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 <linux/perf_event.h>
44 #include <trace/events/kvm.h>
46 #define CREATE_TRACE_POINTS
49 #include <asm/debugreg.h>
50 #include <asm/uaccess.h>
56 #define MAX_IO_MSRS 256
57 #define CR0_RESERVED_BITS \
58 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
59 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
60 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
61 #define CR4_RESERVED_BITS \
62 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
63 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
64 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
65 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
67 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
69 #define KVM_MAX_MCE_BANKS 32
70 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
73 * - enable syscall per default because its emulated by KVM
74 * - enable LME and LMA per default on 64 bit KVM
77 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
79 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
82 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
83 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
85 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
86 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
87 struct kvm_cpuid_entry2 __user *entries);
89 struct kvm_x86_ops *kvm_x86_ops;
90 EXPORT_SYMBOL_GPL(kvm_x86_ops);
93 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
95 #define KVM_NR_SHARED_MSRS 16
97 struct kvm_shared_msrs_global {
99 u32 msrs[KVM_NR_SHARED_MSRS];
102 struct kvm_shared_msrs {
103 struct user_return_notifier urn;
105 struct kvm_shared_msr_values {
108 } values[KVM_NR_SHARED_MSRS];
111 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
112 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
114 struct kvm_stats_debugfs_item debugfs_entries[] = {
115 { "pf_fixed", VCPU_STAT(pf_fixed) },
116 { "pf_guest", VCPU_STAT(pf_guest) },
117 { "tlb_flush", VCPU_STAT(tlb_flush) },
118 { "invlpg", VCPU_STAT(invlpg) },
119 { "exits", VCPU_STAT(exits) },
120 { "io_exits", VCPU_STAT(io_exits) },
121 { "mmio_exits", VCPU_STAT(mmio_exits) },
122 { "signal_exits", VCPU_STAT(signal_exits) },
123 { "irq_window", VCPU_STAT(irq_window_exits) },
124 { "nmi_window", VCPU_STAT(nmi_window_exits) },
125 { "halt_exits", VCPU_STAT(halt_exits) },
126 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
127 { "hypercalls", VCPU_STAT(hypercalls) },
128 { "request_irq", VCPU_STAT(request_irq_exits) },
129 { "irq_exits", VCPU_STAT(irq_exits) },
130 { "host_state_reload", VCPU_STAT(host_state_reload) },
131 { "efer_reload", VCPU_STAT(efer_reload) },
132 { "fpu_reload", VCPU_STAT(fpu_reload) },
133 { "insn_emulation", VCPU_STAT(insn_emulation) },
134 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
135 { "irq_injections", VCPU_STAT(irq_injections) },
136 { "nmi_injections", VCPU_STAT(nmi_injections) },
137 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
138 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
139 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
140 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
141 { "mmu_flooded", VM_STAT(mmu_flooded) },
142 { "mmu_recycled", VM_STAT(mmu_recycled) },
143 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
144 { "mmu_unsync", VM_STAT(mmu_unsync) },
145 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
146 { "largepages", VM_STAT(lpages) },
150 static void kvm_on_user_return(struct user_return_notifier *urn)
153 struct kvm_shared_msrs *locals
154 = container_of(urn, struct kvm_shared_msrs, urn);
155 struct kvm_shared_msr_values *values;
157 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
158 values = &locals->values[slot];
159 if (values->host != values->curr) {
160 wrmsrl(shared_msrs_global.msrs[slot], values->host);
161 values->curr = values->host;
164 locals->registered = false;
165 user_return_notifier_unregister(urn);
168 static void shared_msr_update(unsigned slot, u32 msr)
170 struct kvm_shared_msrs *smsr;
173 smsr = &__get_cpu_var(shared_msrs);
174 /* only read, and nobody should modify it at this time,
175 * so don't need lock */
176 if (slot >= shared_msrs_global.nr) {
177 printk(KERN_ERR "kvm: invalid MSR slot!");
180 rdmsrl_safe(msr, &value);
181 smsr->values[slot].host = value;
182 smsr->values[slot].curr = value;
185 void kvm_define_shared_msr(unsigned slot, u32 msr)
187 if (slot >= shared_msrs_global.nr)
188 shared_msrs_global.nr = slot + 1;
189 shared_msrs_global.msrs[slot] = msr;
190 /* we need ensured the shared_msr_global have been updated */
193 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
195 static void kvm_shared_msr_cpu_online(void)
199 for (i = 0; i < shared_msrs_global.nr; ++i)
200 shared_msr_update(i, shared_msrs_global.msrs[i]);
203 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
205 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
207 if (((value ^ smsr->values[slot].curr) & mask) == 0)
209 smsr->values[slot].curr = value;
210 wrmsrl(shared_msrs_global.msrs[slot], value);
211 if (!smsr->registered) {
212 smsr->urn.on_user_return = kvm_on_user_return;
213 user_return_notifier_register(&smsr->urn);
214 smsr->registered = true;
217 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
219 static void drop_user_return_notifiers(void *ignore)
221 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
223 if (smsr->registered)
224 kvm_on_user_return(&smsr->urn);
227 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
229 if (irqchip_in_kernel(vcpu->kvm))
230 return vcpu->arch.apic_base;
232 return vcpu->arch.apic_base;
234 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
236 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
238 /* TODO: reserve bits check */
239 if (irqchip_in_kernel(vcpu->kvm))
240 kvm_lapic_set_base(vcpu, data);
242 vcpu->arch.apic_base = data;
244 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
246 #define EXCPT_BENIGN 0
247 #define EXCPT_CONTRIBUTORY 1
250 static int exception_class(int vector)
260 return EXCPT_CONTRIBUTORY;
267 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
268 unsigned nr, bool has_error, u32 error_code,
274 if (!vcpu->arch.exception.pending) {
276 vcpu->arch.exception.pending = true;
277 vcpu->arch.exception.has_error_code = has_error;
278 vcpu->arch.exception.nr = nr;
279 vcpu->arch.exception.error_code = error_code;
280 vcpu->arch.exception.reinject = reinject;
284 /* to check exception */
285 prev_nr = vcpu->arch.exception.nr;
286 if (prev_nr == DF_VECTOR) {
287 /* triple fault -> shutdown */
288 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
291 class1 = exception_class(prev_nr);
292 class2 = exception_class(nr);
293 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
294 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
295 /* generate double fault per SDM Table 5-5 */
296 vcpu->arch.exception.pending = true;
297 vcpu->arch.exception.has_error_code = true;
298 vcpu->arch.exception.nr = DF_VECTOR;
299 vcpu->arch.exception.error_code = 0;
301 /* replace previous exception with a new one in a hope
302 that instruction re-execution will regenerate lost
307 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
309 kvm_multiple_exception(vcpu, nr, false, 0, false);
311 EXPORT_SYMBOL_GPL(kvm_queue_exception);
313 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
315 kvm_multiple_exception(vcpu, nr, false, 0, true);
317 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
319 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
322 ++vcpu->stat.pf_guest;
323 vcpu->arch.cr2 = addr;
324 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
327 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
329 vcpu->arch.nmi_pending = 1;
331 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
333 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
335 kvm_multiple_exception(vcpu, nr, true, error_code, false);
337 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
339 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
341 kvm_multiple_exception(vcpu, nr, true, error_code, true);
343 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
346 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
347 * a #GP and return false.
349 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
351 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
353 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
356 EXPORT_SYMBOL_GPL(kvm_require_cpl);
359 * Load the pae pdptrs. Return true is they are all valid.
361 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
363 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
364 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
367 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
369 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
370 offset * sizeof(u64), sizeof(pdpte));
375 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
376 if (is_present_gpte(pdpte[i]) &&
377 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
384 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
385 __set_bit(VCPU_EXREG_PDPTR,
386 (unsigned long *)&vcpu->arch.regs_avail);
387 __set_bit(VCPU_EXREG_PDPTR,
388 (unsigned long *)&vcpu->arch.regs_dirty);
393 EXPORT_SYMBOL_GPL(load_pdptrs);
395 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
397 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
401 if (is_long_mode(vcpu) || !is_pae(vcpu))
404 if (!test_bit(VCPU_EXREG_PDPTR,
405 (unsigned long *)&vcpu->arch.regs_avail))
408 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
411 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
417 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
422 if (cr0 & 0xffffffff00000000UL) {
423 kvm_inject_gp(vcpu, 0);
428 cr0 &= ~CR0_RESERVED_BITS;
430 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
431 kvm_inject_gp(vcpu, 0);
435 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
436 kvm_inject_gp(vcpu, 0);
440 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
442 if ((vcpu->arch.efer & EFER_LME)) {
446 kvm_inject_gp(vcpu, 0);
449 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
451 kvm_inject_gp(vcpu, 0);
457 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
458 kvm_inject_gp(vcpu, 0);
464 kvm_x86_ops->set_cr0(vcpu, cr0);
466 kvm_mmu_reset_context(vcpu);
469 EXPORT_SYMBOL_GPL(kvm_set_cr0);
471 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
473 kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
475 EXPORT_SYMBOL_GPL(kvm_lmsw);
477 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
479 unsigned long old_cr4 = kvm_read_cr4(vcpu);
480 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
482 if (cr4 & CR4_RESERVED_BITS) {
483 kvm_inject_gp(vcpu, 0);
487 if (is_long_mode(vcpu)) {
488 if (!(cr4 & X86_CR4_PAE)) {
489 kvm_inject_gp(vcpu, 0);
492 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
493 && ((cr4 ^ old_cr4) & pdptr_bits)
494 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
495 kvm_inject_gp(vcpu, 0);
499 if (cr4 & X86_CR4_VMXE) {
500 kvm_inject_gp(vcpu, 0);
503 kvm_x86_ops->set_cr4(vcpu, cr4);
504 vcpu->arch.cr4 = cr4;
505 kvm_mmu_reset_context(vcpu);
507 EXPORT_SYMBOL_GPL(kvm_set_cr4);
509 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
511 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
512 kvm_mmu_sync_roots(vcpu);
513 kvm_mmu_flush_tlb(vcpu);
517 if (is_long_mode(vcpu)) {
518 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
519 kvm_inject_gp(vcpu, 0);
524 if (cr3 & CR3_PAE_RESERVED_BITS) {
525 kvm_inject_gp(vcpu, 0);
528 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
529 kvm_inject_gp(vcpu, 0);
534 * We don't check reserved bits in nonpae mode, because
535 * this isn't enforced, and VMware depends on this.
540 * Does the new cr3 value map to physical memory? (Note, we
541 * catch an invalid cr3 even in real-mode, because it would
542 * cause trouble later on when we turn on paging anyway.)
544 * A real CPU would silently accept an invalid cr3 and would
545 * attempt to use it - with largely undefined (and often hard
546 * to debug) behavior on the guest side.
548 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
549 kvm_inject_gp(vcpu, 0);
551 vcpu->arch.cr3 = cr3;
552 vcpu->arch.mmu.new_cr3(vcpu);
555 EXPORT_SYMBOL_GPL(kvm_set_cr3);
557 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
559 if (cr8 & CR8_RESERVED_BITS) {
560 kvm_inject_gp(vcpu, 0);
563 if (irqchip_in_kernel(vcpu->kvm))
564 kvm_lapic_set_tpr(vcpu, cr8);
566 vcpu->arch.cr8 = cr8;
568 EXPORT_SYMBOL_GPL(kvm_set_cr8);
570 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
572 if (irqchip_in_kernel(vcpu->kvm))
573 return kvm_lapic_get_cr8(vcpu);
575 return vcpu->arch.cr8;
577 EXPORT_SYMBOL_GPL(kvm_get_cr8);
579 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
583 vcpu->arch.db[dr] = val;
584 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
585 vcpu->arch.eff_db[dr] = val;
588 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) {
589 kvm_queue_exception(vcpu, UD_VECTOR);
594 if (val & 0xffffffff00000000ULL) {
595 kvm_inject_gp(vcpu, 0);
598 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
601 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) {
602 kvm_queue_exception(vcpu, UD_VECTOR);
607 if (val & 0xffffffff00000000ULL) {
608 kvm_inject_gp(vcpu, 0);
611 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
612 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
613 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
614 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
621 EXPORT_SYMBOL_GPL(kvm_set_dr);
623 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
627 *val = vcpu->arch.db[dr];
630 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) {
631 kvm_queue_exception(vcpu, UD_VECTOR);
636 *val = vcpu->arch.dr6;
639 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) {
640 kvm_queue_exception(vcpu, UD_VECTOR);
645 *val = vcpu->arch.dr7;
651 EXPORT_SYMBOL_GPL(kvm_get_dr);
653 static inline u32 bit(int bitno)
655 return 1 << (bitno & 31);
659 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
660 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
662 * This list is modified at module load time to reflect the
663 * capabilities of the host cpu. This capabilities test skips MSRs that are
664 * kvm-specific. Those are put in the beginning of the list.
667 #define KVM_SAVE_MSRS_BEGIN 7
668 static u32 msrs_to_save[] = {
669 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
670 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
671 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
672 HV_X64_MSR_APIC_ASSIST_PAGE,
673 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
676 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
678 MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
681 static unsigned num_msrs_to_save;
683 static u32 emulated_msrs[] = {
684 MSR_IA32_MISC_ENABLE,
687 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
689 if (efer & efer_reserved_bits)
693 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
696 if (efer & EFER_FFXSR) {
697 struct kvm_cpuid_entry2 *feat;
699 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
700 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
704 if (efer & EFER_SVME) {
705 struct kvm_cpuid_entry2 *feat;
707 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
708 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
713 efer |= vcpu->arch.efer & EFER_LMA;
715 kvm_x86_ops->set_efer(vcpu, efer);
717 vcpu->arch.efer = efer;
719 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
720 kvm_mmu_reset_context(vcpu);
725 void kvm_enable_efer_bits(u64 mask)
727 efer_reserved_bits &= ~mask;
729 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
733 * Writes msr value into into the appropriate "register".
