2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
19 #include "x86_emulate.h"
20 #include "segment_descriptor.h"
23 #include <linux/kvm.h>
24 #include <linux/module.h>
25 #include <linux/errno.h>
26 #include <linux/percpu.h>
27 #include <linux/gfp.h>
29 #include <linux/miscdevice.h>
30 #include <linux/vmalloc.h>
31 #include <linux/reboot.h>
32 #include <linux/debugfs.h>
33 #include <linux/highmem.h>
34 #include <linux/file.h>
35 #include <linux/sysdev.h>
36 #include <linux/cpu.h>
37 #include <linux/sched.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
42 #include <asm/processor.h>
45 #include <asm/uaccess.h>
48 MODULE_AUTHOR("Qumranet");
49 MODULE_LICENSE("GPL");
51 static DEFINE_SPINLOCK(kvm_lock);
52 static LIST_HEAD(vm_list);
54 static cpumask_t cpus_hardware_enabled;
56 struct kvm_arch_ops *kvm_arch_ops;
57 struct kmem_cache *kvm_vcpu_cache;
58 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
60 static __read_mostly struct preempt_ops kvm_preempt_ops;
62 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
64 static struct kvm_stats_debugfs_item {
67 struct dentry *dentry;
68 } debugfs_entries[] = {
69 { "pf_fixed", STAT_OFFSET(pf_fixed) },
70 { "pf_guest", STAT_OFFSET(pf_guest) },
71 { "tlb_flush", STAT_OFFSET(tlb_flush) },
72 { "invlpg", STAT_OFFSET(invlpg) },
73 { "exits", STAT_OFFSET(exits) },
74 { "io_exits", STAT_OFFSET(io_exits) },
75 { "mmio_exits", STAT_OFFSET(mmio_exits) },
76 { "signal_exits", STAT_OFFSET(signal_exits) },
77 { "irq_window", STAT_OFFSET(irq_window_exits) },
78 { "halt_exits", STAT_OFFSET(halt_exits) },
79 { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
80 { "request_irq", STAT_OFFSET(request_irq_exits) },
81 { "irq_exits", STAT_OFFSET(irq_exits) },
82 { "light_exits", STAT_OFFSET(light_exits) },
83 { "efer_reload", STAT_OFFSET(efer_reload) },
87 static struct dentry *debugfs_dir;
89 #define MAX_IO_MSRS 256
91 #define CR0_RESERVED_BITS \
92 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
93 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
94 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
95 #define CR4_RESERVED_BITS \
96 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
97 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
98 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
99 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
101 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
102 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
105 // LDT or TSS descriptor in the GDT. 16 bytes.
106 struct segment_descriptor_64 {
107 struct segment_descriptor s;
114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
117 unsigned long segment_base(u16 selector)
119 struct descriptor_table gdt;
120 struct segment_descriptor *d;
121 unsigned long table_base;
122 typedef unsigned long ul;
128 asm ("sgdt %0" : "=m"(gdt));
129 table_base = gdt.base;
131 if (selector & 4) { /* from ldt */
134 asm ("sldt %0" : "=g"(ldt_selector));
135 table_base = segment_base(ldt_selector);
137 d = (struct segment_descriptor *)(table_base + (selector & ~7));
138 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
141 && (d->type == 2 || d->type == 9 || d->type == 11))
142 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
146 EXPORT_SYMBOL_GPL(segment_base);
148 static inline int valid_vcpu(int n)
150 return likely(n >= 0 && n < KVM_MAX_VCPUS);
153 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
155 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
158 vcpu->guest_fpu_loaded = 1;
159 fx_save(&vcpu->host_fx_image);
160 fx_restore(&vcpu->guest_fx_image);
162 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
164 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
166 if (!vcpu->guest_fpu_loaded)
169 vcpu->guest_fpu_loaded = 0;
170 fx_save(&vcpu->guest_fx_image);
171 fx_restore(&vcpu->host_fx_image);
173 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
176 * Switches to specified vcpu, until a matching vcpu_put()
178 static void vcpu_load(struct kvm_vcpu *vcpu)
182 mutex_lock(&vcpu->mutex);
184 preempt_notifier_register(&vcpu->preempt_notifier);
185 kvm_arch_ops->vcpu_load(vcpu, cpu);
189 static void vcpu_put(struct kvm_vcpu *vcpu)
192 kvm_arch_ops->vcpu_put(vcpu);
193 preempt_notifier_unregister(&vcpu->preempt_notifier);
195 mutex_unlock(&vcpu->mutex);
198 static void ack_flush(void *_completed)
200 atomic_t *completed = _completed;
202 atomic_inc(completed);
205 void kvm_flush_remote_tlbs(struct kvm *kvm)
209 struct kvm_vcpu *vcpu;
212 atomic_set(&completed, 0);
215 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
216 vcpu = kvm->vcpus[i];
219 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
222 if (cpu != -1 && cpu != raw_smp_processor_id())
223 if (!cpu_isset(cpu, cpus)) {
230 * We really want smp_call_function_mask() here. But that's not
231 * available, so ipi all cpus in parallel and wait for them
234 for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
235 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
236 while (atomic_read(&completed) != needed) {
242 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
247 mutex_init(&vcpu->mutex);
249 vcpu->mmu.root_hpa = INVALID_PAGE;
252 init_waitqueue_head(&vcpu->wq);
254 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
259 vcpu->run = page_address(page);
261 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
266 vcpu->pio_data = page_address(page);
268 r = kvm_mmu_create(vcpu);
270 goto fail_free_pio_data;
275 free_page((unsigned long)vcpu->pio_data);
277 free_page((unsigned long)vcpu->run);
281 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
283 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
285 kvm_mmu_destroy(vcpu);
287 hrtimer_cancel(&vcpu->apic->timer.dev);
288 kvm_free_apic(vcpu->apic);
289 free_page((unsigned long)vcpu->pio_data);
290 free_page((unsigned long)vcpu->run);
292 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
294 static struct kvm *kvm_create_vm(void)
296 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
299 return ERR_PTR(-ENOMEM);
301 kvm_io_bus_init(&kvm->pio_bus);
302 mutex_init(&kvm->lock);
303 INIT_LIST_HEAD(&kvm->active_mmu_pages);
304 kvm_io_bus_init(&kvm->mmio_bus);
305 spin_lock(&kvm_lock);
306 list_add(&kvm->vm_list, &vm_list);
307 spin_unlock(&kvm_lock);
312 * Free any memory in @free but not in @dont.
314 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
315 struct kvm_memory_slot *dont)
319 if (!dont || free->phys_mem != dont->phys_mem)
320 if (free->phys_mem) {
321 for (i = 0; i < free->npages; ++i)
322 if (free->phys_mem[i])
323 __free_page(free->phys_mem[i]);
324 vfree(free->phys_mem);
327 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
328 vfree(free->dirty_bitmap);
330 free->phys_mem = NULL;
332 free->dirty_bitmap = NULL;
335 static void kvm_free_physmem(struct kvm *kvm)
339 for (i = 0; i < kvm->nmemslots; ++i)
340 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
343 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
347 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
348 if (vcpu->pio.guest_pages[i]) {
349 __free_page(vcpu->pio.guest_pages[i]);
350 vcpu->pio.guest_pages[i] = NULL;
354 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
357 kvm_mmu_unload(vcpu);
361 static void kvm_free_vcpus(struct kvm *kvm)
366 * Unpin any mmu pages first.
368 for (i = 0; i < KVM_MAX_VCPUS; ++i)
370 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
371 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
373 kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
374 kvm->vcpus[i] = NULL;
380 static void kvm_destroy_vm(struct kvm *kvm)
382 spin_lock(&kvm_lock);
383 list_del(&kvm->vm_list);
384 spin_unlock(&kvm_lock);
385 kvm_io_bus_destroy(&kvm->pio_bus);
386 kvm_io_bus_destroy(&kvm->mmio_bus);
390 kvm_free_physmem(kvm);
394 static int kvm_vm_release(struct inode *inode, struct file *filp)
396 struct kvm *kvm = filp->private_data;
402 static void inject_gp(struct kvm_vcpu *vcpu)
404 kvm_arch_ops->inject_gp(vcpu, 0);
408 * Load the pae pdptrs. Return true is they are all valid.
