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
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Amit Shah <amit.shah@qumranet.com>
14 * Ben-Ami Yassour <benami@il.ibm.com>
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
21 #include <linux/kvm_host.h>
26 #include "kvm_cache_regs.h"
29 #include <linux/clocksource.h>
30 #include <linux/interrupt.h>
31 #include <linux/kvm.h>
33 #include <linux/vmalloc.h>
34 #include <linux/module.h>
35 #include <linux/mman.h>
36 #include <linux/highmem.h>
37 #include <linux/iommu.h>
38 #include <linux/intel-iommu.h>
39 #include <linux/cpufreq.h>
40 #include <linux/user-return-notifier.h>
41 #include <linux/srcu.h>
42 #include <linux/slab.h>
43 #include <trace/events/kvm.h>
45 #define CREATE_TRACE_POINTS
48 #include <asm/debugreg.h>
49 #include <asm/uaccess.h>
55 #define MAX_IO_MSRS 256
56 #define CR0_RESERVED_BITS \
57 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
58 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
59 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
60 #define CR4_RESERVED_BITS \
61 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
62 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
63 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
64 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
66 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
68 #define KVM_MAX_MCE_BANKS 32
69 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
72 * - enable syscall per default because its emulated by KVM
73 * - enable LME and LMA per default on 64 bit KVM
76 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
78 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
81 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
82 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
84 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
85 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
86 struct kvm_cpuid_entry2 __user *entries);
88 struct kvm_x86_ops *kvm_x86_ops;
89 EXPORT_SYMBOL_GPL(kvm_x86_ops);
92 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
94 #define KVM_NR_SHARED_MSRS 16
96 struct kvm_shared_msrs_global {
98 u32 msrs[KVM_NR_SHARED_MSRS];
101 struct kvm_shared_msrs {
102 struct user_return_notifier urn;
104 struct kvm_shared_msr_values {
107 } values[KVM_NR_SHARED_MSRS];
110 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
111 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
113 struct kvm_stats_debugfs_item debugfs_entries[] = {
114 { "pf_fixed", VCPU_STAT(pf_fixed) },
115 { "pf_guest", VCPU_STAT(pf_guest) },
116 { "tlb_flush", VCPU_STAT(tlb_flush) },
117 { "invlpg", VCPU_STAT(invlpg) },
118 { "exits", VCPU_STAT(exits) },
119 { "io_exits", VCPU_STAT(io_exits) },
120 { "mmio_exits", VCPU_STAT(mmio_exits) },
121 { "signal_exits", VCPU_STAT(signal_exits) },
122 { "irq_window", VCPU_STAT(irq_window_exits) },
123 { "nmi_window", VCPU_STAT(nmi_window_exits) },
124 { "halt_exits", VCPU_STAT(halt_exits) },
125 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
126 { "hypercalls", VCPU_STAT(hypercalls) },
127 { "request_irq", VCPU_STAT(request_irq_exits) },
128 { "irq_exits", VCPU_STAT(irq_exits) },
129 { "host_state_reload", VCPU_STAT(host_state_reload) },
130 { "efer_reload", VCPU_STAT(efer_reload) },
131 { "fpu_reload", VCPU_STAT(fpu_reload) },
132 { "insn_emulation", VCPU_STAT(insn_emulation) },
133 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
134 { "irq_injections", VCPU_STAT(irq_injections) },
135 { "nmi_injections", VCPU_STAT(nmi_injections) },
136 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
137 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
138 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
139 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
140 { "mmu_flooded", VM_STAT(mmu_flooded) },
141 { "mmu_recycled", VM_STAT(mmu_recycled) },
142 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
143 { "mmu_unsync", VM_STAT(mmu_unsync) },
144 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
145 { "largepages", VM_STAT(lpages) },
149 static void kvm_on_user_return(struct user_return_notifier *urn)
152 struct kvm_shared_msrs *locals
153 = container_of(urn, struct kvm_shared_msrs, urn);
154 struct kvm_shared_msr_values *values;
156 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
157 values = &locals->values[slot];
158 if (values->host != values->curr) {
159 wrmsrl(shared_msrs_global.msrs[slot], values->host);
160 values->curr = values->host;
163 locals->registered = false;
164 user_return_notifier_unregister(urn);
167 static void shared_msr_update(unsigned slot, u32 msr)
169 struct kvm_shared_msrs *smsr;
172 smsr = &__get_cpu_var(shared_msrs);
173 /* only read, and nobody should modify it at this time,
174 * so don't need lock */
175 if (slot >= shared_msrs_global.nr) {
176 printk(KERN_ERR "kvm: invalid MSR slot!");
179 rdmsrl_safe(msr, &value);
180 smsr->values[slot].host = value;
181 smsr->values[slot].curr = value;
184 void kvm_define_shared_msr(unsigned slot, u32 msr)
186 if (slot >= shared_msrs_global.nr)
187 shared_msrs_global.nr = slot + 1;
188 shared_msrs_global.msrs[slot] = msr;
189 /* we need ensured the shared_msr_global have been updated */
192 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
194 static void kvm_shared_msr_cpu_online(void)
198 for (i = 0; i < shared_msrs_global.nr; ++i)
199 shared_msr_update(i, shared_msrs_global.msrs[i]);
202 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
204 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
206 if (((value ^ smsr->values[slot].curr) & mask) == 0)
208 smsr->values[slot].curr = value;
209 wrmsrl(shared_msrs_global.msrs[slot], value);
210 if (!smsr->registered) {
211 smsr->urn.on_user_return = kvm_on_user_return;
212 user_return_notifier_register(&smsr->urn);
213 smsr->registered = true;
216 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
218 static void drop_user_return_notifiers(void *ignore)
220 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
222 if (smsr->registered)
223 kvm_on_user_return(&smsr->urn);
226 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
228 if (irqchip_in_kernel(vcpu->kvm))
229 return vcpu->arch.apic_base;
231 return vcpu->arch.apic_base;
233 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
235 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
237 /* TODO: reserve bits check */
238 if (irqchip_in_kernel(vcpu->kvm))
239 kvm_lapic_set_base(vcpu, data);
241 vcpu->arch.apic_base = data;
243 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
245 #define EXCPT_BENIGN 0
246 #define EXCPT_CONTRIBUTORY 1
249 static int exception_class(int vector)
259 return EXCPT_CONTRIBUTORY;
266 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
267 unsigned nr, bool has_error, u32 error_code)
272 if (!vcpu->arch.exception.pending) {
274 vcpu->arch.exception.pending = true;
275 vcpu->arch.exception.has_error_code = has_error;
276 vcpu->arch.exception.nr = nr;
277 vcpu->arch.exception.error_code = error_code;
281 /* to check exception */
282 prev_nr = vcpu->arch.exception.nr;
283 if (prev_nr == DF_VECTOR) {
284 /* triple fault -> shutdown */
285 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
288 class1 = exception_class(prev_nr);
289 class2 = exception_class(nr);
290 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
291 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
292 /* generate double fault per SDM Table 5-5 */
293 vcpu->arch.exception.pending = true;
294 vcpu->arch.exception.has_error_code = true;
295 vcpu->arch.exception.nr = DF_VECTOR;
296 vcpu->arch.exception.error_code = 0;
298 /* replace previous exception with a new one in a hope
299 that instruction re-execution will regenerate lost
304 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
306 kvm_multiple_exception(vcpu, nr, false, 0);
308 EXPORT_SYMBOL_GPL(kvm_queue_exception);
310 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
313 ++vcpu->stat.pf_guest;
314 vcpu->arch.cr2 = addr;
315 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
318 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
320 vcpu->arch.nmi_pending = 1;
322 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
324 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
326 kvm_multiple_exception(vcpu, nr, true, error_code);
328 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
331 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
332 * a #GP and return false.
334 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
336 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
338 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
341 EXPORT_SYMBOL_GPL(kvm_require_cpl);
344 * Load the pae pdptrs. Return true is they are all valid.
346 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
348 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
349 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
352 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
354 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
355 offset * sizeof(u64), sizeof(pdpte));
360 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
361 if (is_present_gpte(pdpte[i]) &&
362 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
369 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
370 __set_bit(VCPU_EXREG_PDPTR,
371 (unsigned long *)&vcpu->arch.regs_avail);
372 __set_bit(VCPU_EXREG_PDPTR,
373 (unsigned long *)&vcpu->arch.regs_dirty);
378 EXPORT_SYMBOL_GPL(load_pdptrs);
380 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
382 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
386 if (is_long_mode(vcpu) || !is_pae(vcpu))
389 if (!test_bit(VCPU_EXREG_PDPTR,
390 (unsigned long *)&vcpu->arch.regs_avail))
393 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
396 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
402 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
407 if (cr0 & 0xffffffff00000000UL) {
408 kvm_inject_gp(vcpu, 0);
413 cr0 &= ~CR0_RESERVED_BITS;
415 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
416 kvm_inject_gp(vcpu, 0);
420 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
421 kvm_inject_gp(vcpu, 0);
425 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
427 if ((vcpu->arch.efer & EFER_LME)) {
431 kvm_inject_gp(vcpu, 0);
434 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
436 kvm_inject_gp(vcpu, 0);
442 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
443 kvm_inject_gp(vcpu, 0);
449 kvm_x86_ops->set_cr0(vcpu, cr0);
451 kvm_mmu_reset_context(vcpu);
454 EXPORT_SYMBOL_GPL(kvm_set_cr0);
456 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
458 kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0ful) | (msw & 0x0f));
460 EXPORT_SYMBOL_GPL(kvm_lmsw);
462 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
464 unsigned long old_cr4 = kvm_read_cr4(vcpu);
465 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
467 if (cr4 & CR4_RESERVED_BITS) {
468 kvm_inject_gp(vcpu, 0);
472 if (is_long_mode(vcpu)) {
473 if (!(cr4 & X86_CR4_PAE)) {
474 kvm_inject_gp(vcpu, 0);
477 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
478 && ((cr4 ^ old_cr4) & pdptr_bits)
479 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
480 kvm_inject_gp(vcpu, 0);
484 if (cr4 & X86_CR4_VMXE) {
485 kvm_inject_gp(vcpu, 0);
488 kvm_x86_ops->set_cr4(vcpu, cr4);
489 vcpu->arch.cr4 = cr4;
490 vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
491 kvm_mmu_reset_context(vcpu);
493 EXPORT_SYMBOL_GPL(kvm_set_cr4);
495 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
497 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
498 kvm_mmu_sync_roots(vcpu);
499 kvm_mmu_flush_tlb(vcpu);
503 if (is_long_mode(vcpu)) {
504 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
505 kvm_inject_gp(vcpu, 0);
510 if (cr3 & CR3_PAE_RESERVED_BITS) {
511 kvm_inject_gp(vcpu, 0);
514 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
515 kvm_inject_gp(vcpu, 0);
520 * We don't check reserved bits in nonpae mode, because
521 * this isn't enforced, and VMware depends on this.
526 * Does the new cr3 value map to physical memory? (Note, we
527 * catch an invalid cr3 even in real-mode, because it would
528 * cause trouble later on when we turn on paging anyway.)
530 * A real CPU would silently accept an invalid cr3 and would
531 * attempt to use it - with largely undefined (and often hard
532 * to debug) behavior on the guest side.
534 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
535 kvm_inject_gp(vcpu, 0);
537 vcpu->arch.cr3 = cr3;
538 vcpu->arch.mmu.new_cr3(vcpu);
541 EXPORT_SYMBOL_GPL(kvm_set_cr3);
543 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
545 if (cr8 & CR8_RESERVED_BITS) {
546 kvm_inject_gp(vcpu, 0);
549 if (irqchip_in_kernel(vcpu->kvm))
550 kvm_lapic_set_tpr(vcpu, cr8);
552 vcpu->arch.cr8 = cr8;
554 EXPORT_SYMBOL_GPL(kvm_set_cr8);
556 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
558 if (irqchip_in_kernel(vcpu->kvm))
559 return kvm_lapic_get_cr8(vcpu);
561 return vcpu->arch.cr8;
563 EXPORT_SYMBOL_GPL(kvm_get_cr8);
565 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
569 vcpu->arch.db[dr] = val;
570 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
571 vcpu->arch.eff_db[dr] = val;
574 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) {
575 kvm_queue_exception(vcpu, UD_VECTOR);
580 if (val & 0xffffffff00000000ULL) {
581 kvm_inject_gp(vcpu, 0);
584 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
587 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) {
588 kvm_queue_exception(vcpu, UD_VECTOR);
593 if (val & 0xffffffff00000000ULL) {
594 kvm_inject_gp(vcpu, 0);
597 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
598 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
599 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
600 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
607 EXPORT_SYMBOL_GPL(kvm_set_dr);
609 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
613 *val = vcpu->arch.db[dr];
616 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) {
617 kvm_queue_exception(vcpu, UD_VECTOR);
622 *val = vcpu->arch.dr6;
625 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) {
626 kvm_queue_exception(vcpu, UD_VECTOR);
631 *val = vcpu->arch.dr7;
637 EXPORT_SYMBOL_GPL(kvm_get_dr);
639 static inline u32 bit(int bitno)
641 return 1 << (bitno & 31);
645 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
646 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
648 * This list is modified at module load time to reflect the
649 * capabilities of the host cpu. This capabilities test skips MSRs that are
650 * kvm-specific. Those are put in the beginning of the list.
