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.
19 #include "segment_descriptor.h"
22 #include <linux/kvm.h>
24 #include <linux/vmalloc.h>
25 #include <linux/module.h>
27 #include <asm/uaccess.h>
29 #define MAX_IO_MSRS 256
30 #define CR0_RESERVED_BITS \
31 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
32 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
33 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
34 #define CR4_RESERVED_BITS \
35 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
36 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
37 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
38 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
40 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
41 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
43 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
45 struct kvm_stats_debugfs_item debugfs_entries[] = {
46 { "pf_fixed", STAT_OFFSET(pf_fixed) },
47 { "pf_guest", STAT_OFFSET(pf_guest) },
48 { "tlb_flush", STAT_OFFSET(tlb_flush) },
49 { "invlpg", STAT_OFFSET(invlpg) },
50 { "exits", STAT_OFFSET(exits) },
51 { "io_exits", STAT_OFFSET(io_exits) },
52 { "mmio_exits", STAT_OFFSET(mmio_exits) },
53 { "signal_exits", STAT_OFFSET(signal_exits) },
54 { "irq_window", STAT_OFFSET(irq_window_exits) },
55 { "halt_exits", STAT_OFFSET(halt_exits) },
56 { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
57 { "request_irq", STAT_OFFSET(request_irq_exits) },
58 { "irq_exits", STAT_OFFSET(irq_exits) },
59 { "light_exits", STAT_OFFSET(light_exits) },
60 { "efer_reload", STAT_OFFSET(efer_reload) },
65 unsigned long segment_base(u16 selector)
67 struct descriptor_table gdt;
68 struct segment_descriptor *d;
69 unsigned long table_base;
75 asm("sgdt %0" : "=m"(gdt));
76 table_base = gdt.base;
78 if (selector & 4) { /* from ldt */
81 asm("sldt %0" : "=g"(ldt_selector));
82 table_base = segment_base(ldt_selector);
84 d = (struct segment_descriptor *)(table_base + (selector & ~7));
85 v = d->base_low | ((unsigned long)d->base_mid << 16) |
86 ((unsigned long)d->base_high << 24);
88 if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
89 v |= ((unsigned long) \
90 ((struct segment_descriptor_64 *)d)->base_higher) << 32;
94 EXPORT_SYMBOL_GPL(segment_base);
96 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
98 if (irqchip_in_kernel(vcpu->kvm))
99 return vcpu->apic_base;
101 return vcpu->apic_base;
103 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
105 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
107 /* TODO: reserve bits check */
108 if (irqchip_in_kernel(vcpu->kvm))
109 kvm_lapic_set_base(vcpu, data);
111 vcpu->apic_base = data;
113 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
115 static void inject_gp(struct kvm_vcpu *vcpu)
117 kvm_x86_ops->inject_gp(vcpu, 0);
121 * Load the pae pdptrs. Return true is they are all valid.
123 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
125 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
126 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
129 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
131 mutex_lock(&vcpu->kvm->lock);
132 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
133 offset * sizeof(u64), sizeof(pdpte));
138 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
139 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
146 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
148 mutex_unlock(&vcpu->kvm->lock);
153 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
155 if (cr0 & CR0_RESERVED_BITS) {
156 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
162 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
163 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
168 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
169 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
170 "and a clear PE flag\n");
175 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
177 if ((vcpu->shadow_efer & EFER_LME)) {
181 printk(KERN_DEBUG "set_cr0: #GP, start paging "
182 "in long mode while PAE is disabled\n");
186 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
188 printk(KERN_DEBUG "set_cr0: #GP, start paging "
189 "in long mode while CS.L == 1\n");
196 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
197 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
205 kvm_x86_ops->set_cr0(vcpu, cr0);
208 mutex_lock(&vcpu->kvm->lock);
209 kvm_mmu_reset_context(vcpu);
210 mutex_unlock(&vcpu->kvm->lock);
213 EXPORT_SYMBOL_GPL(set_cr0);
215 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
217 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
219 EXPORT_SYMBOL_GPL(lmsw);
221 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
223 if (cr4 & CR4_RESERVED_BITS) {
224 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
229 if (is_long_mode(vcpu)) {
230 if (!(cr4 & X86_CR4_PAE)) {
231 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
236 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
237 && !load_pdptrs(vcpu, vcpu->cr3)) {
238 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
243 if (cr4 & X86_CR4_VMXE) {
244 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
248 kvm_x86_ops->set_cr4(vcpu, cr4);
250 mutex_lock(&vcpu->kvm->lock);
251 kvm_mmu_reset_context(vcpu);
252 mutex_unlock(&vcpu->kvm->lock);
254 EXPORT_SYMBOL_GPL(set_cr4);
256 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
258 if (is_long_mode(vcpu)) {
259 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
260 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
266 if (cr3 & CR3_PAE_RESERVED_BITS) {
268 "set_cr3: #GP, reserved bits\n");
272 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
273 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
280 * We don't check reserved bits in nonpae mode, because
281 * this isn't enforced, and VMware depends on this.
285 mutex_lock(&vcpu->kvm->lock);
287 * Does the new cr3 value map to physical memory? (Note, we
288 * catch an invalid cr3 even in real-mode, because it would
289 * cause trouble later on when we turn on paging anyway.)
291 * A real CPU would silently accept an invalid cr3 and would
292 * attempt to use it - with largely undefined (and often hard
293 * to debug) behavior on the guest side.