410 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
412 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
413 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
418 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
420 mutex_lock(&vcpu->kvm->lock);
421 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
427 pdpt = kmap_atomic(page, KM_USER0);
428 memcpy(pdpte, pdpt+offset, sizeof(pdpte));
429 kunmap_atomic(pdpt, KM_USER0);
431 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
432 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
439 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
441 mutex_unlock(&vcpu->kvm->lock);
446 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
448 if (cr0 & CR0_RESERVED_BITS) {
449 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
455 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
456 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
461 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
462 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
463 "and a clear PE flag\n");
468 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
470 if ((vcpu->shadow_efer & EFER_LME)) {
474 printk(KERN_DEBUG "set_cr0: #GP, start paging "
475 "in long mode while PAE is disabled\n");
479 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
481 printk(KERN_DEBUG "set_cr0: #GP, start paging "
482 "in long mode while CS.L == 1\n");
489 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
490 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
498 kvm_arch_ops->set_cr0(vcpu, cr0);
501 mutex_lock(&vcpu->kvm->lock);
502 kvm_mmu_reset_context(vcpu);
503 mutex_unlock(&vcpu->kvm->lock);
506 EXPORT_SYMBOL_GPL(set_cr0);
508 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
510 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
512 EXPORT_SYMBOL_GPL(lmsw);
514 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
516 if (cr4 & CR4_RESERVED_BITS) {
517 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
522 if (is_long_mode(vcpu)) {
523 if (!(cr4 & X86_CR4_PAE)) {
524 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
529 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
530 && !load_pdptrs(vcpu, vcpu->cr3)) {
531 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
536 if (cr4 & X86_CR4_VMXE) {
537 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
541 kvm_arch_ops->set_cr4(vcpu, cr4);
542 mutex_lock(&vcpu->kvm->lock);
543 kvm_mmu_reset_context(vcpu);
544 mutex_unlock(&vcpu->kvm->lock);
546 EXPORT_SYMBOL_GPL(set_cr4);
548 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
550 if (is_long_mode(vcpu)) {
551 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
552 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
558 if (cr3 & CR3_PAE_RESERVED_BITS) {
560 "set_cr3: #GP, reserved bits\n");
564 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
565 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
571 if (cr3 & CR3_NONPAE_RESERVED_BITS) {
573 "set_cr3: #GP, reserved bits\n");
580 mutex_lock(&vcpu->kvm->lock);
582 * Does the new cr3 value map to physical memory? (Note, we
583 * catch an invalid cr3 even in real-mode, because it would
584 * cause trouble later on when we turn on paging anyway.)
586 * A real CPU would silently accept an invalid cr3 and would
587 * attempt to use it - with largely undefined (and often hard
588 * to debug) behavior on the guest side.
590 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
594 vcpu->mmu.new_cr3(vcpu);
596 mutex_unlock(&vcpu->kvm->lock);
598 EXPORT_SYMBOL_GPL(set_cr3);
600 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
602 if (cr8 & CR8_RESERVED_BITS) {
603 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
607 if (irqchip_in_kernel(vcpu->kvm))
608 kvm_lapic_set_tpr(vcpu, cr8);
612 EXPORT_SYMBOL_GPL(set_cr8);
614 unsigned long get_cr8(struct kvm_vcpu *vcpu)
616 if (irqchip_in_kernel(vcpu->kvm))
617 return kvm_lapic_get_cr8(vcpu);
621 EXPORT_SYMBOL_GPL(get_cr8);
623 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
625 if (irqchip_in_kernel(vcpu->kvm))
626 return vcpu->apic_base;
628 return vcpu->apic_base;
630 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
632 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
634 /* TODO: reserve bits check */
635 if (irqchip_in_kernel(vcpu->kvm))
636 kvm_lapic_set_base(vcpu, data);
638 vcpu->apic_base = data;
640 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
642 void fx_init(struct kvm_vcpu *vcpu)
644 unsigned after_mxcsr_mask;
646 /* Initialize guest FPU by resetting ours and saving into guest's */
648 fx_save(&vcpu->host_fx_image);
650 fx_save(&vcpu->guest_fx_image);
651 fx_restore(&vcpu->host_fx_image);
654 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
655 vcpu->guest_fx_image.mxcsr = 0x1f80;
656 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
657 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
659 EXPORT_SYMBOL_GPL(fx_init);
662 * Allocate some memory and give it an address in the guest physical address
665 * Discontiguous memory is allowed, mostly for framebuffers.
667 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
668 struct kvm_memory_region *mem)
672 unsigned long npages;
674 struct kvm_memory_slot *memslot;
675 struct kvm_memory_slot old, new;
676 int memory_config_version;
679 /* General sanity checks */
680 if (mem->memory_size & (PAGE_SIZE - 1))
682 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
684 if (mem->slot >= KVM_MEMORY_SLOTS)
686 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
689 memslot = &kvm->memslots[mem->slot];
690 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
691 npages = mem->memory_size >> PAGE_SHIFT;
694 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
697 mutex_lock(&kvm->lock);
699 memory_config_version = kvm->memory_config_version;
700 new = old = *memslot;
702 new.base_gfn = base_gfn;
704 new.flags = mem->flags;
706 /* Disallow changing a memory slot's size. */
708 if (npages && old.npages && npages != old.npages)
711 /* Check for overlaps */
713 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
714 struct kvm_memory_slot *s = &kvm->memslots[i];
718 if (!((base_gfn + npages <= s->base_gfn) ||
719 (base_gfn >= s->base_gfn + s->npages)))
723 * Do memory allocations outside lock. memory_config_version will
726 mutex_unlock(&kvm->lock);
728 /* Deallocate if slot is being removed */
732 /* Free page dirty bitmap if unneeded */
733 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
734 new.dirty_bitmap = NULL;
738 /* Allocate if a slot is being created */
739 if (npages && !new.phys_mem) {
740 new.phys_mem = vmalloc(npages * sizeof(struct page *));
745 memset(new.phys_mem, 0, npages * sizeof(struct page *));
746 for (i = 0; i < npages; ++i) {
747 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
749 if (!new.phys_mem[i])
751 set_page_private(new.phys_mem[i],0);
755 /* Allocate page dirty bitmap if needed */
756 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
757 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
759 new.dirty_bitmap = vmalloc(dirty_bytes);
760 if (!new.dirty_bitmap)
762 memset(new.dirty_bitmap, 0, dirty_bytes);
765 mutex_lock(&kvm->lock);
767 if (memory_config_version != kvm->memory_config_version) {
768 mutex_unlock(&kvm->lock);
769 kvm_free_physmem_slot(&new, &old);
777 if (mem->slot >= kvm->nmemslots)
778 kvm->nmemslots = mem->slot + 1;
781 ++kvm->memory_config_version;
783 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
784 kvm_flush_remote_tlbs(kvm);
786 mutex_unlock(&kvm->lock);
788 kvm_free_physmem_slot(&old, &new);
792 mutex_unlock(&kvm->lock);
794 kvm_free_physmem_slot(&new, &old);
800 * Get (and clear) the dirty memory log for a memory slot.
