reinject;
359 /* to check exception */
360 prev_nr = vcpu->arch.exception.nr;
361 if (prev_nr == DF_VECTOR) {
362 /* triple fault -> shutdown */
363 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
366 class1 = exception_class(prev_nr);
367 class2 = exception_class(nr);
368 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
369 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
370 /* generate double fault per SDM Table 5-5 */
371 vcpu->arch.exception.pending = true;
372 vcpu->arch.exception.has_error_code = true;
373 vcpu->arch.exception.nr = DF_VECTOR;
374 vcpu->arch.exception.error_code = 0;
376 /* replace previous exception with a new one in a hope
377 that instruction re-execution will regenerate lost
382 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
384 kvm_multiple_exception(vcpu, nr, false, 0, false);
386 EXPORT_SYMBOL_GPL(kvm_queue_exception);
388 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
390 kvm_multiple_exception(vcpu, nr, false, 0, true);
392 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
394 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
397 kvm_inject_gp(vcpu, 0);
399 kvm_x86_ops->skip_emulated_instruction(vcpu);
401 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
403 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
405 ++vcpu->stat.pf_guest;
406 vcpu->arch.cr2 = fault->address;
407 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
409 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
411 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
413 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
414 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
416 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
419 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
421 atomic_inc(&vcpu->arch.nmi_queued);
422 kvm_make_request(KVM_REQ_NMI, vcpu);
424 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
426 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
428 kvm_multiple_exception(vcpu, nr, true, error_code, false);
430 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
432 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
434 kvm_multiple_exception(vcpu, nr, true, error_code, true);
436 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
439 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
440 * a #GP and return false.
442 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
444 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
446 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
449 EXPORT_SYMBOL_GPL(kvm_require_cpl);
452 * This function will be used to read from the physical memory of the currently
453 * running guest. The difference to kvm_read_guest_page is that this function
454 * can read from guest physical or from the guest's guest physical memory.
456 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
457 gfn_t ngfn, void *data, int offset, int len,
463 ngpa = gfn_to_gpa(ngfn);
464 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
465 if (real_gfn == UNMAPPED_GVA)
468 real_gfn = gpa_to_gfn(real_gfn);
470 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
472 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
474 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
475 void *data, int offset, int len, u32 access)
477 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
478 data, offset, len, access);
482 * Load the pae pdptrs. Return true is they are all valid.
484 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
486 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
487 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
490 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
492 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
493 offset * sizeof(u64), sizeof(pdpte),
494 PFERR_USER_MASK|PFERR_WRITE_MASK);
499 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
500 if (is_present_gpte(pdpte[i]) &&
501 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
508 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
509 __set_bit(VCPU_EXREG_PDPTR,
510 (unsigned long *)&vcpu->arch.regs_avail);
511 __set_bit(VCPU_EXREG_PDPTR,
512 (unsigned long *)&vcpu->arch.regs_dirty);
517 EXPORT_SYMBOL_GPL(load_pdptrs);
519 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
521 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
527 if (is_long_mode(vcpu) || !is_pae(vcpu))
530 if (!test_bit(VCPU_EXREG_PDPTR,
531 (unsigned long *)&vcpu->arch.regs_avail))
534 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
535 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
536 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
537 PFERR_USER_MASK | PFERR_WRITE_MASK);
540 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
546 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
548 unsigned long old_cr0 = kvm_read_cr0(vcpu);
549 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
550 X86_CR0_CD | X86_CR0_NW;
555 if (cr0 & 0xffffffff00000000UL)
559 cr0 &= ~CR0_RESERVED_BITS;
561 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
564 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
567 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
569 if ((vcpu->arch.efer & EFER_LME)) {
574 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
579 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
584 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
587 kvm_x86_ops->set_cr0(vcpu, cr0);
589 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
590 kvm_clear_async_pf_completion_queue(vcpu);
591 kvm_async_pf_hash_reset(vcpu);
594 if ((cr0 ^ old_cr0) & update_bits)
595 kvm_mmu_reset_context(vcpu);
598 EXPORT_SYMBOL_GPL(kvm_set_cr0);
600 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
602 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
604 EXPORT_SYMBOL_GPL(kvm_lmsw);
606 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
608 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
609 !vcpu->guest_xcr0_loaded) {
610 /* kvm_set_xcr() also depends on this */
611 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
612 vcpu->guest_xcr0_loaded = 1;
616 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
618 if (vcpu->guest_xcr0_loaded) {
619 if (vcpu->arch.xcr0 != host_xcr0)
620 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
621 vcpu->guest_xcr0_loaded = 0;
625 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
628 u64 old_xcr0 = vcpu->arch.xcr0;
631 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
632 if (index != XCR_XFEATURE_ENABLED_MASK)
634 if (!(xcr0 & XSTATE_FP))
636 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
640 * Do not allow the guest to set bits that we do not support
641 * saving. However, xcr0 bit 0 is always set, even if the
642 * emulated CPU does not support XSAVE (see fx_init).
644 valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
645 if (xcr0 & ~valid_bits)
648 if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
651 kvm_put_guest_xcr0(vcpu);
652 vcpu->arch.xcr0 = xcr0;
654 if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
655 kvm_update_cpuid(vcpu);
659 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
661 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
662 __kvm_set_xcr(vcpu, index, xcr)) {
663 kvm_inject_gp(vcpu, 0);
668 EXPORT_SYMBOL_GPL(kvm_set_xcr);
670 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
672 unsigned long old_cr4 = kvm_read_cr4(vcpu);
673 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
674 X86_CR4_PAE | X86_CR4_SMEP;
675 if (cr4 & CR4_RESERVED_BITS)
678 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
681 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
684 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
687 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
690 if (is_long_mode(vcpu)) {
691 if (!(cr4 & X86_CR4_PAE))
693 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
694 && ((cr4 ^ old_cr4) & pdptr_bits)
695 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
699 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
700 if (!guest_cpuid_has_pcid(vcpu))
703 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
704 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
708 if (kvm_x86_ops->set_cr4(vcpu, cr4))
711 if (((cr4 ^ old_cr4) & pdptr_bits) ||
712 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
713 kvm_mmu_reset_context(vcpu);
715 if ((cr4 ^ old_cr4) & X86_CR4_SMAP)
716 update_permission_bitmask(vcpu, vcpu->arch.walk_mmu, false);
718 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
719 kvm_update_cpuid(vcpu);
723 EXPORT_SYMBOL_GPL(kvm_set_cr4);
725 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
727 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
728 kvm_mmu_sync_roots(vcpu);
729 kvm_mmu_flush_tlb(vcpu);
733 if (is_long_mode(vcpu)) {
734 if (cr3 & CR3_L_MODE_RESERVED_BITS)
736 } else if (is_pae(vcpu) && is_paging(vcpu) &&
737 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
740 vcpu->arch.cr3 = cr3;
741 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
742 kvm_mmu_new_cr3(vcpu);
745 EXPORT_SYMBOL_GPL(kvm_set_cr3);
747 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
749 if (cr8 & CR8_RESERVED_BITS)
751 if (irqchip_in_kernel(vcpu->kvm))
752 kvm_lapic_set_tpr(vcpu, cr8);
754 vcpu->arch.cr8 = cr8;
757 EXPORT_SYMBOL_GPL(kvm_set_cr8);
759 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
761 if (irqchip_in_kernel(vcpu->kvm))
762 return kvm_lapic_get_cr8(vcpu);
764 return vcpu->arch.cr8;
766 EXPORT_SYMBOL_GPL(kvm_get_cr8);
768 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
770 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
771 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
774 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
778 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
779 dr7 = vcpu->arch.guest_debug_dr7;
781 dr7 = vcpu->arch.dr7;
782 kvm_x86_ops->set_dr7(vcpu, dr7);
783 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
784 if (dr7 & DR7_BP_EN_MASK)
785 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
788 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
790 u64 fixed = DR6_FIXED_1;
792 if (!guest_cpuid_has_rtm(vcpu))
797 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
801 vcpu->arch.db[dr] = val;
802 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
803 vcpu->arch.eff_db[dr] = val;
806 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
810 if (val & 0xffffffff00000000ULL)
812 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
813 kvm_update_dr6(vcpu);
816 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
820 if (val & 0xffffffff00000000ULL)
822 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
823 kvm_update_dr7(vcpu);
830 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
834 res = __kvm_set_dr(vcpu, dr, val);
836 kvm_queue_exception(vcpu, UD_VECTOR);
838 kvm_inject_gp(vcpu, 0);
842 EXPORT_SYMBOL_GPL(kvm_set_dr);
844 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
848 *val = vcpu->arch.db[dr];
851 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
855 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
856 *val = vcpu->arch.dr6;
858 *val = kvm_x86_ops->get_dr6(vcpu);
861 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
865 *val = vcpu->arch.dr7;
872 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
874 if (_kvm_get_dr(vcpu, dr, val)) {
875 kvm_queue_exception(vcpu, UD_VECTOR);
880 EXPORT_SYMBOL_GPL(kvm_get_dr);
882 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
884 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
888 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
891 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
892 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
895 EXPORT_SYMBOL_GPL(kvm_rdpmc);
898 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
899 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
901 * This list is modified at module load time to reflect the
902 * capabilities of the host cpu. This capabilities test skips MSRs that are
903 * kvm-specific. Those are put in the beginning of the list.
906 #define KVM_SAVE_MSRS_BEGIN 12
907 static u32 msrs_to_save[] = {
908 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
909 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
910 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
911 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
912 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
914 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
917 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
919 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
920 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
923 static unsigned num_msrs_to_save;
925 static const u32 emulated_msrs[] = {
927 MSR_IA32_TSCDEADLINE,
928 MSR_IA32_MISC_ENABLE,
933 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
935 if (efer & efer_reserved_bits)
938 if (efer & EFER_FFXSR) {
939 struct kvm_cpuid_entry2 *feat;
941 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
942 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
946 if (efer & EFER_SVME) {
947 struct kvm_cpuid_entry2 *feat;
949 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
950 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
956 EXPORT_SYMBOL_GPL(kvm_valid_efer);
958 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
960 u64 old_efer = vcpu->arch.efer;
962 if (!kvm_valid_efer(vcpu, efer))
966 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
970 efer |= vcpu->arch.efer & EFER_LMA;
972 kvm_x86_ops->set_efer(vcpu, efer);
974 /* Update reserved bits */
975 if ((efer ^ old_efer) & EFER_NX)
976 kvm_mmu_reset_context(vcpu);
981 void kvm_enable_efer_bits(u64 mask)
983 efer_reserved_bits &= ~mask;
985 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
989 * Writes msr value into into the appropriate "register".
990 * Returns 0 on success, non-0 otherwise.
991 * Assumes vcpu_load() was already called.
993 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
995 return kvm_x86_ops->set_msr(vcpu, msr);
999 * Adapt set_msr() to msr_io()'s calling convention
1001 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1003 struct msr_data msr;
1007 msr.host_initiated = true;
1008 return kvm_set_msr(vcpu, &msr);
1011 #ifdef CONFIG_X86_64
1012 struct pvclock_gtod_data {
1015 struct { /* extract of a clocksource struct */
1023 /* open coded 'struct timespec' */
1024 u64 monotonic_time_snsec;
1025 time_t monotonic_time_sec;
1028 static struct pvclock_gtod_data pvclock_gtod_data;
1030 static void update_pvclock_gtod(struct timekeeper *tk)
1032 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1034 write_seqcount_begin(&vdata->seq);
1036 /* copy pvclock gtod data */
1037 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
1038 vdata->clock.cycle_last = tk->clock->cycle_last;
1039 vdata->clock.mask = tk->clock->mask;
1040 vdata->clock.mult = tk->mult;
1041 vdata->clock.shift = tk->shift;
1043 vdata->monotonic_time_sec = tk->xtime_sec
1044 + tk->wall_to_monotonic.tv_sec;
1045 vdata->monotonic_time_snsec = tk->xtime_nsec
1046 + (tk->wall_to_monotonic.tv_nsec
1048 while (vdata->monotonic_time_snsec >=
1049 (((u64)NSEC_PER_SEC) << tk->shift)) {
1050 vdata->monotonic_time_snsec -=
1051 ((u64)NSEC_PER_SEC) << tk->shift;
1052 vdata->monotonic_time_sec++;
1055 write_seqcount_end(&vdata->seq);
1060 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1064 struct pvclock_wall_clock wc;
1065 struct timespec boot;
1070 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1075 ++version; /* first time write, random junk */
1079 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1082 * The guest calculates current wall clock time by adding
1083 * system time (updated by kvm_guest_time_update below) to the
1084 * wall clock specified here. guest system time equals host
1085 * system time for us, thus we must fill in host boot time here.
1089 if (kvm->arch.kvmclock_offset) {
1090 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1091 boot = timespec_sub(boot, ts);
1093 wc.sec = boot.tv_sec;
1094 wc.nsec = boot.tv_nsec;
1095 wc.version = version;
1097 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1100 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1103 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1105 uint32_t quotient, remainder;
1107 /* Don't try to replace with do_div(), this one calculates
1108 * "(dividend << 32) / divisor" */
1110 : "=a" (quotient), "=d" (remainder)
1111 : "0" (0), "1" (dividend), "r" (divisor) );
1115 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1116 s8 *pshift, u32 *pmultiplier)
1123 tps64 = base_khz * 1000LL;
1124 scaled64 = scaled_khz * 1000LL;
1125 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1130 tps32 = (uint32_t)tps64;
1131 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1132 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1140 *pmultiplier = div_frac(scaled64, tps32);
1142 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1143 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1146 static inline u64 get_kernel_ns(void)
1151 monotonic_to_bootbased(&ts);
1152 return timespec_to_ns(&ts);
1155 #ifdef CONFIG_X86_64
1156 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1159 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1160 unsigned long max_tsc_khz;
1162 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1164 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1165 vcpu->arch.virtual_tsc_shift);
1168 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1170 u64 v = (u64)khz * (1000000 + ppm);
1175 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1177 u32 thresh_lo, thresh_hi;
1178 int use_scaling = 0;
1180 /* tsc_khz can be zero if TSC calibration fails */
1181 if (this_tsc_khz == 0)
1184 /* Compute a scale to convert nanoseconds in TSC cycles */
1185 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1186 &vcpu->arch.virtual_tsc_shift,
1187 &vcpu->arch.virtual_tsc_mult);
1188 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1191 * Compute the variation in TSC rate which is acceptable
1192 * within the range of tolerance and decide if the
1193 * rate being applied is within that bounds of the hardware
1194 * rate. If so, no scaling or compensation need be done.
1196 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1197 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1198 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1199 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1202 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1205 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1207 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1208 vcpu->arch.virtual_tsc_mult,
1209 vcpu->arch.virtual_tsc_shift);
1210 tsc += vcpu->arch.this_tsc_write;
1214 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1216 #ifdef CONFIG_X86_64
1218 bool do_request = false;
1219 struct kvm_arch *ka = &vcpu->kvm->arch;
1220 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1222 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1223 atomic_read(&vcpu->kvm->online_vcpus));
1225 if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1226 if (!ka->use_master_clock)
1229 if (!vcpus_matched && ka->use_master_clock)
1233 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1235 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1236 atomic_read(&vcpu->kvm->online_vcpus),
1237 ka->use_master_clock, gtod->clock.vclock_mode);
1241 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1243 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1244 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1247 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1249 struct kvm *kvm = vcpu->kvm;
1250 u64 offset, ns, elapsed;
1251 unsigned long flags;
1254 bool already_matched;
1255 u64 data = msr->data;
1257 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1258 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1259 ns = get_kernel_ns();
1260 elapsed = ns - kvm->arch.last_tsc_nsec;
1262 if (vcpu->arch.virtual_tsc_khz) {
1265 /* n.b - signed multiplication and division required */
1266 usdiff = data - kvm->arch.last_tsc_write;
1267 #ifdef CONFIG_X86_64
1268 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1270 /* do_div() only does unsigned */
1271 asm("1: idivl %[divisor]\n"
1272 "2: xor %%edx, %%edx\n"
1273 " movl $0, %[faulted]\n"
1275 ".section .fixup,\"ax\"\n"
1276 "4: movl $1, %[faulted]\n"
1280 _ASM_EXTABLE(1b, 4b)
1282 : "=A"(usdiff), [faulted] "=r" (faulted)
1283 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1286 do_div(elapsed, 1000);
1291 /* idivl overflow => difference is larger than USEC_PER_SEC */
1293 usdiff = USEC_PER_SEC;
1295 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1298 * Special case: TSC write with a small delta (1 second) of virtual
1299 * cycle time against real time is interpreted as an attempt to
1300 * synchronize the CPU.
1302 * For a reliable TSC, we can match TSC offsets, and for an unstable
1303 * TSC, we add elapsed time in this computation. We could let the
1304 * compensation code attempt to catch up if we fall behind, but
1305 * it's better to try to match offsets from the beginning.
1307 if (usdiff < USEC_PER_SEC &&
1308 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1309 if (!check_tsc_unstable()) {
1310 offset = kvm->arch.cur_tsc_offset;
1311 pr_debug("kvm: matched tsc offset for %llu\n", data);
1313 u64 delta = nsec_to_cycles(vcpu, elapsed);
1315 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1316 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1319 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1322 * We split periods of matched TSC writes into generations.
1323 * For each generation, we track the original measured
1324 * nanosecond time, offset, and write, so if TSCs are in
1325 * sync, we can match exact offset, and if not, we can match
1326 * exact software computation in compute_guest_tsc()
1328 * These values are tracked in kvm->arch.cur_xxx variables.
1330 kvm->arch.cur_tsc_generation++;
1331 kvm->arch.cur_tsc_nsec = ns;
1332 kvm->arch.cur_tsc_write = data;
1333 kvm->arch.cur_tsc_offset = offset;
1335 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1336 kvm->arch.cur_tsc_generation, data);
1340 * We also track th most recent recorded KHZ, write and time to
1341 * allow the matching interval to be extended at each write.