653 #define KVM_SAVE_MSRS_BEGIN 5
654 static u32 msrs_to_save[] = {
655 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
656 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
657 HV_X64_MSR_APIC_ASSIST_PAGE,
658 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
661 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
663 MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
666 static unsigned num_msrs_to_save;
668 static u32 emulated_msrs[] = {
669 MSR_IA32_MISC_ENABLE,
672 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
674 if (efer & efer_reserved_bits) {
675 kvm_inject_gp(vcpu, 0);
680 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) {
681 kvm_inject_gp(vcpu, 0);
685 if (efer & EFER_FFXSR) {
686 struct kvm_cpuid_entry2 *feat;
688 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
689 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
690 kvm_inject_gp(vcpu, 0);
695 if (efer & EFER_SVME) {
696 struct kvm_cpuid_entry2 *feat;
698 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
699 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
700 kvm_inject_gp(vcpu, 0);
705 kvm_x86_ops->set_efer(vcpu, efer);
708 efer |= vcpu->arch.efer & EFER_LMA;
710 vcpu->arch.efer = efer;
712 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
713 kvm_mmu_reset_context(vcpu);
716 void kvm_enable_efer_bits(u64 mask)
718 efer_reserved_bits &= ~mask;
720 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
724 * Writes msr value into into the appropriate "register".
725 * Returns 0 on success, non-0 otherwise.
726 * Assumes vcpu_load() was already called.
728 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
730 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
734 * Adapt set_msr() to msr_io()'s calling convention
736 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
738 return kvm_set_msr(vcpu, index, *data);
741 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
744 struct pvclock_wall_clock wc;
745 struct timespec boot;
752 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
755 * The guest calculates current wall clock time by adding
756 * system time (updated by kvm_write_guest_time below) to the
757 * wall clock specified here. guest system time equals host
758 * system time for us, thus we must fill in host boot time here.
762 wc.sec = boot.tv_sec;
763 wc.nsec = boot.tv_nsec;
764 wc.version = version;
766 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
769 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
772 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
774 uint32_t quotient, remainder;
776 /* Don't try to replace with do_div(), this one calculates
777 * "(dividend << 32) / divisor" */
779 : "=a" (quotient), "=d" (remainder)
780 : "0" (0), "1" (dividend), "r" (divisor) );
784 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
786 uint64_t nsecs = 1000000000LL;
791 tps64 = tsc_khz * 1000LL;
792 while (tps64 > nsecs*2) {
797 tps32 = (uint32_t)tps64;
798 while (tps32 <= (uint32_t)nsecs) {
803 hv_clock->tsc_shift = shift;
804 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
806 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
807 __func__, tsc_khz, hv_clock->tsc_shift,
808 hv_clock->tsc_to_system_mul);
811 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
813 static void kvm_write_guest_time(struct kvm_vcpu *v)
817 struct kvm_vcpu_arch *vcpu = &v->arch;
819 unsigned long this_tsc_khz;
821 if ((!vcpu->time_page))
824 this_tsc_khz = get_cpu_var(cpu_tsc_khz);
825 if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
826 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
827 vcpu->hv_clock_tsc_khz = this_tsc_khz;
829 put_cpu_var(cpu_tsc_khz);
831 /* Keep irq disabled to prevent changes to the clock */
832 local_irq_save(flags);
833 kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
835 monotonic_to_bootbased(&ts);
836 local_irq_restore(flags);
838 /* With all the info we got, fill in the values */
840 vcpu->hv_clock.system_time = ts.tv_nsec +
841 (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
844 * The interface expects us to write an even number signaling that the
845 * update is finished. Since the guest won't see the intermediate
846 * state, we just increase by 2 at the end.
848 vcpu->hv_clock.version += 2;
850 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
852 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
853 sizeof(vcpu->hv_clock));
855 kunmap_atomic(shared_kaddr, KM_USER0);
857 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
860 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
862 struct kvm_vcpu_arch *vcpu = &v->arch;
864 if (!vcpu->time_page)
866 set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
870 static bool msr_mtrr_valid(unsigned msr)
873 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
874 case MSR_MTRRfix64K_00000:
875 case MSR_MTRRfix16K_80000:
876 case MSR_MTRRfix16K_A0000:
877 case MSR_MTRRfix4K_C0000:
878 case MSR_MTRRfix4K_C8000:
879 case MSR_MTRRfix4K_D0000:
880 case MSR_MTRRfix4K_D8000:
881 case MSR_MTRRfix4K_E0000:
882 case MSR_MTRRfix4K_E8000:
883 case MSR_MTRRfix4K_F0000:
884 case MSR_MTRRfix4K_F8000:
885 case MSR_MTRRdefType:
886 case MSR_IA32_CR_PAT:
894 static bool valid_pat_type(unsigned t)
896 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
899 static bool valid_mtrr_type(unsigned t)
901 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
904 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
908 if (!msr_mtrr_valid(msr))
911 if (msr == MSR_IA32_CR_PAT) {
912 for (i = 0; i < 8; i++)
913 if (!valid_pat_type((data >> (i * 8)) & 0xff))
916 } else if (msr == MSR_MTRRdefType) {
919 return valid_mtrr_type(data & 0xff);
920 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
921 for (i = 0; i < 8 ; i++)
922 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
928 return valid_mtrr_type(data & 0xff);
931 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
933 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
935 if (!mtrr_valid(vcpu, msr, data))
938 if (msr == MSR_MTRRdefType) {
939 vcpu->arch.mtrr_state.def_type = data;
940 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
941 } else if (msr == MSR_MTRRfix64K_00000)
943 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
944 p[1 + msr - MSR_MTRRfix16K_80000] = data;
945 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
946 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
947 else if (msr == MSR_IA32_CR_PAT)
948 vcpu->arch.pat = data;
949 else { /* Variable MTRRs */
950 int idx, is_mtrr_mask;
953 idx = (msr - 0x200) / 2;
954 is_mtrr_mask = msr - 0x200 - 2 * idx;
957 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
960 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
964 kvm_mmu_reset_context(vcpu);
968 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
970 u64 mcg_cap = vcpu->arch.mcg_cap;
971 unsigned bank_num = mcg_cap & 0xff;
974 case MSR_IA32_MCG_STATUS:
975 vcpu->arch.mcg_status = data;
977 case MSR_IA32_MCG_CTL:
978 if (!(mcg_cap & MCG_CTL_P))
980 if (data != 0 && data != ~(u64)0)
982 vcpu->arch.mcg_ctl = data;
985 if (msr >= MSR_IA32_MC0_CTL &&
986 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
987 u32 offset = msr - MSR_IA32_MC0_CTL;
988 /* only 0 or all 1s can be written to IA32_MCi_CTL
989 * some Linux kernels though clear bit 10 in bank 4 to
990 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
991 * this to avoid an uncatched #GP in the guest
993 if ((offset & 0x3) == 0 &&
994 data != 0 && (data | (1 << 10)) != ~(u64)0)
996 vcpu->arch.mce_banks[offset] = data;
1004 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1006 struct kvm *kvm = vcpu->kvm;
1007 int lm = is_long_mode(vcpu);
1008 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1009 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1010 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1011 : kvm->arch.xen_hvm_config.blob_size_32;
1012 u32 page_num = data & ~PAGE_MASK;
1013 u64 page_addr = data & PAGE_MASK;
1018 if (page_num >= blob_size)
1021 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1025 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1027 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1036 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1038 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1041 static bool kvm_hv_msr_partition_wide(u32 msr)
1045 case HV_X64_MSR_GUEST_OS_ID:
1046 case HV_X64_MSR_HYPERCALL:
1054 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1056 struct kvm *kvm = vcpu->kvm;
1059 case HV_X64_MSR_GUEST_OS_ID:
1060 kvm->arch.hv_guest_os_id = data;
1061 /* setting guest os id to zero disables hypercall page */
1062 if (!kvm->arch.hv_guest_os_id)
1063 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1065 case HV_X64_MSR_HYPERCALL: {
1070 /* if guest os id is not set hypercall should remain disabled */
1071 if (!kvm->arch.hv_guest_os_id)
1073 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1074 kvm->arch.hv_hypercall = data;
1077 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1078 addr = gfn_to_hva(kvm, gfn);
1079 if (kvm_is_error_hva(addr))
1081 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1082 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1083 if (copy_to_user((void __user *)addr, instructions, 4))
1085 kvm->arch.hv_hypercall = data;
1089 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1090 "data 0x%llx\n", msr, data);
1096 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1099 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1102 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1103 vcpu->arch.hv_vapic = data;
1106 addr = gfn_to_hva(vcpu->kvm, data >>
1107 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1108 if (kvm_is_error_hva(addr))
1110 if (clear_user((void __user *)addr, PAGE_SIZE))
1112 vcpu->arch.hv_vapic = data;
1115 case HV_X64_MSR_EOI:
1116 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1117 case HV_X64_MSR_ICR:
1118 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1119 case HV_X64_MSR_TPR:
1120 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1122 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1123 "data 0x%llx\n", msr, data);
1130 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1134 set_efer(vcpu, data);
1137 data &= ~(u64)0x40; /* ignore flush filter disable */
1138 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1140 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1145 case MSR_FAM10H_MMIO_CONF_BASE:
1147 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1152 case MSR_AMD64_NB_CFG:
1154 case MSR_IA32_DEBUGCTLMSR:
1156 /* We support the non-activated case already */
1158 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1159 /* Values other than LBR and BTF are vendor-specific,
1160 thus reserved and should throw a #GP */
1163 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1166 case MSR_IA32_UCODE_REV:
1167 case MSR_IA32_UCODE_WRITE:
1168 case MSR_VM_HSAVE_PA:
1169 case MSR_AMD64_PATCH_LOADER:
1171 case 0x200 ... 0x2ff:
1172 return set_msr_mtrr(vcpu, msr, data);
1173 case MSR_IA32_APICBASE:
1174 kvm_set_apic_base(vcpu, data);
1176 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1177 return kvm_x2apic_msr_write(vcpu, msr, data);
1178 case MSR_IA32_MISC_ENABLE:
1179 vcpu->arch.ia32_misc_enable_msr = data;
1181 case MSR_KVM_WALL_CLOCK:
1182 vcpu->kvm->arch.wall_clock = data;
1183 kvm_write_wall_clock(vcpu->kvm, data);
1185 case MSR_KVM_SYSTEM_TIME: {
1186 if (vcpu->arch.time_page) {
1187 kvm_release_page_dirty(vcpu->arch.time_page);
1188 vcpu->arch.time_page = NULL;
1191 vcpu->arch.time = data;
1193 /* we verify if the enable bit is set... */
1197 /* ...but clean it before doing the actual write */
1198 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1200 vcpu->arch.time_page =
1201 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1203 if (is_error_page(vcpu->arch.time_page)) {
1204 kvm_release_page_clean(vcpu->arch.time_page);
1205 vcpu->arch.time_page = NULL;
1208 kvm_request_guest_time_update(vcpu);
1211 case MSR_IA32_MCG_CTL:
1212 case MSR_IA32_MCG_STATUS:
1213 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1214 return set_msr_mce(vcpu, msr, data);
1216 /* Performance counters are not protected by a CPUID bit,
1217 * so we should check all of them in the generic path for the sake of
1218 * cross vendor migration.
1219 * Writing a zero into the event select MSRs disables them,
1220 * which we perfectly emulate ;-). Any other value should be at least
1221 * reported, some guests depend on them.
1223 case MSR_P6_EVNTSEL0:
1224 case MSR_P6_EVNTSEL1:
1225 case MSR_K7_EVNTSEL0:
1226 case MSR_K7_EVNTSEL1:
1227 case MSR_K7_EVNTSEL2:
1228 case MSR_K7_EVNTSEL3:
1230 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1231 "0x%x data 0x%llx\n", msr, data);
1233 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1234 * so we ignore writes to make it happy.
1236 case MSR_P6_PERFCTR0:
1237 case MSR_P6_PERFCTR1:
1238 case MSR_K7_PERFCTR0:
1239 case MSR_K7_PERFCTR1:
1240 case MSR_K7_PERFCTR2:
1241 case MSR_K7_PERFCTR3:
1242 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1243 "0x%x data 0x%llx\n", msr, data);
1245 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1246 if (kvm_hv_msr_partition_wide(msr)) {
1248 mutex_lock(&vcpu->kvm->lock);
1249 r = set_msr_hyperv_pw(vcpu, msr, data);
1250 mutex_unlock(&vcpu->kvm->lock);
1253 return set_msr_hyperv(vcpu, msr, data);
1256 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1257 return xen_hvm_config(vcpu, data);
1259 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1263 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1270 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1274 * Reads an msr value (of 'msr_index') into 'pdata'.
1275 * Returns 0 on success, non-0 otherwise.
1276 * Assumes vcpu_load() was already called.