295 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
299 vcpu->mmu.new_cr3(vcpu);
301 mutex_unlock(&vcpu->kvm->lock);
303 EXPORT_SYMBOL_GPL(set_cr3);
305 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
307 if (cr8 & CR8_RESERVED_BITS) {
308 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
312 if (irqchip_in_kernel(vcpu->kvm))
313 kvm_lapic_set_tpr(vcpu, cr8);
317 EXPORT_SYMBOL_GPL(set_cr8);
319 unsigned long get_cr8(struct kvm_vcpu *vcpu)
321 if (irqchip_in_kernel(vcpu->kvm))
322 return kvm_lapic_get_cr8(vcpu);
326 EXPORT_SYMBOL_GPL(get_cr8);
329 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
330 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
332 * This list is modified at module load time to reflect the
333 * capabilities of the host cpu.
335 static u32 msrs_to_save[] = {
336 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
339 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
341 MSR_IA32_TIME_STAMP_COUNTER,
344 static unsigned num_msrs_to_save;
346 static u32 emulated_msrs[] = {
347 MSR_IA32_MISC_ENABLE,
352 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
354 if (efer & EFER_RESERVED_BITS) {
355 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
362 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
363 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
368 kvm_x86_ops->set_efer(vcpu, efer);
371 efer |= vcpu->shadow_efer & EFER_LMA;
373 vcpu->shadow_efer = efer;
379 * Writes msr value into into the appropriate "register".
380 * Returns 0 on success, non-0 otherwise.
381 * Assumes vcpu_load() was already called.
383 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
385 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
389 * Adapt set_msr() to msr_io()'s calling convention
391 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
393 return kvm_set_msr(vcpu, index, *data);
397 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
402 set_efer(vcpu, data);
405 case MSR_IA32_MC0_STATUS:
406 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
409 case MSR_IA32_MCG_STATUS:
410 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
413 case MSR_IA32_UCODE_REV:
414 case MSR_IA32_UCODE_WRITE:
415 case 0x200 ... 0x2ff: /* MTRRs */
417 case MSR_IA32_APICBASE:
418 kvm_set_apic_base(vcpu, data);
420 case MSR_IA32_MISC_ENABLE:
421 vcpu->ia32_misc_enable_msr = data;
424 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
429 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
433 * Reads an msr value (of 'msr_index') into 'pdata'.
434 * Returns 0 on success, non-0 otherwise.
435 * Assumes vcpu_load() was already called.
437 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
439 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
442 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
447 case 0xc0010010: /* SYSCFG */
448 case 0xc0010015: /* HWCR */
449 case MSR_IA32_PLATFORM_ID:
450 case MSR_IA32_P5_MC_ADDR:
451 case MSR_IA32_P5_MC_TYPE:
452 case MSR_IA32_MC0_CTL:
453 case MSR_IA32_MCG_STATUS:
454 case MSR_IA32_MCG_CAP:
455 case MSR_IA32_MC0_MISC:
456 case MSR_IA32_MC0_MISC+4:
457 case MSR_IA32_MC0_MISC+8:
458 case MSR_IA32_MC0_MISC+12:
459 case MSR_IA32_MC0_MISC+16:
460 case MSR_IA32_UCODE_REV:
461 case MSR_IA32_PERF_STATUS:
462 case MSR_IA32_EBL_CR_POWERON:
465 case 0x200 ... 0x2ff:
468 case 0xcd: /* fsb frequency */
471 case MSR_IA32_APICBASE:
472 data = kvm_get_apic_base(vcpu);
474 case MSR_IA32_MISC_ENABLE:
475 data = vcpu->ia32_misc_enable_msr;
479 data = vcpu->shadow_efer;
483 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
489 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
492 * Read or write a bunch of msrs. All parameters are kernel addresses.
494 * @return number of msrs set successfully.
496 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
497 struct kvm_msr_entry *entries,
498 int (*do_msr)(struct kvm_vcpu *vcpu,
499 unsigned index, u64 *data))
505 for (i = 0; i < msrs->nmsrs; ++i)
506 if (do_msr(vcpu, entries[i].index, &entries[i].data))
515 * Read or write a bunch of msrs. Parameters are user addresses.
517 * @return number of msrs set successfully.