734 * Returns 0 on success, non-0 otherwise.
735 * Assumes vcpu_load() was already called.
737 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
739 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
743 * Adapt set_msr() to msr_io()'s calling convention
745 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
747 return kvm_set_msr(vcpu, index, *data);
750 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
754 struct pvclock_wall_clock wc;
755 struct timespec boot;
760 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
765 ++version; /* first time write, random junk */
769 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
772 * The guest calculates current wall clock time by adding
773 * system time (updated by kvm_write_guest_time below) to the
774 * wall clock specified here. guest system time equals host
775 * system time for us, thus we must fill in host boot time here.
779 wc.sec = boot.tv_sec;
780 wc.nsec = boot.tv_nsec;
781 wc.version = version;
783 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
786 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
789 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
791 uint32_t quotient, remainder;
793 /* Don't try to replace with do_div(), this one calculates
794 * "(dividend << 32) / divisor" */
796 : "=a" (quotient), "=d" (remainder)
797 : "0" (0), "1" (dividend), "r" (divisor) );
801 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
803 uint64_t nsecs = 1000000000LL;
808 tps64 = tsc_khz * 1000LL;
809 while (tps64 > nsecs*2) {
814 tps32 = (uint32_t)tps64;
815 while (tps32 <= (uint32_t)nsecs) {
820 hv_clock->tsc_shift = shift;
821 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
823 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
824 __func__, tsc_khz, hv_clock->tsc_shift,
825 hv_clock->tsc_to_system_mul);
828 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
830 static void kvm_write_guest_time(struct kvm_vcpu *v)
834 struct kvm_vcpu_arch *vcpu = &v->arch;
836 unsigned long this_tsc_khz;
838 if ((!vcpu->time_page))
841 this_tsc_khz = get_cpu_var(cpu_tsc_khz);
842 if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
843 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
844 vcpu->hv_clock_tsc_khz = this_tsc_khz;
846 put_cpu_var(cpu_tsc_khz);
848 /* Keep irq disabled to prevent changes to the clock */
849 local_irq_save(flags);
850 kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
852 monotonic_to_bootbased(&ts);
853 local_irq_restore(flags);
855 /* With all the info we got, fill in the values */
857 vcpu->hv_clock.system_time = ts.tv_nsec +
858 (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
860 vcpu->hv_clock.flags = 0;
863 * The interface expects us to write an even number signaling that the
864 * update is finished. Since the guest won't see the intermediate
865 * state, we just increase by 2 at the end.
867 vcpu->hv_clock.version += 2;
869 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
871 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
872 sizeof(vcpu->hv_clock));
874 kunmap_atomic(shared_kaddr, KM_USER0);
876 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
879 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
881 struct kvm_vcpu_arch *vcpu = &v->arch;
883 if (!vcpu->time_page)
885 set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
889 static bool msr_mtrr_valid(unsigned msr)
892 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
893 case MSR_MTRRfix64K_00000:
894 case MSR_MTRRfix16K_80000:
895 case MSR_MTRRfix16K_A0000:
896 case MSR_MTRRfix4K_C0000:
897 case MSR_MTRRfix4K_C8000:
898 case MSR_MTRRfix4K_D0000:
899 case MSR_MTRRfix4K_D8000:
900 case MSR_MTRRfix4K_E0000:
901 case MSR_MTRRfix4K_E8000:
902 case MSR_MTRRfix4K_F0000:
903 case MSR_MTRRfix4K_F8000:
904 case MSR_MTRRdefType:
905 case MSR_IA32_CR_PAT:
913 static bool valid_pat_type(unsigned t)
915 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
918 static bool valid_mtrr_type(unsigned t)
920 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
923 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
927 if (!msr_mtrr_valid(msr))
930 if (msr == MSR_IA32_CR_PAT) {
931 for (i = 0; i < 8; i++)
932 if (!valid_pat_type((data >> (i * 8)) & 0xff))
935 } else if (msr == MSR_MTRRdefType) {
938 return valid_mtrr_type(data & 0xff);
939 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
940 for (i = 0; i < 8 ; i++)
941 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
947 return valid_mtrr_type(data & 0xff);
950 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
952 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
954 if (!mtrr_valid(vcpu, msr, data))
957 if (msr == MSR_MTRRdefType) {
958 vcpu->arch.mtrr_state.def_type = data;
959 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
960 } else if (msr == MSR_MTRRfix64K_00000)
962 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
963 p[1 + msr - MSR_MTRRfix16K_80000] = data;
964 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
965 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
966 else if (msr == MSR_IA32_CR_PAT)
967 vcpu->arch.pat = data;
968 else { /* Variable MTRRs */
969 int idx, is_mtrr_mask;
972 idx = (msr - 0x200) / 2;
973 is_mtrr_mask = msr - 0x200 - 2 * idx;
976 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
979 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
983 kvm_mmu_reset_context(vcpu);
987 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
989 u64 mcg_cap = vcpu->arch.mcg_cap;
990 unsigned bank_num = mcg_cap & 0xff;
993 case MSR_IA32_MCG_STATUS:
994 vcpu->arch.mcg_status = data;
996 case MSR_IA32_MCG_CTL:
997 if (!(mcg_cap & MCG_CTL_P))
999 if (data != 0 && data != ~(u64)0)
1001 vcpu->arch.mcg_ctl = data;
1004 if (msr >= MSR_IA32_MC0_CTL &&
1005 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1006 u32 offset = msr - MSR_IA32_MC0_CTL;
1007 /* only 0 or all 1s can be written to IA32_MCi_CTL
1008 * some Linux kernels though clear bit 10 in bank 4 to
1009 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1010 * this to avoid an uncatched #GP in the guest
1012 if ((offset & 0x3) == 0 &&
1013 data != 0 && (data | (1 << 10)) != ~(u64)0)
1015 vcpu->arch.mce_banks[offset] = data;
1023 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1025 struct kvm *kvm = vcpu->kvm;
1026 int lm = is_long_mode(vcpu);
1027 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1028 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1029 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1030 : kvm->arch.xen_hvm_config.blob_size_32;
1031 u32 page_num = data & ~PAGE_MASK;
1032 u64 page_addr = data & PAGE_MASK;
1037 if (page_num >= blob_size)
1040 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1044 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1046 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1055 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1057 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1060 static bool kvm_hv_msr_partition_wide(u32 msr)
1064 case HV_X64_MSR_GUEST_OS_ID:
1065 case HV_X64_MSR_HYPERCALL:
1073 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1075 struct kvm *kvm = vcpu->kvm;
1078 case HV_X64_MSR_GUEST_OS_ID:
1079 kvm->arch.hv_guest_os_id = data;
1080 /* setting guest os id to zero disables hypercall page */
1081 if (!kvm->arch.hv_guest_os_id)
1082 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1084 case HV_X64_MSR_HYPERCALL: {
1089 /* if guest os id is not set hypercall should remain disabled */
1090 if (!kvm->arch.hv_guest_os_id)
1092 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1093 kvm->arch.hv_hypercall = data;
1096 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1097 addr = gfn_to_hva(kvm, gfn);
1098 if (kvm_is_error_hva(addr))
1100 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1101 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1102 if (copy_to_user((void __user *)addr, instructions, 4))
1104 kvm->arch.hv_hypercall = data;
1108 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1109 "data 0x%llx\n", msr, data);
1115 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1118 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1121 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1122 vcpu->arch.hv_vapic = data;
1125 addr = gfn_to_hva(vcpu->kvm, data >>
1126 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1127 if (kvm_is_error_hva(addr))
1129 if (clear_user((void __user *)addr, PAGE_SIZE))
1131 vcpu->arch.hv_vapic = data;
1134 case HV_X64_MSR_EOI:
1135 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1136 case HV_X64_MSR_ICR:
1137 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1138 case HV_X64_MSR_TPR:
1139 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1141 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1142 "data 0x%llx\n", msr, data);
1149 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1153 return set_efer(vcpu, data);
1155 data &= ~(u64)0x40; /* ignore flush filter disable */
1156 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1158 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1163 case MSR_FAM10H_MMIO_CONF_BASE:
1165 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1170 case MSR_AMD64_NB_CFG:
1172 case MSR_IA32_DEBUGCTLMSR:
1174 /* We support the non-activated case already */
1176 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1177 /* Values other than LBR and BTF are vendor-specific,
1178 thus reserved and should throw a #GP */
1181 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1184 case MSR_IA32_UCODE_REV:
1185 case MSR_IA32_UCODE_WRITE:
1186 case MSR_VM_HSAVE_PA:
1187 case MSR_AMD64_PATCH_LOADER:
1189 case 0x200 ... 0x2ff:
1190 return set_msr_mtrr(vcpu, msr, data);
1191 case MSR_IA32_APICBASE:
1192 kvm_set_apic_base(vcpu, data);
1194 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1195 return kvm_x2apic_msr_write(vcpu, msr, data);
1196 case MSR_IA32_MISC_ENABLE:
1197 vcpu->arch.ia32_misc_enable_msr = data;
1199 case MSR_KVM_WALL_CLOCK_NEW:
1200 case MSR_KVM_WALL_CLOCK:
1201 vcpu->kvm->arch.wall_clock = data;
1202 kvm_write_wall_clock(vcpu->kvm, data);
1204 case MSR_KVM_SYSTEM_TIME_NEW:
1205 case MSR_KVM_SYSTEM_TIME: {
1206 if (vcpu->arch.time_page) {
1207 kvm_release_page_dirty(vcpu->arch.time_page);
1208 vcpu->arch.time_page = NULL;
1211 vcpu->arch.time = data;
1213 /* we verify if the enable bit is set... */
1217 /* ...but clean it before doing the actual write */
1218 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1220 vcpu->arch.time_page =
1221 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1223 if (is_error_page(vcpu->arch.time_page)) {
1224 kvm_release_page_clean(vcpu->arch.time_page);
1225 vcpu->arch.time_page = NULL;
1228 kvm_request_guest_time_update(vcpu);
1231 case MSR_IA32_MCG_CTL:
1232 case MSR_IA32_MCG_STATUS:
1233 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1234 return set_msr_mce(vcpu, msr, data);
1236 /* Performance counters are not protected by a CPUID bit,
1237 * so we should check all of them in the generic path for the sake of
1238 * cross vendor migration.
1239 * Writing a zero into the event select MSRs disables them,
1240 * which we perfectly emulate ;-). Any other value should be at least
1241 * reported, some guests depend on them.
1243 case MSR_P6_EVNTSEL0:
1244 case MSR_P6_EVNTSEL1:
1245 case MSR_K7_EVNTSEL0:
1246 case MSR_K7_EVNTSEL1:
1247 case MSR_K7_EVNTSEL2:
1248 case MSR_K7_EVNTSEL3:
1250 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1251 "0x%x data 0x%llx\n", msr, data);
1253 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1254 * so we ignore writes to make it happy.
1256 case MSR_P6_PERFCTR0:
1257 case MSR_P6_PERFCTR1:
1258 case MSR_K7_PERFCTR0:
1259 case MSR_K7_PERFCTR1:
1260 case MSR_K7_PERFCTR2:
1261 case MSR_K7_PERFCTR3:
1262 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1263 "0x%x data 0x%llx\n", msr, data);
1265 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1266 if (kvm_hv_msr_partition_wide(msr)) {
1268 mutex_lock(&vcpu->kvm->lock);
1269 r = set_msr_hyperv_pw(vcpu, msr, data);
1270 mutex_unlock(&vcpu->kvm->lock);
1273 return set_msr_hyperv(vcpu, msr, data);
1276 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1277 return xen_hvm_config(vcpu, data);
1279 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1283 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1290 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1294 * Reads an msr value (of 'msr_index') into 'pdata'.
1295 * Returns 0 on success, non-0 otherwise.
1296 * Assumes vcpu_load() was already called.