802 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
803 struct kvm_dirty_log *log)
805 struct kvm_memory_slot *memslot;
808 unsigned long any = 0;
810 mutex_lock(&kvm->lock);
813 * Prevent changes to guest memory configuration even while the lock
817 mutex_unlock(&kvm->lock);
819 if (log->slot >= KVM_MEMORY_SLOTS)
822 memslot = &kvm->memslots[log->slot];
824 if (!memslot->dirty_bitmap)
827 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
829 for (i = 0; !any && i < n/sizeof(long); ++i)
830 any = memslot->dirty_bitmap[i];
833 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
836 /* If nothing is dirty, don't bother messing with page tables. */
838 mutex_lock(&kvm->lock);
839 kvm_mmu_slot_remove_write_access(kvm, log->slot);
840 kvm_flush_remote_tlbs(kvm);
841 memset(memslot->dirty_bitmap, 0, n);
842 mutex_unlock(&kvm->lock);
848 mutex_lock(&kvm->lock);
850 mutex_unlock(&kvm->lock);
855 * Set a new alias region. Aliases map a portion of physical memory into
856 * another portion. This is useful for memory windows, for example the PC
859 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
860 struct kvm_memory_alias *alias)
863 struct kvm_mem_alias *p;
866 /* General sanity checks */
867 if (alias->memory_size & (PAGE_SIZE - 1))
869 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
871 if (alias->slot >= KVM_ALIAS_SLOTS)
873 if (alias->guest_phys_addr + alias->memory_size
874 < alias->guest_phys_addr)
876 if (alias->target_phys_addr + alias->memory_size
877 < alias->target_phys_addr)
880 mutex_lock(&kvm->lock);
882 p = &kvm->aliases[alias->slot];
883 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
884 p->npages = alias->memory_size >> PAGE_SHIFT;
885 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
887 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
888 if (kvm->aliases[n - 1].npages)
892 kvm_mmu_zap_all(kvm);
894 mutex_unlock(&kvm->lock);
902 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
907 switch (chip->chip_id) {
908 case KVM_IRQCHIP_PIC_MASTER:
909 memcpy (&chip->chip.pic,
910 &pic_irqchip(kvm)->pics[0],
911 sizeof(struct kvm_pic_state));
913 case KVM_IRQCHIP_PIC_SLAVE:
914 memcpy (&chip->chip.pic,
915 &pic_irqchip(kvm)->pics[1],
916 sizeof(struct kvm_pic_state));
918 case KVM_IRQCHIP_IOAPIC:
919 memcpy (&chip->chip.ioapic,
921 sizeof(struct kvm_ioapic_state));
930 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
935 switch (chip->chip_id) {
936 case KVM_IRQCHIP_PIC_MASTER:
937 memcpy (&pic_irqchip(kvm)->pics[0],
939 sizeof(struct kvm_pic_state));
941 case KVM_IRQCHIP_PIC_SLAVE:
942 memcpy (&pic_irqchip(kvm)->pics[1],
944 sizeof(struct kvm_pic_state));
946 case KVM_IRQCHIP_IOAPIC:
947 memcpy (ioapic_irqchip(kvm),
949 sizeof(struct kvm_ioapic_state));
955 kvm_pic_update_irq(pic_irqchip(kvm));
959 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
962 struct kvm_mem_alias *alias;
964 for (i = 0; i < kvm->naliases; ++i) {
965 alias = &kvm->aliases[i];
966 if (gfn >= alias->base_gfn
967 && gfn < alias->base_gfn + alias->npages)
968 return alias->target_gfn + gfn - alias->base_gfn;
973 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
977 for (i = 0; i < kvm->nmemslots; ++i) {
978 struct kvm_memory_slot *memslot = &kvm->memslots[i];
980 if (gfn >= memslot->base_gfn
981 && gfn < memslot->base_gfn + memslot->npages)
987 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
989 gfn = unalias_gfn(kvm, gfn);
990 return __gfn_to_memslot(kvm, gfn);
993 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
995 struct kvm_memory_slot *slot;
997 gfn = unalias_gfn(kvm, gfn);
998 slot = __gfn_to_memslot(kvm, gfn);
1001 return slot->phys_mem[gfn - slot->base_gfn];
1003 EXPORT_SYMBOL_GPL(gfn_to_page);
1005 /* WARNING: Does not work on aliased pages. */
1006 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1008 struct kvm_memory_slot *memslot;
1010 memslot = __gfn_to_memslot(kvm, gfn);
1011 if (memslot && memslot->dirty_bitmap) {
1012 unsigned long rel_gfn = gfn - memslot->base_gfn;
1015 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1016 set_bit(rel_gfn, memslot->dirty_bitmap);
1020 int emulator_read_std(unsigned long addr,
1023 struct kvm_vcpu *vcpu)
1028 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1029 unsigned offset = addr & (PAGE_SIZE-1);
1030 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1035 if (gpa == UNMAPPED_GVA)
1036 return X86EMUL_PROPAGATE_FAULT;
1037 pfn = gpa >> PAGE_SHIFT;
1038 page = gfn_to_page(vcpu->kvm, pfn);
1040 return X86EMUL_UNHANDLEABLE;
1041 page_virt = kmap_atomic(page, KM_USER0);
1043 memcpy(data, page_virt + offset, tocopy);
1045 kunmap_atomic(page_virt, KM_USER0);
1052 return X86EMUL_CONTINUE;
1054 EXPORT_SYMBOL_GPL(emulator_read_std);
1056 static int emulator_write_std(unsigned long addr,
1059 struct kvm_vcpu *vcpu)
1061 pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1062 return X86EMUL_UNHANDLEABLE;
1066 * Only apic need an MMIO device hook, so shortcut now..
1068 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1071 struct kvm_io_device *dev;
1074 dev = &vcpu->apic->dev;
1075 if (dev->in_range(dev, addr))
1081 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1084 struct kvm_io_device *dev;
1086 dev = vcpu_find_pervcpu_dev(vcpu, addr);
1088 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1092 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1095 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1098 static int emulator_read_emulated(unsigned long addr,
1101 struct kvm_vcpu *vcpu)
1103 struct kvm_io_device *mmio_dev;
1106 if (vcpu->mmio_read_completed) {
1107 memcpy(val, vcpu->mmio_data, bytes);
1108 vcpu->mmio_read_completed = 0;
1109 return X86EMUL_CONTINUE;
1110 } else if (emulator_read_std(addr, val, bytes, vcpu)
1111 == X86EMUL_CONTINUE)
1112 return X86EMUL_CONTINUE;
1114 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1115 if (gpa == UNMAPPED_GVA)
1116 return X86EMUL_PROPAGATE_FAULT;
1119 * Is this MMIO handled locally?
1121 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1123 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1124 return X86EMUL_CONTINUE;
1127 vcpu->mmio_needed = 1;
1128 vcpu->mmio_phys_addr = gpa;
1129 vcpu->mmio_size = bytes;
1130 vcpu->mmio_is_write = 0;
1132 return X86EMUL_UNHANDLEABLE;
1135 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1136 const void *val, int bytes)
1141 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1143 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1146 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1147 virt = kmap_atomic(page, KM_USER0);
1148 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1149 memcpy(virt + offset_in_page(gpa), val, bytes);
1150 kunmap_atomic(virt, KM_USER0);
1154 static int emulator_write_emulated_onepage(unsigned long addr,
1157 struct kvm_vcpu *vcpu)
1159 struct kvm_io_device *mmio_dev;
1160 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1162 if (gpa == UNMAPPED_GVA) {
1163 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1164 return X86EMUL_PROPAGATE_FAULT;
1167 if (emulator_write_phys(vcpu, gpa, val, bytes))
1168 return X86EMUL_CONTINUE;
1171 * Is this MMIO handled locally?
1173 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1175 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1176 return X86EMUL_CONTINUE;
1179 vcpu->mmio_needed = 1;
1180 vcpu->mmio_phys_addr = gpa;
1181 vcpu->mmio_size = bytes;
1182 vcpu->mmio_is_write = 1;
1183 memcpy(vcpu->mmio_data, val, bytes);
1185 return X86EMUL_CONTINUE;
1188 int emulator_write_emulated(unsigned long addr,
1191 struct kvm_vcpu *vcpu)
1193 /* Crossing a page boundary? */
1194 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1197 now = -addr & ~PAGE_MASK;
1198 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1199 if (rc != X86EMUL_CONTINUE)
1205 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1207 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1209 static int emulator_cmpxchg_emulated(unsigned long addr,
1213 struct kvm_vcpu *vcpu)
1215 static int reported;
1219 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1221 return emulator_write_emulated(addr, new, bytes, vcpu);
1224 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1226 return kvm_arch_ops->get_segment_base(vcpu, seg);
1229 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1231 return X86EMUL_CONTINUE;
1234 int emulate_clts(struct kvm_vcpu *vcpu)
1238 cr0 = vcpu->cr0 & ~X86_CR0_TS;
1239 kvm_arch_ops->set_cr0(vcpu, cr0);
1240 return X86EMUL_CONTINUE;
1243 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1245 struct kvm_vcpu *vcpu = ctxt->vcpu;
1249 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1250 return X86EMUL_CONTINUE;
1252 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1253 return X86EMUL_UNHANDLEABLE;
1257 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1259 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1262 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1264 /* FIXME: better handling */
1265 return X86EMUL_UNHANDLEABLE;
1267 return X86EMUL_CONTINUE;
1270 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1272 static int reported;
1274 unsigned long rip = ctxt->vcpu->rip;
1275 unsigned long rip_linear;
1277 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1282 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
1284 printk(KERN_ERR "emulation failed but !mmio_needed?"
