2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
9 * Avi Kivity <avi@qumranet.com>
10 * Yaniv Kamay <yaniv@qumranet.com>
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
17 #include <linux/kvm_host.h>
18 #include "segment_descriptor.h"
22 #include <linux/kvm.h>
24 #include <linux/vmalloc.h>
25 #include <linux/module.h>
26 #include <linux/mman.h>
27 #include <linux/highmem.h>
29 #include <asm/uaccess.h>
32 #define MAX_IO_MSRS 256
33 #define CR0_RESERVED_BITS \
34 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
35 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
36 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
37 #define CR4_RESERVED_BITS \
38 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
39 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
40 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
41 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
43 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
44 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
46 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
47 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
49 struct kvm_x86_ops *kvm_x86_ops;
51 struct kvm_stats_debugfs_item debugfs_entries[] = {
52 { "pf_fixed", VCPU_STAT(pf_fixed) },
53 { "pf_guest", VCPU_STAT(pf_guest) },
54 { "tlb_flush", VCPU_STAT(tlb_flush) },
55 { "invlpg", VCPU_STAT(invlpg) },
56 { "exits", VCPU_STAT(exits) },
57 { "io_exits", VCPU_STAT(io_exits) },
58 { "mmio_exits", VCPU_STAT(mmio_exits) },
59 { "signal_exits", VCPU_STAT(signal_exits) },
60 { "irq_window", VCPU_STAT(irq_window_exits) },
61 { "halt_exits", VCPU_STAT(halt_exits) },
62 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
63 { "request_irq", VCPU_STAT(request_irq_exits) },
64 { "irq_exits", VCPU_STAT(irq_exits) },
65 { "host_state_reload", VCPU_STAT(host_state_reload) },
66 { "efer_reload", VCPU_STAT(efer_reload) },
67 { "fpu_reload", VCPU_STAT(fpu_reload) },
68 { "insn_emulation", VCPU_STAT(insn_emulation) },
69 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
70 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
71 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
72 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
73 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
74 { "mmu_flooded", VM_STAT(mmu_flooded) },
75 { "mmu_recycled", VM_STAT(mmu_recycled) },
76 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
81 unsigned long segment_base(u16 selector)
83 struct descriptor_table gdt;
84 struct segment_descriptor *d;
85 unsigned long table_base;
91 asm("sgdt %0" : "=m"(gdt));
92 table_base = gdt.base;
94 if (selector & 4) { /* from ldt */
97 asm("sldt %0" : "=g"(ldt_selector));
98 table_base = segment_base(ldt_selector);
100 d = (struct segment_descriptor *)(table_base + (selector & ~7));
101 v = d->base_low | ((unsigned long)d->base_mid << 16) |
102 ((unsigned long)d->base_high << 24);
104 if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
105 v |= ((unsigned long) \
106 ((struct segment_descriptor_64 *)d)->base_higher) << 32;
110 EXPORT_SYMBOL_GPL(segment_base);
112 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
114 if (irqchip_in_kernel(vcpu->kvm))
115 return vcpu->arch.apic_base;
117 return vcpu->arch.apic_base;
119 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
121 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
123 /* TODO: reserve bits check */
124 if (irqchip_in_kernel(vcpu->kvm))
125 kvm_lapic_set_base(vcpu, data);
127 vcpu->arch.apic_base = data;
129 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
131 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
133 WARN_ON(vcpu->arch.exception.pending);
134 vcpu->arch.exception.pending = true;
135 vcpu->arch.exception.has_error_code = false;
136 vcpu->arch.exception.nr = nr;
138 EXPORT_SYMBOL_GPL(kvm_queue_exception);
140 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
143 ++vcpu->stat.pf_guest;
144 if (vcpu->arch.exception.pending && vcpu->arch.exception.nr == PF_VECTOR) {
145 printk(KERN_DEBUG "kvm: inject_page_fault:"
146 " double fault 0x%lx\n", addr);
147 vcpu->arch.exception.nr = DF_VECTOR;
148 vcpu->arch.exception.error_code = 0;
151 vcpu->arch.cr2 = addr;
152 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
155 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
157 WARN_ON(vcpu->arch.exception.pending);
158 vcpu->arch.exception.pending = true;
159 vcpu->arch.exception.has_error_code = true;
160 vcpu->arch.exception.nr = nr;
161 vcpu->arch.exception.error_code = error_code;
163 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
165 static void __queue_exception(struct kvm_vcpu *vcpu)
167 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
168 vcpu->arch.exception.has_error_code,
169 vcpu->arch.exception.error_code);
173 * Load the pae pdptrs. Return true is they are all valid.
175 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
177 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
178 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
181 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
183 mutex_lock(&vcpu->kvm->lock);
184 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
185 offset * sizeof(u64), sizeof(pdpte));
190 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
191 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
198 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
200 mutex_unlock(&vcpu->kvm->lock);
205 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
207 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
211 if (is_long_mode(vcpu) || !is_pae(vcpu))
214 mutex_lock(&vcpu->kvm->lock);
215 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
218 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
220 mutex_unlock(&vcpu->kvm->lock);
225 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
227 if (cr0 & CR0_RESERVED_BITS) {
228 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
229 cr0, vcpu->arch.cr0);
230 kvm_inject_gp(vcpu, 0);
234 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
235 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
236 kvm_inject_gp(vcpu, 0);
240 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
241 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
242 "and a clear PE flag\n");
243 kvm_inject_gp(vcpu, 0);
247 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
249 if ((vcpu->arch.shadow_efer & EFER_LME)) {
253 printk(KERN_DEBUG "set_cr0: #GP, start paging "
254 "in long mode while PAE is disabled\n");
255 kvm_inject_gp(vcpu, 0);
258 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
260 printk(KERN_DEBUG "set_cr0: #GP, start paging "
261 "in long mode while CS.L == 1\n");
262 kvm_inject_gp(vcpu, 0);
268 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
269 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
271 kvm_inject_gp(vcpu, 0);
277 kvm_x86_ops->set_cr0(vcpu, cr0);
278 vcpu->arch.cr0 = cr0;
280 mutex_lock(&vcpu->kvm->lock);
281 kvm_mmu_reset_context(vcpu);
282 mutex_unlock(&vcpu->kvm->lock);
285 EXPORT_SYMBOL_GPL(set_cr0);
287 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
289 set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
291 EXPORT_SYMBOL_GPL(lmsw);
293 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
295 if (cr4 & CR4_RESERVED_BITS) {
296 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
297 kvm_inject_gp(vcpu, 0);
301 if (is_long_mode(vcpu)) {
302 if (!(cr4 & X86_CR4_PAE)) {
303 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
305 kvm_inject_gp(vcpu, 0);
308 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
309 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
310 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
311 kvm_inject_gp(vcpu, 0);
315 if (cr4 & X86_CR4_VMXE) {
316 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
317 kvm_inject_gp(vcpu, 0);
320 kvm_x86_ops->set_cr4(vcpu, cr4);
321 vcpu->arch.cr4 = cr4;
322 mutex_lock(&vcpu->kvm->lock);
323 kvm_mmu_reset_context(vcpu);
324 mutex_unlock(&vcpu->kvm->lock);
326 EXPORT_SYMBOL_GPL(set_cr4);
328 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
330 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
331 kvm_mmu_flush_tlb(vcpu);
335 if (is_long_mode(vcpu)) {
336 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
337 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
338 kvm_inject_gp(vcpu, 0);
343 if (cr3 & CR3_PAE_RESERVED_BITS) {
345 "set_cr3: #GP, reserved bits\n");
346 kvm_inject_gp(vcpu, 0);
349 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
350 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
352 kvm_inject_gp(vcpu, 0);
357 * We don't check reserved bits in nonpae mode, because
358 * this isn't enforced, and VMware depends on this.
362 mutex_lock(&vcpu->kvm->lock);
364 * Does the new cr3 value map to physical memory? (Note, we
365 * catch an invalid cr3 even in real-mode, because it would
366 * cause trouble later on when we turn on paging anyway.)
368 * A real CPU would silently accept an invalid cr3 and would
369 * attempt to use it - with largely undefined (and often hard
370 * to debug) behavior on the guest side.