1298 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1300 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1303 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1305 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1307 if (!msr_mtrr_valid(msr))
1310 if (msr == MSR_MTRRdefType)
1311 *pdata = vcpu->arch.mtrr_state.def_type +
1312 (vcpu->arch.mtrr_state.enabled << 10);
1313 else if (msr == MSR_MTRRfix64K_00000)
1315 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1316 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1317 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1318 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1319 else if (msr == MSR_IA32_CR_PAT)
1320 *pdata = vcpu->arch.pat;
1321 else { /* Variable MTRRs */
1322 int idx, is_mtrr_mask;
1325 idx = (msr - 0x200) / 2;
1326 is_mtrr_mask = msr - 0x200 - 2 * idx;
1329 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1332 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1339 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1342 u64 mcg_cap = vcpu->arch.mcg_cap;
1343 unsigned bank_num = mcg_cap & 0xff;
1346 case MSR_IA32_P5_MC_ADDR:
1347 case MSR_IA32_P5_MC_TYPE:
1350 case MSR_IA32_MCG_CAP:
1351 data = vcpu->arch.mcg_cap;
1353 case MSR_IA32_MCG_CTL:
1354 if (!(mcg_cap & MCG_CTL_P))
1356 data = vcpu->arch.mcg_ctl;
1358 case MSR_IA32_MCG_STATUS:
1359 data = vcpu->arch.mcg_status;
1362 if (msr >= MSR_IA32_MC0_CTL &&
1363 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1364 u32 offset = msr - MSR_IA32_MC0_CTL;
1365 data = vcpu->arch.mce_banks[offset];
1374 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1377 struct kvm *kvm = vcpu->kvm;
1380 case HV_X64_MSR_GUEST_OS_ID:
1381 data = kvm->arch.hv_guest_os_id;
1383 case HV_X64_MSR_HYPERCALL:
1384 data = kvm->arch.hv_hypercall;
1387 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1395 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1400 case HV_X64_MSR_VP_INDEX: {
1403 kvm_for_each_vcpu(r, v, vcpu->kvm)
1408 case HV_X64_MSR_EOI:
1409 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1410 case HV_X64_MSR_ICR:
1411 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1412 case HV_X64_MSR_TPR:
1413 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1415 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1422 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1427 case MSR_IA32_PLATFORM_ID:
1428 case MSR_IA32_UCODE_REV:
1429 case MSR_IA32_EBL_CR_POWERON:
1430 case MSR_IA32_DEBUGCTLMSR:
1431 case MSR_IA32_LASTBRANCHFROMIP:
1432 case MSR_IA32_LASTBRANCHTOIP:
1433 case MSR_IA32_LASTINTFROMIP:
1434 case MSR_IA32_LASTINTTOIP:
1437 case MSR_VM_HSAVE_PA:
1438 case MSR_P6_PERFCTR0:
1439 case MSR_P6_PERFCTR1:
1440 case MSR_P6_EVNTSEL0:
1441 case MSR_P6_EVNTSEL1:
1442 case MSR_K7_EVNTSEL0:
1443 case MSR_K7_PERFCTR0:
1444 case MSR_K8_INT_PENDING_MSG:
1445 case MSR_AMD64_NB_CFG:
1446 case MSR_FAM10H_MMIO_CONF_BASE:
1450 data = 0x500 | KVM_NR_VAR_MTRR;
1452 case 0x200 ... 0x2ff:
1453 return get_msr_mtrr(vcpu, msr, pdata);
1454 case 0xcd: /* fsb frequency */
1457 case MSR_IA32_APICBASE:
1458 data = kvm_get_apic_base(vcpu);
1460 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1461 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1463 case MSR_IA32_MISC_ENABLE:
1464 data = vcpu->arch.ia32_misc_enable_msr;
1466 case MSR_IA32_PERF_STATUS:
1467 /* TSC increment by tick */
1469 /* CPU multiplier */
1470 data |= (((uint64_t)4ULL) << 40);
1473 data = vcpu->arch.efer;
1475 case MSR_KVM_WALL_CLOCK:
1476 case MSR_KVM_WALL_CLOCK_NEW:
1477 data = vcpu->kvm->arch.wall_clock;
1479 case MSR_KVM_SYSTEM_TIME:
1480 case MSR_KVM_SYSTEM_TIME_NEW:
1481 data = vcpu->arch.time;
1483 case MSR_IA32_P5_MC_ADDR:
1484 case MSR_IA32_P5_MC_TYPE:
1485 case MSR_IA32_MCG_CAP:
1486 case MSR_IA32_MCG_CTL:
1487 case MSR_IA32_MCG_STATUS:
1488 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1489 return get_msr_mce(vcpu, msr, pdata);
1490 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1491 if (kvm_hv_msr_partition_wide(msr)) {
1493 mutex_lock(&vcpu->kvm->lock);
1494 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1495 mutex_unlock(&vcpu->kvm->lock);
1498 return get_msr_hyperv(vcpu, msr, pdata);
1502 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1505 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1513 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1516 * Read or write a bunch of msrs. All parameters are kernel addresses.
1518 * @return number of msrs set successfully.
1520 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1521 struct kvm_msr_entry *entries,
1522 int (*do_msr)(struct kvm_vcpu *vcpu,
1523 unsigned index, u64 *data))
1529 idx = srcu_read_lock(&vcpu->kvm->srcu);
1530 for (i = 0; i < msrs->nmsrs; ++i)
1531 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1533 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1541 * Read or write a bunch of msrs. Parameters are user addresses.
1543 * @return number of msrs set successfully.
1545 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1546 int (*do_msr)(struct kvm_vcpu *vcpu,
1547 unsigned index, u64 *data),
1550 struct kvm_msrs msrs;
1551 struct kvm_msr_entry *entries;
1556 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1560 if (msrs.nmsrs >= MAX_IO_MSRS)
1564 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1565 entries = vmalloc(size);
1570 if (copy_from_user(entries, user_msrs->entries, size))
1573 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1578 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1589 int kvm_dev_ioctl_check_extension(long ext)
1594 case KVM_CAP_IRQCHIP:
1596 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1597 case KVM_CAP_SET_TSS_ADDR:
1598 case KVM_CAP_EXT_CPUID:
1599 case KVM_CAP_CLOCKSOURCE:
1601 case KVM_CAP_NOP_IO_DELAY:
1602 case KVM_CAP_MP_STATE:
1603 case KVM_CAP_SYNC_MMU:
1604 case KVM_CAP_REINJECT_CONTROL:
1605 case KVM_CAP_IRQ_INJECT_STATUS:
1606 case KVM_CAP_ASSIGN_DEV_IRQ:
1608 case KVM_CAP_IOEVENTFD:
1610 case KVM_CAP_PIT_STATE2:
1611 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1612 case KVM_CAP_XEN_HVM:
1613 case KVM_CAP_ADJUST_CLOCK:
1614 case KVM_CAP_VCPU_EVENTS:
1615 case KVM_CAP_HYPERV:
1616 case KVM_CAP_HYPERV_VAPIC:
1617 case KVM_CAP_HYPERV_SPIN:
1618 case KVM_CAP_PCI_SEGMENT:
1619 case KVM_CAP_DEBUGREGS:
1620 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1623 case KVM_CAP_COALESCED_MMIO:
1624 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1627 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1629 case KVM_CAP_NR_VCPUS:
1632 case KVM_CAP_NR_MEMSLOTS:
1633 r = KVM_MEMORY_SLOTS;
1635 case KVM_CAP_PV_MMU: /* obsolete */
1642 r = KVM_MAX_MCE_BANKS;
1652 long kvm_arch_dev_ioctl(struct file *filp,
1653 unsigned int ioctl, unsigned long arg)
1655 void __user *argp = (void __user *)arg;
1659 case KVM_GET_MSR_INDEX_LIST: {
1660 struct kvm_msr_list __user *user_msr_list = argp;
1661 struct kvm_msr_list msr_list;
1665 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1668 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1669 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1672 if (n < msr_list.nmsrs)
1675 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1676 num_msrs_to_save * sizeof(u32)))
1678 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1680 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1685 case KVM_GET_SUPPORTED_CPUID: {
1686 struct kvm_cpuid2 __user *cpuid_arg = argp;
1687 struct kvm_cpuid2 cpuid;
1690 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1692 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1693 cpuid_arg->entries);
1698 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1703 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1706 mce_cap = KVM_MCE_CAP_SUPPORTED;
1708 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1720 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1722 kvm_x86_ops->vcpu_load(vcpu, cpu);
1723 if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1724 unsigned long khz = cpufreq_quick_get(cpu);
1727 per_cpu(cpu_tsc_khz, cpu) = khz;
1729 kvm_request_guest_time_update(vcpu);
1732 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1734 kvm_put_guest_fpu(vcpu);
1735 kvm_x86_ops->vcpu_put(vcpu);
1738 static int is_efer_nx(void)
1740 unsigned long long efer = 0;
1742 rdmsrl_safe(MSR_EFER, &efer);
1743 return efer & EFER_NX;
1746 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1749 struct kvm_cpuid_entry2 *e, *entry;
1752 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1753 e = &vcpu->arch.cpuid_entries[i];
1754 if (e->function == 0x80000001) {
1759 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1760 entry->edx &= ~(1 << 20);
1761 printk(KERN_INFO "kvm: guest NX capability removed\n");
1765 /* when an old userspace process fills a new kernel module */
1766 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1767 struct kvm_cpuid *cpuid,
1768 struct kvm_cpuid_entry __user *entries)
1771 struct kvm_cpuid_entry *cpuid_entries;
1774 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1777 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1781 if (copy_from_user(cpuid_entries, entries,
1782 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1785 for (i = 0; i < cpuid->nent; i++) {
1786 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1787 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1788 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1789 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1790 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1791 vcpu->arch.cpuid_entries[i].index = 0;
1792 vcpu->arch.cpuid_entries[i].flags = 0;
1793 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1794 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1795 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1797 vcpu->arch.cpuid_nent = cpuid->nent;
1798 cpuid_fix_nx_cap(vcpu);
1800 kvm_apic_set_version(vcpu);
1801 kvm_x86_ops->cpuid_update(vcpu);
1805 vfree(cpuid_entries);
1810 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1811 struct kvm_cpuid2 *cpuid,
1812 struct kvm_cpuid_entry2 __user *entries)
1817 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1820 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1821 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1824 vcpu->arch.cpuid_nent = cpuid->nent;
1825 kvm_apic_set_version(vcpu);
1826 kvm_x86_ops->cpuid_update(vcpu);
1834 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1835 struct kvm_cpuid2 *cpuid,
1836 struct kvm_cpuid_entry2 __user *entries)
1842 if (cpuid->nent < vcpu->arch.cpuid_nent)
1845 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1846 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1851 cpuid->nent = vcpu->arch.cpuid_nent;
1856 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1859 entry->function = function;
1860 entry->index = index;
1861 cpuid_count(entry->function, entry->index,
1862 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1866 #define F(x) bit(X86_FEATURE_##x)
1868 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1869 u32 index, int *nent, int maxnent)
1871 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1872 #ifdef CONFIG_X86_64
1873 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1875 unsigned f_lm = F(LM);
1877 unsigned f_gbpages = 0;
1880 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1883 const u32 kvm_supported_word0_x86_features =
1884 F(FPU) | F(VME) | F(DE) | F(PSE) |
1885 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1886 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1887 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1888 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1889 0 /* Reserved, DS, ACPI */ | F(MMX) |
1890 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1891 0 /* HTT, TM, Reserved, PBE */;
1892 /* cpuid 0x80000001.edx */
1893 const u32 kvm_supported_word1_x86_features =
1894 F(FPU) | F(VME) | F(DE) | F(PSE) |
1895 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1896 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1897 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1898 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1899 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1900 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1901 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1903 const u32 kvm_supported_word4_x86_features =
1904 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1905 0 /* DS-CPL, VMX, SMX, EST */ |
1906 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1907 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1908 0 /* Reserved, DCA */ | F(XMM4_1) |
1909 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1910 0 /* Reserved, XSAVE, OSXSAVE */;
1911 /* cpuid 0x80000001.ecx */
1912 const u32 kvm_supported_word6_x86_features =
1913 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1914 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1915 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1916 0 /* SKINIT */ | 0 /* WDT */;
1918 /* all calls to cpuid_count() should be made on the same cpu */
1920 do_cpuid_1_ent(entry, function, index);
1925 entry->eax = min(entry->eax, (u32)0xb);
1928 entry->edx &= kvm_supported_word0_x86_features;
1929 entry->ecx &= kvm_supported_word4_x86_features;
1930 /* we support x2apic emulation even if host does not support
1931 * it since we emulate x2apic in software */
1932 entry->ecx |= F(X2APIC);
1934 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1935 * may return different values. This forces us to get_cpu() before
1936 * issuing the first command, and also to emulate this annoying behavior
1937 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1939 int t, times = entry->eax & 0xff;
1941 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1942 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1943 for (t = 1; t < times && *nent < maxnent; ++t) {
1944 do_cpuid_1_ent(&entry[t], function, 0);
1945 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1950 /* function 4 and 0xb have additional index. */
1954 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1955 /* read more entries until cache_type is zero */
1956 for (i = 1; *nent < maxnent; ++i) {
1957 cache_type = entry[i - 1].eax & 0x1f;
1960 do_cpuid_1_ent(&entry[i], function, i);
1962 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1970 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1971 /* read more entries until level_type is zero */
1972 for (i = 1; *nent < maxnent; ++i) {
1973 level_type = entry[i - 1].ecx & 0xff00;
1976 do_cpuid_1_ent(&entry[i], function, i);
1978 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1983 case KVM_CPUID_SIGNATURE: {
1984 char signature[12] = "KVMKVMKVM\0\0";
1985 u32 *sigptr = (u32 *)signature;
1987 entry->ebx = sigptr[0];
1988 entry->ecx = sigptr[1];
1989 entry->edx = sigptr[2];
1992 case KVM_CPUID_FEATURES:
1993 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
1994 (1 << KVM_FEATURE_NOP_IO_DELAY) |
1995 (1 << KVM_FEATURE_CLOCKSOURCE2) |
1996 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2002 entry->eax = min(entry->eax, 0x8000001a);
2005 entry->edx &= kvm_supported_word1_x86_features;
2006 entry->ecx &= kvm_supported_word6_x86_features;
2010 kvm_x86_ops->set_supported_cpuid(function, entry);
2017 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2018 struct kvm_cpuid_entry2 __user *entries)
2020 struct kvm_cpuid_entry2 *cpuid_entries;
2021 int limit, nent = 0, r = -E2BIG;
2024 if (cpuid->nent < 1)
2026 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2027 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2029 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2033 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2034 limit = cpuid_entries[0].eax;