1285 " rip %lx %02x %02x %02x %02x\n",
1286 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1290 struct x86_emulate_ops emulate_ops = {
1291 .read_std = emulator_read_std,
1292 .write_std = emulator_write_std,
1293 .read_emulated = emulator_read_emulated,
1294 .write_emulated = emulator_write_emulated,
1295 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1298 int emulate_instruction(struct kvm_vcpu *vcpu,
1299 struct kvm_run *run,
1303 struct x86_emulate_ctxt emulate_ctxt;
1307 vcpu->mmio_fault_cr2 = cr2;
1308 kvm_arch_ops->cache_regs(vcpu);
1310 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1312 emulate_ctxt.vcpu = vcpu;
1313 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1314 emulate_ctxt.cr2 = cr2;
1315 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1316 ? X86EMUL_MODE_REAL : cs_l
1317 ? X86EMUL_MODE_PROT64 : cs_db
1318 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1320 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1321 emulate_ctxt.cs_base = 0;
1322 emulate_ctxt.ds_base = 0;
1323 emulate_ctxt.es_base = 0;
1324 emulate_ctxt.ss_base = 0;
1326 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1327 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1328 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1329 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1332 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1333 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1335 vcpu->mmio_is_write = 0;
1336 vcpu->pio.string = 0;
1337 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1338 if (vcpu->pio.string)
1339 return EMULATE_DO_MMIO;
1341 if ((r || vcpu->mmio_is_write) && run) {
1342 run->exit_reason = KVM_EXIT_MMIO;
1343 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1344 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1345 run->mmio.len = vcpu->mmio_size;
1346 run->mmio.is_write = vcpu->mmio_is_write;
1350 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1351 return EMULATE_DONE;
1352 if (!vcpu->mmio_needed) {
1353 report_emulation_failure(&emulate_ctxt);
1354 return EMULATE_FAIL;
1356 return EMULATE_DO_MMIO;
1359 kvm_arch_ops->decache_regs(vcpu);
1360 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1362 if (vcpu->mmio_is_write) {
1363 vcpu->mmio_needed = 0;
1364 return EMULATE_DO_MMIO;
1367 return EMULATE_DONE;
1369 EXPORT_SYMBOL_GPL(emulate_instruction);
1372 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1374 static void kvm_vcpu_kernel_halt(struct kvm_vcpu *vcpu)
1376 DECLARE_WAITQUEUE(wait, current);
1378 add_wait_queue(&vcpu->wq, &wait);
1381 * We will block until either an interrupt or a signal wakes us up
1383 while(!(irqchip_in_kernel(vcpu->kvm) && kvm_cpu_has_interrupt(vcpu))
1384 && !vcpu->irq_summary
1385 && !signal_pending(current)) {
1386 set_current_state(TASK_INTERRUPTIBLE);
1392 remove_wait_queue(&vcpu->wq, &wait);
1393 set_current_state(TASK_RUNNING);
1396 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1398 ++vcpu->stat.halt_exits;
1399 if (irqchip_in_kernel(vcpu->kvm)) {
1400 kvm_vcpu_kernel_halt(vcpu);
1403 vcpu->run->exit_reason = KVM_EXIT_HLT;
1407 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1409 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1411 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1413 kvm_arch_ops->cache_regs(vcpu);
1415 #ifdef CONFIG_X86_64
1416 if (is_long_mode(vcpu)) {
1417 nr = vcpu->regs[VCPU_REGS_RAX];
1418 a0 = vcpu->regs[VCPU_REGS_RDI];
1419 a1 = vcpu->regs[VCPU_REGS_RSI];
1420 a2 = vcpu->regs[VCPU_REGS_RDX];
1421 a3 = vcpu->regs[VCPU_REGS_RCX];
1422 a4 = vcpu->regs[VCPU_REGS_R8];
1423 a5 = vcpu->regs[VCPU_REGS_R9];
1427 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1428 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1429 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1430 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1431 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1432 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1433 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1437 run->hypercall.nr = nr;
1438 run->hypercall.args[0] = a0;
1439 run->hypercall.args[1] = a1;
1440 run->hypercall.args[2] = a2;
1441 run->hypercall.args[3] = a3;
1442 run->hypercall.args[4] = a4;
1443 run->hypercall.args[5] = a5;
1444 run->hypercall.ret = ret;
1445 run->hypercall.longmode = is_long_mode(vcpu);
1446 kvm_arch_ops->decache_regs(vcpu);
1449 vcpu->regs[VCPU_REGS_RAX] = ret;
1450 kvm_arch_ops->decache_regs(vcpu);
1453 EXPORT_SYMBOL_GPL(kvm_hypercall);
1455 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1457 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1460 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1462 struct descriptor_table dt = { limit, base };
1464 kvm_arch_ops->set_gdt(vcpu, &dt);
1467 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1469 struct descriptor_table dt = { limit, base };
1471 kvm_arch_ops->set_idt(vcpu, &dt);
1474 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1475 unsigned long *rflags)
1478 *rflags = kvm_arch_ops->get_rflags(vcpu);
1481 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1483 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1494 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1499 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1500 unsigned long *rflags)
1504 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1505 *rflags = kvm_arch_ops->get_rflags(vcpu);
1514 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1517 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1522 * Register the para guest with the host:
1524 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1526 struct kvm_vcpu_para_state *para_state;
1527 hpa_t para_state_hpa, hypercall_hpa;
1528 struct page *para_state_page;
1529 unsigned char *hypercall;
1530 gpa_t hypercall_gpa;
1532 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1533 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1536 * Needs to be page aligned:
1538 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1541 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1542 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1543 if (is_error_hpa(para_state_hpa))
1546 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1547 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1548 para_state = kmap(para_state_page);
1550 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1551 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1553 para_state->host_version = KVM_PARA_API_VERSION;
1555 * We cannot support guests that try to register themselves
1556 * with a newer API version than the host supports:
1558 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1559 para_state->ret = -KVM_EINVAL;
1560 goto err_kunmap_skip;
1563 hypercall_gpa = para_state->hypercall_gpa;
1564 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1565 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1566 if (is_error_hpa(hypercall_hpa)) {
1567 para_state->ret = -KVM_EINVAL;
1568 goto err_kunmap_skip;
1571 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1572 vcpu->para_state_page = para_state_page;
1573 vcpu->para_state_gpa = para_state_gpa;
1574 vcpu->hypercall_gpa = hypercall_gpa;
1576 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1577 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1578 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1579 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1580 kunmap_atomic(hypercall, KM_USER1);
1582 para_state->ret = 0;
1584 kunmap(para_state_page);
1590 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1595 case 0xc0010010: /* SYSCFG */
1596 case 0xc0010015: /* HWCR */
1597 case MSR_IA32_PLATFORM_ID:
1598 case MSR_IA32_P5_MC_ADDR:
1599 case MSR_IA32_P5_MC_TYPE:
1600 case MSR_IA32_MC0_CTL:
1601 case MSR_IA32_MCG_STATUS:
1602 case MSR_IA32_MCG_CAP:
1603 case MSR_IA32_MC0_MISC:
1604 case MSR_IA32_MC0_MISC+4:
1605 case MSR_IA32_MC0_MISC+8:
1606 case MSR_IA32_MC0_MISC+12:
1607 case MSR_IA32_MC0_MISC+16:
1608 case MSR_IA32_UCODE_REV:
1609 case MSR_IA32_PERF_STATUS:
1610 case MSR_IA32_EBL_CR_POWERON:
1611 /* MTRR registers */
1613 case 0x200 ... 0x2ff:
1616 case 0xcd: /* fsb frequency */
1619 case MSR_IA32_APICBASE:
1620 data = kvm_get_apic_base(vcpu);
1622 case MSR_IA32_MISC_ENABLE:
1623 data = vcpu->ia32_misc_enable_msr;
1625 #ifdef CONFIG_X86_64
1627 data = vcpu->shadow_efer;
1631 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1637 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1640 * Reads an msr value (of 'msr_index') into 'pdata'.
1641 * Returns 0 on success, non-0 otherwise.
1642 * Assumes vcpu_load() was already called.
1644 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1646 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1649 #ifdef CONFIG_X86_64
1651 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1653 if (efer & EFER_RESERVED_BITS) {
1654 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1661 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1662 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1667 kvm_arch_ops->set_efer(vcpu, efer);
1670 efer |= vcpu->shadow_efer & EFER_LMA;
1672 vcpu->shadow_efer = efer;
1677 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1680 #ifdef CONFIG_X86_64
1682 set_efer(vcpu, data);
1685 case MSR_IA32_MC0_STATUS:
1686 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1687 __FUNCTION__, data);
1689 case MSR_IA32_MCG_STATUS:
1690 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1691 __FUNCTION__, data);
1693 case MSR_IA32_UCODE_REV:
1694 case MSR_IA32_UCODE_WRITE:
1695 case 0x200 ... 0x2ff: /* MTRRs */
1697 case MSR_IA32_APICBASE:
1698 kvm_set_apic_base(vcpu, data);
1700 case MSR_IA32_MISC_ENABLE:
1701 vcpu->ia32_misc_enable_msr = data;
1704 * This is the 'probe whether the host is KVM' logic:
1706 case MSR_KVM_API_MAGIC:
1707 return vcpu_register_para(vcpu, data);
1710 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1715 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1718 * Writes msr value into into the appropriate "register".