1343 kvm->arch.last_tsc_nsec = ns;
1344 kvm->arch.last_tsc_write = data;
1345 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1347 vcpu->arch.last_guest_tsc = data;
1349 /* Keep track of which generation this VCPU has synchronized to */
1350 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1351 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1352 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1354 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1355 update_ia32_tsc_adjust_msr(vcpu, offset);
1356 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1357 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1359 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1361 kvm->arch.nr_vcpus_matched_tsc = 0;
1362 } else if (!already_matched) {
1363 kvm->arch.nr_vcpus_matched_tsc++;
1366 kvm_track_tsc_matching(vcpu);
1367 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1370 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1372 #ifdef CONFIG_X86_64
1374 static cycle_t read_tsc(void)
1380 * Empirically, a fence (of type that depends on the CPU)
1381 * before rdtsc is enough to ensure that rdtsc is ordered
1382 * with respect to loads. The various CPU manuals are unclear
1383 * as to whether rdtsc can be reordered with later loads,
1384 * but no one has ever seen it happen.
1387 ret = (cycle_t)vget_cycles();
1389 last = pvclock_gtod_data.clock.cycle_last;
1391 if (likely(ret >= last))
1395 * GCC likes to generate cmov here, but this branch is extremely
1396 * predictable (it's just a funciton of time and the likely is
1397 * very likely) and there's a data dependence, so force GCC
1398 * to generate a branch instead. I don't barrier() because
1399 * we don't actually need a barrier, and if this function
1400 * ever gets inlined it will generate worse code.
1406 static inline u64 vgettsc(cycle_t *cycle_now)
1409 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1411 *cycle_now = read_tsc();
1413 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1414 return v * gtod->clock.mult;
1417 static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1422 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1426 seq = read_seqcount_begin(>od->seq);
1427 mode = gtod->clock.vclock_mode;
1428 ts->tv_sec = gtod->monotonic_time_sec;
1429 ns = gtod->monotonic_time_snsec;
1430 ns += vgettsc(cycle_now);
1431 ns >>= gtod->clock.shift;
1432 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1433 timespec_add_ns(ts, ns);
1438 /* returns true if host is using tsc clocksource */
1439 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1443 /* checked again under seqlock below */
1444 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1447 if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1450 monotonic_to_bootbased(&ts);
1451 *kernel_ns = timespec_to_ns(&ts);
1459 * Assuming a stable TSC across physical CPUS, and a stable TSC
1460 * across virtual CPUs, the following condition is possible.
1461 * Each numbered line represents an event visible to both
1462 * CPUs at the next numbered event.
1464 * "timespecX" represents host monotonic time. "tscX" represents
1467 * VCPU0 on CPU0 | VCPU1 on CPU1
1469 * 1. read timespec0,tsc0
1470 * 2. | timespec1 = timespec0 + N
1472 * 3. transition to guest | transition to guest
1473 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1474 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1475 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1477 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1480 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1482 * - 0 < N - M => M < N
1484 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1485 * always the case (the difference between two distinct xtime instances
1486 * might be smaller then the difference between corresponding TSC reads,
1487 * when updating guest vcpus pvclock areas).
1489 * To avoid that problem, do not allow visibility of distinct
1490 * system_timestamp/tsc_timestamp values simultaneously: use a master
1491 * copy of host monotonic time values. Update that master copy
1494 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1498 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1500 #ifdef CONFIG_X86_64
1501 struct kvm_arch *ka = &kvm->arch;
1503 bool host_tsc_clocksource, vcpus_matched;
1505 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1506 atomic_read(&kvm->online_vcpus));
1509 * If the host uses TSC clock, then passthrough TSC as stable
1512 host_tsc_clocksource = kvm_get_time_and_clockread(
1513 &ka->master_kernel_ns,
1514 &ka->master_cycle_now);
1516 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1517 && !backwards_tsc_observed;
1519 if (ka->use_master_clock)
1520 atomic_set(&kvm_guest_has_master_clock, 1);
1522 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1523 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1528 static void kvm_gen_update_masterclock(struct kvm *kvm)
1530 #ifdef CONFIG_X86_64
1532 struct kvm_vcpu *vcpu;
1533 struct kvm_arch *ka = &kvm->arch;
1535 spin_lock(&ka->pvclock_gtod_sync_lock);
1536 kvm_make_mclock_inprogress_request(kvm);
1537 /* no guest entries from this point */
1538 pvclock_update_vm_gtod_copy(kvm);
1540 kvm_for_each_vcpu(i, vcpu, kvm)
1541 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
1543 /* guest entries allowed */
1544 kvm_for_each_vcpu(i, vcpu, kvm)
1545 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1547 spin_unlock(&ka->pvclock_gtod_sync_lock);
1551 static int kvm_guest_time_update(struct kvm_vcpu *v)
1553 unsigned long flags, this_tsc_khz;
1554 struct kvm_vcpu_arch *vcpu = &v->arch;
1555 struct kvm_arch *ka = &v->kvm->arch;
1557 u64 tsc_timestamp, host_tsc;
1558 struct pvclock_vcpu_time_info guest_hv_clock;
1560 bool use_master_clock;
1566 * If the host uses TSC clock, then passthrough TSC as stable
1569 spin_lock(&ka->pvclock_gtod_sync_lock);
1570 use_master_clock = ka->use_master_clock;
1571 if (use_master_clock) {
1572 host_tsc = ka->master_cycle_now;
1573 kernel_ns = ka->master_kernel_ns;
1575 spin_unlock(&ka->pvclock_gtod_sync_lock);
1577 /* Keep irq disabled to prevent changes to the clock */
1578 local_irq_save(flags);
1579 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1580 if (unlikely(this_tsc_khz == 0)) {
1581 local_irq_restore(flags);
1582 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1585 if (!use_master_clock) {
1586 host_tsc = native_read_tsc();
1587 kernel_ns = get_kernel_ns();
1590 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1593 * We may have to catch up the TSC to match elapsed wall clock
1594 * time for two reasons, even if kvmclock is used.
1595 * 1) CPU could have been running below the maximum TSC rate
1596 * 2) Broken TSC compensation resets the base at each VCPU
1597 * entry to avoid unknown leaps of TSC even when running
1598 * again on the same CPU. This may cause apparent elapsed
1599 * time to disappear, and the guest to stand still or run
1602 if (vcpu->tsc_catchup) {
1603 u64 tsc = compute_guest_tsc(v, kernel_ns);
1604 if (tsc > tsc_timestamp) {
1605 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1606 tsc_timestamp = tsc;
1610 local_irq_restore(flags);
1612 if (!vcpu->pv_time_enabled)
1615 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1616 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1617 &vcpu->hv_clock.tsc_shift,
1618 &vcpu->hv_clock.tsc_to_system_mul);
1619 vcpu->hw_tsc_khz = this_tsc_khz;
1622 /* With all the info we got, fill in the values */
1623 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1624 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1625 vcpu->last_guest_tsc = tsc_timestamp;
1628 * The interface expects us to write an even number signaling that the
1629 * update is finished. Since the guest won't see the intermediate
1630 * state, we just increase by 2 at the end.
1632 vcpu->hv_clock.version += 2;
1634 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1635 &guest_hv_clock, sizeof(guest_hv_clock))))
1638 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1639 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1641 if (vcpu->pvclock_set_guest_stopped_request) {
1642 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1643 vcpu->pvclock_set_guest_stopped_request = false;
1646 /* If the host uses TSC clocksource, then it is stable */
1647 if (use_master_clock)
1648 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1650 vcpu->hv_clock.flags = pvclock_flags;
1652 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1654 sizeof(vcpu->hv_clock));
1659 * kvmclock updates which are isolated to a given vcpu, such as
1660 * vcpu->cpu migration, should not allow system_timestamp from
1661 * the rest of the vcpus to remain static. Otherwise ntp frequency
1662 * correction applies to one vcpu's system_timestamp but not
1665 * So in those cases, request a kvmclock update for all vcpus.
1666 * We need to rate-limit these requests though, as they can
1667 * considerably slow guests that have a large number of vcpus.
1668 * The time for a remote vcpu to update its kvmclock is bound
1669 * by the delay we use to rate-limit the updates.
1672 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1674 static void kvmclock_update_fn(struct work_struct *work)
1677 struct delayed_work *dwork = to_delayed_work(work);
1678 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1679 kvmclock_update_work);
1680 struct kvm *kvm = container_of(ka, struct kvm, arch);
1681 struct kvm_vcpu *vcpu;
1683 kvm_for_each_vcpu(i, vcpu, kvm) {
1684 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
1685 kvm_vcpu_kick(vcpu);
1689 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1691 struct kvm *kvm = v->kvm;
1693 set_bit(KVM_REQ_CLOCK_UPDATE, &v->requests);
1694 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1695 KVMCLOCK_UPDATE_DELAY);
1698 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1700 static void kvmclock_sync_fn(struct work_struct *work)
1702 struct delayed_work *dwork = to_delayed_work(work);
1703 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1704 kvmclock_sync_work);
1705 struct kvm *kvm = container_of(ka, struct kvm, arch);
1707 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1708 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1709 KVMCLOCK_SYNC_PERIOD);
1712 static bool msr_mtrr_valid(unsigned msr)
1715 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1716 case MSR_MTRRfix64K_00000:
1717 case MSR_MTRRfix16K_80000:
1718 case MSR_MTRRfix16K_A0000:
1719 case MSR_MTRRfix4K_C0000:
1720 case MSR_MTRRfix4K_C8000:
1721 case MSR_MTRRfix4K_D0000:
1722 case MSR_MTRRfix4K_D8000:
1723 case MSR_MTRRfix4K_E0000:
1724 case MSR_MTRRfix4K_E8000:
1725 case MSR_MTRRfix4K_F0000:
1726 case MSR_MTRRfix4K_F8000:
1727 case MSR_MTRRdefType:
1728 case MSR_IA32_CR_PAT:
1736 static bool valid_pat_type(unsigned t)
1738 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1741 static bool valid_mtrr_type(unsigned t)
1743 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1746 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1750 if (!msr_mtrr_valid(msr))
1753 if (msr == MSR_IA32_CR_PAT) {
1754 for (i = 0; i < 8; i++)
1755 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1758 } else if (msr == MSR_MTRRdefType) {
1761 return valid_mtrr_type(data & 0xff);
1762 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1763 for (i = 0; i < 8 ; i++)
1764 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1769 /* variable MTRRs */
1770 return valid_mtrr_type(data & 0xff);
1773 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1775 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1777 if (!mtrr_valid(vcpu, msr, data))
1780 if (msr == MSR_MTRRdefType) {
1781 vcpu->arch.mtrr_state.def_type = data;
1782 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1783 } else if (msr == MSR_MTRRfix64K_00000)
1785 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1786 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1787 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1788 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1789 else if (msr == MSR_IA32_CR_PAT)
1790 vcpu->arch.pat = data;
1791 else { /* Variable MTRRs */
1792 int idx, is_mtrr_mask;
1795 idx = (msr - 0x200) / 2;
1796 is_mtrr_mask = msr - 0x200 - 2 * idx;
1799 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1802 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1806 kvm_mmu_reset_context(vcpu);
1810 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1812 u64 mcg_cap = vcpu->arch.mcg_cap;
1813 unsigned bank_num = mcg_cap & 0xff;
1816 case MSR_IA32_MCG_STATUS:
1817 vcpu->arch.mcg_status = data;
1819 case MSR_IA32_MCG_CTL:
1820 if (!(mcg_cap & MCG_CTL_P))
1822 if (data != 0 && data != ~(u64)0)
1824 vcpu->arch.mcg_ctl = data;
1827 if (msr >= MSR_IA32_MC0_CTL &&
1828 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1829 u32 offset = msr - MSR_IA32_MC0_CTL;
1830 /* only 0 or all 1s can be written to IA32_MCi_CTL
1831 * some Linux kernels though clear bit 10 in bank 4 to
1832 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1833 * this to avoid an uncatched #GP in the guest
1835 if ((offset & 0x3) == 0 &&
1836 data != 0 && (data | (1 << 10)) != ~(u64)0)
1838 vcpu->arch.mce_banks[offset] = data;
1846 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1848 struct kvm *kvm = vcpu->kvm;
1849 int lm = is_long_mode(vcpu);
1850 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1851 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1852 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1853 : kvm->arch.xen_hvm_config.blob_size_32;
1854 u32 page_num = data & ~PAGE_MASK;
1855 u64 page_addr = data & PAGE_MASK;
1860 if (page_num >= blob_size)
1863 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1868 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1877 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1879 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1882 static bool kvm_hv_msr_partition_wide(u32 msr)
1886 case HV_X64_MSR_GUEST_OS_ID:
1887 case HV_X64_MSR_HYPERCALL:
1888 case HV_X64_MSR_REFERENCE_TSC:
1889 case HV_X64_MSR_TIME_REF_COUNT:
1897 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1899 struct kvm *kvm = vcpu->kvm;
1902 case HV_X64_MSR_GUEST_OS_ID:
1903 kvm->arch.hv_guest_os_id = data;
1904 /* setting guest os id to zero disables hypercall page */
1905 if (!kvm->arch.hv_guest_os_id)
1906 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1908 case HV_X64_MSR_HYPERCALL: {
1913 /* if guest os id is not set hypercall should remain disabled */
1914 if (!kvm->arch.hv_guest_os_id)
1916 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1917 kvm->arch.hv_hypercall = data;
1920 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1921 addr = gfn_to_hva(kvm, gfn);
1922 if (kvm_is_error_hva(addr))
1924 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1925 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1926 if (__copy_to_user((void __user *)addr, instructions, 4))
1928 kvm->arch.hv_hypercall = data;
1929 mark_page_dirty(kvm, gfn);
1932 case HV_X64_MSR_REFERENCE_TSC: {
1934 HV_REFERENCE_TSC_PAGE tsc_ref;
1935 memset(&tsc_ref, 0, sizeof(tsc_ref));
1936 kvm->arch.hv_tsc_page = data;
1937 if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1939 gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1940 if (kvm_write_guest(kvm, gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
1941 &tsc_ref, sizeof(tsc_ref)))
1943 mark_page_dirty(kvm, gfn);
1947 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1948 "data 0x%llx\n", msr, data);
1954 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1957 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1961 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1962 vcpu->arch.hv_vapic = data;
1963 if (kvm_lapic_enable_pv_eoi(vcpu, 0))
1967 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
1968 addr = gfn_to_hva(vcpu->kvm, gfn);
1969 if (kvm_is_error_hva(addr))
1971 if (__clear_user((void __user *)addr, PAGE_SIZE))
1973 vcpu->arch.hv_vapic = data;
1974 mark_page_dirty(vcpu->kvm, gfn);
1975 if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
1979 case HV_X64_MSR_EOI:
1980 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1981 case HV_X64_MSR_ICR:
1982 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1983 case HV_X64_MSR_TPR:
1984 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1986 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1987 "data 0x%llx\n", msr, data);
1994 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1996 gpa_t gpa = data & ~0x3f;
1998 /* Bits 2:5 are reserved, Should be zero */
2002 vcpu->arch.apf.msr_val = data;
2004 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2005 kvm_clear_async_pf_completion_queue(vcpu);
2006 kvm_async_pf_hash_reset(vcpu);
2010 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2014 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2015 kvm_async_pf_wakeup_all(vcpu);
2019 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2021 vcpu->arch.pv_time_enabled = false;
2024 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2028 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2031 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2032 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2033 vcpu->arch.st.accum_steal = delta;
2036 static void record_steal_time(struct kvm_vcpu *vcpu)
2038 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2041 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2042 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2045 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2046 vcpu->arch.st.steal.version += 2;
2047 vcpu->arch.st.accum_steal = 0;
2049 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2050 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2053 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2056 u32 msr = msr_info->index;
2057 u64 data = msr_info->data;
2060 case MSR_AMD64_NB_CFG:
2061 case MSR_IA32_UCODE_REV:
2062 case MSR_IA32_UCODE_WRITE:
2063 case MSR_VM_HSAVE_PA:
2064 case MSR_AMD64_PATCH_LOADER:
2065 case MSR_AMD64_BU_CFG2:
2069 return set_efer(vcpu, data);
2071 data &= ~(u64)0x40; /* ignore flush filter disable */
2072 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2073 data &= ~(u64)0x8; /* ignore TLB cache disable */
2074 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2076 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2081 case MSR_FAM10H_MMIO_CONF_BASE:
2083 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2088 case MSR_IA32_DEBUGCTLMSR:
2090 /* We support the non-activated case already */
2092 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2093 /* Values other than LBR and BTF are vendor-specific,
2094 thus reserved and should throw a #GP */
2097 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2100 case 0x200 ... 0x2ff:
2101 return set_msr_mtrr(vcpu, msr, data);
2102 case MSR_IA32_APICBASE:
2103 return kvm_set_apic_base(vcpu, msr_info);
2104 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2105 return kvm_x2apic_msr_write(vcpu, msr, data);
2106 case MSR_IA32_TSCDEADLINE:
2107 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2109 case MSR_IA32_TSC_ADJUST:
2110 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2111 if (!msr_info->host_initiated) {
2112 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2113 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2115 vcpu->arch.ia32_tsc_adjust_msr = data;
2118 case MSR_IA32_MISC_ENABLE:
2119 vcpu->arch.ia32_misc_enable_msr = 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 kvmclock_reset(vcpu);
2131 vcpu->arch.time = data;
2132 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2134 /* we verify if the enable bit is set... */
2138 gpa_offset = data & ~(PAGE_MASK | 1);
2140 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2141 &vcpu->arch.pv_time, data & ~1ULL,
2142 sizeof(struct pvclock_vcpu_time_info)))
2143 vcpu->arch.pv_time_enabled = false;
2145 vcpu->arch.pv_time_enabled = true;
2149 case MSR_KVM_ASYNC_PF_EN:
2150 if (kvm_pv_enable_async_pf(vcpu, data))
2153 case MSR_KVM_STEAL_TIME:
2155 if (unlikely(!sched_info_on()))
2158 if (data & KVM_STEAL_RESERVED_MASK)
2161 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2162 data & KVM_STEAL_VALID_BITS,
2163 sizeof(struct kvm_steal_time)))
2166 vcpu->arch.st.msr_val = data;
2168 if (!(data & KVM_MSR_ENABLED))
2171 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2174 accumulate_steal_time(vcpu);
2177 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2180 case MSR_KVM_PV_EOI_EN:
2181 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2185 case MSR_IA32_MCG_CTL:
2186 case MSR_IA32_MCG_STATUS:
2187 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2188 return set_msr_mce(vcpu, msr, data);
2190 /* Performance counters are not protected by a CPUID bit,
2191 * so we should check all of them in the generic path for the sake of
2192 * cross vendor migration.