1278 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1280 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1283 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1285 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1287 if (!msr_mtrr_valid(msr))
1290 if (msr == MSR_MTRRdefType)
1291 *pdata = vcpu->arch.mtrr_state.def_type +
1292 (vcpu->arch.mtrr_state.enabled << 10);
1293 else if (msr == MSR_MTRRfix64K_00000)
1295 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1296 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1297 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1298 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1299 else if (msr == MSR_IA32_CR_PAT)
1300 *pdata = vcpu->arch.pat;
1301 else { /* Variable MTRRs */
1302 int idx, is_mtrr_mask;
1305 idx = (msr - 0x200) / 2;
1306 is_mtrr_mask = msr - 0x200 - 2 * idx;
1309 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1312 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1319 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1322 u64 mcg_cap = vcpu->arch.mcg_cap;
1323 unsigned bank_num = mcg_cap & 0xff;
1326 case MSR_IA32_P5_MC_ADDR:
1327 case MSR_IA32_P5_MC_TYPE:
1330 case MSR_IA32_MCG_CAP:
1331 data = vcpu->arch.mcg_cap;
1333 case MSR_IA32_MCG_CTL:
1334 if (!(mcg_cap & MCG_CTL_P))
1336 data = vcpu->arch.mcg_ctl;
1338 case MSR_IA32_MCG_STATUS:
1339 data = vcpu->arch.mcg_status;
1342 if (msr >= MSR_IA32_MC0_CTL &&
1343 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1344 u32 offset = msr - MSR_IA32_MC0_CTL;
1345 data = vcpu->arch.mce_banks[offset];
1354 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1357 struct kvm *kvm = vcpu->kvm;
1360 case HV_X64_MSR_GUEST_OS_ID:
1361 data = kvm->arch.hv_guest_os_id;
1363 case HV_X64_MSR_HYPERCALL:
1364 data = kvm->arch.hv_hypercall;
1367 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1375 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1380 case HV_X64_MSR_VP_INDEX: {
1383 kvm_for_each_vcpu(r, v, vcpu->kvm)
1388 case HV_X64_MSR_EOI:
1389 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1390 case HV_X64_MSR_ICR:
1391 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1392 case HV_X64_MSR_TPR:
1393 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1395 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1402 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1407 case MSR_IA32_PLATFORM_ID:
1408 case MSR_IA32_UCODE_REV:
1409 case MSR_IA32_EBL_CR_POWERON:
1410 case MSR_IA32_DEBUGCTLMSR:
1411 case MSR_IA32_LASTBRANCHFROMIP:
1412 case MSR_IA32_LASTBRANCHTOIP:
1413 case MSR_IA32_LASTINTFROMIP:
1414 case MSR_IA32_LASTINTTOIP:
1417 case MSR_VM_HSAVE_PA:
1418 case MSR_P6_PERFCTR0:
1419 case MSR_P6_PERFCTR1:
1420 case MSR_P6_EVNTSEL0:
1421 case MSR_P6_EVNTSEL1:
1422 case MSR_K7_EVNTSEL0:
1423 case MSR_K7_PERFCTR0:
1424 case MSR_K8_INT_PENDING_MSG:
1425 case MSR_AMD64_NB_CFG:
1426 case MSR_FAM10H_MMIO_CONF_BASE:
1430 data = 0x500 | KVM_NR_VAR_MTRR;
1432 case 0x200 ... 0x2ff:
1433 return get_msr_mtrr(vcpu, msr, pdata);
1434 case 0xcd: /* fsb frequency */
1437 case MSR_IA32_APICBASE:
1438 data = kvm_get_apic_base(vcpu);
1440 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1441 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1443 case MSR_IA32_MISC_ENABLE:
1444 data = vcpu->arch.ia32_misc_enable_msr;
1446 case MSR_IA32_PERF_STATUS:
1447 /* TSC increment by tick */
1449 /* CPU multiplier */
1450 data |= (((uint64_t)4ULL) << 40);
1453 data = vcpu->arch.efer;
1455 case MSR_KVM_WALL_CLOCK:
1456 data = vcpu->kvm->arch.wall_clock;
1458 case MSR_KVM_SYSTEM_TIME:
1459 data = vcpu->arch.time;
1461 case MSR_IA32_P5_MC_ADDR:
1462 case MSR_IA32_P5_MC_TYPE:
1463 case MSR_IA32_MCG_CAP:
1464 case MSR_IA32_MCG_CTL:
1465 case MSR_IA32_MCG_STATUS:
1466 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1467 return get_msr_mce(vcpu, msr, pdata);
1468 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1469 if (kvm_hv_msr_partition_wide(msr)) {
1471 mutex_lock(&vcpu->kvm->lock);
1472 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1473 mutex_unlock(&vcpu->kvm->lock);
1476 return get_msr_hyperv(vcpu, msr, pdata);
1480 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1483 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1491 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1494 * Read or write a bunch of msrs. All parameters are kernel addresses.
1496 * @return number of msrs set successfully.
1498 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1499 struct kvm_msr_entry *entries,
1500 int (*do_msr)(struct kvm_vcpu *vcpu,
1501 unsigned index, u64 *data))
1507 idx = srcu_read_lock(&vcpu->kvm->srcu);
1508 for (i = 0; i < msrs->nmsrs; ++i)
1509 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1511 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1519 * Read or write a bunch of msrs. Parameters are user addresses.
1521 * @return number of msrs set successfully.
1523 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1524 int (*do_msr)(struct kvm_vcpu *vcpu,
1525 unsigned index, u64 *data),
1528 struct kvm_msrs msrs;
1529 struct kvm_msr_entry *entries;
1534 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1538 if (msrs.nmsrs >= MAX_IO_MSRS)
1542 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1543 entries = vmalloc(size);
1548 if (copy_from_user(entries, user_msrs->entries, size))
1551 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1556 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1567 int kvm_dev_ioctl_check_extension(long ext)
1572 case KVM_CAP_IRQCHIP:
1574 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1575 case KVM_CAP_SET_TSS_ADDR:
1576 case KVM_CAP_EXT_CPUID:
1577 case KVM_CAP_CLOCKSOURCE:
1579 case KVM_CAP_NOP_IO_DELAY:
1580 case KVM_CAP_MP_STATE:
1581 case KVM_CAP_SYNC_MMU:
1582 case KVM_CAP_REINJECT_CONTROL:
1583 case KVM_CAP_IRQ_INJECT_STATUS:
1584 case KVM_CAP_ASSIGN_DEV_IRQ:
1586 case KVM_CAP_IOEVENTFD:
1588 case KVM_CAP_PIT_STATE2:
1589 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1590 case KVM_CAP_XEN_HVM:
1591 case KVM_CAP_ADJUST_CLOCK:
1592 case KVM_CAP_VCPU_EVENTS:
1593 case KVM_CAP_HYPERV:
1594 case KVM_CAP_HYPERV_VAPIC:
1595 case KVM_CAP_HYPERV_SPIN:
1596 case KVM_CAP_PCI_SEGMENT:
1597 case KVM_CAP_DEBUGREGS:
1598 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1601 case KVM_CAP_COALESCED_MMIO:
1602 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1605 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1607 case KVM_CAP_NR_VCPUS:
1610 case KVM_CAP_NR_MEMSLOTS:
1611 r = KVM_MEMORY_SLOTS;
1613 case KVM_CAP_PV_MMU: /* obsolete */
1620 r = KVM_MAX_MCE_BANKS;
1630 long kvm_arch_dev_ioctl(struct file *filp,
1631 unsigned int ioctl, unsigned long arg)
1633 void __user *argp = (void __user *)arg;
1637 case KVM_GET_MSR_INDEX_LIST: {
1638 struct kvm_msr_list __user *user_msr_list = argp;
1639 struct kvm_msr_list msr_list;
1643 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1646 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1647 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1650 if (n < msr_list.nmsrs)
1653 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1654 num_msrs_to_save * sizeof(u32)))
1656 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1658 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1663 case KVM_GET_SUPPORTED_CPUID: {
1664 struct kvm_cpuid2 __user *cpuid_arg = argp;
1665 struct kvm_cpuid2 cpuid;
1668 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1670 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1671 cpuid_arg->entries);
1676 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1681 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1684 mce_cap = KVM_MCE_CAP_SUPPORTED;
1686 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1698 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1700 kvm_x86_ops->vcpu_load(vcpu, cpu);
1701 if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1702 unsigned long khz = cpufreq_quick_get(cpu);
1705 per_cpu(cpu_tsc_khz, cpu) = khz;
1707 kvm_request_guest_time_update(vcpu);
1710 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1712 kvm_put_guest_fpu(vcpu);
1713 kvm_x86_ops->vcpu_put(vcpu);
1716 static int is_efer_nx(void)
1718 unsigned long long efer = 0;
1720 rdmsrl_safe(MSR_EFER, &efer);
1721 return efer & EFER_NX;
1724 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1727 struct kvm_cpuid_entry2 *e, *entry;
1730 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1731 e = &vcpu->arch.cpuid_entries[i];
1732 if (e->function == 0x80000001) {
1737 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1738 entry->edx &= ~(1 << 20);
1739 printk(KERN_INFO "kvm: guest NX capability removed\n");
1743 /* when an old userspace process fills a new kernel module */
1744 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1745 struct kvm_cpuid *cpuid,
1746 struct kvm_cpuid_entry __user *entries)
1749 struct kvm_cpuid_entry *cpuid_entries;
1752 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1755 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1759 if (copy_from_user(cpuid_entries, entries,
1760 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1762 for (i = 0; i < cpuid->nent; i++) {
1763 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1764 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1765 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1766 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1767 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1768 vcpu->arch.cpuid_entries[i].index = 0;
1769 vcpu->arch.cpuid_entries[i].flags = 0;
1770 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1771 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1772 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1774 vcpu->arch.cpuid_nent = cpuid->nent;
1775 cpuid_fix_nx_cap(vcpu);
1777 kvm_apic_set_version(vcpu);
1778 kvm_x86_ops->cpuid_update(vcpu);
1781 vfree(cpuid_entries);
1786 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1787 struct kvm_cpuid2 *cpuid,
1788 struct kvm_cpuid_entry2 __user *entries)
1793 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1796 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1797 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1799 vcpu->arch.cpuid_nent = cpuid->nent;
1800 kvm_apic_set_version(vcpu);
1801 kvm_x86_ops->cpuid_update(vcpu);
1808 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1809 struct kvm_cpuid2 *cpuid,
1810 struct kvm_cpuid_entry2 __user *entries)
1815 if (cpuid->nent < vcpu->arch.cpuid_nent)
1818 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1819 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1824 cpuid->nent = vcpu->arch.cpuid_nent;
1828 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1831 entry->function = function;
1832 entry->index = index;
1833 cpuid_count(entry->function, entry->index,
1834 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1838 #define F(x) bit(X86_FEATURE_##x)
1840 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1841 u32 index, int *nent, int maxnent)
1843 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1844 #ifdef CONFIG_X86_64
1845 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1847 unsigned f_lm = F(LM);
1849 unsigned f_gbpages = 0;
1852 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1855 const u32 kvm_supported_word0_x86_features =
1856 F(FPU) | F(VME) | F(DE) | F(PSE) |
1857 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1858 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1859 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1860 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1861 0 /* Reserved, DS, ACPI */ | F(MMX) |
1862 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1863 0 /* HTT, TM, Reserved, PBE */;
1864 /* cpuid 0x80000001.edx */
1865 const u32 kvm_supported_word1_x86_features =
1866 F(FPU) | F(VME) | F(DE) | F(PSE) |
1867 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1868 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1869 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1870 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1871 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1872 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1873 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1875 const u32 kvm_supported_word4_x86_features =
1876 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1877 0 /* DS-CPL, VMX, SMX, EST */ |
1878 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1879 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1880 0 /* Reserved, DCA */ | F(XMM4_1) |
1881 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1882 0 /* Reserved, XSAVE, OSXSAVE */;
1883 /* cpuid 0x80000001.ecx */
1884 const u32 kvm_supported_word6_x86_features =
1885 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1886 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1887 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1888 0 /* SKINIT */ | 0 /* WDT */;
1890 /* all calls to cpuid_count() should be made on the same cpu */
1892 do_cpuid_1_ent(entry, function, index);
1897 entry->eax = min(entry->eax, (u32)0xb);
1900 entry->edx &= kvm_supported_word0_x86_features;
1901 entry->ecx &= kvm_supported_word4_x86_features;
1902 /* we support x2apic emulation even if host does not support
1903 * it since we emulate x2apic in software */
1904 entry->ecx |= F(X2APIC);
1906 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1907 * may return different values. This forces us to get_cpu() before
1908 * issuing the first command, and also to emulate this annoying behavior
1909 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1911 int t, times = entry->eax & 0xff;
1913 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1914 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1915 for (t = 1; t < times && *nent < maxnent; ++t) {
1916 do_cpuid_1_ent(&entry[t], function, 0);
1917 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1922 /* function 4 and 0xb have additional index. */
1926 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1927 /* read more entries until cache_type is zero */
1928 for (i = 1; *nent < maxnent; ++i) {
1929 cache_type = entry[i - 1].eax & 0x1f;
1932 do_cpuid_1_ent(&entry[i], function, i);
1934 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1942 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1943 /* read more entries until level_type is zero */
1944 for (i = 1; *nent < maxnent; ++i) {
1945 level_type = entry[i - 1].ecx & 0xff00;
1948 do_cpuid_1_ent(&entry[i], function, i);
1950 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1956 entry->eax = min(entry->eax, 0x8000001a);
1959 entry->edx &= kvm_supported_word1_x86_features;
1960 entry->ecx &= kvm_supported_word6_x86_features;
1968 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1969 struct kvm_cpuid_entry2 __user *entries)
1971 struct kvm_cpuid_entry2 *cpuid_entries;
1972 int limit, nent = 0, r = -E2BIG;
1975 if (cpuid->nent < 1)
1977 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1978 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1980 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1984 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1985 limit = cpuid_entries[0].eax;
1986 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1987 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1988 &nent, cpuid->nent);
1990 if (nent >= cpuid->nent)
1993 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1994 limit = cpuid_entries[nent - 1].eax;