1719 * Returns 0 on success, non-0 otherwise.
1720 * Assumes vcpu_load() was already called.
1722 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1724 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1727 void kvm_resched(struct kvm_vcpu *vcpu)
1729 if (!need_resched())
1733 EXPORT_SYMBOL_GPL(kvm_resched);
1735 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1739 struct kvm_cpuid_entry *e, *best;
1741 kvm_arch_ops->cache_regs(vcpu);
1742 function = vcpu->regs[VCPU_REGS_RAX];
1743 vcpu->regs[VCPU_REGS_RAX] = 0;
1744 vcpu->regs[VCPU_REGS_RBX] = 0;
1745 vcpu->regs[VCPU_REGS_RCX] = 0;
1746 vcpu->regs[VCPU_REGS_RDX] = 0;
1748 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1749 e = &vcpu->cpuid_entries[i];
1750 if (e->function == function) {
1755 * Both basic or both extended?
1757 if (((e->function ^ function) & 0x80000000) == 0)
1758 if (!best || e->function > best->function)
1762 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1763 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1764 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1765 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1767 kvm_arch_ops->decache_regs(vcpu);
1768 kvm_arch_ops->skip_emulated_instruction(vcpu);
1770 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1772 static int pio_copy_data(struct kvm_vcpu *vcpu)
1774 void *p = vcpu->pio_data;
1777 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1779 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1782 free_pio_guest_pages(vcpu);
1785 q += vcpu->pio.guest_page_offset;
1786 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1788 memcpy(q, p, bytes);
1790 memcpy(p, q, bytes);
1791 q -= vcpu->pio.guest_page_offset;
1793 free_pio_guest_pages(vcpu);
1797 static int complete_pio(struct kvm_vcpu *vcpu)
1799 struct kvm_pio_request *io = &vcpu->pio;
1803 kvm_arch_ops->cache_regs(vcpu);
1807 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1811 r = pio_copy_data(vcpu);
1813 kvm_arch_ops->cache_regs(vcpu);
1820 delta *= io->cur_count;
1822 * The size of the register should really depend on
1823 * current address size.
1825 vcpu->regs[VCPU_REGS_RCX] -= delta;
1831 vcpu->regs[VCPU_REGS_RDI] += delta;
1833 vcpu->regs[VCPU_REGS_RSI] += delta;
1836 kvm_arch_ops->decache_regs(vcpu);
1838 io->count -= io->cur_count;
1842 kvm_arch_ops->skip_emulated_instruction(vcpu);
1846 static void kernel_pio(struct kvm_io_device *pio_dev,
1847 struct kvm_vcpu *vcpu,
1850 /* TODO: String I/O for in kernel device */
1852 mutex_lock(&vcpu->kvm->lock);
1854 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1858 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1861 mutex_unlock(&vcpu->kvm->lock);
1864 static void pio_string_write(struct kvm_io_device *pio_dev,
1865 struct kvm_vcpu *vcpu)
1867 struct kvm_pio_request *io = &vcpu->pio;
1868 void *pd = vcpu->pio_data;
1871 mutex_lock(&vcpu->kvm->lock);
1872 for (i = 0; i < io->cur_count; i++) {
1873 kvm_iodevice_write(pio_dev, io->port,
1878 mutex_unlock(&vcpu->kvm->lock);
1881 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1882 int size, unsigned port)
1884 struct kvm_io_device *pio_dev;
1886 vcpu->run->exit_reason = KVM_EXIT_IO;
1887 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1888 vcpu->run->io.size = vcpu->pio.size = size;
1889 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1890 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1891 vcpu->run->io.port = vcpu->pio.port = port;
1893 vcpu->pio.string = 0;
1895 vcpu->pio.guest_page_offset = 0;
1898 kvm_arch_ops->cache_regs(vcpu);
1899 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1900 kvm_arch_ops->decache_regs(vcpu);
1902 pio_dev = vcpu_find_pio_dev(vcpu, port);
1904 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1910 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1912 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1913 int size, unsigned long count, int down,
1914 gva_t address, int rep, unsigned port)
1916 unsigned now, in_page;
1920 struct kvm_io_device *pio_dev;
1922 vcpu->run->exit_reason = KVM_EXIT_IO;
1923 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1924 vcpu->run->io.size = vcpu->pio.size = size;
1925 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1926 vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1927 vcpu->run->io.port = vcpu->pio.port = port;
1929 vcpu->pio.string = 1;
1930 vcpu->pio.down = down;
1931 vcpu->pio.guest_page_offset = offset_in_page(address);
1932 vcpu->pio.rep = rep;
1935 kvm_arch_ops->skip_emulated_instruction(vcpu);
1940 in_page = PAGE_SIZE - offset_in_page(address);
1942 in_page = offset_in_page(address) + size;
1943 now = min(count, (unsigned long)in_page / size);
1946 * String I/O straddles page boundary. Pin two guest pages
1947 * so that we satisfy atomicity constraints. Do just one
1948 * transaction to avoid complexity.
1955 * String I/O in reverse. Yuck. Kill the guest, fix later.
1957 pr_unimpl(vcpu, "guest string pio down\n");
1961 vcpu->run->io.count = now;
1962 vcpu->pio.cur_count = now;
1964 for (i = 0; i < nr_pages; ++i) {
1965 mutex_lock(&vcpu->kvm->lock);
1966 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1969 vcpu->pio.guest_pages[i] = page;
1970 mutex_unlock(&vcpu->kvm->lock);
1973 free_pio_guest_pages(vcpu);
1978 pio_dev = vcpu_find_pio_dev(vcpu, port);
1979 if (!vcpu->pio.in) {
1980 /* string PIO write */
1981 ret = pio_copy_data(vcpu);
1982 if (ret >= 0 && pio_dev) {
1983 pio_string_write(pio_dev, vcpu);
1985 if (vcpu->pio.count == 0)
1989 pr_unimpl(vcpu, "no string pio read support yet, "
1990 "port %x size %d count %ld\n",
1995 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1997 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2004 if (vcpu->sigset_active)
2005 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2007 /* re-sync apic's tpr */
2008 if (!irqchip_in_kernel(vcpu->kvm))
2009 set_cr8(vcpu, kvm_run->cr8);
2011 if (vcpu->pio.cur_count) {
2012 r = complete_pio(vcpu);
2017 if (vcpu->mmio_needed) {
2018 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2019 vcpu->mmio_read_completed = 1;
2020 vcpu->mmio_needed = 0;
2021 r = emulate_instruction(vcpu, kvm_run,
2022 vcpu->mmio_fault_cr2, 0);
2023 if (r == EMULATE_DO_MMIO) {
2025 * Read-modify-write. Back to userspace.