519 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
520 int (*do_msr)(struct kvm_vcpu *vcpu,
521 unsigned index, u64 *data),
524 struct kvm_msrs msrs;
525 struct kvm_msr_entry *entries;
530 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
534 if (msrs.nmsrs >= MAX_IO_MSRS)
538 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
539 entries = vmalloc(size);
544 if (copy_from_user(entries, user_msrs->entries, size))
547 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
552 if (writeback && copy_to_user(user_msrs->entries, entries, size))
563 long kvm_arch_dev_ioctl(struct file *filp,
564 unsigned int ioctl, unsigned long arg)
566 void __user *argp = (void __user *)arg;
570 case KVM_GET_MSR_INDEX_LIST: {
571 struct kvm_msr_list __user *user_msr_list = argp;
572 struct kvm_msr_list msr_list;
576 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
579 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
580 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
583 if (n < num_msrs_to_save)
586 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
587 num_msrs_to_save * sizeof(u32)))
589 if (copy_to_user(user_msr_list->indices
590 + num_msrs_to_save * sizeof(u32),
592 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
604 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
606 kvm_x86_ops->vcpu_load(vcpu, cpu);
609 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
611 kvm_x86_ops->vcpu_put(vcpu);
614 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
618 struct kvm_cpuid_entry *e, *entry;
620 rdmsrl(MSR_EFER, efer);
622 for (i = 0; i < vcpu->cpuid_nent; ++i) {
623 e = &vcpu->cpuid_entries[i];
624 if (e->function == 0x80000001) {
629 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
630 entry->edx &= ~(1 << 20);
631 printk(KERN_INFO "kvm: guest NX capability removed\n");
635 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
636 struct kvm_cpuid *cpuid,
637 struct kvm_cpuid_entry __user *entries)
642 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
645 if (copy_from_user(&vcpu->cpuid_entries, entries,
646 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
648 vcpu->cpuid_nent = cpuid->nent;
649 cpuid_fix_nx_cap(vcpu);
656 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
657 struct kvm_lapic_state *s)
660 memcpy(s->regs, vcpu->apic->regs, sizeof *s);
666 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
667 struct kvm_lapic_state *s)
670 memcpy(vcpu->apic->regs, s->regs, sizeof *s);
671 kvm_apic_post_state_restore(vcpu);
677 long kvm_arch_vcpu_ioctl(struct file *filp,
678 unsigned int ioctl, unsigned long arg)
680 struct kvm_vcpu *vcpu = filp->private_data;
681 void __user *argp = (void __user *)arg;
685 case KVM_GET_LAPIC: {
686 struct kvm_lapic_state lapic;
688 memset(&lapic, 0, sizeof lapic);
689 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
693 if (copy_to_user(argp, &lapic, sizeof lapic))
698 case KVM_SET_LAPIC: {
699 struct kvm_lapic_state lapic;
702 if (copy_from_user(&lapic, argp, sizeof lapic))
704 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
710 case KVM_SET_CPUID: {
711 struct kvm_cpuid __user *cpuid_arg = argp;
712 struct kvm_cpuid cpuid;
715 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
717 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
723 r = msr_io(vcpu, argp, kvm_get_msr, 1);
726 r = msr_io(vcpu, argp, do_set_msr, 0);
735 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
739 if (addr > (unsigned int)(-3 * PAGE_SIZE))
741 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
745 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
746 u32 kvm_nr_mmu_pages)
748 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
751 mutex_lock(&kvm->lock);
753 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
754 kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
756 mutex_unlock(&kvm->lock);
760 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
762 return kvm->n_alloc_mmu_pages;
766 * Set a new alias region. Aliases map a portion of physical memory into
767 * another portion. This is useful for memory windows, for example the PC
770 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
771 struct kvm_memory_alias *alias)
774 struct kvm_mem_alias *p;
777 /* General sanity checks */
778 if (alias->memory_size & (PAGE_SIZE - 1))
780 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
782 if (alias->slot >= KVM_ALIAS_SLOTS)
784 if (alias->guest_phys_addr + alias->memory_size
785 < alias->guest_phys_addr)
787 if (alias->target_phys_addr + alias->memory_size
788 < alias->target_phys_addr)
791 mutex_lock(&kvm->lock);
793 p = &kvm->aliases[alias->slot];
794 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
795 p->npages = alias->memory_size >> PAGE_SHIFT;
796 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
798 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
799 if (kvm->aliases[n - 1].npages)
803 kvm_mmu_zap_all(kvm);
805 mutex_unlock(&kvm->lock);
813 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
818 switch (chip->chip_id) {
819 case KVM_IRQCHIP_PIC_MASTER:
820 memcpy(&chip->chip.pic,
821 &pic_irqchip(kvm)->pics[0],
822 sizeof(struct kvm_pic_state));
824 case KVM_IRQCHIP_PIC_SLAVE:
825 memcpy(&chip->chip.pic,
826 &pic_irqchip(kvm)->pics[1],
827 sizeof(struct kvm_pic_state));
829 case KVM_IRQCHIP_IOAPIC:
830 memcpy(&chip->chip.ioapic,
832 sizeof(struct kvm_ioapic_state));
841 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
846 switch (chip->chip_id) {
847 case KVM_IRQCHIP_PIC_MASTER:
848 memcpy(&pic_irqchip(kvm)->pics[0],
850 sizeof(struct kvm_pic_state));
852 case KVM_IRQCHIP_PIC_SLAVE:
853 memcpy(&pic_irqchip(kvm)->pics[1],
855 sizeof(struct kvm_pic_state));
857 case KVM_IRQCHIP_IOAPIC:
858 memcpy(ioapic_irqchip(kvm),
860 sizeof(struct kvm_ioapic_state));
866 kvm_pic_update_irq(pic_irqchip(kvm));
870 long kvm_arch_vm_ioctl(struct file *filp,
871 unsigned int ioctl, unsigned long arg)
873 struct kvm *kvm = filp->private_data;
874 void __user *argp = (void __user *)arg;
878 case KVM_SET_TSS_ADDR:
879 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
883 case KVM_SET_MEMORY_REGION: {
884 struct kvm_memory_region kvm_mem;
885 struct kvm_userspace_memory_region kvm_userspace_mem;
888 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
890 kvm_userspace_mem.slot = kvm_mem.slot;
891 kvm_userspace_mem.flags = kvm_mem.flags;
892 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
893 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
894 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
899 case KVM_SET_NR_MMU_PAGES:
900 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
904 case KVM_GET_NR_MMU_PAGES:
905 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
907 case KVM_SET_MEMORY_ALIAS: {
908 struct kvm_memory_alias alias;
911 if (copy_from_user(&alias, argp, sizeof alias))
913 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
918 case KVM_CREATE_IRQCHIP:
920 kvm->vpic = kvm_create_pic(kvm);
922 r = kvm_ioapic_init(kvm);
932 struct kvm_irq_level irq_event;
935 if (copy_from_user(&irq_event, argp, sizeof irq_event))
937 if (irqchip_in_kernel(kvm)) {
938 mutex_lock(&kvm->lock);
939 if (irq_event.irq < 16)
940 kvm_pic_set_irq(pic_irqchip(kvm),
943 kvm_ioapic_set_irq(kvm->vioapic,
946 mutex_unlock(&kvm->lock);
951 case KVM_GET_IRQCHIP: {
952 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
953 struct kvm_irqchip chip;
956 if (copy_from_user(&chip, argp, sizeof chip))
959 if (!irqchip_in_kernel(kvm))
961 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
965 if (copy_to_user(argp, &chip, sizeof chip))
970 case KVM_SET_IRQCHIP: {
971 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
972 struct kvm_irqchip chip;
975 if (copy_from_user(&chip, argp, sizeof chip))
978 if (!irqchip_in_kernel(kvm))
980 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
993 static __init void kvm_init_msr_list(void)
998 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
999 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
1002 msrs_to_save[j] = msrs_to_save[i];
1005 num_msrs_to_save = j;
1009 * Only apic need an MMIO device hook, so shortcut now..