1995 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1996 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1997 &nent, cpuid->nent);
1999 if (nent >= cpuid->nent)
2003 if (copy_to_user(entries, cpuid_entries,
2004 nent * sizeof(struct kvm_cpuid_entry2)))
2010 vfree(cpuid_entries);
2015 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2016 struct kvm_lapic_state *s)
2019 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2025 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2026 struct kvm_lapic_state *s)
2029 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2030 kvm_apic_post_state_restore(vcpu);
2031 update_cr8_intercept(vcpu);
2037 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2038 struct kvm_interrupt *irq)
2040 if (irq->irq < 0 || irq->irq >= 256)
2042 if (irqchip_in_kernel(vcpu->kvm))
2046 kvm_queue_interrupt(vcpu, irq->irq, false);
2053 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2056 kvm_inject_nmi(vcpu);
2062 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2063 struct kvm_tpr_access_ctl *tac)
2067 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2071 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2075 unsigned bank_num = mcg_cap & 0xff, bank;
2078 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2080 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2083 vcpu->arch.mcg_cap = mcg_cap;
2084 /* Init IA32_MCG_CTL to all 1s */
2085 if (mcg_cap & MCG_CTL_P)
2086 vcpu->arch.mcg_ctl = ~(u64)0;
2087 /* Init IA32_MCi_CTL to all 1s */
2088 for (bank = 0; bank < bank_num; bank++)
2089 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2094 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2095 struct kvm_x86_mce *mce)
2097 u64 mcg_cap = vcpu->arch.mcg_cap;
2098 unsigned bank_num = mcg_cap & 0xff;
2099 u64 *banks = vcpu->arch.mce_banks;
2101 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2104 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2105 * reporting is disabled
2107 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2108 vcpu->arch.mcg_ctl != ~(u64)0)
2110 banks += 4 * mce->bank;
2112 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2113 * reporting is disabled for the bank
2115 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2117 if (mce->status & MCI_STATUS_UC) {
2118 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2119 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2120 printk(KERN_DEBUG "kvm: set_mce: "
2121 "injects mce exception while "
2122 "previous one is in progress!\n");
2123 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2126 if (banks[1] & MCI_STATUS_VAL)
2127 mce->status |= MCI_STATUS_OVER;
2128 banks[2] = mce->addr;
2129 banks[3] = mce->misc;
2130 vcpu->arch.mcg_status = mce->mcg_status;
2131 banks[1] = mce->status;
2132 kvm_queue_exception(vcpu, MC_VECTOR);
2133 } else if (!(banks[1] & MCI_STATUS_VAL)
2134 || !(banks[1] & MCI_STATUS_UC)) {
2135 if (banks[1] & MCI_STATUS_VAL)
2136 mce->status |= MCI_STATUS_OVER;
2137 banks[2] = mce->addr;
2138 banks[3] = mce->misc;
2139 banks[1] = mce->status;
2141 banks[1] |= MCI_STATUS_OVER;
2145 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2146 struct kvm_vcpu_events *events)
2150 events->exception.injected =
2151 vcpu->arch.exception.pending &&
2152 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2153 events->exception.nr = vcpu->arch.exception.nr;
2154 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2155 events->exception.error_code = vcpu->arch.exception.error_code;
2157 events->interrupt.injected =
2158 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2159 events->interrupt.nr = vcpu->arch.interrupt.nr;
2160 events->interrupt.soft = 0;
2161 events->interrupt.shadow =
2162 kvm_x86_ops->get_interrupt_shadow(vcpu,
2163 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2165 events->nmi.injected = vcpu->arch.nmi_injected;
2166 events->nmi.pending = vcpu->arch.nmi_pending;
2167 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2169 events->sipi_vector = vcpu->arch.sipi_vector;
2171 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2172 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2173 | KVM_VCPUEVENT_VALID_SHADOW);
2178 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2179 struct kvm_vcpu_events *events)
2181 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2182 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2183 | KVM_VCPUEVENT_VALID_SHADOW))
2188 vcpu->arch.exception.pending = events->exception.injected;
2189 vcpu->arch.exception.nr = events->exception.nr;
2190 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2191 vcpu->arch.exception.error_code = events->exception.error_code;
2193 vcpu->arch.interrupt.pending = events->interrupt.injected;
2194 vcpu->arch.interrupt.nr = events->interrupt.nr;
2195 vcpu->arch.interrupt.soft = events->interrupt.soft;
2196 if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2197 kvm_pic_clear_isr_ack(vcpu->kvm);
2198 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2199 kvm_x86_ops->set_interrupt_shadow(vcpu,
2200 events->interrupt.shadow);
2202 vcpu->arch.nmi_injected = events->nmi.injected;
2203 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2204 vcpu->arch.nmi_pending = events->nmi.pending;
2205 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2207 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2208 vcpu->arch.sipi_vector = events->sipi_vector;
2215 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2216 struct kvm_debugregs *dbgregs)
2220 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2221 dbgregs->dr6 = vcpu->arch.dr6;
2222 dbgregs->dr7 = vcpu->arch.dr7;
2228 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2229 struct kvm_debugregs *dbgregs)
2236 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2237 vcpu->arch.dr6 = dbgregs->dr6;
2238 vcpu->arch.dr7 = dbgregs->dr7;
2245 long kvm_arch_vcpu_ioctl(struct file *filp,
2246 unsigned int ioctl, unsigned long arg)
2248 struct kvm_vcpu *vcpu = filp->private_data;
2249 void __user *argp = (void __user *)arg;
2251 struct kvm_lapic_state *lapic = NULL;
2254 case KVM_GET_LAPIC: {
2256 if (!vcpu->arch.apic)
2258 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2263 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2267 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2272 case KVM_SET_LAPIC: {
2274 if (!vcpu->arch.apic)
2276 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2281 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2283 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2289 case KVM_INTERRUPT: {
2290 struct kvm_interrupt irq;
2293 if (copy_from_user(&irq, argp, sizeof irq))
2295 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2302 r = kvm_vcpu_ioctl_nmi(vcpu);
2308 case KVM_SET_CPUID: {
2309 struct kvm_cpuid __user *cpuid_arg = argp;
2310 struct kvm_cpuid cpuid;
2313 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2315 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2320 case KVM_SET_CPUID2: {
2321 struct kvm_cpuid2 __user *cpuid_arg = argp;
2322 struct kvm_cpuid2 cpuid;
2325 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2327 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2328 cpuid_arg->entries);
2333 case KVM_GET_CPUID2: {
2334 struct kvm_cpuid2 __user *cpuid_arg = argp;
2335 struct kvm_cpuid2 cpuid;
2338 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2340 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2341 cpuid_arg->entries);
2345 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2351 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2354 r = msr_io(vcpu, argp, do_set_msr, 0);
2356 case KVM_TPR_ACCESS_REPORTING: {
2357 struct kvm_tpr_access_ctl tac;
2360 if (copy_from_user(&tac, argp, sizeof tac))
2362 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2366 if (copy_to_user(argp, &tac, sizeof tac))
2371 case KVM_SET_VAPIC_ADDR: {
2372 struct kvm_vapic_addr va;
2375 if (!irqchip_in_kernel(vcpu->kvm))
2378 if (copy_from_user(&va, argp, sizeof va))
2381 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2384 case KVM_X86_SETUP_MCE: {
2388 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2390 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2393 case KVM_X86_SET_MCE: {
2394 struct kvm_x86_mce mce;
2397 if (copy_from_user(&mce, argp, sizeof mce))
2399 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2402 case KVM_GET_VCPU_EVENTS: {
2403 struct kvm_vcpu_events events;
2405 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2408 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2413 case KVM_SET_VCPU_EVENTS: {
2414 struct kvm_vcpu_events events;
2417 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2420 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2423 case KVM_GET_DEBUGREGS: {
2424 struct kvm_debugregs dbgregs;
2426 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2429 if (copy_to_user(argp, &dbgregs,
2430 sizeof(struct kvm_debugregs)))
2435 case KVM_SET_DEBUGREGS: {
2436 struct kvm_debugregs dbgregs;
2439 if (copy_from_user(&dbgregs, argp,
2440 sizeof(struct kvm_debugregs)))
2443 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2454 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2458 if (addr > (unsigned int)(-3 * PAGE_SIZE))
2460 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2464 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2467 kvm->arch.ept_identity_map_addr = ident_addr;
2471 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2472 u32 kvm_nr_mmu_pages)
2474 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2477 mutex_lock(&kvm->slots_lock);
2478 spin_lock(&kvm->mmu_lock);
2480 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2481 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2483 spin_unlock(&kvm->mmu_lock);
2484 mutex_unlock(&kvm->slots_lock);
2488 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2490 return kvm->arch.n_alloc_mmu_pages;
2493 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2496 struct kvm_mem_alias *alias;
2497 struct kvm_mem_aliases *aliases;
2499 aliases = rcu_dereference(kvm->arch.aliases);
2501 for (i = 0; i < aliases->naliases; ++i) {
2502 alias = &aliases->aliases[i];
2503 if (alias->flags & KVM_ALIAS_INVALID)
2505 if (gfn >= alias->base_gfn
2506 && gfn < alias->base_gfn + alias->npages)
2507 return alias->target_gfn + gfn - alias->base_gfn;
2512 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2515 struct kvm_mem_alias *alias;
2516 struct kvm_mem_aliases *aliases;
2518 aliases = rcu_dereference(kvm->arch.aliases);
2520 for (i = 0; i < aliases->naliases; ++i) {
2521 alias = &aliases->aliases[i];
2522 if (gfn >= alias->base_gfn
2523 && gfn < alias->base_gfn + alias->npages)
2524 return alias->target_gfn + gfn - alias->base_gfn;
2530 * Set a new alias region. Aliases map a portion of physical memory into
2531 * another portion. This is useful for memory windows, for example the PC
2534 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2535 struct kvm_memory_alias *alias)
2538 struct kvm_mem_alias *p;
2539 struct kvm_mem_aliases *aliases, *old_aliases;
2542 /* General sanity checks */
2543 if (alias->memory_size & (PAGE_SIZE - 1))
2545 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2547 if (alias->slot >= KVM_ALIAS_SLOTS)
2549 if (alias->guest_phys_addr + alias->memory_size
2550 < alias->guest_phys_addr)
2552 if (alias->target_phys_addr + alias->memory_size
2553 < alias->target_phys_addr)
2557 aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2561 mutex_lock(&kvm->slots_lock);
2563 /* invalidate any gfn reference in case of deletion/shrinking */
2564 memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2565 aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2566 old_aliases = kvm->arch.aliases;
2567 rcu_assign_pointer(kvm->arch.aliases, aliases);
2568 synchronize_srcu_expedited(&kvm->srcu);
2569 kvm_mmu_zap_all(kvm);
2573 aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2577 memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2579 p = &aliases->aliases[alias->slot];
2580 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2581 p->npages = alias->memory_size >> PAGE_SHIFT;
2582 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2583 p->flags &= ~(KVM_ALIAS_INVALID);
2585 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2586 if (aliases->aliases[n - 1].npages)
2588 aliases->naliases = n;
2590 old_aliases = kvm->arch.aliases;
2591 rcu_assign_pointer(kvm->arch.aliases, aliases);
2592 synchronize_srcu_expedited(&kvm->srcu);
2597 mutex_unlock(&kvm->slots_lock);
2602 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2607 switch (chip->chip_id) {
2608 case KVM_IRQCHIP_PIC_MASTER:
2609 memcpy(&chip->chip.pic,
2610 &pic_irqchip(kvm)->pics[0],
2611 sizeof(struct kvm_pic_state));
2613 case KVM_IRQCHIP_PIC_SLAVE:
2614 memcpy(&chip->chip.pic,
2615 &pic_irqchip(kvm)->pics[1],
2616 sizeof(struct kvm_pic_state));
2618 case KVM_IRQCHIP_IOAPIC:
2619 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2628 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2633 switch (chip->chip_id) {
2634 case KVM_IRQCHIP_PIC_MASTER:
2635 raw_spin_lock(&pic_irqchip(kvm)->lock);
2636 memcpy(&pic_irqchip(kvm)->pics[0],
2638 sizeof(struct kvm_pic_state));
2639 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2641 case KVM_IRQCHIP_PIC_SLAVE:
2642 raw_spin_lock(&pic_irqchip(kvm)->lock);
2643 memcpy(&pic_irqchip(kvm)->pics[1],
2645 sizeof(struct kvm_pic_state));
2646 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2648 case KVM_IRQCHIP_IOAPIC:
2649 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2655 kvm_pic_update_irq(pic_irqchip(kvm));
2659 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2663 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2664 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2665 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2669 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2673 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2674 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2675 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2676 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2680 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2684 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2685 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2686 sizeof(ps->channels));
2687 ps->flags = kvm->arch.vpit->pit_state.flags;
2688 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2692 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2694 int r = 0, start = 0;
2695 u32 prev_legacy, cur_legacy;
2696 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2697 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2698 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2699 if (!prev_legacy && cur_legacy)
2701 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2702 sizeof(kvm->arch.vpit->pit_state.channels));
2703 kvm->arch.vpit->pit_state.flags = ps->flags;
2704 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2705 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2709 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2710 struct kvm_reinject_control *control)
2712 if (!kvm->arch.vpit)
2714 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2715 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2716 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2721 * Get (and clear) the dirty memory log for a memory slot.
