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"
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/gfp.h>
28 #include <linux/miscdevice.h>
29 #include <linux/vmalloc.h>
30 #include <linux/reboot.h>
31 #include <linux/debugfs.h>
32 #include <linux/highmem.h>
33 #include <linux/file.h>
34 #include <linux/sysdev.h>
35 #include <linux/cpu.h>
36 #include <linux/sched.h>
37 #include <linux/cpumask.h>
38 #include <linux/smp.h>
39 #include <linux/anon_inodes.h>
41 #include <asm/processor.h>
44 #include <asm/uaccess.h>
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
50 static DEFINE_SPINLOCK(kvm_lock);
51 static LIST_HEAD(vm_list);
53 static cpumask_t cpus_hardware_enabled;
55 struct kvm_arch_ops *kvm_arch_ops;
56 struct kmem_cache *kvm_vcpu_cache;
57 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
59 static __read_mostly struct preempt_ops kvm_preempt_ops;
61 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
63 static struct kvm_stats_debugfs_item {
66 struct dentry *dentry;
67 } debugfs_entries[] = {
68 { "pf_fixed", STAT_OFFSET(pf_fixed) },
69 { "pf_guest", STAT_OFFSET(pf_guest) },
70 { "tlb_flush", STAT_OFFSET(tlb_flush) },
71 { "invlpg", STAT_OFFSET(invlpg) },
72 { "exits", STAT_OFFSET(exits) },
73 { "io_exits", STAT_OFFSET(io_exits) },
74 { "mmio_exits", STAT_OFFSET(mmio_exits) },
75 { "signal_exits", STAT_OFFSET(signal_exits) },
76 { "irq_window", STAT_OFFSET(irq_window_exits) },
77 { "halt_exits", STAT_OFFSET(halt_exits) },
78 { "request_irq", STAT_OFFSET(request_irq_exits) },
79 { "irq_exits", STAT_OFFSET(irq_exits) },
80 { "light_exits", STAT_OFFSET(light_exits) },
81 { "efer_reload", STAT_OFFSET(efer_reload) },
85 static struct dentry *debugfs_dir;
87 #define MAX_IO_MSRS 256
89 #define CR0_RESERVED_BITS \
90 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
91 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
92 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
93 #define CR4_RESERVED_BITS \
94 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
95 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
96 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
97 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
99 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
100 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
103 // LDT or TSS descriptor in the GDT. 16 bytes.
104 struct segment_descriptor_64 {
105 struct segment_descriptor s;
112 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
115 unsigned long segment_base(u16 selector)
117 struct descriptor_table gdt;
118 struct segment_descriptor *d;
119 unsigned long table_base;
120 typedef unsigned long ul;
126 asm ("sgdt %0" : "=m"(gdt));
127 table_base = gdt.base;
129 if (selector & 4) { /* from ldt */
132 asm ("sldt %0" : "=g"(ldt_selector));
133 table_base = segment_base(ldt_selector);
135 d = (struct segment_descriptor *)(table_base + (selector & ~7));
136 v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
139 && (d->type == 2 || d->type == 9 || d->type == 11))
140 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
144 EXPORT_SYMBOL_GPL(segment_base);
146 static inline int valid_vcpu(int n)
148 return likely(n >= 0 && n < KVM_MAX_VCPUS);
151 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
153 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
156 vcpu->guest_fpu_loaded = 1;
157 fx_save(&vcpu->host_fx_image);
158 fx_restore(&vcpu->guest_fx_image);
160 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
162 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
164 if (!vcpu->guest_fpu_loaded)
167 vcpu->guest_fpu_loaded = 0;
168 fx_save(&vcpu->guest_fx_image);
169 fx_restore(&vcpu->host_fx_image);
171 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
174 * Switches to specified vcpu, until a matching vcpu_put()
176 static void vcpu_load(struct kvm_vcpu *vcpu)
180 mutex_lock(&vcpu->mutex);
182 preempt_notifier_register(&vcpu->preempt_notifier);
183 kvm_arch_ops->vcpu_load(vcpu, cpu);
187 static void vcpu_put(struct kvm_vcpu *vcpu)
190 kvm_arch_ops->vcpu_put(vcpu);
191 preempt_notifier_unregister(&vcpu->preempt_notifier);
193 mutex_unlock(&vcpu->mutex);
196 static void ack_flush(void *_completed)
198 atomic_t *completed = _completed;
200 atomic_inc(completed);
203 void kvm_flush_remote_tlbs(struct kvm *kvm)
207 struct kvm_vcpu *vcpu;
210 atomic_set(&completed, 0);
213 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
214 vcpu = kvm->vcpus[i];
217 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
220 if (cpu != -1 && cpu != raw_smp_processor_id())
221 if (!cpu_isset(cpu, cpus)) {
228 * We really want smp_call_function_mask() here. But that's not
229 * available, so ipi all cpus in parallel and wait for them
232 for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
233 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
234 while (atomic_read(&completed) != needed) {
240 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
245 mutex_init(&vcpu->mutex);
247 vcpu->mmu.root_hpa = INVALID_PAGE;
251 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
256 vcpu->run = page_address(page);
258 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
263 vcpu->pio_data = page_address(page);
265 r = kvm_mmu_create(vcpu);
267 goto fail_free_pio_data;
272 free_page((unsigned long)vcpu->pio_data);
274 free_page((unsigned long)vcpu->run);
278 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
280 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
282 kvm_mmu_destroy(vcpu);
283 free_page((unsigned long)vcpu->pio_data);
284 free_page((unsigned long)vcpu->run);
286 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
288 static struct kvm *kvm_create_vm(void)
290 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
293 return ERR_PTR(-ENOMEM);
295 kvm_io_bus_init(&kvm->pio_bus);
296 mutex_init(&kvm->lock);
297 INIT_LIST_HEAD(&kvm->active_mmu_pages);
298 kvm_io_bus_init(&kvm->mmio_bus);
299 spin_lock(&kvm_lock);
300 list_add(&kvm->vm_list, &vm_list);
301 spin_unlock(&kvm_lock);
305 static int kvm_dev_open(struct inode *inode, struct file *filp)
311 * Free any memory in @free but not in @dont.
313 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
314 struct kvm_memory_slot *dont)
318 if (!dont || free->phys_mem != dont->phys_mem)
319 if (free->phys_mem) {
320 for (i = 0; i < free->npages; ++i)
321 if (free->phys_mem[i])
322 __free_page(free->phys_mem[i]);
323 vfree(free->phys_mem);
326 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
327 vfree(free->dirty_bitmap);
329 free->phys_mem = NULL;
331 free->dirty_bitmap = NULL;
334 static void kvm_free_physmem(struct kvm *kvm)
338 for (i = 0; i < kvm->nmemslots; ++i)
339 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
342 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
346 for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
347 if (vcpu->pio.guest_pages[i]) {
348 __free_page(vcpu->pio.guest_pages[i]);
349 vcpu->pio.guest_pages[i] = NULL;
353 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
356 kvm_mmu_unload(vcpu);
360 static void kvm_free_vcpus(struct kvm *kvm)
365 * Unpin any mmu pages first.
367 for (i = 0; i < KVM_MAX_VCPUS; ++i)
369 kvm_unload_vcpu_mmu(kvm->vcpus[i]);
370 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
372 kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
373 kvm->vcpus[i] = NULL;
379 static int kvm_dev_release(struct inode *inode, struct file *filp)
384 static void kvm_destroy_vm(struct kvm *kvm)
386 spin_lock(&kvm_lock);
387 list_del(&kvm->vm_list);
388 spin_unlock(&kvm_lock);
389 kvm_io_bus_destroy(&kvm->pio_bus);
390 kvm_io_bus_destroy(&kvm->mmio_bus);
392 kvm_free_physmem(kvm);
396 static int kvm_vm_release(struct inode *inode, struct file *filp)
398 struct kvm *kvm = filp->private_data;
404 static void inject_gp(struct kvm_vcpu *vcpu)
406 kvm_arch_ops->inject_gp(vcpu, 0);
410 * Load the pae pdptrs. Return true is they are all valid.