2032 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2033 kvm_arch_ops->cache_regs(vcpu);
2034 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2035 kvm_arch_ops->decache_regs(vcpu);
2038 r = kvm_arch_ops->run(vcpu, kvm_run);
2041 if (vcpu->sigset_active)
2042 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2048 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2049 struct kvm_regs *regs)
2053 kvm_arch_ops->cache_regs(vcpu);
2055 regs->rax = vcpu->regs[VCPU_REGS_RAX];
2056 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2057 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2058 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2059 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2060 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2061 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2062 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2063 #ifdef CONFIG_X86_64
2064 regs->r8 = vcpu->regs[VCPU_REGS_R8];
2065 regs->r9 = vcpu->regs[VCPU_REGS_R9];
2066 regs->r10 = vcpu->regs[VCPU_REGS_R10];
2067 regs->r11 = vcpu->regs[VCPU_REGS_R11];
2068 regs->r12 = vcpu->regs[VCPU_REGS_R12];
2069 regs->r13 = vcpu->regs[VCPU_REGS_R13];
2070 regs->r14 = vcpu->regs[VCPU_REGS_R14];
2071 regs->r15 = vcpu->regs[VCPU_REGS_R15];
2074 regs->rip = vcpu->rip;
2075 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
2078 * Don't leak debug flags in case they were set for guest debugging
2080 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2081 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2088 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2089 struct kvm_regs *regs)
2093 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2094 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2095 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2096 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2097 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2098 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2099 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2100 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2101 #ifdef CONFIG_X86_64
2102 vcpu->regs[VCPU_REGS_R8] = regs->r8;
2103 vcpu->regs[VCPU_REGS_R9] = regs->r9;
2104 vcpu->regs[VCPU_REGS_R10] = regs->r10;
2105 vcpu->regs[VCPU_REGS_R11] = regs->r11;
2106 vcpu->regs[VCPU_REGS_R12] = regs->r12;
2107 vcpu->regs[VCPU_REGS_R13] = regs->r13;
2108 vcpu->regs[VCPU_REGS_R14] = regs->r14;
2109 vcpu->regs[VCPU_REGS_R15] = regs->r15;
2112 vcpu->rip = regs->rip;
2113 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2115 kvm_arch_ops->decache_regs(vcpu);
2122 static void get_segment(struct kvm_vcpu *vcpu,
2123 struct kvm_segment *var, int seg)
2125 return kvm_arch_ops->get_segment(vcpu, var, seg);
2128 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2129 struct kvm_sregs *sregs)
2131 struct descriptor_table dt;
2136 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2137 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2138 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2139 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2140 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2141 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2143 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2144 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2146 kvm_arch_ops->get_idt(vcpu, &dt);
2147 sregs->idt.limit = dt.limit;
2148 sregs->idt.base = dt.base;
2149 kvm_arch_ops->get_gdt(vcpu, &dt);
2150 sregs->gdt.limit = dt.limit;
2151 sregs->gdt.base = dt.base;
2153 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2154 sregs->cr0 = vcpu->cr0;
2155 sregs->cr2 = vcpu->cr2;
2156 sregs->cr3 = vcpu->cr3;
2157 sregs->cr4 = vcpu->cr4;
2158 sregs->cr8 = get_cr8(vcpu);
2159 sregs->efer = vcpu->shadow_efer;
2160 sregs->apic_base = kvm_get_apic_base(vcpu);
2162 if (irqchip_in_kernel(vcpu->kvm)) {
2163 memset(sregs->interrupt_bitmap, 0,
2164 sizeof sregs->interrupt_bitmap);
2165 pending_vec = kvm_arch_ops->get_irq(vcpu);
2166 if (pending_vec >= 0)
2167 set_bit(pending_vec, (unsigned long *)sregs->interrupt_bitmap);
2169 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2170 sizeof sregs->interrupt_bitmap);
2177 static void set_segment(struct kvm_vcpu *vcpu,
2178 struct kvm_segment *var, int seg)
2180 return kvm_arch_ops->set_segment(vcpu, var, seg);
2183 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2184 struct kvm_sregs *sregs)
2186 int mmu_reset_needed = 0;
2187 int i, pending_vec, max_bits;
2188 struct descriptor_table dt;
2192 dt.limit = sregs->idt.limit;
2193 dt.base = sregs->idt.base;
2194 kvm_arch_ops->set_idt(vcpu, &dt);
2195 dt.limit = sregs->gdt.limit;
2196 dt.base = sregs->gdt.base;
2197 kvm_arch_ops->set_gdt(vcpu, &dt);
2199 vcpu->cr2 = sregs->cr2;
2200 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2201 vcpu->cr3 = sregs->cr3;
2203 set_cr8(vcpu, sregs->cr8);
2205 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2206 #ifdef CONFIG_X86_64
2207 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2209 kvm_set_apic_base(vcpu, sregs->apic_base);
2211 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2213 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2214 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2216 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2217 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2218 if (!is_long_mode(vcpu) && is_pae(vcpu))
2219 load_pdptrs(vcpu, vcpu->cr3);
2221 if (mmu_reset_needed)
2222 kvm_mmu_reset_context(vcpu);
2224 if (!irqchip_in_kernel(vcpu->kvm)) {
2225 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2226 sizeof vcpu->irq_pending);
2227 vcpu->irq_summary = 0;
2228 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2229 if (vcpu->irq_pending[i])
2230 __set_bit(i, &vcpu->irq_summary);
2232 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2233 pending_vec = find_first_bit(
2234 (const unsigned long *)sregs->interrupt_bitmap,
2236 /* Only pending external irq is handled here */
2237 if (pending_vec < max_bits) {
2238 kvm_arch_ops->set_irq(vcpu, pending_vec);
2239 printk("Set back pending irq %d\n", pending_vec);
2243 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2244 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2245 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2246 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2247 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2248 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2250 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2251 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2259 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2260 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2262 * This list is modified at module load time to reflect the
2263 * capabilities of the host cpu.
2265 static u32 msrs_to_save[] = {
2266 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2268 #ifdef CONFIG_X86_64
2269 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2271 MSR_IA32_TIME_STAMP_COUNTER,
2274 static unsigned num_msrs_to_save;
2276 static u32 emulated_msrs[] = {
2277 MSR_IA32_MISC_ENABLE,
2280 static __init void kvm_init_msr_list(void)
2285 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2286 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2289 msrs_to_save[j] = msrs_to_save[i];
2292 num_msrs_to_save = j;
2296 * Adapt set_msr() to msr_io()'s calling convention
2298 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2300 return kvm_set_msr(vcpu, index, *data);
2304 * Read or write a bunch of msrs. All parameters are kernel addresses.
2306 * @return number of msrs set successfully.
2308 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2309 struct kvm_msr_entry *entries,
2310 int (*do_msr)(struct kvm_vcpu *vcpu,
2311 unsigned index, u64 *data))
2317 for (i = 0; i < msrs->nmsrs; ++i)
2318 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2327 * Read or write a bunch of msrs. Parameters are user addresses.
2329 * @return number of msrs set successfully.
2331 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2332 int (*do_msr)(struct kvm_vcpu *vcpu,
2333 unsigned index, u64 *data),
2336 struct kvm_msrs msrs;
2337 struct kvm_msr_entry *entries;
2342 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2346 if (msrs.nmsrs >= MAX_IO_MSRS)
2350 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2351 entries = vmalloc(size);
2356 if (copy_from_user(entries, user_msrs->entries, size))
2359 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2364 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2376 * Translate a guest virtual address to a guest physical address.
2378 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2379 struct kvm_translation *tr)
2381 unsigned long vaddr = tr->linear_address;
2385 mutex_lock(&vcpu->kvm->lock);
2386 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2387 tr->physical_address = gpa;
2388 tr->valid = gpa != UNMAPPED_GVA;
2391 mutex_unlock(&vcpu->kvm->lock);
2397 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2398 struct kvm_interrupt *irq)
2400 if (irq->irq < 0 || irq->irq >= 256)
2402 if (irqchip_in_kernel(vcpu->kvm))
2406 set_bit(irq->irq, vcpu->irq_pending);
2407 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2414 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2415 struct kvm_debug_guest *dbg)
2421 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2428 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2429 unsigned long address,
2432 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2433 unsigned long pgoff;
2436 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2438 page = virt_to_page(vcpu->run);
2439 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2440 page = virt_to_page(vcpu->pio_data);
2442 return NOPAGE_SIGBUS;
2445 *type = VM_FAULT_MINOR;
2450 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2451 .nopage = kvm_vcpu_nopage,
2454 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2456 vma->vm_ops = &kvm_vcpu_vm_ops;
2460 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2462 struct kvm_vcpu *vcpu = filp->private_data;
2464 fput(vcpu->kvm->filp);
2468 static struct file_operations kvm_vcpu_fops = {
2469 .release = kvm_vcpu_release,
2470 .unlocked_ioctl = kvm_vcpu_ioctl,
2471 .compat_ioctl = kvm_vcpu_ioctl,
2472 .mmap = kvm_vcpu_mmap,
2476 * Allocates an inode for the vcpu.
2478 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2481 struct inode *inode;
2484 r = anon_inode_getfd(&fd, &inode, &file,
2485 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2488 atomic_inc(&vcpu->kvm->filp->f_count);
2493 * Creates some virtual cpus. Good luck creating more than one.