2723 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2724 struct kvm_dirty_log *log)
2727 struct kvm_memory_slot *memslot;
2729 unsigned long is_dirty = 0;
2730 unsigned long *dirty_bitmap = NULL;
2732 mutex_lock(&kvm->slots_lock);
2735 if (log->slot >= KVM_MEMORY_SLOTS)
2738 memslot = &kvm->memslots->memslots[log->slot];
2740 if (!memslot->dirty_bitmap)
2743 n = kvm_dirty_bitmap_bytes(memslot);
2746 dirty_bitmap = vmalloc(n);
2749 memset(dirty_bitmap, 0, n);
2751 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2752 is_dirty = memslot->dirty_bitmap[i];
2754 /* If nothing is dirty, don't bother messing with page tables. */
2756 struct kvm_memslots *slots, *old_slots;
2758 spin_lock(&kvm->mmu_lock);
2759 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2760 spin_unlock(&kvm->mmu_lock);
2762 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2766 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2767 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2769 old_slots = kvm->memslots;
2770 rcu_assign_pointer(kvm->memslots, slots);
2771 synchronize_srcu_expedited(&kvm->srcu);
2772 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2777 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2780 vfree(dirty_bitmap);
2782 mutex_unlock(&kvm->slots_lock);
2786 long kvm_arch_vm_ioctl(struct file *filp,
2787 unsigned int ioctl, unsigned long arg)
2789 struct kvm *kvm = filp->private_data;
2790 void __user *argp = (void __user *)arg;
2793 * This union makes it completely explicit to gcc-3.x
2794 * that these two variables' stack usage should be
2795 * combined, not added together.
2798 struct kvm_pit_state ps;
2799 struct kvm_pit_state2 ps2;
2800 struct kvm_memory_alias alias;
2801 struct kvm_pit_config pit_config;
2805 case KVM_SET_TSS_ADDR:
2806 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2810 case KVM_SET_IDENTITY_MAP_ADDR: {
2814 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2816 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2821 case KVM_SET_MEMORY_REGION: {
2822 struct kvm_memory_region kvm_mem;
2823 struct kvm_userspace_memory_region kvm_userspace_mem;
2826 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2828 kvm_userspace_mem.slot = kvm_mem.slot;
2829 kvm_userspace_mem.flags = kvm_mem.flags;
2830 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2831 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2832 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2837 case KVM_SET_NR_MMU_PAGES:
2838 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2842 case KVM_GET_NR_MMU_PAGES:
2843 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2845 case KVM_SET_MEMORY_ALIAS:
2847 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2849 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2853 case KVM_CREATE_IRQCHIP: {
2854 struct kvm_pic *vpic;
2856 mutex_lock(&kvm->lock);
2859 goto create_irqchip_unlock;
2861 vpic = kvm_create_pic(kvm);
2863 r = kvm_ioapic_init(kvm);
2865 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
2868 goto create_irqchip_unlock;
2871 goto create_irqchip_unlock;
2873 kvm->arch.vpic = vpic;
2875 r = kvm_setup_default_irq_routing(kvm);
2877 mutex_lock(&kvm->irq_lock);
2878 kvm_ioapic_destroy(kvm);
2879 kvm_destroy_pic(kvm);
2880 mutex_unlock(&kvm->irq_lock);
2882 create_irqchip_unlock:
2883 mutex_unlock(&kvm->lock);
2886 case KVM_CREATE_PIT:
2887 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2889 case KVM_CREATE_PIT2:
2891 if (copy_from_user(&u.pit_config, argp,
2892 sizeof(struct kvm_pit_config)))
2895 mutex_lock(&kvm->slots_lock);
2898 goto create_pit_unlock;
2900 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2904 mutex_unlock(&kvm->slots_lock);
2906 case KVM_IRQ_LINE_STATUS:
2907 case KVM_IRQ_LINE: {
2908 struct kvm_irq_level irq_event;
2911 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2914 if (irqchip_in_kernel(kvm)) {
2916 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2917 irq_event.irq, irq_event.level);
2918 if (ioctl == KVM_IRQ_LINE_STATUS) {
2920 irq_event.status = status;
2921 if (copy_to_user(argp, &irq_event,
2929 case KVM_GET_IRQCHIP: {
2930 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2931 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2937 if (copy_from_user(chip, argp, sizeof *chip))
2938 goto get_irqchip_out;
2940 if (!irqchip_in_kernel(kvm))
2941 goto get_irqchip_out;
2942 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2944 goto get_irqchip_out;
2946 if (copy_to_user(argp, chip, sizeof *chip))
2947 goto get_irqchip_out;
2955 case KVM_SET_IRQCHIP: {
2956 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2957 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2963 if (copy_from_user(chip, argp, sizeof *chip))
2964 goto set_irqchip_out;
2966 if (!irqchip_in_kernel(kvm))
2967 goto set_irqchip_out;
2968 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2970 goto set_irqchip_out;
2980 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2983 if (!kvm->arch.vpit)
2985 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2989 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2996 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2999 if (!kvm->arch.vpit)
3001 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3007 case KVM_GET_PIT2: {
3009 if (!kvm->arch.vpit)
3011 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3015 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3020 case KVM_SET_PIT2: {
3022 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3025 if (!kvm->arch.vpit)
3027 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3033 case KVM_REINJECT_CONTROL: {
3034 struct kvm_reinject_control control;
3036 if (copy_from_user(&control, argp, sizeof(control)))
3038 r = kvm_vm_ioctl_reinject(kvm, &control);
3044 case KVM_XEN_HVM_CONFIG: {
3046 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3047 sizeof(struct kvm_xen_hvm_config)))
3050 if (kvm->arch.xen_hvm_config.flags)
3055 case KVM_SET_CLOCK: {
3056 struct timespec now;
3057 struct kvm_clock_data user_ns;
3062 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3071 now_ns = timespec_to_ns(&now);
3072 delta = user_ns.clock - now_ns;
3073 kvm->arch.kvmclock_offset = delta;
3076 case KVM_GET_CLOCK: {
3077 struct timespec now;
3078 struct kvm_clock_data user_ns;
3082 now_ns = timespec_to_ns(&now);
3083 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3087 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3100 static void kvm_init_msr_list(void)
3105 /* skip the first msrs in the list. KVM-specific */
3106 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3107 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3110 msrs_to_save[j] = msrs_to_save[i];
3113 num_msrs_to_save = j;
3116 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3119 if (vcpu->arch.apic &&
3120 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3123 return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3126 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3128 if (vcpu->arch.apic &&
3129 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3132 return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3135 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3136 struct kvm_segment *var, int seg)
3138 kvm_x86_ops->set_segment(vcpu, var, seg);
3141 void kvm_get_segment(struct kvm_vcpu *vcpu,
3142 struct kvm_segment *var, int seg)
3144 kvm_x86_ops->get_segment(vcpu, var, seg);
3147 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3149 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3150 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3153 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3155 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3156 access |= PFERR_FETCH_MASK;
3157 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3160 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3162 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3163 access |= PFERR_WRITE_MASK;
3164 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3167 /* uses this to access any guest's mapped memory without checking CPL */
3168 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3170 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, 0, error);
3173 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3174 struct kvm_vcpu *vcpu, u32 access,
3178 int r = X86EMUL_CONTINUE;
3181 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, access, error);
3182 unsigned offset = addr & (PAGE_SIZE-1);
3183 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3186 if (gpa == UNMAPPED_GVA) {
3187 r = X86EMUL_PROPAGATE_FAULT;
3190 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3192 r = X86EMUL_UNHANDLEABLE;
3204 /* used for instruction fetching */
3205 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3206 struct kvm_vcpu *vcpu, u32 *error)
3208 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3209 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3210 access | PFERR_FETCH_MASK, error);
3213 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3214 struct kvm_vcpu *vcpu, u32 *error)
3216 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3217 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3221 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3222 struct kvm_vcpu *vcpu, u32 *error)
3224 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3227 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3229 struct kvm_vcpu *vcpu,
3233 int r = X86EMUL_CONTINUE;
3236 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr,
3237 PFERR_WRITE_MASK, error);
3238 unsigned offset = addr & (PAGE_SIZE-1);
3239 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3242 if (gpa == UNMAPPED_GVA) {
3243 r = X86EMUL_PROPAGATE_FAULT;
3246 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3248 r = X86EMUL_UNHANDLEABLE;
3260 static int emulator_read_emulated(unsigned long addr,
3263 struct kvm_vcpu *vcpu)
3268 if (vcpu->mmio_read_completed) {
3269 memcpy(val, vcpu->mmio_data, bytes);
3270 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3271 vcpu->mmio_phys_addr, *(u64 *)val);
3272 vcpu->mmio_read_completed = 0;
3273 return X86EMUL_CONTINUE;
3276 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, &error_code);
3278 if (gpa == UNMAPPED_GVA) {
3279 kvm_inject_page_fault(vcpu, addr, error_code);
3280 return X86EMUL_PROPAGATE_FAULT;
3283 /* For APIC access vmexit */
3284 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3287 if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3288 == X86EMUL_CONTINUE)
3289 return X86EMUL_CONTINUE;
3293 * Is this MMIO handled locally?
3295 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3296 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3297 return X86EMUL_CONTINUE;
3300 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3302 vcpu->mmio_needed = 1;
3303 vcpu->mmio_phys_addr = gpa;
3304 vcpu->mmio_size = bytes;
3305 vcpu->mmio_is_write = 0;
3307 return X86EMUL_UNHANDLEABLE;
3310 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3311 const void *val, int bytes)
3315 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3318 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3322 static int emulator_write_emulated_onepage(unsigned long addr,
3325 struct kvm_vcpu *vcpu)
3330 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, &error_code);
3332 if (gpa == UNMAPPED_GVA) {
3333 kvm_inject_page_fault(vcpu, addr, error_code);
3334 return X86EMUL_PROPAGATE_FAULT;
3337 /* For APIC access vmexit */
3338 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3341 if (emulator_write_phys(vcpu, gpa, val, bytes))
3342 return X86EMUL_CONTINUE;
3345 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3347 * Is this MMIO handled locally?