412 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
414 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
415 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
420 u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
422 mutex_lock(&vcpu->kvm->lock);
423 page = gfn_to_page(vcpu->kvm, pdpt_gfn);
429 pdpt = kmap_atomic(page, KM_USER0);
430 memcpy(pdpte, pdpt+offset, sizeof(pdpte));
431 kunmap_atomic(pdpt, KM_USER0);
433 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
434 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
441 memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
443 mutex_unlock(&vcpu->kvm->lock);
448 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
450 if (cr0 & CR0_RESERVED_BITS) {
451 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
457 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
458 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
463 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
464 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
465 "and a clear PE flag\n");
470 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
472 if ((vcpu->shadow_efer & EFER_LME)) {
476 printk(KERN_DEBUG "set_cr0: #GP, start paging "
477 "in long mode while PAE is disabled\n");
481 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
483 printk(KERN_DEBUG "set_cr0: #GP, start paging "
484 "in long mode while CS.L == 1\n");
491 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
492 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
500 kvm_arch_ops->set_cr0(vcpu, cr0);
503 mutex_lock(&vcpu->kvm->lock);
504 kvm_mmu_reset_context(vcpu);
505 mutex_unlock(&vcpu->kvm->lock);
508 EXPORT_SYMBOL_GPL(set_cr0);
510 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
512 set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
514 EXPORT_SYMBOL_GPL(lmsw);
516 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
518 if (cr4 & CR4_RESERVED_BITS) {
519 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
524 if (is_long_mode(vcpu)) {
525 if (!(cr4 & X86_CR4_PAE)) {
526 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
531 } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
532 && !load_pdptrs(vcpu, vcpu->cr3)) {
533 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
538 if (cr4 & X86_CR4_VMXE) {
539 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
543 kvm_arch_ops->set_cr4(vcpu, cr4);
544 mutex_lock(&vcpu->kvm->lock);
545 kvm_mmu_reset_context(vcpu);
546 mutex_unlock(&vcpu->kvm->lock);
548 EXPORT_SYMBOL_GPL(set_cr4);
550 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
552 if (is_long_mode(vcpu)) {
553 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
554 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
560 if (cr3 & CR3_PAE_RESERVED_BITS) {
562 "set_cr3: #GP, reserved bits\n");
566 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
567 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
573 if (cr3 & CR3_NONPAE_RESERVED_BITS) {
575 "set_cr3: #GP, reserved bits\n");
582 mutex_lock(&vcpu->kvm->lock);
584 * Does the new cr3 value map to physical memory? (Note, we
585 * catch an invalid cr3 even in real-mode, because it would
586 * cause trouble later on when we turn on paging anyway.)
588 * A real CPU would silently accept an invalid cr3 and would
589 * attempt to use it - with largely undefined (and often hard
590 * to debug) behavior on the guest side.
592 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
596 vcpu->mmu.new_cr3(vcpu);
598 mutex_unlock(&vcpu->kvm->lock);
600 EXPORT_SYMBOL_GPL(set_cr3);
602 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
604 if (cr8 & CR8_RESERVED_BITS) {
605 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
611 EXPORT_SYMBOL_GPL(set_cr8);
613 void fx_init(struct kvm_vcpu *vcpu)
615 unsigned after_mxcsr_mask;
617 /* Initialize guest FPU by resetting ours and saving into guest's */
619 fx_save(&vcpu->host_fx_image);
621 fx_save(&vcpu->guest_fx_image);
622 fx_restore(&vcpu->host_fx_image);
625 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
626 vcpu->guest_fx_image.mxcsr = 0x1f80;
627 memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
628 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
630 EXPORT_SYMBOL_GPL(fx_init);
633 * Allocate some memory and give it an address in the guest physical address
636 * Discontiguous memory is allowed, mostly for framebuffers.
638 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
639 struct kvm_memory_region *mem)
643 unsigned long npages;
645 struct kvm_memory_slot *memslot;
646 struct kvm_memory_slot old, new;
647 int memory_config_version;
650 /* General sanity checks */
651 if (mem->memory_size & (PAGE_SIZE - 1))
653 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
655 if (mem->slot >= KVM_MEMORY_SLOTS)
657 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
660 memslot = &kvm->memslots[mem->slot];
661 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
662 npages = mem->memory_size >> PAGE_SHIFT;
665 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
668 mutex_lock(&kvm->lock);
670 memory_config_version = kvm->memory_config_version;
671 new = old = *memslot;
673 new.base_gfn = base_gfn;
675 new.flags = mem->flags;
677 /* Disallow changing a memory slot's size. */
679 if (npages && old.npages && npages != old.npages)
682 /* Check for overlaps */
684 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
685 struct kvm_memory_slot *s = &kvm->memslots[i];
689 if (!((base_gfn + npages <= s->base_gfn) ||
690 (base_gfn >= s->base_gfn + s->npages)))
694 * Do memory allocations outside lock. memory_config_version will
697 mutex_unlock(&kvm->lock);
699 /* Deallocate if slot is being removed */
703 /* Free page dirty bitmap if unneeded */
704 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
705 new.dirty_bitmap = NULL;
709 /* Allocate if a slot is being created */
710 if (npages && !new.phys_mem) {
711 new.phys_mem = vmalloc(npages * sizeof(struct page *));
716 memset(new.phys_mem, 0, npages * sizeof(struct page *));
717 for (i = 0; i < npages; ++i) {
718 new.phys_mem[i] = alloc_page(GFP_HIGHUSER
720 if (!new.phys_mem[i])
722 set_page_private(new.phys_mem[i],0);
726 /* Allocate page dirty bitmap if needed */
727 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
728 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
730 new.dirty_bitmap = vmalloc(dirty_bytes);
731 if (!new.dirty_bitmap)
733 memset(new.dirty_bitmap, 0, dirty_bytes);
736 mutex_lock(&kvm->lock);
738 if (memory_config_version != kvm->memory_config_version) {
739 mutex_unlock(&kvm->lock);
740 kvm_free_physmem_slot(&new, &old);
748 if (mem->slot >= kvm->nmemslots)
749 kvm->nmemslots = mem->slot + 1;
752 ++kvm->memory_config_version;
754 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
755 kvm_flush_remote_tlbs(kvm);
757 mutex_unlock(&kvm->lock);
759 kvm_free_physmem_slot(&old, &new);
763 mutex_unlock(&kvm->lock);
765 kvm_free_physmem_slot(&new, &old);
771 * Get (and clear) the dirty memory log for a memory slot.