1011 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1014 struct kvm_io_device *dev;
1017 dev = &vcpu->apic->dev;
1018 if (dev->in_range(dev, addr))
1025 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1028 struct kvm_io_device *dev;
1030 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1032 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1036 int emulator_read_std(unsigned long addr,
1039 struct kvm_vcpu *vcpu)
1044 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1045 unsigned offset = addr & (PAGE_SIZE-1);
1046 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1049 if (gpa == UNMAPPED_GVA)
1050 return X86EMUL_PROPAGATE_FAULT;
1051 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1053 return X86EMUL_UNHANDLEABLE;
1060 return X86EMUL_CONTINUE;
1062 EXPORT_SYMBOL_GPL(emulator_read_std);
1064 static int emulator_write_std(unsigned long addr,
1067 struct kvm_vcpu *vcpu)
1069 pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1070 return X86EMUL_UNHANDLEABLE;
1073 static int emulator_read_emulated(unsigned long addr,
1076 struct kvm_vcpu *vcpu)
1078 struct kvm_io_device *mmio_dev;
1081 if (vcpu->mmio_read_completed) {
1082 memcpy(val, vcpu->mmio_data, bytes);
1083 vcpu->mmio_read_completed = 0;
1084 return X86EMUL_CONTINUE;
1087 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1089 /* For APIC access vmexit */
1090 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1093 if (emulator_read_std(addr, val, bytes, vcpu)
1094 == X86EMUL_CONTINUE)
1095 return X86EMUL_CONTINUE;
1096 if (gpa == UNMAPPED_GVA)
1097 return X86EMUL_PROPAGATE_FAULT;
1101 * Is this MMIO handled locally?
1103 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1105 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1106 return X86EMUL_CONTINUE;
1109 vcpu->mmio_needed = 1;
1110 vcpu->mmio_phys_addr = gpa;
1111 vcpu->mmio_size = bytes;
1112 vcpu->mmio_is_write = 0;
1114 return X86EMUL_UNHANDLEABLE;
1117 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1118 const void *val, int bytes)
1122 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1125 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1129 static int emulator_write_emulated_onepage(unsigned long addr,
1132 struct kvm_vcpu *vcpu)
1134 struct kvm_io_device *mmio_dev;
1135 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1137 if (gpa == UNMAPPED_GVA) {
1138 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1139 return X86EMUL_PROPAGATE_FAULT;
1142 /* For APIC access vmexit */
1143 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1146 if (emulator_write_phys(vcpu, gpa, val, bytes))
1147 return X86EMUL_CONTINUE;
1151 * Is this MMIO handled locally?
1153 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1155 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1156 return X86EMUL_CONTINUE;
1159 vcpu->mmio_needed = 1;
1160 vcpu->mmio_phys_addr = gpa;
1161 vcpu->mmio_size = bytes;
1162 vcpu->mmio_is_write = 1;
1163 memcpy(vcpu->mmio_data, val, bytes);
1165 return X86EMUL_CONTINUE;
1168 int emulator_write_emulated(unsigned long addr,
1171 struct kvm_vcpu *vcpu)
1173 /* Crossing a page boundary? */
1174 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1177 now = -addr & ~PAGE_MASK;
1178 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1179 if (rc != X86EMUL_CONTINUE)
1185 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1187 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1189 static int emulator_cmpxchg_emulated(unsigned long addr,
1193 struct kvm_vcpu *vcpu)
1195 static int reported;
1199 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1201 return emulator_write_emulated(addr, new, bytes, vcpu);
1204 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1206 return kvm_x86_ops->get_segment_base(vcpu, seg);
1209 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1211 return X86EMUL_CONTINUE;
1214 int emulate_clts(struct kvm_vcpu *vcpu)
1216 kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1217 return X86EMUL_CONTINUE;
1220 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1222 struct kvm_vcpu *vcpu = ctxt->vcpu;
1226 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1227 return X86EMUL_CONTINUE;
1229 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1230 return X86EMUL_UNHANDLEABLE;