372 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
373 kvm_inject_gp(vcpu, 0);
375 vcpu->arch.cr3 = cr3;
376 vcpu->arch.mmu.new_cr3(vcpu);
378 mutex_unlock(&vcpu->kvm->lock);
380 EXPORT_SYMBOL_GPL(set_cr3);
382 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
384 if (cr8 & CR8_RESERVED_BITS) {
385 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
386 kvm_inject_gp(vcpu, 0);
389 if (irqchip_in_kernel(vcpu->kvm))
390 kvm_lapic_set_tpr(vcpu, cr8);
392 vcpu->arch.cr8 = cr8;
394 EXPORT_SYMBOL_GPL(set_cr8);
396 unsigned long get_cr8(struct kvm_vcpu *vcpu)
398 if (irqchip_in_kernel(vcpu->kvm))
399 return kvm_lapic_get_cr8(vcpu);
401 return vcpu->arch.cr8;
403 EXPORT_SYMBOL_GPL(get_cr8);
406 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
407 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
409 * This list is modified at module load time to reflect the
410 * capabilities of the host cpu.
412 static u32 msrs_to_save[] = {
413 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
416 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
418 MSR_IA32_TIME_STAMP_COUNTER,
421 static unsigned num_msrs_to_save;
423 static u32 emulated_msrs[] = {
424 MSR_IA32_MISC_ENABLE,
429 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
431 if (efer & EFER_RESERVED_BITS) {
432 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
434 kvm_inject_gp(vcpu, 0);
439 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
440 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
441 kvm_inject_gp(vcpu, 0);
445 kvm_x86_ops->set_efer(vcpu, efer);
448 efer |= vcpu->arch.shadow_efer & EFER_LMA;
450 vcpu->arch.shadow_efer = efer;
456 * Writes msr value into into the appropriate "register".
457 * Returns 0 on success, non-0 otherwise.
458 * Assumes vcpu_load() was already called.
460 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
462 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
466 * Adapt set_msr() to msr_io()'s calling convention
468 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
470 return kvm_set_msr(vcpu, index, *data);
474 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
479 set_efer(vcpu, data);
482 case MSR_IA32_MC0_STATUS:
483 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
486 case MSR_IA32_MCG_STATUS:
487 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
490 case MSR_IA32_UCODE_REV:
491 case MSR_IA32_UCODE_WRITE:
492 case 0x200 ... 0x2ff: /* MTRRs */
494 case MSR_IA32_APICBASE:
495 kvm_set_apic_base(vcpu, data);
497 case MSR_IA32_MISC_ENABLE:
498 vcpu->arch.ia32_misc_enable_msr = data;
501 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
506 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
510 * Reads an msr value (of 'msr_index') into 'pdata'.
511 * Returns 0 on success, non-0 otherwise.
512 * Assumes vcpu_load() was already called.
514 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
516 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
519 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
524 case 0xc0010010: /* SYSCFG */
525 case 0xc0010015: /* HWCR */
526 case MSR_IA32_PLATFORM_ID:
527 case MSR_IA32_P5_MC_ADDR:
528 case MSR_IA32_P5_MC_TYPE:
529 case MSR_IA32_MC0_CTL:
530 case MSR_IA32_MCG_STATUS:
531 case MSR_IA32_MCG_CAP:
532 case MSR_IA32_MC0_MISC:
533 case MSR_IA32_MC0_MISC+4:
534 case MSR_IA32_MC0_MISC+8:
535 case MSR_IA32_MC0_MISC+12:
536 case MSR_IA32_MC0_MISC+16:
537 case MSR_IA32_UCODE_REV:
538 case MSR_IA32_PERF_STATUS:
539 case MSR_IA32_EBL_CR_POWERON:
542 case 0x200 ... 0x2ff:
545 case 0xcd: /* fsb frequency */
548 case MSR_IA32_APICBASE:
549 data = kvm_get_apic_base(vcpu);
551 case MSR_IA32_MISC_ENABLE:
552 data = vcpu->arch.ia32_misc_enable_msr;
556 data = vcpu->arch.shadow_efer;
560 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
566 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
569 * Read or write a bunch of msrs. All parameters are kernel addresses.
571 * @return number of msrs set successfully.
573 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
574 struct kvm_msr_entry *entries,
575 int (*do_msr)(struct kvm_vcpu *vcpu,
576 unsigned index, u64 *data))
582 for (i = 0; i < msrs->nmsrs; ++i)
583 if (do_msr(vcpu, entries[i].index, &entries[i].data))
592 * Read or write a bunch of msrs. Parameters are user addresses.
594 * @return number of msrs set successfully.
596 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
597 int (*do_msr)(struct kvm_vcpu *vcpu,
598 unsigned index, u64 *data),
601 struct kvm_msrs msrs;
602 struct kvm_msr_entry *entries;
607 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
611 if (msrs.nmsrs >= MAX_IO_MSRS)
615 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
616 entries = vmalloc(size);
621 if (copy_from_user(entries, user_msrs->entries, size))
624 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
629 if (writeback && copy_to_user(user_msrs->entries, entries, size))
641 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
644 void decache_vcpus_on_cpu(int cpu)
647 struct kvm_vcpu *vcpu;
650 spin_lock(&kvm_lock);
651 list_for_each_entry(vm, &vm_list, vm_list)
652 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
657 * If the vcpu is locked, then it is running on some
658 * other cpu and therefore it is not cached on the
661 * If it's not locked, check the last cpu it executed
664 if (mutex_trylock(&vcpu->mutex)) {
665 if (vcpu->cpu == cpu) {
666 kvm_x86_ops->vcpu_decache(vcpu);
669 mutex_unlock(&vcpu->mutex);
672 spin_unlock(&kvm_lock);
675 int kvm_dev_ioctl_check_extension(long ext)
680 case KVM_CAP_IRQCHIP:
682 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
683 case KVM_CAP_USER_MEMORY:
684 case KVM_CAP_SET_TSS_ADDR:
685 case KVM_CAP_EXT_CPUID:
696 long kvm_arch_dev_ioctl(struct file *filp,
697 unsigned int ioctl, unsigned long arg)
699 void __user *argp = (void __user *)arg;
703 case KVM_GET_MSR_INDEX_LIST: {
704 struct kvm_msr_list __user *user_msr_list = argp;
705 struct kvm_msr_list msr_list;
709 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
712 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
713 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
716 if (n < num_msrs_to_save)
719 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
720 num_msrs_to_save * sizeof(u32)))
722 if (copy_to_user(user_msr_list->indices
723 + num_msrs_to_save * sizeof(u32),
725 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
737 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
739 kvm_x86_ops->vcpu_load(vcpu, cpu);
742 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
744 kvm_x86_ops->vcpu_put(vcpu);
745 kvm_put_guest_fpu(vcpu);
748 static int is_efer_nx(void)
752 rdmsrl(MSR_EFER, efer);
753 return efer & EFER_NX;
756 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
759 struct kvm_cpuid_entry2 *e, *entry;
762 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
763 e = &vcpu->arch.cpuid_entries[i];
764 if (e->function == 0x80000001) {
769 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
770 entry->edx &= ~(1 << 20);
771 printk(KERN_INFO "kvm: guest NX capability removed\n");
775 /* when an old userspace process fills a new kernel module */
776 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
777 struct kvm_cpuid *cpuid,
778 struct kvm_cpuid_entry __user *entries)
781 struct kvm_cpuid_entry *cpuid_entries;
784 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
787 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
791 if (copy_from_user(cpuid_entries, entries,
792 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
794 for (i = 0; i < cpuid->nent; i++) {
795 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
796 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
797 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
798 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
799 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
800 vcpu->arch.cpuid_entries[i].index = 0;
801 vcpu->arch.cpuid_entries[i].flags = 0;
802 vcpu->arch.cpuid_entries[i].padding[0] = 0;
803 vcpu->arch.cpuid_entries[i].padding[1] = 0;
804 vcpu->arch.cpuid_entries[i].padding[2] = 0;
806 vcpu->arch.cpuid_nent = cpuid->nent;
807 cpuid_fix_nx_cap(vcpu);
811 vfree(cpuid_entries);
816 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
817 struct kvm_cpuid2 *cpuid,
818 struct kvm_cpuid_entry2 __user *entries)
823 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
826 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
827 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
829 vcpu->arch.cpuid_nent = cpuid->nent;
836 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
837 struct kvm_cpuid2 *cpuid,
838 struct kvm_cpuid_entry2 __user *entries)
843 if (cpuid->nent < vcpu->arch.cpuid_nent)
846 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
847 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
852 cpuid->nent = vcpu->arch.cpuid_nent;
856 static inline u32 bit(int bitno)
858 return 1 << (bitno & 31);
861 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
864 entry->function = function;
865 entry->index = index;
866 cpuid_count(entry->function, entry->index,
867 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
871 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