773 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
774 struct kvm_dirty_log *log)
776 struct kvm_memory_slot *memslot;
779 unsigned long any = 0;
781 mutex_lock(&kvm->lock);
784 * Prevent changes to guest memory configuration even while the lock
788 mutex_unlock(&kvm->lock);
790 if (log->slot >= KVM_MEMORY_SLOTS)
793 memslot = &kvm->memslots[log->slot];
795 if (!memslot->dirty_bitmap)
798 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
800 for (i = 0; !any && i < n/sizeof(long); ++i)
801 any = memslot->dirty_bitmap[i];
804 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
807 /* If nothing is dirty, don't bother messing with page tables. */
809 mutex_lock(&kvm->lock);
810 kvm_mmu_slot_remove_write_access(kvm, log->slot);
811 kvm_flush_remote_tlbs(kvm);
812 memset(memslot->dirty_bitmap, 0, n);
813 mutex_unlock(&kvm->lock);
819 mutex_lock(&kvm->lock);
821 mutex_unlock(&kvm->lock);
826 * Set a new alias region. Aliases map a portion of physical memory into
827 * another portion. This is useful for memory windows, for example the PC
830 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
831 struct kvm_memory_alias *alias)
834 struct kvm_mem_alias *p;
837 /* General sanity checks */
838 if (alias->memory_size & (PAGE_SIZE - 1))
840 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
842 if (alias->slot >= KVM_ALIAS_SLOTS)
844 if (alias->guest_phys_addr + alias->memory_size
845 < alias->guest_phys_addr)
847 if (alias->target_phys_addr + alias->memory_size
848 < alias->target_phys_addr)
851 mutex_lock(&kvm->lock);
853 p = &kvm->aliases[alias->slot];
854 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
855 p->npages = alias->memory_size >> PAGE_SHIFT;
856 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
858 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
859 if (kvm->aliases[n - 1].npages)
863 kvm_mmu_zap_all(kvm);
865 mutex_unlock(&kvm->lock);
873 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
876 struct kvm_mem_alias *alias;
878 for (i = 0; i < kvm->naliases; ++i) {
879 alias = &kvm->aliases[i];
880 if (gfn >= alias->base_gfn
881 && gfn < alias->base_gfn + alias->npages)
882 return alias->target_gfn + gfn - alias->base_gfn;
887 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
891 for (i = 0; i < kvm->nmemslots; ++i) {
892 struct kvm_memory_slot *memslot = &kvm->memslots[i];
894 if (gfn >= memslot->base_gfn
895 && gfn < memslot->base_gfn + memslot->npages)
901 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
903 gfn = unalias_gfn(kvm, gfn);
904 return __gfn_to_memslot(kvm, gfn);
907 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
909 struct kvm_memory_slot *slot;
911 gfn = unalias_gfn(kvm, gfn);
912 slot = __gfn_to_memslot(kvm, gfn);
915 return slot->phys_mem[gfn - slot->base_gfn];
917 EXPORT_SYMBOL_GPL(gfn_to_page);
919 /* WARNING: Does not work on aliased pages. */
920 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
922 struct kvm_memory_slot *memslot;
924 memslot = __gfn_to_memslot(kvm, gfn);
925 if (memslot && memslot->dirty_bitmap) {
926 unsigned long rel_gfn = gfn - memslot->base_gfn;
929 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
930 set_bit(rel_gfn, memslot->dirty_bitmap);
934 int emulator_read_std(unsigned long addr,
937 struct kvm_vcpu *vcpu)
942 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
943 unsigned offset = addr & (PAGE_SIZE-1);
944 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
949 if (gpa == UNMAPPED_GVA)
950 return X86EMUL_PROPAGATE_FAULT;
951 pfn = gpa >> PAGE_SHIFT;
952 page = gfn_to_page(vcpu->kvm, pfn);
954 return X86EMUL_UNHANDLEABLE;
955 page_virt = kmap_atomic(page, KM_USER0);
957 memcpy(data, page_virt + offset, tocopy);
959 kunmap_atomic(page_virt, KM_USER0);
966 return X86EMUL_CONTINUE;
968 EXPORT_SYMBOL_GPL(emulator_read_std);
970 static int emulator_write_std(unsigned long addr,
973 struct kvm_vcpu *vcpu)
975 printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
977 return X86EMUL_UNHANDLEABLE;
980 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
984 * Note that its important to have this wrapper function because
985 * in the very near future we will be checking for MMIOs against
986 * the LAPIC as well as the general MMIO bus
988 return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
991 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
994 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
997 static int emulator_read_emulated(unsigned long addr,
1000 struct kvm_vcpu *vcpu)
1002 struct kvm_io_device *mmio_dev;
1005 if (vcpu->mmio_read_completed) {
1006 memcpy(val, vcpu->mmio_data, bytes);
1007 vcpu->mmio_read_completed = 0;
1008 return X86EMUL_CONTINUE;
1009 } else if (emulator_read_std(addr, val, bytes, vcpu)
1010 == X86EMUL_CONTINUE)
1011 return X86EMUL_CONTINUE;
1013 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1014 if (gpa == UNMAPPED_GVA)
1015 return X86EMUL_PROPAGATE_FAULT;
1018 * Is this MMIO handled locally?
1020 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1022 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1023 return X86EMUL_CONTINUE;
1026 vcpu->mmio_needed = 1;
1027 vcpu->mmio_phys_addr = gpa;
1028 vcpu->mmio_size = bytes;
1029 vcpu->mmio_is_write = 0;
1031 return X86EMUL_UNHANDLEABLE;
1034 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1035 const void *val, int bytes)
1040 if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1042 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1045 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1046 virt = kmap_atomic(page, KM_USER0);
1047 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1048 memcpy(virt + offset_in_page(gpa), val, bytes);
1049 kunmap_atomic(virt, KM_USER0);
1053 static int emulator_write_emulated_onepage(unsigned long addr,
1056 struct kvm_vcpu *vcpu)
1058 struct kvm_io_device *mmio_dev;
1059 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1061 if (gpa == UNMAPPED_GVA) {
1062 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1063 return X86EMUL_PROPAGATE_FAULT;
1066 if (emulator_write_phys(vcpu, gpa, val, bytes))
1067 return X86EMUL_CONTINUE;
1070 * Is this MMIO handled locally?
1072 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1074 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1075 return X86EMUL_CONTINUE;
1078 vcpu->mmio_needed = 1;
1079 vcpu->mmio_phys_addr = gpa;
1080 vcpu->mmio_size = bytes;
1081 vcpu->mmio_is_write = 1;
1082 memcpy(vcpu->mmio_data, val, bytes);
1084 return X86EMUL_CONTINUE;
1087 int emulator_write_emulated(unsigned long addr,
1090 struct kvm_vcpu *vcpu)
1092 /* Crossing a page boundary? */
1093 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1096 now = -addr & ~PAGE_MASK;
1097 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1098 if (rc != X86EMUL_CONTINUE)
1104 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1106 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1108 static int emulator_cmpxchg_emulated(unsigned long addr,
1112 struct kvm_vcpu *vcpu)
1114 static int reported;
1118 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1120 return emulator_write_emulated(addr, new, bytes, vcpu);
1123 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1125 return kvm_arch_ops->get_segment_base(vcpu, seg);
1128 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1130 return X86EMUL_CONTINUE;
1133 int emulate_clts(struct kvm_vcpu *vcpu)
1137 cr0 = vcpu->cr0 & ~X86_CR0_TS;
1138 kvm_arch_ops->set_cr0(vcpu, cr0);
1139 return X86EMUL_CONTINUE;
1142 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1144 struct kvm_vcpu *vcpu = ctxt->vcpu;
1148 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1149 return X86EMUL_CONTINUE;
1151 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1153 return X86EMUL_UNHANDLEABLE;
1157 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1159 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1162 kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1164 /* FIXME: better handling */
1165 return X86EMUL_UNHANDLEABLE;
1167 return X86EMUL_CONTINUE;
1170 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1172 static int reported;
1174 unsigned long rip = ctxt->vcpu->rip;
1175 unsigned long rip_linear;
1177 rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1182 emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
1184 printk(KERN_ERR "emulation failed but !mmio_needed?"