3349 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3350 return X86EMUL_CONTINUE;
3352 vcpu->mmio_needed = 1;
3353 vcpu->mmio_phys_addr = gpa;
3354 vcpu->mmio_size = bytes;
3355 vcpu->mmio_is_write = 1;
3356 memcpy(vcpu->mmio_data, val, bytes);
3358 return X86EMUL_CONTINUE;
3361 int emulator_write_emulated(unsigned long addr,
3364 struct kvm_vcpu *vcpu)
3366 /* Crossing a page boundary? */
3367 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3370 now = -addr & ~PAGE_MASK;
3371 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3372 if (rc != X86EMUL_CONTINUE)
3378 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3380 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3382 #define CMPXCHG_TYPE(t, ptr, old, new) \
3383 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3385 #ifdef CONFIG_X86_64
3386 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3388 # define CMPXCHG64(ptr, old, new) \
3389 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3392 static int emulator_cmpxchg_emulated(unsigned long addr,
3396 struct kvm_vcpu *vcpu)
3403 /* guests cmpxchg8b have to be emulated atomically */
3404 if (bytes > 8 || (bytes & (bytes - 1)))
3407 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3409 if (gpa == UNMAPPED_GVA ||
3410 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3413 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3416 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3418 kaddr = kmap_atomic(page, KM_USER0);
3419 kaddr += offset_in_page(gpa);
3422 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3425 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3428 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3431 exchanged = CMPXCHG64(kaddr, old, new);
3436 kunmap_atomic(kaddr, KM_USER0);
3437 kvm_release_page_dirty(page);
3440 return X86EMUL_CMPXCHG_FAILED;
3442 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3444 return X86EMUL_CONTINUE;
3447 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3449 return emulator_write_emulated(addr, new, bytes, vcpu);
3452 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3454 /* TODO: String I/O for in kernel device */
3457 if (vcpu->arch.pio.in)
3458 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3459 vcpu->arch.pio.size, pd);
3461 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3462 vcpu->arch.pio.port, vcpu->arch.pio.size,
3468 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
3469 unsigned int count, struct kvm_vcpu *vcpu)
3471 if (vcpu->arch.pio.count)
3474 trace_kvm_pio(1, port, size, 1);
3476 vcpu->arch.pio.port = port;
3477 vcpu->arch.pio.in = 1;
3478 vcpu->arch.pio.count = count;
3479 vcpu->arch.pio.size = size;
3481 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3483 memcpy(val, vcpu->arch.pio_data, size * count);
3484 vcpu->arch.pio.count = 0;
3488 vcpu->run->exit_reason = KVM_EXIT_IO;
3489 vcpu->run->io.direction = KVM_EXIT_IO_IN;
3490 vcpu->run->io.size = size;
3491 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3492 vcpu->run->io.count = count;
3493 vcpu->run->io.port = port;
3498 static int emulator_pio_out_emulated(int size, unsigned short port,
3499 const void *val, unsigned int count,
3500 struct kvm_vcpu *vcpu)
3502 trace_kvm_pio(0, port, size, 1);
3504 vcpu->arch.pio.port = port;
3505 vcpu->arch.pio.in = 0;
3506 vcpu->arch.pio.count = count;
3507 vcpu->arch.pio.size = size;
3509 memcpy(vcpu->arch.pio_data, val, size * count);
3511 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3512 vcpu->arch.pio.count = 0;
3516 vcpu->run->exit_reason = KVM_EXIT_IO;
3517 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
3518 vcpu->run->io.size = size;
3519 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3520 vcpu->run->io.count = count;
3521 vcpu->run->io.port = port;
3526 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3528 return kvm_x86_ops->get_segment_base(vcpu, seg);
3531 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3533 kvm_mmu_invlpg(vcpu, address);
3534 return X86EMUL_CONTINUE;
3537 int emulate_clts(struct kvm_vcpu *vcpu)
3539 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3540 kvm_x86_ops->fpu_activate(vcpu);
3541 return X86EMUL_CONTINUE;
3544 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3546 return kvm_get_dr(ctxt->vcpu, dr, dest);
3549 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3551 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3553 return kvm_set_dr(ctxt->vcpu, dr, value & mask);
3556 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3559 unsigned long rip = kvm_rip_read(vcpu);
3560 unsigned long rip_linear;
3562 if (!printk_ratelimit())
3565 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3567 kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu, NULL);
3569 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3570 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3572 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3574 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3576 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3579 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
3581 unsigned long value;
3585 value = kvm_read_cr0(vcpu);
3588 value = vcpu->arch.cr2;
3591 value = vcpu->arch.cr3;
3594 value = kvm_read_cr4(vcpu);
3597 value = kvm_get_cr8(vcpu);
3600 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3607 static void emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
3611 kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
3614 vcpu->arch.cr2 = val;
3617 kvm_set_cr3(vcpu, val);
3620 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3623 kvm_set_cr8(vcpu, val & 0xfUL);
3626 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3630 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
3632 return kvm_x86_ops->get_cpl(vcpu);
3635 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
3637 kvm_x86_ops->get_gdt(vcpu, dt);
3640 static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
3641 struct kvm_vcpu *vcpu)
3643 struct kvm_segment var;
3645 kvm_get_segment(vcpu, &var, seg);
3652 set_desc_limit(desc, var.limit);
3653 set_desc_base(desc, (unsigned long)var.base);
3654 desc->type = var.type;
3656 desc->dpl = var.dpl;
3657 desc->p = var.present;
3658 desc->avl = var.avl;
3666 static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg,
3667 struct kvm_vcpu *vcpu)
3669 struct kvm_segment var;
3671 /* needed to preserve selector */
3672 kvm_get_segment(vcpu, &var, seg);
3674 var.base = get_desc_base(desc);
3675 var.limit = get_desc_limit(desc);
3677 var.limit = (var.limit << 12) | 0xfff;
3678 var.type = desc->type;
3679 var.present = desc->p;
3680 var.dpl = desc->dpl;
3685 var.avl = desc->avl;
3686 var.present = desc->p;
3687 var.unusable = !var.present;
3690 kvm_set_segment(vcpu, &var, seg);
3694 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
3696 struct kvm_segment kvm_seg;
3698 kvm_get_segment(vcpu, &kvm_seg, seg);
3699 return kvm_seg.selector;
3702 static void emulator_set_segment_selector(u16 sel, int seg,
3703 struct kvm_vcpu *vcpu)
3705 struct kvm_segment kvm_seg;
3707 kvm_get_segment(vcpu, &kvm_seg, seg);
3708 kvm_seg.selector = sel;
3709 kvm_set_segment(vcpu, &kvm_seg, seg);
3712 static void emulator_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
3714 kvm_x86_ops->set_rflags(vcpu, rflags);
3717 static struct x86_emulate_ops emulate_ops = {
3718 .read_std = kvm_read_guest_virt_system,
3719 .write_std = kvm_write_guest_virt_system,
3720 .fetch = kvm_fetch_guest_virt,
3721 .read_emulated = emulator_read_emulated,
3722 .write_emulated = emulator_write_emulated,
3723 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3724 .pio_in_emulated = emulator_pio_in_emulated,
3725 .pio_out_emulated = emulator_pio_out_emulated,
3726 .get_cached_descriptor = emulator_get_cached_descriptor,
3727 .set_cached_descriptor = emulator_set_cached_descriptor,
3728 .get_segment_selector = emulator_get_segment_selector,
3729 .set_segment_selector = emulator_set_segment_selector,
3730 .get_gdt = emulator_get_gdt,
3731 .get_cr = emulator_get_cr,
3732 .set_cr = emulator_set_cr,
3733 .cpl = emulator_get_cpl,
3734 .set_rflags = emulator_set_rflags,
3737 static void cache_all_regs(struct kvm_vcpu *vcpu)
3739 kvm_register_read(vcpu, VCPU_REGS_RAX);
3740 kvm_register_read(vcpu, VCPU_REGS_RSP);
3741 kvm_register_read(vcpu, VCPU_REGS_RIP);
3742 vcpu->arch.regs_dirty = ~0;
3745 int emulate_instruction(struct kvm_vcpu *vcpu,
3751 struct decode_cache *c;
3752 struct kvm_run *run = vcpu->run;
3754 kvm_clear_exception_queue(vcpu);
3755 vcpu->arch.mmio_fault_cr2 = cr2;
3757 * TODO: fix emulate.c to use guest_read/write_register
3758 * instead of direct ->regs accesses, can save hundred cycles
3759 * on Intel for instructions that don't read/change RSP, for
3762 cache_all_regs(vcpu);
3764 vcpu->mmio_is_write = 0;
3766 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3768 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3770 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3771 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
3772 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
3773 vcpu->arch.emulate_ctxt.mode =
3774 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
3775 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3776 ? X86EMUL_MODE_VM86 : cs_l
3777 ? X86EMUL_MODE_PROT64 : cs_db
3778 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3780 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3781 trace_kvm_emulate_insn_start(vcpu);
3783 /* Only allow emulation of specific instructions on #UD
3784 * (namely VMMCALL, sysenter, sysexit, syscall)*/
3785 c = &vcpu->arch.emulate_ctxt.decode;
3786 if (emulation_type & EMULTYPE_TRAP_UD) {
3788 return EMULATE_FAIL;
3790 case 0x01: /* VMMCALL */
3791 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3792 return EMULATE_FAIL;
3794 case 0x34: /* sysenter */
3795 case 0x35: /* sysexit */
3796 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3797 return EMULATE_FAIL;
3799 case 0x05: /* syscall */
3800 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3801 return EMULATE_FAIL;
3804 return EMULATE_FAIL;
3807 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3808 return EMULATE_FAIL;
3811 ++vcpu->stat.insn_emulation;
3813 ++vcpu->stat.insn_emulation_fail;
3814 trace_kvm_emulate_insn_failed(vcpu);
3815 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3816 return EMULATE_DONE;
3817 return EMULATE_FAIL;
3821 if (emulation_type & EMULTYPE_SKIP) {
3822 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3823 return EMULATE_DONE;
3827 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3828 shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3831 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3833 if (vcpu->arch.pio.count) {
3834 if (!vcpu->arch.pio.in)
3835 vcpu->arch.pio.count = 0;
3836 return EMULATE_DO_MMIO;
3839 if (r || vcpu->mmio_is_write) {
3840 run->exit_reason = KVM_EXIT_MMIO;
3841 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3842 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3843 run->mmio.len = vcpu->mmio_size;
3844 run->mmio.is_write = vcpu->mmio_is_write;
3848 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3850 if (!vcpu->mmio_needed) {
3851 ++vcpu->stat.insn_emulation_fail;
3852 trace_kvm_emulate_insn_failed(vcpu);
3853 kvm_report_emulation_failure(vcpu, "mmio");
3854 return EMULATE_FAIL;
3856 return EMULATE_DO_MMIO;
3859 if (vcpu->mmio_is_write) {
3860 vcpu->mmio_needed = 0;
3861 return EMULATE_DO_MMIO;
3865 if (vcpu->arch.exception.pending)
3866 vcpu->arch.emulate_ctxt.restart = false;
3868 if (vcpu->arch.emulate_ctxt.restart)
3871 return EMULATE_DONE;
3873 EXPORT_SYMBOL_GPL(emulate_instruction);
3875 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
3877 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3878 int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
3879 /* do not return to emulator after return from userspace */
3880 vcpu->arch.pio.count = 0;
3883 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
3885 static void bounce_off(void *info)
3890 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3893 struct cpufreq_freqs *freq = data;
3895 struct kvm_vcpu *vcpu;
3896 int i, send_ipi = 0;
3898 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3900 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3902 per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3904 spin_lock(&kvm_lock);
3905 list_for_each_entry(kvm, &vm_list, vm_list) {
3906 kvm_for_each_vcpu(i, vcpu, kvm) {
3907 if (vcpu->cpu != freq->cpu)
3909 if (!kvm_request_guest_time_update(vcpu))
3911 if (vcpu->cpu != smp_processor_id())
3915 spin_unlock(&kvm_lock);
3917 if (freq->old < freq->new && send_ipi) {
3919 * We upscale the frequency. Must make the guest
3920 * doesn't see old kvmclock values while running with
3921 * the new frequency, otherwise we risk the guest sees
3922 * time go backwards.
3924 * In case we update the frequency for another cpu
3925 * (which might be in guest context) send an interrupt
3926 * to kick the cpu out of guest context. Next time
3927 * guest context is entered kvmclock will be updated,
3928 * so the guest will not see stale values.
3930 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3935 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3936 .notifier_call = kvmclock_cpufreq_notifier
3939 static void kvm_timer_init(void)
3943 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3944 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3945 CPUFREQ_TRANSITION_NOTIFIER);
3946 for_each_online_cpu(cpu) {
3947 unsigned long khz = cpufreq_get(cpu);
3950 per_cpu(cpu_tsc_khz, cpu) = khz;
3953 for_each_possible_cpu(cpu)
3954 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3958 int kvm_arch_init(void *opaque)
3961 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3964 printk(KERN_ERR "kvm: already loaded the other module\n");
3969 if (!ops->cpu_has_kvm_support()) {
3970 printk(KERN_ERR "kvm: no hardware support\n");
3974 if (ops->disabled_by_bios()) {
3975 printk(KERN_ERR "kvm: disabled by bios\n");
3980 r = kvm_mmu_module_init();
3984 kvm_init_msr_list();
3987 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3988 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3989 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3990 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4000 void kvm_arch_exit(void)
4002 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4003 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4004 CPUFREQ_TRANSITION_NOTIFIER);
4006 kvm_mmu_module_exit();
4009 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4011 ++vcpu->stat.halt_exits;
4012 if (irqchip_in_kernel(vcpu->kvm)) {
4013 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4016 vcpu->run->exit_reason = KVM_EXIT_HLT;
4020 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4022 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4025 if (is_long_mode(vcpu))
4028 return a0 | ((gpa_t)a1 << 32);
4031 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4033 u64 param, ingpa, outgpa, ret;
4034 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4035 bool fast, longmode;
4039 * hypercall generates UD from non zero cpl and real mode
4042 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4043 kvm_queue_exception(vcpu, UD_VECTOR);
4047 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4048 longmode = is_long_mode(vcpu) && cs_l == 1;
4051 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4052 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4053 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4054 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4055 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4056 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4058 #ifdef CONFIG_X86_64
4060 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4061 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4062 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4066 code = param & 0xffff;
4067 fast = (param >> 16) & 0x1;
4068 rep_cnt = (param >> 32) & 0xfff;
4069 rep_idx = (param >> 48) & 0xfff;
4071 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4074 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4075 kvm_vcpu_on_spin(vcpu);
4078 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4082 ret = res | (((u64)rep_done & 0xfff) << 32);
4084 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4086 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4087 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4093 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4095 unsigned long nr, a0, a1, a2, a3, ret;
4098 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4099 return kvm_hv_hypercall(vcpu);
4101 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4102 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4103 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4104 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4105 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4107 trace_kvm_hypercall(nr, a0, a1, a2, a3);
4109 if (!is_long_mode(vcpu)) {
4117 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4123 case KVM_HC_VAPIC_POLL_IRQ:
4127 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4134 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4135 ++vcpu->stat.hypercalls;
4138 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4140 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4142 char instruction[3];
4143 unsigned long rip = kvm_rip_read(vcpu);
4146 * Blow out the MMU to ensure that no other VCPU has an active mapping
4147 * to ensure that the updated hypercall appears atomically across all
4150 kvm_mmu_zap_all(vcpu->kvm);
4152 kvm_x86_ops->patch_hypercall(vcpu, instruction);
4154 return emulator_write_emulated(rip, instruction, 3, vcpu);
4157 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4159 struct desc_ptr dt = { limit, base };
4161 kvm_x86_ops->set_gdt(vcpu, &dt);
4164 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4166 struct desc_ptr dt = { limit, base };
4168 kvm_x86_ops->set_idt(vcpu, &dt);
4171 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4173 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4174 int j, nent = vcpu->arch.cpuid_nent;
4176 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4177 /* when no next entry is found, the current entry[i] is reselected */
4178 for (j = i + 1; ; j = (j + 1) % nent) {
4179 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4180 if (ej->function == e->function) {
4181 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4185 return 0; /* silence gcc, even though control never reaches here */
4188 /* find an entry with matching function, matching index (if needed), and that
4189 * should be read next (if it's stateful) */
4190 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4191 u32 function, u32 index)
4193 if (e->function != function)
4195 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4197 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4198 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4203 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4204 u32 function, u32 index)
4207 struct kvm_cpuid_entry2 *best = NULL;
4209 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4210 struct kvm_cpuid_entry2 *e;
4212 e = &vcpu->arch.cpuid_entries[i];
4213 if (is_matching_cpuid_entry(e, function, index)) {
4214 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4215 move_to_next_stateful_cpuid_entry(vcpu, i);
4220 * Both basic or both extended?