872 u32 index, int *nent, int maxnent)
874 const u32 kvm_supported_word0_x86_features = bit(X86_FEATURE_FPU) |
875 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
876 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
877 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
878 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
879 bit(X86_FEATURE_SEP) | bit(X86_FEATURE_PGE) |
880 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
881 bit(X86_FEATURE_CLFLSH) | bit(X86_FEATURE_MMX) |
882 bit(X86_FEATURE_FXSR) | bit(X86_FEATURE_XMM) |
883 bit(X86_FEATURE_XMM2) | bit(X86_FEATURE_SELFSNOOP);
884 const u32 kvm_supported_word1_x86_features = bit(X86_FEATURE_FPU) |
885 bit(X86_FEATURE_VME) | bit(X86_FEATURE_DE) |
886 bit(X86_FEATURE_PSE) | bit(X86_FEATURE_TSC) |
887 bit(X86_FEATURE_MSR) | bit(X86_FEATURE_PAE) |
888 bit(X86_FEATURE_CX8) | bit(X86_FEATURE_APIC) |
889 bit(X86_FEATURE_PGE) |
890 bit(X86_FEATURE_CMOV) | bit(X86_FEATURE_PSE36) |
891 bit(X86_FEATURE_MMX) | bit(X86_FEATURE_FXSR) |
892 bit(X86_FEATURE_SYSCALL) |
893 (bit(X86_FEATURE_NX) && is_efer_nx()) |
895 bit(X86_FEATURE_LM) |
897 bit(X86_FEATURE_MMXEXT) |
898 bit(X86_FEATURE_3DNOWEXT) |
899 bit(X86_FEATURE_3DNOW);
900 const u32 kvm_supported_word3_x86_features =
901 bit(X86_FEATURE_XMM3) | bit(X86_FEATURE_CX16);
902 const u32 kvm_supported_word6_x86_features =
903 bit(X86_FEATURE_LAHF_LM) | bit(X86_FEATURE_CMP_LEGACY);
905 /* all func 2 cpuid_count() should be called on the same cpu */
907 do_cpuid_1_ent(entry, function, index);
912 entry->eax = min(entry->eax, (u32)0xb);
915 entry->edx &= kvm_supported_word0_x86_features;
916 entry->ecx &= kvm_supported_word3_x86_features;
918 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
919 * may return different values. This forces us to get_cpu() before
920 * issuing the first command, and also to emulate this annoying behavior
921 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
923 int t, times = entry->eax & 0xff;
925 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
926 for (t = 1; t < times && *nent < maxnent; ++t) {
927 do_cpuid_1_ent(&entry[t], function, 0);
928 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
933 /* function 4 and 0xb have additional index. */
935 int index, cache_type;
937 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
938 /* read more entries until cache_type is zero */
939 for (index = 1; *nent < maxnent; ++index) {
940 cache_type = entry[index - 1].eax & 0x1f;
943 do_cpuid_1_ent(&entry[index], function, index);
944 entry[index].flags |=
945 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
951 int index, level_type;
953 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
954 /* read more entries until level_type is zero */
955 for (index = 1; *nent < maxnent; ++index) {
956 level_type = entry[index - 1].ecx & 0xff;
959 do_cpuid_1_ent(&entry[index], function, index);
960 entry[index].flags |=
961 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
967 entry->eax = min(entry->eax, 0x8000001a);
970 entry->edx &= kvm_supported_word1_x86_features;
971 entry->ecx &= kvm_supported_word6_x86_features;
977 static int kvm_vm_ioctl_get_supported_cpuid(struct kvm *kvm,
978 struct kvm_cpuid2 *cpuid,
979 struct kvm_cpuid_entry2 __user *entries)
981 struct kvm_cpuid_entry2 *cpuid_entries;
982 int limit, nent = 0, r = -E2BIG;
988 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
992 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
993 limit = cpuid_entries[0].eax;
994 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
995 do_cpuid_ent(&cpuid_entries[nent], func, 0,
998 if (nent >= cpuid->nent)
1001 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1002 limit = cpuid_entries[nent - 1].eax;
1003 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1004 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1005 &nent, cpuid->nent);
1007 if (copy_to_user(entries, cpuid_entries,
1008 nent * sizeof(struct kvm_cpuid_entry2)))
1014 vfree(cpuid_entries);
1019 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1020 struct kvm_lapic_state *s)
1023 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1029 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1030 struct kvm_lapic_state *s)
1033 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1034 kvm_apic_post_state_restore(vcpu);
1040 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1041 struct kvm_interrupt *irq)
1043 if (irq->irq < 0 || irq->irq >= 256)
1045 if (irqchip_in_kernel(vcpu->kvm))
1049 set_bit(irq->irq, vcpu->arch.irq_pending);
1050 set_bit(irq->irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1057 long kvm_arch_vcpu_ioctl(struct file *filp,
1058 unsigned int ioctl, unsigned long arg)
1060 struct kvm_vcpu *vcpu = filp->private_data;
1061 void __user *argp = (void __user *)arg;
1065 case KVM_GET_LAPIC: {
1066 struct kvm_lapic_state lapic;
1068 memset(&lapic, 0, sizeof lapic);
1069 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
1073 if (copy_to_user(argp, &lapic, sizeof lapic))
1078 case KVM_SET_LAPIC: {
1079 struct kvm_lapic_state lapic;
1082 if (copy_from_user(&lapic, argp, sizeof lapic))
1084 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
1090 case KVM_INTERRUPT: {
1091 struct kvm_interrupt irq;
1094 if (copy_from_user(&irq, argp, sizeof irq))
1096 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1102 case KVM_SET_CPUID: {
1103 struct kvm_cpuid __user *cpuid_arg = argp;
1104 struct kvm_cpuid cpuid;
1107 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1109 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1114 case KVM_SET_CPUID2: {
1115 struct kvm_cpuid2 __user *cpuid_arg = argp;
1116 struct kvm_cpuid2 cpuid;
1119 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1121 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1122 cpuid_arg->entries);
1127 case KVM_GET_CPUID2: {
1128 struct kvm_cpuid2 __user *cpuid_arg = argp;
1129 struct kvm_cpuid2 cpuid;
1132 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1134 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1135 cpuid_arg->entries);
1139 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1145 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1148 r = msr_io(vcpu, argp, do_set_msr, 0);
1157 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1161 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1163 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1167 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1168 u32 kvm_nr_mmu_pages)
1170 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1173 mutex_lock(&kvm->lock);
1175 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1176 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1178 mutex_unlock(&kvm->lock);
1182 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1184 return kvm->arch.n_alloc_mmu_pages;
1187 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1190 struct kvm_mem_alias *alias;
1192 for (i = 0; i < kvm->arch.naliases; ++i) {
1193 alias = &kvm->arch.aliases[i];
1194 if (gfn >= alias->base_gfn
1195 && gfn < alias->base_gfn + alias->npages)
1196 return alias->target_gfn + gfn - alias->base_gfn;
1202 * Set a new alias region. Aliases map a portion of physical memory into
1203 * another portion. This is useful for memory windows, for example the PC
1206 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1207 struct kvm_memory_alias *alias)
1210 struct kvm_mem_alias *p;
1213 /* General sanity checks */
1214 if (alias->memory_size & (PAGE_SIZE - 1))
1216 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1218 if (alias->slot >= KVM_ALIAS_SLOTS)
1220 if (alias->guest_phys_addr + alias->memory_size
1221 < alias->guest_phys_addr)
1223 if (alias->target_phys_addr + alias->memory_size
1224 < alias->target_phys_addr)
1227 mutex_lock(&kvm->lock);
1229 p = &kvm->arch.aliases[alias->slot];
1230 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1231 p->npages = alias->memory_size >> PAGE_SHIFT;
1232 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1234 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1235 if (kvm->arch.aliases[n - 1].npages)
1237 kvm->arch.naliases = n;
1239 kvm_mmu_zap_all(kvm);
1241 mutex_unlock(&kvm->lock);
1249 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1254 switch (chip->chip_id) {
1255 case KVM_IRQCHIP_PIC_MASTER:
1256 memcpy(&chip->chip.pic,
1257 &pic_irqchip(kvm)->pics[0],
1258 sizeof(struct kvm_pic_state));
1260 case KVM_IRQCHIP_PIC_SLAVE:
1261 memcpy(&chip->chip.pic,
1262 &pic_irqchip(kvm)->pics[1],
1263 sizeof(struct kvm_pic_state));
1265 case KVM_IRQCHIP_IOAPIC:
1266 memcpy(&chip->chip.ioapic,
1267 ioapic_irqchip(kvm),
1268 sizeof(struct kvm_ioapic_state));
1277 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1282 switch (chip->chip_id) {
1283 case KVM_IRQCHIP_PIC_MASTER:
1284 memcpy(&pic_irqchip(kvm)->pics[0],
1286 sizeof(struct kvm_pic_state));
1288 case KVM_IRQCHIP_PIC_SLAVE:
1289 memcpy(&pic_irqchip(kvm)->pics[1],
1291 sizeof(struct kvm_pic_state));
1293 case KVM_IRQCHIP_IOAPIC:
1294 memcpy(ioapic_irqchip(kvm),
1296 sizeof(struct kvm_ioapic_state));
1302 kvm_pic_update_irq(pic_irqchip(kvm));
1307 * Get (and clear) the dirty memory log for a memory slot.