2193 * Writing a zero into the event select MSRs disables them,
2194 * which we perfectly emulate ;-). Any other value should be at least
2195 * reported, some guests depend on them.
2197 case MSR_K7_EVNTSEL0:
2198 case MSR_K7_EVNTSEL1:
2199 case MSR_K7_EVNTSEL2:
2200 case MSR_K7_EVNTSEL3:
2202 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2203 "0x%x data 0x%llx\n", msr, data);
2205 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2206 * so we ignore writes to make it happy.
2208 case MSR_K7_PERFCTR0:
2209 case MSR_K7_PERFCTR1:
2210 case MSR_K7_PERFCTR2:
2211 case MSR_K7_PERFCTR3:
2212 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2213 "0x%x data 0x%llx\n", msr, data);
2215 case MSR_P6_PERFCTR0:
2216 case MSR_P6_PERFCTR1:
2218 case MSR_P6_EVNTSEL0:
2219 case MSR_P6_EVNTSEL1:
2220 if (kvm_pmu_msr(vcpu, msr))
2221 return kvm_pmu_set_msr(vcpu, msr_info);
2223 if (pr || data != 0)
2224 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2225 "0x%x data 0x%llx\n", msr, data);
2227 case MSR_K7_CLK_CTL:
2229 * Ignore all writes to this no longer documented MSR.
2230 * Writes are only relevant for old K7 processors,
2231 * all pre-dating SVM, but a recommended workaround from
2232 * AMD for these chips. It is possible to specify the
2233 * affected processor models on the command line, hence
2234 * the need to ignore the workaround.
2237 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2238 if (kvm_hv_msr_partition_wide(msr)) {
2240 mutex_lock(&vcpu->kvm->lock);
2241 r = set_msr_hyperv_pw(vcpu, msr, data);
2242 mutex_unlock(&vcpu->kvm->lock);
2245 return set_msr_hyperv(vcpu, msr, data);
2247 case MSR_IA32_BBL_CR_CTL3:
2248 /* Drop writes to this legacy MSR -- see rdmsr
2249 * counterpart for further detail.
2251 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2253 case MSR_AMD64_OSVW_ID_LENGTH:
2254 if (!guest_cpuid_has_osvw(vcpu))
2256 vcpu->arch.osvw.length = data;
2258 case MSR_AMD64_OSVW_STATUS:
2259 if (!guest_cpuid_has_osvw(vcpu))
2261 vcpu->arch.osvw.status = data;
2264 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2265 return xen_hvm_config(vcpu, data);
2266 if (kvm_pmu_msr(vcpu, msr))
2267 return kvm_pmu_set_msr(vcpu, msr_info);
2269 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2273 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2280 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2284 * Reads an msr value (of 'msr_index') into 'pdata'.
2285 * Returns 0 on success, non-0 otherwise.
2286 * Assumes vcpu_load() was already called.
2288 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2290 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2293 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2295 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2297 if (!msr_mtrr_valid(msr))
2300 if (msr == MSR_MTRRdefType)
2301 *pdata = vcpu->arch.mtrr_state.def_type +
2302 (vcpu->arch.mtrr_state.enabled << 10);
2303 else if (msr == MSR_MTRRfix64K_00000)
2305 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2306 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2307 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2308 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2309 else if (msr == MSR_IA32_CR_PAT)
2310 *pdata = vcpu->arch.pat;
2311 else { /* Variable MTRRs */
2312 int idx, is_mtrr_mask;
2315 idx = (msr - 0x200) / 2;
2316 is_mtrr_mask = msr - 0x200 - 2 * idx;
2319 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2322 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2329 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2332 u64 mcg_cap = vcpu->arch.mcg_cap;
2333 unsigned bank_num = mcg_cap & 0xff;
2336 case MSR_IA32_P5_MC_ADDR:
2337 case MSR_IA32_P5_MC_TYPE:
2340 case MSR_IA32_MCG_CAP:
2341 data = vcpu->arch.mcg_cap;
2343 case MSR_IA32_MCG_CTL:
2344 if (!(mcg_cap & MCG_CTL_P))
2346 data = vcpu->arch.mcg_ctl;
2348 case MSR_IA32_MCG_STATUS:
2349 data = vcpu->arch.mcg_status;
2352 if (msr >= MSR_IA32_MC0_CTL &&
2353 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2354 u32 offset = msr - MSR_IA32_MC0_CTL;
2355 data = vcpu->arch.mce_banks[offset];
2364 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2367 struct kvm *kvm = vcpu->kvm;
2370 case HV_X64_MSR_GUEST_OS_ID:
2371 data = kvm->arch.hv_guest_os_id;
2373 case HV_X64_MSR_HYPERCALL:
2374 data = kvm->arch.hv_hypercall;
2376 case HV_X64_MSR_TIME_REF_COUNT: {
2378 div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2381 case HV_X64_MSR_REFERENCE_TSC:
2382 data = kvm->arch.hv_tsc_page;
2385 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2393 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2398 case HV_X64_MSR_VP_INDEX: {
2401 kvm_for_each_vcpu(r, v, vcpu->kvm) {
2409 case HV_X64_MSR_EOI:
2410 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2411 case HV_X64_MSR_ICR:
2412 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2413 case HV_X64_MSR_TPR:
2414 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2415 case HV_X64_MSR_APIC_ASSIST_PAGE:
2416 data = vcpu->arch.hv_vapic;
2419 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2426 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2431 case MSR_IA32_PLATFORM_ID:
2432 case MSR_IA32_EBL_CR_POWERON:
2433 case MSR_IA32_DEBUGCTLMSR:
2434 case MSR_IA32_LASTBRANCHFROMIP:
2435 case MSR_IA32_LASTBRANCHTOIP:
2436 case MSR_IA32_LASTINTFROMIP:
2437 case MSR_IA32_LASTINTTOIP:
2440 case MSR_VM_HSAVE_PA:
2441 case MSR_K7_EVNTSEL0:
2442 case MSR_K7_PERFCTR0:
2443 case MSR_K8_INT_PENDING_MSG:
2444 case MSR_AMD64_NB_CFG:
2445 case MSR_FAM10H_MMIO_CONF_BASE:
2446 case MSR_AMD64_BU_CFG2:
2449 case MSR_P6_PERFCTR0:
2450 case MSR_P6_PERFCTR1:
2451 case MSR_P6_EVNTSEL0:
2452 case MSR_P6_EVNTSEL1:
2453 if (kvm_pmu_msr(vcpu, msr))
2454 return kvm_pmu_get_msr(vcpu, msr, pdata);
2457 case MSR_IA32_UCODE_REV:
2458 data = 0x100000000ULL;
2461 data = 0x500 | KVM_NR_VAR_MTRR;
2463 case 0x200 ... 0x2ff:
2464 return get_msr_mtrr(vcpu, msr, pdata);
2465 case 0xcd: /* fsb frequency */
2469 * MSR_EBC_FREQUENCY_ID
2470 * Conservative value valid for even the basic CPU models.
2471 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2472 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2473 * and 266MHz for model 3, or 4. Set Core Clock
2474 * Frequency to System Bus Frequency Ratio to 1 (bits
2475 * 31:24) even though these are only valid for CPU
2476 * models > 2, however guests may end up dividing or
2477 * multiplying by zero otherwise.
2479 case MSR_EBC_FREQUENCY_ID:
2482 case MSR_IA32_APICBASE:
2483 data = kvm_get_apic_base(vcpu);
2485 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2486 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2488 case MSR_IA32_TSCDEADLINE:
2489 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2491 case MSR_IA32_TSC_ADJUST:
2492 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2494 case MSR_IA32_MISC_ENABLE:
2495 data = vcpu->arch.ia32_misc_enable_msr;
2497 case MSR_IA32_PERF_STATUS:
2498 /* TSC increment by tick */
2500 /* CPU multiplier */
2501 data |= (((uint64_t)4ULL) << 40);
2504 data = vcpu->arch.efer;
2506 case MSR_KVM_WALL_CLOCK:
2507 case MSR_KVM_WALL_CLOCK_NEW:
2508 data = vcpu->kvm->arch.wall_clock;
2510 case MSR_KVM_SYSTEM_TIME:
2511 case MSR_KVM_SYSTEM_TIME_NEW:
2512 data = vcpu->arch.time;
2514 case MSR_KVM_ASYNC_PF_EN:
2515 data = vcpu->arch.apf.msr_val;
2517 case MSR_KVM_STEAL_TIME:
2518 data = vcpu->arch.st.msr_val;
2520 case MSR_KVM_PV_EOI_EN:
2521 data = vcpu->arch.pv_eoi.msr_val;
2523 case MSR_IA32_P5_MC_ADDR:
2524 case MSR_IA32_P5_MC_TYPE:
2525 case MSR_IA32_MCG_CAP:
2526 case MSR_IA32_MCG_CTL:
2527 case MSR_IA32_MCG_STATUS:
2528 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2529 return get_msr_mce(vcpu, msr, pdata);
2530 case MSR_K7_CLK_CTL:
2532 * Provide expected ramp-up count for K7. All other
2533 * are set to zero, indicating minimum divisors for
2536 * This prevents guest kernels on AMD host with CPU
2537 * type 6, model 8 and higher from exploding due to
2538 * the rdmsr failing.
2542 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2543 if (kvm_hv_msr_partition_wide(msr)) {
2545 mutex_lock(&vcpu->kvm->lock);
2546 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2547 mutex_unlock(&vcpu->kvm->lock);
2550 return get_msr_hyperv(vcpu, msr, pdata);
2552 case MSR_IA32_BBL_CR_CTL3:
2553 /* This legacy MSR exists but isn't fully documented in current
2554 * silicon. It is however accessed by winxp in very narrow
2555 * scenarios where it sets bit #19, itself documented as
2556 * a "reserved" bit. Best effort attempt to source coherent
2557 * read data here should the balance of the register be
2558 * interpreted by the guest:
2560 * L2 cache control register 3: 64GB range, 256KB size,
2561 * enabled, latency 0x1, configured
2565 case MSR_AMD64_OSVW_ID_LENGTH:
2566 if (!guest_cpuid_has_osvw(vcpu))
2568 data = vcpu->arch.osvw.length;
2570 case MSR_AMD64_OSVW_STATUS:
2571 if (!guest_cpuid_has_osvw(vcpu))
2573 data = vcpu->arch.osvw.status;
2576 if (kvm_pmu_msr(vcpu, msr))
2577 return kvm_pmu_get_msr(vcpu, msr, pdata);
2579 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2582 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2590 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2593 * Read or write a bunch of msrs. All parameters are kernel addresses.
2595 * @return number of msrs set successfully.
2597 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2598 struct kvm_msr_entry *entries,
2599 int (*do_msr)(struct kvm_vcpu *vcpu,
2600 unsigned index, u64 *data))
2604 idx = srcu_read_lock(&vcpu->kvm->srcu);
2605 for (i = 0; i < msrs->nmsrs; ++i)
2606 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2608 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2614 * Read or write a bunch of msrs. Parameters are user addresses.
2616 * @return number of msrs set successfully.
2618 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2619 int (*do_msr)(struct kvm_vcpu *vcpu,
2620 unsigned index, u64 *data),
2623 struct kvm_msrs msrs;
2624 struct kvm_msr_entry *entries;
2629 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2633 if (msrs.nmsrs >= MAX_IO_MSRS)
2636 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2637 entries = memdup_user(user_msrs->entries, size);
2638 if (IS_ERR(entries)) {
2639 r = PTR_ERR(entries);
2643 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2648 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2659 int kvm_dev_ioctl_check_extension(long ext)
2664 case KVM_CAP_IRQCHIP:
2666 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2667 case KVM_CAP_SET_TSS_ADDR:
2668 case KVM_CAP_EXT_CPUID:
2669 case KVM_CAP_EXT_EMUL_CPUID:
2670 case KVM_CAP_CLOCKSOURCE:
2672 case KVM_CAP_NOP_IO_DELAY:
2673 case KVM_CAP_MP_STATE:
2674 case KVM_CAP_SYNC_MMU:
2675 case KVM_CAP_USER_NMI:
2676 case KVM_CAP_REINJECT_CONTROL:
2677 case KVM_CAP_IRQ_INJECT_STATUS:
2679 case KVM_CAP_IOEVENTFD:
2680 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2682 case KVM_CAP_PIT_STATE2:
2683 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2684 case KVM_CAP_XEN_HVM:
2685 case KVM_CAP_ADJUST_CLOCK:
2686 case KVM_CAP_VCPU_EVENTS:
2687 case KVM_CAP_HYPERV:
2688 case KVM_CAP_HYPERV_VAPIC:
2689 case KVM_CAP_HYPERV_SPIN:
2690 case KVM_CAP_PCI_SEGMENT:
2691 case KVM_CAP_DEBUGREGS:
2692 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2694 case KVM_CAP_ASYNC_PF:
2695 case KVM_CAP_GET_TSC_KHZ:
2696 case KVM_CAP_KVMCLOCK_CTRL:
2697 case KVM_CAP_READONLY_MEM:
2698 case KVM_CAP_HYPERV_TIME:
2699 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2700 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2701 case KVM_CAP_ASSIGN_DEV_IRQ:
2702 case KVM_CAP_PCI_2_3:
2706 case KVM_CAP_COALESCED_MMIO:
2707 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2710 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2712 case KVM_CAP_NR_VCPUS:
2713 r = KVM_SOFT_MAX_VCPUS;
2715 case KVM_CAP_MAX_VCPUS:
2718 case KVM_CAP_NR_MEMSLOTS:
2719 r = KVM_USER_MEM_SLOTS;
2721 case KVM_CAP_PV_MMU: /* obsolete */
2724 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2726 r = iommu_present(&pci_bus_type);
2730 r = KVM_MAX_MCE_BANKS;
2735 case KVM_CAP_TSC_CONTROL:
2736 r = kvm_has_tsc_control;
2738 case KVM_CAP_TSC_DEADLINE_TIMER:
2739 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2749 long kvm_arch_dev_ioctl(struct file *filp,
2750 unsigned int ioctl, unsigned long arg)
2752 void __user *argp = (void __user *)arg;
2756 case KVM_GET_MSR_INDEX_LIST: {
2757 struct kvm_msr_list __user *user_msr_list = argp;
2758 struct kvm_msr_list msr_list;
2762 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2765 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2766 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2769 if (n < msr_list.nmsrs)
2772 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2773 num_msrs_to_save * sizeof(u32)))
2775 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2777 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2782 case KVM_GET_SUPPORTED_CPUID:
2783 case KVM_GET_EMULATED_CPUID: {
2784 struct kvm_cpuid2 __user *cpuid_arg = argp;
2785 struct kvm_cpuid2 cpuid;
2788 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2791 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2797 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2802 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2805 mce_cap = KVM_MCE_CAP_SUPPORTED;
2807 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2819 static void wbinvd_ipi(void *garbage)
2824 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2826 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2829 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2831 /* Address WBINVD may be executed by guest */
2832 if (need_emulate_wbinvd(vcpu)) {
2833 if (kvm_x86_ops->has_wbinvd_exit())
2834 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2835 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2836 smp_call_function_single(vcpu->cpu,
2837 wbinvd_ipi, NULL, 1);
2840 kvm_x86_ops->vcpu_load(vcpu, cpu);
2842 /* Apply any externally detected TSC adjustments (due to suspend) */
2843 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2844 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2845 vcpu->arch.tsc_offset_adjustment = 0;
2846 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2849 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2850 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2851 native_read_tsc() - vcpu->arch.last_host_tsc;
2853 mark_tsc_unstable("KVM discovered backwards TSC");
2854 if (check_tsc_unstable()) {
2855 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2856 vcpu->arch.last_guest_tsc);
2857 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2858 vcpu->arch.tsc_catchup = 1;
2861 * On a host with synchronized TSC, there is no need to update
2862 * kvmclock on vcpu->cpu migration
2864 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2865 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2866 if (vcpu->cpu != cpu)
2867 kvm_migrate_timers(vcpu);
2871 accumulate_steal_time(vcpu);
2872 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2875 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2877 kvm_x86_ops->vcpu_put(vcpu);
2878 kvm_put_guest_fpu(vcpu);
2879 vcpu->arch.last_host_tsc = native_read_tsc();
2882 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2883 struct kvm_lapic_state *s)
2885 kvm_x86_ops->sync_pir_to_irr(vcpu);
2886 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2891 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2892 struct kvm_lapic_state *s)
2894 kvm_apic_post_state_restore(vcpu, s);
2895 update_cr8_intercept(vcpu);
2900 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2901 struct kvm_interrupt *irq)
2903 if (irq->irq >= KVM_NR_INTERRUPTS)
2905 if (irqchip_in_kernel(vcpu->kvm))
2908 kvm_queue_interrupt(vcpu, irq->irq, false);
2909 kvm_make_request(KVM_REQ_EVENT, vcpu);
2914 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2916 kvm_inject_nmi(vcpu);
2921 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2922 struct kvm_tpr_access_ctl *tac)
2926 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2930 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2934 unsigned bank_num = mcg_cap & 0xff, bank;
2937 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2939 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2942 vcpu->arch.mcg_cap = mcg_cap;
2943 /* Init IA32_MCG_CTL to all 1s */
2944 if (mcg_cap & MCG_CTL_P)
2945 vcpu->arch.mcg_ctl = ~(u64)0;
2946 /* Init IA32_MCi_CTL to all 1s */
2947 for (bank = 0; bank < bank_num; bank++)
2948 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2953 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2954 struct kvm_x86_mce *mce)
2956 u64 mcg_cap = vcpu->arch.mcg_cap;
2957 unsigned bank_num = mcg_cap & 0xff;
2958 u64 *banks = vcpu->arch.mce_banks;
2960 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2963 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2964 * reporting is disabled
2966 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2967 vcpu->arch.mcg_ctl != ~(u64)0)
2969 banks += 4 * mce->bank;
2971 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2972 * reporting is disabled for the bank
2974 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2976 if (mce->status & MCI_STATUS_UC) {
2977 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2978 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2979 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2982 if (banks[1] & MCI_STATUS_VAL)
2983 mce->status |= MCI_STATUS_OVER;
2984 banks[2] = mce->addr;
2985 banks[3] = mce->misc;
2986 vcpu->arch.mcg_status = mce->mcg_status;
2987 banks[1] = mce->status;
2988 kvm_queue_exception(vcpu, MC_VECTOR);
2989 } else if (!(banks[1] & MCI_STATUS_VAL)
2990 || !(banks[1] & MCI_STATUS_UC)) {
2991 if (banks[1] & MCI_STATUS_VAL)
2992 mce->status |= MCI_STATUS_OVER;
2993 banks[2] = mce->addr;
2994 banks[3] = mce->misc;
2995 banks[1] = mce->status;
2997 banks[1] |= MCI_STATUS_OVER;
3001 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3002 struct kvm_vcpu_events *events)
3005 events->exception.injected =
3006 vcpu->arch.exception.pending &&
3007 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3008 events->exception.nr = vcpu->arch.exception.nr;
3009 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3010 events->exception.pad = 0;
3011 events->exception.error_code = vcpu->arch.exception.error_code;
3013 events->interrupt.injected =
3014 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3015 events->interrupt.nr = vcpu->arch.interrupt.nr;
3016 events->interrupt.soft = 0;
3017 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3019 events->nmi.injected = vcpu->arch.nmi_injected;
3020 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3021 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3022 events->nmi.pad = 0;
3024 events->sipi_vector = 0; /* never valid when reporting to user space */
3026 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3027 | KVM_VCPUEVENT_VALID_SHADOW);
3028 memset(&events->reserved, 0, sizeof(events->reserved));
3031 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3032 struct kvm_vcpu_events *events)
3034 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3035 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3036 | KVM_VCPUEVENT_VALID_SHADOW))
3040 vcpu->arch.exception.pending = events->exception.injected;
3041 vcpu->arch.exception.nr = events->exception.nr;
3042 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3043 vcpu->arch.exception.error_code = events->exception.error_code;
3045 vcpu->arch.interrupt.pending = events->interrupt.injected;
3046 vcpu->arch.interrupt.nr = events->interrupt.nr;
3047 vcpu->arch.interrupt.soft = events->interrupt.soft;
3048 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3049 kvm_x86_ops->set_interrupt_shadow(vcpu,
3050 events->interrupt.shadow);
3052 vcpu->arch.nmi_injected = events->nmi.injected;
3053 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3054 vcpu->arch.nmi_pending = events->nmi.pending;
3055 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3057 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3058 kvm_vcpu_has_lapic(vcpu))
3059 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3061 kvm_make_request(KVM_REQ_EVENT, vcpu);
3066 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3067 struct kvm_debugregs *dbgregs)
3071 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3072 _kvm_get_dr(vcpu, 6, &val);
3074 dbgregs->dr7 = vcpu->arch.dr7;
3076 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3079 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3080 struct kvm_debugregs *dbgregs)
3085 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3086 vcpu->arch.dr6 = dbgregs->dr6;
3087 kvm_update_dr6(vcpu);
3088 vcpu->arch.dr7 = dbgregs->dr7;
3089 kvm_update_dr7(vcpu);
3094 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3095 struct kvm_xsave *guest_xsave)
3097 if (cpu_has_xsave) {
3098 memcpy(guest_xsave->region,
3099 &vcpu->arch.guest_fpu.state->xsave,
3100 vcpu->arch.guest_xstate_size);
3101 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] &=
3102 vcpu->arch.guest_supported_xcr0 | XSTATE_FPSSE;
3104 memcpy(guest_xsave->region,
3105 &vcpu->arch.guest_fpu.state->fxsave,
3106 sizeof(struct i387_fxsave_struct));
3107 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3112 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3113 struct kvm_xsave *guest_xsave)
3116 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3118 if (cpu_has_xsave) {
3120 * Here we allow setting states that are not present in
3121 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3122 * with old userspace.