2035 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2036 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2037 &nent, cpuid->nent);
2039 if (nent >= cpuid->nent)
2042 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2043 limit = cpuid_entries[nent - 1].eax;
2044 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2045 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2046 &nent, cpuid->nent);
2051 if (nent >= cpuid->nent)
2054 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2058 if (nent >= cpuid->nent)
2061 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2065 if (nent >= cpuid->nent)
2069 if (copy_to_user(entries, cpuid_entries,
2070 nent * sizeof(struct kvm_cpuid_entry2)))
2076 vfree(cpuid_entries);
2081 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2082 struct kvm_lapic_state *s)
2085 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2091 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2092 struct kvm_lapic_state *s)
2095 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2096 kvm_apic_post_state_restore(vcpu);
2097 update_cr8_intercept(vcpu);
2103 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2104 struct kvm_interrupt *irq)
2106 if (irq->irq < 0 || irq->irq >= 256)
2108 if (irqchip_in_kernel(vcpu->kvm))
2112 kvm_queue_interrupt(vcpu, irq->irq, false);
2119 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2122 kvm_inject_nmi(vcpu);
2128 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2129 struct kvm_tpr_access_ctl *tac)
2133 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2137 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2141 unsigned bank_num = mcg_cap & 0xff, bank;
2145 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2147 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2150 vcpu->arch.mcg_cap = mcg_cap;
2151 /* Init IA32_MCG_CTL to all 1s */
2152 if (mcg_cap & MCG_CTL_P)
2153 vcpu->arch.mcg_ctl = ~(u64)0;
2154 /* Init IA32_MCi_CTL to all 1s */
2155 for (bank = 0; bank < bank_num; bank++)
2156 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2162 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2163 struct kvm_x86_mce *mce)
2165 u64 mcg_cap = vcpu->arch.mcg_cap;
2166 unsigned bank_num = mcg_cap & 0xff;
2167 u64 *banks = vcpu->arch.mce_banks;
2169 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2172 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2173 * reporting is disabled
2175 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2176 vcpu->arch.mcg_ctl != ~(u64)0)
2178 banks += 4 * mce->bank;
2180 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2181 * reporting is disabled for the bank
2183 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2185 if (mce->status & MCI_STATUS_UC) {
2186 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2187 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2188 printk(KERN_DEBUG "kvm: set_mce: "
2189 "injects mce exception while "
2190 "previous one is in progress!\n");
2191 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2194 if (banks[1] & MCI_STATUS_VAL)
2195 mce->status |= MCI_STATUS_OVER;
2196 banks[2] = mce->addr;
2197 banks[3] = mce->misc;
2198 vcpu->arch.mcg_status = mce->mcg_status;
2199 banks[1] = mce->status;
2200 kvm_queue_exception(vcpu, MC_VECTOR);
2201 } else if (!(banks[1] & MCI_STATUS_VAL)
2202 || !(banks[1] & MCI_STATUS_UC)) {
2203 if (banks[1] & MCI_STATUS_VAL)
2204 mce->status |= MCI_STATUS_OVER;
2205 banks[2] = mce->addr;
2206 banks[3] = mce->misc;
2207 banks[1] = mce->status;
2209 banks[1] |= MCI_STATUS_OVER;
2213 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2214 struct kvm_vcpu_events *events)
2218 events->exception.injected =
2219 vcpu->arch.exception.pending &&
2220 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2221 events->exception.nr = vcpu->arch.exception.nr;
2222 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2223 events->exception.error_code = vcpu->arch.exception.error_code;
2225 events->interrupt.injected =
2226 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2227 events->interrupt.nr = vcpu->arch.interrupt.nr;
2228 events->interrupt.soft = 0;
2229 events->interrupt.shadow =
2230 kvm_x86_ops->get_interrupt_shadow(vcpu,
2231 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2233 events->nmi.injected = vcpu->arch.nmi_injected;
2234 events->nmi.pending = vcpu->arch.nmi_pending;
2235 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2237 events->sipi_vector = vcpu->arch.sipi_vector;
2239 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2240 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2241 | KVM_VCPUEVENT_VALID_SHADOW);
2246 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2247 struct kvm_vcpu_events *events)
2249 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2250 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2251 | KVM_VCPUEVENT_VALID_SHADOW))
2256 vcpu->arch.exception.pending = events->exception.injected;
2257 vcpu->arch.exception.nr = events->exception.nr;
2258 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2259 vcpu->arch.exception.error_code = events->exception.error_code;
2261 vcpu->arch.interrupt.pending = events->interrupt.injected;
2262 vcpu->arch.interrupt.nr = events->interrupt.nr;
2263 vcpu->arch.interrupt.soft = events->interrupt.soft;
2264 if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2265 kvm_pic_clear_isr_ack(vcpu->kvm);
2266 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2267 kvm_x86_ops->set_interrupt_shadow(vcpu,
2268 events->interrupt.shadow);
2270 vcpu->arch.nmi_injected = events->nmi.injected;
2271 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2272 vcpu->arch.nmi_pending = events->nmi.pending;
2273 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2275 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2276 vcpu->arch.sipi_vector = events->sipi_vector;
2283 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2284 struct kvm_debugregs *dbgregs)
2288 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2289 dbgregs->dr6 = vcpu->arch.dr6;
2290 dbgregs->dr7 = vcpu->arch.dr7;
2296 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2297 struct kvm_debugregs *dbgregs)
2304 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2305 vcpu->arch.dr6 = dbgregs->dr6;
2306 vcpu->arch.dr7 = dbgregs->dr7;
2313 long kvm_arch_vcpu_ioctl(struct file *filp,
2314 unsigned int ioctl, unsigned long arg)
2316 struct kvm_vcpu *vcpu = filp->private_data;
2317 void __user *argp = (void __user *)arg;
2319 struct kvm_lapic_state *lapic = NULL;
2322 case KVM_GET_LAPIC: {
2324 if (!vcpu->arch.apic)
2326 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2331 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2335 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2340 case KVM_SET_LAPIC: {
2342 if (!vcpu->arch.apic)
2344 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2349 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2351 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2357 case KVM_INTERRUPT: {
2358 struct kvm_interrupt irq;
2361 if (copy_from_user(&irq, argp, sizeof irq))
2363 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2370 r = kvm_vcpu_ioctl_nmi(vcpu);
2376 case KVM_SET_CPUID: {
2377 struct kvm_cpuid __user *cpuid_arg = argp;
2378 struct kvm_cpuid cpuid;
2381 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2383 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2388 case KVM_SET_CPUID2: {
2389 struct kvm_cpuid2 __user *cpuid_arg = argp;
2390 struct kvm_cpuid2 cpuid;
2393 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2395 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2396 cpuid_arg->entries);
2401 case KVM_GET_CPUID2: {
2402 struct kvm_cpuid2 __user *cpuid_arg = argp;
2403 struct kvm_cpuid2 cpuid;
2406 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2408 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2409 cpuid_arg->entries);
2413 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2419 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2422 r = msr_io(vcpu, argp, do_set_msr, 0);
2424 case KVM_TPR_ACCESS_REPORTING: {
2425 struct kvm_tpr_access_ctl tac;
2428 if (copy_from_user(&tac, argp, sizeof tac))
2430 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2434 if (copy_to_user(argp, &tac, sizeof tac))
2439 case KVM_SET_VAPIC_ADDR: {
2440 struct kvm_vapic_addr va;
2443 if (!irqchip_in_kernel(vcpu->kvm))
2446 if (copy_from_user(&va, argp, sizeof va))
2449 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2452 case KVM_X86_SETUP_MCE: {
2456 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2458 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2461 case KVM_X86_SET_MCE: {
2462 struct kvm_x86_mce mce;
2465 if (copy_from_user(&mce, argp, sizeof mce))
2468 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2472 case KVM_GET_VCPU_EVENTS: {
2473 struct kvm_vcpu_events events;
2475 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2478 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2483 case KVM_SET_VCPU_EVENTS: {
2484 struct kvm_vcpu_events events;
2487 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2490 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2493 case KVM_GET_DEBUGREGS: {
2494 struct kvm_debugregs dbgregs;
2496 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2499 if (copy_to_user(argp, &dbgregs,
2500 sizeof(struct kvm_debugregs)))
2505 case KVM_SET_DEBUGREGS: {
2506 struct kvm_debugregs dbgregs;
2509 if (copy_from_user(&dbgregs, argp,
2510 sizeof(struct kvm_debugregs)))
2513 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2524 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2528 if (addr > (unsigned int)(-3 * PAGE_SIZE))
2530 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2534 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2537 kvm->arch.ept_identity_map_addr = ident_addr;
2541 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2542 u32 kvm_nr_mmu_pages)
2544 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2547 mutex_lock(&kvm->slots_lock);
2548 spin_lock(&kvm->mmu_lock);
2550 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2551 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2553 spin_unlock(&kvm->mmu_lock);
2554 mutex_unlock(&kvm->slots_lock);
2558 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2560 return kvm->arch.n_alloc_mmu_pages;
2563 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2566 struct kvm_mem_alias *alias;
2567 struct kvm_mem_aliases *aliases;
2569 aliases = kvm_aliases(kvm);
2571 for (i = 0; i < aliases->naliases; ++i) {
2572 alias = &aliases->aliases[i];
2573 if (alias->flags & KVM_ALIAS_INVALID)
2575 if (gfn >= alias->base_gfn
2576 && gfn < alias->base_gfn + alias->npages)
2577 return alias->target_gfn + gfn - alias->base_gfn;
2582 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2585 struct kvm_mem_alias *alias;
2586 struct kvm_mem_aliases *aliases;
2588 aliases = kvm_aliases(kvm);
2590 for (i = 0; i < aliases->naliases; ++i) {
2591 alias = &aliases->aliases[i];
2592 if (gfn >= alias->base_gfn
2593 && gfn < alias->base_gfn + alias->npages)
2594 return alias->target_gfn + gfn - alias->base_gfn;
2600 * Set a new alias region. Aliases map a portion of physical memory into
2601 * another portion. This is useful for memory windows, for example the PC
2604 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2605 struct kvm_memory_alias *alias)
2608 struct kvm_mem_alias *p;
2609 struct kvm_mem_aliases *aliases, *old_aliases;
2612 /* General sanity checks */
2613 if (alias->memory_size & (PAGE_SIZE - 1))
2615 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2617 if (alias->slot >= KVM_ALIAS_SLOTS)
2619 if (alias->guest_phys_addr + alias->memory_size
2620 < alias->guest_phys_addr)
2622 if (alias->target_phys_addr + alias->memory_size
2623 < alias->target_phys_addr)
2627 aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2631 mutex_lock(&kvm->slots_lock);
2633 /* invalidate any gfn reference in case of deletion/shrinking */
2634 memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2635 aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2636 old_aliases = kvm->arch.aliases;
2637 rcu_assign_pointer(kvm->arch.aliases, aliases);
2638 synchronize_srcu_expedited(&kvm->srcu);
2639 kvm_mmu_zap_all(kvm);
2643 aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2647 memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2649 p = &aliases->aliases[alias->slot];
2650 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2651 p->npages = alias->memory_size >> PAGE_SHIFT;
2652 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2653 p->flags &= ~(KVM_ALIAS_INVALID);
2655 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2656 if (aliases->aliases[n - 1].npages)
2658 aliases->naliases = n;
2660 old_aliases = kvm->arch.aliases;
2661 rcu_assign_pointer(kvm->arch.aliases, aliases);
2662 synchronize_srcu_expedited(&kvm->srcu);
2667 mutex_unlock(&kvm->slots_lock);
2672 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2677 switch (chip->chip_id) {
2678 case KVM_IRQCHIP_PIC_MASTER:
2679 memcpy(&chip->chip.pic,
2680 &pic_irqchip(kvm)->pics[0],
2681 sizeof(struct kvm_pic_state));
2683 case KVM_IRQCHIP_PIC_SLAVE:
2684 memcpy(&chip->chip.pic,
2685 &pic_irqchip(kvm)->pics[1],
2686 sizeof(struct kvm_pic_state));
2688 case KVM_IRQCHIP_IOAPIC:
2689 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2698 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2703 switch (chip->chip_id) {
2704 case KVM_IRQCHIP_PIC_MASTER:
2705 raw_spin_lock(&pic_irqchip(kvm)->lock);
2706 memcpy(&pic_irqchip(kvm)->pics[0],
2708 sizeof(struct kvm_pic_state));
2709 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2711 case KVM_IRQCHIP_PIC_SLAVE:
2712 raw_spin_lock(&pic_irqchip(kvm)->lock);
2713 memcpy(&pic_irqchip(kvm)->pics[1],
2715 sizeof(struct kvm_pic_state));
2716 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2718 case KVM_IRQCHIP_IOAPIC:
2719 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2725 kvm_pic_update_irq(pic_irqchip(kvm));
2729 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2733 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2734 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2735 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2739 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2743 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2744 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2745 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2746 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2750 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2754 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2755 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2756 sizeof(ps->channels));
2757 ps->flags = kvm->arch.vpit->pit_state.flags;
2758 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2762 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2764 int r = 0, start = 0;
2765 u32 prev_legacy, cur_legacy;
2766 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2767 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2768 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2769 if (!prev_legacy && cur_legacy)
2771 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2772 sizeof(kvm->arch.vpit->pit_state.channels));
2773 kvm->arch.vpit->pit_state.flags = ps->flags;
2774 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2775 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2779 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2780 struct kvm_reinject_control *control)
2782 if (!kvm->arch.vpit)
2784 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2785 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2786 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2791 * Get (and clear) the dirty memory log for a memory slot.