1185 " rip %lx %02x %02x %02x %02x\n",
1186 rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1190 struct x86_emulate_ops emulate_ops = {
1191 .read_std = emulator_read_std,
1192 .write_std = emulator_write_std,
1193 .read_emulated = emulator_read_emulated,
1194 .write_emulated = emulator_write_emulated,
1195 .cmpxchg_emulated = emulator_cmpxchg_emulated,
1198 int emulate_instruction(struct kvm_vcpu *vcpu,
1199 struct kvm_run *run,
1203 struct x86_emulate_ctxt emulate_ctxt;
1207 vcpu->mmio_fault_cr2 = cr2;
1208 kvm_arch_ops->cache_regs(vcpu);
1210 kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1212 emulate_ctxt.vcpu = vcpu;
1213 emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1214 emulate_ctxt.cr2 = cr2;
1215 emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1216 ? X86EMUL_MODE_REAL : cs_l
1217 ? X86EMUL_MODE_PROT64 : cs_db
1218 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1220 if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1221 emulate_ctxt.cs_base = 0;
1222 emulate_ctxt.ds_base = 0;
1223 emulate_ctxt.es_base = 0;
1224 emulate_ctxt.ss_base = 0;
1226 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1227 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1228 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1229 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1232 emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1233 emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1235 vcpu->mmio_is_write = 0;
1236 r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1238 if ((r || vcpu->mmio_is_write) && run) {
1239 run->exit_reason = KVM_EXIT_MMIO;
1240 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1241 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1242 run->mmio.len = vcpu->mmio_size;
1243 run->mmio.is_write = vcpu->mmio_is_write;
1247 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1248 return EMULATE_DONE;
1249 if (!vcpu->mmio_needed) {
1250 report_emulation_failure(&emulate_ctxt);
1251 return EMULATE_FAIL;
1253 return EMULATE_DO_MMIO;
1256 kvm_arch_ops->decache_regs(vcpu);
1257 kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1259 if (vcpu->mmio_is_write) {
1260 vcpu->mmio_needed = 0;
1261 return EMULATE_DO_MMIO;
1264 return EMULATE_DONE;
1266 EXPORT_SYMBOL_GPL(emulate_instruction);
1268 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1270 if (vcpu->irq_summary)
1273 vcpu->run->exit_reason = KVM_EXIT_HLT;
1274 ++vcpu->stat.halt_exits;
1277 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1279 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1281 unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1283 kvm_arch_ops->cache_regs(vcpu);
1285 #ifdef CONFIG_X86_64
1286 if (is_long_mode(vcpu)) {
1287 nr = vcpu->regs[VCPU_REGS_RAX];
1288 a0 = vcpu->regs[VCPU_REGS_RDI];
1289 a1 = vcpu->regs[VCPU_REGS_RSI];
1290 a2 = vcpu->regs[VCPU_REGS_RDX];
1291 a3 = vcpu->regs[VCPU_REGS_RCX];
1292 a4 = vcpu->regs[VCPU_REGS_R8];
1293 a5 = vcpu->regs[VCPU_REGS_R9];
1297 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1298 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1299 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1300 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1301 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1302 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1303 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1307 run->hypercall.nr = nr;
1308 run->hypercall.args[0] = a0;
1309 run->hypercall.args[1] = a1;
1310 run->hypercall.args[2] = a2;
1311 run->hypercall.args[3] = a3;
1312 run->hypercall.args[4] = a4;
1313 run->hypercall.args[5] = a5;
1314 run->hypercall.ret = ret;
1315 run->hypercall.longmode = is_long_mode(vcpu);
1316 kvm_arch_ops->decache_regs(vcpu);
1319 vcpu->regs[VCPU_REGS_RAX] = ret;
1320 kvm_arch_ops->decache_regs(vcpu);
1323 EXPORT_SYMBOL_GPL(kvm_hypercall);
1325 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1327 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1330 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1332 struct descriptor_table dt = { limit, base };
1334 kvm_arch_ops->set_gdt(vcpu, &dt);
1337 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1339 struct descriptor_table dt = { limit, base };
1341 kvm_arch_ops->set_idt(vcpu, &dt);
1344 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1345 unsigned long *rflags)
1348 *rflags = kvm_arch_ops->get_rflags(vcpu);
1351 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1353 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1364 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1369 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1370 unsigned long *rflags)
1374 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1375 *rflags = kvm_arch_ops->get_rflags(vcpu);
1384 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1387 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1392 * Register the para guest with the host:
1394 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1396 struct kvm_vcpu_para_state *para_state;
1397 hpa_t para_state_hpa, hypercall_hpa;
1398 struct page *para_state_page;
1399 unsigned char *hypercall;
1400 gpa_t hypercall_gpa;
1402 printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1403 printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1406 * Needs to be page aligned:
1408 if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1411 para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1412 printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1413 if (is_error_hpa(para_state_hpa))
1416 mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1417 para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1418 para_state = kmap(para_state_page);
1420 printk(KERN_DEBUG ".... guest version: %d\n", para_state->guest_version);
1421 printk(KERN_DEBUG ".... size: %d\n", para_state->size);
1423 para_state->host_version = KVM_PARA_API_VERSION;
1425 * We cannot support guests that try to register themselves
1426 * with a newer API version than the host supports:
1428 if (para_state->guest_version > KVM_PARA_API_VERSION) {
1429 para_state->ret = -KVM_EINVAL;
1430 goto err_kunmap_skip;
1433 hypercall_gpa = para_state->hypercall_gpa;
1434 hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1435 printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1436 if (is_error_hpa(hypercall_hpa)) {
1437 para_state->ret = -KVM_EINVAL;
1438 goto err_kunmap_skip;
1441 printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1442 vcpu->para_state_page = para_state_page;
1443 vcpu->para_state_gpa = para_state_gpa;
1444 vcpu->hypercall_gpa = hypercall_gpa;
1446 mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1447 hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1448 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1449 kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1450 kunmap_atomic(hypercall, KM_USER1);
1452 para_state->ret = 0;
1454 kunmap(para_state_page);
1460 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1465 case 0xc0010010: /* SYSCFG */
1466 case 0xc0010015: /* HWCR */
1467 case MSR_IA32_PLATFORM_ID:
1468 case MSR_IA32_P5_MC_ADDR:
1469 case MSR_IA32_P5_MC_TYPE:
1470 case MSR_IA32_MC0_CTL:
1471 case MSR_IA32_MCG_STATUS:
1472 case MSR_IA32_MCG_CAP:
1473 case MSR_IA32_MC0_MISC:
1474 case MSR_IA32_MC0_MISC+4:
1475 case MSR_IA32_MC0_MISC+8:
1476 case MSR_IA32_MC0_MISC+12:
1477 case MSR_IA32_MC0_MISC+16:
1478 case MSR_IA32_UCODE_REV:
1479 case MSR_IA32_PERF_STATUS:
1480 case MSR_IA32_EBL_CR_POWERON:
1481 /* MTRR registers */
1483 case 0x200 ... 0x2ff:
1486 case 0xcd: /* fsb frequency */
1489 case MSR_IA32_APICBASE:
1490 data = vcpu->apic_base;
1492 case MSR_IA32_MISC_ENABLE:
1493 data = vcpu->ia32_misc_enable_msr;
1495 #ifdef CONFIG_X86_64
1497 data = vcpu->shadow_efer;
1501 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1507 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1510 * Reads an msr value (of 'msr_index') into 'pdata'.
1511 * Returns 0 on success, non-0 otherwise.
1512 * Assumes vcpu_load() was already called.
1514 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1516 return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1519 #ifdef CONFIG_X86_64
1521 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1523 if (efer & EFER_RESERVED_BITS) {
1524 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1531 && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1532 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1537 kvm_arch_ops->set_efer(vcpu, efer);
1540 efer |= vcpu->shadow_efer & EFER_LMA;
1542 vcpu->shadow_efer = efer;
1547 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1550 #ifdef CONFIG_X86_64
1552 set_efer(vcpu, data);
1555 case MSR_IA32_MC0_STATUS:
1556 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1557 __FUNCTION__, data);
1559 case MSR_IA32_MCG_STATUS:
1560 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1561 __FUNCTION__, data);
1563 case MSR_IA32_UCODE_REV:
1564 case MSR_IA32_UCODE_WRITE:
1565 case 0x200 ... 0x2ff: /* MTRRs */
1567 case MSR_IA32_APICBASE:
1568 vcpu->apic_base = data;
1570 case MSR_IA32_MISC_ENABLE:
1571 vcpu->ia32_misc_enable_msr = data;
1574 * This is the 'probe whether the host is KVM' logic:
1576 case MSR_KVM_API_MAGIC:
1577 return vcpu_register_para(vcpu, data);
1580 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1585 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1588 * Writes msr value into into the appropriate "register".