1309 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
1310 struct kvm_dirty_log *log)
1314 struct kvm_memory_slot *memslot;
1317 mutex_lock(&kvm->lock);
1319 r = kvm_get_dirty_log(kvm, log, &is_dirty);
1323 /* If nothing is dirty, don't bother messing with page tables. */
1325 kvm_mmu_slot_remove_write_access(kvm, log->slot);
1326 kvm_flush_remote_tlbs(kvm);
1327 memslot = &kvm->memslots[log->slot];
1328 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1329 memset(memslot->dirty_bitmap, 0, n);
1333 mutex_unlock(&kvm->lock);
1337 long kvm_arch_vm_ioctl(struct file *filp,
1338 unsigned int ioctl, unsigned long arg)
1340 struct kvm *kvm = filp->private_data;
1341 void __user *argp = (void __user *)arg;
1345 case KVM_SET_TSS_ADDR:
1346 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1350 case KVM_SET_MEMORY_REGION: {
1351 struct kvm_memory_region kvm_mem;
1352 struct kvm_userspace_memory_region kvm_userspace_mem;
1355 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
1357 kvm_userspace_mem.slot = kvm_mem.slot;
1358 kvm_userspace_mem.flags = kvm_mem.flags;
1359 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
1360 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
1361 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
1366 case KVM_SET_NR_MMU_PAGES:
1367 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1371 case KVM_GET_NR_MMU_PAGES:
1372 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
1374 case KVM_SET_MEMORY_ALIAS: {
1375 struct kvm_memory_alias alias;
1378 if (copy_from_user(&alias, argp, sizeof alias))
1380 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
1385 case KVM_CREATE_IRQCHIP:
1387 kvm->arch.vpic = kvm_create_pic(kvm);
1388 if (kvm->arch.vpic) {
1389 r = kvm_ioapic_init(kvm);
1391 kfree(kvm->arch.vpic);
1392 kvm->arch.vpic = NULL;
1398 case KVM_IRQ_LINE: {
1399 struct kvm_irq_level irq_event;
1402 if (copy_from_user(&irq_event, argp, sizeof irq_event))
1404 if (irqchip_in_kernel(kvm)) {
1405 mutex_lock(&kvm->lock);
1406 if (irq_event.irq < 16)
1407 kvm_pic_set_irq(pic_irqchip(kvm),
1410 kvm_ioapic_set_irq(kvm->arch.vioapic,
1413 mutex_unlock(&kvm->lock);
1418 case KVM_GET_IRQCHIP: {
1419 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1420 struct kvm_irqchip chip;
1423 if (copy_from_user(&chip, argp, sizeof chip))
1426 if (!irqchip_in_kernel(kvm))
1428 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
1432 if (copy_to_user(argp, &chip, sizeof chip))
1437 case KVM_SET_IRQCHIP: {
1438 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
1439 struct kvm_irqchip chip;
1442 if (copy_from_user(&chip, argp, sizeof chip))
1445 if (!irqchip_in_kernel(kvm))
1447 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
1453 case KVM_GET_SUPPORTED_CPUID: {
1454 struct kvm_cpuid2 __user *cpuid_arg = argp;
1455 struct kvm_cpuid2 cpuid;
1458 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1460 r = kvm_vm_ioctl_get_supported_cpuid(kvm, &cpuid,
1461 cpuid_arg->entries);
1466 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1478 static void kvm_init_msr_list(void)
1483 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
1484 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1487 msrs_to_save[j] = msrs_to_save[i];
1490 num_msrs_to_save = j;
1494 * Only apic need an MMIO device hook, so shortcut now..
1496 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1499 struct kvm_io_device *dev;
1501 if (vcpu->arch.apic) {
1502 dev = &vcpu->arch.apic->dev;
1503 if (dev->in_range(dev, addr))
1510 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1513 struct kvm_io_device *dev;
1515 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1517 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1521 int emulator_read_std(unsigned long addr,
1524 struct kvm_vcpu *vcpu)
1529 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1530 unsigned offset = addr & (PAGE_SIZE-1);
1531 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1534 if (gpa == UNMAPPED_GVA)
1535 return X86EMUL_PROPAGATE_FAULT;
1536 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1538 return X86EMUL_UNHANDLEABLE;
1545 return X86EMUL_CONTINUE;
1547 EXPORT_SYMBOL_GPL(emulator_read_std);
1549 static int emulator_read_emulated(unsigned long addr,
1552 struct kvm_vcpu *vcpu)
1554 struct kvm_io_device *mmio_dev;
1557 if (vcpu->mmio_read_completed) {
1558 memcpy(val, vcpu->mmio_data, bytes);
1559 vcpu->mmio_read_completed = 0;
1560 return X86EMUL_CONTINUE;
1563 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1565 /* For APIC access vmexit */
1566 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1569 if (emulator_read_std(addr, val, bytes, vcpu)
1570 == X86EMUL_CONTINUE)
1571 return X86EMUL_CONTINUE;
1572 if (gpa == UNMAPPED_GVA)
1573 return X86EMUL_PROPAGATE_FAULT;
1577 * Is this MMIO handled locally?
1579 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1581 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1582 return X86EMUL_CONTINUE;
1585 vcpu->mmio_needed = 1;
1586 vcpu->mmio_phys_addr = gpa;
1587 vcpu->mmio_size = bytes;
1588 vcpu->mmio_is_write = 0;
1590 return X86EMUL_UNHANDLEABLE;
1593 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1594 const void *val, int bytes)
1598 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1601 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1605 static int emulator_write_emulated_onepage(unsigned long addr,
1608 struct kvm_vcpu *vcpu)
1610 struct kvm_io_device *mmio_dev;
1611 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1613 if (gpa == UNMAPPED_GVA) {
1614 kvm_inject_page_fault(vcpu, addr, 2);
1615 return X86EMUL_PROPAGATE_FAULT;
1618 /* For APIC access vmexit */
1619 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1622 if (emulator_write_phys(vcpu, gpa, val, bytes))
1623 return X86EMUL_CONTINUE;
1627 * Is this MMIO handled locally?