2793 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2794 struct kvm_dirty_log *log)
2797 struct kvm_memory_slot *memslot;
2799 unsigned long is_dirty = 0;
2800 unsigned long *dirty_bitmap = NULL;
2802 mutex_lock(&kvm->slots_lock);
2805 if (log->slot >= KVM_MEMORY_SLOTS)
2808 memslot = &kvm->memslots->memslots[log->slot];
2810 if (!memslot->dirty_bitmap)
2813 n = kvm_dirty_bitmap_bytes(memslot);
2816 dirty_bitmap = vmalloc(n);
2819 memset(dirty_bitmap, 0, n);
2821 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2822 is_dirty = memslot->dirty_bitmap[i];
2824 /* If nothing is dirty, don't bother messing with page tables. */
2826 struct kvm_memslots *slots, *old_slots;
2828 spin_lock(&kvm->mmu_lock);
2829 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2830 spin_unlock(&kvm->mmu_lock);
2832 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2836 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2837 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2839 old_slots = kvm->memslots;
2840 rcu_assign_pointer(kvm->memslots, slots);
2841 synchronize_srcu_expedited(&kvm->srcu);
2842 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2847 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2850 vfree(dirty_bitmap);
2852 mutex_unlock(&kvm->slots_lock);
2856 long kvm_arch_vm_ioctl(struct file *filp,
2857 unsigned int ioctl, unsigned long arg)
2859 struct kvm *kvm = filp->private_data;
2860 void __user *argp = (void __user *)arg;
2863 * This union makes it completely explicit to gcc-3.x
2864 * that these two variables' stack usage should be
2865 * combined, not added together.
2868 struct kvm_pit_state ps;
2869 struct kvm_pit_state2 ps2;
2870 struct kvm_memory_alias alias;
2871 struct kvm_pit_config pit_config;
2875 case KVM_SET_TSS_ADDR:
2876 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2880 case KVM_SET_IDENTITY_MAP_ADDR: {
2884 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2886 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2891 case KVM_SET_MEMORY_REGION: {
2892 struct kvm_memory_region kvm_mem;
2893 struct kvm_userspace_memory_region kvm_userspace_mem;
2896 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2898 kvm_userspace_mem.slot = kvm_mem.slot;
2899 kvm_userspace_mem.flags = kvm_mem.flags;
2900 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2901 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2902 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2907 case KVM_SET_NR_MMU_PAGES:
2908 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2912 case KVM_GET_NR_MMU_PAGES:
2913 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2915 case KVM_SET_MEMORY_ALIAS:
2917 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2919 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2923 case KVM_CREATE_IRQCHIP: {
2924 struct kvm_pic *vpic;
2926 mutex_lock(&kvm->lock);
2929 goto create_irqchip_unlock;
2931 vpic = kvm_create_pic(kvm);
2933 r = kvm_ioapic_init(kvm);
2935 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
2938 goto create_irqchip_unlock;
2941 goto create_irqchip_unlock;
2943 kvm->arch.vpic = vpic;
2945 r = kvm_setup_default_irq_routing(kvm);
2947 mutex_lock(&kvm->irq_lock);
2948 kvm_ioapic_destroy(kvm);
2949 kvm_destroy_pic(kvm);
2950 mutex_unlock(&kvm->irq_lock);
2952 create_irqchip_unlock:
2953 mutex_unlock(&kvm->lock);
2956 case KVM_CREATE_PIT:
2957 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2959 case KVM_CREATE_PIT2:
2961 if (copy_from_user(&u.pit_config, argp,
2962 sizeof(struct kvm_pit_config)))
2965 mutex_lock(&kvm->slots_lock);
2968 goto create_pit_unlock;
2970 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2974 mutex_unlock(&kvm->slots_lock);
2976 case KVM_IRQ_LINE_STATUS:
2977 case KVM_IRQ_LINE: {
2978 struct kvm_irq_level irq_event;
2981 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2984 if (irqchip_in_kernel(kvm)) {
2986 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2987 irq_event.irq, irq_event.level);
2988 if (ioctl == KVM_IRQ_LINE_STATUS) {
2990 irq_event.status = status;
2991 if (copy_to_user(argp, &irq_event,
2999 case KVM_GET_IRQCHIP: {
3000 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3001 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3007 if (copy_from_user(chip, argp, sizeof *chip))
3008 goto get_irqchip_out;
3010 if (!irqchip_in_kernel(kvm))
3011 goto get_irqchip_out;
3012 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3014 goto get_irqchip_out;
3016 if (copy_to_user(argp, chip, sizeof *chip))
3017 goto get_irqchip_out;
3025 case KVM_SET_IRQCHIP: {
3026 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3027 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3033 if (copy_from_user(chip, argp, sizeof *chip))
3034 goto set_irqchip_out;
3036 if (!irqchip_in_kernel(kvm))
3037 goto set_irqchip_out;
3038 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3040 goto set_irqchip_out;
3050 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3053 if (!kvm->arch.vpit)
3055 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3059 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3066 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3069 if (!kvm->arch.vpit)
3071 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3077 case KVM_GET_PIT2: {
3079 if (!kvm->arch.vpit)
3081 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3085 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3090 case KVM_SET_PIT2: {
3092 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3095 if (!kvm->arch.vpit)
3097 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3103 case KVM_REINJECT_CONTROL: {
3104 struct kvm_reinject_control control;
3106 if (copy_from_user(&control, argp, sizeof(control)))
3108 r = kvm_vm_ioctl_reinject(kvm, &control);
3114 case KVM_XEN_HVM_CONFIG: {
3116 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3117 sizeof(struct kvm_xen_hvm_config)))
3120 if (kvm->arch.xen_hvm_config.flags)
3125 case KVM_SET_CLOCK: {
3126 struct timespec now;
3127 struct kvm_clock_data user_ns;
3132 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3141 now_ns = timespec_to_ns(&now);
3142 delta = user_ns.clock - now_ns;
3143 kvm->arch.kvmclock_offset = delta;
3146 case KVM_GET_CLOCK: {
3147 struct timespec now;
3148 struct kvm_clock_data user_ns;
3152 now_ns = timespec_to_ns(&now);
3153 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3157 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3170 static void kvm_init_msr_list(void)
3175 /* skip the first msrs in the list. KVM-specific */
3176 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3177 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3180 msrs_to_save[j] = msrs_to_save[i];
3183 num_msrs_to_save = j;
3186 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3189 if (vcpu->arch.apic &&
3190 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3193 return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3196 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3198 if (vcpu->arch.apic &&
3199 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3202 return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3205 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3206 struct kvm_segment *var, int seg)
3208 kvm_x86_ops->set_segment(vcpu, var, seg);
3211 void kvm_get_segment(struct kvm_vcpu *vcpu,
3212 struct kvm_segment *var, int seg)
3214 kvm_x86_ops->get_segment(vcpu, var, seg);
3217 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3219 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3220 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3223 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3225 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3226 access |= PFERR_FETCH_MASK;
3227 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3230 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3232 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3233 access |= PFERR_WRITE_MASK;
3234 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3237 /* uses this to access any guest's mapped memory without checking CPL */
3238 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3240 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, 0, error);
3243 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3244 struct kvm_vcpu *vcpu, u32 access,
3248 int r = X86EMUL_CONTINUE;
3251 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, access, error);
3252 unsigned offset = addr & (PAGE_SIZE-1);
3253 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3256 if (gpa == UNMAPPED_GVA) {
3257 r = X86EMUL_PROPAGATE_FAULT;
3260 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3262 r = X86EMUL_UNHANDLEABLE;
3274 /* used for instruction fetching */
3275 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3276 struct kvm_vcpu *vcpu, u32 *error)
3278 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3279 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3280 access | PFERR_FETCH_MASK, error);
3283 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3284 struct kvm_vcpu *vcpu, u32 *error)
3286 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3287 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3291 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3292 struct kvm_vcpu *vcpu, u32 *error)
3294 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3297 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3299 struct kvm_vcpu *vcpu,
3303 int r = X86EMUL_CONTINUE;
3306 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr,
3307 PFERR_WRITE_MASK, error);
3308 unsigned offset = addr & (PAGE_SIZE-1);
3309 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3312 if (gpa == UNMAPPED_GVA) {
3313 r = X86EMUL_PROPAGATE_FAULT;
3316 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3318 r = X86EMUL_UNHANDLEABLE;
3330 static int emulator_read_emulated(unsigned long addr,
3333 struct kvm_vcpu *vcpu)
3338 if (vcpu->mmio_read_completed) {
3339 memcpy(val, vcpu->mmio_data, bytes);
3340 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3341 vcpu->mmio_phys_addr, *(u64 *)val);
3342 vcpu->mmio_read_completed = 0;
3343 return X86EMUL_CONTINUE;
3346 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, &error_code);
3348 if (gpa == UNMAPPED_GVA) {
3349 kvm_inject_page_fault(vcpu, addr, error_code);
3350 return X86EMUL_PROPAGATE_FAULT;
3353 /* For APIC access vmexit */
3354 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3357 if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3358 == X86EMUL_CONTINUE)
3359 return X86EMUL_CONTINUE;
3363 * Is this MMIO handled locally?
3365 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3366 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3367 return X86EMUL_CONTINUE;
3370 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3372 vcpu->mmio_needed = 1;
3373 vcpu->mmio_phys_addr = gpa;
3374 vcpu->mmio_size = bytes;
3375 vcpu->mmio_is_write = 0;
3377 return X86EMUL_UNHANDLEABLE;
3380 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3381 const void *val, int bytes)
3385 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3388 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3392 static int emulator_write_emulated_onepage(unsigned long addr,
3395 struct kvm_vcpu *vcpu)
3400 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, &error_code);
3402 if (gpa == UNMAPPED_GVA) {
3403 kvm_inject_page_fault(vcpu, addr, error_code);
3404 return X86EMUL_PROPAGATE_FAULT;
3407 /* For APIC access vmexit */
3408 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3411 if (emulator_write_phys(vcpu, gpa, val, bytes))
3412 return X86EMUL_CONTINUE;
3415 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3417 * Is this MMIO handled locally?