2495 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2498 struct kvm_vcpu *vcpu;
2503 vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2505 return PTR_ERR(vcpu);
2507 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2509 /* We do fxsave: this must be aligned. */
2510 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2513 r = kvm_mmu_setup(vcpu);
2518 mutex_lock(&kvm->lock);
2519 if (kvm->vcpus[n]) {
2521 mutex_unlock(&kvm->lock);
2524 kvm->vcpus[n] = vcpu;
2525 mutex_unlock(&kvm->lock);
2527 /* Now it's all set up, let userspace reach it */
2528 r = create_vcpu_fd(vcpu);
2534 mutex_lock(&kvm->lock);
2535 kvm->vcpus[n] = NULL;
2536 mutex_unlock(&kvm->lock);
2540 kvm_mmu_unload(vcpu);
2544 kvm_arch_ops->vcpu_free(vcpu);
2548 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2552 struct kvm_cpuid_entry *e, *entry;
2554 rdmsrl(MSR_EFER, efer);
2556 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2557 e = &vcpu->cpuid_entries[i];
2558 if (e->function == 0x80000001) {
2563 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2564 entry->edx &= ~(1 << 20);
2565 printk(KERN_INFO "kvm: guest NX capability removed\n");
2569 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2570 struct kvm_cpuid *cpuid,
2571 struct kvm_cpuid_entry __user *entries)
2576 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2579 if (copy_from_user(&vcpu->cpuid_entries, entries,
2580 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2582 vcpu->cpuid_nent = cpuid->nent;
2583 cpuid_fix_nx_cap(vcpu);
2590 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2593 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2594 vcpu->sigset_active = 1;
2595 vcpu->sigset = *sigset;
2597 vcpu->sigset_active = 0;
2602 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2603 * we have asm/x86/processor.h
2614 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2615 #ifdef CONFIG_X86_64
2616 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2618 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2622 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2624 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2628 memcpy(fpu->fpr, fxsave->st_space, 128);
2629 fpu->fcw = fxsave->cwd;
2630 fpu->fsw = fxsave->swd;
2631 fpu->ftwx = fxsave->twd;
2632 fpu->last_opcode = fxsave->fop;
2633 fpu->last_ip = fxsave->rip;
2634 fpu->last_dp = fxsave->rdp;
2635 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2642 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2644 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2648 memcpy(fxsave->st_space, fpu->fpr, 128);
2649 fxsave->cwd = fpu->fcw;
2650 fxsave->swd = fpu->fsw;
2651 fxsave->twd = fpu->ftwx;
2652 fxsave->fop = fpu->last_opcode;
2653 fxsave->rip = fpu->last_ip;
2654 fxsave->rdp = fpu->last_dp;
2655 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2662 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2663 struct kvm_lapic_state *s)
2666 memcpy(s->regs, vcpu->apic->regs, sizeof *s);
2672 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2673 struct kvm_lapic_state *s)
2676 memcpy(vcpu->apic->regs, s->regs, sizeof *s);
2677 kvm_apic_post_state_restore(vcpu);
2683 static long kvm_vcpu_ioctl(struct file *filp,
2684 unsigned int ioctl, unsigned long arg)
2686 struct kvm_vcpu *vcpu = filp->private_data;
2687 void __user *argp = (void __user *)arg;
2695 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2697 case KVM_GET_REGS: {
2698 struct kvm_regs kvm_regs;
2700 memset(&kvm_regs, 0, sizeof kvm_regs);
2701 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2705 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2710 case KVM_SET_REGS: {
2711 struct kvm_regs kvm_regs;
2714 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2716 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2722 case KVM_GET_SREGS: {
2723 struct kvm_sregs kvm_sregs;
2725 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2726 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2730 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2735 case KVM_SET_SREGS: {
2736 struct kvm_sregs kvm_sregs;
2739 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2741 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2747 case KVM_TRANSLATE: {
2748 struct kvm_translation tr;
2751 if (copy_from_user(&tr, argp, sizeof tr))
2753 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2757 if (copy_to_user(argp, &tr, sizeof tr))
2762 case KVM_INTERRUPT: {
2763 struct kvm_interrupt irq;
2766 if (copy_from_user(&irq, argp, sizeof irq))
2768 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2774 case KVM_DEBUG_GUEST: {
2775 struct kvm_debug_guest dbg;
2778 if (copy_from_user(&dbg, argp, sizeof dbg))
2780 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2787 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2790 r = msr_io(vcpu, argp, do_set_msr, 0);
2792 case KVM_SET_CPUID: {
2793 struct kvm_cpuid __user *cpuid_arg = argp;
2794 struct kvm_cpuid cpuid;
2797 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2799 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2804 case KVM_SET_SIGNAL_MASK: {
2805 struct kvm_signal_mask __user *sigmask_arg = argp;
2806 struct kvm_signal_mask kvm_sigmask;
2807 sigset_t sigset, *p;
2812 if (copy_from_user(&kvm_sigmask, argp,
2813 sizeof kvm_sigmask))
2816 if (kvm_sigmask.len != sizeof sigset)
2819 if (copy_from_user(&sigset, sigmask_arg->sigset,
2824 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2830 memset(&fpu, 0, sizeof fpu);
2831 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2835 if (copy_to_user(argp, &fpu, sizeof fpu))
2844 if (copy_from_user(&fpu, argp, sizeof fpu))
2846 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2852 case KVM_GET_LAPIC: {
2853 struct kvm_lapic_state lapic;
2855 memset(&lapic, 0, sizeof lapic);
2856 r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
2860 if (copy_to_user(argp, &lapic, sizeof lapic))
2865 case KVM_SET_LAPIC: {
2866 struct kvm_lapic_state lapic;
2869 if (copy_from_user(&lapic, argp, sizeof lapic))
2871 r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
2884 static long kvm_vm_ioctl(struct file *filp,
2885 unsigned int ioctl, unsigned long arg)
2887 struct kvm *kvm = filp->private_data;
2888 void __user *argp = (void __user *)arg;
2892 case KVM_CREATE_VCPU:
2893 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2897 case KVM_SET_MEMORY_REGION: {
2898 struct kvm_memory_region kvm_mem;
2901 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2903 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2908 case KVM_GET_DIRTY_LOG: {
2909 struct kvm_dirty_log log;
2912 if (copy_from_user(&log, argp, sizeof log))
2914 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2919 case KVM_SET_MEMORY_ALIAS: {
2920 struct kvm_memory_alias alias;
2923 if (copy_from_user(&alias, argp, sizeof alias))
2925 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2930 case KVM_CREATE_IRQCHIP:
2932 kvm->vpic = kvm_create_pic(kvm);
2934 r = kvm_ioapic_init(kvm);
2944 case KVM_IRQ_LINE: {
2945 struct kvm_irq_level irq_event;
2948 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2950 if (irqchip_in_kernel(kvm)) {
2951 mutex_lock(&kvm->lock);
2952 if (irq_event.irq < 16)
2953 kvm_pic_set_irq(pic_irqchip(kvm),
2956 kvm_ioapic_set_irq(kvm->vioapic,
2959 mutex_unlock(&kvm->lock);
2964 case KVM_GET_IRQCHIP: {
2965 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2966 struct kvm_irqchip chip;
2969 if (copy_from_user(&chip, argp, sizeof chip))
2972 if (!irqchip_in_kernel(kvm))
2974 r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
2978 if (copy_to_user(argp, &chip, sizeof chip))
2983 case KVM_SET_IRQCHIP: {
2984 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2985 struct kvm_irqchip chip;
2988 if (copy_from_user(&chip, argp, sizeof chip))
2991 if (!irqchip_in_kernel(kvm))
2993 r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
3006 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
3007 unsigned long address,
3010 struct kvm *kvm = vma->vm_file->private_data;
3011 unsigned long pgoff;
3014 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3015 page = gfn_to_page(kvm, pgoff);
3017 return NOPAGE_SIGBUS;
3020 *type = VM_FAULT_MINOR;
3025 static struct vm_operations_struct kvm_vm_vm_ops = {
3026 .nopage = kvm_vm_nopage,
3029 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
3031 vma->vm_ops = &kvm_vm_vm_ops;