1589 * Returns 0 on success, non-0 otherwise.
1590 * Assumes vcpu_load() was already called.
1592 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1594 return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1597 void kvm_resched(struct kvm_vcpu *vcpu)
1599 if (!need_resched())
1603 EXPORT_SYMBOL_GPL(kvm_resched);
1605 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1609 struct kvm_cpuid_entry *e, *best;
1611 kvm_arch_ops->cache_regs(vcpu);
1612 function = vcpu->regs[VCPU_REGS_RAX];
1613 vcpu->regs[VCPU_REGS_RAX] = 0;
1614 vcpu->regs[VCPU_REGS_RBX] = 0;
1615 vcpu->regs[VCPU_REGS_RCX] = 0;
1616 vcpu->regs[VCPU_REGS_RDX] = 0;
1618 for (i = 0; i < vcpu->cpuid_nent; ++i) {
1619 e = &vcpu->cpuid_entries[i];
1620 if (e->function == function) {
1625 * Both basic or both extended?
1627 if (((e->function ^ function) & 0x80000000) == 0)
1628 if (!best || e->function > best->function)
1632 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1633 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1634 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1635 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1637 kvm_arch_ops->decache_regs(vcpu);
1638 kvm_arch_ops->skip_emulated_instruction(vcpu);
1640 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1642 static int pio_copy_data(struct kvm_vcpu *vcpu)
1644 void *p = vcpu->pio_data;
1647 int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1649 q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1652 free_pio_guest_pages(vcpu);
1655 q += vcpu->pio.guest_page_offset;
1656 bytes = vcpu->pio.size * vcpu->pio.cur_count;
1658 memcpy(q, p, bytes);
1660 memcpy(p, q, bytes);
1661 q -= vcpu->pio.guest_page_offset;
1663 free_pio_guest_pages(vcpu);
1667 static int complete_pio(struct kvm_vcpu *vcpu)
1669 struct kvm_pio_request *io = &vcpu->pio;
1673 kvm_arch_ops->cache_regs(vcpu);
1677 memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1681 r = pio_copy_data(vcpu);
1683 kvm_arch_ops->cache_regs(vcpu);
1690 delta *= io->cur_count;
1692 * The size of the register should really depend on
1693 * current address size.
1695 vcpu->regs[VCPU_REGS_RCX] -= delta;
1701 vcpu->regs[VCPU_REGS_RDI] += delta;
1703 vcpu->regs[VCPU_REGS_RSI] += delta;
1706 kvm_arch_ops->decache_regs(vcpu);
1708 io->count -= io->cur_count;
1712 kvm_arch_ops->skip_emulated_instruction(vcpu);
1716 static void kernel_pio(struct kvm_io_device *pio_dev,
1717 struct kvm_vcpu *vcpu,
1720 /* TODO: String I/O for in kernel device */
1723 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1727 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1732 static void pio_string_write(struct kvm_io_device *pio_dev,
1733 struct kvm_vcpu *vcpu)
1735 struct kvm_pio_request *io = &vcpu->pio;
1736 void *pd = vcpu->pio_data;
1739 for (i = 0; i < io->cur_count; i++) {
1740 kvm_iodevice_write(pio_dev, io->port,
1747 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1748 int size, unsigned long count, int string, int down,
1749 gva_t address, int rep, unsigned port)
1751 unsigned now, in_page;
1755 struct kvm_io_device *pio_dev;
1757 vcpu->run->exit_reason = KVM_EXIT_IO;
1758 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1759 vcpu->run->io.size = size;
1760 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1761 vcpu->run->io.count = count;
1762 vcpu->run->io.port = port;
1763 vcpu->pio.count = count;
1764 vcpu->pio.cur_count = count;
1765 vcpu->pio.size = size;
1767 vcpu->pio.port = port;
1768 vcpu->pio.string = string;
1769 vcpu->pio.down = down;
1770 vcpu->pio.guest_page_offset = offset_in_page(address);
1771 vcpu->pio.rep = rep;
1773 pio_dev = vcpu_find_pio_dev(vcpu, port);
1775 kvm_arch_ops->cache_regs(vcpu);
1776 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1777 kvm_arch_ops->decache_regs(vcpu);
1779 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1787 kvm_arch_ops->skip_emulated_instruction(vcpu);
1791 now = min(count, PAGE_SIZE / size);
1794 in_page = PAGE_SIZE - offset_in_page(address);
1796 in_page = offset_in_page(address) + size;
1797 now = min(count, (unsigned long)in_page / size);
1800 * String I/O straddles page boundary. Pin two guest pages
1801 * so that we satisfy atomicity constraints. Do just one
1802 * transaction to avoid complexity.
1809 * String I/O in reverse. Yuck. Kill the guest, fix later.
1811 printk(KERN_ERR "kvm: guest string pio down\n");
1815 vcpu->run->io.count = now;
1816 vcpu->pio.cur_count = now;
1818 for (i = 0; i < nr_pages; ++i) {
1819 mutex_lock(&vcpu->kvm->lock);
1820 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1823 vcpu->pio.guest_pages[i] = page;
1824 mutex_unlock(&vcpu->kvm->lock);
1827 free_pio_guest_pages(vcpu);
1832 if (!vcpu->pio.in) {
1833 /* string PIO write */
1834 ret = pio_copy_data(vcpu);
1835 if (ret >= 0 && pio_dev) {
1836 pio_string_write(pio_dev, vcpu);
1838 if (vcpu->pio.count == 0)
1842 printk(KERN_ERR "no string pio read support yet, "
1843 "port %x size %d count %ld\n",
1848 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1850 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1857 if (vcpu->sigset_active)
1858 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1860 /* re-sync apic's tpr */
1861 vcpu->cr8 = kvm_run->cr8;
1863 if (vcpu->pio.cur_count) {
1864 r = complete_pio(vcpu);
1869 if (vcpu->mmio_needed) {
1870 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1871 vcpu->mmio_read_completed = 1;
1872 vcpu->mmio_needed = 0;
1873 r = emulate_instruction(vcpu, kvm_run,
1874 vcpu->mmio_fault_cr2, 0);
1875 if (r == EMULATE_DO_MMIO) {
1877 * Read-modify-write. Back to userspace.