3419 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3420 return X86EMUL_CONTINUE;
3422 vcpu->mmio_needed = 1;
3423 vcpu->mmio_phys_addr = gpa;
3424 vcpu->mmio_size = bytes;
3425 vcpu->mmio_is_write = 1;
3426 memcpy(vcpu->mmio_data, val, bytes);
3428 return X86EMUL_CONTINUE;
3431 int emulator_write_emulated(unsigned long addr,
3434 struct kvm_vcpu *vcpu)
3436 /* Crossing a page boundary? */
3437 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3440 now = -addr & ~PAGE_MASK;
3441 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3442 if (rc != X86EMUL_CONTINUE)
3448 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3450 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3452 #define CMPXCHG_TYPE(t, ptr, old, new) \
3453 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3455 #ifdef CONFIG_X86_64
3456 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3458 # define CMPXCHG64(ptr, old, new) \
3459 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3462 static int emulator_cmpxchg_emulated(unsigned long addr,
3466 struct kvm_vcpu *vcpu)
3473 /* guests cmpxchg8b have to be emulated atomically */
3474 if (bytes > 8 || (bytes & (bytes - 1)))
3477 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3479 if (gpa == UNMAPPED_GVA ||
3480 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3483 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3486 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3488 kaddr = kmap_atomic(page, KM_USER0);
3489 kaddr += offset_in_page(gpa);
3492 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3495 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3498 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3501 exchanged = CMPXCHG64(kaddr, old, new);
3506 kunmap_atomic(kaddr, KM_USER0);
3507 kvm_release_page_dirty(page);
3510 return X86EMUL_CMPXCHG_FAILED;
3512 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3514 return X86EMUL_CONTINUE;
3517 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3519 return emulator_write_emulated(addr, new, bytes, vcpu);
3522 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3524 /* TODO: String I/O for in kernel device */
3527 if (vcpu->arch.pio.in)
3528 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3529 vcpu->arch.pio.size, pd);
3531 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3532 vcpu->arch.pio.port, vcpu->arch.pio.size,
3538 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
3539 unsigned int count, struct kvm_vcpu *vcpu)
3541 if (vcpu->arch.pio.count)
3544 trace_kvm_pio(1, port, size, 1);
3546 vcpu->arch.pio.port = port;
3547 vcpu->arch.pio.in = 1;
3548 vcpu->arch.pio.count = count;
3549 vcpu->arch.pio.size = size;
3551 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3553 memcpy(val, vcpu->arch.pio_data, size * count);
3554 vcpu->arch.pio.count = 0;
3558 vcpu->run->exit_reason = KVM_EXIT_IO;
3559 vcpu->run->io.direction = KVM_EXIT_IO_IN;
3560 vcpu->run->io.size = size;
3561 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3562 vcpu->run->io.count = count;
3563 vcpu->run->io.port = port;
3568 static int emulator_pio_out_emulated(int size, unsigned short port,
3569 const void *val, unsigned int count,
3570 struct kvm_vcpu *vcpu)
3572 trace_kvm_pio(0, port, size, 1);
3574 vcpu->arch.pio.port = port;
3575 vcpu->arch.pio.in = 0;
3576 vcpu->arch.pio.count = count;
3577 vcpu->arch.pio.size = size;
3579 memcpy(vcpu->arch.pio_data, val, size * count);
3581 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3582 vcpu->arch.pio.count = 0;
3586 vcpu->run->exit_reason = KVM_EXIT_IO;
3587 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
3588 vcpu->run->io.size = size;
3589 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3590 vcpu->run->io.count = count;
3591 vcpu->run->io.port = port;
3596 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3598 return kvm_x86_ops->get_segment_base(vcpu, seg);
3601 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3603 kvm_mmu_invlpg(vcpu, address);
3604 return X86EMUL_CONTINUE;
3607 int emulate_clts(struct kvm_vcpu *vcpu)
3609 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3610 kvm_x86_ops->fpu_activate(vcpu);
3611 return X86EMUL_CONTINUE;
3614 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3616 return kvm_get_dr(ctxt->vcpu, dr, dest);
3619 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3621 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3623 return kvm_set_dr(ctxt->vcpu, dr, value & mask);
3626 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3629 unsigned long rip = kvm_rip_read(vcpu);
3630 unsigned long rip_linear;
3632 if (!printk_ratelimit())
3635 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3637 kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu, NULL);
3639 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3640 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3642 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3644 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3646 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3649 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
3651 unsigned long value;
3655 value = kvm_read_cr0(vcpu);
3658 value = vcpu->arch.cr2;
3661 value = vcpu->arch.cr3;
3664 value = kvm_read_cr4(vcpu);
3667 value = kvm_get_cr8(vcpu);
3670 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3677 static void emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
3681 kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
3684 vcpu->arch.cr2 = val;
3687 kvm_set_cr3(vcpu, val);
3690 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
3693 kvm_set_cr8(vcpu, val & 0xfUL);
3696 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3700 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
3702 return kvm_x86_ops->get_cpl(vcpu);
3705 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
3707 kvm_x86_ops->get_gdt(vcpu, dt);
3710 static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
3711 struct kvm_vcpu *vcpu)
3713 struct kvm_segment var;
3715 kvm_get_segment(vcpu, &var, seg);
3722 set_desc_limit(desc, var.limit);
3723 set_desc_base(desc, (unsigned long)var.base);
3724 desc->type = var.type;
3726 desc->dpl = var.dpl;
3727 desc->p = var.present;
3728 desc->avl = var.avl;
3736 static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg,
3737 struct kvm_vcpu *vcpu)
3739 struct kvm_segment var;
3741 /* needed to preserve selector */
3742 kvm_get_segment(vcpu, &var, seg);
3744 var.base = get_desc_base(desc);
3745 var.limit = get_desc_limit(desc);
3747 var.limit = (var.limit << 12) | 0xfff;
3748 var.type = desc->type;
3749 var.present = desc->p;
3750 var.dpl = desc->dpl;
3755 var.avl = desc->avl;
3756 var.present = desc->p;
3757 var.unusable = !var.present;
3760 kvm_set_segment(vcpu, &var, seg);
3764 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
3766 struct kvm_segment kvm_seg;
3768 kvm_get_segment(vcpu, &kvm_seg, seg);
3769 return kvm_seg.selector;
3772 static void emulator_set_segment_selector(u16 sel, int seg,
3773 struct kvm_vcpu *vcpu)
3775 struct kvm_segment kvm_seg;
3777 kvm_get_segment(vcpu, &kvm_seg, seg);
3778 kvm_seg.selector = sel;
3779 kvm_set_segment(vcpu, &kvm_seg, seg);
3782 static void emulator_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
3784 kvm_x86_ops->set_rflags(vcpu, rflags);
3787 static struct x86_emulate_ops emulate_ops = {
3788 .read_std = kvm_read_guest_virt_system,
3789 .write_std = kvm_write_guest_virt_system,
3790 .fetch = kvm_fetch_guest_virt,
3791 .read_emulated = emulator_read_emulated,
3792 .write_emulated = emulator_write_emulated,
3793 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3794 .pio_in_emulated = emulator_pio_in_emulated,
3795 .pio_out_emulated = emulator_pio_out_emulated,
3796 .get_cached_descriptor = emulator_get_cached_descriptor,
3797 .set_cached_descriptor = emulator_set_cached_descriptor,
3798 .get_segment_selector = emulator_get_segment_selector,
3799 .set_segment_selector = emulator_set_segment_selector,
3800 .get_gdt = emulator_get_gdt,
3801 .get_cr = emulator_get_cr,
3802 .set_cr = emulator_set_cr,
3803 .cpl = emulator_get_cpl,
3804 .set_rflags = emulator_set_rflags,
3807 static void cache_all_regs(struct kvm_vcpu *vcpu)
3809 kvm_register_read(vcpu, VCPU_REGS_RAX);
3810 kvm_register_read(vcpu, VCPU_REGS_RSP);
3811 kvm_register_read(vcpu, VCPU_REGS_RIP);
3812 vcpu->arch.regs_dirty = ~0;
3815 int emulate_instruction(struct kvm_vcpu *vcpu,
3821 struct decode_cache *c;
3822 struct kvm_run *run = vcpu->run;
3824 kvm_clear_exception_queue(vcpu);
3825 vcpu->arch.mmio_fault_cr2 = cr2;
3827 * TODO: fix emulate.c to use guest_read/write_register
3828 * instead of direct ->regs accesses, can save hundred cycles
3829 * on Intel for instructions that don't read/change RSP, for
3832 cache_all_regs(vcpu);
3834 vcpu->mmio_is_write = 0;
3836 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3838 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3840 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3841 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
3842 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
3843 vcpu->arch.emulate_ctxt.mode =
3844 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
3845 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3846 ? X86EMUL_MODE_VM86 : cs_l
3847 ? X86EMUL_MODE_PROT64 : cs_db
3848 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3850 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3851 trace_kvm_emulate_insn_start(vcpu);
3853 /* Only allow emulation of specific instructions on #UD
3854 * (namely VMMCALL, sysenter, sysexit, syscall)*/
3855 c = &vcpu->arch.emulate_ctxt.decode;
3856 if (emulation_type & EMULTYPE_TRAP_UD) {
3858 return EMULATE_FAIL;
3860 case 0x01: /* VMMCALL */
3861 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3862 return EMULATE_FAIL;
3864 case 0x34: /* sysenter */
3865 case 0x35: /* sysexit */
3866 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3867 return EMULATE_FAIL;
3869 case 0x05: /* syscall */
3870 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3871 return EMULATE_FAIL;
3874 return EMULATE_FAIL;
3877 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3878 return EMULATE_FAIL;
3881 ++vcpu->stat.insn_emulation;
3883 ++vcpu->stat.insn_emulation_fail;
3884 trace_kvm_emulate_insn_failed(vcpu);
3885 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3886 return EMULATE_DONE;
3887 return EMULATE_FAIL;
3891 if (emulation_type & EMULTYPE_SKIP) {
3892 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3893 return EMULATE_DONE;
3897 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3898 shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3901 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3903 if (vcpu->arch.pio.count) {
3904 if (!vcpu->arch.pio.in)
3905 vcpu->arch.pio.count = 0;
3906 return EMULATE_DO_MMIO;
3909 if (r || vcpu->mmio_is_write) {
3910 run->exit_reason = KVM_EXIT_MMIO;
3911 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3912 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3913 run->mmio.len = vcpu->mmio_size;
3914 run->mmio.is_write = vcpu->mmio_is_write;
3918 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3920 if (!vcpu->mmio_needed) {
3921 ++vcpu->stat.insn_emulation_fail;
3922 trace_kvm_emulate_insn_failed(vcpu);
3923 kvm_report_emulation_failure(vcpu, "mmio");
3924 return EMULATE_FAIL;
3926 return EMULATE_DO_MMIO;
3929 if (vcpu->mmio_is_write) {
3930 vcpu->mmio_needed = 0;
3931 return EMULATE_DO_MMIO;
3935 if (vcpu->arch.exception.pending)
3936 vcpu->arch.emulate_ctxt.restart = false;
3938 if (vcpu->arch.emulate_ctxt.restart)
3941 return EMULATE_DONE;
3943 EXPORT_SYMBOL_GPL(emulate_instruction);
3945 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
3947 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3948 int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
3949 /* do not return to emulator after return from userspace */
3950 vcpu->arch.pio.count = 0;
3953 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
3955 static void bounce_off(void *info)
3960 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3963 struct cpufreq_freqs *freq = data;
3965 struct kvm_vcpu *vcpu;
3966 int i, send_ipi = 0;
3968 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3970 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3972 per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3974 spin_lock(&kvm_lock);
3975 list_for_each_entry(kvm, &vm_list, vm_list) {
3976 kvm_for_each_vcpu(i, vcpu, kvm) {
3977 if (vcpu->cpu != freq->cpu)
3979 if (!kvm_request_guest_time_update(vcpu))
3981 if (vcpu->cpu != smp_processor_id())
3985 spin_unlock(&kvm_lock);
3987 if (freq->old < freq->new && send_ipi) {
3989 * We upscale the frequency. Must make the guest
3990 * doesn't see old kvmclock values while running with
3991 * the new frequency, otherwise we risk the guest sees
3992 * time go backwards.
3994 * In case we update the frequency for another cpu
3995 * (which might be in guest context) send an interrupt
3996 * to kick the cpu out of guest context. Next time
3997 * guest context is entered kvmclock will be updated,
3998 * so the guest will not see stale values.
4000 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
4005 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4006 .notifier_call = kvmclock_cpufreq_notifier
4009 static void kvm_timer_init(void)
4013 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4014 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4015 CPUFREQ_TRANSITION_NOTIFIER);
4016 for_each_online_cpu(cpu) {
4017 unsigned long khz = cpufreq_get(cpu);
4020 per_cpu(cpu_tsc_khz, cpu) = khz;
4023 for_each_possible_cpu(cpu)
4024 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
4028 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4030 static int kvm_is_in_guest(void)
4032 return percpu_read(current_vcpu) != NULL;
4035 static int kvm_is_user_mode(void)
4039 if (percpu_read(current_vcpu))
4040 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4042 return user_mode != 0;
4045 static unsigned long kvm_get_guest_ip(void)
4047 unsigned long ip = 0;
4049 if (percpu_read(current_vcpu))
4050 ip = kvm_rip_read(percpu_read(current_vcpu));
4055 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4056 .is_in_guest = kvm_is_in_guest,
4057 .is_user_mode = kvm_is_user_mode,
4058 .get_guest_ip = kvm_get_guest_ip,
4061 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4063 percpu_write(current_vcpu, vcpu);
4065 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4067 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4069 percpu_write(current_vcpu, NULL);
4071 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4073 int kvm_arch_init(void *opaque)
4076 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4079 printk(KERN_ERR "kvm: already loaded the other module\n");
4084 if (!ops->cpu_has_kvm_support()) {
4085 printk(KERN_ERR "kvm: no hardware support\n");
4089 if (ops->disabled_by_bios()) {
4090 printk(KERN_ERR "kvm: disabled by bios\n");
4095 r = kvm_mmu_module_init();
4099 kvm_init_msr_list();
4102 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4103 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
4104 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4105 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4109 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4117 void kvm_arch_exit(void)
4119 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4121 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4122 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4123 CPUFREQ_TRANSITION_NOTIFIER);
4125 kvm_mmu_module_exit();
4128 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4130 ++vcpu->stat.halt_exits;
4131 if (irqchip_in_kernel(vcpu->kvm)) {
4132 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4135 vcpu->run->exit_reason = KVM_EXIT_HLT;
4139 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4141 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4144 if (is_long_mode(vcpu))
4147 return a0 | ((gpa_t)a1 << 32);
4150 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4152 u64 param, ingpa, outgpa, ret;
4153 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4154 bool fast, longmode;
4158 * hypercall generates UD from non zero cpl and real mode
4161 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4162 kvm_queue_exception(vcpu, UD_VECTOR);
4166 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4167 longmode = is_long_mode(vcpu) && cs_l == 1;
4170 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4171 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4172 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4173 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4174 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4175 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4177 #ifdef CONFIG_X86_64
4179 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4180 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4181 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4185 code = param & 0xffff;
4186 fast = (param >> 16) & 0x1;
4187 rep_cnt = (param >> 32) & 0xfff;
4188 rep_idx = (param >> 48) & 0xfff;
4190 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4193 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4194 kvm_vcpu_on_spin(vcpu);
4197 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4201 ret = res | (((u64)rep_done & 0xfff) << 32);
4203 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4205 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4206 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4212 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4214 unsigned long nr, a0, a1, a2, a3, ret;
4217 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4218 return kvm_hv_hypercall(vcpu);
4220 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4221 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4222 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4223 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4224 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4226 trace_kvm_hypercall(nr, a0, a1, a2, a3);
4228 if (!is_long_mode(vcpu)) {
4236 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4242 case KVM_HC_VAPIC_POLL_IRQ:
4246 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4253 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4254 ++vcpu->stat.hypercalls;
4257 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4259 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4261 char instruction[3];
4262 unsigned long rip = kvm_rip_read(vcpu);
4265 * Blow out the MMU to ensure that no other VCPU has an active mapping
4266 * to ensure that the updated hypercall appears atomically across all
4269 kvm_mmu_zap_all(vcpu->kvm);
4271 kvm_x86_ops->patch_hypercall(vcpu, instruction);
4273 return emulator_write_emulated(rip, instruction, 3, vcpu);
4276 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4278 struct desc_ptr dt = { limit, base };
4280 kvm_x86_ops->set_gdt(vcpu, &dt);
4283 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4285 struct desc_ptr dt = { limit, base };
4287 kvm_x86_ops->set_idt(vcpu, &dt);
4290 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4292 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4293 int j, nent = vcpu->arch.cpuid_nent;
4295 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4296 /* when no next entry is found, the current entry[i] is reselected */
4297 for (j = i + 1; ; j = (j + 1) % nent) {
4298 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4299 if (ej->function == e->function) {
4300 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4304 return 0; /* silence gcc, even though control never reaches here */
4307 /* find an entry with matching function, matching index (if needed), and that
4308 * should be read next (if it's stateful) */
4309 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4310 u32 function, u32 index)
4312 if (e->function != function)
4314 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4316 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4317 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4322 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4323 u32 function, u32 index)
4326 struct kvm_cpuid_entry2 *best = NULL;
4328 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4329 struct kvm_cpuid_entry2 *e;
4331 e = &vcpu->arch.cpuid_entries[i];
4332 if (is_matching_cpuid_entry(e, function, index)) {
4333 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4334 move_to_next_stateful_cpuid_entry(vcpu, i);
4339 * Both basic or both extended?