3124 if (xstate_bv & ~kvm_supported_xcr0())
3126 memcpy(&vcpu->arch.guest_fpu.state->xsave,
3127 guest_xsave->region, vcpu->arch.guest_xstate_size);
3129 if (xstate_bv & ~XSTATE_FPSSE)
3131 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3132 guest_xsave->region, sizeof(struct i387_fxsave_struct));
3137 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3138 struct kvm_xcrs *guest_xcrs)
3140 if (!cpu_has_xsave) {
3141 guest_xcrs->nr_xcrs = 0;
3145 guest_xcrs->nr_xcrs = 1;
3146 guest_xcrs->flags = 0;
3147 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3148 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3151 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3152 struct kvm_xcrs *guest_xcrs)
3159 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3162 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3163 /* Only support XCR0 currently */
3164 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3165 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3166 guest_xcrs->xcrs[i].value);
3175 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3176 * stopped by the hypervisor. This function will be called from the host only.
3177 * EINVAL is returned when the host attempts to set the flag for a guest that
3178 * does not support pv clocks.
3180 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3182 if (!vcpu->arch.pv_time_enabled)
3184 vcpu->arch.pvclock_set_guest_stopped_request = true;
3185 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3189 long kvm_arch_vcpu_ioctl(struct file *filp,
3190 unsigned int ioctl, unsigned long arg)
3192 struct kvm_vcpu *vcpu = filp->private_data;
3193 void __user *argp = (void __user *)arg;
3196 struct kvm_lapic_state *lapic;
3197 struct kvm_xsave *xsave;
3198 struct kvm_xcrs *xcrs;
3204 case KVM_GET_LAPIC: {
3206 if (!vcpu->arch.apic)
3208 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3213 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3217 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3222 case KVM_SET_LAPIC: {
3224 if (!vcpu->arch.apic)
3226 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3227 if (IS_ERR(u.lapic))
3228 return PTR_ERR(u.lapic);
3230 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3233 case KVM_INTERRUPT: {
3234 struct kvm_interrupt irq;
3237 if (copy_from_user(&irq, argp, sizeof irq))
3239 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3243 r = kvm_vcpu_ioctl_nmi(vcpu);
3246 case KVM_SET_CPUID: {
3247 struct kvm_cpuid __user *cpuid_arg = argp;
3248 struct kvm_cpuid cpuid;
3251 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3253 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3256 case KVM_SET_CPUID2: {
3257 struct kvm_cpuid2 __user *cpuid_arg = argp;
3258 struct kvm_cpuid2 cpuid;
3261 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3263 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3264 cpuid_arg->entries);
3267 case KVM_GET_CPUID2: {
3268 struct kvm_cpuid2 __user *cpuid_arg = argp;
3269 struct kvm_cpuid2 cpuid;
3272 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3274 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3275 cpuid_arg->entries);
3279 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3285 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3288 r = msr_io(vcpu, argp, do_set_msr, 0);
3290 case KVM_TPR_ACCESS_REPORTING: {
3291 struct kvm_tpr_access_ctl tac;
3294 if (copy_from_user(&tac, argp, sizeof tac))
3296 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3300 if (copy_to_user(argp, &tac, sizeof tac))
3305 case KVM_SET_VAPIC_ADDR: {
3306 struct kvm_vapic_addr va;
3309 if (!irqchip_in_kernel(vcpu->kvm))
3312 if (copy_from_user(&va, argp, sizeof va))
3314 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3317 case KVM_X86_SETUP_MCE: {
3321 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3323 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3326 case KVM_X86_SET_MCE: {
3327 struct kvm_x86_mce mce;
3330 if (copy_from_user(&mce, argp, sizeof mce))
3332 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3335 case KVM_GET_VCPU_EVENTS: {
3336 struct kvm_vcpu_events events;
3338 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3341 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3346 case KVM_SET_VCPU_EVENTS: {
3347 struct kvm_vcpu_events events;
3350 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3353 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3356 case KVM_GET_DEBUGREGS: {
3357 struct kvm_debugregs dbgregs;
3359 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3362 if (copy_to_user(argp, &dbgregs,
3363 sizeof(struct kvm_debugregs)))
3368 case KVM_SET_DEBUGREGS: {
3369 struct kvm_debugregs dbgregs;
3372 if (copy_from_user(&dbgregs, argp,
3373 sizeof(struct kvm_debugregs)))
3376 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3379 case KVM_GET_XSAVE: {
3380 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3385 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3388 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3393 case KVM_SET_XSAVE: {
3394 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3395 if (IS_ERR(u.xsave))
3396 return PTR_ERR(u.xsave);
3398 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3401 case KVM_GET_XCRS: {
3402 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3407 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3410 if (copy_to_user(argp, u.xcrs,
3411 sizeof(struct kvm_xcrs)))
3416 case KVM_SET_XCRS: {
3417 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3419 return PTR_ERR(u.xcrs);
3421 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3424 case KVM_SET_TSC_KHZ: {
3428 user_tsc_khz = (u32)arg;
3430 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3433 if (user_tsc_khz == 0)
3434 user_tsc_khz = tsc_khz;
3436 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3441 case KVM_GET_TSC_KHZ: {
3442 r = vcpu->arch.virtual_tsc_khz;
3445 case KVM_KVMCLOCK_CTRL: {
3446 r = kvm_set_guest_paused(vcpu);
3457 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3459 return VM_FAULT_SIGBUS;
3462 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3466 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3468 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3472 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3475 kvm->arch.ept_identity_map_addr = ident_addr;
3479 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3480 u32 kvm_nr_mmu_pages)
3482 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3485 mutex_lock(&kvm->slots_lock);
3487 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3488 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3490 mutex_unlock(&kvm->slots_lock);
3494 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3496 return kvm->arch.n_max_mmu_pages;
3499 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3504 switch (chip->chip_id) {
3505 case KVM_IRQCHIP_PIC_MASTER:
3506 memcpy(&chip->chip.pic,
3507 &pic_irqchip(kvm)->pics[0],
3508 sizeof(struct kvm_pic_state));
3510 case KVM_IRQCHIP_PIC_SLAVE:
3511 memcpy(&chip->chip.pic,
3512 &pic_irqchip(kvm)->pics[1],
3513 sizeof(struct kvm_pic_state));
3515 case KVM_IRQCHIP_IOAPIC:
3516 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3525 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3530 switch (chip->chip_id) {
3531 case KVM_IRQCHIP_PIC_MASTER:
3532 spin_lock(&pic_irqchip(kvm)->lock);
3533 memcpy(&pic_irqchip(kvm)->pics[0],
3535 sizeof(struct kvm_pic_state));
3536 spin_unlock(&pic_irqchip(kvm)->lock);
3538 case KVM_IRQCHIP_PIC_SLAVE:
3539 spin_lock(&pic_irqchip(kvm)->lock);
3540 memcpy(&pic_irqchip(kvm)->pics[1],
3542 sizeof(struct kvm_pic_state));
3543 spin_unlock(&pic_irqchip(kvm)->lock);
3545 case KVM_IRQCHIP_IOAPIC:
3546 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3552 kvm_pic_update_irq(pic_irqchip(kvm));
3556 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3560 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3561 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3562 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3566 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3570 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3571 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3572 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3573 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3577 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3581 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3582 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3583 sizeof(ps->channels));
3584 ps->flags = kvm->arch.vpit->pit_state.flags;
3585 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3586 memset(&ps->reserved, 0, sizeof(ps->reserved));
3590 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3592 int r = 0, start = 0;
3593 u32 prev_legacy, cur_legacy;
3594 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3595 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3596 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3597 if (!prev_legacy && cur_legacy)
3599 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3600 sizeof(kvm->arch.vpit->pit_state.channels));
3601 kvm->arch.vpit->pit_state.flags = ps->flags;
3602 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3603 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3607 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3608 struct kvm_reinject_control *control)
3610 if (!kvm->arch.vpit)
3612 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3613 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3614 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3619 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3620 * @kvm: kvm instance
3621 * @log: slot id and address to which we copy the log
3623 * We need to keep it in mind that VCPU threads can write to the bitmap
3624 * concurrently. So, to avoid losing data, we keep the following order for
3627 * 1. Take a snapshot of the bit and clear it if needed.
3628 * 2. Write protect the corresponding page.
3629 * 3. Flush TLB's if needed.
3630 * 4. Copy the snapshot to the userspace.
3632 * Between 2 and 3, the guest may write to the page using the remaining TLB
3633 * entry. This is not a problem because the page will be reported dirty at
3634 * step 4 using the snapshot taken before and step 3 ensures that successive
3635 * writes will be logged for the next call.
3637 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3640 struct kvm_memory_slot *memslot;
3642 unsigned long *dirty_bitmap;
3643 unsigned long *dirty_bitmap_buffer;
3644 bool is_dirty = false;
3646 mutex_lock(&kvm->slots_lock);
3649 if (log->slot >= KVM_USER_MEM_SLOTS)
3652 memslot = id_to_memslot(kvm->memslots, log->slot);
3654 dirty_bitmap = memslot->dirty_bitmap;
3659 n = kvm_dirty_bitmap_bytes(memslot);
3661 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3662 memset(dirty_bitmap_buffer, 0, n);
3664 spin_lock(&kvm->mmu_lock);
3666 for (i = 0; i < n / sizeof(long); i++) {
3670 if (!dirty_bitmap[i])
3675 mask = xchg(&dirty_bitmap[i], 0);
3676 dirty_bitmap_buffer[i] = mask;
3678 offset = i * BITS_PER_LONG;
3679 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3682 spin_unlock(&kvm->mmu_lock);
3684 /* See the comments in kvm_mmu_slot_remove_write_access(). */
3685 lockdep_assert_held(&kvm->slots_lock);
3688 * All the TLBs can be flushed out of mmu lock, see the comments in
3689 * kvm_mmu_slot_remove_write_access().
3692 kvm_flush_remote_tlbs(kvm);
3695 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3700 mutex_unlock(&kvm->slots_lock);
3704 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3707 if (!irqchip_in_kernel(kvm))
3710 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3711 irq_event->irq, irq_event->level,
3716 long kvm_arch_vm_ioctl(struct file *filp,
3717 unsigned int ioctl, unsigned long arg)
3719 struct kvm *kvm = filp->private_data;
3720 void __user *argp = (void __user *)arg;
3723 * This union makes it completely explicit to gcc-3.x
3724 * that these two variables' stack usage should be
3725 * combined, not added together.
3728 struct kvm_pit_state ps;
3729 struct kvm_pit_state2 ps2;
3730 struct kvm_pit_config pit_config;
3734 case KVM_SET_TSS_ADDR:
3735 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3737 case KVM_SET_IDENTITY_MAP_ADDR: {
3741 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3743 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3746 case KVM_SET_NR_MMU_PAGES:
3747 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3749 case KVM_GET_NR_MMU_PAGES:
3750 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3752 case KVM_CREATE_IRQCHIP: {
3753 struct kvm_pic *vpic;
3755 mutex_lock(&kvm->lock);
3758 goto create_irqchip_unlock;
3760 if (atomic_read(&kvm->online_vcpus))
3761 goto create_irqchip_unlock;
3763 vpic = kvm_create_pic(kvm);
3765 r = kvm_ioapic_init(kvm);
3767 mutex_lock(&kvm->slots_lock);
3768 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3770 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3772 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3774 mutex_unlock(&kvm->slots_lock);
3776 goto create_irqchip_unlock;
3779 goto create_irqchip_unlock;
3781 kvm->arch.vpic = vpic;
3783 r = kvm_setup_default_irq_routing(kvm);
3785 mutex_lock(&kvm->slots_lock);
3786 mutex_lock(&kvm->irq_lock);
3787 kvm_ioapic_destroy(kvm);
3788 kvm_destroy_pic(kvm);
3789 mutex_unlock(&kvm->irq_lock);
3790 mutex_unlock(&kvm->slots_lock);
3792 create_irqchip_unlock:
3793 mutex_unlock(&kvm->lock);
3796 case KVM_CREATE_PIT:
3797 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3799 case KVM_CREATE_PIT2:
3801 if (copy_from_user(&u.pit_config, argp,
3802 sizeof(struct kvm_pit_config)))
3805 mutex_lock(&kvm->slots_lock);
3808 goto create_pit_unlock;
3810 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3814 mutex_unlock(&kvm->slots_lock);
3816 case KVM_GET_IRQCHIP: {
3817 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3818 struct kvm_irqchip *chip;
3820 chip = memdup_user(argp, sizeof(*chip));
3827 if (!irqchip_in_kernel(kvm))
3828 goto get_irqchip_out;
3829 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3831 goto get_irqchip_out;
3833 if (copy_to_user(argp, chip, sizeof *chip))
3834 goto get_irqchip_out;
3840 case KVM_SET_IRQCHIP: {
3841 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3842 struct kvm_irqchip *chip;
3844 chip = memdup_user(argp, sizeof(*chip));
3851 if (!irqchip_in_kernel(kvm))
3852 goto set_irqchip_out;
3853 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3855 goto set_irqchip_out;
3863 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3866 if (!kvm->arch.vpit)
3868 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3872 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3879 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3882 if (!kvm->arch.vpit)
3884 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3887 case KVM_GET_PIT2: {
3889 if (!kvm->arch.vpit)
3891 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3895 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3900 case KVM_SET_PIT2: {
3902 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3905 if (!kvm->arch.vpit)
3907 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3910 case KVM_REINJECT_CONTROL: {
3911 struct kvm_reinject_control control;
3913 if (copy_from_user(&control, argp, sizeof(control)))
3915 r = kvm_vm_ioctl_reinject(kvm, &control);
3918 case KVM_XEN_HVM_CONFIG: {
3920 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3921 sizeof(struct kvm_xen_hvm_config)))
3924 if (kvm->arch.xen_hvm_config.flags)
3929 case KVM_SET_CLOCK: {
3930 struct kvm_clock_data user_ns;
3935 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3943 local_irq_disable();
3944 now_ns = get_kernel_ns();
3945 delta = user_ns.clock - now_ns;
3947 kvm->arch.kvmclock_offset = delta;
3948 kvm_gen_update_masterclock(kvm);
3951 case KVM_GET_CLOCK: {
3952 struct kvm_clock_data user_ns;
3955 local_irq_disable();
3956 now_ns = get_kernel_ns();
3957 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3960 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3963 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3976 static void kvm_init_msr_list(void)
3981 /* skip the first msrs in the list. KVM-specific */
3982 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3983 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3987 * Even MSRs that are valid in the host may not be exposed
3988 * to the guests in some cases. We could work around this
3989 * in VMX with the generic MSR save/load machinery, but it
3990 * is not really worthwhile since it will really only
3991 * happen with nested virtualization.