1884 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1885 kvm_arch_ops->cache_regs(vcpu);
1886 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1887 kvm_arch_ops->decache_regs(vcpu);
1890 r = kvm_arch_ops->run(vcpu, kvm_run);
1893 if (vcpu->sigset_active)
1894 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1900 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1901 struct kvm_regs *regs)
1905 kvm_arch_ops->cache_regs(vcpu);
1907 regs->rax = vcpu->regs[VCPU_REGS_RAX];
1908 regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1909 regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1910 regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1911 regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1912 regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1913 regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1914 regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1915 #ifdef CONFIG_X86_64
1916 regs->r8 = vcpu->regs[VCPU_REGS_R8];
1917 regs->r9 = vcpu->regs[VCPU_REGS_R9];
1918 regs->r10 = vcpu->regs[VCPU_REGS_R10];
1919 regs->r11 = vcpu->regs[VCPU_REGS_R11];
1920 regs->r12 = vcpu->regs[VCPU_REGS_R12];
1921 regs->r13 = vcpu->regs[VCPU_REGS_R13];
1922 regs->r14 = vcpu->regs[VCPU_REGS_R14];
1923 regs->r15 = vcpu->regs[VCPU_REGS_R15];
1926 regs->rip = vcpu->rip;
1927 regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1930 * Don't leak debug flags in case they were set for guest debugging
1932 if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1933 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1940 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1941 struct kvm_regs *regs)
1945 vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1946 vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1947 vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1948 vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1949 vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1950 vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1951 vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1952 vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1953 #ifdef CONFIG_X86_64
1954 vcpu->regs[VCPU_REGS_R8] = regs->r8;
1955 vcpu->regs[VCPU_REGS_R9] = regs->r9;
1956 vcpu->regs[VCPU_REGS_R10] = regs->r10;
1957 vcpu->regs[VCPU_REGS_R11] = regs->r11;
1958 vcpu->regs[VCPU_REGS_R12] = regs->r12;
1959 vcpu->regs[VCPU_REGS_R13] = regs->r13;
1960 vcpu->regs[VCPU_REGS_R14] = regs->r14;
1961 vcpu->regs[VCPU_REGS_R15] = regs->r15;
1964 vcpu->rip = regs->rip;
1965 kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1967 kvm_arch_ops->decache_regs(vcpu);
1974 static void get_segment(struct kvm_vcpu *vcpu,
1975 struct kvm_segment *var, int seg)
1977 return kvm_arch_ops->get_segment(vcpu, var, seg);
1980 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1981 struct kvm_sregs *sregs)
1983 struct descriptor_table dt;
1987 get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1988 get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1989 get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1990 get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1991 get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1992 get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1994 get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1995 get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1997 kvm_arch_ops->get_idt(vcpu, &dt);
1998 sregs->idt.limit = dt.limit;
1999 sregs->idt.base = dt.base;
2000 kvm_arch_ops->get_gdt(vcpu, &dt);
2001 sregs->gdt.limit = dt.limit;
2002 sregs->gdt.base = dt.base;
2004 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2005 sregs->cr0 = vcpu->cr0;
2006 sregs->cr2 = vcpu->cr2;
2007 sregs->cr3 = vcpu->cr3;
2008 sregs->cr4 = vcpu->cr4;
2009 sregs->cr8 = vcpu->cr8;
2010 sregs->efer = vcpu->shadow_efer;
2011 sregs->apic_base = vcpu->apic_base;
2013 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2014 sizeof sregs->interrupt_bitmap);
2021 static void set_segment(struct kvm_vcpu *vcpu,
2022 struct kvm_segment *var, int seg)
2024 return kvm_arch_ops->set_segment(vcpu, var, seg);
2027 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2028 struct kvm_sregs *sregs)
2030 int mmu_reset_needed = 0;
2032 struct descriptor_table dt;
2036 dt.limit = sregs->idt.limit;
2037 dt.base = sregs->idt.base;
2038 kvm_arch_ops->set_idt(vcpu, &dt);
2039 dt.limit = sregs->gdt.limit;
2040 dt.base = sregs->gdt.base;
2041 kvm_arch_ops->set_gdt(vcpu, &dt);
2043 vcpu->cr2 = sregs->cr2;
2044 mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2045 vcpu->cr3 = sregs->cr3;
2047 vcpu->cr8 = sregs->cr8;
2049 mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2050 #ifdef CONFIG_X86_64
2051 kvm_arch_ops->set_efer(vcpu, sregs->efer);
2053 vcpu->apic_base = sregs->apic_base;
2055 kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2057 mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2058 kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2060 mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2061 kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2062 if (!is_long_mode(vcpu) && is_pae(vcpu))
2063 load_pdptrs(vcpu, vcpu->cr3);
2065 if (mmu_reset_needed)
2066 kvm_mmu_reset_context(vcpu);
2068 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2069 sizeof vcpu->irq_pending);
2070 vcpu->irq_summary = 0;
2071 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2072 if (vcpu->irq_pending[i])
2073 __set_bit(i, &vcpu->irq_summary);
2075 set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2076 set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2077 set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2078 set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2079 set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2080 set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2082 set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2083 set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2091 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2092 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2094 * This list is modified at module load time to reflect the
2095 * capabilities of the host cpu.
2097 static u32 msrs_to_save[] = {
2098 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2100 #ifdef CONFIG_X86_64
2101 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2103 MSR_IA32_TIME_STAMP_COUNTER,
2106 static unsigned num_msrs_to_save;
2108 static u32 emulated_msrs[] = {
2109 MSR_IA32_MISC_ENABLE,
2112 static __init void kvm_init_msr_list(void)
2117 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2118 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2121 msrs_to_save[j] = msrs_to_save[i];
2124 num_msrs_to_save = j;
2128 * Adapt set_msr() to msr_io()'s calling convention
2130 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2132 return kvm_set_msr(vcpu, index, *data);
2136 * Read or write a bunch of msrs. All parameters are kernel addresses.
2138 * @return number of msrs set successfully.
2140 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2141 struct kvm_msr_entry *entries,
2142 int (*do_msr)(struct kvm_vcpu *vcpu,
2143 unsigned index, u64 *data))
2149 for (i = 0; i < msrs->nmsrs; ++i)
2150 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2159 * Read or write a bunch of msrs. Parameters are user addresses.
2161 * @return number of msrs set successfully.
2163 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2164 int (*do_msr)(struct kvm_vcpu *vcpu,
2165 unsigned index, u64 *data),
2168 struct kvm_msrs msrs;
2169 struct kvm_msr_entry *entries;
2174 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2178 if (msrs.nmsrs >= MAX_IO_MSRS)
2182 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2183 entries = vmalloc(size);
2188 if (copy_from_user(entries, user_msrs->entries, size))
2191 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2196 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2208 * Translate a guest virtual address to a guest physical address.
2210 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2211 struct kvm_translation *tr)
2213 unsigned long vaddr = tr->linear_address;
2217 mutex_lock(&vcpu->kvm->lock);
2218 gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2219 tr->physical_address = gpa;
2220 tr->valid = gpa != UNMAPPED_GVA;
2223 mutex_unlock(&vcpu->kvm->lock);
2229 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2230 struct kvm_interrupt *irq)
2232 if (irq->irq < 0 || irq->irq >= 256)
2236 set_bit(irq->irq, vcpu->irq_pending);
2237 set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2244 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2245 struct kvm_debug_guest *dbg)
2251 r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2258 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2259 unsigned long address,
2262 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2263 unsigned long pgoff;
2266 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2268 page = virt_to_page(vcpu->run);
2269 else if (pgoff == KVM_PIO_PAGE_OFFSET)
2270 page = virt_to_page(vcpu->pio_data);
2272 return NOPAGE_SIGBUS;
2275 *type = VM_FAULT_MINOR;
2280 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2281 .nopage = kvm_vcpu_nopage,
2284 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2286 vma->vm_ops = &kvm_vcpu_vm_ops;
2290 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2292 struct kvm_vcpu *vcpu = filp->private_data;
2294 fput(vcpu->kvm->filp);
2298 static struct file_operations kvm_vcpu_fops = {
2299 .release = kvm_vcpu_release,
2300 .unlocked_ioctl = kvm_vcpu_ioctl,
2301 .compat_ioctl = kvm_vcpu_ioctl,
2302 .mmap = kvm_vcpu_mmap,
2306 * Allocates an inode for the vcpu.
2308 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2311 struct inode *inode;
2314 r = anon_inode_getfd(&fd, &inode, &file,
2315 "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2318 atomic_inc(&vcpu->kvm->filp->f_count);
2323 * Creates some virtual cpus. Good luck creating more than one.