3035 static struct file_operations kvm_vm_fops = {
3036 .release = kvm_vm_release,
3037 .unlocked_ioctl = kvm_vm_ioctl,
3038 .compat_ioctl = kvm_vm_ioctl,
3039 .mmap = kvm_vm_mmap,
3042 static int kvm_dev_ioctl_create_vm(void)
3045 struct inode *inode;
3049 kvm = kvm_create_vm();
3051 return PTR_ERR(kvm);
3052 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
3054 kvm_destroy_vm(kvm);
3063 static long kvm_dev_ioctl(struct file *filp,
3064 unsigned int ioctl, unsigned long arg)
3066 void __user *argp = (void __user *)arg;
3070 case KVM_GET_API_VERSION:
3074 r = KVM_API_VERSION;
3080 r = kvm_dev_ioctl_create_vm();
3082 case KVM_GET_MSR_INDEX_LIST: {
3083 struct kvm_msr_list __user *user_msr_list = argp;
3084 struct kvm_msr_list msr_list;
3088 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3091 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
3092 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3095 if (n < num_msrs_to_save)
3098 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3099 num_msrs_to_save * sizeof(u32)))
3101 if (copy_to_user(user_msr_list->indices
3102 + num_msrs_to_save * sizeof(u32),
3104 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
3109 case KVM_CHECK_EXTENSION: {
3110 int ext = (long)argp;
3113 case KVM_CAP_IRQCHIP:
3123 case KVM_GET_VCPU_MMAP_SIZE:
3136 static struct file_operations kvm_chardev_ops = {
3137 .unlocked_ioctl = kvm_dev_ioctl,
3138 .compat_ioctl = kvm_dev_ioctl,
3141 static struct miscdevice kvm_dev = {
3148 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
3151 static void decache_vcpus_on_cpu(int cpu)
3154 struct kvm_vcpu *vcpu;
3157 spin_lock(&kvm_lock);
3158 list_for_each_entry(vm, &vm_list, vm_list)
3159 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3160 vcpu = vm->vcpus[i];
3164 * If the vcpu is locked, then it is running on some
3165 * other cpu and therefore it is not cached on the
3168 * If it's not locked, check the last cpu it executed
3171 if (mutex_trylock(&vcpu->mutex)) {
3172 if (vcpu->cpu == cpu) {
3173 kvm_arch_ops->vcpu_decache(vcpu);
3176 mutex_unlock(&vcpu->mutex);
3179 spin_unlock(&kvm_lock);
3182 static void hardware_enable(void *junk)
3184 int cpu = raw_smp_processor_id();
3186 if (cpu_isset(cpu, cpus_hardware_enabled))
3188 cpu_set(cpu, cpus_hardware_enabled);
3189 kvm_arch_ops->hardware_enable(NULL);
3192 static void hardware_disable(void *junk)
3194 int cpu = raw_smp_processor_id();
3196 if (!cpu_isset(cpu, cpus_hardware_enabled))
3198 cpu_clear(cpu, cpus_hardware_enabled);
3199 decache_vcpus_on_cpu(cpu);
3200 kvm_arch_ops->hardware_disable(NULL);
3203 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3210 case CPU_DYING_FROZEN:
3211 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3213 hardware_disable(NULL);
3215 case CPU_UP_CANCELED:
3216 case CPU_UP_CANCELED_FROZEN:
3217 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3219 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3222 case CPU_ONLINE_FROZEN:
3223 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3225 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3231 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3234 if (val == SYS_RESTART) {
3236 * Some (well, at least mine) BIOSes hang on reboot if
3239 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3240 on_each_cpu(hardware_disable, NULL, 0, 1);
3245 static struct notifier_block kvm_reboot_notifier = {
3246 .notifier_call = kvm_reboot,
3250 void kvm_io_bus_init(struct kvm_io_bus *bus)
3252 memset(bus, 0, sizeof(*bus));
3255 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3259 for (i = 0; i < bus->dev_count; i++) {
3260 struct kvm_io_device *pos = bus->devs[i];
3262 kvm_iodevice_destructor(pos);
3266 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3270 for (i = 0; i < bus->dev_count; i++) {
3271 struct kvm_io_device *pos = bus->devs[i];
3273 if (pos->in_range(pos, addr))
3280 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3282 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3284 bus->devs[bus->dev_count++] = dev;
3287 static struct notifier_block kvm_cpu_notifier = {
3288 .notifier_call = kvm_cpu_hotplug,
3289 .priority = 20, /* must be > scheduler priority */
3292 static u64 stat_get(void *_offset)
3294 unsigned offset = (long)_offset;
3297 struct kvm_vcpu *vcpu;
3300 spin_lock(&kvm_lock);
3301 list_for_each_entry(kvm, &vm_list, vm_list)
3302 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3303 vcpu = kvm->vcpus[i];
3305 total += *(u32 *)((void *)vcpu + offset);
3307 spin_unlock(&kvm_lock);
3311 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3313 static __init void kvm_init_debug(void)
3315 struct kvm_stats_debugfs_item *p;
3317 debugfs_dir = debugfs_create_dir("kvm", NULL);
3318 for (p = debugfs_entries; p->name; ++p)
3319 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3320 (void *)(long)p->offset,
3324 static void kvm_exit_debug(void)
3326 struct kvm_stats_debugfs_item *p;
3328 for (p = debugfs_entries; p->name; ++p)
3329 debugfs_remove(p->dentry);
3330 debugfs_remove(debugfs_dir);
3333 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3335 hardware_disable(NULL);
3339 static int kvm_resume(struct sys_device *dev)
3341 hardware_enable(NULL);
3345 static struct sysdev_class kvm_sysdev_class = {
3346 set_kset_name("kvm"),
3347 .suspend = kvm_suspend,
3348 .resume = kvm_resume,
3351 static struct sys_device kvm_sysdev = {
3353 .cls = &kvm_sysdev_class,
3356 hpa_t bad_page_address;
3359 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3361 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3364 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3366 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3368 kvm_arch_ops->vcpu_load(vcpu, cpu);
3371 static void kvm_sched_out(struct preempt_notifier *pn,
3372 struct task_struct *next)
3374 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3376 kvm_arch_ops->vcpu_put(vcpu);
3379 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3380 struct module *module)
3386 printk(KERN_ERR "kvm: already loaded the other module\n");
3390 if (!ops->cpu_has_kvm_support()) {
3391 printk(KERN_ERR "kvm: no hardware support\n");
3394 if (ops->disabled_by_bios()) {
3395 printk(KERN_ERR "kvm: disabled by bios\n");
3401 r = kvm_arch_ops->hardware_setup();
3405 for_each_online_cpu(cpu) {
3406 smp_call_function_single(cpu,
3407 kvm_arch_ops->check_processor_compatibility,
3413 on_each_cpu(hardware_enable, NULL, 0, 1);
3414 r = register_cpu_notifier(&kvm_cpu_notifier);
3417 register_reboot_notifier(&kvm_reboot_notifier);
3419 r = sysdev_class_register(&kvm_sysdev_class);
3423 r = sysdev_register(&kvm_sysdev);
3427 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3428 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3429 __alignof__(struct kvm_vcpu), 0, 0);
3430 if (!kvm_vcpu_cache) {
3435 kvm_chardev_ops.owner = module;
3437 r = misc_register(&kvm_dev);
3439 printk (KERN_ERR "kvm: misc device register failed\n");
3443 kvm_preempt_ops.sched_in = kvm_sched_in;
3444 kvm_preempt_ops.sched_out = kvm_sched_out;
3449 kmem_cache_destroy(kvm_vcpu_cache);
3451 sysdev_unregister(&kvm_sysdev);
3453 sysdev_class_unregister(&kvm_sysdev_class);
3455 unregister_reboot_notifier(&kvm_reboot_notifier);
3456 unregister_cpu_notifier(&kvm_cpu_notifier);
3458 on_each_cpu(hardware_disable, NULL, 0, 1);
3460 kvm_arch_ops->hardware_unsetup();
3462 kvm_arch_ops = NULL;
3466 void kvm_exit_arch(void)
3468 misc_deregister(&kvm_dev);
3469 kmem_cache_destroy(kvm_vcpu_cache);
3470 sysdev_unregister(&kvm_sysdev);
3471 sysdev_class_unregister(&kvm_sysdev_class);
3472 unregister_reboot_notifier(&kvm_reboot_notifier);
3473 unregister_cpu_notifier(&kvm_cpu_notifier);
3474 on_each_cpu(hardware_disable, NULL, 0, 1);
3475 kvm_arch_ops->hardware_unsetup();
3476 kvm_arch_ops = NULL;
3479 static __init int kvm_init(void)
3481 static struct page *bad_page;
3484 r = kvm_mmu_module_init();
3490 kvm_init_msr_list();
3492 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3497 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3498 memset(__va(bad_page_address), 0, PAGE_SIZE);
3504 kvm_mmu_module_exit();
3509 static __exit void kvm_exit(void)
3512 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3513 kvm_mmu_module_exit();
3516 module_init(kvm_init)
3517 module_exit(kvm_exit)
3519 EXPORT_SYMBOL_GPL(kvm_init_arch);
3520 EXPORT_SYMBOL_GPL(kvm_exit_arch);