3993 switch (msrs_to_save[i]) {
3994 case MSR_IA32_BNDCFGS:
3995 if (!kvm_x86_ops->mpx_supported())
4003 msrs_to_save[j] = msrs_to_save[i];
4006 num_msrs_to_save = j;
4009 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4017 if (!(vcpu->arch.apic &&
4018 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
4019 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
4030 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4037 if (!(vcpu->arch.apic &&
4038 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
4039 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
4041 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4051 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4052 struct kvm_segment *var, int seg)
4054 kvm_x86_ops->set_segment(vcpu, var, seg);
4057 void kvm_get_segment(struct kvm_vcpu *vcpu,
4058 struct kvm_segment *var, int seg)
4060 kvm_x86_ops->get_segment(vcpu, var, seg);
4063 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
4066 struct x86_exception exception;
4068 BUG_ON(!mmu_is_nested(vcpu));
4070 /* NPT walks are always user-walks */
4071 access |= PFERR_USER_MASK;
4072 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
4077 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4078 struct x86_exception *exception)
4080 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4081 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4084 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4085 struct x86_exception *exception)
4087 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4088 access |= PFERR_FETCH_MASK;
4089 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4092 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4093 struct x86_exception *exception)
4095 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4096 access |= PFERR_WRITE_MASK;
4097 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4100 /* uses this to access any guest's mapped memory without checking CPL */
4101 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4102 struct x86_exception *exception)
4104 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4107 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4108 struct kvm_vcpu *vcpu, u32 access,
4109 struct x86_exception *exception)
4112 int r = X86EMUL_CONTINUE;
4115 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4117 unsigned offset = addr & (PAGE_SIZE-1);
4118 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4121 if (gpa == UNMAPPED_GVA)
4122 return X86EMUL_PROPAGATE_FAULT;
4123 ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, data,
4126 r = X86EMUL_IO_NEEDED;
4138 /* used for instruction fetching */
4139 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4140 gva_t addr, void *val, unsigned int bytes,
4141 struct x86_exception *exception)
4143 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4144 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4148 /* Inline kvm_read_guest_virt_helper for speed. */
4149 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4151 if (unlikely(gpa == UNMAPPED_GVA))
4152 return X86EMUL_PROPAGATE_FAULT;
4154 offset = addr & (PAGE_SIZE-1);
4155 if (WARN_ON(offset + bytes > PAGE_SIZE))
4156 bytes = (unsigned)PAGE_SIZE - offset;
4157 ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, val,
4159 if (unlikely(ret < 0))
4160 return X86EMUL_IO_NEEDED;
4162 return X86EMUL_CONTINUE;
4165 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4166 gva_t addr, void *val, unsigned int bytes,
4167 struct x86_exception *exception)
4169 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4170 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4172 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4175 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4177 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4178 gva_t addr, void *val, unsigned int bytes,
4179 struct x86_exception *exception)
4181 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4182 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4185 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4186 gva_t addr, void *val,
4188 struct x86_exception *exception)
4190 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4192 int r = X86EMUL_CONTINUE;
4195 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4198 unsigned offset = addr & (PAGE_SIZE-1);
4199 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4202 if (gpa == UNMAPPED_GVA)
4203 return X86EMUL_PROPAGATE_FAULT;
4204 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4206 r = X86EMUL_IO_NEEDED;
4217 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4219 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4220 gpa_t *gpa, struct x86_exception *exception,
4223 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4224 | (write ? PFERR_WRITE_MASK : 0);
4226 if (vcpu_match_mmio_gva(vcpu, gva)
4227 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4228 vcpu->arch.access, access)) {
4229 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4230 (gva & (PAGE_SIZE - 1));
4231 trace_vcpu_match_mmio(gva, *gpa, write, false);
4235 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4237 if (*gpa == UNMAPPED_GVA)
4240 /* For APIC access vmexit */
4241 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4244 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4245 trace_vcpu_match_mmio(gva, *gpa, write, true);
4252 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4253 const void *val, int bytes)
4257 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4260 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4264 struct read_write_emulator_ops {
4265 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4267 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4268 void *val, int bytes);
4269 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4270 int bytes, void *val);
4271 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4272 void *val, int bytes);
4276 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4278 if (vcpu->mmio_read_completed) {
4279 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4280 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4281 vcpu->mmio_read_completed = 0;
4288 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4289 void *val, int bytes)
4291 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4294 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4295 void *val, int bytes)
4297 return emulator_write_phys(vcpu, gpa, val, bytes);
4300 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4302 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4303 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4306 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4307 void *val, int bytes)
4309 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4310 return X86EMUL_IO_NEEDED;
4313 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4314 void *val, int bytes)
4316 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4318 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4319 return X86EMUL_CONTINUE;
4322 static const struct read_write_emulator_ops read_emultor = {
4323 .read_write_prepare = read_prepare,
4324 .read_write_emulate = read_emulate,
4325 .read_write_mmio = vcpu_mmio_read,
4326 .read_write_exit_mmio = read_exit_mmio,
4329 static const struct read_write_emulator_ops write_emultor = {
4330 .read_write_emulate = write_emulate,
4331 .read_write_mmio = write_mmio,
4332 .read_write_exit_mmio = write_exit_mmio,
4336 static int emulator_read_write_onepage(unsigned long addr, void *val,
4338 struct x86_exception *exception,
4339 struct kvm_vcpu *vcpu,
4340 const struct read_write_emulator_ops *ops)
4344 bool write = ops->write;
4345 struct kvm_mmio_fragment *frag;
4347 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4350 return X86EMUL_PROPAGATE_FAULT;
4352 /* For APIC access vmexit */
4356 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4357 return X86EMUL_CONTINUE;
4361 * Is this MMIO handled locally?
4363 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4364 if (handled == bytes)
4365 return X86EMUL_CONTINUE;
4371 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4372 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4376 return X86EMUL_CONTINUE;
4379 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4380 void *val, unsigned int bytes,
4381 struct x86_exception *exception,
4382 const struct read_write_emulator_ops *ops)
4384 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4388 if (ops->read_write_prepare &&
4389 ops->read_write_prepare(vcpu, val, bytes))
4390 return X86EMUL_CONTINUE;
4392 vcpu->mmio_nr_fragments = 0;
4394 /* Crossing a page boundary? */
4395 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4398 now = -addr & ~PAGE_MASK;
4399 rc = emulator_read_write_onepage(addr, val, now, exception,
4402 if (rc != X86EMUL_CONTINUE)
4409 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4411 if (rc != X86EMUL_CONTINUE)
4414 if (!vcpu->mmio_nr_fragments)
4417 gpa = vcpu->mmio_fragments[0].gpa;
4419 vcpu->mmio_needed = 1;
4420 vcpu->mmio_cur_fragment = 0;
4422 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4423 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4424 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4425 vcpu->run->mmio.phys_addr = gpa;
4427 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4430 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4434 struct x86_exception *exception)
4436 return emulator_read_write(ctxt, addr, val, bytes,
4437 exception, &read_emultor);
4440 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4444 struct x86_exception *exception)
4446 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4447 exception, &write_emultor);
4450 #define CMPXCHG_TYPE(t, ptr, old, new) \
4451 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4453 #ifdef CONFIG_X86_64
4454 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4456 # define CMPXCHG64(ptr, old, new) \
4457 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4460 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4465 struct x86_exception *exception)
4467 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4473 /* guests cmpxchg8b have to be emulated atomically */
4474 if (bytes > 8 || (bytes & (bytes - 1)))
4477 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4479 if (gpa == UNMAPPED_GVA ||
4480 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4483 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4486 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4487 if (is_error_page(page))
4490 kaddr = kmap_atomic(page);
4491 kaddr += offset_in_page(gpa);
4494 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4497 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4500 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4503 exchanged = CMPXCHG64(kaddr, old, new);
4508 kunmap_atomic(kaddr);
4509 kvm_release_page_dirty(page);
4512 return X86EMUL_CMPXCHG_FAILED;
4514 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
4515 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4517 return X86EMUL_CONTINUE;
4520 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4522 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4525 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4527 /* TODO: String I/O for in kernel device */
4530 if (vcpu->arch.pio.in)
4531 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4532 vcpu->arch.pio.size, pd);
4534 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4535 vcpu->arch.pio.port, vcpu->arch.pio.size,
4540 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4541 unsigned short port, void *val,
4542 unsigned int count, bool in)
4544 vcpu->arch.pio.port = port;
4545 vcpu->arch.pio.in = in;
4546 vcpu->arch.pio.count = count;
4547 vcpu->arch.pio.size = size;
4549 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4550 vcpu->arch.pio.count = 0;
4554 vcpu->run->exit_reason = KVM_EXIT_IO;
4555 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4556 vcpu->run->io.size = size;
4557 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4558 vcpu->run->io.count = count;
4559 vcpu->run->io.port = port;
4564 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4565 int size, unsigned short port, void *val,
4568 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4571 if (vcpu->arch.pio.count)
4574 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4577 memcpy(val, vcpu->arch.pio_data, size * count);
4578 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4579 vcpu->arch.pio.count = 0;
4586 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4587 int size, unsigned short port,
4588 const void *val, unsigned int count)
4590 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4592 memcpy(vcpu->arch.pio_data, val, size * count);
4593 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4594 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4597 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4599 return kvm_x86_ops->get_segment_base(vcpu, seg);
4602 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4604 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4607 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4609 if (!need_emulate_wbinvd(vcpu))
4610 return X86EMUL_CONTINUE;
4612 if (kvm_x86_ops->has_wbinvd_exit()) {
4613 int cpu = get_cpu();
4615 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4616 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4617 wbinvd_ipi, NULL, 1);
4619 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4622 return X86EMUL_CONTINUE;
4624 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4626 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4628 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4631 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4633 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4636 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4639 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4642 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4644 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4647 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4649 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4650 unsigned long value;
4654 value = kvm_read_cr0(vcpu);
4657 value = vcpu->arch.cr2;
4660 value = kvm_read_cr3(vcpu);
4663 value = kvm_read_cr4(vcpu);
4666 value = kvm_get_cr8(vcpu);
4669 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4676 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4678 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4683 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4686 vcpu->arch.cr2 = val;
4689 res = kvm_set_cr3(vcpu, val);
4692 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4695 res = kvm_set_cr8(vcpu, val);
4698 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4705 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4707 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4710 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4712 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4715 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4717 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4720 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4722 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4725 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4727 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4730 static unsigned long emulator_get_cached_segment_base(
4731 struct x86_emulate_ctxt *ctxt, int seg)
4733 return get_segment_base(emul_to_vcpu(ctxt), seg);
4736 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4737 struct desc_struct *desc, u32 *base3,
4740 struct kvm_segment var;
4742 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4743 *selector = var.selector;
4746 memset(desc, 0, sizeof(*desc));
4752 set_desc_limit(desc, var.limit);
4753 set_desc_base(desc, (unsigned long)var.base);
4754 #ifdef CONFIG_X86_64
4756 *base3 = var.base >> 32;
4758 desc->type = var.type;
4760 desc->dpl = var.dpl;
4761 desc->p = var.present;
4762 desc->avl = var.avl;
4770 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4771 struct desc_struct *desc, u32 base3,
4774 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4775 struct kvm_segment var;
4777 var.selector = selector;
4778 var.base = get_desc_base(desc);
4779 #ifdef CONFIG_X86_64
4780 var.base |= ((u64)base3) << 32;
4782 var.limit = get_desc_limit(desc);
4784 var.limit = (var.limit << 12) | 0xfff;
4785 var.type = desc->type;
4786 var.dpl = desc->dpl;
4791 var.avl = desc->avl;
4792 var.present = desc->p;
4793 var.unusable = !var.present;
4796 kvm_set_segment(vcpu, &var, seg);
4800 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4801 u32 msr_index, u64 *pdata)
4803 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4806 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4807 u32 msr_index, u64 data)
4809 struct msr_data msr;
4812 msr.index = msr_index;
4813 msr.host_initiated = false;
4814 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4817 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4820 return kvm_pmu_check_pmc(emul_to_vcpu(ctxt), pmc);
4823 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4824 u32 pmc, u64 *pdata)
4826 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4829 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4831 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4834 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4837 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4839 * CR0.TS may reference the host fpu state, not the guest fpu state,
4840 * so it may be clear at this point.
4845 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4850 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4851 struct x86_instruction_info *info,
4852 enum x86_intercept_stage stage)
4854 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4857 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4858 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4860 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4863 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4865 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4868 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4870 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4873 static const struct x86_emulate_ops emulate_ops = {
4874 .read_gpr = emulator_read_gpr,
4875 .write_gpr = emulator_write_gpr,
4876 .read_std = kvm_read_guest_virt_system,
4877 .write_std = kvm_write_guest_virt_system,
4878 .fetch = kvm_fetch_guest_virt,
4879 .read_emulated = emulator_read_emulated,
4880 .write_emulated = emulator_write_emulated,
4881 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4882 .invlpg = emulator_invlpg,
4883 .pio_in_emulated = emulator_pio_in_emulated,
4884 .pio_out_emulated = emulator_pio_out_emulated,
4885 .get_segment = emulator_get_segment,
4886 .set_segment = emulator_set_segment,
4887 .get_cached_segment_base = emulator_get_cached_segment_base,
4888 .get_gdt = emulator_get_gdt,
4889 .get_idt = emulator_get_idt,
4890 .set_gdt = emulator_set_gdt,
4891 .set_idt = emulator_set_idt,
4892 .get_cr = emulator_get_cr,
4893 .set_cr = emulator_set_cr,
4894 .cpl = emulator_get_cpl,
4895 .get_dr = emulator_get_dr,
4896 .set_dr = emulator_set_dr,
4897 .set_msr = emulator_set_msr,
4898 .get_msr = emulator_get_msr,
4899 .check_pmc = emulator_check_pmc,
4900 .read_pmc = emulator_read_pmc,
4901 .halt = emulator_halt,
4902 .wbinvd = emulator_wbinvd,
4903 .fix_hypercall = emulator_fix_hypercall,
4904 .get_fpu = emulator_get_fpu,
4905 .put_fpu = emulator_put_fpu,
4906 .intercept = emulator_intercept,
4907 .get_cpuid = emulator_get_cpuid,
4910 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4912 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
4914 * an sti; sti; sequence only disable interrupts for the first
4915 * instruction. So, if the last instruction, be it emulated or
4916 * not, left the system with the INT_STI flag enabled, it
4917 * means that the last instruction is an sti. We should not
4918 * leave the flag on in this case. The same goes for mov ss
4920 if (int_shadow & mask)
4922 if (unlikely(int_shadow || mask)) {
4923 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4925 kvm_make_request(KVM_REQ_EVENT, vcpu);
4929 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4931 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4932 if (ctxt->exception.vector == PF_VECTOR)
4933 kvm_propagate_fault(vcpu, &ctxt->exception);
4934 else if (ctxt->exception.error_code_valid)
4935 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4936 ctxt->exception.error_code);
4938 kvm_queue_exception(vcpu, ctxt->exception.vector);
4941 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4943 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4946 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4948 ctxt->eflags = kvm_get_rflags(vcpu);
4949 ctxt->eip = kvm_rip_read(vcpu);
4950 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4951 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4952 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
4953 cs_db ? X86EMUL_MODE_PROT32 :
4954 X86EMUL_MODE_PROT16;
4955 ctxt->guest_mode = is_guest_mode(vcpu);
4957 init_decode_cache(ctxt);
4958 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4961 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4963 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4966 init_emulate_ctxt(vcpu);
4970 ctxt->_eip = ctxt->eip + inc_eip;
4971 ret = emulate_int_real(ctxt, irq);
4973 if (ret != X86EMUL_CONTINUE)
4974 return EMULATE_FAIL;
4976 ctxt->eip = ctxt->_eip;
4977 kvm_rip_write(vcpu, ctxt->eip);
4978 kvm_set_rflags(vcpu, ctxt->eflags);
4980 if (irq == NMI_VECTOR)
4981 vcpu->arch.nmi_pending = 0;
4983 vcpu->arch.interrupt.pending = false;
4985 return EMULATE_DONE;
4987 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4989 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4991 int r = EMULATE_DONE;
4993 ++vcpu->stat.insn_emulation_fail;
4994 trace_kvm_emulate_insn_failed(vcpu);
4995 if (!is_guest_mode(vcpu)) {
4996 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4997 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4998 vcpu->run->internal.ndata = 0;
5001 kvm_queue_exception(vcpu, UD_VECTOR);
5006 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5007 bool write_fault_to_shadow_pgtable,
5013 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5016 if (!vcpu->arch.mmu.direct_map) {
5018 * Write permission should be allowed since only
5019 * write access need to be emulated.
5021 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5024 * If the mapping is invalid in guest, let cpu retry
5025 * it to generate fault.
5027 if (gpa == UNMAPPED_GVA)
5032 * Do not retry the unhandleable instruction if it faults on the
5033 * readonly host memory, otherwise it will goto a infinite loop:
5034 * retry instruction -> write #PF -> emulation fail -> retry
5035 * instruction -> ...
5037 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5040 * If the instruction failed on the error pfn, it can not be fixed,
5041 * report the error to userspace.
5043 if (is_error_noslot_pfn(pfn))
5046 kvm_release_pfn_clean(pfn);
5048 /* The instructions are well-emulated on direct mmu. */
5049 if (vcpu->arch.mmu.direct_map) {
5050 unsigned int indirect_shadow_pages;
5052 spin_lock(&vcpu->kvm->mmu_lock);
5053 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5054 spin_unlock(&vcpu->kvm->mmu_lock);
5056 if (indirect_shadow_pages)
5057 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5063 * if emulation was due to access to shadowed page table
5064 * and it failed try to unshadow page and re-enter the
5065 * guest to let CPU execute the instruction.
5067 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5070 * If the access faults on its page table, it can not
5071 * be fixed by unprotecting shadow page and it should
5072 * be reported to userspace.