2325 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2328 struct kvm_vcpu *vcpu;
2333 vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2335 return PTR_ERR(vcpu);
2337 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2339 /* We do fxsave: this must be aligned. */
2340 BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2343 r = kvm_mmu_setup(vcpu);
2348 mutex_lock(&kvm->lock);
2349 if (kvm->vcpus[n]) {
2351 mutex_unlock(&kvm->lock);
2354 kvm->vcpus[n] = vcpu;
2355 mutex_unlock(&kvm->lock);
2357 /* Now it's all set up, let userspace reach it */
2358 r = create_vcpu_fd(vcpu);
2364 mutex_lock(&kvm->lock);
2365 kvm->vcpus[n] = NULL;
2366 mutex_unlock(&kvm->lock);
2370 kvm_mmu_unload(vcpu);
2374 kvm_arch_ops->vcpu_free(vcpu);
2378 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2382 struct kvm_cpuid_entry *e, *entry;
2384 rdmsrl(MSR_EFER, efer);
2386 for (i = 0; i < vcpu->cpuid_nent; ++i) {
2387 e = &vcpu->cpuid_entries[i];
2388 if (e->function == 0x80000001) {
2393 if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2394 entry->edx &= ~(1 << 20);
2395 printk(KERN_INFO "kvm: guest NX capability removed\n");
2399 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2400 struct kvm_cpuid *cpuid,
2401 struct kvm_cpuid_entry __user *entries)
2406 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2409 if (copy_from_user(&vcpu->cpuid_entries, entries,
2410 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2412 vcpu->cpuid_nent = cpuid->nent;
2413 cpuid_fix_nx_cap(vcpu);
2420 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2423 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2424 vcpu->sigset_active = 1;
2425 vcpu->sigset = *sigset;
2427 vcpu->sigset_active = 0;
2432 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2433 * we have asm/x86/processor.h
2444 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2445 #ifdef CONFIG_X86_64
2446 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2448 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2452 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2454 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2458 memcpy(fpu->fpr, fxsave->st_space, 128);
2459 fpu->fcw = fxsave->cwd;
2460 fpu->fsw = fxsave->swd;
2461 fpu->ftwx = fxsave->twd;
2462 fpu->last_opcode = fxsave->fop;
2463 fpu->last_ip = fxsave->rip;
2464 fpu->last_dp = fxsave->rdp;
2465 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2472 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2474 struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2478 memcpy(fxsave->st_space, fpu->fpr, 128);
2479 fxsave->cwd = fpu->fcw;
2480 fxsave->swd = fpu->fsw;
2481 fxsave->twd = fpu->ftwx;
2482 fxsave->fop = fpu->last_opcode;
2483 fxsave->rip = fpu->last_ip;
2484 fxsave->rdp = fpu->last_dp;
2485 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2492 static long kvm_vcpu_ioctl(struct file *filp,
2493 unsigned int ioctl, unsigned long arg)
2495 struct kvm_vcpu *vcpu = filp->private_data;
2496 void __user *argp = (void __user *)arg;
2504 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2506 case KVM_GET_REGS: {
2507 struct kvm_regs kvm_regs;
2509 memset(&kvm_regs, 0, sizeof kvm_regs);
2510 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2514 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2519 case KVM_SET_REGS: {
2520 struct kvm_regs kvm_regs;
2523 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2525 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2531 case KVM_GET_SREGS: {
2532 struct kvm_sregs kvm_sregs;
2534 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2535 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2539 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2544 case KVM_SET_SREGS: {
2545 struct kvm_sregs kvm_sregs;
2548 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2550 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2556 case KVM_TRANSLATE: {
2557 struct kvm_translation tr;
2560 if (copy_from_user(&tr, argp, sizeof tr))
2562 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2566 if (copy_to_user(argp, &tr, sizeof tr))
2571 case KVM_INTERRUPT: {
2572 struct kvm_interrupt irq;
2575 if (copy_from_user(&irq, argp, sizeof irq))
2577 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2583 case KVM_DEBUG_GUEST: {
2584 struct kvm_debug_guest dbg;
2587 if (copy_from_user(&dbg, argp, sizeof dbg))
2589 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2596 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2599 r = msr_io(vcpu, argp, do_set_msr, 0);
2601 case KVM_SET_CPUID: {
2602 struct kvm_cpuid __user *cpuid_arg = argp;
2603 struct kvm_cpuid cpuid;
2606 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2608 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2613 case KVM_SET_SIGNAL_MASK: {
2614 struct kvm_signal_mask __user *sigmask_arg = argp;
2615 struct kvm_signal_mask kvm_sigmask;
2616 sigset_t sigset, *p;
2621 if (copy_from_user(&kvm_sigmask, argp,
2622 sizeof kvm_sigmask))
2625 if (kvm_sigmask.len != sizeof sigset)
2628 if (copy_from_user(&sigset, sigmask_arg->sigset,
2633 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2639 memset(&fpu, 0, sizeof fpu);
2640 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2644 if (copy_to_user(argp, &fpu, sizeof fpu))
2653 if (copy_from_user(&fpu, argp, sizeof fpu))
2655 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2668 static long kvm_vm_ioctl(struct file *filp,
2669 unsigned int ioctl, unsigned long arg)
2671 struct kvm *kvm = filp->private_data;
2672 void __user *argp = (void __user *)arg;
2676 case KVM_CREATE_VCPU:
2677 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2681 case KVM_SET_MEMORY_REGION: {
2682 struct kvm_memory_region kvm_mem;
2685 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2687 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2692 case KVM_GET_DIRTY_LOG: {
2693 struct kvm_dirty_log log;
2696 if (copy_from_user(&log, argp, sizeof log))
2698 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2703 case KVM_SET_MEMORY_ALIAS: {
2704 struct kvm_memory_alias alias;
2707 if (copy_from_user(&alias, argp, sizeof alias))
2709 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2721 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2722 unsigned long address,
2725 struct kvm *kvm = vma->vm_file->private_data;
2726 unsigned long pgoff;
2729 pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2730 page = gfn_to_page(kvm, pgoff);
2732 return NOPAGE_SIGBUS;
2735 *type = VM_FAULT_MINOR;
2740 static struct vm_operations_struct kvm_vm_vm_ops = {
2741 .nopage = kvm_vm_nopage,
2744 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2746 vma->vm_ops = &kvm_vm_vm_ops;
2750 static struct file_operations kvm_vm_fops = {
2751 .release = kvm_vm_release,
2752 .unlocked_ioctl = kvm_vm_ioctl,
2753 .compat_ioctl = kvm_vm_ioctl,
2754 .mmap = kvm_vm_mmap,
2757 static int kvm_dev_ioctl_create_vm(void)
2760 struct inode *inode;
2764 kvm = kvm_create_vm();
2766 return PTR_ERR(kvm);
2767 r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2769 kvm_destroy_vm(kvm);
2778 static long kvm_dev_ioctl(struct file *filp,
2779 unsigned int ioctl, unsigned long arg)
2781 void __user *argp = (void __user *)arg;
2785 case KVM_GET_API_VERSION:
2789 r = KVM_API_VERSION;
2795 r = kvm_dev_ioctl_create_vm();
2797 case KVM_GET_MSR_INDEX_LIST: {
2798 struct kvm_msr_list __user *user_msr_list = argp;
2799 struct kvm_msr_list msr_list;
2803 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2806 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2807 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2810 if (n < num_msrs_to_save)
2813 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2814 num_msrs_to_save * sizeof(u32)))
2816 if (copy_to_user(user_msr_list->indices
2817 + num_msrs_to_save * sizeof(u32),
2819 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2824 case KVM_CHECK_EXTENSION:
2826 * No extensions defined at present.