1629 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1631 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1632 return X86EMUL_CONTINUE;
1635 vcpu->mmio_needed = 1;
1636 vcpu->mmio_phys_addr = gpa;
1637 vcpu->mmio_size = bytes;
1638 vcpu->mmio_is_write = 1;
1639 memcpy(vcpu->mmio_data, val, bytes);
1641 return X86EMUL_CONTINUE;
1644 int emulator_write_emulated(unsigned long addr,
1647 struct kvm_vcpu *vcpu)
1649 /* Crossing a page boundary? */
1650 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1653 now = -addr & ~PAGE_MASK;
1654 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1655 if (rc != X86EMUL_CONTINUE)
1661 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1663 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1665 static int emulator_cmpxchg_emulated(unsigned long addr,
1669 struct kvm_vcpu *vcpu)
1671 static int reported;
1675 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1677 #ifndef CONFIG_X86_64
1678 /* guests cmpxchg8b have to be emulated atomically */
1680 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
1685 if (gpa == UNMAPPED_GVA ||
1686 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1689 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
1693 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1694 addr = kmap_atomic(page, KM_USER0);
1695 set_64bit((u64 *)(addr + offset_in_page(gpa)), val);
1696 kunmap_atomic(addr, KM_USER0);
1697 kvm_release_page_dirty(page);
1702 return emulator_write_emulated(addr, new, bytes, vcpu);
1705 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1707 return kvm_x86_ops->get_segment_base(vcpu, seg);
1710 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1712 return X86EMUL_CONTINUE;
1715 int emulate_clts(struct kvm_vcpu *vcpu)
1717 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
1718 return X86EMUL_CONTINUE;
1721 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1723 struct kvm_vcpu *vcpu = ctxt->vcpu;
1727 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1728 return X86EMUL_CONTINUE;
1730 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1731 return X86EMUL_UNHANDLEABLE;
1735 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1737 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1740 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1742 /* FIXME: better handling */
1743 return X86EMUL_UNHANDLEABLE;
1745 return X86EMUL_CONTINUE;
1748 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1750 static int reported;
1752 unsigned long rip = vcpu->arch.rip;
1753 unsigned long rip_linear;
1755 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1760 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1762 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1763 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1766 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1768 struct x86_emulate_ops emulate_ops = {
1769 .read_std = emulator_read_std,
1770 .read_emulated = emulator_read_emulated,
1771 .write_emulated = emulator_write_emulated,
1772 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1775 int emulate_instruction(struct kvm_vcpu *vcpu,
1776 struct kvm_run *run,
1783 vcpu->arch.mmio_fault_cr2 = cr2;
1784 kvm_x86_ops->cache_regs(vcpu);
1786 vcpu->mmio_is_write = 0;
1787 vcpu->arch.pio.string = 0;
1791 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1793 vcpu->arch.emulate_ctxt.vcpu = vcpu;
1794 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1795 vcpu->arch.emulate_ctxt.mode =
1796 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
1797 ? X86EMUL_MODE_REAL : cs_l
1798 ? X86EMUL_MODE_PROT64 : cs_db
1799 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1801 if (vcpu->arch.emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1802 vcpu->arch.emulate_ctxt.cs_base = 0;
1803 vcpu->arch.emulate_ctxt.ds_base = 0;
1804 vcpu->arch.emulate_ctxt.es_base = 0;
1805 vcpu->arch.emulate_ctxt.ss_base = 0;
1807 vcpu->arch.emulate_ctxt.cs_base =
1808 get_segment_base(vcpu, VCPU_SREG_CS);
1809 vcpu->arch.emulate_ctxt.ds_base =
1810 get_segment_base(vcpu, VCPU_SREG_DS);
1811 vcpu->arch.emulate_ctxt.es_base =
1812 get_segment_base(vcpu, VCPU_SREG_ES);
1813 vcpu->arch.emulate_ctxt.ss_base =
1814 get_segment_base(vcpu, VCPU_SREG_SS);
1817 vcpu->arch.emulate_ctxt.gs_base =
1818 get_segment_base(vcpu, VCPU_SREG_GS);
1819 vcpu->arch.emulate_ctxt.fs_base =
1820 get_segment_base(vcpu, VCPU_SREG_FS);
1822 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
1823 ++vcpu->stat.insn_emulation;
1825 ++vcpu->stat.insn_emulation_fail;
1826 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1827 return EMULATE_DONE;
1828 return EMULATE_FAIL;
1832 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
1834 if (vcpu->arch.pio.string)
1835 return EMULATE_DO_MMIO;
1837 if ((r || vcpu->mmio_is_write) && run) {
1838 run->exit_reason = KVM_EXIT_MMIO;
1839 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1840 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1841 run->mmio.len = vcpu->mmio_size;
1842 run->mmio.is_write = vcpu->mmio_is_write;
1846 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1847 return EMULATE_DONE;
1848 if (!vcpu->mmio_needed) {
1849 kvm_report_emulation_failure(vcpu, "mmio");
1850 return EMULATE_FAIL;
1852 return EMULATE_DO_MMIO;
1855 kvm_x86_ops->decache_regs(vcpu);
1856 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
1858 if (vcpu->mmio_is_write) {
1859 vcpu->mmio_needed = 0;
1860 return EMULATE_DO_MMIO;
1863 return EMULATE_DONE;
1865 EXPORT_SYMBOL_GPL(emulate_instruction);
1867 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1871 for (i = 0; i < ARRAY_SIZE(vcpu->arch.pio.guest_pages); ++i)
1872 if (vcpu->arch.pio.guest_pages[i]) {
1873 kvm_release_page_dirty(vcpu->arch.pio.guest_pages[i]);
1874 vcpu->arch.pio.guest_pages[i] = NULL;
1878 static int pio_copy_data(struct kvm_vcpu *vcpu)
1880 void *p = vcpu->arch.pio_data;
1883 int nr_pages = vcpu->arch.pio.guest_pages[1] ? 2 : 1;
1885 q = vmap(vcpu->arch.pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1888 free_pio_guest_pages(vcpu);
1891 q += vcpu->arch.pio.guest_page_offset;
1892 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
1893 if (vcpu->arch.pio.in)
1894 memcpy(q, p, bytes);
1896 memcpy(p, q, bytes);
1897 q -= vcpu->arch.pio.guest_page_offset;
1899 free_pio_guest_pages(vcpu);
1903 int complete_pio(struct kvm_vcpu *vcpu)
1905 struct kvm_pio_request *io = &vcpu->arch.pio;
1909 kvm_x86_ops->cache_regs(vcpu);
1913 memcpy(&vcpu->arch.regs[VCPU_REGS_RAX], vcpu->arch.pio_data,
1917 r = pio_copy_data(vcpu);
1919 kvm_x86_ops->cache_regs(vcpu);
1926 delta *= io->cur_count;
1928 * The size of the register should really depend on
1929 * current address size.
1931 vcpu->arch.regs[VCPU_REGS_RCX] -= delta;
1937 vcpu->arch.regs[VCPU_REGS_RDI] += delta;
1939 vcpu->arch.regs[VCPU_REGS_RSI] += delta;
1942 kvm_x86_ops->decache_regs(vcpu);
1944 io->count -= io->cur_count;
1950 static void kernel_pio(struct kvm_io_device *pio_dev,
1951 struct kvm_vcpu *vcpu,
1954 /* TODO: String I/O for in kernel device */
1956 mutex_lock(&vcpu->kvm->lock);
1957 if (vcpu->arch.pio.in)
1958 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
1959 vcpu->arch.pio.size,
1962 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
1963 vcpu->arch.pio.size,
1965 mutex_unlock(&vcpu->kvm->lock);
1968 static void pio_string_write(struct kvm_io_device *pio_dev,
1969 struct kvm_vcpu *vcpu)
1971 struct kvm_pio_request *io = &vcpu->arch.pio;
1972 void *pd = vcpu->arch.pio_data;
1975 mutex_lock(&vcpu->kvm->lock);
1976 for (i = 0; i < io->cur_count; i++) {
1977 kvm_iodevice_write(pio_dev, io->port,
1982 mutex_unlock(&vcpu->kvm->lock);
1985 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1988 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1991 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1992 int size, unsigned port)
1994 struct kvm_io_device *pio_dev;
1996 vcpu->run->exit_reason = KVM_EXIT_IO;
1997 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1998 vcpu->run->io.size = vcpu->arch.pio.size = size;
1999 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2000 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2001 vcpu->run->io.port = vcpu->arch.pio.port = port;
2002 vcpu->arch.pio.in = in;
2003 vcpu->arch.pio.string = 0;
2004 vcpu->arch.pio.down = 0;
2005 vcpu->arch.pio.guest_page_offset = 0;
2006 vcpu->arch.pio.rep = 0;
2008 kvm_x86_ops->cache_regs(vcpu);
2009 memcpy(vcpu->arch.pio_data, &vcpu->arch.regs[VCPU_REGS_RAX], 4);
2010 kvm_x86_ops->decache_regs(vcpu);
2012 kvm_x86_ops->skip_emulated_instruction(vcpu);
2014 pio_dev = vcpu_find_pio_dev(vcpu, port);
2016 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2022 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2024 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2025 int size, unsigned long count, int down,
2026 gva_t address, int rep, unsigned port)
2028 unsigned now, in_page;
2032 struct kvm_io_device *pio_dev;
2034 vcpu->run->exit_reason = KVM_EXIT_IO;
2035 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2036 vcpu->run->io.size = vcpu->arch.pio.size = size;
2037 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2038 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2039 vcpu->run->io.port = vcpu->arch.pio.port = port;
2040 vcpu->arch.pio.in = in;
2041 vcpu->arch.pio.string = 1;
2042 vcpu->arch.pio.down = down;
2043 vcpu->arch.pio.guest_page_offset = offset_in_page(address);
2044 vcpu->arch.pio.rep = rep;
2047 kvm_x86_ops->skip_emulated_instruction(vcpu);
2052 in_page = PAGE_SIZE - offset_in_page(address);
2054 in_page = offset_in_page(address) + size;
2055 now = min(count, (unsigned long)in_page / size);
2058 * String I/O straddles page boundary. Pin two guest pages
2059 * so that we satisfy atomicity constraints. Do just one
2060 * transaction to avoid complexity.