5074 return !write_fault_to_shadow_pgtable;
5077 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5078 unsigned long cr2, int emulation_type)
5080 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5081 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5083 last_retry_eip = vcpu->arch.last_retry_eip;
5084 last_retry_addr = vcpu->arch.last_retry_addr;
5087 * If the emulation is caused by #PF and it is non-page_table
5088 * writing instruction, it means the VM-EXIT is caused by shadow
5089 * page protected, we can zap the shadow page and retry this
5090 * instruction directly.
5092 * Note: if the guest uses a non-page-table modifying instruction
5093 * on the PDE that points to the instruction, then we will unmap
5094 * the instruction and go to an infinite loop. So, we cache the
5095 * last retried eip and the last fault address, if we meet the eip
5096 * and the address again, we can break out of the potential infinite
5099 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5101 if (!(emulation_type & EMULTYPE_RETRY))
5104 if (x86_page_table_writing_insn(ctxt))
5107 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5110 vcpu->arch.last_retry_eip = ctxt->eip;
5111 vcpu->arch.last_retry_addr = cr2;
5113 if (!vcpu->arch.mmu.direct_map)
5114 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5116 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5121 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5122 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5124 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5133 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5134 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5139 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5141 struct kvm_run *kvm_run = vcpu->run;
5144 * rflags is the old, "raw" value of the flags. The new value has
5145 * not been saved yet.
5147 * This is correct even for TF set by the guest, because "the
5148 * processor will not generate this exception after the instruction
5149 * that sets the TF flag".
5151 if (unlikely(rflags & X86_EFLAGS_TF)) {
5152 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5153 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5155 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5156 kvm_run->debug.arch.exception = DB_VECTOR;
5157 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5158 *r = EMULATE_USER_EXIT;
5160 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5162 * "Certain debug exceptions may clear bit 0-3. The
5163 * remaining contents of the DR6 register are never
5164 * cleared by the processor".
5166 vcpu->arch.dr6 &= ~15;
5167 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5168 kvm_queue_exception(vcpu, DB_VECTOR);
5173 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5175 struct kvm_run *kvm_run = vcpu->run;
5176 unsigned long eip = vcpu->arch.emulate_ctxt.eip;
5179 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5180 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5181 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5182 vcpu->arch.guest_debug_dr7,
5186 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5187 kvm_run->debug.arch.pc = kvm_rip_read(vcpu) +
5188 get_segment_base(vcpu, VCPU_SREG_CS);
5190 kvm_run->debug.arch.exception = DB_VECTOR;
5191 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5192 *r = EMULATE_USER_EXIT;
5197 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5198 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5199 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5204 vcpu->arch.dr6 &= ~15;
5205 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5206 kvm_queue_exception(vcpu, DB_VECTOR);
5215 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5222 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5223 bool writeback = true;
5224 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5227 * Clear write_fault_to_shadow_pgtable here to ensure it is
5230 vcpu->arch.write_fault_to_shadow_pgtable = false;
5231 kvm_clear_exception_queue(vcpu);
5233 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5234 init_emulate_ctxt(vcpu);
5237 * We will reenter on the same instruction since
5238 * we do not set complete_userspace_io. This does not
5239 * handle watchpoints yet, those would be handled in
5242 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5245 ctxt->interruptibility = 0;
5246 ctxt->have_exception = false;
5247 ctxt->perm_ok = false;
5249 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5251 r = x86_decode_insn(ctxt, insn, insn_len);
5253 trace_kvm_emulate_insn_start(vcpu);
5254 ++vcpu->stat.insn_emulation;
5255 if (r != EMULATION_OK) {
5256 if (emulation_type & EMULTYPE_TRAP_UD)
5257 return EMULATE_FAIL;
5258 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5260 return EMULATE_DONE;
5261 if (emulation_type & EMULTYPE_SKIP)
5262 return EMULATE_FAIL;
5263 return handle_emulation_failure(vcpu);
5267 if (emulation_type & EMULTYPE_SKIP) {
5268 kvm_rip_write(vcpu, ctxt->_eip);
5269 if (ctxt->eflags & X86_EFLAGS_RF)
5270 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5271 return EMULATE_DONE;
5274 if (retry_instruction(ctxt, cr2, emulation_type))
5275 return EMULATE_DONE;
5277 /* this is needed for vmware backdoor interface to work since it
5278 changes registers values during IO operation */
5279 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5280 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5281 emulator_invalidate_register_cache(ctxt);
5285 r = x86_emulate_insn(ctxt);
5287 if (r == EMULATION_INTERCEPTED)
5288 return EMULATE_DONE;
5290 if (r == EMULATION_FAILED) {
5291 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5293 return EMULATE_DONE;
5295 return handle_emulation_failure(vcpu);
5298 if (ctxt->have_exception) {
5299 inject_emulated_exception(vcpu);
5301 } else if (vcpu->arch.pio.count) {
5302 if (!vcpu->arch.pio.in) {
5303 /* FIXME: return into emulator if single-stepping. */
5304 vcpu->arch.pio.count = 0;
5307 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5309 r = EMULATE_USER_EXIT;
5310 } else if (vcpu->mmio_needed) {
5311 if (!vcpu->mmio_is_write)
5313 r = EMULATE_USER_EXIT;
5314 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5315 } else if (r == EMULATION_RESTART)
5321 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5322 toggle_interruptibility(vcpu, ctxt->interruptibility);
5323 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5324 kvm_rip_write(vcpu, ctxt->eip);
5325 if (r == EMULATE_DONE)
5326 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5327 __kvm_set_rflags(vcpu, ctxt->eflags);
5330 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5331 * do nothing, and it will be requested again as soon as
5332 * the shadow expires. But we still need to check here,
5333 * because POPF has no interrupt shadow.
5335 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5336 kvm_make_request(KVM_REQ_EVENT, vcpu);
5338 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5342 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5344 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5346 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5347 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5348 size, port, &val, 1);
5349 /* do not return to emulator after return from userspace */
5350 vcpu->arch.pio.count = 0;
5353 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5355 static void tsc_bad(void *info)
5357 __this_cpu_write(cpu_tsc_khz, 0);
5360 static void tsc_khz_changed(void *data)
5362 struct cpufreq_freqs *freq = data;
5363 unsigned long khz = 0;
5367 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5368 khz = cpufreq_quick_get(raw_smp_processor_id());
5371 __this_cpu_write(cpu_tsc_khz, khz);
5374 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5377 struct cpufreq_freqs *freq = data;
5379 struct kvm_vcpu *vcpu;
5380 int i, send_ipi = 0;
5383 * We allow guests to temporarily run on slowing clocks,
5384 * provided we notify them after, or to run on accelerating
5385 * clocks, provided we notify them before. Thus time never
5388 * However, we have a problem. We can't atomically update
5389 * the frequency of a given CPU from this function; it is
5390 * merely a notifier, which can be called from any CPU.
5391 * Changing the TSC frequency at arbitrary points in time
5392 * requires a recomputation of local variables related to
5393 * the TSC for each VCPU. We must flag these local variables
5394 * to be updated and be sure the update takes place with the
5395 * new frequency before any guests proceed.
5397 * Unfortunately, the combination of hotplug CPU and frequency
5398 * change creates an intractable locking scenario; the order
5399 * of when these callouts happen is undefined with respect to
5400 * CPU hotplug, and they can race with each other. As such,
5401 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5402 * undefined; you can actually have a CPU frequency change take
5403 * place in between the computation of X and the setting of the
5404 * variable. To protect against this problem, all updates of
5405 * the per_cpu tsc_khz variable are done in an interrupt
5406 * protected IPI, and all callers wishing to update the value
5407 * must wait for a synchronous IPI to complete (which is trivial
5408 * if the caller is on the CPU already). This establishes the
5409 * necessary total order on variable updates.
5411 * Note that because a guest time update may take place
5412 * anytime after the setting of the VCPU's request bit, the
5413 * correct TSC value must be set before the request. However,
5414 * to ensure the update actually makes it to any guest which
5415 * starts running in hardware virtualization between the set
5416 * and the acquisition of the spinlock, we must also ping the
5417 * CPU after setting the request bit.
5421 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5423 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5426 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5428 spin_lock(&kvm_lock);
5429 list_for_each_entry(kvm, &vm_list, vm_list) {
5430 kvm_for_each_vcpu(i, vcpu, kvm) {
5431 if (vcpu->cpu != freq->cpu)
5433 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5434 if (vcpu->cpu != smp_processor_id())
5438 spin_unlock(&kvm_lock);
5440 if (freq->old < freq->new && send_ipi) {
5442 * We upscale the frequency. Must make the guest
5443 * doesn't see old kvmclock values while running with
5444 * the new frequency, otherwise we risk the guest sees
5445 * time go backwards.
5447 * In case we update the frequency for another cpu
5448 * (which might be in guest context) send an interrupt
5449 * to kick the cpu out of guest context. Next time
5450 * guest context is entered kvmclock will be updated,
5451 * so the guest will not see stale values.
5453 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5458 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5459 .notifier_call = kvmclock_cpufreq_notifier
5462 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5463 unsigned long action, void *hcpu)
5465 unsigned int cpu = (unsigned long)hcpu;
5469 case CPU_DOWN_FAILED:
5470 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5472 case CPU_DOWN_PREPARE:
5473 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5479 static struct notifier_block kvmclock_cpu_notifier_block = {
5480 .notifier_call = kvmclock_cpu_notifier,
5481 .priority = -INT_MAX
5484 static void kvm_timer_init(void)
5488 max_tsc_khz = tsc_khz;
5490 cpu_notifier_register_begin();
5491 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5492 #ifdef CONFIG_CPU_FREQ
5493 struct cpufreq_policy policy;
5494 memset(&policy, 0, sizeof(policy));
5496 cpufreq_get_policy(&policy, cpu);
5497 if (policy.cpuinfo.max_freq)
5498 max_tsc_khz = policy.cpuinfo.max_freq;
5501 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5502 CPUFREQ_TRANSITION_NOTIFIER);
5504 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5505 for_each_online_cpu(cpu)
5506 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5508 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5509 cpu_notifier_register_done();
5513 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5515 int kvm_is_in_guest(void)
5517 return __this_cpu_read(current_vcpu) != NULL;
5520 static int kvm_is_user_mode(void)
5524 if (__this_cpu_read(current_vcpu))
5525 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5527 return user_mode != 0;
5530 static unsigned long kvm_get_guest_ip(void)
5532 unsigned long ip = 0;
5534 if (__this_cpu_read(current_vcpu))
5535 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5540 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5541 .is_in_guest = kvm_is_in_guest,
5542 .is_user_mode = kvm_is_user_mode,
5543 .get_guest_ip = kvm_get_guest_ip,
5546 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5548 __this_cpu_write(current_vcpu, vcpu);
5550 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5552 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5554 __this_cpu_write(current_vcpu, NULL);
5556 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5558 static void kvm_set_mmio_spte_mask(void)
5561 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5564 * Set the reserved bits and the present bit of an paging-structure
5565 * entry to generate page fault with PFER.RSV = 1.
5567 /* Mask the reserved physical address bits. */
5568 mask = ((1ull << (51 - maxphyaddr + 1)) - 1) << maxphyaddr;
5570 /* Bit 62 is always reserved for 32bit host. */
5571 mask |= 0x3ull << 62;
5573 /* Set the present bit. */
5576 #ifdef CONFIG_X86_64
5578 * If reserved bit is not supported, clear the present bit to disable
5581 if (maxphyaddr == 52)
5585 kvm_mmu_set_mmio_spte_mask(mask);
5588 #ifdef CONFIG_X86_64
5589 static void pvclock_gtod_update_fn(struct work_struct *work)
5593 struct kvm_vcpu *vcpu;
5596 spin_lock(&kvm_lock);
5597 list_for_each_entry(kvm, &vm_list, vm_list)
5598 kvm_for_each_vcpu(i, vcpu, kvm)
5599 set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5600 atomic_set(&kvm_guest_has_master_clock, 0);
5601 spin_unlock(&kvm_lock);
5604 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5607 * Notification about pvclock gtod data update.
5609 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5612 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5613 struct timekeeper *tk = priv;
5615 update_pvclock_gtod(tk);
5617 /* disable master clock if host does not trust, or does not
5618 * use, TSC clocksource
5620 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5621 atomic_read(&kvm_guest_has_master_clock) != 0)
5622 queue_work(system_long_wq, &pvclock_gtod_work);
5627 static struct notifier_block pvclock_gtod_notifier = {
5628 .notifier_call = pvclock_gtod_notify,
5632 int kvm_arch_init(void *opaque)
5635 struct kvm_x86_ops *ops = opaque;
5638 printk(KERN_ERR "kvm: already loaded the other module\n");
5643 if (!ops->cpu_has_kvm_support()) {
5644 printk(KERN_ERR "kvm: no hardware support\n");
5648 if (ops->disabled_by_bios()) {
5649 printk(KERN_ERR "kvm: disabled by bios\n");
5655 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5657 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5661 r = kvm_mmu_module_init();
5663 goto out_free_percpu;
5665 kvm_set_mmio_spte_mask();
5668 kvm_init_msr_list();
5670 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5671 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5675 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5678 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5681 #ifdef CONFIG_X86_64
5682 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5688 free_percpu(shared_msrs);
5693 void kvm_arch_exit(void)
5695 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5697 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5698 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5699 CPUFREQ_TRANSITION_NOTIFIER);
5700 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5701 #ifdef CONFIG_X86_64
5702 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5705 kvm_mmu_module_exit();
5706 free_percpu(shared_msrs);
5709 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5711 ++vcpu->stat.halt_exits;
5712 if (irqchip_in_kernel(vcpu->kvm)) {
5713 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5716 vcpu->run->exit_reason = KVM_EXIT_HLT;
5720 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5722 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5724 u64 param, ingpa, outgpa, ret;
5725 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5726 bool fast, longmode;
5729 * hypercall generates UD from non zero cpl and real mode
5732 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5733 kvm_queue_exception(vcpu, UD_VECTOR);
5737 longmode = is_64_bit_mode(vcpu);
5740 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5741 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5742 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5743 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5744 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5745 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5747 #ifdef CONFIG_X86_64
5749 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5750 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5751 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5755 code = param & 0xffff;
5756 fast = (param >> 16) & 0x1;
5757 rep_cnt = (param >> 32) & 0xfff;
5758 rep_idx = (param >> 48) & 0xfff;
5760 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5763 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5764 kvm_vcpu_on_spin(vcpu);
5767 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5771 ret = res | (((u64)rep_done & 0xfff) << 32);
5773 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5775 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5776 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5783 * kvm_pv_kick_cpu_op: Kick a vcpu.
5785 * @apicid - apicid of vcpu to be kicked.
5787 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5789 struct kvm_lapic_irq lapic_irq;
5791 lapic_irq.shorthand = 0;
5792 lapic_irq.dest_mode = 0;
5793 lapic_irq.dest_id = apicid;
5795 lapic_irq.delivery_mode = APIC_DM_REMRD;
5796 kvm_irq_delivery_to_apic(kvm, 0, &lapic_irq, NULL);
5799 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5801 unsigned long nr, a0, a1, a2, a3, ret;
5802 int op_64_bit, r = 1;
5804 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5805 return kvm_hv_hypercall(vcpu);
5807 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5808 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5809 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5810 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5811 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5813 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5815 op_64_bit = is_64_bit_mode(vcpu);
5824 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5830 case KVM_HC_VAPIC_POLL_IRQ:
5833 case KVM_HC_KICK_CPU:
5834 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5844 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5845 ++vcpu->stat.hypercalls;
5848 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5850 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5852 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5853 char instruction[3];
5854 unsigned long rip = kvm_rip_read(vcpu);
5856 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5858 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5862 * Check if userspace requested an interrupt window, and that the
5863 * interrupt window is open.
5865 * No need to exit to userspace if we already have an interrupt queued.
5867 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5869 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5870 vcpu->run->request_interrupt_window &&
5871 kvm_arch_interrupt_allowed(vcpu));
5874 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5876 struct kvm_run *kvm_run = vcpu->run;
5878 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5879 kvm_run->cr8 = kvm_get_cr8(vcpu);
5880 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5881 if (irqchip_in_kernel(vcpu->kvm))
5882 kvm_run->ready_for_interrupt_injection = 1;
5884 kvm_run->ready_for_interrupt_injection =
5885 kvm_arch_interrupt_allowed(vcpu) &&
5886 !kvm_cpu_has_interrupt(vcpu) &&
5887 !kvm_event_needs_reinjection(vcpu);
5890 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5894 if (!kvm_x86_ops->update_cr8_intercept)
5897 if (!vcpu->arch.apic)
5900 if (!vcpu->arch.apic->vapic_addr)
5901 max_irr = kvm_lapic_find_highest_irr(vcpu);
5908 tpr = kvm_lapic_get_cr8(vcpu);
5910 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5913 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
5917 /* try to reinject previous events if any */
5918 if (vcpu->arch.exception.pending) {
5919 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5920 vcpu->arch.exception.has_error_code,
5921 vcpu->arch.exception.error_code);
5923 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
5924 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
5927 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5928 vcpu->arch.exception.has_error_code,
5929 vcpu->arch.exception.error_code,
5930 vcpu->arch.exception.reinject);
5934 if (vcpu->arch.nmi_injected) {
5935 kvm_x86_ops->set_nmi(vcpu);
5939 if (vcpu->arch.interrupt.pending) {
5940 kvm_x86_ops->set_irq(vcpu);
5944 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5945 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5950 /* try to inject new event if pending */
5951 if (vcpu->arch.nmi_pending) {
5952 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5953 --vcpu->arch.nmi_pending;
5954 vcpu->arch.nmi_injected = true;
5955 kvm_x86_ops->set_nmi(vcpu);
5957 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
5959 * Because interrupts can be injected asynchronously, we are
5960 * calling check_nested_events again here to avoid a race condition.
5961 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
5962 * proposal and current concerns. Perhaps we should be setting
5963 * KVM_REQ_EVENT only on certain events and not unconditionally?
5965 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5966 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5970 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5971 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5973 kvm_x86_ops->set_irq(vcpu);
5979 static void process_nmi(struct kvm_vcpu *vcpu)
5984 * x86 is limited to one NMI running, and one NMI pending after it.