2830 case KVM_GET_VCPU_MMAP_SIZE:
2843 static struct file_operations kvm_chardev_ops = {
2844 .open = kvm_dev_open,
2845 .release = kvm_dev_release,
2846 .unlocked_ioctl = kvm_dev_ioctl,
2847 .compat_ioctl = kvm_dev_ioctl,
2850 static struct miscdevice kvm_dev = {
2857 * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2860 static void decache_vcpus_on_cpu(int cpu)
2863 struct kvm_vcpu *vcpu;
2866 spin_lock(&kvm_lock);
2867 list_for_each_entry(vm, &vm_list, vm_list)
2868 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2869 vcpu = vm->vcpus[i];
2873 * If the vcpu is locked, then it is running on some
2874 * other cpu and therefore it is not cached on the
2877 * If it's not locked, check the last cpu it executed
2880 if (mutex_trylock(&vcpu->mutex)) {
2881 if (vcpu->cpu == cpu) {
2882 kvm_arch_ops->vcpu_decache(vcpu);
2885 mutex_unlock(&vcpu->mutex);
2888 spin_unlock(&kvm_lock);
2891 static void hardware_enable(void *junk)
2893 int cpu = raw_smp_processor_id();
2895 if (cpu_isset(cpu, cpus_hardware_enabled))
2897 cpu_set(cpu, cpus_hardware_enabled);
2898 kvm_arch_ops->hardware_enable(NULL);
2901 static void hardware_disable(void *junk)
2903 int cpu = raw_smp_processor_id();
2905 if (!cpu_isset(cpu, cpus_hardware_enabled))
2907 cpu_clear(cpu, cpus_hardware_enabled);
2908 decache_vcpus_on_cpu(cpu);
2909 kvm_arch_ops->hardware_disable(NULL);
2912 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2919 case CPU_DYING_FROZEN:
2920 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2922 hardware_disable(NULL);
2924 case CPU_UP_CANCELED:
2925 case CPU_UP_CANCELED_FROZEN:
2926 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2928 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2931 case CPU_ONLINE_FROZEN:
2932 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2934 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2940 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2943 if (val == SYS_RESTART) {
2945 * Some (well, at least mine) BIOSes hang on reboot if
2948 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2949 on_each_cpu(hardware_disable, NULL, 0, 1);
2954 static struct notifier_block kvm_reboot_notifier = {
2955 .notifier_call = kvm_reboot,
2959 void kvm_io_bus_init(struct kvm_io_bus *bus)
2961 memset(bus, 0, sizeof(*bus));
2964 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2968 for (i = 0; i < bus->dev_count; i++) {
2969 struct kvm_io_device *pos = bus->devs[i];
2971 kvm_iodevice_destructor(pos);
2975 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
2979 for (i = 0; i < bus->dev_count; i++) {
2980 struct kvm_io_device *pos = bus->devs[i];
2982 if (pos->in_range(pos, addr))
2989 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2991 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2993 bus->devs[bus->dev_count++] = dev;
2996 static struct notifier_block kvm_cpu_notifier = {
2997 .notifier_call = kvm_cpu_hotplug,
2998 .priority = 20, /* must be > scheduler priority */
3001 static u64 stat_get(void *_offset)
3003 unsigned offset = (long)_offset;
3006 struct kvm_vcpu *vcpu;
3009 spin_lock(&kvm_lock);
3010 list_for_each_entry(kvm, &vm_list, vm_list)
3011 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3012 vcpu = kvm->vcpus[i];
3014 total += *(u32 *)((void *)vcpu + offset);
3016 spin_unlock(&kvm_lock);
3020 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3022 static __init void kvm_init_debug(void)
3024 struct kvm_stats_debugfs_item *p;
3026 debugfs_dir = debugfs_create_dir("kvm", NULL);
3027 for (p = debugfs_entries; p->name; ++p)
3028 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3029 (void *)(long)p->offset,
3033 static void kvm_exit_debug(void)
3035 struct kvm_stats_debugfs_item *p;
3037 for (p = debugfs_entries; p->name; ++p)
3038 debugfs_remove(p->dentry);
3039 debugfs_remove(debugfs_dir);
3042 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3044 hardware_disable(NULL);
3048 static int kvm_resume(struct sys_device *dev)
3050 hardware_enable(NULL);
3054 static struct sysdev_class kvm_sysdev_class = {
3055 set_kset_name("kvm"),
3056 .suspend = kvm_suspend,
3057 .resume = kvm_resume,
3060 static struct sys_device kvm_sysdev = {
3062 .cls = &kvm_sysdev_class,
3065 hpa_t bad_page_address;
3068 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3070 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3073 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3075 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3077 kvm_arch_ops->vcpu_load(vcpu, cpu);
3080 static void kvm_sched_out(struct preempt_notifier *pn,
3081 struct task_struct *next)
3083 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3085 kvm_arch_ops->vcpu_put(vcpu);
3088 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3089 struct module *module)
3095 printk(KERN_ERR "kvm: already loaded the other module\n");
3099 if (!ops->cpu_has_kvm_support()) {
3100 printk(KERN_ERR "kvm: no hardware support\n");
3103 if (ops->disabled_by_bios()) {
3104 printk(KERN_ERR "kvm: disabled by bios\n");
3110 r = kvm_arch_ops->hardware_setup();
3114 for_each_online_cpu(cpu) {
3115 smp_call_function_single(cpu,
3116 kvm_arch_ops->check_processor_compatibility,
3122 on_each_cpu(hardware_enable, NULL, 0, 1);
3123 r = register_cpu_notifier(&kvm_cpu_notifier);
3126 register_reboot_notifier(&kvm_reboot_notifier);
3128 r = sysdev_class_register(&kvm_sysdev_class);
3132 r = sysdev_register(&kvm_sysdev);
3136 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3137 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3138 __alignof__(struct kvm_vcpu), 0, 0);
3139 if (!kvm_vcpu_cache) {
3144 kvm_chardev_ops.owner = module;
3146 r = misc_register(&kvm_dev);
3148 printk (KERN_ERR "kvm: misc device register failed\n");
3152 kvm_preempt_ops.sched_in = kvm_sched_in;
3153 kvm_preempt_ops.sched_out = kvm_sched_out;
3158 kmem_cache_destroy(kvm_vcpu_cache);
3160 sysdev_unregister(&kvm_sysdev);
3162 sysdev_class_unregister(&kvm_sysdev_class);
3164 unregister_reboot_notifier(&kvm_reboot_notifier);
3165 unregister_cpu_notifier(&kvm_cpu_notifier);
3167 on_each_cpu(hardware_disable, NULL, 0, 1);
3169 kvm_arch_ops->hardware_unsetup();
3171 kvm_arch_ops = NULL;
3175 void kvm_exit_arch(void)
3177 misc_deregister(&kvm_dev);
3178 kmem_cache_destroy(kvm_vcpu_cache);
3179 sysdev_unregister(&kvm_sysdev);
3180 sysdev_class_unregister(&kvm_sysdev_class);
3181 unregister_reboot_notifier(&kvm_reboot_notifier);
3182 unregister_cpu_notifier(&kvm_cpu_notifier);
3183 on_each_cpu(hardware_disable, NULL, 0, 1);
3184 kvm_arch_ops->hardware_unsetup();
3185 kvm_arch_ops = NULL;
3188 static __init int kvm_init(void)
3190 static struct page *bad_page;
3193 r = kvm_mmu_module_init();
3199 kvm_init_msr_list();
3201 if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3206 bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3207 memset(__va(bad_page_address), 0, PAGE_SIZE);
3213 kvm_mmu_module_exit();
3218 static __exit void kvm_exit(void)
3221 __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3222 kvm_mmu_module_exit();
3225 module_init(kvm_init)
3226 module_exit(kvm_exit)
3228 EXPORT_SYMBOL_GPL(kvm_init_arch);
3229 EXPORT_SYMBOL_GPL(kvm_exit_arch);