4222 if (((e->function ^ function) & 0x80000000) == 0)
4223 if (!best || e->function > best->function)
4228 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4230 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4232 struct kvm_cpuid_entry2 *best;
4234 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
4235 if (!best || best->eax < 0x80000008)
4237 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4239 return best->eax & 0xff;
4244 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4246 u32 function, index;
4247 struct kvm_cpuid_entry2 *best;
4249 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4250 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4251 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4252 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4253 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4254 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4255 best = kvm_find_cpuid_entry(vcpu, function, index);
4257 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4258 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4259 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4260 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4262 kvm_x86_ops->skip_emulated_instruction(vcpu);
4263 trace_kvm_cpuid(function,
4264 kvm_register_read(vcpu, VCPU_REGS_RAX),
4265 kvm_register_read(vcpu, VCPU_REGS_RBX),
4266 kvm_register_read(vcpu, VCPU_REGS_RCX),
4267 kvm_register_read(vcpu, VCPU_REGS_RDX));
4269 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4272 * Check if userspace requested an interrupt window, and that the
4273 * interrupt window is open.
4275 * No need to exit to userspace if we already have an interrupt queued.
4277 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4279 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4280 vcpu->run->request_interrupt_window &&
4281 kvm_arch_interrupt_allowed(vcpu));
4284 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4286 struct kvm_run *kvm_run = vcpu->run;
4288 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4289 kvm_run->cr8 = kvm_get_cr8(vcpu);
4290 kvm_run->apic_base = kvm_get_apic_base(vcpu);
4291 if (irqchip_in_kernel(vcpu->kvm))
4292 kvm_run->ready_for_interrupt_injection = 1;
4294 kvm_run->ready_for_interrupt_injection =
4295 kvm_arch_interrupt_allowed(vcpu) &&
4296 !kvm_cpu_has_interrupt(vcpu) &&
4297 !kvm_event_needs_reinjection(vcpu);
4300 static void vapic_enter(struct kvm_vcpu *vcpu)
4302 struct kvm_lapic *apic = vcpu->arch.apic;
4305 if (!apic || !apic->vapic_addr)
4308 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4310 vcpu->arch.apic->vapic_page = page;
4313 static void vapic_exit(struct kvm_vcpu *vcpu)
4315 struct kvm_lapic *apic = vcpu->arch.apic;
4318 if (!apic || !apic->vapic_addr)
4321 idx = srcu_read_lock(&vcpu->kvm->srcu);
4322 kvm_release_page_dirty(apic->vapic_page);
4323 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4324 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4327 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4331 if (!kvm_x86_ops->update_cr8_intercept)
4334 if (!vcpu->arch.apic)
4337 if (!vcpu->arch.apic->vapic_addr)
4338 max_irr = kvm_lapic_find_highest_irr(vcpu);
4345 tpr = kvm_lapic_get_cr8(vcpu);
4347 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4350 static void inject_pending_event(struct kvm_vcpu *vcpu)
4352 /* try to reinject previous events if any */
4353 if (vcpu->arch.exception.pending) {
4354 trace_kvm_inj_exception(vcpu->arch.exception.nr,
4355 vcpu->arch.exception.has_error_code,
4356 vcpu->arch.exception.error_code);
4357 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4358 vcpu->arch.exception.has_error_code,
4359 vcpu->arch.exception.error_code);
4363 if (vcpu->arch.nmi_injected) {
4364 kvm_x86_ops->set_nmi(vcpu);
4368 if (vcpu->arch.interrupt.pending) {
4369 kvm_x86_ops->set_irq(vcpu);
4373 /* try to inject new event if pending */
4374 if (vcpu->arch.nmi_pending) {
4375 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4376 vcpu->arch.nmi_pending = false;
4377 vcpu->arch.nmi_injected = true;
4378 kvm_x86_ops->set_nmi(vcpu);
4380 } else if (kvm_cpu_has_interrupt(vcpu)) {
4381 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4382 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4384 kvm_x86_ops->set_irq(vcpu);
4389 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4392 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4393 vcpu->run->request_interrupt_window;
4396 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4397 kvm_mmu_unload(vcpu);
4399 r = kvm_mmu_reload(vcpu);
4403 if (vcpu->requests) {
4404 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4405 __kvm_migrate_timers(vcpu);
4406 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4407 kvm_write_guest_time(vcpu);
4408 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4409 kvm_mmu_sync_roots(vcpu);
4410 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4411 kvm_x86_ops->tlb_flush(vcpu);
4412 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4414 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4418 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4419 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4423 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4424 vcpu->fpu_active = 0;
4425 kvm_x86_ops->fpu_deactivate(vcpu);
4431 kvm_x86_ops->prepare_guest_switch(vcpu);
4432 if (vcpu->fpu_active)
4433 kvm_load_guest_fpu(vcpu);
4435 local_irq_disable();
4437 clear_bit(KVM_REQ_KICK, &vcpu->requests);
4438 smp_mb__after_clear_bit();
4440 if (vcpu->requests || need_resched() || signal_pending(current)) {
4441 set_bit(KVM_REQ_KICK, &vcpu->requests);
4448 inject_pending_event(vcpu);
4450 /* enable NMI/IRQ window open exits if needed */
4451 if (vcpu->arch.nmi_pending)
4452 kvm_x86_ops->enable_nmi_window(vcpu);
4453 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4454 kvm_x86_ops->enable_irq_window(vcpu);
4456 if (kvm_lapic_enabled(vcpu)) {
4457 update_cr8_intercept(vcpu);
4458 kvm_lapic_sync_to_vapic(vcpu);
4461 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4465 if (unlikely(vcpu->arch.switch_db_regs)) {
4467 set_debugreg(vcpu->arch.eff_db[0], 0);
4468 set_debugreg(vcpu->arch.eff_db[1], 1);
4469 set_debugreg(vcpu->arch.eff_db[2], 2);
4470 set_debugreg(vcpu->arch.eff_db[3], 3);
4473 trace_kvm_entry(vcpu->vcpu_id);
4474 kvm_x86_ops->run(vcpu);
4477 * If the guest has used debug registers, at least dr7
4478 * will be disabled while returning to the host.
4479 * If we don't have active breakpoints in the host, we don't
4480 * care about the messed up debug address registers. But if
4481 * we have some of them active, restore the old state.
4483 if (hw_breakpoint_active())
4484 hw_breakpoint_restore();
4486 set_bit(KVM_REQ_KICK, &vcpu->requests);
4492 * We must have an instruction between local_irq_enable() and
4493 * kvm_guest_exit(), so the timer interrupt isn't delayed by
4494 * the interrupt shadow. The stat.exits increment will do nicely.
4495 * But we need to prevent reordering, hence this barrier():
4503 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4506 * Profile KVM exit RIPs:
4508 if (unlikely(prof_on == KVM_PROFILING)) {
4509 unsigned long rip = kvm_rip_read(vcpu);
4510 profile_hit(KVM_PROFILING, (void *)rip);
4514 kvm_lapic_sync_from_vapic(vcpu);
4516 r = kvm_x86_ops->handle_exit(vcpu);
4522 static int __vcpu_run(struct kvm_vcpu *vcpu)
4525 struct kvm *kvm = vcpu->kvm;
4527 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4528 pr_debug("vcpu %d received sipi with vector # %x\n",
4529 vcpu->vcpu_id, vcpu->arch.sipi_vector);
4530 kvm_lapic_reset(vcpu);
4531 r = kvm_arch_vcpu_reset(vcpu);
4534 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4537 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4542 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4543 r = vcpu_enter_guest(vcpu);
4545 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4546 kvm_vcpu_block(vcpu);
4547 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4548 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4550 switch(vcpu->arch.mp_state) {
4551 case KVM_MP_STATE_HALTED:
4552 vcpu->arch.mp_state =
4553 KVM_MP_STATE_RUNNABLE;
4554 case KVM_MP_STATE_RUNNABLE:
4556 case KVM_MP_STATE_SIPI_RECEIVED:
4567 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4568 if (kvm_cpu_has_pending_timer(vcpu))
4569 kvm_inject_pending_timer_irqs(vcpu);
4571 if (dm_request_for_irq_injection(vcpu)) {
4573 vcpu->run->exit_reason = KVM_EXIT_INTR;
4574 ++vcpu->stat.request_irq_exits;
4576 if (signal_pending(current)) {
4578 vcpu->run->exit_reason = KVM_EXIT_INTR;
4579 ++vcpu->stat.signal_exits;
4581 if (need_resched()) {
4582 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4584 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4588 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4589 post_kvm_run_save(vcpu);
4596 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4603 if (vcpu->sigset_active)
4604 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4606 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4607 kvm_vcpu_block(vcpu);
4608 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4613 /* re-sync apic's tpr */
4614 if (!irqchip_in_kernel(vcpu->kvm))
4615 kvm_set_cr8(vcpu, kvm_run->cr8);
4617 if (vcpu->arch.pio.count || vcpu->mmio_needed ||
4618 vcpu->arch.emulate_ctxt.restart) {
4619 if (vcpu->mmio_needed) {
4620 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4621 vcpu->mmio_read_completed = 1;
4622 vcpu->mmio_needed = 0;
4624 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4625 r = emulate_instruction(vcpu, 0, 0, EMULTYPE_NO_DECODE);
4626 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4627 if (r == EMULATE_DO_MMIO) {
4632 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4633 kvm_register_write(vcpu, VCPU_REGS_RAX,
4634 kvm_run->hypercall.ret);
4636 r = __vcpu_run(vcpu);
4639 if (vcpu->sigset_active)
4640 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4646 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4650 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4651 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4652 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4653 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4654 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4655 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4656 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4657 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4658 #ifdef CONFIG_X86_64
4659 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4660 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4661 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4662 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4663 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4664 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4665 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4666 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4669 regs->rip = kvm_rip_read(vcpu);
4670 regs->rflags = kvm_get_rflags(vcpu);
4677 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4681 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4682 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4683 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4684 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4685 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4686 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4687 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4688 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4689 #ifdef CONFIG_X86_64
4690 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4691 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4692 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4693 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4694 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4695 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4696 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4697 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4700 kvm_rip_write(vcpu, regs->rip);
4701 kvm_set_rflags(vcpu, regs->rflags);
4703 vcpu->arch.exception.pending = false;
4710 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4712 struct kvm_segment cs;
4714 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4718 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4720 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4721 struct kvm_sregs *sregs)
4727 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4728 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4729 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4730 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4731 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4732 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4734 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4735 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4737 kvm_x86_ops->get_idt(vcpu, &dt);
4738 sregs->idt.limit = dt.size;
4739 sregs->idt.base = dt.address;
4740 kvm_x86_ops->get_gdt(vcpu, &dt);
4741 sregs->gdt.limit = dt.size;
4742 sregs->gdt.base = dt.address;
4744 sregs->cr0 = kvm_read_cr0(vcpu);
4745 sregs->cr2 = vcpu->arch.cr2;
4746 sregs->cr3 = vcpu->arch.cr3;
4747 sregs->cr4 = kvm_read_cr4(vcpu);
4748 sregs->cr8 = kvm_get_cr8(vcpu);
4749 sregs->efer = vcpu->arch.efer;
4750 sregs->apic_base = kvm_get_apic_base(vcpu);
4752 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4754 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4755 set_bit(vcpu->arch.interrupt.nr,
4756 (unsigned long *)sregs->interrupt_bitmap);
4763 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4764 struct kvm_mp_state *mp_state)
4767 mp_state->mp_state = vcpu->arch.mp_state;
4772 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4773 struct kvm_mp_state *mp_state)
4776 vcpu->arch.mp_state = mp_state->mp_state;
4781 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
4782 bool has_error_code, u32 error_code)
4784 int cs_db, cs_l, ret;
4785 cache_all_regs(vcpu);
4787 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4789 vcpu->arch.emulate_ctxt.vcpu = vcpu;
4790 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
4791 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4792 vcpu->arch.emulate_ctxt.mode =
4793 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4794 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4795 ? X86EMUL_MODE_VM86 : cs_l
4796 ? X86EMUL_MODE_PROT64 : cs_db
4797 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4799 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt, &emulate_ops,
4800 tss_selector, reason, has_error_code,
4803 if (ret == X86EMUL_CONTINUE)
4804 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4806 return (ret != X86EMUL_CONTINUE);
4808 EXPORT_SYMBOL_GPL(kvm_task_switch);
4810 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4811 struct kvm_sregs *sregs)
4813 int mmu_reset_needed = 0;
4814 int pending_vec, max_bits;
4819 dt.size = sregs->idt.limit;
4820 dt.address = sregs->idt.base;
4821 kvm_x86_ops->set_idt(vcpu, &dt);
4822 dt.size = sregs->gdt.limit;
4823 dt.address = sregs->gdt.base;
4824 kvm_x86_ops->set_gdt(vcpu, &dt);
4826 vcpu->arch.cr2 = sregs->cr2;
4827 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4828 vcpu->arch.cr3 = sregs->cr3;
4830 kvm_set_cr8(vcpu, sregs->cr8);
4832 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
4833 kvm_x86_ops->set_efer(vcpu, sregs->efer);
4834 kvm_set_apic_base(vcpu, sregs->apic_base);
4836 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
4837 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4838 vcpu->arch.cr0 = sregs->cr0;
4840 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
4841 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4842 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
4843 load_pdptrs(vcpu, vcpu->arch.cr3);
4844 mmu_reset_needed = 1;
4847 if (mmu_reset_needed)
4848 kvm_mmu_reset_context(vcpu);
4850 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4851 pending_vec = find_first_bit(
4852 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4853 if (pending_vec < max_bits) {
4854 kvm_queue_interrupt(vcpu, pending_vec, false);
4855 pr_debug("Set back pending irq %d\n", pending_vec);
4856 if (irqchip_in_kernel(vcpu->kvm))
4857 kvm_pic_clear_isr_ack(vcpu->kvm);
4860 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4861 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4862 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4863 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4864 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4865 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4867 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4868 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4870 update_cr8_intercept(vcpu);
4872 /* Older userspace won't unhalt the vcpu on reset. */
4873 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4874 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4876 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4883 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4884 struct kvm_guest_debug *dbg)
4886 unsigned long rflags;
4891 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
4893 if (vcpu->arch.exception.pending)
4895 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4896 kvm_queue_exception(vcpu, DB_VECTOR);
4898 kvm_queue_exception(vcpu, BP_VECTOR);
4902 * Read rflags as long as potentially injected trace flags are still
4905 rflags = kvm_get_rflags(vcpu);
4907 vcpu->guest_debug = dbg->control;
4908 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
4909 vcpu->guest_debug = 0;
4911 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4912 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4913 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4914 vcpu->arch.switch_db_regs =
4915 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4917 for (i = 0; i < KVM_NR_DB_REGS; i++)
4918 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4919 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4922 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
4923 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
4924 get_segment_base(vcpu, VCPU_SREG_CS);
4927 * Trigger an rflags update that will inject or remove the trace
4930 kvm_set_rflags(vcpu, rflags);
4932 kvm_x86_ops->set_guest_debug(vcpu, dbg);
4943 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4944 * we have asm/x86/processor.h
4955 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4956 #ifdef CONFIG_X86_64
4957 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4959 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4964 * Translate a guest virtual address to a guest physical address.