5985 * If an NMI is already in progress, limit further NMIs to just one.
5986 * Otherwise, allow two (and we'll inject the first one immediately).
5988 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5991 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5992 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5993 kvm_make_request(KVM_REQ_EVENT, vcpu);
5996 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
5998 u64 eoi_exit_bitmap[4];
6001 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6004 memset(eoi_exit_bitmap, 0, 32);
6007 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6008 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6009 kvm_apic_update_tmr(vcpu, tmr);
6013 * Returns 1 to let __vcpu_run() continue the guest execution loop without
6014 * exiting to the userspace. Otherwise, the value will be returned to the
6017 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6020 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6021 vcpu->run->request_interrupt_window;
6022 bool req_immediate_exit = false;
6024 if (vcpu->requests) {
6025 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6026 kvm_mmu_unload(vcpu);
6027 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6028 __kvm_migrate_timers(vcpu);
6029 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6030 kvm_gen_update_masterclock(vcpu->kvm);
6031 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6032 kvm_gen_kvmclock_update(vcpu);
6033 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6034 r = kvm_guest_time_update(vcpu);
6038 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6039 kvm_mmu_sync_roots(vcpu);
6040 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6041 kvm_x86_ops->tlb_flush(vcpu);
6042 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6043 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6047 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6048 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6052 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6053 vcpu->fpu_active = 0;
6054 kvm_x86_ops->fpu_deactivate(vcpu);
6056 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6057 /* Page is swapped out. Do synthetic halt */
6058 vcpu->arch.apf.halted = true;
6062 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6063 record_steal_time(vcpu);
6064 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6066 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6067 kvm_handle_pmu_event(vcpu);
6068 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6069 kvm_deliver_pmi(vcpu);
6070 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6071 vcpu_scan_ioapic(vcpu);
6074 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6075 kvm_apic_accept_events(vcpu);
6076 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6081 if (inject_pending_event(vcpu, req_int_win) != 0)
6082 req_immediate_exit = true;
6083 /* enable NMI/IRQ window open exits if needed */
6084 else if (vcpu->arch.nmi_pending)
6085 kvm_x86_ops->enable_nmi_window(vcpu);
6086 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6087 kvm_x86_ops->enable_irq_window(vcpu);
6089 if (kvm_lapic_enabled(vcpu)) {
6091 * Update architecture specific hints for APIC
6092 * virtual interrupt delivery.
6094 if (kvm_x86_ops->hwapic_irr_update)
6095 kvm_x86_ops->hwapic_irr_update(vcpu,
6096 kvm_lapic_find_highest_irr(vcpu));
6097 update_cr8_intercept(vcpu);
6098 kvm_lapic_sync_to_vapic(vcpu);
6102 r = kvm_mmu_reload(vcpu);
6104 goto cancel_injection;
6109 kvm_x86_ops->prepare_guest_switch(vcpu);
6110 if (vcpu->fpu_active)
6111 kvm_load_guest_fpu(vcpu);
6112 kvm_load_guest_xcr0(vcpu);
6114 vcpu->mode = IN_GUEST_MODE;
6116 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6118 /* We should set ->mode before check ->requests,
6119 * see the comment in make_all_cpus_request.
6121 smp_mb__after_srcu_read_unlock();
6123 local_irq_disable();
6125 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6126 || need_resched() || signal_pending(current)) {
6127 vcpu->mode = OUTSIDE_GUEST_MODE;
6131 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6133 goto cancel_injection;
6136 if (req_immediate_exit)
6137 smp_send_reschedule(vcpu->cpu);
6141 if (unlikely(vcpu->arch.switch_db_regs)) {
6143 set_debugreg(vcpu->arch.eff_db[0], 0);
6144 set_debugreg(vcpu->arch.eff_db[1], 1);
6145 set_debugreg(vcpu->arch.eff_db[2], 2);
6146 set_debugreg(vcpu->arch.eff_db[3], 3);
6147 set_debugreg(vcpu->arch.dr6, 6);
6150 trace_kvm_entry(vcpu->vcpu_id);
6151 kvm_x86_ops->run(vcpu);
6154 * Do this here before restoring debug registers on the host. And
6155 * since we do this before handling the vmexit, a DR access vmexit
6156 * can (a) read the correct value of the debug registers, (b) set
6157 * KVM_DEBUGREG_WONT_EXIT again.
6159 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6162 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6163 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6164 for (i = 0; i < KVM_NR_DB_REGS; i++)
6165 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6169 * If the guest has used debug registers, at least dr7
6170 * will be disabled while returning to the host.
6171 * If we don't have active breakpoints in the host, we don't
6172 * care about the messed up debug address registers. But if
6173 * we have some of them active, restore the old state.
6175 if (hw_breakpoint_active())
6176 hw_breakpoint_restore();
6178 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6181 vcpu->mode = OUTSIDE_GUEST_MODE;
6184 /* Interrupt is enabled by handle_external_intr() */
6185 kvm_x86_ops->handle_external_intr(vcpu);
6190 * We must have an instruction between local_irq_enable() and
6191 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6192 * the interrupt shadow. The stat.exits increment will do nicely.
6193 * But we need to prevent reordering, hence this barrier():
6201 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6204 * Profile KVM exit RIPs:
6206 if (unlikely(prof_on == KVM_PROFILING)) {
6207 unsigned long rip = kvm_rip_read(vcpu);
6208 profile_hit(KVM_PROFILING, (void *)rip);
6211 if (unlikely(vcpu->arch.tsc_always_catchup))
6212 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6214 if (vcpu->arch.apic_attention)
6215 kvm_lapic_sync_from_vapic(vcpu);
6217 r = kvm_x86_ops->handle_exit(vcpu);
6221 kvm_x86_ops->cancel_injection(vcpu);
6222 if (unlikely(vcpu->arch.apic_attention))
6223 kvm_lapic_sync_from_vapic(vcpu);
6229 static int __vcpu_run(struct kvm_vcpu *vcpu)
6232 struct kvm *kvm = vcpu->kvm;
6234 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6238 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6239 !vcpu->arch.apf.halted)
6240 r = vcpu_enter_guest(vcpu);
6242 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6243 kvm_vcpu_block(vcpu);
6244 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6245 if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
6246 kvm_apic_accept_events(vcpu);
6247 switch(vcpu->arch.mp_state) {
6248 case KVM_MP_STATE_HALTED:
6249 vcpu->arch.pv.pv_unhalted = false;
6250 vcpu->arch.mp_state =
6251 KVM_MP_STATE_RUNNABLE;
6252 case KVM_MP_STATE_RUNNABLE:
6253 vcpu->arch.apf.halted = false;
6255 case KVM_MP_STATE_INIT_RECEIVED:
6267 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6268 if (kvm_cpu_has_pending_timer(vcpu))
6269 kvm_inject_pending_timer_irqs(vcpu);
6271 if (dm_request_for_irq_injection(vcpu)) {
6273 vcpu->run->exit_reason = KVM_EXIT_INTR;
6274 ++vcpu->stat.request_irq_exits;
6277 kvm_check_async_pf_completion(vcpu);
6279 if (signal_pending(current)) {
6281 vcpu->run->exit_reason = KVM_EXIT_INTR;
6282 ++vcpu->stat.signal_exits;
6284 if (need_resched()) {
6285 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6287 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6291 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6296 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6299 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6300 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6301 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6302 if (r != EMULATE_DONE)
6307 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6309 BUG_ON(!vcpu->arch.pio.count);
6311 return complete_emulated_io(vcpu);
6315 * Implements the following, as a state machine:
6319 * for each mmio piece in the fragment
6327 * for each mmio piece in the fragment
6332 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6334 struct kvm_run *run = vcpu->run;
6335 struct kvm_mmio_fragment *frag;
6338 BUG_ON(!vcpu->mmio_needed);
6340 /* Complete previous fragment */
6341 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6342 len = min(8u, frag->len);
6343 if (!vcpu->mmio_is_write)
6344 memcpy(frag->data, run->mmio.data, len);
6346 if (frag->len <= 8) {
6347 /* Switch to the next fragment. */
6349 vcpu->mmio_cur_fragment++;
6351 /* Go forward to the next mmio piece. */
6357 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6358 vcpu->mmio_needed = 0;
6360 /* FIXME: return into emulator if single-stepping. */
6361 if (vcpu->mmio_is_write)
6363 vcpu->mmio_read_completed = 1;
6364 return complete_emulated_io(vcpu);
6367 run->exit_reason = KVM_EXIT_MMIO;
6368 run->mmio.phys_addr = frag->gpa;
6369 if (vcpu->mmio_is_write)
6370 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6371 run->mmio.len = min(8u, frag->len);
6372 run->mmio.is_write = vcpu->mmio_is_write;
6373 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6378 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6383 if (!tsk_used_math(current) && init_fpu(current))
6386 if (vcpu->sigset_active)
6387 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6389 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6390 kvm_vcpu_block(vcpu);
6391 kvm_apic_accept_events(vcpu);
6392 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6397 /* re-sync apic's tpr */
6398 if (!irqchip_in_kernel(vcpu->kvm)) {
6399 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6405 if (unlikely(vcpu->arch.complete_userspace_io)) {
6406 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6407 vcpu->arch.complete_userspace_io = NULL;
6412 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6414 r = __vcpu_run(vcpu);
6417 post_kvm_run_save(vcpu);
6418 if (vcpu->sigset_active)
6419 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6424 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6426 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6428 * We are here if userspace calls get_regs() in the middle of
6429 * instruction emulation. Registers state needs to be copied
6430 * back from emulation context to vcpu. Userspace shouldn't do
6431 * that usually, but some bad designed PV devices (vmware
6432 * backdoor interface) need this to work
6434 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6435 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6437 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6438 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6439 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6440 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6441 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6442 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6443 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6444 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6445 #ifdef CONFIG_X86_64
6446 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6447 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6448 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6449 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6450 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6451 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6452 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6453 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6456 regs->rip = kvm_rip_read(vcpu);
6457 regs->rflags = kvm_get_rflags(vcpu);
6462 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6464 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6465 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6467 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6468 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6469 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6470 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6471 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6472 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6473 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6474 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6475 #ifdef CONFIG_X86_64
6476 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6477 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6478 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6479 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6480 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6481 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6482 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6483 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6486 kvm_rip_write(vcpu, regs->rip);
6487 kvm_set_rflags(vcpu, regs->rflags);
6489 vcpu->arch.exception.pending = false;
6491 kvm_make_request(KVM_REQ_EVENT, vcpu);
6496 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6498 struct kvm_segment cs;
6500 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6504 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6506 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6507 struct kvm_sregs *sregs)
6511 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6512 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6513 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6514 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6515 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6516 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6518 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6519 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6521 kvm_x86_ops->get_idt(vcpu, &dt);
6522 sregs->idt.limit = dt.size;
6523 sregs->idt.base = dt.address;
6524 kvm_x86_ops->get_gdt(vcpu, &dt);
6525 sregs->gdt.limit = dt.size;
6526 sregs->gdt.base = dt.address;
6528 sregs->cr0 = kvm_read_cr0(vcpu);
6529 sregs->cr2 = vcpu->arch.cr2;
6530 sregs->cr3 = kvm_read_cr3(vcpu);
6531 sregs->cr4 = kvm_read_cr4(vcpu);
6532 sregs->cr8 = kvm_get_cr8(vcpu);
6533 sregs->efer = vcpu->arch.efer;
6534 sregs->apic_base = kvm_get_apic_base(vcpu);
6536 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6538 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6539 set_bit(vcpu->arch.interrupt.nr,
6540 (unsigned long *)sregs->interrupt_bitmap);
6545 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6546 struct kvm_mp_state *mp_state)
6548 kvm_apic_accept_events(vcpu);
6549 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6550 vcpu->arch.pv.pv_unhalted)
6551 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6553 mp_state->mp_state = vcpu->arch.mp_state;
6558 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6559 struct kvm_mp_state *mp_state)
6561 if (!kvm_vcpu_has_lapic(vcpu) &&
6562 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6565 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6566 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6567 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6569 vcpu->arch.mp_state = mp_state->mp_state;
6570 kvm_make_request(KVM_REQ_EVENT, vcpu);
6574 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6575 int reason, bool has_error_code, u32 error_code)
6577 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6580 init_emulate_ctxt(vcpu);
6582 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6583 has_error_code, error_code);
6586 return EMULATE_FAIL;
6588 kvm_rip_write(vcpu, ctxt->eip);
6589 kvm_set_rflags(vcpu, ctxt->eflags);
6590 kvm_make_request(KVM_REQ_EVENT, vcpu);
6591 return EMULATE_DONE;
6593 EXPORT_SYMBOL_GPL(kvm_task_switch);
6595 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6596 struct kvm_sregs *sregs)
6598 struct msr_data apic_base_msr;
6599 int mmu_reset_needed = 0;
6600 int pending_vec, max_bits, idx;
6603 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6606 dt.size = sregs->idt.limit;
6607 dt.address = sregs->idt.base;
6608 kvm_x86_ops->set_idt(vcpu, &dt);
6609 dt.size = sregs->gdt.limit;
6610 dt.address = sregs->gdt.base;
6611 kvm_x86_ops->set_gdt(vcpu, &dt);
6613 vcpu->arch.cr2 = sregs->cr2;
6614 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6615 vcpu->arch.cr3 = sregs->cr3;
6616 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6618 kvm_set_cr8(vcpu, sregs->cr8);
6620 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6621 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6622 apic_base_msr.data = sregs->apic_base;
6623 apic_base_msr.host_initiated = true;
6624 kvm_set_apic_base(vcpu, &apic_base_msr);
6626 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6627 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6628 vcpu->arch.cr0 = sregs->cr0;
6630 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6631 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6632 if (sregs->cr4 & X86_CR4_OSXSAVE)
6633 kvm_update_cpuid(vcpu);
6635 idx = srcu_read_lock(&vcpu->kvm->srcu);
6636 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6637 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6638 mmu_reset_needed = 1;
6640 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6642 if (mmu_reset_needed)
6643 kvm_mmu_reset_context(vcpu);
6645 max_bits = KVM_NR_INTERRUPTS;
6646 pending_vec = find_first_bit(
6647 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6648 if (pending_vec < max_bits) {
6649 kvm_queue_interrupt(vcpu, pending_vec, false);
6650 pr_debug("Set back pending irq %d\n", pending_vec);
6653 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6654 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6655 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6656 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6657 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6658 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6660 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6661 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6663 update_cr8_intercept(vcpu);
6665 /* Older userspace won't unhalt the vcpu on reset. */
6666 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6667 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6669 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6671 kvm_make_request(KVM_REQ_EVENT, vcpu);
6676 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6677 struct kvm_guest_debug *dbg)
6679 unsigned long rflags;
6682 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6684 if (vcpu->arch.exception.pending)
6686 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6687 kvm_queue_exception(vcpu, DB_VECTOR);
6689 kvm_queue_exception(vcpu, BP_VECTOR);
6693 * Read rflags as long as potentially injected trace flags are still
6696 rflags = kvm_get_rflags(vcpu);
6698 vcpu->guest_debug = dbg->control;
6699 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6700 vcpu->guest_debug = 0;
6702 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6703 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6704 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6705 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6707 for (i = 0; i < KVM_NR_DB_REGS; i++)
6708 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6710 kvm_update_dr7(vcpu);
6712 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6713 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6714 get_segment_base(vcpu, VCPU_SREG_CS);
6717 * Trigger an rflags update that will inject or remove the trace
6720 kvm_set_rflags(vcpu, rflags);
6722 kvm_x86_ops->update_db_bp_intercept(vcpu);
6732 * Translate a guest virtual address to a guest physical address.
6734 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6735 struct kvm_translation *tr)
6737 unsigned long vaddr = tr->linear_address;
6741 idx = srcu_read_lock(&vcpu->kvm->srcu);
6742 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6743 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6744 tr->physical_address = gpa;
6745 tr->valid = gpa != UNMAPPED_GVA;
6752 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6754 struct i387_fxsave_struct *fxsave =
6755 &vcpu->arch.guest_fpu.state->fxsave;
6757 memcpy(fpu->fpr, fxsave->st_space, 128);
6758 fpu->fcw = fxsave->cwd;
6759 fpu->fsw = fxsave->swd;
6760 fpu->ftwx = fxsave->twd;
6761 fpu->last_opcode = fxsave->fop;
6762 fpu->last_ip = fxsave->rip;
6763 fpu->last_dp = fxsave->rdp;
6764 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6769 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6771 struct i387_fxsave_struct *fxsave =
6772 &vcpu->arch.guest_fpu.state->fxsave;
6774 memcpy(fxsave->st_space, fpu->fpr, 128);
6775 fxsave->cwd = fpu->fcw;
6776 fxsave->swd = fpu->fsw;
6777 fxsave->twd = fpu->ftwx;
6778 fxsave->fop = fpu->last_opcode;
6779 fxsave->rip = fpu->last_ip;
6780 fxsave->rdp = fpu->last_dp;
6781 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6786 int fx_init(struct kvm_vcpu *vcpu)
6790 err = fpu_alloc(&vcpu->arch.guest_fpu);
6794 fpu_finit(&vcpu->arch.guest_fpu);
6797 * Ensure guest xcr0 is valid for loading
6799 vcpu->arch.xcr0 = XSTATE_FP;
6801 vcpu->arch.cr0 |= X86_CR0_ET;
6805 EXPORT_SYMBOL_GPL(fx_init);
6807 static void fx_free(struct kvm_vcpu *vcpu)
6809 fpu_free(&vcpu->arch.guest_fpu);
6812 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6814 if (vcpu->guest_fpu_loaded)
6818 * Restore all possible states in the guest,
6819 * and assume host would use all available bits.
6820 * Guest xcr0 would be loaded later.