4341 if (((e->function ^ function) & 0x80000000) == 0)
4342 if (!best || e->function > best->function)
4347 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4349 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4351 struct kvm_cpuid_entry2 *best;
4353 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
4354 if (!best || best->eax < 0x80000008)
4356 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4358 return best->eax & 0xff;
4363 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4365 u32 function, index;
4366 struct kvm_cpuid_entry2 *best;
4368 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4369 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4370 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4371 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4372 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4373 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4374 best = kvm_find_cpuid_entry(vcpu, function, index);
4376 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4377 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4378 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4379 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4381 kvm_x86_ops->skip_emulated_instruction(vcpu);
4382 trace_kvm_cpuid(function,
4383 kvm_register_read(vcpu, VCPU_REGS_RAX),
4384 kvm_register_read(vcpu, VCPU_REGS_RBX),
4385 kvm_register_read(vcpu, VCPU_REGS_RCX),
4386 kvm_register_read(vcpu, VCPU_REGS_RDX));
4388 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4391 * Check if userspace requested an interrupt window, and that the
4392 * interrupt window is open.
4394 * No need to exit to userspace if we already have an interrupt queued.
4396 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4398 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4399 vcpu->run->request_interrupt_window &&
4400 kvm_arch_interrupt_allowed(vcpu));
4403 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4405 struct kvm_run *kvm_run = vcpu->run;
4407 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4408 kvm_run->cr8 = kvm_get_cr8(vcpu);
4409 kvm_run->apic_base = kvm_get_apic_base(vcpu);
4410 if (irqchip_in_kernel(vcpu->kvm))
4411 kvm_run->ready_for_interrupt_injection = 1;
4413 kvm_run->ready_for_interrupt_injection =
4414 kvm_arch_interrupt_allowed(vcpu) &&
4415 !kvm_cpu_has_interrupt(vcpu) &&
4416 !kvm_event_needs_reinjection(vcpu);
4419 static void vapic_enter(struct kvm_vcpu *vcpu)
4421 struct kvm_lapic *apic = vcpu->arch.apic;
4424 if (!apic || !apic->vapic_addr)
4427 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4429 vcpu->arch.apic->vapic_page = page;
4432 static void vapic_exit(struct kvm_vcpu *vcpu)
4434 struct kvm_lapic *apic = vcpu->arch.apic;
4437 if (!apic || !apic->vapic_addr)
4440 idx = srcu_read_lock(&vcpu->kvm->srcu);
4441 kvm_release_page_dirty(apic->vapic_page);
4442 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4443 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4446 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4450 if (!kvm_x86_ops->update_cr8_intercept)
4453 if (!vcpu->arch.apic)
4456 if (!vcpu->arch.apic->vapic_addr)
4457 max_irr = kvm_lapic_find_highest_irr(vcpu);
4464 tpr = kvm_lapic_get_cr8(vcpu);
4466 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4469 static void inject_pending_event(struct kvm_vcpu *vcpu)
4471 /* try to reinject previous events if any */
4472 if (vcpu->arch.exception.pending) {
4473 trace_kvm_inj_exception(vcpu->arch.exception.nr,
4474 vcpu->arch.exception.has_error_code,
4475 vcpu->arch.exception.error_code);
4476 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4477 vcpu->arch.exception.has_error_code,
4478 vcpu->arch.exception.error_code,
4479 vcpu->arch.exception.reinject);
4483 if (vcpu->arch.nmi_injected) {
4484 kvm_x86_ops->set_nmi(vcpu);
4488 if (vcpu->arch.interrupt.pending) {
4489 kvm_x86_ops->set_irq(vcpu);
4493 /* try to inject new event if pending */
4494 if (vcpu->arch.nmi_pending) {
4495 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4496 vcpu->arch.nmi_pending = false;
4497 vcpu->arch.nmi_injected = true;
4498 kvm_x86_ops->set_nmi(vcpu);
4500 } else if (kvm_cpu_has_interrupt(vcpu)) {
4501 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4502 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4504 kvm_x86_ops->set_irq(vcpu);
4509 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4512 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4513 vcpu->run->request_interrupt_window;
4516 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4517 kvm_mmu_unload(vcpu);
4519 r = kvm_mmu_reload(vcpu);
4523 if (vcpu->requests) {
4524 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4525 __kvm_migrate_timers(vcpu);
4526 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4527 kvm_write_guest_time(vcpu);
4528 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4529 kvm_mmu_sync_roots(vcpu);
4530 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4531 kvm_x86_ops->tlb_flush(vcpu);
4532 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4534 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4538 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4539 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4543 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4544 vcpu->fpu_active = 0;
4545 kvm_x86_ops->fpu_deactivate(vcpu);
4551 kvm_x86_ops->prepare_guest_switch(vcpu);
4552 if (vcpu->fpu_active)
4553 kvm_load_guest_fpu(vcpu);
4555 local_irq_disable();
4557 clear_bit(KVM_REQ_KICK, &vcpu->requests);
4558 smp_mb__after_clear_bit();
4560 if (vcpu->requests || need_resched() || signal_pending(current)) {
4561 set_bit(KVM_REQ_KICK, &vcpu->requests);
4568 inject_pending_event(vcpu);
4570 /* enable NMI/IRQ window open exits if needed */
4571 if (vcpu->arch.nmi_pending)
4572 kvm_x86_ops->enable_nmi_window(vcpu);
4573 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4574 kvm_x86_ops->enable_irq_window(vcpu);
4576 if (kvm_lapic_enabled(vcpu)) {
4577 update_cr8_intercept(vcpu);
4578 kvm_lapic_sync_to_vapic(vcpu);
4581 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4585 if (unlikely(vcpu->arch.switch_db_regs)) {
4587 set_debugreg(vcpu->arch.eff_db[0], 0);
4588 set_debugreg(vcpu->arch.eff_db[1], 1);
4589 set_debugreg(vcpu->arch.eff_db[2], 2);
4590 set_debugreg(vcpu->arch.eff_db[3], 3);
4593 trace_kvm_entry(vcpu->vcpu_id);
4594 kvm_x86_ops->run(vcpu);
4597 * If the guest has used debug registers, at least dr7
4598 * will be disabled while returning to the host.
4599 * If we don't have active breakpoints in the host, we don't
4600 * care about the messed up debug address registers. But if
4601 * we have some of them active, restore the old state.
4603 if (hw_breakpoint_active())
4604 hw_breakpoint_restore();
4606 set_bit(KVM_REQ_KICK, &vcpu->requests);
4612 * We must have an instruction between local_irq_enable() and
4613 * kvm_guest_exit(), so the timer interrupt isn't delayed by
4614 * the interrupt shadow. The stat.exits increment will do nicely.
4615 * But we need to prevent reordering, hence this barrier():
4623 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4626 * Profile KVM exit RIPs:
4628 if (unlikely(prof_on == KVM_PROFILING)) {
4629 unsigned long rip = kvm_rip_read(vcpu);
4630 profile_hit(KVM_PROFILING, (void *)rip);
4634 kvm_lapic_sync_from_vapic(vcpu);
4636 r = kvm_x86_ops->handle_exit(vcpu);
4642 static int __vcpu_run(struct kvm_vcpu *vcpu)
4645 struct kvm *kvm = vcpu->kvm;
4647 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4648 pr_debug("vcpu %d received sipi with vector # %x\n",
4649 vcpu->vcpu_id, vcpu->arch.sipi_vector);
4650 kvm_lapic_reset(vcpu);
4651 r = kvm_arch_vcpu_reset(vcpu);
4654 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4657 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4662 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4663 r = vcpu_enter_guest(vcpu);
4665 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4666 kvm_vcpu_block(vcpu);
4667 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4668 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4670 switch(vcpu->arch.mp_state) {
4671 case KVM_MP_STATE_HALTED:
4672 vcpu->arch.mp_state =
4673 KVM_MP_STATE_RUNNABLE;
4674 case KVM_MP_STATE_RUNNABLE:
4676 case KVM_MP_STATE_SIPI_RECEIVED:
4687 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4688 if (kvm_cpu_has_pending_timer(vcpu))
4689 kvm_inject_pending_timer_irqs(vcpu);
4691 if (dm_request_for_irq_injection(vcpu)) {
4693 vcpu->run->exit_reason = KVM_EXIT_INTR;
4694 ++vcpu->stat.request_irq_exits;
4696 if (signal_pending(current)) {
4698 vcpu->run->exit_reason = KVM_EXIT_INTR;
4699 ++vcpu->stat.signal_exits;
4701 if (need_resched()) {
4702 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4704 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4708 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4715 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4722 if (vcpu->sigset_active)
4723 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4725 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4726 kvm_vcpu_block(vcpu);
4727 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4732 /* re-sync apic's tpr */
4733 if (!irqchip_in_kernel(vcpu->kvm))
4734 kvm_set_cr8(vcpu, kvm_run->cr8);
4736 if (vcpu->arch.pio.count || vcpu->mmio_needed ||
4737 vcpu->arch.emulate_ctxt.restart) {
4738 if (vcpu->mmio_needed) {
4739 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4740 vcpu->mmio_read_completed = 1;
4741 vcpu->mmio_needed = 0;
4743 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4744 r = emulate_instruction(vcpu, 0, 0, EMULTYPE_NO_DECODE);
4745 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4746 if (r == EMULATE_DO_MMIO) {
4751 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4752 kvm_register_write(vcpu, VCPU_REGS_RAX,
4753 kvm_run->hypercall.ret);
4755 r = __vcpu_run(vcpu);
4758 post_kvm_run_save(vcpu);
4759 if (vcpu->sigset_active)
4760 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4766 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4770 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4771 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4772 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4773 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4774 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4775 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4776 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4777 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4778 #ifdef CONFIG_X86_64
4779 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4780 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4781 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4782 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4783 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4784 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4785 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4786 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4789 regs->rip = kvm_rip_read(vcpu);
4790 regs->rflags = kvm_get_rflags(vcpu);
4797 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4801 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4802 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4803 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4804 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4805 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4806 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4807 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4808 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4809 #ifdef CONFIG_X86_64
4810 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4811 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4812 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4813 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4814 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4815 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4816 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4817 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4820 kvm_rip_write(vcpu, regs->rip);
4821 kvm_set_rflags(vcpu, regs->rflags);
4823 vcpu->arch.exception.pending = false;
4830 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4832 struct kvm_segment cs;
4834 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4838 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4840 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4841 struct kvm_sregs *sregs)
4847 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4848 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4849 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4850 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4851 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4852 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4854 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4855 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4857 kvm_x86_ops->get_idt(vcpu, &dt);
4858 sregs->idt.limit = dt.size;
4859 sregs->idt.base = dt.address;
4860 kvm_x86_ops->get_gdt(vcpu, &dt);
4861 sregs->gdt.limit = dt.size;
4862 sregs->gdt.base = dt.address;
4864 sregs->cr0 = kvm_read_cr0(vcpu);
4865 sregs->cr2 = vcpu->arch.cr2;
4866 sregs->cr3 = vcpu->arch.cr3;
4867 sregs->cr4 = kvm_read_cr4(vcpu);
4868 sregs->cr8 = kvm_get_cr8(vcpu);
4869 sregs->efer = vcpu->arch.efer;
4870 sregs->apic_base = kvm_get_apic_base(vcpu);
4872 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4874 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4875 set_bit(vcpu->arch.interrupt.nr,
4876 (unsigned long *)sregs->interrupt_bitmap);
4883 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4884 struct kvm_mp_state *mp_state)
4887 mp_state->mp_state = vcpu->arch.mp_state;
4892 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4893 struct kvm_mp_state *mp_state)
4896 vcpu->arch.mp_state = mp_state->mp_state;
4901 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
4902 bool has_error_code, u32 error_code)
4904 int cs_db, cs_l, ret;
4905 cache_all_regs(vcpu);
4907 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4909 vcpu->arch.emulate_ctxt.vcpu = vcpu;
4910 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
4911 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4912 vcpu->arch.emulate_ctxt.mode =
4913 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4914 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4915 ? X86EMUL_MODE_VM86 : cs_l
4916 ? X86EMUL_MODE_PROT64 : cs_db
4917 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4919 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt, &emulate_ops,
4920 tss_selector, reason, has_error_code,
4924 return EMULATE_FAIL;
4926 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4927 return EMULATE_DONE;
4929 EXPORT_SYMBOL_GPL(kvm_task_switch);
4931 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4932 struct kvm_sregs *sregs)
4934 int mmu_reset_needed = 0;
4935 int pending_vec, max_bits;
4940 dt.size = sregs->idt.limit;
4941 dt.address = sregs->idt.base;
4942 kvm_x86_ops->set_idt(vcpu, &dt);
4943 dt.size = sregs->gdt.limit;
4944 dt.address = sregs->gdt.base;
4945 kvm_x86_ops->set_gdt(vcpu, &dt);
4947 vcpu->arch.cr2 = sregs->cr2;
4948 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4949 vcpu->arch.cr3 = sregs->cr3;
4951 kvm_set_cr8(vcpu, sregs->cr8);
4953 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
4954 kvm_x86_ops->set_efer(vcpu, sregs->efer);
4955 kvm_set_apic_base(vcpu, sregs->apic_base);
4957 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
4958 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4959 vcpu->arch.cr0 = sregs->cr0;
4961 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
4962 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4963 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
4964 load_pdptrs(vcpu, vcpu->arch.cr3);
4965 mmu_reset_needed = 1;
4968 if (mmu_reset_needed)
4969 kvm_mmu_reset_context(vcpu);
4971 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4972 pending_vec = find_first_bit(
4973 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4974 if (pending_vec < max_bits) {
4975 kvm_queue_interrupt(vcpu, pending_vec, false);
4976 pr_debug("Set back pending irq %d\n", pending_vec);
4977 if (irqchip_in_kernel(vcpu->kvm))
4978 kvm_pic_clear_isr_ack(vcpu->kvm);
4981 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4982 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4983 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4984 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4985 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4986 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4988 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4989 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4991 update_cr8_intercept(vcpu);
4993 /* Older userspace won't unhalt the vcpu on reset. */
4994 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4995 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4997 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5004 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5005 struct kvm_guest_debug *dbg)
5007 unsigned long rflags;
5012 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5014 if (vcpu->arch.exception.pending)
5016 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5017 kvm_queue_exception(vcpu, DB_VECTOR);
5019 kvm_queue_exception(vcpu, BP_VECTOR);
5023 * Read rflags as long as potentially injected trace flags are still
5026 rflags = kvm_get_rflags(vcpu);
5028 vcpu->guest_debug = dbg->control;
5029 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5030 vcpu->guest_debug = 0;
5032 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5033 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5034 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5035 vcpu->arch.switch_db_regs =
5036 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5038 for (i = 0; i < KVM_NR_DB_REGS; i++)
5039 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5040 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5043 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5044 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5045 get_segment_base(vcpu, VCPU_SREG_CS);
5048 * Trigger an rflags update that will inject or remove the trace
5051 kvm_set_rflags(vcpu, rflags);
5053 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5064 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
5065 * we have asm/x86/processor.h
5076 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
5077 #ifdef CONFIG_X86_64
5078 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
5080 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
5085 * Translate a guest virtual address to a guest physical address.