2067 * String I/O in reverse. Yuck. Kill the guest, fix later.
2069 pr_unimpl(vcpu, "guest string pio down\n");
2070 kvm_inject_gp(vcpu, 0);
2073 vcpu->run->io.count = now;
2074 vcpu->arch.pio.cur_count = now;
2076 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2077 kvm_x86_ops->skip_emulated_instruction(vcpu);
2079 for (i = 0; i < nr_pages; ++i) {
2080 mutex_lock(&vcpu->kvm->lock);
2081 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
2082 vcpu->arch.pio.guest_pages[i] = page;
2083 mutex_unlock(&vcpu->kvm->lock);
2085 kvm_inject_gp(vcpu, 0);
2086 free_pio_guest_pages(vcpu);
2091 pio_dev = vcpu_find_pio_dev(vcpu, port);
2092 if (!vcpu->arch.pio.in) {
2093 /* string PIO write */
2094 ret = pio_copy_data(vcpu);
2095 if (ret >= 0 && pio_dev) {
2096 pio_string_write(pio_dev, vcpu);
2098 if (vcpu->arch.pio.count == 0)
2102 pr_unimpl(vcpu, "no string pio read support yet, "
2103 "port %x size %d count %ld\n",
2108 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2110 int kvm_arch_init(void *opaque)
2113 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
2115 r = kvm_mmu_module_init();
2119 kvm_init_msr_list();
2122 printk(KERN_ERR "kvm: already loaded the other module\n");
2127 if (!ops->cpu_has_kvm_support()) {
2128 printk(KERN_ERR "kvm: no hardware support\n");
2132 if (ops->disabled_by_bios()) {
2133 printk(KERN_ERR "kvm: disabled by bios\n");
2139 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2143 kvm_mmu_module_exit();
2148 void kvm_arch_exit(void)
2151 kvm_mmu_module_exit();
2154 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
2156 ++vcpu->stat.halt_exits;
2157 if (irqchip_in_kernel(vcpu->kvm)) {
2158 vcpu->arch.mp_state = VCPU_MP_STATE_HALTED;
2159 kvm_vcpu_block(vcpu);
2160 if (vcpu->arch.mp_state != VCPU_MP_STATE_RUNNABLE)
2164 vcpu->run->exit_reason = KVM_EXIT_HLT;
2168 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
2170 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
2172 unsigned long nr, a0, a1, a2, a3, ret;
2174 kvm_x86_ops->cache_regs(vcpu);
2176 nr = vcpu->arch.regs[VCPU_REGS_RAX];
2177 a0 = vcpu->arch.regs[VCPU_REGS_RBX];
2178 a1 = vcpu->arch.regs[VCPU_REGS_RCX];
2179 a2 = vcpu->arch.regs[VCPU_REGS_RDX];
2180 a3 = vcpu->arch.regs[VCPU_REGS_RSI];
2182 if (!is_long_mode(vcpu)) {
2195 vcpu->arch.regs[VCPU_REGS_RAX] = ret;
2196 kvm_x86_ops->decache_regs(vcpu);
2199 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
2201 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
2203 char instruction[3];
2206 mutex_lock(&vcpu->kvm->lock);
2209 * Blow out the MMU to ensure that no other VCPU has an active mapping
2210 * to ensure that the updated hypercall appears atomically across all
2213 kvm_mmu_zap_all(vcpu->kvm);
2215 kvm_x86_ops->cache_regs(vcpu);
2216 kvm_x86_ops->patch_hypercall(vcpu, instruction);
2217 if (emulator_write_emulated(vcpu->arch.rip, instruction, 3, vcpu)
2218 != X86EMUL_CONTINUE)
2221 mutex_unlock(&vcpu->kvm->lock);
2226 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
2228 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
2231 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2233 struct descriptor_table dt = { limit, base };
2235 kvm_x86_ops->set_gdt(vcpu, &dt);
2238 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
2240 struct descriptor_table dt = { limit, base };
2242 kvm_x86_ops->set_idt(vcpu, &dt);
2245 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
2246 unsigned long *rflags)
2249 *rflags = kvm_x86_ops->get_rflags(vcpu);
2252 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
2254 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2257 return vcpu->arch.cr0;
2259 return vcpu->arch.cr2;
2261 return vcpu->arch.cr3;
2263 return vcpu->arch.cr4;
2265 return get_cr8(vcpu);
2267 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2272 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
2273 unsigned long *rflags)
2277 set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
2278 *rflags = kvm_x86_ops->get_rflags(vcpu);
2281 vcpu->arch.cr2 = val;
2287 set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
2290 set_cr8(vcpu, val & 0xfUL);
2293 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
2297 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
2299 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
2300 int j, nent = vcpu->arch.cpuid_nent;
2302 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
2303 /* when no next entry is found, the current entry[i] is reselected */
2304 for (j = i + 1; j == i; j = (j + 1) % nent) {
2305 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
2306 if (ej->function == e->function) {
2307 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2311 return 0; /* silence gcc, even though control never reaches here */
2314 /* find an entry with matching function, matching index (if needed), and that
2315 * should be read next (if it's stateful) */
2316 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
2317 u32 function, u32 index)
2319 if (e->function != function)
2321 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
2323 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
2324 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
2329 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
2332 u32 function, index;
2333 struct kvm_cpuid_entry2 *e, *best;
2335 kvm_x86_ops->cache_regs(vcpu);
2336 function = vcpu->arch.regs[VCPU_REGS_RAX];
2337 index = vcpu->arch.regs[VCPU_REGS_RCX];
2338 vcpu->arch.regs[VCPU_REGS_RAX] = 0;
2339 vcpu->arch.regs[VCPU_REGS_RBX] = 0;
2340 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2341 vcpu->arch.regs[VCPU_REGS_RDX] = 0;
2343 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2344 e = &vcpu->arch.cpuid_entries[i];
2345 if (is_matching_cpuid_entry(e, function, index)) {
2346 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
2347 move_to_next_stateful_cpuid_entry(vcpu, i);
2352 * Both basic or both extended?
2354 if (((e->function ^ function) & 0x80000000) == 0)
2355 if (!best || e->function > best->function)
2359 vcpu->arch.regs[VCPU_REGS_RAX] = best->eax;
2360 vcpu->arch.regs[VCPU_REGS_RBX] = best->ebx;
2361 vcpu->arch.regs[VCPU_REGS_RCX] = best->ecx;
2362 vcpu->arch.regs[VCPU_REGS_RDX] = best->edx;
2364 kvm_x86_ops->decache_regs(vcpu);
2365 kvm_x86_ops->skip_emulated_instruction(vcpu);
2367 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
2370 * Check if userspace requested an interrupt window, and that the
2371 * interrupt window is open.
2373 * No need to exit to userspace if we already have an interrupt queued.
2375 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
2376 struct kvm_run *kvm_run)
2378 return (!vcpu->arch.irq_summary &&
2379 kvm_run->request_interrupt_window &&
2380 vcpu->arch.interrupt_window_open &&
2381 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
2384 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
2385 struct kvm_run *kvm_run)
2387 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
2388 kvm_run->cr8 = get_cr8(vcpu);
2389 kvm_run->apic_base = kvm_get_apic_base(vcpu);
2390 if (irqchip_in_kernel(vcpu->kvm))
2391 kvm_run->ready_for_interrupt_injection = 1;
2393 kvm_run->ready_for_interrupt_injection =
2394 (vcpu->arch.interrupt_window_open &&
2395 vcpu->arch.irq_summary == 0);
2398 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2402 if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
2403 pr_debug("vcpu %d received sipi with vector # %x\n",
2404 vcpu->vcpu_id, vcpu->arch.sipi_vector);
2405 kvm_lapic_reset(vcpu);
2406 r = kvm_x86_ops->vcpu_reset(vcpu);
2409 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
2413 if (vcpu->guest_debug.enabled)
2414 kvm_x86_ops->guest_debug_pre(vcpu);
2417 r = kvm_mmu_reload(vcpu);
2421 kvm_inject_pending_timer_irqs(vcpu);
2425 kvm_x86_ops->prepare_guest_switch(vcpu);
2426 kvm_load_guest_fpu(vcpu);
2428 local_irq_disable();
2430 if (signal_pending(current)) {
2434 kvm_run->exit_reason = KVM_EXIT_INTR;
2435 ++vcpu->stat.signal_exits;
2439 if (vcpu->arch.exception.pending)
2440 __queue_exception(vcpu);
2441 else if (irqchip_in_kernel(vcpu->kvm))
2442 kvm_x86_ops->inject_pending_irq(vcpu);
2444 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2446 vcpu->guest_mode = 1;
2450 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2451 kvm_x86_ops->tlb_flush(vcpu);
2453 kvm_x86_ops->run(vcpu, kvm_run);
2455 vcpu->guest_mode = 0;
2461 * We must have an instruction between local_irq_enable() and
2462 * kvm_guest_exit(), so the timer interrupt isn't delayed by
2463 * the interrupt shadow. The stat.exits increment will do nicely.