1234 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1236 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1239 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1241 /* FIXME: better handling */
1242 return X86EMUL_UNHANDLEABLE;
1244 return X86EMUL_CONTINUE;
1247 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1249 static int reported;
1251 unsigned long rip = vcpu->rip;
1252 unsigned long rip_linear;
1254 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1259 emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1261 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1262 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1265 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1267 struct x86_emulate_ops emulate_ops = {
1268 .read_std = emulator_read_std,
1269 .write_std = emulator_write_std,
1270 .read_emulated = emulator_read_emulated,
1271 .write_emulated = emulator_write_emulated,
1272 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1275 int emulate_instruction(struct kvm_vcpu *vcpu,
1276 struct kvm_run *run,
1283 vcpu->mmio_fault_cr2 = cr2;
1284 kvm_x86_ops->cache_regs(vcpu);
1286 vcpu->mmio_is_write = 0;
1287 vcpu->pio.string = 0;
1291 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1293 vcpu->emulate_ctxt.vcpu = vcpu;
1294 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1295 vcpu->emulate_ctxt.cr2 = cr2;
1296 vcpu->emulate_ctxt.mode =
1297 (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1298 ? X86EMUL_MODE_REAL : cs_l
1299 ? X86EMUL_MODE_PROT64 : cs_db
1300 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1302 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1303 vcpu->emulate_ctxt.cs_base = 0;
1304 vcpu->emulate_ctxt.ds_base = 0;
1305 vcpu->emulate_ctxt.es_base = 0;
1306 vcpu->emulate_ctxt.ss_base = 0;
1308 vcpu->emulate_ctxt.cs_base =
1309 get_segment_base(vcpu, VCPU_SREG_CS);
1310 vcpu->emulate_ctxt.ds_base =
1311 get_segment_base(vcpu, VCPU_SREG_DS);
1312 vcpu->emulate_ctxt.es_base =
1313 get_segment_base(vcpu, VCPU_SREG_ES);
1314 vcpu->emulate_ctxt.ss_base =
1315 get_segment_base(vcpu, VCPU_SREG_SS);
1318 vcpu->emulate_ctxt.gs_base =
1319 get_segment_base(vcpu, VCPU_SREG_GS);
1320 vcpu->emulate_ctxt.fs_base =
1321 get_segment_base(vcpu, VCPU_SREG_FS);
1323 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1325 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1326 return EMULATE_DONE;
1327 return EMULATE_FAIL;
1331 r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1333 if (vcpu->pio.string)
1334 return EMULATE_DO_MMIO;
1336 if ((r || vcpu->mmio_is_write) && run) {
1337 run->exit_reason = KVM_EXIT_MMIO;
1338 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1339 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1340 run->mmio.len = vcpu->mmio_size;
1341 run->mmio.is_write = vcpu->mmio_is_write;
1345 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1346 return EMULATE_DONE;
1347 if (!vcpu->mmio_needed) {
1348 kvm_report_emulation_failure(vcpu, "mmio");
1349 return EMULATE_FAIL;
1351 return EMULATE_DO_MMIO;
1354 kvm_x86_ops->decache_regs(vcpu);
1355 kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1357 if (vcpu->mmio_is_write) {
1358 vcpu->mmio_needed = 0;
1359 return EMULATE_DO_MMIO;
1362 return EMULATE_DONE;
1364 EXPORT_SYMBOL_GPL(emulate_instruction);
1366 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1370 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
1371 if (vcpu->pio.guest_pages[i]) {
1372 kvm_release_page(vcpu->pio.guest_pages[i]);
1373 vcpu->pio.guest_pages[i] = NULL;
1377 static int pio_copy_data(struct kvm_vcpu *vcpu)
1379 void *p = vcpu->pio_data;
1382 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1384 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1387 free_pio_guest_pages(vcpu);
1390 q += vcpu->pio.guest_page_offset;
1391 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1393 memcpy(q, p, bytes);
1395 memcpy(p, q, bytes);
1396 q -= vcpu->pio.guest_page_offset;
1398 free_pio_guest_pages(vcpu);
1402 int complete_pio(struct kvm_vcpu *vcpu)
1404 struct kvm_pio_request *io = &vcpu->pio;
1408 kvm_x86_ops->cache_regs(vcpu);
1412 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1416 r = pio_copy_data(vcpu);
1418 kvm_x86_ops->cache_regs(vcpu);
1425 delta *= io->cur_count;
1427 * The size of the register should really depend on
1428 * current address size.