4966 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4967 struct kvm_translation *tr)
4969 unsigned long vaddr = tr->linear_address;
4974 idx = srcu_read_lock(&vcpu->kvm->srcu);
4975 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
4976 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4977 tr->physical_address = gpa;
4978 tr->valid = gpa != UNMAPPED_GVA;
4986 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4988 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4992 memcpy(fpu->fpr, fxsave->st_space, 128);
4993 fpu->fcw = fxsave->cwd;
4994 fpu->fsw = fxsave->swd;
4995 fpu->ftwx = fxsave->twd;
4996 fpu->last_opcode = fxsave->fop;
4997 fpu->last_ip = fxsave->rip;
4998 fpu->last_dp = fxsave->rdp;
4999 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5006 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5008 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5012 memcpy(fxsave->st_space, fpu->fpr, 128);
5013 fxsave->cwd = fpu->fcw;
5014 fxsave->swd = fpu->fsw;
5015 fxsave->twd = fpu->ftwx;
5016 fxsave->fop = fpu->last_opcode;
5017 fxsave->rip = fpu->last_ip;
5018 fxsave->rdp = fpu->last_dp;
5019 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5026 void fx_init(struct kvm_vcpu *vcpu)
5028 unsigned after_mxcsr_mask;
5031 * Touch the fpu the first time in non atomic context as if
5032 * this is the first fpu instruction the exception handler
5033 * will fire before the instruction returns and it'll have to
5034 * allocate ram with GFP_KERNEL.
5037 kvm_fx_save(&vcpu->arch.host_fx_image);
5039 /* Initialize guest FPU by resetting ours and saving into guest's */
5041 kvm_fx_save(&vcpu->arch.host_fx_image);
5043 kvm_fx_save(&vcpu->arch.guest_fx_image);
5044 kvm_fx_restore(&vcpu->arch.host_fx_image);
5047 vcpu->arch.cr0 |= X86_CR0_ET;
5048 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5049 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5050 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5051 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5053 EXPORT_SYMBOL_GPL(fx_init);
5055 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5057 if (vcpu->guest_fpu_loaded)
5060 vcpu->guest_fpu_loaded = 1;
5061 kvm_fx_save(&vcpu->arch.host_fx_image);
5062 kvm_fx_restore(&vcpu->arch.guest_fx_image);
5066 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5068 if (!vcpu->guest_fpu_loaded)
5071 vcpu->guest_fpu_loaded = 0;
5072 kvm_fx_save(&vcpu->arch.guest_fx_image);
5073 kvm_fx_restore(&vcpu->arch.host_fx_image);
5074 ++vcpu->stat.fpu_reload;
5075 set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5079 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5081 if (vcpu->arch.time_page) {
5082 kvm_release_page_dirty(vcpu->arch.time_page);
5083 vcpu->arch.time_page = NULL;
5086 kvm_x86_ops->vcpu_free(vcpu);
5089 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5092 return kvm_x86_ops->vcpu_create(kvm, id);
5095 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5099 /* We do fxsave: this must be aligned. */
5100 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5102 vcpu->arch.mtrr_state.have_fixed = 1;
5104 r = kvm_arch_vcpu_reset(vcpu);
5106 r = kvm_mmu_setup(vcpu);
5113 kvm_x86_ops->vcpu_free(vcpu);
5117 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5120 kvm_mmu_unload(vcpu);
5123 kvm_x86_ops->vcpu_free(vcpu);
5126 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5128 vcpu->arch.nmi_pending = false;
5129 vcpu->arch.nmi_injected = false;
5131 vcpu->arch.switch_db_regs = 0;
5132 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5133 vcpu->arch.dr6 = DR6_FIXED_1;
5134 vcpu->arch.dr7 = DR7_FIXED_1;
5136 return kvm_x86_ops->vcpu_reset(vcpu);
5139 int kvm_arch_hardware_enable(void *garbage)
5142 * Since this may be called from a hotplug notifcation,
5143 * we can't get the CPU frequency directly.
5145 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5146 int cpu = raw_smp_processor_id();
5147 per_cpu(cpu_tsc_khz, cpu) = 0;
5150 kvm_shared_msr_cpu_online();
5152 return kvm_x86_ops->hardware_enable(garbage);
5155 void kvm_arch_hardware_disable(void *garbage)
5157 kvm_x86_ops->hardware_disable(garbage);
5158 drop_user_return_notifiers(garbage);
5161 int kvm_arch_hardware_setup(void)
5163 return kvm_x86_ops->hardware_setup();
5166 void kvm_arch_hardware_unsetup(void)
5168 kvm_x86_ops->hardware_unsetup();
5171 void kvm_arch_check_processor_compat(void *rtn)
5173 kvm_x86_ops->check_processor_compatibility(rtn);
5176 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5182 BUG_ON(vcpu->kvm == NULL);
5185 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5186 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5187 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5189 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5191 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5196 vcpu->arch.pio_data = page_address(page);
5198 r = kvm_mmu_create(vcpu);
5200 goto fail_free_pio_data;
5202 if (irqchip_in_kernel(kvm)) {
5203 r = kvm_create_lapic(vcpu);
5205 goto fail_mmu_destroy;
5208 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5210 if (!vcpu->arch.mce_banks) {
5212 goto fail_free_lapic;
5214 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5218 kvm_free_lapic(vcpu);
5220 kvm_mmu_destroy(vcpu);
5222 free_page((unsigned long)vcpu->arch.pio_data);
5227 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5231 kfree(vcpu->arch.mce_banks);
5232 kvm_free_lapic(vcpu);
5233 idx = srcu_read_lock(&vcpu->kvm->srcu);
5234 kvm_mmu_destroy(vcpu);
5235 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5236 free_page((unsigned long)vcpu->arch.pio_data);
5239 struct kvm *kvm_arch_create_vm(void)
5241 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5244 return ERR_PTR(-ENOMEM);
5246 kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5247 if (!kvm->arch.aliases) {
5249 return ERR_PTR(-ENOMEM);
5252 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5253 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5255 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5256 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5258 rdtscll(kvm->arch.vm_init_tsc);
5263 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5266 kvm_mmu_unload(vcpu);
5270 static void kvm_free_vcpus(struct kvm *kvm)
5273 struct kvm_vcpu *vcpu;
5276 * Unpin any mmu pages first.
5278 kvm_for_each_vcpu(i, vcpu, kvm)
5279 kvm_unload_vcpu_mmu(vcpu);
5280 kvm_for_each_vcpu(i, vcpu, kvm)
5281 kvm_arch_vcpu_free(vcpu);
5283 mutex_lock(&kvm->lock);
5284 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5285 kvm->vcpus[i] = NULL;
5287 atomic_set(&kvm->online_vcpus, 0);
5288 mutex_unlock(&kvm->lock);
5291 void kvm_arch_sync_events(struct kvm *kvm)
5293 kvm_free_all_assigned_devices(kvm);
5296 void kvm_arch_destroy_vm(struct kvm *kvm)
5298 kvm_iommu_unmap_guest(kvm);
5300 kfree(kvm->arch.vpic);
5301 kfree(kvm->arch.vioapic);
5302 kvm_free_vcpus(kvm);
5303 kvm_free_physmem(kvm);
5304 if (kvm->arch.apic_access_page)
5305 put_page(kvm->arch.apic_access_page);
5306 if (kvm->arch.ept_identity_pagetable)
5307 put_page(kvm->arch.ept_identity_pagetable);
5308 cleanup_srcu_struct(&kvm->srcu);
5309 kfree(kvm->arch.aliases);
5313 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5314 struct kvm_memory_slot *memslot,
5315 struct kvm_memory_slot old,
5316 struct kvm_userspace_memory_region *mem,
5319 int npages = memslot->npages;
5321 /*To keep backward compatibility with older userspace,
5322 *x86 needs to hanlde !user_alloc case.
5325 if (npages && !old.rmap) {
5326 unsigned long userspace_addr;
5328 down_write(¤t->mm->mmap_sem);
5329 userspace_addr = do_mmap(NULL, 0,
5331 PROT_READ | PROT_WRITE,
5332 MAP_PRIVATE | MAP_ANONYMOUS,
5334 up_write(¤t->mm->mmap_sem);
5336 if (IS_ERR((void *)userspace_addr))
5337 return PTR_ERR((void *)userspace_addr);
5339 memslot->userspace_addr = userspace_addr;
5347 void kvm_arch_commit_memory_region(struct kvm *kvm,
5348 struct kvm_userspace_memory_region *mem,
5349 struct kvm_memory_slot old,
5353 int npages = mem->memory_size >> PAGE_SHIFT;
5355 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5358 down_write(¤t->mm->mmap_sem);
5359 ret = do_munmap(current->mm, old.userspace_addr,
5360 old.npages * PAGE_SIZE);
5361 up_write(¤t->mm->mmap_sem);
5364 "kvm_vm_ioctl_set_memory_region: "
5365 "failed to munmap memory\n");
5368 spin_lock(&kvm->mmu_lock);
5369 if (!kvm->arch.n_requested_mmu_pages) {
5370 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5371 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5374 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5375 spin_unlock(&kvm->mmu_lock);
5378 void kvm_arch_flush_shadow(struct kvm *kvm)
5380 kvm_mmu_zap_all(kvm);
5381 kvm_reload_remote_mmus(kvm);
5384 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5386 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5387 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5388 || vcpu->arch.nmi_pending ||
5389 (kvm_arch_interrupt_allowed(vcpu) &&
5390 kvm_cpu_has_interrupt(vcpu));
5393 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5396 int cpu = vcpu->cpu;
5398 if (waitqueue_active(&vcpu->wq)) {
5399 wake_up_interruptible(&vcpu->wq);
5400 ++vcpu->stat.halt_wakeup;
5404 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5405 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5406 smp_send_reschedule(cpu);
5410 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5412 return kvm_x86_ops->interrupt_allowed(vcpu);
5415 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
5417 unsigned long current_rip = kvm_rip_read(vcpu) +
5418 get_segment_base(vcpu, VCPU_SREG_CS);
5420 return current_rip == linear_rip;
5422 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
5424 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5426 unsigned long rflags;
5428 rflags = kvm_x86_ops->get_rflags(vcpu);
5429 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5430 rflags &= ~X86_EFLAGS_TF;
5433 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5435 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5437 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5438 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
5439 rflags |= X86_EFLAGS_TF;
5440 kvm_x86_ops->set_rflags(vcpu, rflags);
5442 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5444 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5445 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5446 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5447 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5448 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5449 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5450 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5451 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5452 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5453 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5454 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
5455 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);