2464 * But we need to prevent reordering, hence this barrier():
2473 * Profile KVM exit RIPs:
2475 if (unlikely(prof_on == KVM_PROFILING)) {
2476 kvm_x86_ops->cache_regs(vcpu);
2477 profile_hit(KVM_PROFILING, (void *)vcpu->arch.rip);
2480 if (vcpu->arch.exception.pending && kvm_x86_ops->exception_injected(vcpu))
2481 vcpu->arch.exception.pending = false;
2483 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2486 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2488 kvm_run->exit_reason = KVM_EXIT_INTR;
2489 ++vcpu->stat.request_irq_exits;
2492 if (!need_resched())
2502 post_kvm_run_save(vcpu, kvm_run);
2507 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2514 if (unlikely(vcpu->arch.mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2515 kvm_vcpu_block(vcpu);
2520 if (vcpu->sigset_active)
2521 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2523 /* re-sync apic's tpr */
2524 if (!irqchip_in_kernel(vcpu->kvm))
2525 set_cr8(vcpu, kvm_run->cr8);
2527 if (vcpu->arch.pio.cur_count) {
2528 r = complete_pio(vcpu);
2532 #if CONFIG_HAS_IOMEM
2533 if (vcpu->mmio_needed) {
2534 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2535 vcpu->mmio_read_completed = 1;
2536 vcpu->mmio_needed = 0;
2537 r = emulate_instruction(vcpu, kvm_run,
2538 vcpu->arch.mmio_fault_cr2, 0, 1);
2539 if (r == EMULATE_DO_MMIO) {
2541 * Read-modify-write. Back to userspace.
2548 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2549 kvm_x86_ops->cache_regs(vcpu);
2550 vcpu->arch.regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2551 kvm_x86_ops->decache_regs(vcpu);
2554 r = __vcpu_run(vcpu, kvm_run);
2557 if (vcpu->sigset_active)
2558 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2564 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2568 kvm_x86_ops->cache_regs(vcpu);
2570 regs->rax = vcpu->arch.regs[VCPU_REGS_RAX];
2571 regs->rbx = vcpu->arch.regs[VCPU_REGS_RBX];
2572 regs->rcx = vcpu->arch.regs[VCPU_REGS_RCX];
2573 regs->rdx = vcpu->arch.regs[VCPU_REGS_RDX];
2574 regs->rsi = vcpu->arch.regs[VCPU_REGS_RSI];
2575 regs->rdi = vcpu->arch.regs[VCPU_REGS_RDI];
2576 regs->rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2577 regs->rbp = vcpu->arch.regs[VCPU_REGS_RBP];
2578 #ifdef CONFIG_X86_64
2579 regs->r8 = vcpu->arch.regs[VCPU_REGS_R8];
2580 regs->r9 = vcpu->arch.regs[VCPU_REGS_R9];
2581 regs->r10 = vcpu->arch.regs[VCPU_REGS_R10];
2582 regs->r11 = vcpu->arch.regs[VCPU_REGS_R11];
2583 regs->r12 = vcpu->arch.regs[VCPU_REGS_R12];
2584 regs->r13 = vcpu->arch.regs[VCPU_REGS_R13];
2585 regs->r14 = vcpu->arch.regs[VCPU_REGS_R14];
2586 regs->r15 = vcpu->arch.regs[VCPU_REGS_R15];
2589 regs->rip = vcpu->arch.rip;
2590 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2593 * Don't leak debug flags in case they were set for guest debugging
2595 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2596 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2603 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2607 vcpu->arch.regs[VCPU_REGS_RAX] = regs->rax;
2608 vcpu->arch.regs[VCPU_REGS_RBX] = regs->rbx;
2609 vcpu->arch.regs[VCPU_REGS_RCX] = regs->rcx;
2610 vcpu->arch.regs[VCPU_REGS_RDX] = regs->rdx;
2611 vcpu->arch.regs[VCPU_REGS_RSI] = regs->rsi;
2612 vcpu->arch.regs[VCPU_REGS_RDI] = regs->rdi;
2613 vcpu->arch.regs[VCPU_REGS_RSP] = regs->rsp;
2614 vcpu->arch.regs[VCPU_REGS_RBP] = regs->rbp;
2615 #ifdef CONFIG_X86_64
2616 vcpu->arch.regs[VCPU_REGS_R8] = regs->r8;
2617 vcpu->arch.regs[VCPU_REGS_R9] = regs->r9;
2618 vcpu->arch.regs[VCPU_REGS_R10] = regs->r10;
2619 vcpu->arch.regs[VCPU_REGS_R11] = regs->r11;
2620 vcpu->arch.regs[VCPU_REGS_R12] = regs->r12;
2621 vcpu->arch.regs[VCPU_REGS_R13] = regs->r13;
2622 vcpu->arch.regs[VCPU_REGS_R14] = regs->r14;
2623 vcpu->arch.regs[VCPU_REGS_R15] = regs->r15;
2626 vcpu->arch.rip = regs->rip;
2627 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2629 kvm_x86_ops->decache_regs(vcpu);
2636 static void get_segment(struct kvm_vcpu *vcpu,
2637 struct kvm_segment *var, int seg)
2639 return kvm_x86_ops->get_segment(vcpu, var, seg);
2642 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2644 struct kvm_segment cs;
2646 get_segment(vcpu, &cs, VCPU_SREG_CS);
2650 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2652 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2653 struct kvm_sregs *sregs)
2655 struct descriptor_table dt;
2660 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2661 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2662 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2663 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2664 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2665 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2667 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2668 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2670 kvm_x86_ops->get_idt(vcpu, &dt);
2671 sregs->idt.limit = dt.limit;
2672 sregs->idt.base = dt.base;
2673 kvm_x86_ops->get_gdt(vcpu, &dt);
2674 sregs->gdt.limit = dt.limit;
2675 sregs->gdt.base = dt.base;
2677 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2678 sregs->cr0 = vcpu->arch.cr0;
2679 sregs->cr2 = vcpu->arch.cr2;
2680 sregs->cr3 = vcpu->arch.cr3;
2681 sregs->cr4 = vcpu->arch.cr4;
2682 sregs->cr8 = get_cr8(vcpu);
2683 sregs->efer = vcpu->arch.shadow_efer;
2684 sregs->apic_base = kvm_get_apic_base(vcpu);
2686 if (irqchip_in_kernel(vcpu->kvm)) {
2687 memset(sregs->interrupt_bitmap, 0,
2688 sizeof sregs->interrupt_bitmap);
2689 pending_vec = kvm_x86_ops->get_irq(vcpu);
2690 if (pending_vec >= 0)
2691 set_bit(pending_vec,
2692 (unsigned long *)sregs->interrupt_bitmap);
2694 memcpy(sregs->interrupt_bitmap, vcpu->arch.irq_pending,
2695 sizeof sregs->interrupt_bitmap);
2702 static void set_segment(struct kvm_vcpu *vcpu,
2703 struct kvm_segment *var, int seg)
2705 return kvm_x86_ops->set_segment(vcpu, var, seg);
2708 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2709 struct kvm_sregs *sregs)
2711 int mmu_reset_needed = 0;
2712 int i, pending_vec, max_bits;
2713 struct descriptor_table dt;
2717 dt.limit = sregs->idt.limit;
2718 dt.base = sregs->idt.base;
2719 kvm_x86_ops->set_idt(vcpu, &dt);
2720 dt.limit = sregs->gdt.limit;
2721 dt.base = sregs->gdt.base;
2722 kvm_x86_ops->set_gdt(vcpu, &dt);
2724 vcpu->arch.cr2 = sregs->cr2;
2725 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
2726 vcpu->arch.cr3 = sregs->cr3;
2728 set_cr8(vcpu, sregs->cr8);
2730 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
2731 #ifdef CONFIG_X86_64
2732 kvm_x86_ops->set_efer(vcpu, sregs->efer);
2734 kvm_set_apic_base(vcpu, sregs->apic_base);
2736 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2738 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
2739 vcpu->arch.cr0 = sregs->cr0;
2740 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2742 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
2743 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2744 if (!is_long_mode(vcpu) && is_pae(vcpu))
2745 load_pdptrs(vcpu, vcpu->arch.cr3);
2747 if (mmu_reset_needed)
2748 kvm_mmu_reset_context(vcpu);
2750 if (!irqchip_in_kernel(vcpu->kvm)) {
2751 memcpy(vcpu->arch.irq_pending, sregs->interrupt_bitmap,
2752 sizeof vcpu->arch.irq_pending);
2753 vcpu->arch.irq_summary = 0;
2754 for (i = 0; i < ARRAY_SIZE(vcpu->arch.irq_pending); ++i)
2755 if (vcpu->arch.irq_pending[i])
2756 __set_bit(i, &vcpu->arch.irq_summary);
2758 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2759 pending_vec = find_first_bit(
2760 (const unsigned long *)sregs->interrupt_bitmap,
2762 /* Only pending external irq is handled here */
2763 if (pending_vec < max_bits) {
2764 kvm_x86_ops->set_irq(vcpu, pending_vec);
2765 pr_debug("Set back pending irq %d\n",
2770 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2771 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2772 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2773 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2774 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2775 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2777 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2778 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2785 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2786 struct kvm_debug_guest *dbg)
2792 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2800 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2801 * we have asm/x86/processor.h
2812 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2813 #ifdef CONFIG_X86_64
2814 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2816 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2821 * Translate a guest virtual address to a guest physical address.