1430 vcpu->regs[VCPU_REGS_RCX] -= delta;
1436 vcpu->regs[VCPU_REGS_RDI] += delta;
1438 vcpu->regs[VCPU_REGS_RSI] += delta;
1441 kvm_x86_ops->decache_regs(vcpu);
1443 io->count -= io->cur_count;
1449 static void kernel_pio(struct kvm_io_device *pio_dev,
1450 struct kvm_vcpu *vcpu,
1453 /* TODO: String I/O for in kernel device */
1455 mutex_lock(&vcpu->kvm->lock);
1457 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1461 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1464 mutex_unlock(&vcpu->kvm->lock);
1467 static void pio_string_write(struct kvm_io_device *pio_dev,
1468 struct kvm_vcpu *vcpu)
1470 struct kvm_pio_request *io = &vcpu->pio;
1471 void *pd = vcpu->pio_data;
1474 mutex_lock(&vcpu->kvm->lock);
1475 for (i = 0; i < io->cur_count; i++) {
1476 kvm_iodevice_write(pio_dev, io->port,
1481 mutex_unlock(&vcpu->kvm->lock);
1484 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1487 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1490 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1491 int size, unsigned port)
1493 struct kvm_io_device *pio_dev;
1495 vcpu->run->exit_reason = KVM_EXIT_IO;
1496 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1497 vcpu->run->io.size = vcpu->pio.size = size;
1498 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1499 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1500 vcpu->run->io.port = vcpu->pio.port = port;
1502 vcpu->pio.string = 0;
1504 vcpu->pio.guest_page_offset = 0;
1507 kvm_x86_ops->cache_regs(vcpu);
1508 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1509 kvm_x86_ops->decache_regs(vcpu);
1511 kvm_x86_ops->skip_emulated_instruction(vcpu);
1513 pio_dev = vcpu_find_pio_dev(vcpu, port);
1515 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1521 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1523 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1524 int size, unsigned long count, int down,
1525 gva_t address, int rep, unsigned port)
1527 unsigned now, in_page;
1531 struct kvm_io_device *pio_dev;
1533 vcpu->run->exit_reason = KVM_EXIT_IO;
1534 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1535 vcpu->run->io.size = vcpu->pio.size = size;
1536 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1537 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1538 vcpu->run->io.port = vcpu->pio.port = port;
1540 vcpu->pio.string = 1;
1541 vcpu->pio.down = down;
1542 vcpu->pio.guest_page_offset = offset_in_page(address);
1543 vcpu->pio.rep = rep;
1546 kvm_x86_ops->skip_emulated_instruction(vcpu);
1551 in_page = PAGE_SIZE - offset_in_page(address);
1553 in_page = offset_in_page(address) + size;
1554 now = min(count, (unsigned long)in_page / size);
1557 * String I/O straddles page boundary. Pin two guest pages
1558 * so that we satisfy atomicity constraints. Do just one
1559 * transaction to avoid complexity.
1566 * String I/O in reverse. Yuck. Kill the guest, fix later.
1568 pr_unimpl(vcpu, "guest string pio down\n");
1572 vcpu->run->io.count = now;
1573 vcpu->pio.cur_count = now;
1575 if (vcpu->pio.cur_count == vcpu->pio.count)
1576 kvm_x86_ops->skip_emulated_instruction(vcpu);
1578 for (i = 0; i < nr_pages; ++i) {
1579 mutex_lock(&vcpu->kvm->lock);
1580 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1581 vcpu->pio.guest_pages[i] = page;
1582 mutex_unlock(&vcpu->kvm->lock);
1585 free_pio_guest_pages(vcpu);
1590 pio_dev = vcpu_find_pio_dev(vcpu, port);
1591 if (!vcpu->pio.in) {
1592 /* string PIO write */
1593 ret = pio_copy_data(vcpu);
1594 if (ret >= 0 && pio_dev) {
1595 pio_string_write(pio_dev, vcpu);
1597 if (vcpu->pio.count == 0)
1601 pr_unimpl(vcpu, "no string pio read support yet, "
1602 "port %x size %d count %ld\n",
1607 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1609 __init void kvm_arch_init(void)
1611 kvm_init_msr_list();
1614 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1616 ++vcpu->stat.halt_exits;
1617 if (irqchip_in_kernel(vcpu->kvm)) {
1618 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1619 kvm_vcpu_block(vcpu);
1620 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1624 vcpu->run->exit_reason = KVM_EXIT_HLT;
1628 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1630 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1632 unsigned long nr, a0, a1, a2, a3, ret;
1634 kvm_x86_ops->cache_regs(vcpu);
1636 nr = vcpu->regs[VCPU_REGS_RAX];
1637 a0 = vcpu->regs[VCPU_REGS_RBX];
1638 a1 = vcpu->regs[VCPU_REGS_RCX];
1639 a2 = vcpu->regs[VCPU_REGS_RDX];
1640 a3 = vcpu->regs[VCPU_REGS_RSI];
1642 if (!is_long_mode(vcpu)) {
1655 vcpu->regs[VCPU_REGS_RAX] = ret;
1656 kvm_x86_ops->decache_regs(vcpu);
1659 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1661 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1663 char instruction[3];
1666 mutex_lock(&vcpu->kvm->lock);
1669 * Blow out the MMU to ensure that no other VCPU has an active mapping
1670 * to ensure that the updated hypercall appears atomically across all
1673 kvm_mmu_zap_all(vcpu->kvm);
1675 kvm_x86_ops->cache_regs(vcpu);
1676 kvm_x86_ops->patch_hypercall(vcpu, instruction);
1677 if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1678 != X86EMUL_CONTINUE)
1681 mutex_unlock(&vcpu->kvm->lock);
1686 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1688 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1691 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1693 struct descriptor_table dt = { limit, base };
1695 kvm_x86_ops->set_gdt(vcpu, &dt);
1698 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1700 struct descriptor_table dt = { limit, base };
1702 kvm_x86_ops->set_idt(vcpu, &dt);
1705 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1706 unsigned long *rflags)
1709 *rflags = kvm_x86_ops->get_rflags(vcpu);
1712 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1714 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1725 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1730 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1731 unsigned long *rflags)
1735 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1736 *rflags = kvm_x86_ops->get_rflags(vcpu);
1745 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1748 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1752 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1756 struct kvm_cpuid_entry *e, *best;
1758 kvm_x86_ops->cache_regs(vcpu);
1759 function = vcpu->regs[VCPU_REGS_RAX];
1760 vcpu->regs[VCPU_REGS_RAX] = 0;
1761 vcpu->regs[VCPU_REGS_RBX] = 0;
1762 vcpu->regs[VCPU_REGS_RCX] = 0;
1763 vcpu->regs[VCPU_REGS_RDX] = 0;
1765 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1766 e = &vcpu->cpuid_entries[i];
1767 if (e->function == function) {
1772 * Both basic or both extended?