5087 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5088 struct kvm_translation *tr)
5090 unsigned long vaddr = tr->linear_address;
5095 idx = srcu_read_lock(&vcpu->kvm->srcu);
5096 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5097 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5098 tr->physical_address = gpa;
5099 tr->valid = gpa != UNMAPPED_GVA;
5107 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5109 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5113 memcpy(fpu->fpr, fxsave->st_space, 128);
5114 fpu->fcw = fxsave->cwd;
5115 fpu->fsw = fxsave->swd;
5116 fpu->ftwx = fxsave->twd;
5117 fpu->last_opcode = fxsave->fop;
5118 fpu->last_ip = fxsave->rip;
5119 fpu->last_dp = fxsave->rdp;
5120 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5127 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5129 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5133 memcpy(fxsave->st_space, fpu->fpr, 128);
5134 fxsave->cwd = fpu->fcw;
5135 fxsave->swd = fpu->fsw;
5136 fxsave->twd = fpu->ftwx;
5137 fxsave->fop = fpu->last_opcode;
5138 fxsave->rip = fpu->last_ip;
5139 fxsave->rdp = fpu->last_dp;
5140 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5147 void fx_init(struct kvm_vcpu *vcpu)
5149 unsigned after_mxcsr_mask;
5152 * Touch the fpu the first time in non atomic context as if
5153 * this is the first fpu instruction the exception handler
5154 * will fire before the instruction returns and it'll have to
5155 * allocate ram with GFP_KERNEL.
5158 kvm_fx_save(&vcpu->arch.host_fx_image);
5160 /* Initialize guest FPU by resetting ours and saving into guest's */
5162 kvm_fx_save(&vcpu->arch.host_fx_image);
5164 kvm_fx_save(&vcpu->arch.guest_fx_image);
5165 kvm_fx_restore(&vcpu->arch.host_fx_image);
5168 vcpu->arch.cr0 |= X86_CR0_ET;
5169 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5170 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5171 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5172 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5174 EXPORT_SYMBOL_GPL(fx_init);
5176 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5178 if (vcpu->guest_fpu_loaded)
5181 vcpu->guest_fpu_loaded = 1;
5182 kvm_fx_save(&vcpu->arch.host_fx_image);
5183 kvm_fx_restore(&vcpu->arch.guest_fx_image);
5187 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5189 if (!vcpu->guest_fpu_loaded)
5192 vcpu->guest_fpu_loaded = 0;
5193 kvm_fx_save(&vcpu->arch.guest_fx_image);
5194 kvm_fx_restore(&vcpu->arch.host_fx_image);
5195 ++vcpu->stat.fpu_reload;
5196 set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5200 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5202 if (vcpu->arch.time_page) {
5203 kvm_release_page_dirty(vcpu->arch.time_page);
5204 vcpu->arch.time_page = NULL;
5207 kvm_x86_ops->vcpu_free(vcpu);
5210 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5213 return kvm_x86_ops->vcpu_create(kvm, id);
5216 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5220 /* We do fxsave: this must be aligned. */
5221 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5223 vcpu->arch.mtrr_state.have_fixed = 1;
5225 r = kvm_arch_vcpu_reset(vcpu);
5227 r = kvm_mmu_setup(vcpu);
5234 kvm_x86_ops->vcpu_free(vcpu);
5238 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5241 kvm_mmu_unload(vcpu);
5244 kvm_x86_ops->vcpu_free(vcpu);
5247 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5249 vcpu->arch.nmi_pending = false;
5250 vcpu->arch.nmi_injected = false;
5252 vcpu->arch.switch_db_regs = 0;
5253 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5254 vcpu->arch.dr6 = DR6_FIXED_1;
5255 vcpu->arch.dr7 = DR7_FIXED_1;
5257 return kvm_x86_ops->vcpu_reset(vcpu);
5260 int kvm_arch_hardware_enable(void *garbage)
5263 * Since this may be called from a hotplug notifcation,
5264 * we can't get the CPU frequency directly.
5266 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5267 int cpu = raw_smp_processor_id();
5268 per_cpu(cpu_tsc_khz, cpu) = 0;
5271 kvm_shared_msr_cpu_online();
5273 return kvm_x86_ops->hardware_enable(garbage);
5276 void kvm_arch_hardware_disable(void *garbage)
5278 kvm_x86_ops->hardware_disable(garbage);
5279 drop_user_return_notifiers(garbage);
5282 int kvm_arch_hardware_setup(void)
5284 return kvm_x86_ops->hardware_setup();
5287 void kvm_arch_hardware_unsetup(void)
5289 kvm_x86_ops->hardware_unsetup();
5292 void kvm_arch_check_processor_compat(void *rtn)
5294 kvm_x86_ops->check_processor_compatibility(rtn);
5297 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5303 BUG_ON(vcpu->kvm == NULL);
5306 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5307 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5308 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5310 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5312 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5317 vcpu->arch.pio_data = page_address(page);
5319 r = kvm_mmu_create(vcpu);
5321 goto fail_free_pio_data;
5323 if (irqchip_in_kernel(kvm)) {
5324 r = kvm_create_lapic(vcpu);
5326 goto fail_mmu_destroy;
5329 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5331 if (!vcpu->arch.mce_banks) {
5333 goto fail_free_lapic;
5335 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5339 kvm_free_lapic(vcpu);
5341 kvm_mmu_destroy(vcpu);
5343 free_page((unsigned long)vcpu->arch.pio_data);
5348 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5352 kfree(vcpu->arch.mce_banks);
5353 kvm_free_lapic(vcpu);
5354 idx = srcu_read_lock(&vcpu->kvm->srcu);
5355 kvm_mmu_destroy(vcpu);
5356 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5357 free_page((unsigned long)vcpu->arch.pio_data);
5360 struct kvm *kvm_arch_create_vm(void)
5362 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5365 return ERR_PTR(-ENOMEM);
5367 kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5368 if (!kvm->arch.aliases) {
5370 return ERR_PTR(-ENOMEM);
5373 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5374 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5376 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5377 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5379 rdtscll(kvm->arch.vm_init_tsc);
5384 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5387 kvm_mmu_unload(vcpu);
5391 static void kvm_free_vcpus(struct kvm *kvm)
5394 struct kvm_vcpu *vcpu;
5397 * Unpin any mmu pages first.
5399 kvm_for_each_vcpu(i, vcpu, kvm)
5400 kvm_unload_vcpu_mmu(vcpu);
5401 kvm_for_each_vcpu(i, vcpu, kvm)
5402 kvm_arch_vcpu_free(vcpu);
5404 mutex_lock(&kvm->lock);
5405 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5406 kvm->vcpus[i] = NULL;
5408 atomic_set(&kvm->online_vcpus, 0);
5409 mutex_unlock(&kvm->lock);
5412 void kvm_arch_sync_events(struct kvm *kvm)
5414 kvm_free_all_assigned_devices(kvm);
5417 void kvm_arch_destroy_vm(struct kvm *kvm)
5419 kvm_iommu_unmap_guest(kvm);
5421 kfree(kvm->arch.vpic);
5422 kfree(kvm->arch.vioapic);
5423 kvm_free_vcpus(kvm);
5424 kvm_free_physmem(kvm);
5425 if (kvm->arch.apic_access_page)
5426 put_page(kvm->arch.apic_access_page);
5427 if (kvm->arch.ept_identity_pagetable)
5428 put_page(kvm->arch.ept_identity_pagetable);
5429 cleanup_srcu_struct(&kvm->srcu);
5430 kfree(kvm->arch.aliases);
5434 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5435 struct kvm_memory_slot *memslot,
5436 struct kvm_memory_slot old,
5437 struct kvm_userspace_memory_region *mem,
5440 int npages = memslot->npages;
5442 /*To keep backward compatibility with older userspace,
5443 *x86 needs to hanlde !user_alloc case.
5446 if (npages && !old.rmap) {
5447 unsigned long userspace_addr;
5449 down_write(¤t->mm->mmap_sem);
5450 userspace_addr = do_mmap(NULL, 0,
5452 PROT_READ | PROT_WRITE,
5453 MAP_PRIVATE | MAP_ANONYMOUS,
5455 up_write(¤t->mm->mmap_sem);
5457 if (IS_ERR((void *)userspace_addr))
5458 return PTR_ERR((void *)userspace_addr);
5460 memslot->userspace_addr = userspace_addr;
5468 void kvm_arch_commit_memory_region(struct kvm *kvm,
5469 struct kvm_userspace_memory_region *mem,
5470 struct kvm_memory_slot old,
5474 int npages = mem->memory_size >> PAGE_SHIFT;
5476 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5479 down_write(¤t->mm->mmap_sem);
5480 ret = do_munmap(current->mm, old.userspace_addr,
5481 old.npages * PAGE_SIZE);
5482 up_write(¤t->mm->mmap_sem);
5485 "kvm_vm_ioctl_set_memory_region: "
5486 "failed to munmap memory\n");
5489 spin_lock(&kvm->mmu_lock);
5490 if (!kvm->arch.n_requested_mmu_pages) {
5491 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5492 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5495 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5496 spin_unlock(&kvm->mmu_lock);
5499 void kvm_arch_flush_shadow(struct kvm *kvm)
5501 kvm_mmu_zap_all(kvm);
5502 kvm_reload_remote_mmus(kvm);
5505 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5507 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5508 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5509 || vcpu->arch.nmi_pending ||
5510 (kvm_arch_interrupt_allowed(vcpu) &&
5511 kvm_cpu_has_interrupt(vcpu));
5514 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5517 int cpu = vcpu->cpu;
5519 if (waitqueue_active(&vcpu->wq)) {
5520 wake_up_interruptible(&vcpu->wq);
5521 ++vcpu->stat.halt_wakeup;
5525 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5526 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5527 smp_send_reschedule(cpu);
5531 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5533 return kvm_x86_ops->interrupt_allowed(vcpu);
5536 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
5538 unsigned long current_rip = kvm_rip_read(vcpu) +
5539 get_segment_base(vcpu, VCPU_SREG_CS);
5541 return current_rip == linear_rip;
5543 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
5545 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5547 unsigned long rflags;
5549 rflags = kvm_x86_ops->get_rflags(vcpu);
5550 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5551 rflags &= ~X86_EFLAGS_TF;
5554 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5556 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5558 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5559 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
5560 rflags |= X86_EFLAGS_TF;
5561 kvm_x86_ops->set_rflags(vcpu, rflags);
5563 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5565 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5566 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5567 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5568 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5569 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5570 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5571 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5572 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5573 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5574 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5575 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
5576 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);