2823 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2824 struct kvm_translation *tr)
2826 unsigned long vaddr = tr->linear_address;
2830 mutex_lock(&vcpu->kvm->lock);
2831 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
2832 tr->physical_address = gpa;
2833 tr->valid = gpa != UNMAPPED_GVA;
2836 mutex_unlock(&vcpu->kvm->lock);
2842 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2844 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
2848 memcpy(fpu->fpr, fxsave->st_space, 128);
2849 fpu->fcw = fxsave->cwd;
2850 fpu->fsw = fxsave->swd;
2851 fpu->ftwx = fxsave->twd;
2852 fpu->last_opcode = fxsave->fop;
2853 fpu->last_ip = fxsave->rip;
2854 fpu->last_dp = fxsave->rdp;
2855 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2862 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2864 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
2868 memcpy(fxsave->st_space, fpu->fpr, 128);
2869 fxsave->cwd = fpu->fcw;
2870 fxsave->swd = fpu->fsw;
2871 fxsave->twd = fpu->ftwx;
2872 fxsave->fop = fpu->last_opcode;
2873 fxsave->rip = fpu->last_ip;
2874 fxsave->rdp = fpu->last_dp;
2875 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2882 void fx_init(struct kvm_vcpu *vcpu)
2884 unsigned after_mxcsr_mask;
2886 /* Initialize guest FPU by resetting ours and saving into guest's */
2888 fx_save(&vcpu->arch.host_fx_image);
2890 fx_save(&vcpu->arch.guest_fx_image);
2891 fx_restore(&vcpu->arch.host_fx_image);
2894 vcpu->arch.cr0 |= X86_CR0_ET;
2895 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
2896 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
2897 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
2898 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
2900 EXPORT_SYMBOL_GPL(fx_init);
2902 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
2904 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
2907 vcpu->guest_fpu_loaded = 1;
2908 fx_save(&vcpu->arch.host_fx_image);
2909 fx_restore(&vcpu->arch.guest_fx_image);
2911 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
2913 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
2915 if (!vcpu->guest_fpu_loaded)
2918 vcpu->guest_fpu_loaded = 0;
2919 fx_save(&vcpu->arch.guest_fx_image);
2920 fx_restore(&vcpu->arch.host_fx_image);
2921 ++vcpu->stat.fpu_reload;
2923 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
2925 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
2927 kvm_x86_ops->vcpu_free(vcpu);
2930 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2933 return kvm_x86_ops->vcpu_create(kvm, id);
2936 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
2940 /* We do fxsave: this must be aligned. */
2941 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
2944 r = kvm_arch_vcpu_reset(vcpu);
2946 r = kvm_mmu_setup(vcpu);
2953 kvm_x86_ops->vcpu_free(vcpu);
2957 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
2960 kvm_mmu_unload(vcpu);
2963 kvm_x86_ops->vcpu_free(vcpu);
2966 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
2968 return kvm_x86_ops->vcpu_reset(vcpu);
2971 void kvm_arch_hardware_enable(void *garbage)
2973 kvm_x86_ops->hardware_enable(garbage);
2976 void kvm_arch_hardware_disable(void *garbage)
2978 kvm_x86_ops->hardware_disable(garbage);
2981 int kvm_arch_hardware_setup(void)
2983 return kvm_x86_ops->hardware_setup();
2986 void kvm_arch_hardware_unsetup(void)
2988 kvm_x86_ops->hardware_unsetup();
2991 void kvm_arch_check_processor_compat(void *rtn)
2993 kvm_x86_ops->check_processor_compatibility(rtn);
2996 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
3002 BUG_ON(vcpu->kvm == NULL);
3005 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3006 if (!irqchip_in_kernel(kvm) || vcpu->vcpu_id == 0)
3007 vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
3009 vcpu->arch.mp_state = VCPU_MP_STATE_UNINITIALIZED;
3011 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3016 vcpu->arch.pio_data = page_address(page);
3018 r = kvm_mmu_create(vcpu);
3020 goto fail_free_pio_data;
3022 if (irqchip_in_kernel(kvm)) {
3023 r = kvm_create_lapic(vcpu);
3025 goto fail_mmu_destroy;
3031 kvm_mmu_destroy(vcpu);
3033 free_page((unsigned long)vcpu->arch.pio_data);
3038 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
3040 kvm_free_lapic(vcpu);
3041 kvm_mmu_destroy(vcpu);
3042 free_page((unsigned long)vcpu->arch.pio_data);
3045 struct kvm *kvm_arch_create_vm(void)
3047 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
3050 return ERR_PTR(-ENOMEM);
3052 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
3057 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
3060 kvm_mmu_unload(vcpu);
3064 static void kvm_free_vcpus(struct kvm *kvm)
3069 * Unpin any mmu pages first.
3071 for (i = 0; i < KVM_MAX_VCPUS; ++i)
3073 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
3074 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3075 if (kvm->vcpus[i]) {
3076 kvm_arch_vcpu_free(kvm->vcpus[i]);
3077 kvm->vcpus[i] = NULL;
3083 void kvm_arch_destroy_vm(struct kvm *kvm)
3085 kfree(kvm->arch.vpic);
3086 kfree(kvm->arch.vioapic);
3087 kvm_free_vcpus(kvm);
3088 kvm_free_physmem(kvm);
3092 int kvm_arch_set_memory_region(struct kvm *kvm,
3093 struct kvm_userspace_memory_region *mem,
3094 struct kvm_memory_slot old,
3097 int npages = mem->memory_size >> PAGE_SHIFT;
3098 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
3100 /*To keep backward compatibility with older userspace,
3101 *x86 needs to hanlde !user_alloc case.
3104 if (npages && !old.rmap) {
3105 down_write(¤t->mm->mmap_sem);
3106 memslot->userspace_addr = do_mmap(NULL, 0,
3108 PROT_READ | PROT_WRITE,
3109 MAP_SHARED | MAP_ANONYMOUS,
3111 up_write(¤t->mm->mmap_sem);
3113 if (IS_ERR((void *)memslot->userspace_addr))
3114 return PTR_ERR((void *)memslot->userspace_addr);
3116 if (!old.user_alloc && old.rmap) {
3119 down_write(¤t->mm->mmap_sem);
3120 ret = do_munmap(current->mm, old.userspace_addr,
3121 old.npages * PAGE_SIZE);
3122 up_write(¤t->mm->mmap_sem);
3125 "kvm_vm_ioctl_set_memory_region: "
3126 "failed to munmap memory\n");
3131 if (!kvm->arch.n_requested_mmu_pages) {
3132 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
3133 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
3136 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
3137 kvm_flush_remote_tlbs(kvm);
3142 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3144 return vcpu->arch.mp_state == VCPU_MP_STATE_RUNNABLE
3145 || vcpu->arch.mp_state == VCPU_MP_STATE_SIPI_RECEIVED;