1774 if (((e->function ^ function) & 0x80000000) == 0)
1775 if (!best || e->function > best->function)
1779 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1780 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1781 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1782 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1784 kvm_x86_ops->decache_regs(vcpu);
1785 kvm_x86_ops->skip_emulated_instruction(vcpu);
1787 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1790 * Check if userspace requested an interrupt window, and that the
1791 * interrupt window is open.
1793 * No need to exit to userspace if we already have an interrupt queued.
1795 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1796 struct kvm_run *kvm_run)
1798 return (!vcpu->irq_summary &&
1799 kvm_run->request_interrupt_window &&
1800 vcpu->interrupt_window_open &&
1801 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1804 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1805 struct kvm_run *kvm_run)
1807 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1808 kvm_run->cr8 = get_cr8(vcpu);
1809 kvm_run->apic_base = kvm_get_apic_base(vcpu);
1810 if (irqchip_in_kernel(vcpu->kvm))
1811 kvm_run->ready_for_interrupt_injection = 1;
1813 kvm_run->ready_for_interrupt_injection =
1814 (vcpu->interrupt_window_open &&
1815 vcpu->irq_summary == 0);
1818 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1822 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1823 pr_debug("vcpu %d received sipi with vector # %x\n",
1824 vcpu->vcpu_id, vcpu->sipi_vector);
1825 kvm_lapic_reset(vcpu);
1826 r = kvm_x86_ops->vcpu_reset(vcpu);
1829 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1833 if (vcpu->guest_debug.enabled)
1834 kvm_x86_ops->guest_debug_pre(vcpu);
1837 r = kvm_mmu_reload(vcpu);
1841 kvm_inject_pending_timer_irqs(vcpu);
1845 kvm_x86_ops->prepare_guest_switch(vcpu);
1846 kvm_load_guest_fpu(vcpu);
1848 local_irq_disable();
1850 if (signal_pending(current)) {
1854 kvm_run->exit_reason = KVM_EXIT_INTR;
1855 ++vcpu->stat.signal_exits;
1859 if (irqchip_in_kernel(vcpu->kvm))
1860 kvm_x86_ops->inject_pending_irq(vcpu);
1861 else if (!vcpu->mmio_read_completed)
1862 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1864 vcpu->guest_mode = 1;
1868 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
1869 kvm_x86_ops->tlb_flush(vcpu);
1871 kvm_x86_ops->run(vcpu, kvm_run);
1873 vcpu->guest_mode = 0;
1879 * We must have an instruction between local_irq_enable() and
1880 * kvm_guest_exit(), so the timer interrupt isn't delayed by
1881 * the interrupt shadow. The stat.exits increment will do nicely.
1882 * But we need to prevent reordering, hence this barrier():
1891 * Profile KVM exit RIPs:
1893 if (unlikely(prof_on == KVM_PROFILING)) {
1894 kvm_x86_ops->cache_regs(vcpu);
1895 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
1898 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
1901 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1903 kvm_run->exit_reason = KVM_EXIT_INTR;
1904 ++vcpu->stat.request_irq_exits;
1907 if (!need_resched()) {
1908 ++vcpu->stat.light_exits;
1919 post_kvm_run_save(vcpu, kvm_run);
1924 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1931 if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
1932 kvm_vcpu_block(vcpu);
1937 if (vcpu->sigset_active)
1938 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1940 /* re-sync apic's tpr */
1941 if (!irqchip_in_kernel(vcpu->kvm))
1942 set_cr8(vcpu, kvm_run->cr8);
1944 if (vcpu->pio.cur_count) {
1945 r = complete_pio(vcpu);
1949 #if CONFIG_HAS_IOMEM
1950 if (vcpu->mmio_needed) {
1951 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1952 vcpu->mmio_read_completed = 1;
1953 vcpu->mmio_needed = 0;
1954 r = emulate_instruction(vcpu, kvm_run,
1955 vcpu->mmio_fault_cr2, 0, 1);
1956 if (r == EMULATE_DO_MMIO) {
1958 * Read-modify-write. Back to userspace.