6822 kvm_put_guest_xcr0(vcpu);
6823 vcpu->guest_fpu_loaded = 1;
6824 __kernel_fpu_begin();
6825 fpu_restore_checking(&vcpu->arch.guest_fpu);
6829 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6831 kvm_put_guest_xcr0(vcpu);
6833 if (!vcpu->guest_fpu_loaded)
6836 vcpu->guest_fpu_loaded = 0;
6837 fpu_save_init(&vcpu->arch.guest_fpu);
6839 ++vcpu->stat.fpu_reload;
6840 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6844 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6846 kvmclock_reset(vcpu);
6848 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6850 kvm_x86_ops->vcpu_free(vcpu);
6853 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6856 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6857 printk_once(KERN_WARNING
6858 "kvm: SMP vm created on host with unstable TSC; "
6859 "guest TSC will not be reliable\n");
6860 return kvm_x86_ops->vcpu_create(kvm, id);
6863 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6867 vcpu->arch.mtrr_state.have_fixed = 1;
6868 r = vcpu_load(vcpu);
6871 kvm_vcpu_reset(vcpu);
6872 kvm_mmu_setup(vcpu);
6878 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6881 struct msr_data msr;
6882 struct kvm *kvm = vcpu->kvm;
6884 r = vcpu_load(vcpu);
6888 msr.index = MSR_IA32_TSC;
6889 msr.host_initiated = true;
6890 kvm_write_tsc(vcpu, &msr);
6893 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
6894 KVMCLOCK_SYNC_PERIOD);
6899 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6902 vcpu->arch.apf.msr_val = 0;
6904 r = vcpu_load(vcpu);
6906 kvm_mmu_unload(vcpu);
6910 kvm_x86_ops->vcpu_free(vcpu);
6913 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6915 atomic_set(&vcpu->arch.nmi_queued, 0);
6916 vcpu->arch.nmi_pending = 0;
6917 vcpu->arch.nmi_injected = false;
6918 kvm_clear_interrupt_queue(vcpu);
6919 kvm_clear_exception_queue(vcpu);
6921 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6922 vcpu->arch.dr6 = DR6_INIT;
6923 kvm_update_dr6(vcpu);
6924 vcpu->arch.dr7 = DR7_FIXED_1;
6925 kvm_update_dr7(vcpu);
6927 kvm_make_request(KVM_REQ_EVENT, vcpu);
6928 vcpu->arch.apf.msr_val = 0;
6929 vcpu->arch.st.msr_val = 0;
6931 kvmclock_reset(vcpu);
6933 kvm_clear_async_pf_completion_queue(vcpu);
6934 kvm_async_pf_hash_reset(vcpu);
6935 vcpu->arch.apf.halted = false;
6937 kvm_pmu_reset(vcpu);
6939 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6940 vcpu->arch.regs_avail = ~0;
6941 vcpu->arch.regs_dirty = ~0;
6943 kvm_x86_ops->vcpu_reset(vcpu);
6946 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, unsigned int vector)
6948 struct kvm_segment cs;
6950 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6951 cs.selector = vector << 8;
6952 cs.base = vector << 12;
6953 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6954 kvm_rip_write(vcpu, 0);
6957 int kvm_arch_hardware_enable(void *garbage)
6960 struct kvm_vcpu *vcpu;
6965 bool stable, backwards_tsc = false;
6967 kvm_shared_msr_cpu_online();
6968 ret = kvm_x86_ops->hardware_enable(garbage);
6972 local_tsc = native_read_tsc();
6973 stable = !check_tsc_unstable();
6974 list_for_each_entry(kvm, &vm_list, vm_list) {
6975 kvm_for_each_vcpu(i, vcpu, kvm) {
6976 if (!stable && vcpu->cpu == smp_processor_id())
6977 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6978 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6979 backwards_tsc = true;
6980 if (vcpu->arch.last_host_tsc > max_tsc)
6981 max_tsc = vcpu->arch.last_host_tsc;
6987 * Sometimes, even reliable TSCs go backwards. This happens on
6988 * platforms that reset TSC during suspend or hibernate actions, but
6989 * maintain synchronization. We must compensate. Fortunately, we can
6990 * detect that condition here, which happens early in CPU bringup,
6991 * before any KVM threads can be running. Unfortunately, we can't
6992 * bring the TSCs fully up to date with real time, as we aren't yet far
6993 * enough into CPU bringup that we know how much real time has actually
6994 * elapsed; our helper function, get_kernel_ns() will be using boot
6995 * variables that haven't been updated yet.
6997 * So we simply find the maximum observed TSC above, then record the
6998 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6999 * the adjustment will be applied. Note that we accumulate
7000 * adjustments, in case multiple suspend cycles happen before some VCPU
7001 * gets a chance to run again. In the event that no KVM threads get a
7002 * chance to run, we will miss the entire elapsed period, as we'll have
7003 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7004 * loose cycle time. This isn't too big a deal, since the loss will be
7005 * uniform across all VCPUs (not to mention the scenario is extremely
7006 * unlikely). It is possible that a second hibernate recovery happens
7007 * much faster than a first, causing the observed TSC here to be
7008 * smaller; this would require additional padding adjustment, which is
7009 * why we set last_host_tsc to the local tsc observed here.
7011 * N.B. - this code below runs only on platforms with reliable TSC,
7012 * as that is the only way backwards_tsc is set above. Also note
7013 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7014 * have the same delta_cyc adjustment applied if backwards_tsc
7015 * is detected. Note further, this adjustment is only done once,
7016 * as we reset last_host_tsc on all VCPUs to stop this from being
7017 * called multiple times (one for each physical CPU bringup).
7019 * Platforms with unreliable TSCs don't have to deal with this, they
7020 * will be compensated by the logic in vcpu_load, which sets the TSC to
7021 * catchup mode. This will catchup all VCPUs to real time, but cannot
7022 * guarantee that they stay in perfect synchronization.
7024 if (backwards_tsc) {
7025 u64 delta_cyc = max_tsc - local_tsc;
7026 backwards_tsc_observed = true;
7027 list_for_each_entry(kvm, &vm_list, vm_list) {
7028 kvm_for_each_vcpu(i, vcpu, kvm) {
7029 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7030 vcpu->arch.last_host_tsc = local_tsc;
7031 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
7036 * We have to disable TSC offset matching.. if you were
7037 * booting a VM while issuing an S4 host suspend....
7038 * you may have some problem. Solving this issue is
7039 * left as an exercise to the reader.
7041 kvm->arch.last_tsc_nsec = 0;
7042 kvm->arch.last_tsc_write = 0;
7049 void kvm_arch_hardware_disable(void *garbage)
7051 kvm_x86_ops->hardware_disable(garbage);
7052 drop_user_return_notifiers(garbage);
7055 int kvm_arch_hardware_setup(void)
7057 return kvm_x86_ops->hardware_setup();
7060 void kvm_arch_hardware_unsetup(void)
7062 kvm_x86_ops->hardware_unsetup();
7065 void kvm_arch_check_processor_compat(void *rtn)
7067 kvm_x86_ops->check_processor_compatibility(rtn);
7070 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7072 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7075 struct static_key kvm_no_apic_vcpu __read_mostly;
7077 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7083 BUG_ON(vcpu->kvm == NULL);
7086 vcpu->arch.pv.pv_unhalted = false;
7087 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7088 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
7089 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7091 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7093 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7098 vcpu->arch.pio_data = page_address(page);
7100 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7102 r = kvm_mmu_create(vcpu);
7104 goto fail_free_pio_data;
7106 if (irqchip_in_kernel(kvm)) {
7107 r = kvm_create_lapic(vcpu);
7109 goto fail_mmu_destroy;
7111 static_key_slow_inc(&kvm_no_apic_vcpu);
7113 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7115 if (!vcpu->arch.mce_banks) {
7117 goto fail_free_lapic;
7119 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7121 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7123 goto fail_free_mce_banks;
7128 goto fail_free_wbinvd_dirty_mask;
7130 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7131 vcpu->arch.pv_time_enabled = false;
7133 vcpu->arch.guest_supported_xcr0 = 0;
7134 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7136 kvm_async_pf_hash_reset(vcpu);
7140 fail_free_wbinvd_dirty_mask:
7141 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7142 fail_free_mce_banks:
7143 kfree(vcpu->arch.mce_banks);
7145 kvm_free_lapic(vcpu);
7147 kvm_mmu_destroy(vcpu);
7149 free_page((unsigned long)vcpu->arch.pio_data);
7154 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7158 kvm_pmu_destroy(vcpu);
7159 kfree(vcpu->arch.mce_banks);
7160 kvm_free_lapic(vcpu);
7161 idx = srcu_read_lock(&vcpu->kvm->srcu);
7162 kvm_mmu_destroy(vcpu);
7163 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7164 free_page((unsigned long)vcpu->arch.pio_data);
7165 if (!irqchip_in_kernel(vcpu->kvm))
7166 static_key_slow_dec(&kvm_no_apic_vcpu);
7169 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7174 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7175 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7176 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7177 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7179 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7180 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7181 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7182 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7183 &kvm->arch.irq_sources_bitmap);
7185 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7186 mutex_init(&kvm->arch.apic_map_lock);
7187 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7189 pvclock_update_vm_gtod_copy(kvm);
7191 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7192 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7197 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7200 r = vcpu_load(vcpu);
7202 kvm_mmu_unload(vcpu);
7206 static void kvm_free_vcpus(struct kvm *kvm)
7209 struct kvm_vcpu *vcpu;
7212 * Unpin any mmu pages first.
7214 kvm_for_each_vcpu(i, vcpu, kvm) {
7215 kvm_clear_async_pf_completion_queue(vcpu);
7216 kvm_unload_vcpu_mmu(vcpu);
7218 kvm_for_each_vcpu(i, vcpu, kvm)
7219 kvm_arch_vcpu_free(vcpu);
7221 mutex_lock(&kvm->lock);
7222 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7223 kvm->vcpus[i] = NULL;
7225 atomic_set(&kvm->online_vcpus, 0);
7226 mutex_unlock(&kvm->lock);
7229 void kvm_arch_sync_events(struct kvm *kvm)
7231 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7232 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7233 kvm_free_all_assigned_devices(kvm);
7237 void kvm_arch_destroy_vm(struct kvm *kvm)
7239 if (current->mm == kvm->mm) {
7241 * Free memory regions allocated on behalf of userspace,
7242 * unless the the memory map has changed due to process exit
7245 struct kvm_userspace_memory_region mem;
7246 memset(&mem, 0, sizeof(mem));
7247 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7248 kvm_set_memory_region(kvm, &mem);
7250 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7251 kvm_set_memory_region(kvm, &mem);
7253 mem.slot = TSS_PRIVATE_MEMSLOT;
7254 kvm_set_memory_region(kvm, &mem);
7256 kvm_iommu_unmap_guest(kvm);
7257 kfree(kvm->arch.vpic);
7258 kfree(kvm->arch.vioapic);
7259 kvm_free_vcpus(kvm);
7260 if (kvm->arch.apic_access_page)
7261 put_page(kvm->arch.apic_access_page);
7262 if (kvm->arch.ept_identity_pagetable)
7263 put_page(kvm->arch.ept_identity_pagetable);
7264 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7267 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7268 struct kvm_memory_slot *dont)
7272 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7273 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7274 kvm_kvfree(free->arch.rmap[i]);
7275 free->arch.rmap[i] = NULL;
7280 if (!dont || free->arch.lpage_info[i - 1] !=
7281 dont->arch.lpage_info[i - 1]) {
7282 kvm_kvfree(free->arch.lpage_info[i - 1]);
7283 free->arch.lpage_info[i - 1] = NULL;
7288 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7289 unsigned long npages)
7293 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7298 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7299 slot->base_gfn, level) + 1;
7301 slot->arch.rmap[i] =
7302 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7303 if (!slot->arch.rmap[i])
7308 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7309 sizeof(*slot->arch.lpage_info[i - 1]));
7310 if (!slot->arch.lpage_info[i - 1])
7313 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7314 slot->arch.lpage_info[i - 1][0].write_count = 1;
7315 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7316 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7317 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7319 * If the gfn and userspace address are not aligned wrt each
7320 * other, or if explicitly asked to, disable large page
7321 * support for this slot
7323 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7324 !kvm_largepages_enabled()) {
7327 for (j = 0; j < lpages; ++j)
7328 slot->arch.lpage_info[i - 1][j].write_count = 1;
7335 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7336 kvm_kvfree(slot->arch.rmap[i]);
7337 slot->arch.rmap[i] = NULL;
7341 kvm_kvfree(slot->arch.lpage_info[i - 1]);
7342 slot->arch.lpage_info[i - 1] = NULL;
7347 void kvm_arch_memslots_updated(struct kvm *kvm)
7350 * memslots->generation has been incremented.
7351 * mmio generation may have reached its maximum value.
7353 kvm_mmu_invalidate_mmio_sptes(kvm);
7356 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7357 struct kvm_memory_slot *memslot,
7358 struct kvm_userspace_memory_region *mem,
7359 enum kvm_mr_change change)
7362 * Only private memory slots need to be mapped here since
7363 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7365 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7366 unsigned long userspace_addr;
7369 * MAP_SHARED to prevent internal slot pages from being moved
7372 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7373 PROT_READ | PROT_WRITE,
7374 MAP_SHARED | MAP_ANONYMOUS, 0);
7376 if (IS_ERR((void *)userspace_addr))
7377 return PTR_ERR((void *)userspace_addr);
7379 memslot->userspace_addr = userspace_addr;
7385 void kvm_arch_commit_memory_region(struct kvm *kvm,
7386 struct kvm_userspace_memory_region *mem,
7387 const struct kvm_memory_slot *old,
7388 enum kvm_mr_change change)
7391 int nr_mmu_pages = 0;
7393 if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7396 ret = vm_munmap(old->userspace_addr,
7397 old->npages * PAGE_SIZE);
7400 "kvm_vm_ioctl_set_memory_region: "
7401 "failed to munmap memory\n");
7404 if (!kvm->arch.n_requested_mmu_pages)
7405 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7408 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7410 * Write protect all pages for dirty logging.
7412 * All the sptes including the large sptes which point to this
7413 * slot are set to readonly. We can not create any new large
7414 * spte on this slot until the end of the logging.
7416 * See the comments in fast_page_fault().
7418 if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
7419 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7422 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7424 kvm_mmu_invalidate_zap_all_pages(kvm);
7427 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7428 struct kvm_memory_slot *slot)
7430 kvm_mmu_invalidate_zap_all_pages(kvm);
7433 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7435 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7436 kvm_x86_ops->check_nested_events(vcpu, false);
7438 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7439 !vcpu->arch.apf.halted)
7440 || !list_empty_careful(&vcpu->async_pf.done)
7441 || kvm_apic_has_events(vcpu)
7442 || vcpu->arch.pv.pv_unhalted
7443 || atomic_read(&vcpu->arch.nmi_queued) ||
7444 (kvm_arch_interrupt_allowed(vcpu) &&
7445 kvm_cpu_has_interrupt(vcpu));
7448 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7450 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7453 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7455 return kvm_x86_ops->interrupt_allowed(vcpu);
7458 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7460 unsigned long current_rip = kvm_rip_read(vcpu) +
7461 get_segment_base(vcpu, VCPU_SREG_CS);
7463 return current_rip == linear_rip;
7465 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7467 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7469 unsigned long rflags;
7471 rflags = kvm_x86_ops->get_rflags(vcpu);
7472 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7473 rflags &= ~X86_EFLAGS_TF;
7476 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7478 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7480 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7481 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7482 rflags |= X86_EFLAGS_TF;
7483 kvm_x86_ops->set_rflags(vcpu, rflags);
7486 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7488 __kvm_set_rflags(vcpu, rflags);
7489 kvm_make_request(KVM_REQ_EVENT, vcpu);
7491 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7493 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7497 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7501 r = kvm_mmu_reload(vcpu);
7505 if (!vcpu->arch.mmu.direct_map &&
7506 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7509 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7512 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7514 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7517 static inline u32 kvm_async_pf_next_probe(u32 key)
7519 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7522 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7524 u32 key = kvm_async_pf_hash_fn(gfn);
7526 while (vcpu->arch.apf.gfns[key] != ~0)
7527 key = kvm_async_pf_next_probe(key);
7529 vcpu->arch.apf.gfns[key] = gfn;
7532 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7535 u32 key = kvm_async_pf_hash_fn(gfn);
7537 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7538 (vcpu->arch.apf.gfns[key] != gfn &&
7539 vcpu->arch.apf.gfns[key] != ~0); i++)
7540 key = kvm_async_pf_next_probe(key);
7545 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7547 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7550 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7554 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7556 vcpu->arch.apf.gfns[i] = ~0;
7558 j = kvm_async_pf_next_probe(j);
7559 if (vcpu->arch.apf.gfns[j] == ~0)
7561 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7563 * k lies cyclically in ]i,j]
7565 * |....j i.k.| or |.k..j i...|
7567 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7568 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7573 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7576 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7580 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7581 struct kvm_async_pf *work)
7583 struct x86_exception fault;
7585 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7586 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7588 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7589 (vcpu->arch.apf.send_user_only &&
7590 kvm_x86_ops->get_cpl(vcpu) == 0))
7591 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7592 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7593 fault.vector = PF_VECTOR;
7594 fault.error_code_valid = true;
7595 fault.error_code = 0;
7596 fault.nested_page_fault = false;
7597 fault.address = work->arch.token;
7598 kvm_inject_page_fault(vcpu, &fault);
7602 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7603 struct kvm_async_pf *work)
7605 struct x86_exception fault;
7607 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7608 if (work->wakeup_all)
7609 work->arch.token = ~0; /* broadcast wakeup */
7611 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7613 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7614 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7615 fault.vector = PF_VECTOR;
7616 fault.error_code_valid = true;
7617 fault.error_code = 0;
7618 fault.nested_page_fault = false;
7619 fault.address = work->arch.token;
7620 kvm_inject_page_fault(vcpu, &fault);
7622 vcpu->arch.apf.halted = false;
7623 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7626 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7628 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7631 return !kvm_event_needs_reinjection(vcpu) &&
7632 kvm_x86_ops->interrupt_allowed(vcpu);
7635 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
7637 atomic_inc(&kvm->arch.noncoherent_dma_count);
7639 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
7641 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
7643 atomic_dec(&kvm->arch.noncoherent_dma_count);
7645 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
7647 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
7649 return atomic_read(&kvm->arch.noncoherent_dma_count);
7651 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
7653 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7654 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7655 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7656 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7657 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7658 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7659 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7660 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7661 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7662 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7663 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7664 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7665 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);