1965 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1966 kvm_x86_ops->cache_regs(vcpu);
1967 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1968 kvm_x86_ops->decache_regs(vcpu);
1971 r = __vcpu_run(vcpu, kvm_run);
1974 if (vcpu->sigset_active)
1975 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1981 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1985 kvm_x86_ops->cache_regs(vcpu);
1987 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1988 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1989 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1990 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1991 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1992 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1993 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1994 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1995 #ifdef CONFIG_X86_64
1996 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1997 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1998 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1999 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2000 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2001 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2002 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2003 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2006 regs->rip = vcpu->rip;
2007 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2010 * Don't leak debug flags in case they were set for guest debugging
2012 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2013 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2020 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2024 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2025 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2026 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2027 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2028 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2029 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2030 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2031 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2032 #ifdef CONFIG_X86_64
2033 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2034 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2035 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2036 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2037 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2038 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2039 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2040 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2043 vcpu->rip = regs->rip;
2044 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2046 kvm_x86_ops->decache_regs(vcpu);
2053 static void get_segment(struct kvm_vcpu *vcpu,
2054 struct kvm_segment *var, int seg)
2056 return kvm_x86_ops->get_segment(vcpu, var, seg);
2059 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2061 struct kvm_segment cs;
2063 get_segment(vcpu, &cs, VCPU_SREG_CS);
2067 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2069 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2070 struct kvm_sregs *sregs)
2072 struct descriptor_table dt;
2077 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2078 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2079 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2080 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2081 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2082 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2084 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2085 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2087 kvm_x86_ops->get_idt(vcpu, &dt);
2088 sregs->idt.limit = dt.limit;
2089 sregs->idt.base = dt.base;
2090 kvm_x86_ops->get_gdt(vcpu, &dt);
2091 sregs->gdt.limit = dt.limit;
2092 sregs->gdt.base = dt.base;
2094 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2095 sregs->cr0 = vcpu->cr0;
2096 sregs->cr2 = vcpu->cr2;
2097 sregs->cr3 = vcpu->cr3;
2098 sregs->cr4 = vcpu->cr4;
2099 sregs->cr8 = get_cr8(vcpu);
2100 sregs->efer = vcpu->shadow_efer;
2101 sregs->apic_base = kvm_get_apic_base(vcpu);
2103 if (irqchip_in_kernel(vcpu->kvm)) {
2104 memset(sregs->interrupt_bitmap, 0,
2105 sizeof sregs->interrupt_bitmap);
2106 pending_vec = kvm_x86_ops->get_irq(vcpu);
2107 if (pending_vec >= 0)
2108 set_bit(pending_vec,
2109 (unsigned long *)sregs->interrupt_bitmap);
2111 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2112 sizeof sregs->interrupt_bitmap);
2119 static void set_segment(struct kvm_vcpu *vcpu,
2120 struct kvm_segment *var, int seg)
2122 return kvm_x86_ops->set_segment(vcpu, var, seg);
2125 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2126 struct kvm_sregs *sregs)
2128 int mmu_reset_needed = 0;
2129 int i, pending_vec, max_bits;
2130 struct descriptor_table dt;
2134 dt.limit = sregs->idt.limit;
2135 dt.base = sregs->idt.base;
2136 kvm_x86_ops->set_idt(vcpu, &dt);
2137 dt.limit = sregs->gdt.limit;
2138 dt.base = sregs->gdt.base;
2139 kvm_x86_ops->set_gdt(vcpu, &dt);
2141 vcpu->cr2 = sregs->cr2;
2142 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2143 vcpu->cr3 = sregs->cr3;
2145 set_cr8(vcpu, sregs->cr8);
2147 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2148 #ifdef CONFIG_X86_64
2149 kvm_x86_ops->set_efer(vcpu, sregs->efer);
2151 kvm_set_apic_base(vcpu, sregs->apic_base);
2153 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2155 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2156 vcpu->cr0 = sregs->cr0;
2157 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2159 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2160 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2161 if (!is_long_mode(vcpu) && is_pae(vcpu))
2162 load_pdptrs(vcpu, vcpu->cr3);
2164 if (mmu_reset_needed)
2165 kvm_mmu_reset_context(vcpu);
2167 if (!irqchip_in_kernel(vcpu->kvm)) {
2168 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2169 sizeof vcpu->irq_pending);
2170 vcpu->irq_summary = 0;
2171 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2172 if (vcpu->irq_pending[i])
2173 __set_bit(i, &vcpu->irq_summary);
2175 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2176 pending_vec = find_first_bit(
2177 (const unsigned long *)sregs->interrupt_bitmap,
2179 /* Only pending external irq is handled here */
2180 if (pending_vec < max_bits) {
2181 kvm_x86_ops->set_irq(vcpu, pending_vec);
2182 pr_debug("Set back pending irq %d\n",
2187 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2188 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2189 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2190 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2191 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2192 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2194 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2195 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2202 int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2203 struct kvm_debug_guest *dbg)
2209 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2217 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2218 * we have asm/x86/processor.h
2229 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2230 #ifdef CONFIG_X86_64
2231 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2233 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2237 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2239 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2243 memcpy(fpu->fpr, fxsave->st_space, 128);
2244 fpu->fcw = fxsave->cwd;
2245 fpu->fsw = fxsave->swd;
2246 fpu->ftwx = fxsave->twd;
2247 fpu->last_opcode = fxsave->fop;
2248 fpu->last_ip = fxsave->rip;
2249 fpu->last_dp = fxsave->rdp;
2250 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2257 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2259 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2263 memcpy(fxsave->st_space, fpu->fpr, 128);
2264 fxsave->cwd = fpu->fcw;
2265 fxsave->swd = fpu->fsw;
2266 fxsave->twd = fpu->ftwx;
2267 fxsave->fop = fpu->last_opcode;
2268 fxsave->rip = fpu->last_ip;
2269 fxsave->rdp = fpu->last_dp;
2270 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2277 void fx_init(struct kvm_vcpu *vcpu)
2279 unsigned after_mxcsr_mask;
2281 /* Initialize guest FPU by resetting ours and saving into guest's */
2283 fx_save(&vcpu->host_fx_image);
2285 fx_save(&vcpu->guest_fx_image);
2286 fx_restore(&vcpu->host_fx_image);
2289 vcpu->cr0 |= X86_CR0_ET;
2290 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
2291 vcpu->guest_fx_image.mxcsr = 0x1f80;
2292 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
2293 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
2295 EXPORT_SYMBOL_GPL(fx_init);
2297 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
2299 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
2302 vcpu->guest_fpu_loaded = 1;
2303 fx_save(&vcpu->host_fx_image);
2304 fx_restore(&vcpu->guest_fx_image);
2306 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
2308 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
2310 if (!vcpu->guest_fpu_loaded)
2313 vcpu->guest_fpu_loaded = 0;
2314 fx_save(&vcpu->guest_fx_image);
2315 fx_restore(&vcpu->host_fx_image);
2317 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);